[AMDGPU][NewPM] Port SILowerWWMCopies to NPM

Author: optimisan

llvm/lib/Target/AMDGPU/AMDGPU.h

@@ -180,8 +180,8 @@ extern char &SIFixSGPRCopiesLegacyID;
 void initializeSIFixVGPRCopiesPass(PassRegistry &);
 extern char &SIFixVGPRCopiesID;
 
-void initializeSILowerWWMCopiesPass(PassRegistry &);
-extern char &SILowerWWMCopiesID;
+void initializeSILowerWWMCopiesLegacyPass(PassRegistry &);
+extern char &SILowerWWMCopiesLegacyID;
 
 void initializeSILowerI1CopiesLegacyPass(PassRegistry &);
 extern char &SILowerI1CopiesLegacyID;

llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def

@@ -104,6 +104,7 @@ MACHINE_FUNCTION_PASS("gcn-dpp-combine", GCNDPPCombinePass())
 MACHINE_FUNCTION_PASS("si-load-store-opt", SILoadStoreOptimizerPass())
 MACHINE_FUNCTION_PASS("si-lower-control-flow", SILowerControlFlowPass())
 MACHINE_FUNCTION_PASS("si-lower-sgpr-spills", SILowerSGPRSpillsPass())
+MACHINE_FUNCTION_PASS("si-lower-wwm-copies", SILowerWWMCopiesPass())
 MACHINE_FUNCTION_PASS("si-opt-vgpr-liverange", SIOptimizeVGPRLiveRangePass())
 MACHINE_FUNCTION_PASS("si-optimize-exec-masking", SIOptimizeExecMaskingPass())
 MACHINE_FUNCTION_PASS("si-peephole-sdwa", SIPeepholeSDWAPass())

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -40,6 +40,7 @@
 #include "SILoadStoreOptimizer.h"
 #include "SILowerControlFlow.h"
 #include "SILowerSGPRSpills.h"
+#include "SILowerWWMCopies.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIMachineScheduler.h"
 #include "SIOptimizeExecMasking.h"
@@ -482,7 +483,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   initializeAMDGPUGlobalISelDivergenceLoweringPass(*PR);
   initializeAMDGPURegBankSelectPass(*PR);
   initializeAMDGPURegBankLegalizePass(*PR);
-  initializeSILowerWWMCopiesPass(*PR);
+  initializeSILowerWWMCopiesLegacyPass(*PR);
   initializeAMDGPUMarkLastScratchLoadPass(*PR);
   initializeSILowerSGPRSpillsLegacyPass(*PR);
   initializeSIFixSGPRCopiesLegacyPass(*PR);
@@ -1581,7 +1582,7 @@ bool GCNPassConfig::addRegAssignAndRewriteFast() {
   // For allocating other wwm register operands.
   addPass(createWWMRegAllocPass(false));
 
-  addPass(&SILowerWWMCopiesID);
+  addPass(&SILowerWWMCopiesLegacyID);
   addPass(&AMDGPUReserveWWMRegsID);
 
   // For allocating per-thread VGPRs.
@@ -1617,7 +1618,7 @@ bool GCNPassConfig::addRegAssignAndRewriteOptimized() {
 
   // For allocating other whole wave mode registers.
   addPass(createWWMRegAllocPass(true));
-  addPass(&SILowerWWMCopiesID);
+  addPass(&SILowerWWMCopiesLegacyID);
   addPass(createVirtRegRewriter(false));
   addPass(&AMDGPUReserveWWMRegsID);
 

llvm/lib/Target/AMDGPU/SILowerWWMCopies.cpp

@@ -15,6 +15,7 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "SILowerWWMCopies.h"
 #include "AMDGPU.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
@@ -30,12 +31,30 @@ using namespace llvm;
 
