a33a84ee63
[AMDGPU][NFC] Replace gfx940 and gfx941 with gfx942 in llvm/test gfx940 and gfx941 are no longer supported. This is one of a series of PRs to remove them from the code base. This PR uses gfx942 instead of gfx940 and gfx941 in the test RUN-lines (unless there is already a RUN-line for gfx942). The only notable difference in the test output is that gfx942 does not force the use of sc0 and sc1 on stores while gfx940 and gfx941 do (cf. https://reviews.llvm.org/D149986). For SWDEV-512631
7330 lines
256 KiB
LLVM
7330 lines
256 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_v2bf16_v4bf16__u_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v2bf16_v4bf16__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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> poison
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__0_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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__0_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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__0_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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__2_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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__2_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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__2_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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__3_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, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__3_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, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__3_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, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v2bf16_v4bf16__4_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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_u(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 poison>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v2, 16
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v3, 16
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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: 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 v0, v0, v3, 16
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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 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_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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 v0, v0, v3, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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: s_nop 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v1, v3, 16
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v1, v3, 16
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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: 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 v0, v1, v3, 16
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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, 0x7060302
|
|
; 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_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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, 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, v1, v3, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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, 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, v1, v3, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_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, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_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, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_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, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_6:
|
|
; 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, v1, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_6:
|
|
; 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, v1, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_6:
|
|
; 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, v1, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__7_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__7_7:
|
|
; 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 v0, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__7_7:
|
|
; 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 v0, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__7_7:
|
|
; 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 v0, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__u_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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__u_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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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 v0, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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 v0, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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 v0, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> zeroinitializer
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v1, 16
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v2, 16
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: 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 v0, v0, v2, 16
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_0(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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 v0, v0, v3, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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 v0, v0, v3, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 0>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__u_1:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__u_1:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__u_1:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__0_1:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__0_1:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__0_1:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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 v0, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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 v0, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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 v0, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v0
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__4_1:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__4_1:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__4_1:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: s_mov_b32 s4, 0x7060302
|
|
; 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, v1, s4
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 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, v0, v2, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 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, v0, v2, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_1(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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_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_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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_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_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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_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_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 1>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v1, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v1, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v1, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v1, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v1, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v1, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v1, v2, 16
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v1, v2, 16
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: 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 v0, v1, v2, 16
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_2(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 2>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__u_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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__u_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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__u_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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: s_mov_b32 s4, 0xffff
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v1
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 0xffff
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v1
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 0xffff
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v1
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: s_mov_b32 s4, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__2_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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__2_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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__2_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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__4_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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__4_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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__4_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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; 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, v2, s4
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__5_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, 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, v1, v2, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__5_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, 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, v1, v2, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_3(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v1
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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: 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, v1
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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: 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, v1
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 3>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v2bf16_v4bf16__u_4:
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__0_4:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__0_4:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__0_4:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__2_4:
|
|
; 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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__2_4:
|
|
; 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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__2_4:
|
|
; 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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__3_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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, s4, v1, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__3_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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, s4, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__3_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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, s0, v1, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX9-LABEL: v_shuffle_v2bf16_v4bf16__4_4:
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__5_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: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__5_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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__5_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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v0, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_4(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 4>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__u_5:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__u_5:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__u_5:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 poison, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: 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, v0, v1
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 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, v0, v2
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 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, v0, v2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 0, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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: s_mov_b32 s4, 0x7060302
|
|
; 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, v1, v0, s4
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 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: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 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: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 1, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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, 0xffff
|
|
; GFX900-NEXT: v_mov_b32_e32 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v2
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 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, v1, v2
