[AMDGPU] Use GlobalISel MatchTable Combiner Backend

[Pierre-vh]

Summary

Use the new matchtable-based combiner backend for all AMDGPU combiners.
This drop-in from the user's perspective; there are no test changes, the new combiner behaves exactly like the old one.

Depends on D153757

NOTE: This would land iff D153757 (RFC) lands too.

[Pierre-vh]

Rebase

[Pierre-vh]

llvm/lib/Target/AMDGPU/AMDGPUPostLegalizerCombiner.cpp

22–23

Some small detail I'm unsure about: Should I put these in the GET_GICOMBINER_DEPS and also do a typedef of GIMatchTableExecutor to something like ClassNameBase ? That way users don't even need to know about the match table executor.

[Pierre-vh]

Rebase

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUPostLegalizerCombiner.cpp

85

function_ref(I guess this was wrong before)

[Pierre-vh]

llvm/lib/Target/AMDGPU/AMDGPUPostLegalizerCombiner.cpp

85

It doesn't work because of the lifetime of the function unfortunately, there's a segfault

[Pierre-vh]

Rebase + comments

[Pierre-vh]

Add missing setupMF call

Currently it's called on every call to combine() which is annoying. I'd prefer if we only instantiated everything once.
It's a cheap function though so I believe it's fine to start with.

I think when/if this all lands and the dust settles a bit, we need to remove CombinerInfo and replace it with something
closer to InstructionSelector, e.g. have a InstructionCombiner base class and instantiate it once.
It's a bigger refactor though so I think it's better to do it later to make the review stack less confusing.

llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h

public:
   MachineBasicBlock *CurMBB = nullptr;
 
   virtual void setupGeneratedPerFunctionState(MachineFunction &MF) {
     llvm_unreachable("TableGen should have emitted implementation");
   }
 
   /// Setup per-MF executor state.
   virtual void setupMF(MachineFunction &mf, GISelKnownBits *KB,
-                       CodeGenCoverage &covinfo, ProfileSummaryInfo *psi,
-                       BlockFrequencyInfo *bfi) {
-    CoverageInfo = &covinfo;
+                       CodeGenCoverage *covinfo = nullptr,
+                       ProfileSummaryInfo *psi = nullptr,
+                       BlockFrequencyInfo *bfi = nullptr) {
+    CoverageInfo = covinfo;
     KnownBits = KB;
     MF = &mf;
     PSI = psi;
     BFI = bfi;
     CurMBB = nullptr;
     setupGeneratedPerFunctionState(mf);
   }
 

llvm/include/llvm/Target/GlobalISel/Combine.td

def p2i_to_i2p: GICombineRule<
   (apply [{ Helper.applyCombineI2PToP2I(*${root}, ${info}); }])
 >;
 
 // Fold ptr2int(int2ptr(x)) -> x
 def i2p_to_p2i: GICombineRule<
   (defs root:$dst, register_matchinfo:$info),
   (match (G_INTTOPTR $t, $ptr),
          (G_PTRTOINT $dst, $t):$mi,
-    [{ ${info} = ${ptr}.getReg(); }]),
+    [{ ${info} = ${ptr}.getReg(); return true; }]),
   (apply [{ Helper.applyCombineP2IToI2P(*${mi}, ${info}); }])
 >;
 
 // Fold add ptrtoint(x), y -> ptrtoint (ptr_add x), y
 def add_p2i_to_ptradd_matchinfo : GIDefMatchData<"std::pair<Register, bool>">;
 def add_p2i_to_ptradd : GICombineRule<
   (defs root:$root, add_p2i_to_ptradd_matchinfo:$info),
   (match (wip_match_opcode G_ADD):$root,
def not_cmp_fold : GICombineRule<
   (apply [{ Helper.applyNotCmp(*${d}, ${info}); }])
 >;
 
 // Fold (fneg (fneg x)) -> x.
 def fneg_fneg_fold: GICombineRule <
   (defs root:$dst, register_matchinfo:$matchinfo),
   (match (G_FNEG $t, $src),
          (G_FNEG $dst, $t):$mi,
-         [{ ${matchinfo} = ${src}.getReg(); }]),
+         [{ ${matchinfo} = ${src}.getReg(); return true; }]),
   (apply [{ Helper.replaceSingleDefInstWithReg(*${mi}, ${matchinfo}); }])
 >;
 
 // Fold (unmerge(merge x, y, z)) -> z, y, z.
 def unmerge_merge_matchinfo : GIDefMatchData<"SmallVector<Register, 8>">;
 def unmerge_merge : GICombineRule<
   (defs root:$d, unmerge_merge_matchinfo:$info),
   (match (wip_match_opcode G_UNMERGE_VALUES): $d,

llvm/lib/CodeGen/GlobalISel/InstructionSelect.cpp

bool InstructionSelect::runOnMachineFunction(MachineFunction &MF) {
   if (OptLevel != CodeGenOpt::None) {
     PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI();
     if (PSI && PSI->hasProfileSummary())
       BFI = &getAnalysis<LazyBlockFrequencyInfoPass>().getBFI();
   }
 
   CodeGenCoverage CoverageInfo;
   assert(ISel && "Cannot work without InstructionSelector");
-  ISel->setupMF(MF, KB, CoverageInfo, PSI, BFI);
+  ISel->setupMF(MF, KB, &CoverageInfo, PSI, BFI);
 
   // An optimization remark emitter. Used to report failures.
   MachineOptimizationRemarkEmitter MORE(MF, /*MBFI=*/nullptr);
 
   // FIXME: There are many other MF/MFI fields we need to initialize.
 
   MachineRegisterInfo &MRI = MF.getRegInfo();
 #ifndef NDEBUG

llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

public:
   AArch64InstructionSelector(const AArch64TargetMachine &TM,
                              const AArch64Subtarget &STI,
                              const AArch64RegisterBankInfo &RBI);
 
   bool select(MachineInstr &I) override;
   static const char *getName() { return DEBUG_TYPE; }
 
   void setupMF(MachineFunction &MF, GISelKnownBits *KB,
-               CodeGenCoverage &CoverageInfo, ProfileSummaryInfo *PSI,
+               CodeGenCoverage *CoverageInfo, ProfileSummaryInfo *PSI,
                BlockFrequencyInfo *BFI) override {
     InstructionSelector::setupMF(MF, KB, CoverageInfo, PSI, BFI);
     MIB.setMF(MF);
 
     // hasFnAttribute() is expensive to call on every BRCOND selection, so
     // cache it here for each run of the selector.
     ProduceNonFlagSettingCondBr =
         !MF.getFunction().hasFnAttribute(Attribute::SpeculativeLoadHardening);

llvm/lib/Target/AMDGPU/AMDGPUCombine.td

 //=- AMDGPUCombine.td - Define AMDGPU Combine Rules ----------*- tablegen -*-=//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 
 include "llvm/Target/GlobalISel/Combine.td"
 
 // TODO: This really belongs after legalization after scalarization.
 
-def fmin_fmax_legacy_matchdata : GIDefMatchData<"AMDGPUPostLegalizerCombinerHelper::FMinFMaxLegacyInfo">;
+def fmin_fmax_legacy_matchdata : GIDefMatchData<"FMinFMaxLegacyInfo">;
 
 let Predicates = [HasFminFmaxLegacy] in
 def fcmp_select_to_fmin_fmax_legacy : GICombineRule<
   (defs root:$select, fmin_fmax_legacy_matchdata:$matchinfo),
   (match (wip_match_opcode G_SELECT):$select,
-         [{ return PostLegalizerHelper.matchFMinFMaxLegacy(*${select}, ${matchinfo}); }]),
-  (apply [{ PostLegalizerHelper.applySelectFCmpToFMinToFMaxLegacy(*${select}, ${matchinfo}); }])>;
+         [{ return matchFMinFMaxLegacy(*${select}, ${matchinfo}); }]),
+  (apply [{ applySelectFCmpToFMinToFMaxLegacy(*${select}, ${matchinfo}); }])>;
 
 
 def uchar_to_float : GICombineRule<
   (defs root:$itofp),
   (match (wip_match_opcode G_UITOFP, G_SITOFP):$itofp,
-         [{ return PostLegalizerHelper.matchUCharToFloat(*${itofp}); }]),
-  (apply [{ PostLegalizerHelper.applyUCharToFloat(*${itofp}); }])>;
+         [{ return matchUCharToFloat(*${itofp}); }]),
+  (apply [{ applyUCharToFloat(*${itofp}); }])>;
 
 
 def rcp_sqrt_to_rsq : GICombineRule<
   (defs root:$rcp, build_fn_matchinfo:$matchinfo),
   (match (wip_match_opcode G_INTRINSIC, G_FSQRT):$rcp,
-         [{ return PostLegalizerHelper.matchRcpSqrtToRsq(*${rcp}, ${matchinfo}); }]),
+         [{ return matchRcpSqrtToRsq(*${rcp}, ${matchinfo}); }]),
   (apply [{ Helper.applyBuildFn(*${rcp}, ${matchinfo}); }])>;
 
