2502e3b7ba
Convert "denormal-fp-math" and "denormal-fp-math-f32" into a first class denormal_fpenv attribute. Previously the query for the effective denormal mode involved two string attribute queries with parsing. I'm introducing more uses of this, so it makes sense to convert this to a more efficient encoding. The old representation was also awkward since it was split across two separate attributes. The new encoding just stores the default and float modes as bitfields, largely avoiding the need to consider if the other mode is set. The syntax in the common cases looks like this: `denormal_fpenv(preservesign,preservesign)` `denormal_fpenv(float: preservesign,preservesign)` `denormal_fpenv(dynamic,dynamic float: preservesign,preservesign)` I wasn't sure about reusing the float type name instead of adding a new keyword. It's parsed as a type but only accepts float. I'm also debating switching the name to subnormal to match the current preferred IEEE terminology (also used by nofpclass and other contexts). This has a behavior change when using the command flag debug options to set the denormal mode. The behavior of the flag ignored functions with an explicit attribute set, per the default and f32 version. Now that these are one attribute, the flag logic can't distinguish which of the two components were explicitly set on the function. Only one test appeared to rely on this behavior, so I just avoided using the flags in it. This also does not perform all the code cleanups this enables. In particular the attributor handling could be cleaned up. I also guessed at how to support this in MLIR. I followed MemoryEffects as a reference; it appears bitfields are expanded into arguments to attributes, so the representation there is a bit uglier with the 2 2-element fields flattened into 4 arguments.
4046 lines
160 KiB
LLVM
4046 lines
160 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -enable-var-scope -check-prefixes=VI %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=kaveri < %s | FileCheck -enable-var-scope -check-prefixes=CI %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-TRUE16 %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-FAKE16 %s
|
|
|
|
declare half @llvm.fabs.f16(half) #0
|
|
declare half @llvm.canonicalize.f16(half) #0
|
|
declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #0
|
|
declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>) #0
|
|
declare <3 x half> @llvm.canonicalize.v3f16(<3 x half>) #0
|
|
declare <4 x half> @llvm.canonicalize.v4f16(<4 x half>) #0
|
|
declare <6 x half> @llvm.canonicalize.v6f16(<6 x half>) #0
|
|
declare <8 x half> @llvm.canonicalize.v8f16(<8 x half>) #0
|
|
declare <12 x half> @llvm.canonicalize.v12f16(<12 x half>) #0
|
|
declare <16 x half> @llvm.canonicalize.v16f16(<16 x half>) #0
|
|
declare <32 x half> @llvm.canonicalize.v32f16(<32 x half>) #0
|
|
declare <64 x half> @llvm.canonicalize.v64f16(<64 x half>) #0
|
|
declare i32 @llvm.amdgcn.workitem.id.x() #0
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_undef_value_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_undef_value_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_undef_value_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_undef_value_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_undef_value_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_undef_value_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half undef)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_var_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_load_ushort v0, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: flat_store_short v[0:1], v0
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX9-NEXT: global_store_short v[0:1], v0, off
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v[0:1], v0, off
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v0, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v[0:1], v0, off
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = load half, ptr addrspace(1) %out
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val)
|
|
store half %canonicalized, ptr addrspace(1) poison
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @s_test_canonicalize_var_f16(ptr addrspace(1) %out, i16 zeroext %val.arg) #1 {
|
|
; VI-LABEL: s_test_canonicalize_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dword s2, s[4:5], 0x2c
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_e64 v2, s2, s2
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: s_test_canonicalize_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e64 v1, s2, s2
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: s_test_canonicalize_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dword s0, s[4:5], 0xb
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, s0
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: s_test_canonicalize_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_clause 0x1
|
|
; GFX11-TRUE16-NEXT: s_load_b32 s2, s[4:5], 0x2c
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e64 v0.l, s2, s2
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: s_test_canonicalize_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_clause 0x1
|
|
; GFX11-FAKE16-NEXT: s_load_b32 s2, s[4:5], 0x2c
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e64 v1, s2, s2
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = bitcast i16 %val.arg to half
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_build_vector_v2f16(half %lo, half %hi) #1 {
|
|
; VI-LABEL: v_test_canonicalize_build_vector_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_build_vector_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_build_vector_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_build_vector_v2f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-TRUE16-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_build_vector_v2f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%ins0 = insertelement <2 x half> poison, half %lo, i32 0
|
|
%ins1 = insertelement <2 x half> %ins0, half %hi, i32 1
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %ins1)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_fabs_var_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_fabs_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_load_ushort v2, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_e64 v2, |v2|, |v2|
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_fabs_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e64 v1, |v1|, |v1|
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_fabs_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_fabs_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e64 v0.l, |v0.l|, |v0.l|
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_fabs_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e64 v1, |v1|, |v1|
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = load half, ptr addrspace(1) %out
|
|
%val.fabs = call half @llvm.fabs.f16(half %val)
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val.fabs)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_load_ushort v2, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_e64 v2, -|v2|, -|v2|
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e64 v1, -|v1|, -|v1|
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v0, -|v0|
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e64 v0.l, -|v0.l|, -|v0.l|
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e64 v1, -|v1|, -|v1|
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = load half, ptr addrspace(1) %out
|
|
%val.fabs = call half @llvm.fabs.f16(half %val)
|
|
%val.fabs.fneg = fneg half %val.fabs
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val.fabs.fneg)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_fneg_var_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_fneg_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_load_ushort v2, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_e64 v2, -v2, -v2
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_fneg_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e64 v1, -v1, -v1
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_fneg_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_fneg_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e64 v0.l, -v0.l, -v0.l
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_fneg_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e64 v1, -v1, -v1
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = load half, ptr addrspace(1) %out
|
|
%val.fneg = fneg half %val
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val.fneg)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_var_f16(ptr addrspace(1) %out) #2 {
|
|
; VI-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_load_ushort v2, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_mul_f16_e32 v2, -1.0, v2
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e64 v1, -v1, -v1
