[llvm] [AMDGPU][GlobalISel] Lower G_FMINIMUM and G_FMAXIMUM (PR #151122)
Mirko BrkuĊĦanin via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 29 04:08:43 PDT 2025
https://github.com/mbrkusanin updated https://github.com/llvm/llvm-project/pull/151122
>From 61fbbb49537c3e3546102737c48b5f58bb85dc44 Mon Sep 17 00:00:00 2001
From: Mirko Brkusanin <Mirko.Brkusanin at amd.com>
Date: Tue, 29 Jul 2025 12:31:30 +0200
Subject: [PATCH 1/2] [AMDGPU][GlobalISel] Lower G_FMINIMUM and G_FMAXIMUM
Add GlobalISel lowering of G_FMINIMUM and G_FMAXIMUM following the same
logic as in SDag's expandFMINIMUM_FMAXIMUM.
Update AMDGPU legalization rules: Pre GFX12 now uses new lowering method
and make G_FMINNUM_IEEE and G_FMAXNUM_IEEE legal to match SDag.
---
.../llvm/CodeGen/GlobalISel/LegalizerHelper.h | 1 +
.../CodeGen/GlobalISel/LegalizerHelper.cpp | 72 ++
.../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp | 55 +-
.../AMDGPU/GlobalISel/legalize-fmaximum.mir | 275 ++++++
.../AMDGPU/GlobalISel/legalize-fminimum.mir | 275 ++++++
llvm/test/CodeGen/AMDGPU/fmaximum.ll | 921 +++++++++++++++---
llvm/test/CodeGen/AMDGPU/fminimum.ll | 921 +++++++++++++++---
7 files changed, 2235 insertions(+), 285 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h
index ea0873f41ebba..e100a2c69ffbf 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h
@@ -456,6 +456,7 @@ class LegalizerHelper {
LLVM_ABI LegalizeResult lowerMinMax(MachineInstr &MI);
LLVM_ABI LegalizeResult lowerFCopySign(MachineInstr &MI);
LLVM_ABI LegalizeResult lowerFMinNumMaxNum(MachineInstr &MI);
+ LLVM_ABI LegalizeResult lowerFMinimumMaximum(MachineInstr &MI);
LLVM_ABI LegalizeResult lowerFMad(MachineInstr &MI);
LLVM_ABI LegalizeResult lowerIntrinsicRound(MachineInstr &MI);
LLVM_ABI LegalizeResult lowerFFloor(MachineInstr &MI);
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index d9d3569affa39..669590707cb6c 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -4674,6 +4674,9 @@ LegalizerHelper::lower(MachineInstr &MI, unsigned TypeIdx, LLT LowerHintTy) {
case G_FMINIMUMNUM:
case G_FMAXIMUMNUM:
return lowerFMinNumMaxNum(MI);
+ case G_FMINIMUM:
+ case G_FMAXIMUM:
+ return lowerFMinimumMaximum(MI);
case G_MERGE_VALUES:
return lowerMergeValues(MI);
case G_UNMERGE_VALUES:
@@ -8294,6 +8297,75 @@ LegalizerHelper::lowerFMinNumMaxNum(MachineInstr &MI) {
return Legalized;
}
+LegalizerHelper::LegalizeResult
+LegalizerHelper::lowerFMinimumMaximum(MachineInstr &MI) {
+ unsigned Opc = MI.getOpcode();
+ auto [Dst, Src0, Src1] = MI.getFirst3Regs();
+ LLT Ty = MRI.getType(Dst);
+ LLT CmpTy =
+ Ty.isScalar() ? LLT::scalar(1) : LLT::vector(Ty.getElementCount(), 1);
+
+ bool IsMax = (Opc == TargetOpcode::G_FMAXIMUM);
+ unsigned OpcIeee =
+ IsMax ? TargetOpcode::G_FMAXNUM_IEEE : TargetOpcode::G_FMINNUM_IEEE;
+ unsigned OpcNonIeee =
+ IsMax ? TargetOpcode::G_FMAXNUM : TargetOpcode::G_FMINNUM;
+ bool MinMaxMustRespectOrderedZero = false;
+ Register Res;
+
+ // IEEE variants don't need canonicalization
+ if (LI.isLegalOrCustom({OpcIeee, Ty})) {
+ Res = MIRBuilder.buildInstr(OpcIeee, {Ty}, {Src0, Src1}).getReg(0);
+ MinMaxMustRespectOrderedZero = true;
+ } else if (LI.isLegalOrCustom({OpcNonIeee, Ty})) {
+ Res = MIRBuilder.buildInstr(OpcNonIeee, {Ty}, {Src0, Src1}).getReg(0);
+ } else {
+ auto Compare = MIRBuilder.buildFCmp(
+ IsMax ? CmpInst::FCMP_OGT : CmpInst::FCMP_OLT, CmpTy, Src0, Src1);
+ Res = MIRBuilder.buildSelect(Ty, Compare, Src0, Src1).getReg(0);
+ }
+
+ // Propagate any NaN of both operands
+ if (!MI.getFlag(MachineInstr::FmNoNans) &&
+ (!isKnownNeverNaN(Src0, MRI) || isKnownNeverNaN(Src1, MRI))) {
+ auto IsOrdered = MIRBuilder.buildFCmp(CmpInst::FCMP_ORD, CmpTy, Src0, Src1);
+
+ LLT ElementTy = Ty.isScalar() ? Ty : Ty.getElementType();
+ APFloat NaNValue = APFloat::getNaN(getFltSemanticForLLT(ElementTy));
+ Register NaN = MIRBuilder.buildFConstant(ElementTy, NaNValue).getReg(0);
+ if (Ty.isVector())
+ NaN = MIRBuilder.buildSplatBuildVector(Ty, NaN).getReg(0);
+
+ Res = MIRBuilder.buildSelect(Ty, IsOrdered, Res, NaN).getReg(0);
+ }
+
+ // fminimum/fmaximum requires -0.0 less than +0.0
+ if (!MinMaxMustRespectOrderedZero && !MI.getFlag(MachineInstr::FmNsz)) {
+ GISelValueTracking VT(MIRBuilder.getMF());
+ KnownFPClass Src0Info = VT.computeKnownFPClass(Src0, fcZero);
+ KnownFPClass Src1Info = VT.computeKnownFPClass(Src1, fcZero);
+
+ if (!Src0Info.isKnownNeverZero() && !Src1Info.isKnownNeverZero()) {
+ Register Zero = MIRBuilder.buildFConstant(Ty, 0.0).getReg(0);
+ auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_OEQ, CmpTy, Res, Zero);
+
+ unsigned TestClass = IsMax ? fcPosZero : fcNegZero;
+
+ auto LHSTestZero = MIRBuilder.buildIsFPClass(CmpTy, Src0, TestClass);
+ auto LHSSelect = MIRBuilder.buildSelect(Ty, LHSTestZero, Src0, Res);
+
+ auto RHSTestZero = MIRBuilder.buildIsFPClass(CmpTy, Src1, TestClass);
+ auto RHSSelect = MIRBuilder.buildSelect(Ty, RHSTestZero, Src1, LHSSelect);
+
+ Res = MIRBuilder.buildSelect(Ty, IsZero, RHSSelect, Res).getReg(0);
+ }
+ }
+
+ MIRBuilder.buildCopy(Dst, Res);
+ MI.eraseFromParent();
+ return Legalized;
+}
+
LegalizerHelper::LegalizeResult LegalizerHelper::lowerFMad(MachineInstr &MI) {
// Expand G_FMAD a, b, c -> G_FADD (G_FMUL a, b), c
Register DstReg = MI.getOperand(0).getReg();
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index fedfa3f9dd900..2dfa07ca9cfe2 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -957,9 +957,27 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
FPOpActions.clampMaxNumElementsStrict(0, S32, 2);
}
+ auto &MinNumMaxNumIeee = getActionDefinitionsBuilder(
+ {G_FMINNUM_IEEE, G_FMAXNUM_IEEE});
+
+ if (ST.hasVOP3PInsts()) {
+ MinNumMaxNumIeee.legalFor(FPTypesPK16)
+ .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
+ .clampMaxNumElements(0, S16, 2)
+ .clampScalar(0, S16, S64)
+ .scalarize(0);
+ } else if (ST.has16BitInsts()) {
+ MinNumMaxNumIeee.legalFor(FPTypes16)
+ .clampScalar(0, S16, S64)
+ .scalarize(0);
+ } else {
+ MinNumMaxNumIeee.legalFor(FPTypesBase)
+ .clampScalar(0, S32, S64)
+ .scalarize(0);
+ }
+
auto &MinNumMaxNum = getActionDefinitionsBuilder(
- {G_FMINNUM, G_FMAXNUM, G_FMINIMUMNUM, G_FMAXIMUMNUM, G_FMINNUM_IEEE,
- G_FMAXNUM_IEEE});
+ {G_FMINNUM, G_FMAXNUM, G_FMINIMUMNUM, G_FMAXIMUMNUM});
if (ST.hasVOP3PInsts()) {
MinNumMaxNum.customFor(FPTypesPK16)
@@ -2100,9 +2118,17 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
.legalFor(FPTypesPK16)
.clampMaxNumElements(0, S16, 2)
.scalarize(0);
+ } else if (ST.hasVOP3PInsts()){
+ getActionDefinitionsBuilder({G_FMINIMUM, G_FMAXIMUM})
+ .lowerFor({V2S16})
+ .clampMaxNumElementsStrict(0, S16, 2)
+ .scalarize(0)
+ .lower();
} else {
- // TODO: Implement
- getActionDefinitionsBuilder({G_FMINIMUM, G_FMAXIMUM}).lower();
+ getActionDefinitionsBuilder({G_FMINIMUM, G_FMAXIMUM})
+ .scalarize(0)
+ .clampScalar(0, S32, S64)
+ .lower();
}
getActionDefinitionsBuilder({G_MEMCPY, G_MEMCPY_INLINE, G_MEMMOVE, G_MEMSET})
@@ -2159,8 +2185,6 @@ bool AMDGPULegalizerInfo::legalizeCustom(
case TargetOpcode::G_FMAXNUM:
case TargetOpcode::G_FMINIMUMNUM:
case TargetOpcode::G_FMAXIMUMNUM:
- case TargetOpcode::G_FMINNUM_IEEE:
- case TargetOpcode::G_FMAXNUM_IEEE:
return legalizeMinNumMaxNum(Helper, MI);
case TargetOpcode::G_EXTRACT_VECTOR_ELT:
return legalizeExtractVectorElt(MI, MRI, B);
@@ -2734,23 +2758,8 @@ bool AMDGPULegalizerInfo::legalizeMinNumMaxNum(LegalizerHelper &Helper,
MachineFunction &MF = Helper.MIRBuilder.getMF();
const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
- const bool IsIEEEOp = MI.getOpcode() == AMDGPU::G_FMINNUM_IEEE ||
- MI.getOpcode() == AMDGPU::G_FMAXNUM_IEEE;
-
- // With ieee_mode disabled, the instructions have the correct behavior
- // already for G_FMINIMUMNUM/G_FMAXIMUMNUM.