 
-def cvt_f32_ubyteN_matchdata : GIDefMatchData<"AMDGPUPostLegalizerCombinerHelper::CvtF32UByteMatchInfo">;
+def cvt_f32_ubyteN_matchdata : GIDefMatchData<"CvtF32UByteMatchInfo">;
 
 def cvt_f32_ubyteN : GICombineRule<
   (defs root:$cvt_f32_ubyteN, cvt_f32_ubyteN_matchdata:$matchinfo),
   (match (wip_match_opcode G_AMDGPU_CVT_F32_UBYTE0,
                            G_AMDGPU_CVT_F32_UBYTE1,
                            G_AMDGPU_CVT_F32_UBYTE2,
                            G_AMDGPU_CVT_F32_UBYTE3):$cvt_f32_ubyteN,
-         [{ return PostLegalizerHelper.matchCvtF32UByteN(*${cvt_f32_ubyteN}, ${matchinfo}); }]),
-  (apply [{ PostLegalizerHelper.applyCvtF32UByteN(*${cvt_f32_ubyteN}, ${matchinfo}); }])>;
+         [{ return matchCvtF32UByteN(*${cvt_f32_ubyteN}, ${matchinfo}); }]),
+  (apply [{ applyCvtF32UByteN(*${cvt_f32_ubyteN}, ${matchinfo}); }])>;
 
-def clamp_i64_to_i16_matchdata : GIDefMatchData<"AMDGPUPreLegalizerCombinerHelper::ClampI64ToI16MatchInfo">;
+def clamp_i64_to_i16_matchdata : GIDefMatchData<"ClampI64ToI16MatchInfo">;
 
 def clamp_i64_to_i16 : GICombineRule<
   (defs root:$clamp_i64_to_i16, clamp_i64_to_i16_matchdata:$matchinfo),
   (match (wip_match_opcode G_TRUNC):$clamp_i64_to_i16,
-      [{ return PreLegalizerHelper.matchClampI64ToI16(*${clamp_i64_to_i16}, MRI, *MF, ${matchinfo}); }]),
-  (apply [{ PreLegalizerHelper.applyClampI64ToI16(*${clamp_i64_to_i16}, ${matchinfo}); }])>;
+      [{ return matchClampI64ToI16(*${clamp_i64_to_i16}, MRI, MF, ${matchinfo}); }]),
+  (apply [{ applyClampI64ToI16(*${clamp_i64_to_i16}, ${matchinfo}); }])>;
 
-def med3_matchdata : GIDefMatchData<"AMDGPURegBankCombinerHelper::Med3MatchInfo">;
+def med3_matchdata : GIDefMatchData<"Med3MatchInfo">;
 
 def int_minmax_to_med3 : GICombineRule<
   (defs root:$min_or_max, med3_matchdata:$matchinfo),
   (match (wip_match_opcode G_SMAX,
                            G_SMIN,
                            G_UMAX,
                            G_UMIN):$min_or_max,
-         [{ return RegBankHelper.matchIntMinMaxToMed3(*${min_or_max}, ${matchinfo}); }]),
-  (apply [{ RegBankHelper.applyMed3(*${min_or_max}, ${matchinfo}); }])>;
+         [{ return matchIntMinMaxToMed3(*${min_or_max}, ${matchinfo}); }]),
+  (apply [{ applyMed3(*${min_or_max}, ${matchinfo}); }])>;
 
 def fp_minmax_to_med3 : GICombineRule<
   (defs root:$min_or_max, med3_matchdata:$matchinfo),
   (match (wip_match_opcode G_FMAXNUM,
                            G_FMINNUM,
                            G_FMAXNUM_IEEE,
                            G_FMINNUM_IEEE):$min_or_max,
-         [{ return RegBankHelper.matchFPMinMaxToMed3(*${min_or_max}, ${matchinfo}); }]),
-  (apply [{ RegBankHelper.applyMed3(*${min_or_max}, ${matchinfo}); }])>;
+         [{ return matchFPMinMaxToMed3(*${min_or_max}, ${matchinfo}); }]),
+  (apply [{ applyMed3(*${min_or_max}, ${matchinfo}); }])>;
 
 def fp_minmax_to_clamp : GICombineRule<
   (defs root:$min_or_max, register_matchinfo:$matchinfo),
   (match (wip_match_opcode G_FMAXNUM,
                            G_FMINNUM,
                            G_FMAXNUM_IEEE,
                            G_FMINNUM_IEEE):$min_or_max,
-         [{ return RegBankHelper.matchFPMinMaxToClamp(*${min_or_max}, ${matchinfo}); }]),
-  (apply [{ RegBankHelper.applyClamp(*${min_or_max}, ${matchinfo}); }])>;
+         [{ return matchFPMinMaxToClamp(*${min_or_max}, ${matchinfo}); }]),
+  (apply [{ applyClamp(*${min_or_max}, ${matchinfo}); }])>;
 
 def fmed3_intrinsic_to_clamp : GICombineRule<
   (defs root:$fmed3, register_matchinfo:$matchinfo),
   (match (wip_match_opcode G_AMDGPU_FMED3):$fmed3,
-         [{ return RegBankHelper.matchFPMed3ToClamp(*${fmed3}, ${matchinfo}); }]),
-  (apply [{ RegBankHelper.applyClamp(*${fmed3}, ${matchinfo}); }])>;
+         [{ return matchFPMed3ToClamp(*${fmed3}, ${matchinfo}); }]),
+  (apply [{ applyClamp(*${fmed3}, ${matchinfo}); }])>;
 
 def remove_fcanonicalize_matchinfo : GIDefMatchData<"Register">;
 
 def remove_fcanonicalize : GICombineRule<
   (defs root:$fcanonicalize, remove_fcanonicalize_matchinfo:$matchinfo),
   (match (wip_match_opcode G_FCANONICALIZE):$fcanonicalize,
-         [{ return PostLegalizerHelper.matchRemoveFcanonicalize(*${fcanonicalize}, ${matchinfo}); }]),
+         [{ return matchRemoveFcanonicalize(*${fcanonicalize}, ${matchinfo}); }]),
   (apply [{ Helper.replaceSingleDefInstWithReg(*${fcanonicalize}, ${matchinfo}); }])>;
 
 def foldable_fneg_matchdata : GIDefMatchData<"MachineInstr *">;
 
 def foldable_fneg : GICombineRule<
   (defs root:$ffn, foldable_fneg_matchdata:$matchinfo),
   (match (wip_match_opcode G_FNEG):$ffn,
          [{ return Helper.matchFoldableFneg(*${ffn}, ${matchinfo}); }]),
   (apply [{ Helper.applyFoldableFneg(*${ffn}, ${matchinfo}); }])>;
 
 def sign_exension_in_reg_matchdata : GIDefMatchData<"MachineInstr *">;
 
 def sign_extension_in_reg : GICombineRule<
   (defs root:$sign_inreg, sign_exension_in_reg_matchdata:$matchinfo),
   (match (wip_match_opcode G_SEXT_INREG):$sign_inreg,
-         [{ return PostLegalizerHelper.matchCombineSignExtendInReg(*${sign_inreg}, ${matchinfo}); }]),
-  (apply [{ PostLegalizerHelper.applyCombineSignExtendInReg(*${sign_inreg}, ${matchinfo}); }])>;
+         [{ return matchCombineSignExtendInReg(*${sign_inreg}, ${matchinfo}); }]),
+  (apply [{ applyCombineSignExtendInReg(*${sign_inreg}, ${matchinfo}); }])>;
 
 
 let Predicates = [Has16BitInsts, NotHasMed3_16] in {
 // For gfx8, expand f16-fmed3-as-f32 into a min/max f16 sequence. This
 // saves one instruction compared to the promotion.
 //
 // FIXME: Should have ComplexPattern like in/out matchers
 //

 } // End Predicates = [NotHasMed3_16]
 
 // Combines which should only apply on SI/CI
 def gfx6gfx7_combines : GICombineGroup<[fcmp_select_to_fmin_fmax_legacy]>;
 
 // Combines which should only apply on VI
 def gfx8_combines : GICombineGroup<[expand_promoted_fmed3]>;
 
-def AMDGPUPreLegalizerCombinerHelper: GICombinerHelper<
-  "AMDGPUGenPreLegalizerCombinerHelper",
+def AMDGPUPreLegalizerCombiner: GICombinerHelper<
+  "AMDGPUPreLegalizerCombinerImpl",
   [all_combines, clamp_i64_to_i16, foldable_fneg]> {
-  let DisableRuleOption = "amdgpuprelegalizercombiner-disable-rule";
-  let StateClass = "AMDGPUPreLegalizerCombinerHelperState";
-  let AdditionalArguments = [];
 }
 