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e64 v0.l, -v0.l, -v0.l
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e64 v1, -v1, -v1
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = load half, ptr addrspace(1) %out
|
|
%val.fneg = fneg half %val
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val.fneg)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_fabs_var_f16(ptr addrspace(1) %out) #2 {
|
|
; VI-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_load_ushort v2, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_mul_f16_e64 v2, -1.0, |v2|
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e64 v1, -|v1|, -|v1|
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v0, -|v0|
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e64 v0.l, -|v0.l|, -|v0.l|
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e64 v1, -|v1|, -|v1|
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%val = load half, ptr addrspace(1) %out
|
|
%val.fabs = call half @llvm.fabs.f16(half %val)
|
|
%val.fabs.fneg = fneg half %val.fabs
|
|
%canonicalized = call half @llvm.canonicalize.f16(half %val.fabs.fneg)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_p0_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_p0_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_p0_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_p0_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_p0_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_p0_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v0, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0.0)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_n0_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_n0_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0xffff8000
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_n0_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff8000
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_n0_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x8000
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_n0_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x8000
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_n0_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0xffff8000
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half -0.0)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_p1_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_p1_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x3c00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_p1_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3c00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_p1_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3c00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_p1_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3c00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_p1_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x3c00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 1.0)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_n1_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_n1_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0xffffbc00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_n1_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffffbc00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_n1_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0xbc00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_n1_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0xbc00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_n1_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0xffffbc00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half -1.0)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_literal_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_literal_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x4c00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_literal_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x4c00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_literal_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x4c00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_literal_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x4c00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_literal_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x4c00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 16.0)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal0_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3ff
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x3ff
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH03FF)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_f16(ptr addrspace(1) %out) #3 {
|
|
; VI-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3ff
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x3ff
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH03FF)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal1_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0xffff83ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff83ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x83ff
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0xffff83ff
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH83FF)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_f16(ptr addrspace(1) %out) #3 {
|
|
; VI-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0xffff83ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff83ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x83ff
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0xffff83ff
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH83FF)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_qnan_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_qnan_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7c00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_qnan_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7c00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_qnan_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7c00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_qnan_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7c00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_qnan_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7c00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH7C00)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7e00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half bitcast (i16 -1 to half))
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7e00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half bitcast (i16 -2 to half))
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan0_value_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan0_value_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan0_value_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan0_value_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_snan0_value_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7e00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_snan0_value_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH7C01)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan1_value_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan1_value_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan1_value_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan1_value_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_snan1_value_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7e00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_snan1_value_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xH7DFF)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan2_value_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan2_value_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan2_value_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan2_value_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_snan2_value_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7e00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_snan2_value_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xHFDFF)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan3_value_f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan3_value_f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_short v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan3_value_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan3_value_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-TRUE16-LABEL: test_fold_canonicalize_snan3_value_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7e00
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; GFX11-TRUE16-NEXT: s_endpgm
|
|
;
|
|
; GFX11-FAKE16-LABEL: test_fold_canonicalize_snan3_value_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e00
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
|
|
; GFX11-FAKE16-NEXT: s_endpgm
|
|
%canonicalized = call half @llvm.canonicalize.f16(half 0xHFC01)
|
|