- //
- // FIXME: G_FMINNUM/G_FMAXNUM should match the behavior with ieee_mode
- // enabled.
- if (!MFI->getMode().IEEE) {
- if (MI.getOpcode() == AMDGPU::G_FMINIMUMNUM ||
- MI.getOpcode() == AMDGPU::G_FMAXIMUMNUM)
- return true;
-
- return !IsIEEEOp;
- }
-
- if (IsIEEEOp)
+ // With ieee_mode disabled, the instructions have the correct behavior.
+ if (!MFI->getMode().IEEE)
return true;
return Helper.lowerFMinNumMaxNum(MI) == LegalizerHelper::Legalized;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir
new file mode 100644
index 0000000000000..4b214e66ea994
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir
@@ -0,0 +1,275 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9 %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 -mattr=-real-true16 -run-pass=legalizer %s -o - | FileCheck -check-prefixes=GFX12 %s
+
+---
+name: test_fmaximum_f16
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fmaximum_f16
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s16) = G_FMAXNUM_IEEE [[TRUNC]], [[TRUNC1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s16) = COPY [[SELECT]](s16)
+ ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16)
+ ; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_f16
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s16) = G_FMAXIMUM [[TRUNC]], [[TRUNC1]]
+ ; GFX12-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMAXIMUM]](s16)
+ ; GFX12-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s16) = G_TRUNC %0(s32)
+ %2:_(s32) = COPY $vgpr1
+ %3:_(s16) = G_TRUNC %2(s32)
+ %4:_(s16) = G_FMAXIMUM %1, %3
+ %5:_(s32) = G_ANYEXT %4(s16)
+ $vgpr0 = COPY %5(s32)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fmaximum_f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fmaximum_f32
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_f32
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = G_FMAXIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = G_FMAXIMUM %0, %1
+ $vgpr0 = COPY %2(s32)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fmaximum_f64
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+ ; GFX9-LABEL: name: test_fmaximum_f64
+ ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s64) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s64), [[COPY1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY [[SELECT]](s64)
+ ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[COPY2]](s64)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0_vgpr1
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_f64
+ ; GFX12: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s64) = G_FMAXIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[FMAXIMUM]](s64)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0_vgpr1
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s64) = COPY $vgpr2_vgpr3
+ %2:_(s64) = G_FMAXIMUM %0, %1
+ $vgpr0_vgpr1 = COPY %2(s64)
+ SI_RETURN implicit $vgpr0_vgpr1
+...
+---
+name: test_fmaximum_v2f16
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fmaximum_v2f16
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(<2 x s16>) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+ ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+ ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+ ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+ ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
+ ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
+ ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
+ ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC2]]
+ ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC1]](s16), [[TRUNC3]]
+ ; GFX9-NEXT: [[C1:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00
+ ; GFX9-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[FMAXNUM_IEEE]](<2 x s16>)
+ ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
+ ; GFX9-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
+ ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[TRUNC4]], [[C1]]
+ ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s16) = G_SELECT [[FCMP1]](s1), [[TRUNC5]], [[C1]]
+ ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[SELECT]](s16), [[SELECT1]](s16)
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY [[BUILD_VECTOR]](<2 x s16>)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](<2 x s16>)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_v2f16
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(<2 x s16>) = G_FMAXIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](<2 x s16>)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(<2 x s16>) = COPY $vgpr0
+ %1:_(<2 x s16>) = COPY $vgpr1
+ %2:_(<2 x s16>) = G_FMAXIMUM %0, %1
+ $vgpr0 = COPY %2(<2 x s16>)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fmaximum_v2f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+
+ ; GFX9-LABEL: name: test_fmaximum_v2f32
+ ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+ ; GFX9-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY2]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY2]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32)
+ ; GFX9-NEXT: [[FMAXNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY1]], [[COPY3]]
+ ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY1]](s32), [[COPY3]]
+ ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[FCMP1]](s1), [[FMAXNUM_IEEE1]], [[C]]
+ ; GFX9-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY [[SELECT1]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY4]](s32)
+ ; GFX9-NEXT: $vgpr1 = COPY [[COPY5]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_v2f32
+ ; GFX12: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+ ; GFX12-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = G_FMAXIMUM [[COPY]], [[COPY2]]
+ ; GFX12-NEXT: [[FMAXIMUM1:%[0-9]+]]:_(s32) = G_FMAXIMUM [[COPY1]], [[COPY3]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32)
+ ; GFX12-NEXT: $vgpr1 = COPY [[FMAXIMUM1]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(<2 x s32>) = G_BUILD_VECTOR %0(s32), %1(s32)
+ %3:_(s32) = COPY $vgpr2
+ %4:_(s32) = COPY $vgpr3
+ %5:_(<2 x s32>) = G_BUILD_VECTOR %3(s32), %4(s32)
+ %6:_(<2 x s32>) = G_FMAXIMUM %2, %5
+ %7:_(s32), %8:_(s32) = G_UNMERGE_VALUES %6(<2 x s32>)
+ $vgpr0 = COPY %7(s32)
+ $vgpr1 = COPY %8(s32)
+ SI_RETURN implicit $vgpr0, implicit $vgpr1
+...
+---
+name: test_fmaximum_nsz_f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fmaximum_nsz_f32
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_nsz_f32
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = nsz G_FMAXIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = nsz G_FMAXIMUM %0, %1
+ $vgpr0 = COPY %2(s32)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fmaximum_nnan_f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fmaximum_nnan_f32
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[FMAXNUM_IEEE]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fmaximum_nnan_f32
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = nnan G_FMAXIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = nnan G_FMAXIMUM %0, %1
+ $vgpr0 = COPY %2(s32)
+ SI_RETURN implicit $vgpr0
+...
+
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir
new file mode 100644
index 0000000000000..8ba0794be3069
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir
@@ -0,0 +1,275 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9 %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 -mattr=-real-true16 -run-pass=legalizer %s -o - | FileCheck -check-prefixes=GFX12 %s
+
+---
+name: test_fminimum_f16
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fminimum_f16
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[TRUNC]], [[TRUNC1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s16) = COPY [[SELECT]](s16)
+ ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16)
+ ; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fminimum_f16
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s16) = G_FMINIMUM [[TRUNC]], [[TRUNC1]]
+ ; GFX12-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMINIMUM]](s16)
+ ; GFX12-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s16) = G_TRUNC %0(s32)
+ %2:_(s32) = COPY $vgpr1
+ %3:_(s16) = G_TRUNC %2(s32)
+ %4:_(s16) = G_FMINIMUM %1, %3
+ %5:_(s32) = G_ANYEXT %4(s16)
+ $vgpr0 = COPY %5(s32)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fminimum_f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fminimum_f32
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fminimum_f32
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = G_FMINIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = G_FMINIMUM %0, %1
+ $vgpr0 = COPY %2(s32)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fminimum_f64
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+ ; GFX9-LABEL: name: test_fminimum_f64
+ ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s64) = G_FMINNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s64), [[COPY1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY [[SELECT]](s64)
+ ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[COPY2]](s64)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0_vgpr1
+ ;
+ ; GFX12-LABEL: name: test_fminimum_f64
+ ; GFX12: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s64) = G_FMINIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[FMINIMUM]](s64)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0_vgpr1
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s64) = COPY $vgpr2_vgpr3
+ %2:_(s64) = G_FMINIMUM %0, %1
+ $vgpr0_vgpr1 = COPY %2(s64)
+ SI_RETURN implicit $vgpr0_vgpr1
+...
+---
+name: test_fminimum_v2f16
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fminimum_v2f16
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(<2 x s16>) = G_FMINNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+ ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+ ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+ ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+ ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
+ ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
+ ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
+ ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC2]]
+ ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC1]](s16), [[TRUNC3]]
+ ; GFX9-NEXT: [[C1:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00
+ ; GFX9-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[FMINNUM_IEEE]](<2 x s16>)
+ ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
+ ; GFX9-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
+ ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[TRUNC4]], [[C1]]
+ ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s16) = G_SELECT [[FCMP1]](s1), [[TRUNC5]], [[C1]]
+ ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[SELECT]](s16), [[SELECT1]](s16)
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY [[BUILD_VECTOR]](<2 x s16>)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](<2 x s16>)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fminimum_v2f16
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(<2 x s16>) = G_FMINIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](<2 x s16>)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(<2 x s16>) = COPY $vgpr0
+ %1:_(<2 x s16>) = COPY $vgpr1
+ %2:_(<2 x s16>) = G_FMINIMUM %0, %1
+ $vgpr0 = COPY %2(<2 x s16>)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fminimum_v2f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+
+ ; GFX9-LABEL: name: test_fminimum_v2f32
+ ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+ ; GFX9-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY2]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY2]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32)
+ ; GFX9-NEXT: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY1]], [[COPY3]]
+ ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY1]](s32), [[COPY3]]
+ ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[FCMP1]](s1), [[FMINNUM_IEEE1]], [[C]]
+ ; GFX9-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY [[SELECT1]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY4]](s32)
+ ; GFX9-NEXT: $vgpr1 = COPY [[COPY5]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
+ ;
+ ; GFX12-LABEL: name: test_fminimum_v2f32
+ ; GFX12: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+ ; GFX12-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = G_FMINIMUM [[COPY]], [[COPY2]]
+ ; GFX12-NEXT: [[FMINIMUM1:%[0-9]+]]:_(s32) = G_FMINIMUM [[COPY1]], [[COPY3]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32)
+ ; GFX12-NEXT: $vgpr1 = COPY [[FMINIMUM1]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(<2 x s32>) = G_BUILD_VECTOR %0(s32), %1(s32)
+ %3:_(s32) = COPY $vgpr2
+ %4:_(s32) = COPY $vgpr3
+ %5:_(<2 x s32>) = G_BUILD_VECTOR %3(s32), %4(s32)
+ %6:_(<2 x s32>) = G_FMINIMUM %2, %5
+ %7:_(s32), %8:_(s32) = G_UNMERGE_VALUES %6(<2 x s32>)
+ $vgpr0 = COPY %7(s32)
+ $vgpr1 = COPY %8(s32)
+ SI_RETURN implicit $vgpr0, implicit $vgpr1
+...