-def AMDGPUPostLegalizerCombinerHelper: GICombinerHelper<
-  "AMDGPUGenPostLegalizerCombinerHelper",
+def AMDGPUPostLegalizerCombiner: GICombinerHelper<
+  "AMDGPUPostLegalizerCombinerImpl",
   [all_combines, gfx6gfx7_combines, gfx8_combines,
    uchar_to_float, cvt_f32_ubyteN, remove_fcanonicalize, foldable_fneg,
    rcp_sqrt_to_rsq, sign_extension_in_reg]> {
-  let DisableRuleOption = "amdgpupostlegalizercombiner-disable-rule";
-  let StateClass = "AMDGPUPostLegalizerCombinerHelperState";
-  let AdditionalArguments = [];
 }
 
-def AMDGPURegBankCombinerHelper : GICombinerHelper<
-  "AMDGPUGenRegBankCombinerHelper",
+def AMDGPURegBankCombiner : GICombinerHelper<
+  "AMDGPURegBankCombinerImpl",
   [unmerge_merge, unmerge_cst, unmerge_undef,
    zext_trunc_fold, int_minmax_to_med3, ptr_add_immed_chain,
    fp_minmax_to_clamp, fp_minmax_to_med3, fmed3_intrinsic_to_clamp]> {
-  let DisableRuleOption = "amdgpuregbankcombiner-disable-rule";
-  let StateClass = "AMDGPURegBankCombinerHelperState";
-  let AdditionalArguments = [];
 }

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h

public:
   AMDGPUInstructionSelector(const GCNSubtarget &STI,
                             const AMDGPURegisterBankInfo &RBI,
                             const AMDGPUTargetMachine &TM);
 
   bool select(MachineInstr &I) override;
   static const char *getName();
 
   void setupMF(MachineFunction &MF, GISelKnownBits *KB,
-               CodeGenCoverage &CoverageInfo, ProfileSummaryInfo *PSI,
+               CodeGenCoverage *CoverageInfo, ProfileSummaryInfo *PSI,
                BlockFrequencyInfo *BFI) override;
 
 private:
   struct GEPInfo {
     SmallVector<unsigned, 2> SgprParts;
     SmallVector<unsigned, 2> VgprParts;
     int64_t Imm = 0;
   };

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

 #include "AMDGPUGenGlobalISel.inc"
 #undef GET_GLOBALISEL_TEMPORARIES_INIT
 {
 }
 
 const char *AMDGPUInstructionSelector::getName() { return DEBUG_TYPE; }
 
 void AMDGPUInstructionSelector::setupMF(MachineFunction &MF, GISelKnownBits *KB,
-                                        CodeGenCoverage &CoverageInfo,
+                                        CodeGenCoverage *CoverageInfo,
                                         ProfileSummaryInfo *PSI,
                                         BlockFrequencyInfo *BFI) {
   MRI = &MF.getRegInfo();
   Subtarget = &MF.getSubtarget<GCNSubtarget>();
   InstructionSelector::setupMF(MF, KB, CoverageInfo, PSI, BFI);
 }
 
 bool AMDGPUInstructionSelector::isVCC(Register Reg,

llvm/lib/Target/AMDGPU/AMDGPUPostLegalizerCombiner.cpp

-//=== lib/CodeGen/GlobalISel/AMDGPUPostLegalizerCombiner.cpp ---------------===//
+//=== lib/CodeGen/GlobalISel/AMDGPUPostLegalizerCombiner.cpp --------------===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // This pass does combining of machine instructions at the generic MI level,
 // after the legalizer.
 //
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPU.h"
 #include "AMDGPUCombinerHelper.h"
 #include "AMDGPULegalizerInfo.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "llvm/CodeGen/GlobalISel/Combiner.h"
 #include "llvm/CodeGen/GlobalISel/CombinerHelper.h"
 #include "llvm/CodeGen/GlobalISel/CombinerInfo.h"
+#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h"
+#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h"
 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/Target/TargetMachine.h"
 
+#define GET_GICOMBINER_DEPS
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_DEPS
+
 #define DEBUG_TYPE "amdgpu-postlegalizer-combiner"
 
 using namespace llvm;
 using namespace MIPatternMatch;
 
-class AMDGPUPostLegalizerCombinerHelper {
+namespace {
+#define GET_GICOMBINER_TYPES
+#include "AMDGPUGenPostLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_TYPES
+
+class AMDGPUPostLegalizerCombinerImpl : public GIMatchTableExecutor {
 protected:
+  const AMDGPUPostLegalizerCombinerImplRuleConfig &RuleConfig;
+
   MachineIRBuilder &B;
   MachineFunction &MF;
   MachineRegisterInfo &MRI;
-  const GCNSubtarget &Subtarget;
+  const GCNSubtarget &STI;
   const SIInstrInfo &TII;
   AMDGPUCombinerHelper &Helper;
+  GISelChangeObserver &Observer;
 
 public:
-  AMDGPUPostLegalizerCombinerHelper(MachineIRBuilder &B,
-                                    AMDGPUCombinerHelper &Helper)
-      : B(B), MF(B.getMF()), MRI(*B.getMRI()),
-        Subtarget(MF.getSubtarget<GCNSubtarget>()),
-        TII(*Subtarget.getInstrInfo()), Helper(Helper){};
+  AMDGPUPostLegalizerCombinerImpl(
+      const AMDGPUPostLegalizerCombinerImplRuleConfig &RuleConfig,
+      MachineIRBuilder &B, AMDGPUCombinerHelper &Helper,
+      GISelChangeObserver &Observer);
+
+  static const char *getName() { return "AMDGPUPostLegalizerCombinerImpl"; }
+
+  bool tryCombineAll(MachineInstr &I) const;
 
   struct FMinFMaxLegacyInfo {
     Register LHS;
     Register RHS;
     Register True;
     Register False;
     CmpInst::Predicate Pred;
   };
 
   // TODO: Make sure fmin_legacy/fmax_legacy don't canonicalize
-  bool matchFMinFMaxLegacy(MachineInstr &MI, FMinFMaxLegacyInfo &Info);
+  bool matchFMinFMaxLegacy(MachineInstr &MI, FMinFMaxLegacyInfo &Info) const;
   void applySelectFCmpToFMinToFMaxLegacy(MachineInstr &MI,
-                                         const FMinFMaxLegacyInfo &Info);
+                                         const FMinFMaxLegacyInfo &Info) const;
 
-  bool matchUCharToFloat(MachineInstr &MI);
-  void applyUCharToFloat(MachineInstr &MI);
+  bool matchUCharToFloat(MachineInstr &MI) const;
+  void applyUCharToFloat(MachineInstr &MI) const;
 
-  bool matchRcpSqrtToRsq(MachineInstr &MI,
-                         std::function<void(MachineIRBuilder &)> &MatchInfo);
+  bool
+  matchRcpSqrtToRsq(MachineInstr &MI,
+                    std::function<void(MachineIRBuilder &)> &MatchInfo) const;
 
   // FIXME: Should be able to have 2 separate matchdatas rather than custom
   // struct boilerplate.
   struct CvtF32UByteMatchInfo {
     Register CvtVal;
     unsigned ShiftOffset;
   };
 
-  bool matchCvtF32UByteN(MachineInstr &MI, CvtF32UByteMatchInfo &MatchInfo);
+  bool matchCvtF32UByteN(MachineInstr &MI,
+                         CvtF32UByteMatchInfo &MatchInfo) const;
   void applyCvtF32UByteN(MachineInstr &MI,
-                         const CvtF32UByteMatchInfo &MatchInfo);
+                         const CvtF32UByteMatchInfo &MatchInfo) const;
 
-  bool matchRemoveFcanonicalize(MachineInstr &MI, Register &Reg);
+  bool matchRemoveFcanonicalize(MachineInstr &MI, Register &Reg) const;
 
   // Combine unsigned buffer load and signed extension instructions to generate
   // signed buffer laod instructions.
-  bool matchCombineSignExtendInReg(MachineInstr &MI, MachineInstr *&MatchInfo);
-  void applyCombineSignExtendInReg(MachineInstr &MI, MachineInstr *&MatchInfo);
+  bool matchCombineSignExtendInReg(MachineInstr &MI,
+                                   MachineInstr *&MatchInfo) const;
+  void applyCombineSignExtendInReg(MachineInstr &MI,
+                                   MachineInstr *&MatchInfo) const;
+
+private:
+#define GET_GICOMBINER_CLASS_MEMBERS
+#define AMDGPUSubtarget GCNSubtarget
+#include "AMDGPUGenPostLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_CLASS_MEMBERS
+#undef AMDGPUSubtarget
 };
 