store half %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_var_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v1, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s6, 0
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
|
|
; CI-NEXT: v_mov_b32_e32 v1, 0
|
|
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_load_b32 v0, v0, s[0:1]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: global_store_b32 v1, v0, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
|
|
%val = load <2 x half>, ptr addrspace(1) %gep
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_fabs_var_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v1, |v0|, |v0| dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e64 v0, |v0|, |v0|
|
|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_fabs_var_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_fabs_var_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s6, 0
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
|
|
; CI-NEXT: v_mov_b32_e32 v1, 0
|
|
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v1, |v1|
|
|
; CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_fabs_var_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_load_b32 v0, v0, s[0:1]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: global_store_b32 v1, v0, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
|
|
%val = load <2 x half>, ptr addrspace(1) %gep
|
|
%val.fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %val)
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val.fabs)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v1, -|v0|, -|v0| dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e64 v0, -|v0|, -|v0|
|
|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0 neg_lo:[1,1] neg_hi:[1,1]
|
|
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s6, 0
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
|
|
; CI-NEXT: v_mov_b32_e32 v1, 0
|
|
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_or_b32_e32 v0, 0x80008000, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
|
|
; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_load_b32 v0, v0, s[0:1]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0 neg_lo:[1,1] neg_hi:[1,1]
|
|
; GFX11-NEXT: global_store_b32 v1, v0, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
|
|
%val = load <2 x half>, ptr addrspace(1) %gep
|
|
%val.fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %val)
|
|
%val.fabs.fneg = fneg <2 x half> %val.fabs
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val.fabs.fneg)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @v_test_canonicalize_fneg_var_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: v_test_canonicalize_fneg_var_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
|
|
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
; VI-NEXT: flat_load_dword v0, v[0:1]
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v1, -v0, -v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e64 v0, -v0, -v0
|
|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_fneg_var_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0 neg_lo:[1,1] neg_hi:[1,1]
|
|
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_fneg_var_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s6, 0
|
|
; CI-NEXT: s_mov_b32 s7, s3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
|
|
; CI-NEXT: v_mov_b32_e32 v1, 0
|
|
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_fneg_var_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_load_b32 v0, v0, s[0:1]
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0 neg_lo:[1,1] neg_hi:[1,1]
|
|
; GFX11-NEXT: global_store_b32 v1, v0, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
|
|
%val = load <2 x half>, ptr addrspace(1) %gep
|
|
%fneg.val = fneg <2 x half> %val
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %fneg.val)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @s_test_canonicalize_var_v2f16(ptr addrspace(1) %out, i32 zeroext %val.arg) #1 {
|
|
; VI-LABEL: s_test_canonicalize_var_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dword s2, s[4:5], 0x2c
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: s_lshr_b32 s3, s2, 16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s3
|
|
; VI-NEXT: v_max_f16_e64 v0, s2, s2
|
|
; VI-NEXT: v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: s_test_canonicalize_var_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v1, s2, s2
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: s_test_canonicalize_var_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dword s0, s[4:5], 0xb
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_lshr_b32 s1, s0, 16
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, s1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, s0
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: s_test_canonicalize_var_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_clause 0x1
|
|
; GFX11-NEXT: s_load_b32 s2, s[4:5], 0x2c
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v1, s2, s2
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%val = bitcast i32 %val.arg to <2 x half>
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_p0_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_p0_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_p0_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_p0_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_p0_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v0, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> zeroinitializer)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_n0_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_n0_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x80008000
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_n0_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x80008000
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_n0_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x80008000
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_n0_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x80008000
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half -0.0, half -0.0>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_p1_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_p1_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x3c003c00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_p1_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3c003c00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_p1_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3c003c00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_p1_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x3c003c00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 1.0, half 1.0>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_n1_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_n1_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0xbc00bc00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_n1_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0xbc00bc00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_n1_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0xbc00bc00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_n1_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0xbc00bc00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half -1.0, half -1.0>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_literal_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_literal_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x4c004c00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_literal_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x4c004c00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_literal_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x4c004c00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_literal_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x4c004c00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 16.0, half 16.0>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal0_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff03ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff03ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff03ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x3ff03ff
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH03FF, half 0xH03FF>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_v2f16(ptr addrspace(1) %out) #3 {
|
|
; VI-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff03ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff03ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff03ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x3ff03ff
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH03FF, half 0xH03FF>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal1_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x83ff83ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x83ff83ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff83ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x83ff83ff
