[llvm] [AMDGPU] Remove unused VGPRSingleUseHintInsts feature (PR #109769)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Sep 24 01:59:35 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Scott Egerton (ScottEgerton)
<details>
<summary>Changes</summary>
---
Patch is 94.50 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/109769.diff
25 Files Affected:
- (modified) llvm/docs/AMDGPUUsage.rst (+1-3)
- (modified) llvm/lib/Target/AMDGPU/AMDGPU.h (-3)
- (modified) llvm/lib/Target/AMDGPU/AMDGPU.td (-13)
- (removed) llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp (-245)
- (modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (-10)
- (modified) llvm/lib/Target/AMDGPU/CMakeLists.txt (-1)
- (modified) llvm/lib/Target/AMDGPU/GCNSubtarget.h (-3)
- (modified) llvm/lib/Target/AMDGPU/SIInstrInfo.td (-2)
- (modified) llvm/lib/Target/AMDGPU/SOPInstructions.td (-11)
- (modified) llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp (-18)
- (modified) llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h (+2)
- (modified) llvm/lib/Target/AMDGPU/VOP1Instructions.td (+3-15)
- (modified) llvm/lib/Target/AMDGPU/VOP2Instructions.td (+2-4)
- (modified) llvm/lib/Target/AMDGPU/VOP3Instructions.td (+16-19)
- (modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+4-8)
- (modified) llvm/lib/Target/AMDGPU/VOPCInstructions.td (+5-8)
- (modified) llvm/lib/Target/AMDGPU/VOPInstructions.td (+1-19)
- (removed) llvm/test/CodeGen/AMDGPU/insert-singleuse-vdst.mir (-1420)
- (removed) llvm/test/MC/AMDGPU/gfx1150_asm_sopp.s (-10)
- (modified) llvm/test/MC/AMDGPU/gfx11_unsupported.s (-3)
- (modified) llvm/test/MC/AMDGPU/gfx12_asm_sopp.s (-9)
- (modified) llvm/test/MC/Disassembler/AMDGPU/decode-err.txt (-5)
- (removed) llvm/test/MC/Disassembler/AMDGPU/gfx1150_dasm_sopp.txt (-10)
- (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_sopp.txt (-8)
- (modified) llvm/utils/gn/secondary/llvm/lib/Target/AMDGPU/BUILD.gn (-1)
``````````diff
diff --git a/llvm/docs/AMDGPUUsage.rst b/llvm/docs/AMDGPUUsage.rst
index 4b48b54b18bb99..9e11b13c101d47 100644
--- a/llvm/docs/AMDGPUUsage.rst
+++ b/llvm/docs/AMDGPUUsage.rst
@@ -611,9 +611,7 @@ Generic processor code objects are versioned. See :ref:`amdgpu-generic-processor
- ``gfx1152``
SALU floating point instructions
- and single-use VGPR hint
- instructions are not available
- on:
+ are not available on:
- ``gfx1150``
- ``gfx1151``
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.h b/llvm/lib/Target/AMDGPU/AMDGPU.h
index b2dd354e496a2e..4abb5a63ab6d2c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -405,9 +405,6 @@ extern char &SIModeRegisterID;
void initializeAMDGPUInsertDelayAluPass(PassRegistry &);
extern char &AMDGPUInsertDelayAluID;
-void initializeAMDGPUInsertSingleUseVDSTPass(PassRegistry &);
-extern char &AMDGPUInsertSingleUseVDSTID;
-
void initializeSIInsertHardClausesPass(PassRegistry &);
extern char &SIInsertHardClausesID;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index 5757ac0d4454d0..66474dfcbb5cc0 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -929,12 +929,6 @@ def FeatureSALUFloatInsts : SubtargetFeature<"salu-float",
"Has SALU floating point instructions"
>;
-def FeatureVGPRSingleUseHintInsts : SubtargetFeature<"vgpr-singleuse-hint",
- "HasVGPRSingleUseHintInsts",
- "true",
- "Has single-use VGPR hint instructions"
->;
-