-bool AMDGPUPostLegalizerCombinerHelper::matchFMinFMaxLegacy(
-    MachineInstr &MI, FMinFMaxLegacyInfo &Info) {
+#define GET_GICOMBINER_IMPL
+#define AMDGPUSubtarget GCNSubtarget
+#include "AMDGPUGenPostLegalizeGICombiner.inc"
+#undef AMDGPUSubtarget
+#undef GET_GICOMBINER_IMPL
+
+AMDGPUPostLegalizerCombinerImpl::AMDGPUPostLegalizerCombinerImpl(
+    const AMDGPUPostLegalizerCombinerImplRuleConfig &RuleConfig,
+    MachineIRBuilder &B, AMDGPUCombinerHelper &Helper,
+    GISelChangeObserver &Observer)
+    : RuleConfig(RuleConfig), B(B), MF(B.getMF()), MRI(*B.getMRI()),
+      STI(MF.getSubtarget<GCNSubtarget>()), TII(*STI.getInstrInfo()),
+      Helper(Helper), Observer(Observer),
+#define GET_GICOMBINER_CONSTRUCTOR_INITS
+#include "AMDGPUGenPostLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_CONSTRUCTOR_INITS
+{
+}
+
+bool AMDGPUPostLegalizerCombinerImpl::matchFMinFMaxLegacy(
+    MachineInstr &MI, FMinFMaxLegacyInfo &Info) const {
   // FIXME: Type predicate on pattern
   if (MRI.getType(MI.getOperand(0).getReg()) != LLT::scalar(32))
     return false;
 
   Register Cond = MI.getOperand(1).getReg();
   if (!MRI.hasOneNonDBGUse(Cond) ||
       !mi_match(Cond, MRI,
                 m_GFCmp(m_Pred(Info.Pred), m_Reg(Info.LHS), m_Reg(Info.RHS))))
bool AMDGPUPostLegalizerCombinerImpl::matchFMinFMaxLegacy(
   case CmpInst::FCMP_UNE:
   case CmpInst::FCMP_TRUE:
     return false;
   default:
     return true;
   }
 }
 
-void AMDGPUPostLegalizerCombinerHelper::applySelectFCmpToFMinToFMaxLegacy(
-    MachineInstr &MI, const FMinFMaxLegacyInfo &Info) {
+void AMDGPUPostLegalizerCombinerImpl::applySelectFCmpToFMinToFMaxLegacy(
+    MachineInstr &MI, const FMinFMaxLegacyInfo &Info) const {
   B.setInstrAndDebugLoc(MI);
   auto buildNewInst = [&MI, this](unsigned Opc, Register X, Register Y) {
     B.buildInstr(Opc, {MI.getOperand(0)}, {X, Y}, MI.getFlags());
   };
 
   switch (Info.Pred) {
   case CmpInst::FCMP_ULT:
   case CmpInst::FCMP_ULE:
void AMDGPUPostLegalizerCombinerImpl::applySelectFCmpToFMinToFMaxLegacy(
   }
   default:
     llvm_unreachable("predicate should not have matched");
   }
 
   MI.eraseFromParent();
 }
 
-bool AMDGPUPostLegalizerCombinerHelper::matchUCharToFloat(MachineInstr &MI) {
+bool AMDGPUPostLegalizerCombinerImpl::matchUCharToFloat(
+    MachineInstr &MI) const {
   Register DstReg = MI.getOperand(0).getReg();
 
   // TODO: We could try to match extracting the higher bytes, which would be
   // easier if i8 vectors weren't promoted to i32 vectors, particularly after
   // types are legalized. v4i8 -> v4f32 is probably the only case to worry
   // about in practice.
   LLT Ty = MRI.getType(DstReg);
   if (Ty == LLT::scalar(32) || Ty == LLT::scalar(16)) {
     Register SrcReg = MI.getOperand(1).getReg();
     unsigned SrcSize = MRI.getType(SrcReg).getSizeInBits();
     assert(SrcSize == 16 || SrcSize == 32 || SrcSize == 64);
     const APInt Mask = APInt::getHighBitsSet(SrcSize, SrcSize - 8);
     return Helper.getKnownBits()->maskedValueIsZero(SrcReg, Mask);
   }
 
   return false;
 }
 
-void AMDGPUPostLegalizerCombinerHelper::applyUCharToFloat(MachineInstr &MI) {
+void AMDGPUPostLegalizerCombinerImpl::applyUCharToFloat(
+    MachineInstr &MI) const {
   B.setInstrAndDebugLoc(MI);
 
   const LLT S32 = LLT::scalar(32);
 
   Register DstReg = MI.getOperand(0).getReg();
   Register SrcReg = MI.getOperand(1).getReg();
   LLT Ty = MRI.getType(DstReg);
   LLT SrcTy = MRI.getType(SrcReg);
   if (SrcTy != S32)
     SrcReg = B.buildAnyExtOrTrunc(S32, SrcReg).getReg(0);
 
   if (Ty == S32) {
-    B.buildInstr(AMDGPU::G_AMDGPU_CVT_F32_UBYTE0, {DstReg},
-                   {SrcReg}, MI.getFlags());
+    B.buildInstr(AMDGPU::G_AMDGPU_CVT_F32_UBYTE0, {DstReg}, {SrcReg},
+                 MI.getFlags());
   } else {
-    auto Cvt0 = B.buildInstr(AMDGPU::G_AMDGPU_CVT_F32_UBYTE0, {S32},
-                             {SrcReg}, MI.getFlags());
+    auto Cvt0 = B.buildInstr(AMDGPU::G_AMDGPU_CVT_F32_UBYTE0, {S32}, {SrcReg},
+                             MI.getFlags());
     B.buildFPTrunc(DstReg, Cvt0, MI.getFlags());
   }
 
   MI.eraseFromParent();
 }
 
-bool AMDGPUPostLegalizerCombinerHelper::matchRcpSqrtToRsq(
-    MachineInstr &MI, std::function<void(MachineIRBuilder &)> &MatchInfo) {
+bool AMDGPUPostLegalizerCombinerImpl::matchRcpSqrtToRsq(
+    MachineInstr &MI,
+    std::function<void(MachineIRBuilder &)> &MatchInfo) const {
 
   auto getRcpSrc = [=](const MachineInstr &MI) {
     MachineInstr *ResMI = nullptr;
     if (MI.getOpcode() == TargetOpcode::G_INTRINSIC &&
         MI.getIntrinsicID() == Intrinsic::amdgcn_rcp)
       ResMI = MRI.getVRegDef(MI.getOperand(2).getReg());
 
     return ResMI;
MatchInfo = [RcpSrcMI, &MI](MachineIRBuilder &B) {
           .setMIFlags(MI.getFlags());
     };
     return true;
   }
 
   return false;
 }
 
-bool AMDGPUPostLegalizerCombinerHelper::matchCvtF32UByteN(
-    MachineInstr &MI, CvtF32UByteMatchInfo &MatchInfo) {
+bool AMDGPUPostLegalizerCombinerImpl::matchCvtF32UByteN(
+    MachineInstr &MI, CvtF32UByteMatchInfo &MatchInfo) const {
   Register SrcReg = MI.getOperand(1).getReg();
 
   // Look through G_ZEXT.
   bool IsShr = mi_match(SrcReg, MRI, m_GZExt(m_Reg(SrcReg)));
 
   Register Src0;
   int64_t ShiftAmt;
   IsShr = mi_match(SrcReg, MRI, m_GLShr(m_Reg(Src0), m_ICst(ShiftAmt)));
if (IsShr || mi_match(SrcReg, MRI, m_GShl(m_Reg(Src0), m_ICst(ShiftAmt)))) {
     MatchInfo.ShiftOffset = ShiftOffset;
     return ShiftOffset < 32 && ShiftOffset >= 8 && (ShiftOffset % 8) == 0;
   }
 
   // TODO: Simplify demanded bits.
   return false;
 }
 
-void AMDGPUPostLegalizerCombinerHelper::applyCvtF32UByteN(
-    MachineInstr &MI, const CvtF32UByteMatchInfo &MatchInfo) {
+void AMDGPUPostLegalizerCombinerImpl::applyCvtF32UByteN(
+    MachineInstr &MI, const CvtF32UByteMatchInfo &MatchInfo) const {
   B.setInstrAndDebugLoc(MI);
   unsigned NewOpc = AMDGPU::G_AMDGPU_CVT_F32_UBYTE0 + MatchInfo.ShiftOffset / 8;
 
   const LLT S32 = LLT::scalar(32);
   Register CvtSrc = MatchInfo.CvtVal;
   LLT SrcTy = MRI.getType(MatchInfo.CvtVal);
   if (SrcTy != S32) {
     assert(SrcTy.isScalar() && SrcTy.getSizeInBits() >= 8);
     CvtSrc = B.buildAnyExt(S32, CvtSrc).getReg(0);
   }
 
   assert(MI.getOpcode() != NewOpc);
   B.buildInstr(NewOpc, {MI.getOperand(0)}, {CvtSrc}, MI.getFlags());
   MI.eraseFromParent();
 }
 