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH83FF, half 0xH83FF>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_v2f16(ptr addrspace(1) %out) #3 {
|
|
; VI-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x83ff83ff
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x83ff83ff
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff83ff
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x83ff83ff
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH83FF, half 0xH83FF>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_qnan_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_qnan_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7c007c00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_qnan_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7c007c00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_qnan_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7c007c00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_qnan_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7c007c00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7C00, half 0xH7C00>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e007e00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> bitcast (i32 -1 to <2 x half>))
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e007e00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half bitcast (i16 -2 to half), half bitcast (i16 -2 to half)>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan0_value_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan0_value_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan0_value_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan0_value_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_snan0_value_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e007e00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7C01, half 0xH7C01>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan1_value_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan1_value_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan1_value_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan1_value_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_snan1_value_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e007e00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7DFF, half 0xH7DFF>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan2_value_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan2_value_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan2_value_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan2_value_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_snan2_value_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e007e00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xHFDFF, half 0xHFDFF>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_fold_canonicalize_snan3_value_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: test_fold_canonicalize_snan3_value_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: test_fold_canonicalize_snan3_value_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: test_fold_canonicalize_snan3_value_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: test_fold_canonicalize_snan3_value_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x7e007e00
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xHFC01, half 0xHFC01>)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define <3 x half> @v_test_canonicalize_var_v3f16(<3 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v3f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v2, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v2
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v3f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v3f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v2
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v3f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> %val)
|
|
ret <3 x half> %canonicalized
|
|
}
|
|
|
|
define <4 x half> @v_test_canonicalize_var_v4f16(<4 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v4f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v2, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v3, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v3
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v2
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v4f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v4f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v2
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v3
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v4f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %val)
|
|
ret <4 x half> %canonicalized
|
|
}
|
|
|
|
define amdgpu_kernel void @s_test_canonicalize_undef_v2f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: s_test_canonicalize_undef_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v2, 0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, s0
|
|
; VI-NEXT: v_mov_b32_e32 v1, s1
|
|
; VI-NEXT: flat_store_dword v[0:1], v2
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: s_test_canonicalize_undef_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dword v0, v0, s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: s_test_canonicalize_undef_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: s_test_canonicalize_undef_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b32 v0, v0, s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> undef)
|
|
store <2 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_reg_undef_v2f16(half %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_reg_undef_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_reg_undef_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_reg_undef_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_reg_undef_v2f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.h, 0
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_reg_undef_v2f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, 0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <2 x half> poison, half %val, i32 0
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_undef_reg_v2f16(half %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_undef_reg_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v0, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_undef_reg_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_sdwa v0, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_undef_reg_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_undef_reg_v2f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.h, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_undef_reg_v2f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <2 x half> poison, half %val, i32 1
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_undef_lo_imm_hi_v2f16() #1 {
|
|
; VI-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_bfrev_b32_e32 v0, 60
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_bfrev_b32_e32 v0, 60
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_bfrev_b32_e32 v0, 60
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_bfrev_b32_e32 v0, 60
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <2 x half> undef, half 1.0, i32 1
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_imm_lo_undef_hi_v2f16() #1 {
|
|
; VI-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, 0x3c00
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0x3c00
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x3c00
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0x3c00
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <2 x half> undef, half 1.0, i32 0
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_undef_lo_k_hi_v2f16() #1 {
|
|
; VI-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_bfrev_b32_e32 v0, 50
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_bfrev_b32_e32 v0, 50
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_bfrev_b32_e32 v0, 50
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_bfrev_b32_e32 v0, 50
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <2 x half> undef, half 16.0, i32 1
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_k_lo_undef_hi_v2f16() #1 {
|
|
; VI-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v0, 0x4c00
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0x4c00
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0x4c00
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0x4c00
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <2 x half> undef, half 16.0, i32 0
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_reg_k_v2f16(half %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_reg_k_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, 2.0, v0
