[llvm] 053478b - [AMDGPU] Switch to using real True16 operands.
Ivan Kosarev via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 25 10:21:20 PDT 2023
Author: Ivan Kosarev
Date: 2023-09-25T18:21:13+01:00
New Revision: 053478bbd0ae5329ea7993261225e6541f728858
URL: https://github.com/llvm/llvm-project/commit/053478bbd0ae5329ea7993261225e6541f728858
DIFF: https://github.com/llvm/llvm-project/commit/053478bbd0ae5329ea7993261225e6541f728858.diff
LOG: [AMDGPU] Switch to using real True16 operands.
The DPP source and e64 destination operands remain unchanged for now.
Reviewed By: Joe_Nash
Differential Revision: https://reviews.llvm.org/D156104
Added:
Modified:
llvm/lib/Target/AMDGPU/SIInstrInfo.td
llvm/lib/Target/AMDGPU/SIInstructions.td
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
Removed:
llvm/test/CodeGen/AMDGPU/true16-ra-f128-fail.mir
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index 008f586db30860b..ea6df380153bb86 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -2471,15 +2471,24 @@ class VOPProfile_True16<VOPProfile P> : VOPProfile<P.ArgVT> {
let IsTrue16 = 1;
let IsRealTrue16 = 1;
// Most DstVT are 16-bit, but not all.
- let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
+ let DstRC = getVALUDstForVT<DstVT, 1 /*IsTrue16*/, 0 /*IsVOP3Encoding*/>.ret;
let DstRC64 = getVALUDstForVT<DstVT>.ret;
- let Src1RC32 = RegisterOperand<getVregSrcForVT_t16<Src1VT>.ret>;
+ let Src0RC32 = getVOPSrc0ForVT<Src0VT, 1 /*IsTrue16*/, 0 /*IsFake16*/>.ret;
+ let Src1RC32 = getVregSrcForVT_t16<Src1VT, 0 /*IsFake16*/>.op;
let Src0DPP = getVregSrcForVT_t16<Src0VT>.ret;
let Src1DPP = getVregSrcForVT_t16<Src1VT>.ret;
let Src2DPP = getVregSrcForVT_t16<Src2VT>.ret;
let Src0ModDPP = getSrcModDPP_t16<Src0VT>.ret;
let Src1ModDPP = getSrcModDPP_t16<Src1VT>.ret;
let Src2ModDPP = getSrcModDPP_t16<Src2VT>.ret;
+
+ let DstRC64 = getVALUDstForVT<DstVT, 1 /*IsTrue16*/, 1 /*IsVOP3Encoding*/>.ret;
+ let Src0RC64 = getVOP3SrcForVT<Src0VT, 1 /*IsTrue16*/>.ret;
+ let Src1RC64 = getVOP3SrcForVT<Src1VT, 1 /*IsTrue16*/>.ret;
+ let Src2RC64 = getVOP3SrcForVT<Src2VT, 1 /*IsTrue16*/>.ret;
+ let Src0Mod = getSrcMod<Src0VT, 1 /*IsTrue16*/>.ret;
+ let Src1Mod = getSrcMod<Src1VT, 1 /*IsTrue16*/>.ret;
+ let Src2Mod = getSrcMod<Src2VT, 1 /*IsTrue16*/>.ret;
}
class VOPProfile_Fake16<VOPProfile P> : VOPProfile<P.ArgVT> {
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 245808fc22a9c99..2e4708205523bca 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -2760,7 +2760,7 @@ multiclass SelectCanonicalizeAsMax<
def : GCNPat<
(fcanonicalize (f16 (VOP3Mods f16:$src, i32:$src_mods))),
- (V_MAX_F16_t16_e64 $src_mods, $src, $src_mods, $src, 0, 0)> {
+ (V_MAX_F16_fake16_e64 $src_mods, $src, $src_mods, $src, 0, 0)> {
let OtherPredicates = !listconcat(f16_preds, [Has16BitInsts, HasTrue16BitInsts]);
}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
index b72c4d64d9e0c93..160088b1c210d83 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
@@ -21,26 +21,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX8-NEXT: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_f16_denorm
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX9-NEXT: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_f16_denorm
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX10-NEXT: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_f16_denorm
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_t16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX11-NEXT: [[V_MAX_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_fake16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_fake16_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
%2:vgpr(s16) = G_FCANONICALIZE %1
@@ -64,26 +67,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %2:vgpr_32 = nofpexcept V_MUL_F16_e64 0, 15360, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX8-NEXT: [[V_MUL_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F16_e64 0, 15360, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F16_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_f16_flush
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX9-NEXT: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_f16_flush
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX10-NEXT: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_f16_flush
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F16_t16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX11-NEXT: [[V_MAX_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F16_fake16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F16_fake16_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s16) = G_TRUNC %0
%2:vgpr(s16) = G_FCANONICALIZE %1
@@ -108,26 +114,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %1:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_f32_denorm
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_f32_denorm
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_f32_denorm
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FCANONICALIZE %0
S_ENDPGM 0, implicit %1
@@ -151,26 +160,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %1:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_f32_flush