def FeaturePseudoScalarTrans : SubtargetFeature<"pseudo-scalar-trans",
"HasPseudoScalarTrans",
"true",
@@ -1615,14 +1609,12 @@ def FeatureISAVersion11_5_0 : FeatureSet<
!listconcat(FeatureISAVersion11_Common.Features,
[FeatureSALUFloatInsts,
FeatureDPPSrc1SGPR,
- FeatureVGPRSingleUseHintInsts,
FeatureRequiredExportPriority])>;
def FeatureISAVersion11_5_1 : FeatureSet<
!listconcat(FeatureISAVersion11_Common.Features,
[FeatureSALUFloatInsts,
FeatureDPPSrc1SGPR,
- FeatureVGPRSingleUseHintInsts,
Feature1_5xVGPRs,
FeatureRequiredExportPriority])>;
@@ -1630,7 +1622,6 @@ def FeatureISAVersion11_5_2 : FeatureSet<
!listconcat(FeatureISAVersion11_Common.Features,
[FeatureSALUFloatInsts,
FeatureDPPSrc1SGPR,
- FeatureVGPRSingleUseHintInsts,
FeatureRequiredExportPriority])>;
def FeatureISAVersion12 : FeatureSet<
@@ -1663,7 +1654,6 @@ def FeatureISAVersion12 : FeatureSet<
FeatureSALUFloatInsts,
FeaturePseudoScalarTrans,
FeatureHasRestrictedSOffset,
- FeatureVGPRSingleUseHintInsts,
FeatureScalarDwordx3Loads,
FeatureDPPSrc1SGPR,
FeatureMaxHardClauseLength32,
@@ -2267,9 +2257,6 @@ def HasNotMADIntraFwdBug : Predicate<"!Subtarget->hasMADIntraFwdBug()">;
def HasSALUFloatInsts : Predicate<"Subtarget->hasSALUFloatInsts()">,
AssemblerPredicate<(all_of FeatureSALUFloatInsts)>;
-def HasVGPRSingleUseHintInsts : Predicate<"Subtarget->hasVGPRSingleUseHintInsts()">,
- AssemblerPredicate<(all_of FeatureVGPRSingleUseHintInsts)>;
-
def HasPseudoScalarTrans : Predicate<"Subtarget->hasPseudoScalarTrans()">,
AssemblerPredicate<(all_of FeaturePseudoScalarTrans)>;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp
deleted file mode 100644
index 43b3bf43fe56db..00000000000000
--- a/llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp
+++ /dev/null
@@ -1,245 +0,0 @@
-//===- AMDGPUInsertSingleUseVDST.cpp - Insert s_singleuse_vdst instructions ==//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-//
-/// \file
-/// Insert s_singleuse_vdst instructions on GFX11.5+ to mark regions of VALU
-/// instructions that produce single-use VGPR values. If the value is forwarded
-/// to the consumer instruction prior to VGPR writeback, the hardware can
-/// then skip (kill) the VGPR write.
-//
-//===----------------------------------------------------------------------===//
-
-#include "AMDGPU.h"
-#include "AMDGPUGenSearchableTables.inc"
-#include "GCNSubtarget.h"
-#include "SIInstrInfo.h"
-#include "SIRegisterInfo.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/StringRef.h"
-#include "llvm/CodeGen/MachineBasicBlock.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineInstr.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineOperand.h"
-#include "llvm/CodeGen/Register.h"
-#include "llvm/IR/DebugLoc.h"
-#include "llvm/MC/MCRegister.h"
-#include "llvm/MC/MCRegisterInfo.h"
-#include "llvm/Pass.h"
-#include <array>
-
-using namespace llvm;
-
-#define DEBUG_TYPE "amdgpu-insert-single-use-vdst"
-
-namespace {
-class AMDGPUInsertSingleUseVDST : public MachineFunctionPass {
-private:
- const SIInstrInfo *SII;
- class SingleUseInstruction {
- private:
- static const unsigned MaxSkipRange = 0b111;
- static const unsigned MaxNumberOfSkipRegions = 2;
-
- unsigned LastEncodedPositionEnd;
- MachineInstr *ProducerInstr;
-
- std::array<unsigned, MaxNumberOfSkipRegions + 1> SingleUseRegions;
- SmallVector<unsigned, MaxNumberOfSkipRegions> SkipRegions;
-
- // Adds a skip region into the instruction.