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_reg_k_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 2.0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_reg_k_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, 2.0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_reg_k_v2f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.h, 0x4000
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_reg_k_v2f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, 2.0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = insertelement <2 x half> poison, half %val, i32 0
|
|
%vec1 = insertelement <2 x half> %vec0, half 2.0, i32 1
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec1)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define <2 x half> @v_test_canonicalize_k_reg_v2f16(half %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_k_reg_v2f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v0, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v0, 0x4000, v0
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_k_reg_v2f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX9-NEXT: v_pack_b32_f16 v0, 2.0, v0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_k_reg_v2f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, 2.0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; CI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_k_reg_v2f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.h, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x4000
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_k_reg_v2f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, 2.0, v0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = insertelement <2 x half> poison, half 2.0, i32 0
|
|
%vec1 = insertelement <2 x half> %vec0, half %val, i32 1
|
|
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec1)
|
|
ret <2 x half> %canonicalized
|
|
}
|
|
|
|
define amdgpu_kernel void @s_test_canonicalize_undef_v4f16(ptr addrspace(1) %out) #1 {
|
|
; VI-LABEL: s_test_canonicalize_undef_v4f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; VI-NEXT: v_mov_b32_e32 v0, 0
|
|
; VI-NEXT: v_mov_b32_e32 v1, v0
|
|
; VI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; VI-NEXT: v_mov_b32_e32 v3, s1
|
|
; VI-NEXT: v_mov_b32_e32 v2, s0
|
|
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
|
|
; VI-NEXT: s_endpgm
|
|
;
|
|
; GFX9-LABEL: s_test_canonicalize_undef_v4f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: global_store_dwordx2 v0, v[0:1], s[0:1]
|
|
; GFX9-NEXT: s_endpgm
|
|
;
|
|
; CI-LABEL: s_test_canonicalize_undef_v4f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
|
|
; CI-NEXT: v_mov_b32_e32 v0, 0
|
|
; CI-NEXT: s_mov_b32 s3, 0xf000
|
|
; CI-NEXT: s_mov_b32 s2, -1
|
|
; CI-NEXT: v_mov_b32_e32 v1, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
|
|
; CI-NEXT: s_endpgm
|
|
;
|
|
; GFX11-LABEL: s_test_canonicalize_undef_v4f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
|
|
; GFX11-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
|
|
; GFX11-NEXT: v_mov_b32_e32 v1, v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: global_store_b64 v0, v[0:1], s[0:1]
|
|
; GFX11-NEXT: s_endpgm
|
|
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> undef)
|
|
store <4 x half> %canonicalized, ptr addrspace(1) %out
|
|
ret void
|
|
}
|
|
|
|
define <4 x half> @v_test_canonicalize_reg_undef_undef_undef_v4f16(half %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, 0x7e000000, v0
|
|
; VI-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v1
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v2
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v2
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.h, 0
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, 0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec = insertelement <4 x half> poison, half %val, i32 0
|
|
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec)
|
|
ret <4 x half> %canonicalized
|
|
}
|
|
|
|
define <4 x half> @v_test_canonicalize_reg_reg_undef_undef_v4f16(half %val0, half %val1) #1 {
|
|
; VI-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
|
|
; CI-NEXT: v_or_b32_e32 v1, v2, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v3
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v2
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
|
|
; GFX11-TRUE16-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100
|
|
; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, 0x7e007e00
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = insertelement <4 x half> poison, half %val0, i32 0
|
|
%vec1 = insertelement <4 x half> %vec0, half %val1, i32 1
|
|
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec1)
|
|
ret <4 x half> %canonicalized
|
|
}
|
|
|
|
define <4 x half> @v_test_canonicalize_reg_undef_reg_reg_v4f16(half %val0, half %val1, half %val2) #1 {
|
|
; VI-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_max_f16_sdwa v2, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, 0x7e000000, v0
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v2
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
|
|
; GFX9-NEXT: v_perm_b32 v1, v2, v1, s4
|
|
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, 0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v3
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v2
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
|
|
; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.h, 0
|
|
; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v0.l
|
|
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
|
|
; GFX11-TRUE16-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; GFX11-FAKE16-NEXT: v_perm_b32 v1, v2, v1, 0x5040100
|
|
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
|
|
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, 0
|
|
; GFX11-FAKE16-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%vec0 = insertelement <4 x half> poison, half %val0, i32 0
|
|
%vec1 = insertelement <4 x half> %vec0, half %val1, i32 2
|
|
%vec2 = insertelement <4 x half> %vec1, half %val2, i32 3
|
|
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec2)
|
|
ret <4 x half> %canonicalized
|
|
}
|
|
|
|
define <6 x half> @v_test_canonicalize_var_v6f16(<6 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v6f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v3, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v4, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v5, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v2, v2, v2
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v5
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v4
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v3
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v6f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v6f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_lshrrev_b32_e32 v4, 16, v2
|
|
; CI-NEXT: v_lshrrev_b32_e32 v5, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v5
|
|
; CI-NEXT: v_or_b32_e32 v2, v2, v3
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v6f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <6 x half> @llvm.canonicalize.v6f16(<6 x half> %val)
|
|
ret <6 x half> %canonicalized
|
|
}
|
|
|
|
define <8 x half> @v_test_canonicalize_var_v8f16(<8 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v8f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v4, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v5, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v6, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v7, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v3, v3, v3
|
|
; VI-NEXT: v_max_f16_e32 v2, v2, v2
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v7
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v6
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v5
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v4
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v8f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v8f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v4, 16, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v5, 16, v3
|
|
; CI-NEXT: v_lshrrev_b32_e32 v6, 16, v1
|
|
; CI-NEXT: v_lshrrev_b32_e32 v7, 16, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_lshlrev_b32_e32 v4, 16, v4
|
|
; CI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
|
|
; CI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v4
|
|
; CI-NEXT: v_lshlrev_b32_e32 v4, 16, v7
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v6
|
|
; CI-NEXT: v_or_b32_e32 v2, v2, v4
|
|
; CI-NEXT: v_or_b32_e32 v3, v3, v5
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v8f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX11-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <8 x half> @llvm.canonicalize.v8f16(<8 x half> %val)
|
|
ret <8 x half> %canonicalized
|
|
}
|
|
|
|