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_f32_flush
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_f32_flush
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FCANONICALIZE %0
S_ENDPGM 0, implicit %1
@@ -194,26 +206,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX8-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_v2f16_denorm
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX9-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_v2f16_denorm
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX10-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_v2f16_denorm
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX11-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(<2 x s16>) = G_FCANONICALIZE %0
S_ENDPGM 0, implicit %1
@@ -237,26 +252,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %1:vgpr_32 = nofpexcept V_PK_MUL_F16 0, 15360, 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX8-NEXT: [[V_PK_MUL_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MUL_F16 0, 15360, 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_PK_MUL_F16_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_v2f16_flush
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX9-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_v2f16_flush
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX10-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_v2f16_flush
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX11-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_PK_MAX_F16_]]
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(<2 x s16>) = G_FCANONICALIZE %0
S_ENDPGM 0, implicit %1
@@ -280,26 +298,29 @@ body: |
; GFX8: liveins: $vgpr0_vgpr1
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX8-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX8-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_f64_denorm
; GFX9: liveins: $vgpr0_vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX9-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX9-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_f64_denorm
; GFX10: liveins: $vgpr0_vgpr1
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX10-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX10-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_f64_denorm
; GFX11: liveins: $vgpr0_vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX11-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX11-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
%0:vgpr(s64) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = G_FCANONICALIZE %0
S_ENDPGM 0, implicit %1
@@ -323,26 +344,29 @@ body: |
; GFX8: liveins: $vgpr0_vgpr1
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX8-NEXT: %1:vreg_64 = nofpexcept V_MUL_F64_e64 0, 4607182418800017408, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX8-NEXT: [[V_MUL_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MUL_F64_e64 0, 4607182418800017408, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F64_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_f64_flush
; GFX9: liveins: $vgpr0_vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX9-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX9-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_f64_flush
; GFX10: liveins: $vgpr0_vgpr1
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX10-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX10-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_f64_flush
; GFX11: liveins: $vgpr0_vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
- ; GFX11-NEXT: %1:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %1
+ ; GFX11-NEXT: [[V_MAX_F64_e64_:%[0-9]+]]:vreg_64 = nofpexcept V_MAX_F64_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F64_e64_]]
%0:vgpr(s64) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = G_FCANONICALIZE %0
S_ENDPGM 0, implicit %1
@@ -365,26 +389,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_fabs_f32_denorm
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_fabs_f32_denorm
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_fabs_f32_denorm
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FABS %0
%2:vgpr(s32) = G_FCANONICALIZE %1
@@ -409,26 +436,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_fabs_f32_flush
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_fabs_f32_flush
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_fabs_f32_flush
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FABS %0
%2:vgpr(s32) = G_FCANONICALIZE %1
@@ -452,26 +482,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_fneg_f32_denorm
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_fneg_f32_denorm
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_fneg_f32_denorm
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FNEG %0
%2:vgpr(s32) = G_FCANONICALIZE %1
@@ -495,26 +528,29 @@ body: |
; GFX8: liveins: $vgpr0
; GFX8-NEXT: {{ $}}
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX8-NEXT: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_fneg_f32_flush
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX9-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_fneg_f32_flush