-bool AMDGPUPostLegalizerCombinerHelper::matchRemoveFcanonicalize(
-    MachineInstr &MI, Register &Reg) {
+bool AMDGPUPostLegalizerCombinerImpl::matchRemoveFcanonicalize(
+    MachineInstr &MI, Register &Reg) const {
   const SITargetLowering *TLI = static_cast<const SITargetLowering *>(
       MF.getSubtarget().getTargetLowering());
   Reg = MI.getOperand(1).getReg();
   return TLI->isCanonicalized(Reg, MF);
 }
 
 // The buffer_load_{i8, i16} intrinsics are intially lowered as buffer_load_{u8,
 // u16} instructions. Here, the buffer_load_{u8, u16} instructions are combined
 // with sign extension instrucions in order to generate buffer_load_{i8, i16}
 // instructions.
 
 // Identify buffer_load_{u8, u16}.
-bool AMDGPUPostLegalizerCombinerHelper::matchCombineSignExtendInReg(
-    MachineInstr &MI, MachineInstr *&SubwordBufferLoad) {
+bool AMDGPUPostLegalizerCombinerImpl::matchCombineSignExtendInReg(
+    MachineInstr &MI, MachineInstr *&SubwordBufferLoad) const {
   Register Op0Reg = MI.getOperand(1).getReg();
   SubwordBufferLoad = MRI.getVRegDef(Op0Reg);
 
   if (!MRI.hasOneNonDBGUse(Op0Reg))
     return false;
 
   // Check if the first operand of the sign extension is a subword buffer load
   // instruction.
   return SubwordBufferLoad->getOpcode() == AMDGPU::G_AMDGPU_BUFFER_LOAD_UBYTE ||
          SubwordBufferLoad->getOpcode() == AMDGPU::G_AMDGPU_BUFFER_LOAD_USHORT;
 }
 
 // Combine buffer_load_{u8, u16} and the sign extension instruction to generate
 // buffer_load_{i8, i16}.
-void AMDGPUPostLegalizerCombinerHelper::applyCombineSignExtendInReg(
-    MachineInstr &MI, MachineInstr *&SubwordBufferLoad) {
+void AMDGPUPostLegalizerCombinerImpl::applyCombineSignExtendInReg(
+    MachineInstr &MI, MachineInstr *&SubwordBufferLoad) const {
   // Modify the opcode and the destination of buffer_load_{u8, u16}:
   // Replace the opcode.
   unsigned Opc =
       SubwordBufferLoad->getOpcode() == AMDGPU::G_AMDGPU_BUFFER_LOAD_UBYTE
           ? AMDGPU::G_AMDGPU_BUFFER_LOAD_SBYTE
           : AMDGPU::G_AMDGPU_BUFFER_LOAD_SSHORT;
   SubwordBufferLoad->setDesc(TII.get(Opc));
   // Update the destination register of SubwordBufferLoad with the destination
   // register of the sign extension.
   Register SignExtendInsnDst = MI.getOperand(0).getReg();
   SubwordBufferLoad->getOperand(0).setReg(SignExtendInsnDst);
   // Remove the sign extension.
   MI.eraseFromParent();
 }
 
-class AMDGPUPostLegalizerCombinerHelperState {
-protected:
-  AMDGPUCombinerHelper &Helper;
-  AMDGPUPostLegalizerCombinerHelper &PostLegalizerHelper;
-
-  // Note: pointer is necessary because Target Predicates use
-  //   "Subtarget->"
-  const GCNSubtarget *Subtarget;
-
-public:
-  AMDGPUPostLegalizerCombinerHelperState(
-      AMDGPUCombinerHelper &Helper,
-      AMDGPUPostLegalizerCombinerHelper &PostLegalizerHelper,
-      const GCNSubtarget &Subtarget)
-      : Helper(Helper), PostLegalizerHelper(PostLegalizerHelper),
-        Subtarget(&Subtarget) {}
-};
-
-#define AMDGPUPOSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_DEPS
-#include "AMDGPUGenPostLegalizeGICombiner.inc"
-#undef AMDGPUPOSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_DEPS
-
-namespace {
-#define AMDGPUPOSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_H
-#include "AMDGPUGenPostLegalizeGICombiner.inc"
-#undef AMDGPUPOSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_H
-
 class AMDGPUPostLegalizerCombinerInfo final : public CombinerInfo {
   GISelKnownBits *KB;
   MachineDominatorTree *MDT;
-  const GCNSubtarget &Subtarget;
+  AMDGPUPostLegalizerCombinerImplRuleConfig RuleConfig;
 
 public:
-  AMDGPUGenPostLegalizerCombinerHelperRuleConfig GeneratedRuleCfg;
-
-  AMDGPUPostLegalizerCombinerInfo(const GCNSubtarget &Subtarget, bool EnableOpt,
-                                  bool OptSize, bool MinSize,
+  AMDGPUPostLegalizerCombinerInfo(bool EnableOpt, bool OptSize, bool MinSize,
                                   const AMDGPULegalizerInfo *LI,
                                   GISelKnownBits *KB, MachineDominatorTree *MDT)
       : CombinerInfo(/*AllowIllegalOps*/ false, /*ShouldLegalizeIllegal*/ true,
                      /*LegalizerInfo*/ LI, EnableOpt, OptSize, MinSize),
-        KB(KB), MDT(MDT), Subtarget(Subtarget) {
-    if (!GeneratedRuleCfg.parseCommandLineOption())
+        KB(KB), MDT(MDT) {
+    if (!RuleConfig.parseCommandLineOption())
       report_fatal_error("Invalid rule identifier");
   }
 
   bool combine(GISelChangeObserver &Observer, MachineInstr &MI,
                MachineIRBuilder &B) const override;
 };
 
 bool AMDGPUPostLegalizerCombinerInfo::combine(GISelChangeObserver &Observer,
                                               MachineInstr &MI,
                                               MachineIRBuilder &B) const {
   AMDGPUCombinerHelper Helper(Observer, B, /*IsPreLegalize*/ false, KB, MDT,
                               LInfo);
-  AMDGPUPostLegalizerCombinerHelper PostLegalizerHelper(B, Helper);
-  AMDGPUGenPostLegalizerCombinerHelper Generated(
-      GeneratedRuleCfg, Helper, PostLegalizerHelper, Subtarget);
+  // TODO: Do not re-create the Impl on every inst, it should be per function.
+  AMDGPUPostLegalizerCombinerImpl Impl(RuleConfig, B, Helper, Observer);
+  Impl.setupMF(*MI.getMF(), KB);
 
-  if (Generated.tryCombineAll(Observer, MI, B))
+  if (Impl.tryCombineAll(MI))
     return true;
 
   switch (MI.getOpcode()) {
   case TargetOpcode::G_SHL:
   case TargetOpcode::G_LSHR:
   case TargetOpcode::G_ASHR:
     // On some subtargets, 64-bit shift is a quarter rate instruction. In the
     // common case, splitting this into a move and a 32-bit shift is faster and
     // the same code size.
     return Helper.tryCombineShiftToUnmerge(MI, 32);
   }
 
   return false;
 }
 
-#define AMDGPUPOSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_CPP
-#include "AMDGPUGenPostLegalizeGICombiner.inc"
-#undef AMDGPUPOSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_CPP
-
 // Pass boilerplate
 // ================
 
 class AMDGPUPostLegalizerCombiner : public MachineFunctionPass {
 public:
   static char ID;
 
   AMDGPUPostLegalizerCombiner(bool IsOptNone = false);
void AMDGPUPostLegalizerCombiner::getAnalysisUsage(AnalysisUsage &AU) const {
   if (!IsOptNone) {
     AU.addRequired<MachineDominatorTree>();
     AU.addPreserved<MachineDominatorTree>();
   }
   MachineFunctionPass::getAnalysisUsage(AU);
 }
 
 AMDGPUPostLegalizerCombiner::AMDGPUPostLegalizerCombiner(bool IsOptNone)
     : MachineFunctionPass(ID), IsOptNone(IsOptNone) {
   initializeAMDGPUPostLegalizerCombinerPass(*PassRegistry::getPassRegistry());
 }
 
 bool AMDGPUPostLegalizerCombiner::runOnMachineFunction(MachineFunction &MF) {
   if (MF.getProperties().hasProperty(
           MachineFunctionProperties::Property::FailedISel))
     return false;
   auto *TPC = &getAnalysis<TargetPassConfig>();
   const Function &F = MF.getFunction();
   bool EnableOpt =
       MF.getTarget().getOptLevel() != CodeGenOpt::None && !skipFunction(F);
 