+---
+name: test_fminimum_nsz_f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fminimum_nsz_f32
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]]
+ ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000
+ ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fminimum_nsz_f32
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = nsz G_FMINIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = nsz G_FMINIMUM %0, %1
+ $vgpr0 = COPY %2(s32)
+ SI_RETURN implicit $vgpr0
+...
+---
+name: test_fminimum_nnan_f32
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+
+ ; GFX9-LABEL: name: test_fminimum_nnan_f32
+ ; GFX9: liveins: $vgpr0, $vgpr1
+ ; GFX9-NEXT: {{ $}}
+ ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY1]]
+ ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[FMINNUM_IEEE]](s32)
+ ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32)
+ ; GFX9-NEXT: SI_RETURN implicit $vgpr0
+ ;
+ ; GFX12-LABEL: name: test_fminimum_nnan_f32
+ ; GFX12: liveins: $vgpr0, $vgpr1
+ ; GFX12-NEXT: {{ $}}
+ ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+ ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+ ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = nnan G_FMINIMUM [[COPY]], [[COPY1]]
+ ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32)
+ ; GFX12-NEXT: SI_RETURN implicit $vgpr0
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = nnan G_FMINIMUM %0, %1
+ $vgpr0 = COPY %2(s32)
+ SI_RETURN implicit $vgpr0
+...
+
diff --git a/llvm/test/CodeGen/AMDGPU/fmaximum.ll b/llvm/test/CodeGen/AMDGPU/fmaximum.ll
index 0adbecd952ae3..ac632aca52c5a 100644
--- a/llvm/test/CodeGen/AMDGPU/fmaximum.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmaximum.ll
@@ -1,117 +1,296 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-TRUE16 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-FAKE16 %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-TRUE16 %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s
define amdgpu_ps float @test_fmaximum_f32_vv(float %a, float %b) {
-; GCN-LABEL: test_fmaximum_f32_vv:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, v1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_f32_vv:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f32_vv:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.maximum.f32(float %a, float %b)
ret float %val
}
define amdgpu_ps float @test_fmaximum_f32_ss(float inreg %a, float inreg %b) {
-; GCN-LABEL: test_fmaximum_f32_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_maximum_f32 s0, s0, s1
-; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GCN-NEXT: v_mov_b32_e32 v0, s0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_f32_ss:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-NEXT: v_max_f32_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f32_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_maximum_f32 s0, s0, s1
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.maximum.f32(float %a, float %b)
ret float %val
}
define amdgpu_ps float @test_fmaximum_f32_vs(float %a, float inreg %b) {
-; GCN-LABEL: test_fmaximum_f32_vs:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, s0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_f32_vs:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f32_vs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.maximum.f32(float %a, float %b)
ret float %val
}
define amdgpu_ps float @test_fmaximum_nnan_f32(float %a, float %b) {
-; GCN-LABEL: test_fmaximum_nnan_f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, v1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_nnan_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_nnan_f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
%val = call nnan float @llvm.maximum.f32(float %a, float %b)
ret float %val
}
+define amdgpu_ps float @test_fmaximum_nsz_f32(float %a, float %b) {
+; GFX9-LABEL: test_fmaximum_nsz_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_nsz_f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call nsz float @llvm.maximum.f32(float %a, float %b)
+ ret float %val
+}
+
+define amdgpu_ps float @test_fmaximum_signed_zero_f32() {
+; GFX9-LABEL: test_fmaximum_signed_zero_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_signed_zero_f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call float @llvm.maximum.f32(float -0.0, float 0.0)
+ ret float %val
+}
+
define amdgpu_ps <2 x float> @test_fmaximum_v2f32(<2 x float> %a, <2 x float> %b) {
-; GCN-LABEL: test_fmaximum_v2f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, v2
-; GCN-NEXT: v_maximum_f32 v1, v1, v3
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_v2f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v4, v0, v2
+; GFX9-NEXT: v_mov_b32_e32 v5, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v5, v4, vcc
+; GFX9-NEXT: v_max_f32_e32 v2, v1, v3
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v3
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v2f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v2
+; GFX12-NEXT: v_maximum_f32 v1, v1, v3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x float> @llvm.maximum.v2f32(<2 x float> %a, <2 x float> %b)
ret <2 x float> %val
}
define amdgpu_ps <2 x float> @test_fmaximum_v2f32_ss(<2 x float> inreg %a, <2 x float> inreg %b) {
-; GCN-LABEL: test_fmaximum_v2f32_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_maximum_f32 s0, s0, s2
-; GCN-NEXT: s_maximum_f32 s1, s1, s3
-; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GCN-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_v2f32_ss:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_max_f32_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: v_max_f32_e32 v3, s1, v1
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s1, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v2f32_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_maximum_f32 s0, s0, s2
+; GFX12-NEXT: s_maximum_f32 s1, s1, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x float> @llvm.maximum.v2f32(<2 x float> %a, <2 x float> %b)
ret <2 x float> %val
}
define amdgpu_ps <3 x float> @test_fmaximum_v3f32(<3 x float> %a, <3 x float> %b) {
-; GCN-LABEL: test_fmaximum_v3f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, v3
-; GCN-NEXT: v_maximum_f32 v1, v1, v4
-; GCN-NEXT: v_maximum_f32 v2, v2, v5
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_v3f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v6, v0, v3
+; GFX9-NEXT: v_mov_b32_e32 v7, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v3
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v7, v6, vcc
+; GFX9-NEXT: v_max_f32_e32 v3, v1, v4
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v4
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v7, v3, vcc
+; GFX9-NEXT: v_max_f32_e32 v3, v2, v5
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v5
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v7, v3, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v3f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v3
+; GFX12-NEXT: v_maximum_f32 v1, v1, v4
+; GFX12-NEXT: v_maximum_f32 v2, v2, v5
+; GFX12-NEXT: ; return to shader part epilog
%val = call <3 x float> @llvm.maximum.v3f32(<3 x float> %a, <3 x float> %b)
ret <3 x float> %val
}
define amdgpu_ps <4 x float> @test_fmaximum_v4f32(<4 x float> %a, <4 x float> %b) {
-; GCN-LABEL: test_fmaximum_v4f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, v4
-; GCN-NEXT: v_maximum_f32 v1, v1, v5
-; GCN-NEXT: v_maximum_f32 v2, v2, v6
-; GCN-NEXT: v_maximum_f32 v3, v3, v7
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_v4f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v8, v0, v4
+; GFX9-NEXT: v_mov_b32_e32 v9, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v4
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v9, v8, vcc
+; GFX9-NEXT: v_max_f32_e32 v4, v1, v5
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v5
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v9, v4, vcc
+; GFX9-NEXT: v_max_f32_e32 v4, v2, v6
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v6
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v9, v4, vcc
+; GFX9-NEXT: v_max_f32_e32 v4, v3, v7
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v3, v7
+; GFX9-NEXT: v_cndmask_b32_e32 v3, v9, v4, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v4f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v4
+; GFX12-NEXT: v_maximum_f32 v1, v1, v5
+; GFX12-NEXT: v_maximum_f32 v2, v2, v6
+; GFX12-NEXT: v_maximum_f32 v3, v3, v7
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x float> @llvm.maximum.v4f32(<4 x float> %a, <4 x float> %b)
ret <4 x float> %val
}
define amdgpu_ps <16 x float> @test_fmaximum_v16f32(<16 x float> %a, <16 x float> %b) {
-; GCN-LABEL: test_fmaximum_v16f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f32 v0, v0, v16
-; GCN-NEXT: v_maximum_f32 v1, v1, v17
-; GCN-NEXT: v_maximum_f32 v2, v2, v18
-; GCN-NEXT: v_maximum_f32 v3, v3, v19
-; GCN-NEXT: v_maximum_f32 v4, v4, v20
-; GCN-NEXT: v_maximum_f32 v5, v5, v21
-; GCN-NEXT: v_maximum_f32 v6, v6, v22
-; GCN-NEXT: v_maximum_f32 v7, v7, v23
-; GCN-NEXT: v_maximum_f32 v8, v8, v24
-; GCN-NEXT: v_maximum_f32 v9, v9, v25
-; GCN-NEXT: v_maximum_f32 v10, v10, v26
-; GCN-NEXT: v_maximum_f32 v11, v11, v27
-; GCN-NEXT: v_maximum_f32 v12, v12, v28
-; GCN-NEXT: v_maximum_f32 v13, v13, v29
-; GCN-NEXT: v_maximum_f32 v14, v14, v30
-; GCN-NEXT: v_maximum_f32 v15, v15, v31
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_v16f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v32, v1, v17
+; GFX9-NEXT: v_mov_b32_e32 v33, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v17
+; GFX9-NEXT: v_max_f32_e32 v1, v0, v16
+; GFX9-NEXT: v_cmp_o_f32_e64 s[12:13], v0, v16
+; GFX9-NEXT: v_max_f32_e32 v17, v2, v18
+; GFX9-NEXT: v_cmp_o_f32_e64 s[0:1], v2, v18
+; GFX9-NEXT: v_max_f32_e32 v18, v3, v19
+; GFX9-NEXT: v_cmp_o_f32_e64 s[2:3], v3, v19
+; GFX9-NEXT: v_max_f32_e32 v19, v4, v20
+; GFX9-NEXT: v_cmp_o_f32_e64 s[4:5], v4, v20
+; GFX9-NEXT: v_max_f32_e32 v20, v5, v21
+; GFX9-NEXT: v_cmp_o_f32_e64 s[6:7], v5, v21
+; GFX9-NEXT: v_max_f32_e32 v21, v6, v22
+; GFX9-NEXT: v_cmp_o_f32_e64 s[8:9], v6, v22
+; GFX9-NEXT: v_max_f32_e32 v22, v7, v23
+; GFX9-NEXT: v_cmp_o_f32_e64 s[10:11], v7, v23
+; GFX9-NEXT: v_max_f32_e32 v23, v8, v24
+; GFX9-NEXT: v_cndmask_b32_e64 v0, v33, v1, s[12:13]
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v33, v32, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v8, v24
+; GFX9-NEXT: v_max_f32_e32 v34, v9, v25
+; GFX9-NEXT: v_cndmask_b32_e32 v8, v33, v23, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v9, v25
+; GFX9-NEXT: v_max_f32_e32 v35, v10, v26
+; GFX9-NEXT: v_cndmask_b32_e32 v9, v33, v34, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v10, v26
+; GFX9-NEXT: v_max_f32_e32 v36, v11, v27
+; GFX9-NEXT: v_cndmask_b32_e32 v10, v33, v35, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v11, v27
+; GFX9-NEXT: v_max_f32_e32 v37, v12, v28
+; GFX9-NEXT: v_cndmask_b32_e32 v11, v33, v36, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v12, v28
+; GFX9-NEXT: v_max_f32_e32 v16, v13, v29
+; GFX9-NEXT: v_cndmask_b32_e32 v12, v33, v37, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v13, v29
+; GFX9-NEXT: v_cndmask_b32_e32 v13, v33, v16, vcc
+; GFX9-NEXT: v_max_f32_e32 v16, v14, v30
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v14, v30
+; GFX9-NEXT: v_cndmask_b32_e32 v14, v33, v16, vcc
+; GFX9-NEXT: v_max_f32_e32 v16, v15, v31
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v15, v31
+; GFX9-NEXT: v_cndmask_b32_e64 v2, v33, v17, s[0:1]
+; GFX9-NEXT: v_cndmask_b32_e64 v3, v33, v18, s[2:3]
+; GFX9-NEXT: v_cndmask_b32_e64 v4, v33, v19, s[4:5]
