bbde792786
Allow widening up to 128-bit registers or if the new register class is at least as large as one of the existing register classes. This was artificially limiting. In particular this was doing the wrong thing with sequences involving copies between VGPRs and AV registers. Nearly all test changes are improvements. The coalescer does not just widen registers out of nowhere. If it's trying to "widen" a register, it's generally packing a register into an existing register tuple, or in a situation where the constraints imply the wider class anyway. 067a11015 addressed the allocation failure concern by rejecting coalescing if there are no available registers. The original change in a4e63ead4b didn't include a realistic testcase to judge if this is harmful for pressure. I would expect any issues from this to be of garden variety subreg handling issue. We could use more dynamic state information here if it really is an issue. I get the best results by removing this override completely. This is a smaller step for patch splitting purposes.
14242 lines
548 KiB
LLVM
14242 lines
548 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX900 %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a < %s | FileCheck -check-prefixes=GFX9,GFX90APLUS,GFX90A %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX90APLUS,GFX942 %s
|
|
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v4bf16_v3bf16__u_u_u_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> poison
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 poison, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v0, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v0, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 poison, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v4bf16_v3bf16__3_u_u_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 poison, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 poison, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v0, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v0, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v3, v3, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v2, v3, v3, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v1, s4, v0, 16
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, s4, v0, 16
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_alignbit_b32 v1, s0, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v2, s4, v0, 16
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v3, s4, v0, 16
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v3, s0, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 poison>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v2, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v2, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v1, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v1, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v2, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v3, s4, v1, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v3, s2, v1, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__u_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__u_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__u_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 0, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> zeroinitializer
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v0, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v0, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 0, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__3_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__3_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__3_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 0, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v0, v1, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v2, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v2, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 0, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v2, s4
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v0, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v0, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v2, v2, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v0, v3
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v0, v3
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v2, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v2, s4
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v2, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v2, v2, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v2, v1
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v3, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v3, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v2, 16, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v3, v3, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v2, v3, v3, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v0, v0, 16
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, v0, v0, 16
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_alignbit_b32 v1, v0, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v2, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v2, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v0, v1, 16
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, v0, v2, 16
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v1, v0, v2, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 0>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__u_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__u_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__u_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v1, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v1, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__3_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__3_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__3_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v0, v2, s4
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v0, v2, s2
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v2, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v2, v2, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v0, v3
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v0, v3
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v2, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v0, v3, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v0, v3, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v3, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v3, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v3, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v2, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_perm_b32 v2, v2, v3, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v2, v2, v3, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v2, v1
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v3, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v3, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v3, v3, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v2, v3, v3, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v3, v3, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v2, v3, v3, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: v_bfi_b32 v1, s4, v1, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: v_bfi_b32 v1, s2, v1, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v1, s4, v2, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v1, s2, v2, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v2, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v2, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 1>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__u_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__u_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__u_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v1, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v2, v1, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v2, v1, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__3_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__3_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__3_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v2, v1, v2, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v2, v1, v2, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v2, v1, v2, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v2, v2, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v0, v3
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v0, v3
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v2, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v2, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v2
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v1, v0, 16
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, v1, v0, 16
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_alignbit_b32 v1, v1, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v2, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v2, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v2, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v1, v2, 16
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, v1, v2, 16
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v1, v1, v2, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 2>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v4bf16_v3bf16__u_3_3_3:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v0, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v0, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v4bf16_v3bf16__3_3_3_3:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v0, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v0, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v1, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v1, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v2, 16, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v1, v0, 16
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v2, v0, v0, 16
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v3, v0, v0, 16
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v3, v0, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 3>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__u_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__u_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__u_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 poison, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v1
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 0, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v2, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v2, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 1, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v2
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 2, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__3_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__3_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__3_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 3, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v2, s4, v1, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v2, s2, v1, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v2, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_perm_b32 v1, v0, v1, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v0, v1, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v0, v1, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v2, v0, v0, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v0, v0, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v0, v0, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v0, v1
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v1, s4, v0, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v1, s2, v0, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v0, s4
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x7060302
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v2, v0, s4
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x7060302
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v2, v0, s2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v1, s4, v1, v2
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v1, s4, v1, v2
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v1, s2, v1, v2
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v3, v0
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 4>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__u_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__u_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__u_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__u_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 poison, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__0_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__0_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__0_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__0_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 0, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__1_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__1_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v2, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__1_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v3, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__1_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v3, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v3, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 1, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__2_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__2_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__2_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__2_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 2, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__3_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__3_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__3_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__3_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 3, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__4_5_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__4_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v1, v0, 16
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__4_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_alignbit_b32 v2, v1, v0, 16
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__4_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_alignbit_b32 v2, v1, v0, 16
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 5, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_u_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_u_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_u_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v1, s4
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_u_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v1, s2
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_0_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_0_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_0_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_0_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_1_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_1_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_1_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_1_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_2_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_2_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_2_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_2_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_3_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_3_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_3_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_3_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_4_5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_4_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_4_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_4_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v0
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 5, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_u_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX900-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX90A-NEXT: global_store_dwordx2 v2, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_u_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; GFX942-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_0_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v1, v2, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_0_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_1_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v1, v3, v0, 16
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_1_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v1, v3, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_2_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v1, v3, v1, s4
|
|
; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v1, v3, v1, s4
|
|
; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[0:1], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_2_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v1, v3, v1, s2
|
|
; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[0:1], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_3_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_perm_b32 v2, v1, v0, s4
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_perm_b32 v3, v1, v0, s4
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_3_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v3, v1, v0, s2
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v4bf16_v3bf16__5_5_4_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[0:1]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX900-NEXT: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v2, v1, v0, 16
|
|
; GFX900-NEXT: v_perm_b32 v1, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dwordx2 v3, v[1:2], s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX90A-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[0:1]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v3, v1, v0, 16