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
-  const AMDGPULegalizerInfo *LI
-    = static_cast<const AMDGPULegalizerInfo *>(ST.getLegalizerInfo());
+  const AMDGPULegalizerInfo *LI =
+      static_cast<const AMDGPULegalizerInfo *>(ST.getLegalizerInfo());
 
   GISelKnownBits *KB = &getAnalysis<GISelKnownBitsAnalysis>().get(MF);
   MachineDominatorTree *MDT =
       IsOptNone ? nullptr : &getAnalysis<MachineDominatorTree>();
-  AMDGPUPostLegalizerCombinerInfo PCInfo(ST, EnableOpt, F.hasOptSize(),
+  AMDGPUPostLegalizerCombinerInfo PCInfo(EnableOpt, F.hasOptSize(),
                                          F.hasMinSize(), LI, KB, MDT);
   Combiner C(PCInfo, TPC);
   return C.combineMachineInstrs(MF, /*CSEInfo*/ nullptr);
 }
 
 char AMDGPUPostLegalizerCombiner::ID = 0;
 INITIALIZE_PASS_BEGIN(AMDGPUPostLegalizerCombiner, DEBUG_TYPE,
-                      "Combine AMDGPU machine instrs after legalization",
-                      false, false)
+                      "Combine AMDGPU machine instrs after legalization", false,
+                      false)
 INITIALIZE_PASS_DEPENDENCY(TargetPassConfig)
 INITIALIZE_PASS_DEPENDENCY(GISelKnownBitsAnalysis)
 INITIALIZE_PASS_END(AMDGPUPostLegalizerCombiner, DEBUG_TYPE,
                     "Combine AMDGPU machine instrs after legalization", false,
                     false)
 
 namespace llvm {
 FunctionPass *createAMDGPUPostLegalizeCombiner(bool IsOptNone) {
   return new AMDGPUPostLegalizerCombiner(IsOptNone);
 }
 } // end namespace llvm

llvm/lib/Target/AMDGPU/AMDGPUPreLegalizerCombiner.cpp

 #include "AMDGPUCombinerHelper.h"
 #include "AMDGPULegalizerInfo.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "llvm/CodeGen/GlobalISel/CSEInfo.h"
 #include "llvm/CodeGen/GlobalISel/Combiner.h"
 #include "llvm/CodeGen/GlobalISel/CombinerHelper.h"
 #include "llvm/CodeGen/GlobalISel/CombinerInfo.h"
+#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h"
+#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h"
 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/Target/TargetMachine.h"
 
+#define GET_GICOMBINER_DEPS
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_DEPS
+
 #define DEBUG_TYPE "amdgpu-prelegalizer-combiner"
 
 using namespace llvm;
 using namespace MIPatternMatch;
+namespace {
+
+#define GET_GICOMBINER_TYPES
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_TYPES
 
-class AMDGPUPreLegalizerCombinerHelper {
+class AMDGPUPreLegalizerCombinerImpl : public GIMatchTableExecutor {
 protected:
+  const AMDGPUPreLegalizerCombinerImplRuleConfig &RuleConfig;
+  const GCNSubtarget &STI;
+
+  GISelChangeObserver &Observer;
   MachineIRBuilder &B;
   MachineFunction &MF;
   MachineRegisterInfo &MRI;
   AMDGPUCombinerHelper &Helper;
 
 public:
-  AMDGPUPreLegalizerCombinerHelper(MachineIRBuilder &B,
-                                   AMDGPUCombinerHelper &Helper)
-      : B(B), MF(B.getMF()), MRI(*B.getMRI()), Helper(Helper){};
+  AMDGPUPreLegalizerCombinerImpl(
+      const AMDGPUPreLegalizerCombinerImplRuleConfig &RuleConfig,
+      const GCNSubtarget &STI, GISelChangeObserver &Observer,
+      MachineIRBuilder &B, AMDGPUCombinerHelper &Helper);
+
+  static const char *getName() { return "AMDGPUPreLegalizerCombinerImpl"; }
+
+  bool tryCombineAll(MachineInstr &I) const;
 
   struct ClampI64ToI16MatchInfo {
     int64_t Cmp1 = 0;
     int64_t Cmp2 = 0;
     Register Origin;
   };
 
-  bool matchClampI64ToI16(MachineInstr &MI, MachineRegisterInfo &MRI,
-                          MachineFunction &MF,
-                          ClampI64ToI16MatchInfo &MatchInfo);
+  bool matchClampI64ToI16(MachineInstr &MI, const MachineRegisterInfo &MRI,
+                          const MachineFunction &MF,
+                          ClampI64ToI16MatchInfo &MatchInfo) const;
 
   void applyClampI64ToI16(MachineInstr &MI,
-                          const ClampI64ToI16MatchInfo &MatchInfo);
+                          const ClampI64ToI16MatchInfo &MatchInfo) const;
+
+private:
+#define GET_GICOMBINER_CLASS_MEMBERS
+#define AMDGPUSubtarget GCNSubtarget
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_CLASS_MEMBERS
+#undef AMDGPUSubtarget
 };
 
-bool AMDGPUPreLegalizerCombinerHelper::matchClampI64ToI16(
-    MachineInstr &MI, MachineRegisterInfo &MRI, MachineFunction &MF,
-    ClampI64ToI16MatchInfo &MatchInfo) {
+#define GET_GICOMBINER_IMPL
+#define AMDGPUSubtarget GCNSubtarget
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef AMDGPUSubtarget
+#undef GET_GICOMBINER_IMPL
+
+AMDGPUPreLegalizerCombinerImpl::AMDGPUPreLegalizerCombinerImpl(
+    const AMDGPUPreLegalizerCombinerImplRuleConfig &RuleConfig,
+    const GCNSubtarget &STI, GISelChangeObserver &Observer, MachineIRBuilder &B,
+    AMDGPUCombinerHelper &Helper)
+    : RuleConfig(RuleConfig), STI(STI), Observer(Observer), B(B), MF(B.getMF()),
+      MRI(*B.getMRI()), Helper(Helper),
+#define GET_GICOMBINER_CONSTRUCTOR_INITS
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_CONSTRUCTOR_INITS
+{
+}
+
+bool AMDGPUPreLegalizerCombinerImpl::matchClampI64ToI16(
+    MachineInstr &MI, const MachineRegisterInfo &MRI, const MachineFunction &MF,
+    ClampI64ToI16MatchInfo &MatchInfo) const {
   assert(MI.getOpcode() == TargetOpcode::G_TRUNC && "Invalid instruction!");
 
   // Try to find a pattern where an i64 value should get clamped to short.
   const LLT SrcType = MRI.getType(MI.getOperand(1).getReg());
   if (SrcType != LLT::scalar(64))
     return false;
 
   const LLT DstType = MRI.getType(MI.getOperand(0).getReg());

 
 // We want to find a combination of instructions that
 // gets generated when an i64 gets clamped to i16.
 // The corresponding pattern is:
 // G_MAX / G_MAX for i16 <= G_TRUNC i64.
 // This can be efficiently written as following:
 // v_cvt_pk_i16_i32 v0, v0, v1
 // v_med3_i32 v0, Clamp_Min, v0, Clamp_Max
-void AMDGPUPreLegalizerCombinerHelper::applyClampI64ToI16(
-    MachineInstr &MI, const ClampI64ToI16MatchInfo &MatchInfo) {
+void AMDGPUPreLegalizerCombinerImpl::applyClampI64ToI16(
+    MachineInstr &MI, const ClampI64ToI16MatchInfo &MatchInfo) const {
 
   Register Src = MatchInfo.Origin;
   assert(MI.getParent()->getParent()->getRegInfo().getType(Src) ==
          LLT::scalar(64));
   const LLT S32 = LLT::scalar(32);
 
   B.setInstrAndDebugLoc(MI);
 
auto Med3 = B.buildInstr(
       {MinBoundaryDst.getReg(0), Bitcast.getReg(0), MaxBoundaryDst.getReg(0)},
       MI.getFlags());
 
   B.buildTrunc(MI.getOperand(0).getReg(), Med3);
 