+; GFX9-NEXT: v_cndmask_b32_e64 v5, v33, v20, s[6:7]
+; GFX9-NEXT: v_cndmask_b32_e64 v6, v33, v21, s[8:9]
+; GFX9-NEXT: v_cndmask_b32_e64 v7, v33, v22, s[10:11]
+; GFX9-NEXT: v_cndmask_b32_e32 v15, v33, v16, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v16f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v16
+; GFX12-NEXT: v_maximum_f32 v1, v1, v17
+; GFX12-NEXT: v_maximum_f32 v2, v2, v18
+; GFX12-NEXT: v_maximum_f32 v3, v3, v19
+; GFX12-NEXT: v_maximum_f32 v4, v4, v20
+; GFX12-NEXT: v_maximum_f32 v5, v5, v21
+; GFX12-NEXT: v_maximum_f32 v6, v6, v22
+; GFX12-NEXT: v_maximum_f32 v7, v7, v23
+; GFX12-NEXT: v_maximum_f32 v8, v8, v24
+; GFX12-NEXT: v_maximum_f32 v9, v9, v25
+; GFX12-NEXT: v_maximum_f32 v10, v10, v26
+; GFX12-NEXT: v_maximum_f32 v11, v11, v27
+; GFX12-NEXT: v_maximum_f32 v12, v12, v28
+; GFX12-NEXT: v_maximum_f32 v13, v13, v29
+; GFX12-NEXT: v_maximum_f32 v14, v14, v30
+; GFX12-NEXT: v_maximum_f32 v15, v15, v31
+; GFX12-NEXT: ; return to shader part epilog
%val = call <16 x float> @llvm.maximum.v16f32(<16 x float> %a, <16 x float> %b)
ret <16 x float> %val
}
define amdgpu_ps half @test_fmaximum_f16_vv(half %a, half %b) {
+; GFX9-LABEL: test_fmaximum_f16_vv:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f16_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
; GFX12-SDAG-TRUE16-LABEL: test_fmaximum_f16_vv:
; GFX12-SDAG-TRUE16: ; %bb.0:
; GFX12-SDAG-TRUE16-NEXT: v_maximum_f16 v0.l, v0.l, v1.l
@@ -136,35 +315,131 @@ define amdgpu_ps half @test_fmaximum_f16_vv(half %a, half %b) {
}
define amdgpu_ps half @test_fmaximum_f16_ss(half inreg %a, half inreg %b) {
-; GCN-LABEL: test_fmaximum_f16_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_maximum_f16 s0, s0, s1
-; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GCN-NEXT: v_mov_b32_e32 v0, s0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fmaximum_f16_ss:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-NEXT: v_max_f16_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f16_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_maximum_f16 s0, s0, s1
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call half @llvm.maximum.f16(half %a, half %b)
ret half %val
}
define amdgpu_ps <2 x half> @test_fmaximum_v2f16_vv(<2 x half> %a, <2 x half> %b) {
-; GCN-LABEL: test_fmaximum_v2f16_vv:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_maximum_f16 v0, v0, v1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v2f16_vv:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_pk_max_f16 v2, v0, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v3, v2, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v2f16_vv:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_pk_max_f16 v2, v0, v1
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v1
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v2f16_vv:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_maximum_f16 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)
ret <2 x half> %val
}
define amdgpu_ps <2 x half> @test_fmaximum_v2f16_ss(<2 x half> inreg %a, <2 x half> inreg %b) {
-; GCN-LABEL: test_fmaximum_v2f16_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_maximum_f16 v0, s0, s1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v2f16_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16
+; GFX9-SDAG-NEXT: v_pk_max_f16 v1, s0, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v3
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v2f16_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-GISEL-NEXT: s_lshr_b32 s1, s1, 16
+; GFX9-GISEL-NEXT: s_lshr_b32 s2, s0, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-GISEL-NEXT: v_pk_max_f16 v1, s0, v0
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s2, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s0, v0
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v2f16_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_maximum_f16 v0, s0, s1
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)
ret <2 x half> %val
}
define amdgpu_ps <3 x half> @test_fmaximum_v3f16_vv(<3 x half> %a, <3 x half> %b) {
+; GFX9-SDAG-LABEL: test_fmaximum_v3f16_vv:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_pk_max_f16 v4, v1, v3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc
+; GFX9-SDAG-NEXT: v_pk_max_f16 v3, v0, v2
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v3f16_vv:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_pk_max_f16 v4, v0, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v5, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v2
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v5, v4, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_pk_max_f16 v4, v1, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v1, v3
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
; GFX12-SDAG-LABEL: test_fmaximum_v3f16_vv:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: v_pk_maximum_f16 v0, v0, v2
@@ -187,6 +462,49 @@ define amdgpu_ps <3 x half> @test_fmaximum_v3f16_vv(<3 x half> %a, <3 x half> %b
}
define amdgpu_ps <3 x half> @test_fmaximum_v3f16_ss(<3 x half> inreg %a, <3 x half> inreg %b) {
+; GFX9-SDAG-LABEL: test_fmaximum_v3f16_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-SDAG-NEXT: v_pk_max_f16 v1, s1, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s2
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_pk_max_f16 v3, s0, v3
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v4
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v3f16_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: s_lshr_b32 s5, s2, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16
+; GFX9-GISEL-NEXT: v_pk_max_f16 v1, s0, v0
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s5
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v2, v4, v3, vcc
+; GFX9-GISEL-NEXT: v_pk_max_f16 v3, s1, v1
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s1, v1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
; GFX12-SDAG-LABEL: test_fmaximum_v3f16_ss:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: v_pk_maximum_f16 v0, s0, s2
@@ -206,97 +524,384 @@ define amdgpu_ps <3 x half> @test_fmaximum_v3f16_ss(<3 x half> inreg %a, <3 x ha
}
define amdgpu_ps <4 x half> @test_fmaximum_v4f16(<4 x half> %a, <4 x half> %b) {
-; GCN-LABEL: test_fmaximum_v4f16:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_maximum_f16 v0, v0, v2
-; GCN-NEXT: v_pk_maximum_f16 v1, v1, v3
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v4f16:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_pk_max_f16 v4, v1, v3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v6, v5, v4, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_pk_max_f16 v3, v0, v2
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0
+; GFX9-SDAG-NEXT: v_perm_b32 v1, v1, v6, s0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v4f16:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_pk_max_f16 v4, v0, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v0, v2
+; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v4
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v6, v5, vcc
+; GFX9-GISEL-NEXT: v_and_b32_e32 v2, 0xffff, v4
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v0, 16, v2
+; GFX9-GISEL-NEXT: v_pk_max_f16 v2, v1, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v1, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v6, v2, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v6, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v4f16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_maximum_f16 v0, v0, v2
+; GFX12-NEXT: v_pk_maximum_f16 v1, v1, v3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x half> @llvm.maximum.v4f16(<4 x half> %a, <4 x half> %b)
ret <4 x half> %val
}
define amdgpu_ps <4 x half> @test_fmaximum_v4f16_ss(<4 x half> inreg %a, <4 x half> inreg %b) {
-; GCN-LABEL: test_fmaximum_v4f16_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_maximum_f16 v0, s0, s2
-; GCN-NEXT: v_pk_maximum_f16 v1, s1, s3
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v4f16_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-SDAG-NEXT: s_lshr_b32 s3, s3, 16
+; GFX9-SDAG-NEXT: v_pk_max_f16 v1, s1, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v3, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_lshrrev_b32_e32 v1, 16, v1
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s2
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_pk_max_f16 v4, s0, v4
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, s1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v4, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v5
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-SDAG-NEXT: v_and_b32_e32 v2, 0xffff, v3
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v4f16_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT: s_lshr_b32 s2, s2, 16
+; GFX9-GISEL-NEXT: v_pk_max_f16 v1, s0, v0
+; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: s_lshr_b32 s2, s3, 16
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT: s_lshr_b32 s0, s1, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, s2
+; GFX9-GISEL-NEXT: v_pk_max_f16 v2, s1, v1
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s1, v1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v4, v2, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v4, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v4f16_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_maximum_f16 v0, s0, s2
+; GFX12-NEXT: v_pk_maximum_f16 v1, s1, s3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x half> @llvm.maximum.v4f16(<4 x half> %a, <4 x half> %b)
ret <4 x half> %val
}
define amdgpu_ps <2 x float> @test_fmaximum_f64_vv(double %a, double %b) {
-; GCN-LABEL: test_fmaximum_f64_vv:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f64 v[0:1], v[0:1], v[2:3]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_f64_vv:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_max_f64 v[4:5], v[0:1], v[2:3]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[2:3]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v4, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_f64_vv:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_max_f64 v[4:5], v[0:1], v[2:3]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[2:3]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f64_vv:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f64 v[0:1], v[0:1], v[2:3]
+; GFX12-NEXT: ; return to shader part epilog
%val = call double @llvm.maximum.f64(double %a, double %b)
%ret = bitcast double %val to <2 x float>
ret <2 x float> %ret
}
define amdgpu_ps <2 x float> @test_fmaximum_f64_ss(double inreg %a, double inreg %b) {
-; GCN-LABEL: test_fmaximum_f64_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f64 v[0:1], s[0:1], s[2:3]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_f64_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-SDAG-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_f64_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f64_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f64 v[0:1], s[0:1], s[2:3]
+; GFX12-NEXT: ; return to shader part epilog
%val = call double @llvm.maximum.f64(double %a, double %b)
%ret = bitcast double %val to <2 x float>
ret <2 x float> %ret
}
define amdgpu_ps <4 x float> @test_fmaximum_v2f64_ss(<2 x double> inreg %a, <2 x double> inreg %b) {
-; GCN-LABEL: test_fmaximum_v2f64_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f64 v[0:1], s[0:1], s[4:5]