|
|
; GFX90A-NEXT: v_perm_b32 v2, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dwordx2 v4, v[2:3], s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v4bf16_v3bf16__5_5_4_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: s_mov_b32 s2, 0x5040100
|
|
; GFX942-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v3, v1, v0, 16
|
|
; GFX942-NEXT: v_perm_b32 v2, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dwordx2 v4, v[2:3], s[0:1]
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=v"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 5>
|
|
store <4 x bfloat> %shuf, ptr addrspace(1) %ptr, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_u_u_u() {
|
|
; GFX9-LABEL: s_shuffle_v4bf16_v3bf16__u_u_u_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s[8:9]
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> poison
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_u_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 poison, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_u_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_u_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 poison, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_u_u_u() {
|
|
; GFX9-LABEL: s_shuffle_v4bf16_v3bf16__3_u_u_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s[8:9]
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 poison, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_u_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 poison, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: s_mov_b32 s9, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: s_mov_b32 s9, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: s_mov_b32 s9, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s9, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s9, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s9, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s9, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s9, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s9, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_u() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_u:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s9, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_u:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s9, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_u:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s9, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_u() {
|
|
; GFX9-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; def s[8:9]
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_pack_ll_b32_b16 s8, s9, s9
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s[8:9]
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s9, s8
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s9, s8
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s9, s8
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__u_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_lshl_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__u_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_lshl_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__u_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_lshl_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 0, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s9, s8
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s9, s8
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s9, s8
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> zeroinitializer
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 0, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__3_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_lshl_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__3_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_lshl_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__3_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_lshl_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 0, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 0, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s8, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s8, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s8, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshl_b32 s9, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshl_b32 s9, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshl_b32 s9, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_0() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_0:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_0:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_0:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 0>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__u_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__u_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__u_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s8, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s8, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s9, s8
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s9, s8
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s9, s8
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__3_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__3_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__3_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s8, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s8, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s8, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s8, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: s_mov_b32 s9, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: s_mov_b32 s9, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: s_mov_b32 s9, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: s_mov_b32 s9, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: s_mov_b32 s9, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: s_mov_b32 s9, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_1() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_1:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_1:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_1:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 1>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__u_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: s_lshl_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__u_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: s_lshl_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__u_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: s_lshl_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s9, s8
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s9, s8
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s9, s8
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__3_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: s_lshl_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__3_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: s_lshl_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__3_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: s_lshl_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s8, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s8, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s8, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshl_b32 s9, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshl_b32 s9, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshl_b32 s9, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_2() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_2:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_2:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_2:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 2>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_3_3_3() {
|
|
; GFX9-LABEL: s_shuffle_v4bf16_v3bf16__u_3_3_3:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s[8:9]
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 poison, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_3_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 0, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_3_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 1, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_3_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 2, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_3_3_3() {
|
|
; GFX9-LABEL: s_shuffle_v4bf16_v3bf16__3_3_3_3:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s[8:9]
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_3_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s6
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s6
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s2
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s6
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s6
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s2
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s6
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s6
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s2
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s2, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshl_b32 s9, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshl_b32 s9, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshl_b32 s9, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s6
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s6
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s2
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_3() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_3:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s6, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_3:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s6, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_3:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s2, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s2, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 3>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__u_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__u_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__u_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s8, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 poison, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 0, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 1, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 2, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__3_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__3_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__3_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s8, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 3, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s9, s8
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s9, s8
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s9, s8
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s9, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s9, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s9, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_4() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_4:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s9, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_4:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s9, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_4:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s9, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 4>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__u_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__u_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: s_lshl_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__u_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: s_lshl_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__u_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: s_lshl_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 poison, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__0_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__0_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__0_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__0_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 0, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__1_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__1_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__1_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__1_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 1, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__2_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__2_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s7
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__2_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s7
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__2_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s3
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 2, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__3_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__3_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__3_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__3_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 3, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__4_5_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__4_5_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__4_5_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__4_5_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 4, i32 5, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_u_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_u_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_u_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_u_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 poison, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_0_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_0_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_0_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_0_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 0, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_1_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_1_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_1_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_1_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 1, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_2_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_2_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_2_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_2_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 2, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_3_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_3_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_3_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_3_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 3, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_4_5_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_4_5_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_4_5_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_4_5_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 4, i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_u_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshl_b32 s9, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshl_b32 s9, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_u_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshl_b32 s9, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 poison, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_0_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s7
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s7
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_0_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s3
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 0, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_1_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s7
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s7
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_1_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s3
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 1, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_2_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s5, s7
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s5, s7
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_2_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s1, s3
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 2, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_3_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_3_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 3, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v4bf16_v3bf16__5_5_4_5() {
|
|
; GFX900-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_5:
|
|
; GFX900: ; %bb.0:
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[4:5]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s[8:9]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_5:
|
|
; GFX90A: ; %bb.0:
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[4:5]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s9, s4, s5
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s[8:9]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v4bf16_v3bf16__5_5_4_5:
|
|
; GFX942: ; %bb.0:
|
|
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[0:1]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s9, s0, s1
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s[8:9]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%vec1 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%extract3 = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%extract31 = shufflevector <4 x bfloat> %vec1, <4 x bfloat> poison, <3 x i32> <i32 0, i32 1, i32 2>
|
|
%shuf = shufflevector <3 x bfloat> %extract3, <3 x bfloat> %extract31, <4 x i32> <i32 5, i32 5, i32 4, i32 5>
|
|
call void asm sideeffect "; use $0", "{s[8:9]}"(<4 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
|
; GFX90APLUS: {{.*}}
|