   MI.eraseFromParent();
 }
 
-class AMDGPUPreLegalizerCombinerHelperState {
-protected:
-  AMDGPUCombinerHelper &Helper;
-  AMDGPUPreLegalizerCombinerHelper &PreLegalizerHelper;
-
-public:
-  AMDGPUPreLegalizerCombinerHelperState(
-      AMDGPUCombinerHelper &Helper,
-      AMDGPUPreLegalizerCombinerHelper &PreLegalizerHelper)
-      : Helper(Helper), PreLegalizerHelper(PreLegalizerHelper) {}
-};
-
-#define AMDGPUPRELEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_DEPS
-#include "AMDGPUGenPreLegalizeGICombiner.inc"
-#undef AMDGPUPRELEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_DEPS
-
-namespace {
-#define AMDGPUPRELEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_H
-#include "AMDGPUGenPreLegalizeGICombiner.inc"
-#undef AMDGPUPRELEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_H
-
 class AMDGPUPreLegalizerCombinerInfo final : public CombinerInfo {
   GISelKnownBits *KB;
   MachineDominatorTree *MDT;
+  AMDGPUPreLegalizerCombinerImplRuleConfig RuleConfig;
 
 public:
-  AMDGPUGenPreLegalizerCombinerHelperRuleConfig GeneratedRuleCfg;
-
   AMDGPUPreLegalizerCombinerInfo(bool EnableOpt, bool OptSize, bool MinSize,
-                                  GISelKnownBits *KB, MachineDominatorTree *MDT)
       : CombinerInfo(/*AllowIllegalOps*/ true, /*ShouldLegalizeIllegal*/ false,
                      /*LegalizerInfo*/ nullptr, EnableOpt, OptSize, MinSize),
         KB(KB), MDT(MDT) {
-    if (!GeneratedRuleCfg.parseCommandLineOption())
+    if (!RuleConfig.parseCommandLineOption())
       report_fatal_error("Invalid rule identifier");
   }
 
   bool combine(GISelChangeObserver &Observer, MachineInstr &MI,
                MachineIRBuilder &B) const override;
 };
 
 bool AMDGPUPreLegalizerCombinerInfo::combine(GISelChangeObserver &Observer,
-                                              MachineInstr &MI,
-                                              MachineIRBuilder &B) const {
   const auto *LI = MI.getMF()->getSubtarget().getLegalizerInfo();
   AMDGPUCombinerHelper Helper(Observer, B, /*IsPreLegalize*/ true, KB, MDT, LI);
-  AMDGPUPreLegalizerCombinerHelper PreLegalizerHelper(B, Helper);
-  AMDGPUGenPreLegalizerCombinerHelper Generated(GeneratedRuleCfg, Helper,
-                                                PreLegalizerHelper);
 
-  if (Generated.tryCombineAll(Observer, MI, B))
+  const GCNSubtarget &STI = MI.getMF()->getSubtarget<GCNSubtarget>();
+  // TODO: Do not re-create the Impl on every inst, it should be per function.
+  AMDGPUPreLegalizerCombinerImpl Impl(RuleConfig, STI, Observer, B, Helper);
+  Impl.setupMF(*MI.getMF(), KB);
+
+  if (Impl.tryCombineAll(MI))
     return true;
 
   switch (MI.getOpcode()) {
   case TargetOpcode::G_CONCAT_VECTORS:
     return Helper.tryCombineConcatVectors(MI);
   case TargetOpcode::G_SHUFFLE_VECTOR:
     return Helper.tryCombineShuffleVector(MI);
   }
 
   return false;
 }
 
-#define AMDGPUPRELEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_CPP
-#include "AMDGPUGenPreLegalizeGICombiner.inc"
-#undef AMDGPUPRELEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_CPP
-
 // Pass boilerplate
 // ================
 
 class AMDGPUPreLegalizerCombiner : public MachineFunctionPass {
 public:
   static char ID;
 
   AMDGPUPreLegalizerCombiner(bool IsOptNone = false);
 
   StringRef getPassName() const override {
     return "AMDGPUPreLegalizerCombiner";
   }
 
   bool runOnMachineFunction(MachineFunction &MF) override;
 
   void getAnalysisUsage(AnalysisUsage &AU) const override;
+
 private:
   bool IsOptNone;
 };
 } // end anonymous namespace
 
 void AMDGPUPreLegalizerCombiner::getAnalysisUsage(AnalysisUsage &AU) const {
   AU.addRequired<TargetPassConfig>();
   AU.setPreservesCFG();
   getSelectionDAGFallbackAnalysisUsage(AU);
   AU.addRequired<GISelKnownBitsAnalysis>();
   AU.addPreserved<GISelKnownBitsAnalysis>();
   if (!IsOptNone) {
     AU.addRequired<MachineDominatorTree>();
     AU.addPreserved<MachineDominatorTree>();
   }
 
   AU.addRequired<GISelCSEAnalysisWrapperPass>();
   AU.addPreserved<GISelCSEAnalysisWrapperPass>();
   MachineFunctionPass::getAnalysisUsage(AU);
 }
 
 AMDGPUPreLegalizerCombiner::AMDGPUPreLegalizerCombiner(bool IsOptNone)
     : MachineFunctionPass(ID), IsOptNone(IsOptNone) {
   initializeAMDGPUPreLegalizerCombinerPass(*PassRegistry::getPassRegistry());
 }
 
 bool AMDGPUPreLegalizerCombiner::runOnMachineFunction(MachineFunction &MF) {
   if (MF.getProperties().hasProperty(
           MachineFunctionProperties::Property::FailedISel))
     return false;
   auto *TPC = &getAnalysis<TargetPassConfig>();

llvm/lib/Target/AMDGPU/AMDGPURegBankCombiner.cpp

 #include "AMDGPULegalizerInfo.h"
 #include "AMDGPURegisterBankInfo.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/GlobalISel/Combiner.h"
 #include "llvm/CodeGen/GlobalISel/CombinerHelper.h"
 #include "llvm/CodeGen/GlobalISel/CombinerInfo.h"
+#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h"
+#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h"
 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/Target/TargetMachine.h"
+
+#define GET_GICOMBINER_DEPS
+#include "AMDGPUGenPreLegalizeGICombiner.inc"
+#undef GET_GICOMBINER_DEPS
+
 #define DEBUG_TYPE "amdgpu-regbank-combiner"
 
 using namespace llvm;
 using namespace MIPatternMatch;
 
-class AMDGPURegBankCombinerHelper {
+namespace {
+#define GET_GICOMBINER_TYPES
+#include "AMDGPUGenRegBankGICombiner.inc"
+#undef GET_GICOMBINER_TYPES
+
+class AMDGPURegBankCombinerImpl : public GIMatchTableExecutor {
 protected:
+  const AMDGPURegBankCombinerImplRuleConfig &RuleConfig;
+
   MachineIRBuilder &B;
   MachineFunction &MF;
   MachineRegisterInfo &MRI;
-  const GCNSubtarget &Subtarget;
+  const GCNSubtarget &STI;
   const RegisterBankInfo &RBI;
   const TargetRegisterInfo &TRI;
   const SIInstrInfo &TII;
   CombinerHelper &Helper;
+  GISelChangeObserver &Observer;
 
 public:
-  AMDGPURegBankCombinerHelper(MachineIRBuilder &B, CombinerHelper &Helper)
-      : B(B), MF(B.getMF()), MRI(*B.getMRI()),
-        Subtarget(MF.getSubtarget<GCNSubtarget>()),
-        RBI(*Subtarget.getRegBankInfo()), TRI(*Subtarget.getRegisterInfo()),
-        TII(*Subtarget.getInstrInfo()), Helper(Helper){};
+  AMDGPURegBankCombinerImpl(
+      const AMDGPURegBankCombinerImplRuleConfig &RuleConfig,
+      MachineIRBuilder &B, CombinerHelper &Helper,
+      GISelChangeObserver &Observer);
+
+  static const char *getName() { return "AMDGPURegBankCombinerImpl"; }
 
-  bool isVgprRegBank(Register Reg);
-  Register getAsVgpr(Register Reg);
+  bool tryCombineAll(MachineInstr &I) const;
+
+  bool isVgprRegBank(Register Reg) const;
+  Register getAsVgpr(Register Reg) const;
 
   struct MinMaxMedOpc {
     unsigned Min, Max, Med;
   };
 
   struct Med3MatchInfo {
     unsigned Opc;
     Register Val0, Val1, Val2;
   };
 
-  MinMaxMedOpc getMinMaxPair(unsigned Opc);
+  MinMaxMedOpc getMinMaxPair(unsigned Opc) const;
 
   template <class m_Cst, typename CstTy>
   bool matchMed(MachineInstr &MI, MachineRegisterInfo &MRI, MinMaxMedOpc MMMOpc,
-                Register &Val, CstTy &K0, CstTy &K1);
+                Register &Val, CstTy &K0, CstTy &K1) const;
 