-; GCN-NEXT: v_maximum_f64 v[2:3], s[2:3], s[6:7]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v2f64_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-SDAG-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-SDAG-NEXT: v_max_f64 v[4:5], s[2:3], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v6, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v6, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v6, s[0:1]
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v2f64_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-GISEL-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-GISEL-NEXT: v_max_f64 v[4:5], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v6, v3, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v6, v5, s[0:1]
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v2f64_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f64 v[0:1], s[0:1], s[4:5]
+; GFX12-NEXT: v_maximum_f64 v[2:3], s[2:3], s[6:7]
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x double> @llvm.maximum.v2f64(<2 x double> %a, <2 x double> %b)
%ret = bitcast <2 x double> %val to <4 x float>
ret <4 x float> %ret
}
define amdgpu_ps <8 x float> @test_fmaximum_v4f64(<4 x double> %a, <4 x double> %b) {
-; GCN-LABEL: test_fmaximum_v4f64:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f64 v[0:1], v[0:1], v[8:9]
-; GCN-NEXT: v_maximum_f64 v[2:3], v[2:3], v[10:11]
-; GCN-NEXT: v_maximum_f64 v[4:5], v[4:5], v[12:13]
-; GCN-NEXT: v_maximum_f64 v[6:7], v[6:7], v[14:15]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v4f64:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_max_f64 v[16:17], v[0:1], v[8:9]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[8:9]
+; GFX9-SDAG-NEXT: v_max_f64 v[8:9], v[2:3], v[10:11]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], v[2:3], v[10:11]
+; GFX9-SDAG-NEXT: v_max_f64 v[10:11], v[4:5], v[12:13]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], v[4:5], v[12:13]
+; GFX9-SDAG-NEXT: v_max_f64 v[12:13], v[6:7], v[14:15]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], v[6:7], v[14:15]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v7, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v16, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v17, v7, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v8, 0, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v9, v7, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v10, 0, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v11, v7, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v12, 0, s[4:5]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v13, v7, s[4:5]
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v4f64:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_max_f64 v[16:17], v[0:1], v[8:9]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[8:9]
+; GFX9-GISEL-NEXT: v_max_f64 v[8:9], v[2:3], v[10:11]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], v[2:3], v[10:11]
+; GFX9-GISEL-NEXT: v_max_f64 v[10:11], v[4:5], v[12:13]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], v[4:5], v[12:13]
+; GFX9-GISEL-NEXT: v_max_f64 v[12:13], v[6:7], v[14:15]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], v[6:7], v[14:15]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v18, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v16, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v18, v17, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v8, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v9, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v10, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v18, v11, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v12, s[4:5]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v18, v13, s[4:5]
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v4f64:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f64 v[0:1], v[0:1], v[8:9]
+; GFX12-NEXT: v_maximum_f64 v[2:3], v[2:3], v[10:11]
+; GFX12-NEXT: v_maximum_f64 v[4:5], v[4:5], v[12:13]
+; GFX12-NEXT: v_maximum_f64 v[6:7], v[6:7], v[14:15]
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x double> @llvm.maximum.v4f64(<4 x double> %a, <4 x double> %b)
%ret = bitcast <4 x double> %val to <8 x float>
ret <8 x float> %ret
}
define amdgpu_ps <8 x float> @test_fmaximum_v4f64_ss(<4 x double> inreg %a, <4 x double> inreg %b) {
-; GCN-LABEL: test_fmaximum_v4f64_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_maximum_f64 v[0:1], s[0:1], s[8:9]
-; GCN-NEXT: v_maximum_f64 v[2:3], s[2:3], s[10:11]
-; GCN-NEXT: v_maximum_f64 v[4:5], s[4:5], s[12:13]
-; GCN-NEXT: v_maximum_f64 v[6:7], s[6:7], s[14:15]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fmaximum_v4f64_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-SDAG-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v10, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s10
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s11
+; GFX9-SDAG-NEXT: v_max_f64 v[4:5], s[2:3], v[1:2]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[1:2]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s12
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s13
+; GFX9-SDAG-NEXT: v_max_f64 v[6:7], s[4:5], v[1:2]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], s[4:5], v[1:2]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s14
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s15
+; GFX9-SDAG-NEXT: v_max_f64 v[8:9], s[6:7], v[1:2]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], s[6:7], v[1:2]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v10, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v10, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v6, 0, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v7, v10, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v8, 0, s[4:5]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v9, v10, s[4:5]
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fmaximum_v4f64_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-GISEL-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s10
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s11
+; GFX9-GISEL-NEXT: v_max_f64 v[4:5], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s12
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s13
+; GFX9-GISEL-NEXT: v_max_f64 v[6:7], s[4:5], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], s[4:5], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s14
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s15
+; GFX9-GISEL-NEXT: v_max_f64 v[8:9], s[6:7], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], s[6:7], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v10, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v3, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v10, v5, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v6, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v10, v7, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v8, s[4:5]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v10, v9, s[4:5]
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_v4f64_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f64 v[0:1], s[0:1], s[8:9]
+; GFX12-NEXT: v_maximum_f64 v[2:3], s[2:3], s[10:11]
+; GFX12-NEXT: v_maximum_f64 v[4:5], s[4:5], s[12:13]
+; GFX12-NEXT: v_maximum_f64 v[6:7], s[6:7], s[14:15]
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x double> @llvm.maximum.v4f64(<4 x double> %a, <4 x double> %b)
%ret = bitcast <4 x double> %val to <8 x float>
ret <8 x float> %ret
}
define amdgpu_kernel void @fmaximumi_f32_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) {
-; GCN-LABEL: fmaximumi_f32_move_to_valu:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_clause 0x1
-; GCN-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
-; GCN-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
-; GCN-NEXT: v_mov_b32_e32 v0, 0
-; GCN-NEXT: s_wait_kmcnt 0x0
-; GCN-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
-; GCN-NEXT: s_wait_loadcnt 0x0
-; GCN-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS
-; GCN-NEXT: s_wait_loadcnt 0x0
-; GCN-NEXT: v_maximum_f32 v1, v1, v2
-; GCN-NEXT: global_store_b32 v0, v1, s[0:1]
-; GCN-NEXT: s_endpgm
+; GFX9-LABEL: fmaximumi_f32_move_to_valu:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: global_load_dword v1, v0, s[2:3] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_load_dword v2, v0, s[6:7] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f32_e32 v4, v1, v2
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT: s_endpgm
+;
+; GFX12-LABEL: fmaximumi_f32_move_to_valu:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX12-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
+; GFX12-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
+; GFX12-NEXT: s_wait_loadcnt 0x0
+; GFX12-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS
+; GFX12-NEXT: s_wait_loadcnt 0x0
+; GFX12-NEXT: v_maximum_f32 v1, v1, v2
+; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-NEXT: s_endpgm
%a = load volatile float, ptr addrspace(1) %aptr, align 4
%b = load volatile float, ptr addrspace(1) %bptr, align 4
%v = call float @llvm.maximum.f32(float %a, float %b)
@@ -305,6 +910,23 @@ define amdgpu_kernel void @fmaximumi_f32_move_to_valu(ptr addrspace(1) %out, ptr
}
define amdgpu_kernel void @fmaximum_f16_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) {
+; GFX9-LABEL: fmaximum_f16_move_to_valu:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: global_load_ushort v1, v0, s[2:3] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f16_e32 v4, v1, v2
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v1, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT: global_store_short v0, v1, s[0:1]
+; GFX9-NEXT: s_endpgm
+;
; GFX12-SDAG-TRUE16-LABEL: fmaximum_f16_move_to_valu:
; GFX12-SDAG-TRUE16: ; %bb.0:
; GFX12-SDAG-TRUE16-NEXT: s_clause 0x1
@@ -371,6 +993,40 @@ define amdgpu_kernel void @fmaximum_f16_move_to_valu(ptr addrspace(1) %out, ptr
ret void
}
+define amdgpu_ps float @test_fmaximum_f32_ieee_on(float %a, float %b) #0 {
+; GFX9-LABEL: test_fmaximum_f32_ieee_on:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f32_ieee_on:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call float @llvm.maximum.f32(float %a, float %b)
+ ret float %val
+}
+
+define amdgpu_ps float @test_fmaximum_f32_ieee_off(float %a, float %b) #1 {
+; GFX9-LABEL: test_fmaximum_f32_ieee_off:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_max_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fmaximum_f32_ieee_off:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_maximum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call float @llvm.maximum.f32(float %a, float %b)
+ ret float %val
+}
+
declare float @llvm.maximum.f32(float, float)
declare <2 x float> @llvm.maximum.v2f32(<2 x float>, <2 x float>)
declare <3 x float> @llvm.maximum.v3f32(<3 x float>, <3 x float>)
@@ -383,3 +1039,6 @@ declare <4 x half> @llvm.maximum.v4f16(<4 x half>, <4 x half>)
declare double @llvm.maximum.f64(double, double)
declare <2 x double> @llvm.maximum.v2f64(<2 x double>, <2 x double>)
declare <4 x double> @llvm.maximum.v4f64(<4 x double>, <4 x double>)
+
+attributes #0 = { nounwind "amdgpu-ieee"="true" }
+attributes #1 = { nounwind "amdgpu-ieee"="false" }
diff --git a/llvm/test/CodeGen/AMDGPU/fminimum.ll b/llvm/test/CodeGen/AMDGPU/fminimum.ll
index e1d35b52defee..2baca0aee87ff 100644
--- a/llvm/test/CodeGen/AMDGPU/fminimum.ll
+++ b/llvm/test/CodeGen/AMDGPU/fminimum.ll
@@ -1,117 +1,296 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-TRUE16 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-FAKE16 %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-TRUE16 %s
-; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s
define amdgpu_ps float @test_fminimum_f32_vv(float %a, float %b) {