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 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, v1, v2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 2, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; 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, v1, s4
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__3_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, 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, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__3_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, 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, v1, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 3, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__4_5:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__4_5:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__4_5:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 4, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX900-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__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, 0x7060302
|
|
; GFX942-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_5(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_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_mov_b32_e32 v2, 0
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_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_mov_b32_e32 v2, 0
|
|
; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v0
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_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_mov_b32_e32 v2, 0
|
|
; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v0
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 5>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__u_6:
|
|
; 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_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__u_6:
|
|
; 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_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__u_6:
|
|
; 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_lshlrev_b32_e32 v0, 16, v1
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 poison, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__0_6:
|
|
; 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, v2, v0, s4
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__0_6:
|
|
; 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: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__0_6:
|
|
; 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: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 0, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__1_6:
|
|
; 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 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v2, v0, 16
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__1_6:
|
|
; 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 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v3, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__1_6:
|
|
; 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 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v3, v0, 16
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 1, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__2_6:
|
|
; 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: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__2_6:
|
|
; 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: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__2_6:
|
|
; 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: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 2, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__3_6:
|
|
; 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 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_alignbit_b32 v0, v3, v1, 16
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__3_6:
|
|
; 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 v4, 0
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def v[2:3]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: v_alignbit_b32 v0, v3, v1, 16
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__3_6:
|
|
; 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 v4, 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def v[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_nop 0
|
|
; GFX942-NEXT: v_alignbit_b32 v0, v3, v1, 16
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 3, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__4_6:
|
|
; 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: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__4_6:
|
|
; 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: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__4_6:
|
|
; 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: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 4, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__5_6:
|
|
; 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, v1, v0, 16
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__5_6:
|
|
; 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, v1, v0, 16
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__5_6:
|
|
; 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, v1, v0, 16
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_6(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_6:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_6:
|
|
; 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_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_6:
|
|
; 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_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 6>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__u_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__u_7:
|
|
; 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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__u_7:
|
|
; 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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__u_7:
|
|
; 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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 poison, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__0_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__0_7:
|
|
; 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, v0, v2
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__0_7:
|
|
; 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, v0, v3
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__0_7:
|
|
; 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, v0, v3
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 0, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__1_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__1_7:
|
|
; 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 v3, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[1:2]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4
|
|
; GFX900-NEXT: global_store_dword v3, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__1_7:
|
|
; 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, v3, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__1_7:
|
|
; 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, v3, v0, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 1, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__2_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__2_7:
|
|
; 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 v4, 0
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def v[2:3]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v3
|
|
; GFX900-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__2_7:
|
|
; 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, v1, v3
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__2_7:
|
|
; 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, v1, v3
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 2, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__3_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__3_7:
|
|
; 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 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: global_store_dword v4, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__3_7:
|
|
; 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, v3, v1, s4
|
|
; GFX90A-NEXT: global_store_dword v4, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__3_7:
|
|
; 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, v3, v1, s2
|
|
; GFX942-NEXT: global_store_dword v4, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 3, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__4_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__4_7:
|
|
; 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, v0, v1
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__4_7:
|
|
; 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, v0, v1
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__4_7:
|
|
; 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, v0, v1
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 4, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__5_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__5_7:
|
|
; 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 v0, v1, v0, s4
|
|
; GFX900-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__5_7:
|
|
; 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 v0, v1, v0, s4
|
|
; GFX90A-NEXT: global_store_dword v2, v0, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__5_7:
|
|
; 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 v0, v1, v0, s2
|
|
; GFX942-NEXT: global_store_dword v2, v0, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @v_shuffle_v2bf16_v4bf16__6_7(ptr addrspace(1) inreg %ptr) {
|
|
; GFX900-LABEL: v_shuffle_v2bf16_v4bf16__6_7:
|
|
; 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_dword v2, v1, s[16:17]
|
|
; GFX900-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: v_shuffle_v2bf16_v4bf16__6_7:
|
|
; 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_dword v2, v1, s[16:17]
|
|
; GFX90A-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: v_shuffle_v2bf16_v4bf16__6_7:
|
|
; 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_dword v2, v1, 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 7>
|
|
store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_u() {
|
|
; GFX9-LABEL: s_shuffle_v2bf16_v4bf16__u_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s8
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> poison
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__0_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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__0_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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__0_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 s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__2_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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__2_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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__2_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 s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_u() {
|
|
; GFX9-LABEL: s_shuffle_v2bf16_v4bf16__4_u:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s8
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__5_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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__5_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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__5_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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_u() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 poison>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s7, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s7, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s3, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s7, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s7, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s3, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s7, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s7, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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, s3, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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_lshr_b32 s4, s5, 16
|
|
; GFX900-NEXT: s_lshr_b32 s5, s7, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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_lshr_b32 s4, s5, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s7, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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_lshr_b32 s0, s1, 16
|
|
; GFX942-NEXT: s_lshr_b32 s1, s3, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_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_lshr_b32 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_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_lshr_b32 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_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_lshr_b32 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_6:
|
|
; 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, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_6:
|
|
; 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, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_6:
|
|
; 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, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__7_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__7_7:
|
|
; 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, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__7_7:
|
|
; 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, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__7_7:
|
|
