[llvm] [AMDGPU] New AMDGPUInsertSingleUseVDST pass (PR #72388)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Nov 15 05:48:54 PST 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Jay Foad (jayfoad)
<details>
<summary>Changes</summary>
Add support for emitting GFX11.5 s_singleuse_vdst instructions. This is
a power saving feature whereby the compiler can annotate VALU
instructions whose results are known to have only a single use, so the
hardware can in some cases avoid writing the result back to VGPR RAM.
To begin with the pass is disabled by default because of one missing
feature: we need an exclusion list of opcodes that never qualify as
single-use producers and/or consumers. A future patch will implement
this and enable the pass by default.
---
Patch is 44.05 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/72388.diff
5 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/AMDGPU.h (+3)
- (added) llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp (+120)
- (modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (+10)
- (modified) llvm/lib/Target/AMDGPU/CMakeLists.txt (+1)
- (added) llvm/test/CodeGen/AMDGPU/insert-singleuse_vdst.mir (+1023)
``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.h b/llvm/lib/Target/AMDGPU/AMDGPU.h
index 403014db56171ac..323560a46f31de2 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -335,6 +335,9 @@ 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/AMDGPUInsertSingleUseVDST.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp
new file mode 100644
index 000000000000000..88a77b22d989127
--- /dev/null
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInsertSingleUseVDST.cpp
@@ -0,0 +1,120 @@
+//===- 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 "GCNSubtarget.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "SIInstrInfo.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/STLExtras.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/CodeGen/TargetSubtargetInfo.h"
+#include "llvm/IR/DebugLoc.h"
+#include "llvm/MC/MCRegister.h"
+#include "llvm/Pass.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "amdgpu-insert-single-use-vdst"
+
+namespace {
+class AMDGPUInsertSingleUseVDST : public MachineFunctionPass {
+private:
+ const SIInstrInfo *SII;
+
+public:
+ static char ID;
+
+ AMDGPUInsertSingleUseVDST() : MachineFunctionPass(ID) {}
+
+ void emitSingleUseVDST(MachineInstr &MI) const {
+ BuildMI(*MI.getParent(), MI, DebugLoc(), SII->get(AMDGPU::S_SINGLEUSE_VDST))
+ .addImm(0x1);
+ }
+
+ bool runOnMachineFunction(MachineFunction &MF) override {
+ const auto &ST = MF.getSubtarget<GCNSubtarget>();
+ if (!ST.hasVGPRSingleUseHintInsts())
+ return false;
+
+ SII = ST.getInstrInfo();
+ const auto *TRI = MF.getSubtarget().getRegisterInfo();
+ bool InstructionEmitted = false;
+
+ for (MachineBasicBlock &MBB : MF) {
+ DenseMap<MCPhysReg, unsigned> RegisterUseCount; // TODO: MCRegUnits
+
+ // 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 &Liveouts : MBB.liveouts())
+ RegisterUseCount[Liveouts.PhysReg] = 2;
+
+ 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;
+
+ for (const auto &Operand : MI.operands()) {
+ if (!Operand.isReg())
+ continue;
+ const auto Reg = Operand.getReg();
+
+ // Count the number of times each register is read.
+ if (Operand.readsReg())
+ RegisterUseCount[Reg]++;
+
+ // Do not attempt to optimise across exec mask changes.
+ if (MI.modifiesRegister(AMDGPU::EXEC, TRI)) {
+ for (auto &UsedReg : RegisterUseCount)
+ UsedReg.second = 2;
+ }
+
+ // If we are at the point where the register first became live,
+ // check if the operands are single use.
+ if (!MI.modifiesRegister(Reg, TRI))
+ continue;
+ if (RegisterUseCount[Reg] > 1)
+ AllProducerOperandsAreSingleUse = false;
+ // Reset uses count when a register is no longer live.
+ RegisterUseCount.erase(Reg);
+ }
+ if (AllProducerOperandsAreSingleUse && SIInstrInfo::isVALU(MI)) {
+ // TODO: Replace with candidate logging for instruction grouping
+ // later.