-; GCN-LABEL: test_fminimum_f32_vv:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, v1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_f32_vv:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f32_vv:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.minimum.f32(float %a, float %b)
ret float %val
}
define amdgpu_ps float @test_fminimum_f32_ss(float inreg %a, float inreg %b) {
-; GCN-LABEL: test_fminimum_f32_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_minimum_f32 s0, s0, s1
-; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GCN-NEXT: v_mov_b32_e32 v0, s0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_f32_ss:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-NEXT: v_min_f32_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f32_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_minimum_f32 s0, s0, s1
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.minimum.f32(float %a, float %b)
ret float %val
}
define amdgpu_ps float @test_fminimum_f32_vs(float %a, float inreg %b) {
-; GCN-LABEL: test_fminimum_f32_vs:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, s0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_f32_vs:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f32_vs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.minimum.f32(float %a, float %b)
ret float %val
}
define amdgpu_ps float @test_fminimum_nnan_f32(float %a, float %b) {
-; GCN-LABEL: test_fminimum_nnan_f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, v1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_nnan_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_nnan_f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
%val = call nnan float @llvm.minimum.f32(float %a, float %b)
ret float %val
}
+define amdgpu_ps float @test_fminimum_nsz_f32(float %a, float %b) {
+; GFX9-LABEL: test_fminimum_nsz_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_nsz_f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call nsz float @llvm.minimum.f32(float %a, float %b)
+ ret float %val
+}
+
+define amdgpu_ps float @test_fminimum_signed_zero_f32() {
+; GFX9-LABEL: test_fminimum_signed_zero_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_bfrev_b32_e32 v0, 1
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_signed_zero_f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_bfrev_b32_e32 v0, 1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call float @llvm.minimum.f32(float -0.0, float 0.0)
+ ret float %val
+}
+
define amdgpu_ps <2 x float> @test_fminimum_v2f32(<2 x float> %a, <2 x float> %b) {
-; GCN-LABEL: test_fminimum_v2f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, v2
-; GCN-NEXT: v_minimum_f32 v1, v1, v3
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_v2f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v4, v0, v2
+; GFX9-NEXT: v_mov_b32_e32 v5, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v5, v4, vcc
+; GFX9-NEXT: v_min_f32_e32 v2, v1, v3
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v3
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v2f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v2
+; GFX12-NEXT: v_minimum_f32 v1, v1, v3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x float> @llvm.minimum.v2f32(<2 x float> %a, <2 x float> %b)
ret <2 x float> %val
}
define amdgpu_ps <2 x float> @test_fminimum_v2f32_ss(<2 x float> inreg %a, <2 x float> inreg %b) {
-; GCN-LABEL: test_fminimum_v2f32_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_minimum_f32 s0, s0, s2
-; GCN-NEXT: s_minimum_f32 s1, s1, s3
-; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GCN-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_v2f32_ss:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_min_f32_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: v_min_f32_e32 v3, s1, v1
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s1, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v2f32_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_minimum_f32 s0, s0, s2
+; GFX12-NEXT: s_minimum_f32 s1, s1, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x float> @llvm.minimum.v2f32(<2 x float> %a, <2 x float> %b)
ret <2 x float> %val
}
define amdgpu_ps <3 x float> @test_fminimum_v3f32(<3 x float> %a, <3 x float> %b) {
-; GCN-LABEL: test_fminimum_v3f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, v3
-; GCN-NEXT: v_minimum_f32 v1, v1, v4
-; GCN-NEXT: v_minimum_f32 v2, v2, v5
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_v3f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v6, v0, v3
+; GFX9-NEXT: v_mov_b32_e32 v7, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v3
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v7, v6, vcc
+; GFX9-NEXT: v_min_f32_e32 v3, v1, v4
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v4
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v7, v3, vcc
+; GFX9-NEXT: v_min_f32_e32 v3, v2, v5
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v5
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v7, v3, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v3f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v3
+; GFX12-NEXT: v_minimum_f32 v1, v1, v4
+; GFX12-NEXT: v_minimum_f32 v2, v2, v5
+; GFX12-NEXT: ; return to shader part epilog
%val = call <3 x float> @llvm.minimum.v3f32(<3 x float> %a, <3 x float> %b)
ret <3 x float> %val
}
define amdgpu_ps <4 x float> @test_fminimum_v4f32(<4 x float> %a, <4 x float> %b) {
-; GCN-LABEL: test_fminimum_v4f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, v4
-; GCN-NEXT: v_minimum_f32 v1, v1, v5
-; GCN-NEXT: v_minimum_f32 v2, v2, v6
-; GCN-NEXT: v_minimum_f32 v3, v3, v7
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_v4f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v8, v0, v4
+; GFX9-NEXT: v_mov_b32_e32 v9, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v4
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v9, v8, vcc
+; GFX9-NEXT: v_min_f32_e32 v4, v1, v5
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v5
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v9, v4, vcc
+; GFX9-NEXT: v_min_f32_e32 v4, v2, v6
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v6
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v9, v4, vcc
+; GFX9-NEXT: v_min_f32_e32 v4, v3, v7
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v3, v7
+; GFX9-NEXT: v_cndmask_b32_e32 v3, v9, v4, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v4f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v4
+; GFX12-NEXT: v_minimum_f32 v1, v1, v5
+; GFX12-NEXT: v_minimum_f32 v2, v2, v6
+; GFX12-NEXT: v_minimum_f32 v3, v3, v7
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x float> @llvm.minimum.v4f32(<4 x float> %a, <4 x float> %b)
ret <4 x float> %val
}
define amdgpu_ps <16 x float> @test_fminimum_v16f32(<16 x float> %a, <16 x float> %b) {
-; GCN-LABEL: test_fminimum_v16f32:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f32 v0, v0, v16
-; GCN-NEXT: v_minimum_f32 v1, v1, v17
-; GCN-NEXT: v_minimum_f32 v2, v2, v18
-; GCN-NEXT: v_minimum_f32 v3, v3, v19
-; GCN-NEXT: v_minimum_f32 v4, v4, v20
-; GCN-NEXT: v_minimum_f32 v5, v5, v21
-; GCN-NEXT: v_minimum_f32 v6, v6, v22
-; GCN-NEXT: v_minimum_f32 v7, v7, v23
-; GCN-NEXT: v_minimum_f32 v8, v8, v24
-; GCN-NEXT: v_minimum_f32 v9, v9, v25
-; GCN-NEXT: v_minimum_f32 v10, v10, v26
-; GCN-NEXT: v_minimum_f32 v11, v11, v27
-; GCN-NEXT: v_minimum_f32 v12, v12, v28
-; GCN-NEXT: v_minimum_f32 v13, v13, v29
-; GCN-NEXT: v_minimum_f32 v14, v14, v30
-; GCN-NEXT: v_minimum_f32 v15, v15, v31
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_v16f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v32, v1, v17
+; GFX9-NEXT: v_mov_b32_e32 v33, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v17
+; GFX9-NEXT: v_min_f32_e32 v1, v0, v16
+; GFX9-NEXT: v_cmp_o_f32_e64 s[12:13], v0, v16
+; GFX9-NEXT: v_min_f32_e32 v17, v2, v18
+; GFX9-NEXT: v_cmp_o_f32_e64 s[0:1], v2, v18
+; GFX9-NEXT: v_min_f32_e32 v18, v3, v19
+; GFX9-NEXT: v_cmp_o_f32_e64 s[2:3], v3, v19
+; GFX9-NEXT: v_min_f32_e32 v19, v4, v20
+; GFX9-NEXT: v_cmp_o_f32_e64 s[4:5], v4, v20
+; GFX9-NEXT: v_min_f32_e32 v20, v5, v21
+; GFX9-NEXT: v_cmp_o_f32_e64 s[6:7], v5, v21
+; GFX9-NEXT: v_min_f32_e32 v21, v6, v22
+; GFX9-NEXT: v_cmp_o_f32_e64 s[8:9], v6, v22
+; GFX9-NEXT: v_min_f32_e32 v22, v7, v23
+; GFX9-NEXT: v_cmp_o_f32_e64 s[10:11], v7, v23
+; GFX9-NEXT: v_min_f32_e32 v23, v8, v24
+; GFX9-NEXT: v_cndmask_b32_e64 v0, v33, v1, s[12:13]
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v33, v32, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v8, v24
+; GFX9-NEXT: v_min_f32_e32 v34, v9, v25
+; GFX9-NEXT: v_cndmask_b32_e32 v8, v33, v23, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v9, v25
+; GFX9-NEXT: v_min_f32_e32 v35, v10, v26
+; GFX9-NEXT: v_cndmask_b32_e32 v9, v33, v34, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v10, v26
+; GFX9-NEXT: v_min_f32_e32 v36, v11, v27
+; GFX9-NEXT: v_cndmask_b32_e32 v10, v33, v35, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v11, v27
+; GFX9-NEXT: v_min_f32_e32 v37, v12, v28
+; GFX9-NEXT: v_cndmask_b32_e32 v11, v33, v36, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v12, v28
+; GFX9-NEXT: v_min_f32_e32 v16, v13, v29
+; GFX9-NEXT: v_cndmask_b32_e32 v12, v33, v37, vcc
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v13, v29
+; GFX9-NEXT: v_cndmask_b32_e32 v13, v33, v16, vcc
+; GFX9-NEXT: v_min_f32_e32 v16, v14, v30
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v14, v30
+; GFX9-NEXT: v_cndmask_b32_e32 v14, v33, v16, vcc
+; GFX9-NEXT: v_min_f32_e32 v16, v15, v31
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v15, v31
+; GFX9-NEXT: v_cndmask_b32_e64 v2, v33, v17, s[0:1]
+; GFX9-NEXT: v_cndmask_b32_e64 v3, v33, v18, s[2:3]
+; GFX9-NEXT: v_cndmask_b32_e64 v4, v33, v19, s[4:5]
+; GFX9-NEXT: v_cndmask_b32_e64 v5, v33, v20, s[6:7]
+; GFX9-NEXT: v_cndmask_b32_e64 v6, v33, v21, s[8:9]
+; GFX9-NEXT: v_cndmask_b32_e64 v7, v33, v22, s[10:11]
+; GFX9-NEXT: v_cndmask_b32_e32 v15, v33, v16, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v16f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v16
+; GFX12-NEXT: v_minimum_f32 v1, v1, v17
+; GFX12-NEXT: v_minimum_f32 v2, v2, v18
+; GFX12-NEXT: v_minimum_f32 v3, v3, v19
+; GFX12-NEXT: v_minimum_f32 v4, v4, v20
+; GFX12-NEXT: v_minimum_f32 v5, v5, v21
+; GFX12-NEXT: v_minimum_f32 v6, v6, v22
+; GFX12-NEXT: v_minimum_f32 v7, v7, v23
+; GFX12-NEXT: v_minimum_f32 v8, v8, v24
+; GFX12-NEXT: v_minimum_f32 v9, v9, v25
+; GFX12-NEXT: v_minimum_f32 v10, v10, v26
+; GFX12-NEXT: v_minimum_f32 v11, v11, v27
+; GFX12-NEXT: v_minimum_f32 v12, v12, v28
+; GFX12-NEXT: v_minimum_f32 v13, v13, v29
+; GFX12-NEXT: v_minimum_f32 v14, v14, v30
+; GFX12-NEXT: v_minimum_f32 v15, v15, v31
+; GFX12-NEXT: ; return to shader part epilog
%val = call <16 x float> @llvm.minimum.v16f32(<16 x float> %a, <16 x float> %b)
ret <16 x float> %val
}
define amdgpu_ps half @test_fminimum_f16_vv(half %a, half %b) {
+; GFX9-LABEL: test_fminimum_f16_vv:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f16_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
; GFX12-SDAG-TRUE16-LABEL: test_fminimum_f16_vv:
; GFX12-SDAG-TRUE16: ; %bb.0:
; GFX12-SDAG-TRUE16-NEXT: v_minimum_f16 v0.l, v0.l, v1.l
@@ -136,35 +315,131 @@ define amdgpu_ps half @test_fminimum_f16_vv(half %a, half %b) {
}
define amdgpu_ps half @test_fminimum_f16_ss(half inreg %a, half inreg %b) {
-; GCN-LABEL: test_fminimum_f16_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_minimum_f16 s0, s0, s1
-; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GCN-NEXT: v_mov_b32_e32 v0, s0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: test_fminimum_f16_ss:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-NEXT: v_min_f16_e32 v1, s0, v0