- void skip(const unsigned ProducerPosition) {
- while (LastEncodedPositionEnd + MaxSkipRange < ProducerPosition) {
- SkipRegions.push_back(MaxSkipRange);
- LastEncodedPositionEnd += MaxSkipRange;
- }
- SkipRegions.push_back(ProducerPosition - LastEncodedPositionEnd);
- LastEncodedPositionEnd = ProducerPosition;
- }
-
- bool currentRegionHasSpace() {
- const auto Region = SkipRegions.size();
- // The first region has an extra bit of encoding space.
- return SingleUseRegions[Region] <
- ((Region == MaxNumberOfSkipRegions) ? 0b1111U : 0b111U);
- }
-
- unsigned encodeImm() {
- // Handle the first Single Use Region separately as it has an extra bit
- // of encoding space.
- unsigned Imm = SingleUseRegions[SkipRegions.size()];
- unsigned ShiftAmount = 4;
- for (unsigned i = SkipRegions.size(); i > 0; i--) {
- Imm |= SkipRegions[i - 1] << ShiftAmount;
- ShiftAmount += 3;
- Imm |= SingleUseRegions[i - 1] << ShiftAmount;
- ShiftAmount += 3;
- }
- return Imm;
- }
-
- public:
- SingleUseInstruction(const unsigned ProducerPosition,
- MachineInstr *Producer)
- : LastEncodedPositionEnd(ProducerPosition + 1), ProducerInstr(Producer),
- SingleUseRegions({1, 0, 0}) {}
-
- // Returns false if adding a new single use producer failed. This happens
- // because it could not be encoded, either because there is no room to
- // encode another single use producer region or that this single use
- // producer is too far away to encode the amount of instructions to skip.
- bool tryAddProducer(const unsigned ProducerPosition, MachineInstr *MI) {
- // Producer is too far away to encode into this instruction or another
- // skip region is needed and SkipRegions.size() = 2 so there's no room for
- // another skip region, therefore a new instruction is needed.
- if (LastEncodedPositionEnd +
- (MaxSkipRange * (MaxNumberOfSkipRegions - SkipRegions.size())) <
- ProducerPosition)
- return false;
-
- // If a skip region is needed.
- if (LastEncodedPositionEnd != ProducerPosition ||
- !currentRegionHasSpace()) {
- // If the current region is out of space therefore a skip region would
- // be needed, but there is no room for another skip region.
- if (SkipRegions.size() == MaxNumberOfSkipRegions)
- return false;
- skip(ProducerPosition);
- }
-
- SingleUseRegions[SkipRegions.size()]++;
- LastEncodedPositionEnd = ProducerPosition + 1;
- ProducerInstr = MI;
- return true;
- }
-
- auto emit(const SIInstrInfo *SII) {
- return BuildMI(*ProducerInstr->getParent(), ProducerInstr, DebugLoc(),
- SII->get(AMDGPU::S_SINGLEUSE_VDST))
- .addImm(encodeImm());
- }
- };
-
-public:
- static char ID;
-
- AMDGPUInsertSingleUseVDST() : MachineFunctionPass(ID) {}
-
- void insertSingleUseInstructions(
- ArrayRef<std::pair<unsigned, MachineInstr *>> SingleUseProducers) const {
- SmallVector<SingleUseInstruction> Instructions;
-
- for (auto &[Position, MI] : SingleUseProducers) {
- // Encode this position into the last single use instruction if possible.