define <12 x half> @v_test_canonicalize_var_v12f16(<12 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v12f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v6, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v7, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v8, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v9, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v10, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v11, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v5, v5, v5
|
|
; VI-NEXT: v_max_f16_e32 v4, v4, v4
|
|
; VI-NEXT: v_max_f16_e32 v3, v3, v3
|
|
; VI-NEXT: v_max_f16_e32 v2, v2, v2
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v11
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v10
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v9
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v8
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v7
|
|
; VI-NEXT: v_or_b32_e32 v5, v5, v6
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v12f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v12f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v7, 16, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v9, 16, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; CI-NEXT: v_lshrrev_b32_e32 v6, 16, v5
|
|
; CI-NEXT: v_lshrrev_b32_e32 v8, 16, v1
|
|
; CI-NEXT: v_lshrrev_b32_e32 v10, 16, v3
|
|
; CI-NEXT: v_lshrrev_b32_e32 v11, 16, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v7
|
|
; CI-NEXT: v_lshlrev_b32_e32 v7, 16, v9
|
|
; CI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
|
|
; CI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
|
|
; CI-NEXT: v_lshlrev_b32_e32 v8, 16, v8
|
|
; CI-NEXT: v_or_b32_e32 v2, v2, v7
|
|
; CI-NEXT: v_lshlrev_b32_e32 v7, 16, v11
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v8
|
|
; CI-NEXT: v_or_b32_e32 v3, v3, v10
|
|
; CI-NEXT: v_or_b32_e32 v4, v4, v7
|
|
; CI-NEXT: v_or_b32_e32 v5, v5, v6
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v12f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX11-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX11-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX11-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <12 x half> @llvm.canonicalize.v12f16(<12 x half> %val)
|
|
ret <12 x half> %canonicalized
|
|
}
|
|
|
|
define <16 x half> @v_test_canonicalize_var_v16f16(<16 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v16f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v8, v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v9, v6, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v10, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v11, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v12, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v13, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v14, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v15, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v7, v7, v7
|
|
; VI-NEXT: v_max_f16_e32 v6, v6, v6
|
|
; VI-NEXT: v_max_f16_e32 v5, v5, v5
|
|
; VI-NEXT: v_max_f16_e32 v4, v4, v4
|
|
; VI-NEXT: v_max_f16_e32 v3, v3, v3
|
|
; VI-NEXT: v_max_f16_e32 v2, v2, v2
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v15
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v14
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v13
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v12
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v11
|
|
; VI-NEXT: v_or_b32_e32 v5, v5, v10
|
|
; VI-NEXT: v_or_b32_e32 v6, v6, v9
|
|
; VI-NEXT: v_or_b32_e32 v7, v7, v8
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v16f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX9-NEXT: v_pk_max_f16 v6, v6, v6
|
|
; GFX9-NEXT: v_pk_max_f16 v7, v7, v7
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v16f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v8, 16, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v10, 16, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; CI-NEXT: v_lshrrev_b32_e32 v15, 16, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v9, 16, v1
|
|
; CI-NEXT: v_lshrrev_b32_e32 v11, 16, v7
|
|
; CI-NEXT: v_lshrrev_b32_e32 v12, 16, v3
|
|
; CI-NEXT: v_lshrrev_b32_e32 v13, 16, v5
|
|
; CI-NEXT: v_lshrrev_b32_e32 v14, 16, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v15, v15
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v13, v13
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v14, v14
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v12, v12
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v15, v15
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v13, v13
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v14, v14
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v12, v12
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v8, 16, v8
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v8
|
|
; CI-NEXT: v_lshlrev_b32_e32 v8, 16, v10
|
|
; CI-NEXT: v_or_b32_e32 v2, v2, v8
|
|
; CI-NEXT: v_lshlrev_b32_e32 v8, 16, v15
|
|
; CI-NEXT: v_lshlrev_b32_e32 v11, 16, v11
|
|
; CI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
|
|
; CI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
|
|
; CI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
|
|
; CI-NEXT: v_or_b32_e32 v4, v4, v8
|
|
; CI-NEXT: v_lshlrev_b32_e32 v8, 16, v14
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v9
|
|
; CI-NEXT: v_or_b32_e32 v3, v3, v12
|
|
; CI-NEXT: v_or_b32_e32 v5, v5, v13
|
|
; CI-NEXT: v_or_b32_e32 v6, v6, v8
|
|
; CI-NEXT: v_or_b32_e32 v7, v7, v11
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v16f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX11-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX11-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX11-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX11-NEXT: v_pk_max_f16 v6, v6, v6
|
|
; GFX11-NEXT: v_pk_max_f16 v7, v7, v7
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <16 x half> @llvm.canonicalize.v16f16(<16 x half> %val)
|
|
ret <16 x half> %canonicalized
|
|
}
|
|
|
|
define <32 x half> @v_test_canonicalize_var_v32f16(<32 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v32f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v19, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v2, v2, v2
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v3, v3, v3
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v4, v4, v4
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v5, v5, v5
|
|
; VI-NEXT: v_or_b32_e32 v5, v5, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v6, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v6, v6, v6
|
|
; VI-NEXT: v_or_b32_e32 v6, v6, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v7, v7, v7
|
|
; VI-NEXT: v_or_b32_e32 v7, v7, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v8, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v8, v8, v8
|
|
; VI-NEXT: v_or_b32_e32 v8, v8, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v9, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v9, v9, v9
|
|
; VI-NEXT: v_or_b32_e32 v9, v9, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v10, v10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v10, v10, v10
|
|
; VI-NEXT: v_or_b32_e32 v10, v10, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v11, v11 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v11, v11, v11
|
|
; VI-NEXT: v_max_f16_sdwa v16, v15, v15 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v17, v14, v14 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_sdwa v18, v13, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_or_b32_e32 v11, v11, v19
|
|
; VI-NEXT: v_max_f16_sdwa v19, v12, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v15, v15, v15
|
|
; VI-NEXT: v_max_f16_e32 v14, v14, v14
|
|
; VI-NEXT: v_max_f16_e32 v13, v13, v13
|
|
; VI-NEXT: v_max_f16_e32 v12, v12, v12
|
|
; VI-NEXT: v_or_b32_e32 v12, v12, v19
|
|
; VI-NEXT: v_or_b32_e32 v13, v13, v18
|
|
; VI-NEXT: v_or_b32_e32 v14, v14, v17
|
|
; VI-NEXT: v_or_b32_e32 v15, v15, v16
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v32f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX9-NEXT: v_pk_max_f16 v6, v6, v6
|
|
; GFX9-NEXT: v_pk_max_f16 v7, v7, v7
|
|
; GFX9-NEXT: v_pk_max_f16 v8, v8, v8
|
|
; GFX9-NEXT: v_pk_max_f16 v9, v9, v9
|
|
; GFX9-NEXT: v_pk_max_f16 v10, v10, v10
|
|
; GFX9-NEXT: v_pk_max_f16 v11, v11, v11
|
|
; GFX9-NEXT: v_pk_max_f16 v12, v12, v12
|
|
; GFX9-NEXT: v_pk_max_f16 v13, v13, v13
|
|
; GFX9-NEXT: v_pk_max_f16 v14, v14, v14
|
|
; GFX9-NEXT: v_pk_max_f16 v15, v15, v15
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v32f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v16, 16, v0
|
|
; CI-NEXT: v_lshrrev_b32_e32 v17, 16, v1
|
|
; CI-NEXT: v_lshrrev_b32_e32 v18, 16, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v16, v16
|
|
; CI-NEXT: v_lshrrev_b32_e32 v19, 16, v3
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v16
|
|
; CI-NEXT: v_lshrrev_b32_e32 v16, 16, v4
|
|
; CI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
|
|
; CI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
|
|
; CI-NEXT: v_lshlrev_b32_e32 v19, 16, v19
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v17
|
|
; CI-NEXT: v_lshrrev_b32_e32 v17, 16, v5