+; GFX9-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f16_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_minimum_f16 s0, s0, s1
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call half @llvm.minimum.f16(half %a, half %b)
ret half %val
}
define amdgpu_ps <2 x half> @test_fminimum_v2f16_vv(<2 x half> %a, <2 x half> %b) {
-; GCN-LABEL: test_fminimum_v2f16_vv:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_minimum_f16 v0, v0, v1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v2f16_vv:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_pk_min_f16 v2, v0, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v3, v2, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v2f16_vv:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_pk_min_f16 v2, v0, v1
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v1
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v2f16_vv:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_minimum_f16 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.minimum.v2f16(<2 x half> %a, <2 x half> %b)
ret <2 x half> %val
}
define amdgpu_ps <2 x half> @test_fminimum_v2f16_ss(<2 x half> inreg %a, <2 x half> inreg %b) {
-; GCN-LABEL: test_fminimum_v2f16_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_minimum_f16 v0, s0, s1
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v2f16_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16
+; GFX9-SDAG-NEXT: v_pk_min_f16 v1, s0, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v3
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v2f16_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-GISEL-NEXT: s_lshr_b32 s1, s1, 16
+; GFX9-GISEL-NEXT: s_lshr_b32 s2, s0, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-GISEL-NEXT: v_pk_min_f16 v1, s0, v0
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s2, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s0, v0
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v2f16_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_minimum_f16 v0, s0, s1
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.minimum.v2f16(<2 x half> %a, <2 x half> %b)
ret <2 x half> %val
}
define amdgpu_ps <3 x half> @test_fminimum_v3f16_vv(<3 x half> %a, <3 x half> %b) {
+; GFX9-SDAG-LABEL: test_fminimum_v3f16_vv:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_pk_min_f16 v4, v1, v3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc
+; GFX9-SDAG-NEXT: v_pk_min_f16 v3, v0, v2
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v3f16_vv:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_pk_min_f16 v4, v0, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v5, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v2
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v5, v4, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_pk_min_f16 v4, v1, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v1, v3
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
; GFX12-SDAG-LABEL: test_fminimum_v3f16_vv:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: v_pk_minimum_f16 v0, v0, v2
@@ -187,6 +462,49 @@ define amdgpu_ps <3 x half> @test_fminimum_v3f16_vv(<3 x half> %a, <3 x half> %b
}
define amdgpu_ps <3 x half> @test_fminimum_v3f16_ss(<3 x half> inreg %a, <3 x half> inreg %b) {
+; GFX9-SDAG-LABEL: test_fminimum_v3f16_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-SDAG-NEXT: v_pk_min_f16 v1, s1, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s2
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_pk_min_f16 v3, s0, v3
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v4
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v3f16_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: s_lshr_b32 s5, s2, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16
+; GFX9-GISEL-NEXT: v_pk_min_f16 v1, s0, v0
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s5
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v2, v4, v3, vcc
+; GFX9-GISEL-NEXT: v_pk_min_f16 v3, s1, v1
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s1, v1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
; GFX12-SDAG-LABEL: test_fminimum_v3f16_ss:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: v_pk_minimum_f16 v0, s0, s2
@@ -206,97 +524,384 @@ define amdgpu_ps <3 x half> @test_fminimum_v3f16_ss(<3 x half> inreg %a, <3 x ha
}
define amdgpu_ps <4 x half> @test_fminimum_v4f16(<4 x half> %a, <4 x half> %b) {
-; GCN-LABEL: test_fminimum_v4f16:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_minimum_f16 v0, v0, v2
-; GCN-NEXT: v_pk_minimum_f16 v1, v1, v3
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v4f16:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_pk_min_f16 v4, v1, v3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v6, v5, v4, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_pk_min_f16 v3, v0, v2
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0
+; GFX9-SDAG-NEXT: v_perm_b32 v1, v1, v6, s0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v4f16:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_pk_min_f16 v4, v0, v2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v0, v2
+; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v4
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v6, v5, vcc
+; GFX9-GISEL-NEXT: v_and_b32_e32 v2, 0xffff, v4
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v0, 16, v2
+; GFX9-GISEL-NEXT: v_pk_min_f16 v2, v1, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v1, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v6, v2, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v6, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v4f16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_minimum_f16 v0, v0, v2
+; GFX12-NEXT: v_pk_minimum_f16 v1, v1, v3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x half> @llvm.minimum.v4f16(<4 x half> %a, <4 x half> %b)
ret <4 x half> %val
}
define amdgpu_ps <4 x half> @test_fminimum_v4f16_ss(<4 x half> inreg %a, <4 x half> inreg %b) {
-; GCN-LABEL: test_fminimum_v4f16_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_pk_minimum_f16 v0, s0, s2
-; GCN-NEXT: v_pk_minimum_f16 v1, s1, s3
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v4f16_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-SDAG-NEXT: s_lshr_b32 s3, s3, 16
+; GFX9-SDAG-NEXT: v_pk_min_f16 v1, s1, v1
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v3, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_lshrrev_b32_e32 v1, 16, v1
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s2
+; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc
+; GFX9-SDAG-NEXT: v_pk_min_f16 v4, s0, v4
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, s1
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v4, vcc
+; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v5
+; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-SDAG-NEXT: v_and_b32_e32 v2, 0xffff, v3
+; GFX9-SDAG-NEXT: v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v4f16_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT: s_lshr_b32 s2, s2, 16
+; GFX9-GISEL-NEXT: v_pk_min_f16 v1, s0, v0
+; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0
+; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc
+; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX9-GISEL-NEXT: s_lshr_b32 s2, s3, 16
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT: s_lshr_b32 s0, s1, 16
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, s2
+; GFX9-GISEL-NEXT: v_pk_min_f16 v2, s1, v1
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v3
+; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s1, v1
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v4, v2, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v4, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v4f16_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_minimum_f16 v0, s0, s2
+; GFX12-NEXT: v_pk_minimum_f16 v1, s1, s3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x half> @llvm.minimum.v4f16(<4 x half> %a, <4 x half> %b)
ret <4 x half> %val
}
define amdgpu_ps <2 x float> @test_fminimum_f64_vv(double %a, double %b) {
-; GCN-LABEL: test_fminimum_f64_vv:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f64 v[0:1], v[0:1], v[2:3]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_f64_vv:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_min_f64 v[4:5], v[0:1], v[2:3]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[2:3]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v4, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_f64_vv:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_min_f64 v[4:5], v[0:1], v[2:3]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[2:3]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f64_vv:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f64 v[0:1], v[0:1], v[2:3]
+; GFX12-NEXT: ; return to shader part epilog
%val = call double @llvm.minimum.f64(double %a, double %b)
%ret = bitcast double %val to <2 x float>
ret <2 x float> %ret
}
define amdgpu_ps <2 x float> @test_fminimum_f64_ss(double inreg %a, double inreg %b) {
-; GCN-LABEL: test_fminimum_f64_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f64 v[0:1], s[0:1], s[2:3]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_f64_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-SDAG-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_f64_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f64_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f64 v[0:1], s[0:1], s[2:3]
+; GFX12-NEXT: ; return to shader part epilog
%val = call double @llvm.minimum.f64(double %a, double %b)
%ret = bitcast double %val to <2 x float>
ret <2 x float> %ret
}
define amdgpu_ps <4 x float> @test_fminimum_v2f64_ss(<2 x double> inreg %a, <2 x double> inreg %b) {
-; GCN-LABEL: test_fminimum_v2f64_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f64 v[0:1], s[0:1], s[4:5]
-; GCN-NEXT: v_minimum_f64 v[2:3], s[2:3], s[6:7]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v2f64_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-SDAG-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-SDAG-NEXT: v_min_f64 v[4:5], s[2:3], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v6, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v6, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v6, s[0:1]
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v2f64_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-GISEL-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-GISEL-NEXT: v_min_f64 v[4:5], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v6, v3, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v6, v5, s[0:1]
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v2f64_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f64 v[0:1], s[0:1], s[4:5]
+; GFX12-NEXT: v_minimum_f64 v[2:3], s[2:3], s[6:7]
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x double> @llvm.minimum.v2f64(<2 x double> %a, <2 x double> %b)
%ret = bitcast <2 x double> %val to <4 x float>
ret <4 x float> %ret
}
define amdgpu_ps <8 x float> @test_fminimum_v4f64(<4 x double> %a, <4 x double> %b) {
-; GCN-LABEL: test_fminimum_v4f64:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f64 v[0:1], v[0:1], v[8:9]
-; GCN-NEXT: v_minimum_f64 v[2:3], v[2:3], v[10:11]
-; GCN-NEXT: v_minimum_f64 v[4:5], v[4:5], v[12:13]
-; GCN-NEXT: v_minimum_f64 v[6:7], v[6:7], v[14:15]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v4f64:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_min_f64 v[16:17], v[0:1], v[8:9]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[8:9]