- if (Instructions.empty() ||
- !Instructions.back().tryAddProducer(Position, MI)) {
- // If not, add a new instruction.
- Instructions.push_back(SingleUseInstruction(Position, MI));
- }
- }
-
- for (auto &Instruction : Instructions)
- Instruction.emit(SII);
- }
-
- bool runOnMachineFunction(MachineFunction &MF) override {
- const auto &ST = MF.getSubtarget<GCNSubtarget>();
- if (!ST.hasVGPRSingleUseHintInsts())
- return false;
-
- SII = ST.getInstrInfo();
- const auto *TRI = &SII->getRegisterInfo();
- bool InstructionEmitted = false;
-
- for (MachineBasicBlock &MBB : MF) {
- DenseMap<MCRegUnit, unsigned> RegisterUseCount;
-
- // Handle boundaries at the end of basic block separately to avoid
- // false positives. If they are live at the end of a basic block then
- // assume it has more uses later on.
- for (const auto &Liveout : MBB.liveouts()) {
- for (MCRegUnitMaskIterator Units(Liveout.PhysReg, TRI); Units.isValid();
- ++Units) {
- const auto [Unit, Mask] = *Units;
- if ((Mask & Liveout.LaneMask).any())
- RegisterUseCount[Unit] = 2;
- }
- }
-
- SmallVector<std::pair<unsigned, MachineInstr *>>
- SingleUseProducerPositions;
-
- unsigned VALUInstrCount = 0;
- for (MachineInstr &MI : reverse(MBB.instrs())) {
- // All registers in all operands need to be single use for an
- // instruction to be marked as a single use producer.
- bool AllProducerOperandsAreSingleUse = true;
-
- // Gather a list of Registers used before updating use counts to avoid
- // double counting registers that appear multiple times in a single
- // MachineInstr.
- SmallVector<MCRegUnit> RegistersUsed;
-
- for (const auto &Operand : MI.all_defs()) {
- const auto Reg = Operand.getReg();
-
- const auto RegUnits = TRI->regunits(Reg);
- if (any_of(RegUnits, [&RegisterUseCount](const MCRegUnit Unit) {
- return RegisterUseCount[Unit] > 1;
- }))
- AllProducerOperandsAreSingleUse = false;
-
- // Reset uses count when a register is no longer live.
- for (const MCRegUnit Unit : RegUnits)
- RegisterUseCount.erase(Unit);
- }
-
- for (const auto &Operand : MI.all_uses()) {
- const auto Reg = Operand.getReg();
-
- // Count the number of times each register is read.
- for (const MCRegUnit Unit : TRI->regunits(Reg)) {
- if (!is_contained(RegistersUsed, Unit))
- RegistersUsed.push_back(Unit);
- }
- }
- for (const MCRegUnit Unit : RegistersUsed)
- RegisterUseCount[Unit]++;
-
- // Do not attempt to optimise across exec mask changes.