+ emitSingleUseVDST(MI);
+ InstructionEmitted = true;
+ }
+ }
+ }
+ 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 951ed9420594b19..0c38fa32c6f33a8 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -286,6 +286,12 @@ 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",
@@ -404,6 +410,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
initializeAMDGPURewriteUndefForPHILegacyPass(*PR);
initializeAMDGPUUnifyMetadataPass(*PR);
initializeSIAnnotateControlFlowPass(*PR);
+ initializeAMDGPUInsertSingleUseVDSTPass(*PR);
initializeAMDGPUInsertDelayAluPass(*PR);
initializeSIInsertHardClausesPass(*PR);
initializeSIInsertWaitcntsPass(*PR);
@@ -1448,6 +1455,9 @@ 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 0c0720890794b66..53a33f8210d2a84 100644
--- a/llvm/lib/Target/AMDGPU/CMakeLists.txt
+++ b/llvm/lib/Target/AMDGPU/CMakeLists.txt
@@ -77,6 +77,7 @@ add_llvm_target(AMDGPUCodeGen
AMDGPUMacroFusion.cpp
AMDGPUMCInstLower.cpp
AMDGPUIGroupLP.cpp
+ AMDGPUInsertSingleUseVDST.cpp
AMDGPUMIRFormatter.cpp
AMDGPUOpenCLEnqueuedBlockLowering.cpp
AMDGPUPerfHintAnalysis.cpp
diff --git a/llvm/test/CodeGen/AMDGPU/insert-singleuse_vdst.mir b/llvm/test/CodeGen/AMDGPU/insert-singleuse_vdst.mir
new file mode 100644
index 000000000000000..8480f8d211ebac1
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/insert-singleuse_vdst.mir
@@ -0,0 +1,1023 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+# RUN: llc -march=amdgcn -mcpu=gfx1150 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs -run-pass=amdgpu-insert-single-use-vdst %s -o - | FileCheck %s
+
+---
+name: valu_dep_1
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; CHECK-LABEL: name: valu_dep_1
+ ; CHECK: liveins: $vgpr0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_2
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; CHECK-LABEL: name: valu_dep_2
+ ; CHECK: liveins: $vgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_3
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-LABEL: name: valu_dep_3
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-LABEL: name: valu_dep_4
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_5
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+ ; CHECK-LABEL: name: valu_dep_5
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr4 = V_ADD_U32_e32 $vgpr4, $vgpr4, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+ $vgpr4 = V_ADD_U32_e32 $vgpr4, $vgpr4, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: multiple_uses_1
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-LABEL: name: multiple_uses_1
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: multiple_uses_2
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-LABEL: name: multiple_uses_2
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: multiple_uses_3
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-LABEL: name: multiple_uses_3
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr1, $vgpr0, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr3 = V_ADD_U32_e32 $vgpr1, $vgpr0, implicit $exec
+...
+
+---
+name: basic_block_1
+tracksRegLiveness: true
+
+body: |
+ ; CHECK-LABEL: name: basic_block_1
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ successors: %bb.1
+
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ bb.1:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: basic_block_2
+tracksRegLiveness: true
+body: |
+ ; CHECK-LABEL: name: basic_block_2
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = IMPLICIT_DEF
+ ; CHECK-NEXT: $vgpr1 = IMPLICIT_DEF
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ successors: %bb.1
+
+ $vgpr0 = IMPLICIT_DEF
+ $vgpr1 = IMPLICIT_DEF
+
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ bb.1:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+
+ $vgpr3 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+ $vgpr3 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+...
+
+---
+name: basic_block_3
+tracksRegLiveness: true
+body: |
+ ; CHECK-LABEL: name: basic_block_3
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = IMPLICIT_DEF
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ successors: %bb.1
+
+ $vgpr0 = IMPLICIT_DEF
+
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ bb.1:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+...
+
+---
+name: exec_mask_1
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-LABEL: name: exec_mask_1
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1234
+ ; CHECK-NEXT: $exec = COPY $vgpr0
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+ $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ %9:_(s64) = G_CONSTANT i64 1234
+ $exec = COPY $vgpr0
+ $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ $vgpr3 = V_ADD_U32_e32 $vgpr0, $vgpr1, implicit $exec
+...
+
+---
+name: exec_mask_2
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; CHECK-LABEL: name: exec_mask_2
+ ; CHECK: liveins: $vgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1234
+ ; CHECK-NEXT: $exec = COPY $vgpr0
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+ ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+ ; CHECK-NEXT: $vgpr3 = V...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/72388
More information about the llvm-commits
mailing list