 namespace {
 
-class SILowerWWMCopies : public MachineFunctionPass {
+class SILowerWWMCopies {
+public:
+  SILowerWWMCopies(LiveIntervals *LIS, SlotIndexes *SI, VirtRegMap *VRM)
+      : LIS(LIS), Indexes(SI), VRM(VRM) {}
+  bool run(MachineFunction &MF);
+
+private:
+  bool isSCCLiveAtMI(const MachineInstr &MI);
+  void addToWWMSpills(MachineFunction &MF, Register Reg);
+
+  LiveIntervals *LIS;
+  SlotIndexes *Indexes;
+  VirtRegMap *VRM;
+  const SIRegisterInfo *TRI;
+  const MachineRegisterInfo *MRI;
+  SIMachineFunctionInfo *MFI;
+};
+
+class SILowerWWMCopiesLegacy : public MachineFunctionPass {
 public:
   static char ID;
 
-  SILowerWWMCopies() : MachineFunctionPass(ID) {
-    initializeSILowerWWMCopiesPass(*PassRegistry::getPassRegistry());
+  SILowerWWMCopiesLegacy() : MachineFunctionPass(ID) {
+    initializeSILowerWWMCopiesLegacyPass(*PassRegistry::getPassRegistry());
   }
 
   bool runOnMachineFunction(MachineFunction &MF) override;
@@ -46,31 +65,20 @@ class SILowerWWMCopies : public MachineFunctionPass {
     AU.setPreservesAll();
     MachineFunctionPass::getAnalysisUsage(AU);
   }
-
-private:
-  bool isSCCLiveAtMI(const MachineInstr &MI);
-  void addToWWMSpills(MachineFunction &MF, Register Reg);
-
-  LiveIntervals *LIS;
-  SlotIndexes *Indexes;
-  VirtRegMap *VRM;
-  const SIRegisterInfo *TRI;
-  const MachineRegisterInfo *MRI;
-  SIMachineFunctionInfo *MFI;
 };
 
 } // End anonymous namespace.
 
-INITIALIZE_PASS_BEGIN(SILowerWWMCopies, DEBUG_TYPE, "SI Lower WWM Copies",
+INITIALIZE_PASS_BEGIN(SILowerWWMCopiesLegacy, DEBUG_TYPE, "SI Lower WWM Copies",
                       false, false)
 INITIALIZE_PASS_DEPENDENCY(LiveIntervalsWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(VirtRegMapWrapperLegacy)
-INITIALIZE_PASS_END(SILowerWWMCopies, DEBUG_TYPE, "SI Lower WWM Copies", false,
-                    false)
+INITIALIZE_PASS_END(SILowerWWMCopiesLegacy, DEBUG_TYPE, "SI Lower WWM Copies",
+                    false, false)
 
-char SILowerWWMCopies::ID = 0;
+char SILowerWWMCopiesLegacy::ID = 0;
 
-char &llvm::SILowerWWMCopiesID = SILowerWWMCopies::ID;
+char &llvm::SILowerWWMCopiesLegacyID = SILowerWWMCopiesLegacy::ID;
 
 bool SILowerWWMCopies::isSCCLiveAtMI(const MachineInstr &MI) {
   // We can't determine the liveness info if LIS isn't available. Early return
@@ -90,23 +98,44 @@ void SILowerWWMCopies::addToWWMSpills(MachineFunction &MF, Register Reg) {
   if (Reg.isPhysical())
     return;
 
+  // FIXME: VRM is not an optional requirement, we check it for assignment here
   MCRegister PhysReg = VRM->getPhys(Reg);
   assert(PhysReg && "should have allocated a physical register");
 