- if (MI.modifiesRegister(AMDGPU::EXEC, TRI) ||
- AMDGPU::isInvalidSingleUseConsumerInst(MI.getOpcode())) {
- for (auto &UsedReg : RegisterUseCount)
- UsedReg.second = 2;
- }
-
- if (!SIInstrInfo::isVALU(MI) ||
- AMDGPU::isInvalidSingleUseProducerInst(MI.getOpcode()))
- continue;
- if (AllProducerOperandsAreSingleUse) {
- SingleUseProducerPositions.push_back({VALUInstrCount, &MI});
- InstructionEmitted = true;
- }
- VALUInstrCount++;
- }
- insertSingleUseInstructions(SingleUseProducerPositions);
- }
- return InstructionEmitted;
- }
-};
-} // namespace
-
-char AMDGPUInsertSingleUseVDST::ID = 0;
-
-char &llvm::AMDGPUInsertSingleUseVDSTID = AMDGPUInsertSingleUseVDST::ID;
-
-INITIALIZE_PASS(AMDGPUInsertSingleUseVDST, DEBUG_TYPE,
- "AMDGPU Insert SingleUseVDST", false, false)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 04fdee0819b502..abd50748f2cc05 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -311,12 +311,6 @@ static cl::opt<bool> EnableSIModeRegisterPass(
cl::init(true),
cl::Hidden);
-// Enable GFX11.5+ s_singleuse_vdst insertion
-static cl::opt<bool>
- EnableInsertSingleUseVDST("amdgpu-enable-single-use-vdst",
- cl::desc("Enable s_singleuse_vdst insertion"),
- cl::init(false), cl::Hidden);
-
// Enable GFX11+ s_delay_alu insertion
static cl::opt<bool>
EnableInsertDelayAlu("amdgpu-enable-delay-alu",
@@ -450,7 +444,6 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
initializeAMDGPURewriteUndefForPHILegacyPass(*PR);
initializeAMDGPUUnifyMetadataPass(*PR);
initializeSIAnnotateControlFlowLegacyPass(*PR);
- initializeAMDGPUInsertSingleUseVDSTPass(*PR);
initializeAMDGPUInsertDelayAluPass(*PR);
initializeSIInsertHardClausesPass(*PR);
initializeSIInsertWaitcntsPass(*PR);
@@ -1518,9 +1511,6 @@ void GCNPassConfig::addPreEmitPass() {
// cases.
addPass(&PostRAHazardRecognizerID);
- if (isPassEnabled(EnableInsertSingleUseVDST, CodeGenOptLevel::Less))
- addPass(&AMDGPUInsertSingleUseVDSTID);
-
if (isPassEnabled(EnableInsertDelayAlu, CodeGenOptLevel::Less))
addPass(&AMDGPUInsertDelayAluID);
diff --git a/llvm/lib/Target/AMDGPU/CMakeLists.txt b/llvm/lib/Target/AMDGPU/CMakeLists.txt
index e813653158e5d9..7c883cc2017ddd 100644
--- a/llvm/lib/Target/AMDGPU/CMakeLists.txt
+++ b/llvm/lib/Target/AMDGPU/CMakeLists.txt
@@ -81,7 +81,6 @@ add_llvm_target(AMDGPUCodeGen
AMDGPUMCInstLower.cpp
AMDGPUMemoryUtils.cpp
AMDGPUIGroupLP.cpp
- AMDGPUInsertSingleUseVDST.cpp
AMDGPUMarkLastScratchLoad.cpp
AMDGPUMIRFormatter.cpp
AMDGPUOpenCLEnqueuedBlockLowering.cpp
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index a4ae8a1be32258..e6b7342d5fffcf 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -215,7 +215,6 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
bool HasPackedTID = false;
bool ScalarizeGlobal = false;
bool HasSALUFloatInsts = false;
- bool HasVGPRSingleUseHintInsts = false;
bool HasPseudoScalarTrans = false;
bool HasRestrictedSOffset = false;
@@ -1280,8 +1279,6 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
bool hasSALUFloatInsts() const { return HasSALUFloatInsts; }
- bool hasVGPRSingleUseHintInsts() const { return HasVGPRSingleUseHintInsts; }
-
bool hasPseudoScalarTrans() const { return HasPseudoScalarTrans; }
bool hasRestrictedSOffset() const { return HasRestrictedSOffset; }
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index c016be2fc6c0fb..087ca1f954464d 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -2409,8 +2409,6 @@ class VOPProfile <list<ValueType> _ArgVT, bit _EnableClamp = 0> {
field bit EnableClamp = _EnableClamp;
field bit IsTrue16 = 0;
field bit IsRealTrue16 = 0;
- field bit IsInvalidSingleUseConsumer = 0;
- field bit IsInvalidSingleUseProducer = 0;
field ValueType DstVT = ArgVT[0];
field ValueType Src0VT = ArgVT[1];
diff --git a/llvm/lib/Target/AMDGPU/SOPInstructions.td b/llvm/lib/Target/AMDGPU/SOPInstructions.td
index 2e73a1a15f6b32..9da27a7c7ee7d6 100644
--- a/llvm/lib/Target/AMDGPU/SOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SOPInstructions.td
@@ -1752,11 +1752,6 @@ let OtherPredicates = [HasExportInsts] in
"$simm16">;
} // End SubtargetPredicate = isGFX11Plus
-let SubtargetPredicate = HasVGPRSingleUseHintInsts in {
- def S_SINGLEUSE_VDST :
- SOPP_Pseudo<"s_singleuse_vdst", (ins s16imm:$simm16), "$simm16">;
-} // End SubtargetPredicate = HasVGPRSingeUseHintInsts
-
let SubtargetPredicate = isGFX12Plus, hasSideEffects = 1 in {
def S_WAIT_LOADCNT :
SOPP_Pseudo<"s_wait_loadcnt", (ins s16imm:$simm16), "$simm16",
@@ -2676,12 +2671,6 @@ defm S_ICACHE_INV : SOPP_Real_32_gfx11_gfx12<0x03c>;
defm S_BARRIER : SOPP_Real_32_gfx11<0x03d>;
-//===----------------------------------------------------------------------===//
-// SOPP - GFX1150, GFX12.