; 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, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 7, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> zeroinitializer
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_0() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 0>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__u_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__u_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__u_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_nop 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__0_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__0_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__0_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_nop 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__4_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__4_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__4_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_nop 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_1() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 1>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s4, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s4, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s0, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshl_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_2() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 2>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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 s5, s5, 16
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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 s5, s5, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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 s1, s1, 16
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s5, 16
|
|
; GFX900-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s5, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s1, 16
|
|
; GFX942-NEXT: s_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_3() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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, s5, 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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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, s5, 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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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, s1, 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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 3>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_4() {
|
|
; GFX9-LABEL: s_shuffle_v2bf16_v4bf16__u_4:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s8
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 poison, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__0_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__0_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__0_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_nop 0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 0, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s8, s4, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s8, s4, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_lshr_b32 s8, s0, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 1, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: s_mov_b32 s8, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: s_mov_b32 s8, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: s_mov_b32 s8, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 2, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 3, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_4() {
|
|
; GFX9-LABEL: s_shuffle_v2bf16_v4bf16__4_4:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ;;#ASMSTART
|
|
; GFX9-NEXT: ; use s8
|
|
; GFX9-NEXT: ;;#ASMEND
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = call <4 x bfloat> asm "; def $0", "=s"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> poison, <2 x i32> <i32 4, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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 s5, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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 s5, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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 s1, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_4() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 4>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__u_5:
|
|
; 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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__u_5:
|
|
; 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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__u_5:
|
|
; 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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 poison, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s5, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s5, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s1, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 0, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 1, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 2, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; def s[6:7]
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX900-NEXT: s_lshr_b32 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; def s[6:7]
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s6, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; def s[2:3]
|
|
; GFX942-NEXT: ;;#ASMEND
|
|
; GFX942-NEXT: s_lshr_b32 s0, s2, 16
|
|
; GFX942-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 3, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__4_5:
|
|
; 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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__4_5:
|
|
; 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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__4_5:
|
|
; 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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 4, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__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, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__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, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__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, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_5() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 5>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__u_6:
|
|
; 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 s8, s5, 16
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__u_6:
|
|
; 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 s8, s5, 16
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__u_6:
|
|
; 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 s8, s1, 16
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 poison, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__0_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__0_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__0_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 0, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__1_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__1_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__1_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 1, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__2_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__2_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__2_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 2, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__3_6:
|
|
; 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, s5, 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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__3_6:
|
|
; 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, s5, 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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__3_6:
|
|
; 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, s1, 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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 3, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__4_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__4_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__4_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 4, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__5_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__5_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__5_6:
|
|
; 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: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_6() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_6:
|
|
; 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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_6:
|
|
; 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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_6:
|
|
; 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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 6>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__u_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__u_7:
|
|
; 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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__u_7:
|
|
; 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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__u_7:
|
|
; 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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 poison, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__0_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__0_7:
|
|
; 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, s7, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__0_7:
|
|
; 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, s7, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__0_7:
|
|
; 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, s3, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 0, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__1_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__1_7:
|
|
; 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, s7, 16
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__1_7:
|
|
; 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, s7, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__1_7:
|
|
; 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, s3, 16
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 1, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__2_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__2_7:
|
|
; 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, s7, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__2_7:
|
|
; 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, s7, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__2_7:
|
|
; 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, s3, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 2, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__3_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__3_7:
|
|
; 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, s7, 16
|
|
; GFX900-NEXT: s_lshr_b32 s5, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__3_7:
|
|
; 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, s7, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s5, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__3_7:
|
|
; 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, s3, 16
|
|
; GFX942-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 3, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__4_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__4_7:
|
|
; 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, s5, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__4_7:
|
|
; 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, s5, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__4_7:
|
|
; 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, s1, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 4, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__5_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__5_7:
|
|
; 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, s5, 16
|
|
; GFX900-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX900-NEXT: ;;#ASMSTART
|
|
; GFX900-NEXT: ; use s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__5_7:
|
|
; 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, s5, 16
|
|
; GFX90A-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5
|
|
; GFX90A-NEXT: ;;#ASMSTART
|
|
; GFX90A-NEXT: ; use s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__5_7:
|
|
; 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, s1, 16
|
|
; GFX942-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1
|
|
; GFX942-NEXT: ;;#ASMSTART
|
|
; GFX942-NEXT: ; use s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 5, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
|
|
define void @s_shuffle_v2bf16_v4bf16__6_7() {
|
|
; GFX900-LABEL: s_shuffle_v2bf16_v4bf16__6_7:
|
|
; 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 s8
|
|
; GFX900-NEXT: ;;#ASMEND
|
|
; GFX900-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX90A-LABEL: s_shuffle_v2bf16_v4bf16__6_7:
|
|
; 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 s8
|
|
; GFX90A-NEXT: ;;#ASMEND
|
|
; GFX90A-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX942-LABEL: s_shuffle_v2bf16_v4bf16__6_7:
|
|
; 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 s8
|
|
; 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"()
|
|
%shuf = shufflevector <4 x bfloat> %vec0, <4 x bfloat> %vec1, <2 x i32> <i32 6, i32 7>
|
|
call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf)
|
|
ret void
|
|
}
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
|
; GFX90APLUS: {{.*}}
|