   MFI->allocateWWMSpill(MF, PhysReg);
 }
 
-bool SILowerWWMCopies::runOnMachineFunction(MachineFunction &MF) {
+bool SILowerWWMCopiesLegacy::runOnMachineFunction(MachineFunction &MF) {
+  auto *LISWrapper = getAnalysisIfAvailable<LiveIntervalsWrapperPass>();
+  auto *LIS = LISWrapper ? &LISWrapper->getLIS() : nullptr;
+
+  auto *SIWrapper = getAnalysisIfAvailable<SlotIndexesWrapperPass>();
+  auto *Indexes = SIWrapper ? &SIWrapper->getSI() : nullptr;
+
+  auto *VRMWrapper = getAnalysisIfAvailable<VirtRegMapWrapperLegacy>();
+  auto *VRM = VRMWrapper ? &VRMWrapper->getVRM() : nullptr;
+
+  SILowerWWMCopies Impl(LIS, Indexes, VRM);
+  return Impl.run(MF);
+}
+
+PreservedAnalyses
+SILowerWWMCopiesPass::run(MachineFunction &MF,
+                          MachineFunctionAnalysisManager &MFAM) {
+  auto *LIS = MFAM.getCachedResult<LiveIntervalsAnalysis>(MF);
+  auto *Indexes = MFAM.getCachedResult<SlotIndexesAnalysis>(MF);
+  auto *VRM = MFAM.getCachedResult<VirtRegMapAnalysis>(MF);
+
+  SILowerWWMCopies Impl(LIS, Indexes, VRM);
+  Impl.run(MF);
+  return PreservedAnalyses::all();
+}
+
+bool SILowerWWMCopies::run(MachineFunction &MF) {
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   const SIInstrInfo *TII = ST.getInstrInfo();
 
   MFI = MF.getInfo<SIMachineFunctionInfo>();
-  auto *LISWrapper = getAnalysisIfAvailable<LiveIntervalsWrapperPass>();
-  LIS = LISWrapper ? &LISWrapper->getLIS() : nullptr;
-  auto *SIWrapper = getAnalysisIfAvailable<SlotIndexesWrapperPass>();
-  Indexes = SIWrapper ? &SIWrapper->getSI() : nullptr;
-  auto *VRMWrapper = getAnalysisIfAvailable<VirtRegMapWrapperLegacy>();
-  VRM = VRMWrapper ? &VRMWrapper->getVRM() : nullptr;
   TRI = ST.getRegisterInfo();
   MRI = &MF.getRegInfo();
 

llvm/test/CodeGen/AMDGPU/si-lower-wwm-copies.mir

@@ -0,0 +1,43 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=liveintervals,virtregmap,si-lower-wwm-copies -o - %s | FileCheck %s
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes="require<live-intervals>,require<virtregmap>,si-lower-wwm-copies" -o - %s | FileCheck %s
+
+# Check for two cases of $scc being live and dead.
+---
+name: lower-wwm-copies
+registers:
+  - { id: 1, class: vgpr_32, flags: [ WWM_REG ]}
+machineFunctionInfo:
+  sgprForEXECCopy: '$sgpr2_sgpr3'
+tracksRegLiveness: true
+body: |
+  ; CHECK-LABEL: name: lower-wwm-copies
+  ; CHECK: bb.0:
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $scc
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   S_CMP_EQ_U32 [[DEF]], 0, implicit-def $scc
+  ; CHECK-NEXT:   $sgpr2_sgpr3 = S_MOV_B64 killed $exec
+  ; CHECK-NEXT:   $exec = S_MOV_B64 -1
+  ; CHECK-NEXT:   $vgpr1 = COPY $vgpr0
+  ; CHECK-NEXT:   $exec = S_MOV_B64 killed $sgpr2_sgpr3
+  ; CHECK-NEXT:   S_CBRANCH_SCC1 %bb.1, implicit $scc
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   liveins: $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   $sgpr2_sgpr3 = S_OR_SAVEEXEC_B64 -1, implicit-def $exec, implicit-def dead $scc, implicit $exec
+  ; CHECK-NEXT:   $vgpr2 = COPY $vgpr1
+  ; CHECK-NEXT:   $exec = S_MOV_B64 killed $sgpr2_sgpr3
+  bb.0:
+    liveins: $vgpr0, $scc
+    %0:sgpr_32 = IMPLICIT_DEF
+    S_CMP_EQ_U32 %0, 0, implicit-def $scc
+    $vgpr1 = WWM_COPY $vgpr0
+    S_CBRANCH_SCC1 %bb.1, implicit killed $scc
+
+  bb.1:
+    liveins: $vgpr1
+    $vgpr2 = WWM_COPY $vgpr1