-//===----------------------------------------------------------------------===//
-
-defm S_SINGLEUSE_VDST : SOPP_Real_32_gfx11_gfx12<0x013>;
-
//===----------------------------------------------------------------------===//
// SOPP - GFX6, GFX7, GFX8, GFX9, GFX10
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index 8b5ec8793d84a2..f32c82f1e4ba4c 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -379,12 +379,6 @@ struct VOPTrue16Info {
bool IsTrue16;
};
-struct SingleUseExceptionInfo {
- uint16_t Opcode;
- bool IsInvalidSingleUseConsumer;
- bool IsInvalidSingleUseProducer;
-};
-
struct FP8DstByteSelInfo {
uint16_t Opcode;
bool HasFP8DstByteSel;
@@ -396,8 +390,6 @@ struct FP8DstByteSelInfo {
#define GET_MTBUFInfoTable_IMPL
#define GET_MUBUFInfoTable_DECL
#define GET_MUBUFInfoTable_IMPL
-#define GET_SingleUseExceptionTable_DECL
-#define GET_SingleUseExceptionTable_IMPL
#define GET_SMInfoTable_DECL
#define GET_SMInfoTable_IMPL
#define GET_VOP1InfoTable_DECL
@@ -626,16 +618,6 @@ bool isTrue16Inst(unsigned Opc) {
return Info ? Info->IsTrue16 : false;
}
-bool isInvalidSingleUseConsumerInst(unsigned Opc) {
- const SingleUseExceptionInfo *Info = getSingleUseExceptionHelper(Opc);
- return Info && Info->IsInvalidSingleUseConsumer;
-}
-
-bool isInvalidSingleUseProducerInst(unsigned Opc) {
- const SingleUseExceptionInfo *Info = getSingleUseExceptionHelper(Opc);
- return Info && Info->IsInvalidSingleUseProducer;
-}
-
bool isFP8DstSelInst(unsigned Opc) {
const FP8DstByteSelInfo *Info = getFP8DstByteSelHelper(Opc);
return Info ? Info->HasFP8DstByteSel : false;
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
index 35c080d8e0bebc..da37534f2fa4ff 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -870,6 +870,8 @@ bool isInvalidSingleUseConsumerInst(unsigned Opc);
LLVM_READONLY
bool isInvalidSingleUseProducerInst(unsigned Opc);
+bool isDPMACCInstruction(unsigned Opc);
+
LLVM_READONLY
unsigned mapWMMA2AddrTo3AddrOpcode(unsigned Opc);
diff --git a/llvm/lib/Target/AMDGPU/VOP1Instructions.td b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
index 33f2f9f1f5c5b9..bd805059705783 100644
--- a/llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ b/llvm/lib/Target/AMDGPU...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/109769
More information about the llvm-commits
mailing list