|
|
; CI-NEXT: v_or_b32_e32 v2, v2, v18
|
|
; CI-NEXT: v_lshrrev_b32_e32 v18, 16, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v16, v16
|
|
; CI-NEXT: v_or_b32_e32 v3, v3, v19
|
|
; CI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; CI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
|
|
; CI-NEXT: v_or_b32_e32 v4, v4, v16
|
|
; CI-NEXT: v_lshrrev_b32_e32 v16, 16, v8
|
|
; CI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
|
|
; CI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
|
|
; CI-NEXT: v_lshlrev_b32_e32 v19, 16, v19
|
|
; CI-NEXT: v_or_b32_e32 v5, v5, v17
|
|
; CI-NEXT: v_lshrrev_b32_e32 v17, 16, v9
|
|
; CI-NEXT: v_or_b32_e32 v6, v6, v18
|
|
; CI-NEXT: v_lshrrev_b32_e32 v18, 16, v10
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v16, v16
|
|
; CI-NEXT: v_or_b32_e32 v7, v7, v19
|
|
; CI-NEXT: v_lshrrev_b32_e32 v19, 16, v11
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
|
|
; CI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
|
|
; CI-NEXT: v_or_b32_e32 v8, v8, v16
|
|
; CI-NEXT: v_lshrrev_b32_e32 v16, 16, v12
|
|
; CI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
|
|
; CI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
|
|
; CI-NEXT: v_lshlrev_b32_e32 v19, 16, v19
|
|
; CI-NEXT: v_or_b32_e32 v9, v9, v17
|
|
; CI-NEXT: v_lshrrev_b32_e32 v17, 16, v13
|
|
; CI-NEXT: v_or_b32_e32 v10, v10, v18
|
|
; CI-NEXT: v_lshrrev_b32_e32 v18, 16, v15
|
|
; CI-NEXT: v_or_b32_e32 v11, v11, v19
|
|
; CI-NEXT: v_lshrrev_b32_e32 v19, 16, v14
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v16, v16
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v12, v12
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v15, v15
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v13, v13
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v14, v14
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v12, v12
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v15, v15
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v13, v13
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v14, v14
|
|
; CI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
|
|
; CI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
|
|
; CI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
|
|
; CI-NEXT: v_or_b32_e32 v12, v12, v16
|
|
; CI-NEXT: v_lshlrev_b32_e32 v16, 16, v19
|
|
; CI-NEXT: v_or_b32_e32 v13, v13, v17
|
|
; CI-NEXT: v_or_b32_e32 v14, v14, v16
|
|
; CI-NEXT: v_or_b32_e32 v15, v15, v18
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v32f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX11-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX11-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX11-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX11-NEXT: v_pk_max_f16 v6, v6, v6
|
|
; GFX11-NEXT: v_pk_max_f16 v7, v7, v7
|
|
; GFX11-NEXT: v_pk_max_f16 v8, v8, v8
|
|
; GFX11-NEXT: v_pk_max_f16 v9, v9, v9
|
|
; GFX11-NEXT: v_pk_max_f16 v10, v10, v10
|
|
; GFX11-NEXT: v_pk_max_f16 v11, v11, v11
|
|
; GFX11-NEXT: v_pk_max_f16 v12, v12, v12
|
|
; GFX11-NEXT: v_pk_max_f16 v13, v13, v13
|
|
; GFX11-NEXT: v_pk_max_f16 v14, v14, v14
|
|
; GFX11-NEXT: v_pk_max_f16 v15, v15, v15
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <32 x half> @llvm.canonicalize.v32f16(<32 x half> %val)
|
|
ret <32 x half> %canonicalized
|
|
}
|
|
|
|
define <64 x half> @v_test_canonicalize_var_v64f16(<64 x half> %val) #1 {
|
|
; VI-LABEL: v_test_canonicalize_var_v64f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v31, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v0, v0, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v1, v1, v1
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v2, v2, v2
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v3, v3, v3
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v4, v4, v4
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v5, v5, v5
|
|
; VI-NEXT: v_or_b32_e32 v5, v5, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v6, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v6, v6, v6
|
|
; VI-NEXT: v_or_b32_e32 v6, v6, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v7, v7, v7
|
|
; VI-NEXT: v_or_b32_e32 v7, v7, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v8, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v8, v8, v8
|
|
; VI-NEXT: v_or_b32_e32 v8, v8, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v9, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v9, v9, v9
|
|
; VI-NEXT: v_or_b32_e32 v9, v9, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v10, v10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v10, v10, v10
|
|
; VI-NEXT: v_or_b32_e32 v10, v10, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v11, v11 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v11, v11, v11
|
|
; VI-NEXT: v_or_b32_e32 v11, v11, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v12, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v12, v12, v12
|
|
; VI-NEXT: v_or_b32_e32 v12, v12, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v13, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v13, v13, v13
|
|
; VI-NEXT: v_or_b32_e32 v13, v13, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v14, v14 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v14, v14, v14
|
|
; VI-NEXT: v_or_b32_e32 v14, v14, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v15, v15 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v15, v15, v15
|
|
; VI-NEXT: v_or_b32_e32 v15, v15, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v16, v16 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v16, v16, v16
|
|
; VI-NEXT: v_or_b32_e32 v16, v16, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v17, v17 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v17, v17, v17
|
|
; VI-NEXT: v_or_b32_e32 v17, v17, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v18, v18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v18, v18, v18
|
|
; VI-NEXT: v_or_b32_e32 v18, v18, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v19, v19 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v19, v19, v19
|
|
; VI-NEXT: v_or_b32_e32 v19, v19, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v20, v20 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v20, v20, v20
|
|
; VI-NEXT: v_or_b32_e32 v20, v20, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v21, v21 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v21, v21, v21
|
|
; VI-NEXT: v_or_b32_e32 v21, v21, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v22, v22 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v22, v22, v22
|
|
; VI-NEXT: v_or_b32_e32 v22, v22, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v23, v23 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v23, v23, v23
|
|
; VI-NEXT: v_or_b32_e32 v23, v23, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v24, v24 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v24, v24, v24
|
|
; VI-NEXT: v_or_b32_e32 v24, v24, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v25, v25 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v25, v25, v25
|
|
; VI-NEXT: v_or_b32_e32 v25, v25, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v26, v26 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v26, v26, v26
|
|
; VI-NEXT: v_or_b32_e32 v26, v26, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v27, v27 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v27, v27, v27
|
|
; VI-NEXT: v_or_b32_e32 v27, v27, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v28, v28 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v28, v28, v28
|
|
; VI-NEXT: v_or_b32_e32 v28, v28, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v29, v29 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v29, v29, v29
|
|
; VI-NEXT: v_or_b32_e32 v29, v29, v31
|
|
; VI-NEXT: v_max_f16_sdwa v31, v30, v30 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v30, v30, v30
|
|
; VI-NEXT: v_or_b32_e32 v30, v30, v31
|
|
; VI-NEXT: buffer_load_dword v31, off, s[0:3], s32
|
|
; VI-NEXT: s_waitcnt vmcnt(0)
|
|
; VI-NEXT: v_max_f16_sdwa v32, v31, v31 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
|
|
; VI-NEXT: v_max_f16_e32 v31, v31, v31
|
|
; VI-NEXT: v_or_b32_e32 v31, v31, v32
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: v_test_canonicalize_var_v64f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: buffer_load_dword v31, off, s[0:3], s32
|
|
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX9-NEXT: v_pk_max_f16 v6, v6, v6
|
|
; GFX9-NEXT: v_pk_max_f16 v7, v7, v7
|
|
; GFX9-NEXT: v_pk_max_f16 v8, v8, v8
|
|
; GFX9-NEXT: v_pk_max_f16 v9, v9, v9
|
|
; GFX9-NEXT: v_pk_max_f16 v10, v10, v10
|
|
; GFX9-NEXT: v_pk_max_f16 v11, v11, v11
|
|
; GFX9-NEXT: v_pk_max_f16 v12, v12, v12
|
|
; GFX9-NEXT: v_pk_max_f16 v13, v13, v13
|
|
; GFX9-NEXT: v_pk_max_f16 v14, v14, v14
|
|
; GFX9-NEXT: v_pk_max_f16 v15, v15, v15
|
|
; GFX9-NEXT: v_pk_max_f16 v16, v16, v16
|
|
; GFX9-NEXT: v_pk_max_f16 v17, v17, v17