+; GFX9-SDAG-NEXT: v_min_f64 v[8:9], v[2:3], v[10:11]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], v[2:3], v[10:11]
+; GFX9-SDAG-NEXT: v_min_f64 v[10:11], v[4:5], v[12:13]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], v[4:5], v[12:13]
+; GFX9-SDAG-NEXT: v_min_f64 v[12:13], v[6:7], v[14:15]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], v[6:7], v[14:15]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v7, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v16, 0, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v17, v7, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v8, 0, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v9, v7, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v10, 0, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v11, v7, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v12, 0, s[4:5]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v13, v7, s[4:5]
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v4f64:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_min_f64 v[16:17], v[0:1], v[8:9]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[8:9]
+; GFX9-GISEL-NEXT: v_min_f64 v[8:9], v[2:3], v[10:11]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], v[2:3], v[10:11]
+; GFX9-GISEL-NEXT: v_min_f64 v[10:11], v[4:5], v[12:13]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], v[4:5], v[12:13]
+; GFX9-GISEL-NEXT: v_min_f64 v[12:13], v[6:7], v[14:15]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], v[6:7], v[14:15]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v18, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v16, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v18, v17, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v8, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v9, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v10, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v18, v11, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v12, s[4:5]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v18, v13, s[4:5]
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v4f64:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f64 v[0:1], v[0:1], v[8:9]
+; GFX12-NEXT: v_minimum_f64 v[2:3], v[2:3], v[10:11]
+; GFX12-NEXT: v_minimum_f64 v[4:5], v[4:5], v[12:13]
+; GFX12-NEXT: v_minimum_f64 v[6:7], v[6:7], v[14:15]
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x double> @llvm.minimum.v4f64(<4 x double> %a, <4 x double> %b)
%ret = bitcast <4 x double> %val to <8 x float>
ret <8 x float> %ret
}
define amdgpu_ps <8 x float> @test_fminimum_v4f64_ss(<4 x double> inreg %a, <4 x double> inreg %b) {
-; GCN-LABEL: test_fminimum_v4f64_ss:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_minimum_f64 v[0:1], s[0:1], s[8:9]
-; GCN-NEXT: v_minimum_f64 v[2:3], s[2:3], s[10:11]
-; GCN-NEXT: v_minimum_f64 v[4:5], s[4:5], s[12:13]
-; GCN-NEXT: v_minimum_f64 v[6:7], s[6:7], s[14:15]
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-SDAG-LABEL: test_fminimum_v4f64_ss:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-SDAG-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v10, 0x7ff80000
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s10
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s11
+; GFX9-SDAG-NEXT: v_min_f64 v[4:5], s[2:3], v[1:2]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[1:2]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s12
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s13
+; GFX9-SDAG-NEXT: v_min_f64 v[6:7], s[4:5], v[1:2]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], s[4:5], v[1:2]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s14
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s15
+; GFX9-SDAG-NEXT: v_min_f64 v[8:9], s[6:7], v[1:2]
+; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], s[6:7], v[1:2]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v10, vcc
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v10, s[0:1]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v6, 0, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v7, v10, s[2:3]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v8, 0, s[4:5]
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v9, v10, s[4:5]
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: test_fminimum_v4f64_ss:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-GISEL-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s10
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s11
+; GFX9-GISEL-NEXT: v_min_f64 v[4:5], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s12
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s13
+; GFX9-GISEL-NEXT: v_min_f64 v[6:7], s[4:5], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], s[4:5], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s14
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s15
+; GFX9-GISEL-NEXT: v_min_f64 v[8:9], s[6:7], v[0:1]
+; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], s[6:7], v[0:1]
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v10, 0x7ff80000
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v3, vcc
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v10, v5, s[0:1]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v6, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v10, v7, s[2:3]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v8, s[4:5]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v10, v9, s[4:5]
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_v4f64_ss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f64 v[0:1], s[0:1], s[8:9]
+; GFX12-NEXT: v_minimum_f64 v[2:3], s[2:3], s[10:11]
+; GFX12-NEXT: v_minimum_f64 v[4:5], s[4:5], s[12:13]
+; GFX12-NEXT: v_minimum_f64 v[6:7], s[6:7], s[14:15]
+; GFX12-NEXT: ; return to shader part epilog
%val = call <4 x double> @llvm.minimum.v4f64(<4 x double> %a, <4 x double> %b)
%ret = bitcast <4 x double> %val to <8 x float>
ret <8 x float> %ret
}
define amdgpu_kernel void @fminimumi_f32_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) {
-; GCN-LABEL: fminimumi_f32_move_to_valu:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_clause 0x1
-; GCN-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
-; GCN-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
-; GCN-NEXT: v_mov_b32_e32 v0, 0
-; GCN-NEXT: s_wait_kmcnt 0x0
-; GCN-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
-; GCN-NEXT: s_wait_loadcnt 0x0
-; GCN-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS
-; GCN-NEXT: s_wait_loadcnt 0x0
-; GCN-NEXT: v_minimum_f32 v1, v1, v2
-; GCN-NEXT: global_store_b32 v0, v1, s[0:1]
-; GCN-NEXT: s_endpgm
+; GFX9-LABEL: fminimumi_f32_move_to_valu:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: global_load_dword v1, v0, s[2:3] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_load_dword v2, v0, s[6:7] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_min_f32_e32 v4, v1, v2
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT: s_endpgm
+;
+; GFX12-LABEL: fminimumi_f32_move_to_valu:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX12-NEXT: s_load_b64 s[4:5], s[4:5], 0x34
+; GFX12-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
+; GFX12-NEXT: s_wait_loadcnt 0x0
+; GFX12-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS
+; GFX12-NEXT: s_wait_loadcnt 0x0
+; GFX12-NEXT: v_minimum_f32 v1, v1, v2
+; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-NEXT: s_endpgm
%a = load volatile float, ptr addrspace(1) %aptr, align 4
%b = load volatile float, ptr addrspace(1) %bptr, align 4
%v = call float @llvm.minimum.f32(float %a, float %b)
@@ -305,6 +910,23 @@ define amdgpu_kernel void @fminimumi_f32_move_to_valu(ptr addrspace(1) %out, ptr
}
define amdgpu_kernel void @fminimum_f16_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) {
+; GFX9-LABEL: fminimum_f16_move_to_valu:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: global_load_ushort v1, v0, s[2:3] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_min_f16_e32 v4, v1, v2
+; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v1, v2
+; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT: global_store_short v0, v1, s[0:1]
+; GFX9-NEXT: s_endpgm
+;
; GFX12-SDAG-TRUE16-LABEL: fminimum_f16_move_to_valu:
; GFX12-SDAG-TRUE16: ; %bb.0:
; GFX12-SDAG-TRUE16-NEXT: s_clause 0x1
@@ -371,6 +993,40 @@ define amdgpu_kernel void @fminimum_f16_move_to_valu(ptr addrspace(1) %out, ptr
ret void
}
+define amdgpu_ps float @test_fminimum_f32_ieee_on(float %a, float %b) #0 {
+; GFX9-LABEL: test_fminimum_f32_ieee_on:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f32_ieee_on:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call float @llvm.minimum.f32(float %a, float %b)
+ ret float %val
+}
+
+define amdgpu_ps float @test_fminimum_f32_ieee_off(float %a, float %b) #1 {
+; GFX9-LABEL: test_fminimum_f32_ieee_off:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_min_f32_e32 v2, v0, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000
+; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: test_fminimum_f32_ieee_off:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_minimum_f32 v0, v0, v1
+; GFX12-NEXT: ; return to shader part epilog
+ %val = call float @llvm.minimum.f32(float %a, float %b)
+ ret float %val
+}
+
declare float @llvm.minimum.f32(float, float)
declare <2 x float> @llvm.minimum.v2f32(<2 x float>, <2 x float>)
declare <3 x float> @llvm.minimum.v3f32(<3 x float>, <3 x float>)
@@ -383,3 +1039,6 @@ declare <4 x half> @llvm.minimum.v4f16(<4 x half>, <4 x half>)
declare double @llvm.minimum.f64(double, double)
declare <2 x double> @llvm.minimum.v2f64(<2 x double>, <2 x double>)
declare <4 x double> @llvm.minimum.v4f64(<4 x double>, <4 x double>)
+
+attributes #0 = { nounwind "amdgpu-ieee"="true" }
+attributes #1 = { nounwind "amdgpu-ieee"="false" }
>From 2d11a0fe54ee835bd3febf91acac2272128f22bc Mon Sep 17 00:00:00 2001
From: Mirko Brkusanin <Mirko.Brkusanin at amd.com>
Date: Tue, 29 Jul 2025 13:05:22 +0200
Subject: [PATCH 2/2] clang-format
---
.../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp | 22 +++++++++----------
1 file changed, 10 insertions(+), 12 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 2dfa07ca9cfe2..53db9bc19d9cd 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -957,23 +957,21 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
FPOpActions.clampMaxNumElementsStrict(0, S32, 2);
}
- auto &MinNumMaxNumIeee = getActionDefinitionsBuilder(
- {G_FMINNUM_IEEE, G_FMAXNUM_IEEE});
+ auto &MinNumMaxNumIeee =
+ getActionDefinitionsBuilder({G_FMINNUM_IEEE, G_FMAXNUM_IEEE});
if (ST.hasVOP3PInsts()) {
MinNumMaxNumIeee.legalFor(FPTypesPK16)
- .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
- .clampMaxNumElements(0, S16, 2)
- .clampScalar(0, S16, S64)
- .scalarize(0);
+ .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
+ .clampMaxNumElements(0, S16, 2)
+ .clampScalar(0, S16, S64)
+ .scalarize(0);
} else if (ST.has16BitInsts()) {
- MinNumMaxNumIeee.legalFor(FPTypes16)
- .clampScalar(0, S16, S64)
- .scalarize(0);
+ MinNumMaxNumIeee.legalFor(FPTypes16).clampScalar(0, S16, S64).scalarize(0);
} else {
MinNumMaxNumIeee.legalFor(FPTypesBase)
- .clampScalar(0, S32, S64)
- .scalarize(0);
+ .clampScalar(0, S32, S64)
+ .scalarize(0);
}
auto &MinNumMaxNum = getActionDefinitionsBuilder(
@@ -2118,7 +2116,7 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
.legalFor(FPTypesPK16)
.clampMaxNumElements(0, S16, 2)
.scalarize(0);
- } else if (ST.hasVOP3PInsts()){
+ } else if (ST.hasVOP3PInsts()) {
getActionDefinitionsBuilder({G_FMINIMUM, G_FMAXIMUM})
.lowerFor({V2S16})
.clampMaxNumElementsStrict(0, S16, 2)
More information about the llvm-commits
mailing list