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX10-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_fneg_f32_flush
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
- ; GFX11-NEXT: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %2
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FNEG %0
%2:vgpr(s32) = G_FCANONICALIZE %1
@@ -540,32 +576,35 @@ body: |
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX8-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX8-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX8-NEXT: %3:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_fneg_fabs_f32_denorm
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX9-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX9-NEXT: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_fneg_fabs_f32_denorm
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX10-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX10-NEXT: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_fneg_fabs_f32_denorm
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX11-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX11-NEXT: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FNEG %0
%2:vgpr(s32) = G_FABS %1
@@ -592,32 +631,35 @@ body: |
; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX8-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX8-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX8-NEXT: %3:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX8-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX8-NEXT: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX8-NEXT: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ ;
; GFX9-LABEL: name: fcanonicalize_fneg_fabs_f32_flush
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX9-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX9-NEXT: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX9-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX9-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX9-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX10-LABEL: name: fcanonicalize_fneg_fabs_f32_flush
; GFX10: liveins: $vgpr0
; GFX10-NEXT: {{ $}}
; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX10-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX10-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX10-NEXT: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX10-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX10-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ;
; GFX11-LABEL: name: fcanonicalize_fneg_fabs_f32_flush
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
; GFX11-NEXT: [[V_XOR_B32_e64_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e64 [[S_MOV_B32_]], [[COPY]], implicit $exec
- ; GFX11-NEXT: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
- ; GFX11-NEXT: S_ENDPGM 0, implicit %3
+ ; GFX11-NEXT: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e64_]], 2, [[V_XOR_B32_e64_]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_FNEG %0
%2:vgpr(s32) = G_FABS %1
diff --git a/llvm/test/CodeGen/AMDGPU/true16-ra-f128-fail.mir b/llvm/test/CodeGen/AMDGPU/true16-ra-f128-fail.mir
deleted file mode 100644
index 7cd4f3904cdb89e..000000000000000
--- a/llvm/test/CodeGen/AMDGPU/true16-ra-f128-fail.mir
+++ /dev/null
@@ -1,34 +0,0 @@
-# RUN: not llc -march=amdgcn -mcpu=gfx1100 -debug-only=regalloc -start-before=greedy,0 -stop-after=virtregrewriter,1 -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck --check-prefixes=CHECK %s
-# REQUIRES: asserts
-
---- |
- define amdgpu_ps void @e32() {
- ret void
- }
-...
-
-
----
-name: e32
-tracksRegLiveness: true
-machineFunctionInfo:
- stackPtrOffsetReg: $sgpr32
-
-body: |
- bb.0:
- liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79, $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95, $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111, $vgpr112, $vgpr113, $vgpr114, $vgpr115, $vgpr116, $vgpr117, $vgpr118, $vgpr119, $vgpr120, $vgpr121, $vgpr122, $vgpr123, $vgpr124, $vgpr125, $vgpr126, $vgpr127
-
- ; CHECK: error: ran out of registers during register allocation
- ; CHECK: [[REG1:vgpr[0-9]+]] = V_ADD_F16_t16_e32
- ; CHECK: SI_SPILL_V32_SAVE $[[REG1]]
- %0:vgpr_32_lo128 = V_ADD_F16_t16_e32 $vgpr0, $vgpr1, implicit $exec, implicit $mode
- S_NOP 0, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
- S_NOP 0, implicit $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47
- S_NOP 0, implicit $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63
- S_NOP 0, implicit $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79
- S_NOP 0, implicit $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95
- S_NOP 0, implicit $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111
- S_NOP 0, implicit $vgpr112, implicit $vgpr113, implicit $vgpr114, implicit $vgpr115, implicit $vgpr116, implicit $vgpr117, implicit $vgpr118, implicit $vgpr119, implicit $vgpr120, implicit $vgpr121, implicit $vgpr122, implicit $vgpr123, implicit $vgpr124, implicit $vgpr125, implicit $vgpr126, implicit $vgpr127
- S_ENDPGM 0, implicit %0
-...
-
More information about the llvm-commits
mailing list