|
|
; GFX9-NEXT: v_pk_max_f16 v18, v18, v18
|
|
; GFX9-NEXT: v_pk_max_f16 v19, v19, v19
|
|
; GFX9-NEXT: v_pk_max_f16 v20, v20, v20
|
|
; GFX9-NEXT: v_pk_max_f16 v21, v21, v21
|
|
; GFX9-NEXT: v_pk_max_f16 v22, v22, v22
|
|
; GFX9-NEXT: v_pk_max_f16 v23, v23, v23
|
|
; GFX9-NEXT: v_pk_max_f16 v24, v24, v24
|
|
; GFX9-NEXT: v_pk_max_f16 v25, v25, v25
|
|
; GFX9-NEXT: v_pk_max_f16 v26, v26, v26
|
|
; GFX9-NEXT: v_pk_max_f16 v27, v27, v27
|
|
; GFX9-NEXT: v_pk_max_f16 v28, v28, v28
|
|
; GFX9-NEXT: v_pk_max_f16 v29, v29, v29
|
|
; GFX9-NEXT: v_pk_max_f16 v30, v30, v30
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NEXT: v_pk_max_f16 v31, v31, v31
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; CI-LABEL: v_test_canonicalize_var_v64f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v0
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v0, v0, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v1
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v1, v1, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v2
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v2, v2, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v3
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v3, v3, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v4
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v4, v4, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v5
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v5, v5, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v6
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v6, v6, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v7
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v7, v7, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v8
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v8, v8, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v9
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v9, v9, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v10
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v10, v10, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v11
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v11, v11, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v12
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v12, v12
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v12, v12
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v12, v12, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v13
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v13, v13
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v13, v13
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v13, v13, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v14
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v14, v14
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v14, v14
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v14, v14, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v15
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v15, v15
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v15, v15
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v15, v15, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v16
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v16, v16
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v16, v16, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v17
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v17, v17
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v17, v17, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v18
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v18, v18
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v18, v18, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v19
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v19, v19
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v19, v19, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v20
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v20, v20
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v20, v20
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v20, v20, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v21
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v21, v21
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v21, v21
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v21, v21, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v22
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v22, v22
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v22, v22
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v22, v22, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v23
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v23, v23
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v23, v23
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v23, v23, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v24
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v24, v24
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v24, v24
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v24, v24, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v25
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v25, v25
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v25, v25
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v25, v25, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v26
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v26, v26
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v26, v26, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v27
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v27, v27
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v27, v27
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v27, v27, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v28
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v28, v28
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v28, v28
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v28, v28, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v29
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v29, v29
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v29, v29
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v29, v29, v31
|
|
; CI-NEXT: v_lshrrev_b32_e32 v31, 16, v30
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v30, v30
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v30, v30
|
|
; CI-NEXT: v_lshlrev_b32_e32 v31, 16, v31
|
|
; CI-NEXT: v_or_b32_e32 v30, v30, v31
|
|
; CI-NEXT: buffer_load_dword v31, off, s[0:3], s32
|
|
; CI-NEXT: s_waitcnt vmcnt(0)
|
|
; CI-NEXT: v_lshrrev_b32_e32 v32, 16, v31
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v32, v32
|
|
; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v32, v32
|
|
; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
|
|
; CI-NEXT: v_lshlrev_b32_e32 v32, 16, v32
|
|
; CI-NEXT: v_or_b32_e32 v31, v31, v32
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_test_canonicalize_var_v64f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: scratch_load_b32 v31, off, s32
|
|
; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
|
|
; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
|
|
; GFX11-NEXT: v_pk_max_f16 v2, v2, v2
|
|
; GFX11-NEXT: v_pk_max_f16 v3, v3, v3
|
|
; GFX11-NEXT: v_pk_max_f16 v4, v4, v4
|
|
; GFX11-NEXT: v_pk_max_f16 v5, v5, v5
|
|
; GFX11-NEXT: v_pk_max_f16 v6, v6, v6
|
|
; GFX11-NEXT: v_pk_max_f16 v7, v7, v7
|
|
; GFX11-NEXT: v_pk_max_f16 v8, v8, v8
|
|
; GFX11-NEXT: v_pk_max_f16 v9, v9, v9
|
|
; GFX11-NEXT: v_pk_max_f16 v10, v10, v10
|
|
; GFX11-NEXT: v_pk_max_f16 v11, v11, v11
|
|
; GFX11-NEXT: v_pk_max_f16 v12, v12, v12
|
|
; GFX11-NEXT: v_pk_max_f16 v13, v13, v13
|
|
; GFX11-NEXT: v_pk_max_f16 v14, v14, v14
|
|
; GFX11-NEXT: v_pk_max_f16 v15, v15, v15
|
|
; GFX11-NEXT: v_pk_max_f16 v16, v16, v16
|
|
; GFX11-NEXT: v_pk_max_f16 v17, v17, v17
|
|
; GFX11-NEXT: v_pk_max_f16 v18, v18, v18
|
|
; GFX11-NEXT: v_pk_max_f16 v19, v19, v19
|
|
; GFX11-NEXT: v_pk_max_f16 v20, v20, v20
|
|
; GFX11-NEXT: v_pk_max_f16 v21, v21, v21
|
|
; GFX11-NEXT: v_pk_max_f16 v22, v22, v22
|
|
; GFX11-NEXT: v_pk_max_f16 v23, v23, v23
|
|
; GFX11-NEXT: v_pk_max_f16 v24, v24, v24
|
|
; GFX11-NEXT: v_pk_max_f16 v25, v25, v25
|
|
; GFX11-NEXT: v_pk_max_f16 v26, v26, v26
|
|
; GFX11-NEXT: v_pk_max_f16 v27, v27, v27
|
|
; GFX11-NEXT: v_pk_max_f16 v28, v28, v28
|
|
; GFX11-NEXT: v_pk_max_f16 v29, v29, v29
|
|
; GFX11-NEXT: v_pk_max_f16 v30, v30, v30
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX11-NEXT: v_pk_max_f16 v31, v31, v31
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%canonicalized = call <64 x half> @llvm.canonicalize.v64f16(<64 x half> %val)
|
|
ret <64 x half> %canonicalized
|
|
}
|
|
|
|
attributes #0 = { nounwind readnone }
|
|
attributes #1 = { nounwind denormal_fpenv(float: preservesign) }
|
|
attributes #2 = { nounwind denormal_fpenv(preservesign) }
|
|
attributes #3 = { nounwind denormal_fpenv(float: preservesign) }
|