-  bool matchIntMinMaxToMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo);
-  bool matchFPMinMaxToMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo);
-  bool matchFPMinMaxToClamp(MachineInstr &MI, Register &Reg);
-  bool matchFPMed3ToClamp(MachineInstr &MI, Register &Reg);
-  void applyMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo);
-  void applyClamp(MachineInstr &MI, Register &Reg);
+  bool matchIntMinMaxToMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo) const;
+  bool matchFPMinMaxToMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo) const;
+  bool matchFPMinMaxToClamp(MachineInstr &MI, Register &Reg) const;
+  bool matchFPMed3ToClamp(MachineInstr &MI, Register &Reg) const;
+  void applyMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo) const;
+  void applyClamp(MachineInstr &MI, Register &Reg) const;
 
 private:
-  SIModeRegisterDefaults getMode();
-  bool getIEEE();
-  bool getDX10Clamp();
-  bool isFminnumIeee(const MachineInstr &MI);
-  bool isFCst(MachineInstr *MI);
-  bool isClampZeroToOne(MachineInstr *K0, MachineInstr *K1);
+  SIModeRegisterDefaults getMode() const;
+  bool getIEEE() const;
+  bool getDX10Clamp() const;
+  bool isFminnumIeee(const MachineInstr &MI) const;
+  bool isFCst(MachineInstr *MI) const;
+  bool isClampZeroToOne(MachineInstr *K0, MachineInstr *K1) const;
+
+#define GET_GICOMBINER_CLASS_MEMBERS
+#define AMDGPUSubtarget GCNSubtarget
+#include "AMDGPUGenRegBankGICombiner.inc"
+#undef GET_GICOMBINER_CLASS_MEMBERS
+#undef AMDGPUSubtarget
 };
 
-bool AMDGPURegBankCombinerHelper::isVgprRegBank(Register Reg) {
+#define GET_GICOMBINER_IMPL
+#define AMDGPUSubtarget GCNSubtarget
+#include "AMDGPUGenRegBankGICombiner.inc"
+#undef AMDGPUSubtarget
+#undef GET_GICOMBINER_IMPL
+
+AMDGPURegBankCombinerImpl::AMDGPURegBankCombinerImpl(
+    const AMDGPURegBankCombinerImplRuleConfig &RuleConfig, MachineIRBuilder &B,
+    CombinerHelper &Helper, GISelChangeObserver &Observer)
+    : RuleConfig(RuleConfig), B(B), MF(B.getMF()), MRI(*B.getMRI()),
+      STI(MF.getSubtarget<GCNSubtarget>()), RBI(*STI.getRegBankInfo()),
+      TRI(*STI.getRegisterInfo()), TII(*STI.getInstrInfo()), Helper(Helper),
+      Observer(Observer),
+#define GET_GICOMBINER_CONSTRUCTOR_INITS
+#include "AMDGPUGenRegBankGICombiner.inc"
+#undef GET_GICOMBINER_CONSTRUCTOR_INITS
+{
+}
+
+bool AMDGPURegBankCombinerImpl::isVgprRegBank(Register Reg) const {
   return RBI.getRegBank(Reg, MRI, TRI)->getID() == AMDGPU::VGPRRegBankID;
 }
 
-Register AMDGPURegBankCombinerHelper::getAsVgpr(Register Reg) {
+Register AMDGPURegBankCombinerImpl::getAsVgpr(Register Reg) const {
   if (isVgprRegBank(Reg))
     return Reg;
 
   // Search for existing copy of Reg to vgpr.
   for (MachineInstr &Use : MRI.use_instructions(Reg)) {
     Register Def = Use.getOperand(0).getReg();
     if (Use.getOpcode() == AMDGPU::COPY && isVgprRegBank(Def))
       return Def;
   }
 
   // Copy Reg to vgpr.
   Register VgprReg = B.buildCopy(MRI.getType(Reg), Reg).getReg(0);
   MRI.setRegBank(VgprReg, RBI.getRegBank(AMDGPU::VGPRRegBankID));
   return VgprReg;
 }
 
-AMDGPURegBankCombinerHelper::MinMaxMedOpc
-AMDGPURegBankCombinerHelper::getMinMaxPair(unsigned Opc) {
+AMDGPURegBankCombinerImpl::MinMaxMedOpc
+AMDGPURegBankCombinerImpl::getMinMaxPair(unsigned Opc) const {
   switch (Opc) {
   default:
     llvm_unreachable("Unsupported opcode");
   case AMDGPU::G_SMAX:
   case AMDGPU::G_SMIN:
     return {AMDGPU::G_SMIN, AMDGPU::G_SMAX, AMDGPU::G_AMDGPU_SMED3};
   case AMDGPU::G_UMAX:
   case AMDGPU::G_UMIN:
     return {AMDGPU::G_UMIN, AMDGPU::G_UMAX, AMDGPU::G_AMDGPU_UMED3};
   case AMDGPU::G_FMAXNUM:
   case AMDGPU::G_FMINNUM:
     return {AMDGPU::G_FMINNUM, AMDGPU::G_FMAXNUM, AMDGPU::G_AMDGPU_FMED3};
   case AMDGPU::G_FMAXNUM_IEEE:
   case AMDGPU::G_FMINNUM_IEEE:
     return {AMDGPU::G_FMINNUM_IEEE, AMDGPU::G_FMAXNUM_IEEE,
             AMDGPU::G_AMDGPU_FMED3};
   }
 }
 
 template <class m_Cst, typename CstTy>
-bool AMDGPURegBankCombinerHelper::matchMed(MachineInstr &MI,
-                                           MachineRegisterInfo &MRI,
-                                           MinMaxMedOpc MMMOpc, Register &Val,
-                                           CstTy &K0, CstTy &K1) {
+bool AMDGPURegBankCombinerImpl::matchMed(MachineInstr &MI,
+                                         CstTy &K0, CstTy &K1) const {
   // 4 operand commutes of: min(max(Val, K0), K1).
   // Find K1 from outer instr: min(max(...), K1) or min(K1, max(...)).
   // Find K0 and Val from inner instr: max(K0, Val) or max(Val, K0).
   // 4 operand commutes of: max(min(Val, K1), K0).
   // Find K0 from outer instr: max(min(...), K0) or max(K0, min(...)).
   // Find K1 and Val from inner instr: min(K1, Val) or min(Val, K1).
   return mi_match(
       MI, MRI,
       m_any_of(
           m_CommutativeBinOp(
               MMMOpc.Min, m_CommutativeBinOp(MMMOpc.Max, m_Reg(Val), m_Cst(K0)),
               m_Cst(K1)),
           m_CommutativeBinOp(
               MMMOpc.Max, m_CommutativeBinOp(MMMOpc.Min, m_Reg(Val), m_Cst(K1)),
               m_Cst(K0))));
 }
 
-bool AMDGPURegBankCombinerHelper::matchIntMinMaxToMed3(
-    MachineInstr &MI, Med3MatchInfo &MatchInfo) {
+bool AMDGPURegBankCombinerImpl::matchIntMinMaxToMed3(
+    MachineInstr &MI, Med3MatchInfo &MatchInfo) const {
   Register Dst = MI.getOperand(0).getReg();
   if (!isVgprRegBank(Dst))
     return false;
 
   // med3 for i16 is only available on gfx9+, and not available for v2i16.
   LLT Ty = MRI.getType(Dst);
-  if ((Ty != LLT::scalar(16) || !Subtarget.hasMed3_16()) &&
-      Ty != LLT::scalar(32))
+  if ((Ty != LLT::scalar(16) || !STI.hasMed3_16()) && Ty != LLT::scalar(32))
     return false;
 
   MinMaxMedOpc OpcodeTriple = getMinMaxPair(MI.getOpcode());
   Register Val;
   std::optional<ValueAndVReg> K0, K1;
   // Match min(max(Val, K0), K1) or max(min(Val, K1), K0). Then see if K0 <= K1.
   if (!matchMed<GCstAndRegMatch>(MI, MRI, OpcodeTriple, Val, K0, K1))
     return false;

 // Val = SNaN only for ieee = true
 // fmed3(SNaN, K0, K1) = min(min(SNaN, K0), K1) = min(QNaN, K1) = K1
 // min(max(SNaN, K0), K1) = min(QNaN, K1) = K1
 // max(min(SNaN, K1), K0) = max(K1, K0) = K1
 // Val = NaN,ieee = false or Val = QNaN,ieee = true
 // fmed3(NaN, K0, K1) = min(min(NaN, K0), K1) = min(K0, K1) = K0
 // min(max(NaN, K0), K1) = min(K0, K1) = K0 (can clamp when dx10_clamp = true)
 // max(min(NaN, K1), K0) = max(K1, K0) = K1 != K0
-bool AMDGPURegBankCombinerHelper::matchFPMinMaxToMed3(
-    MachineInstr &MI, Med3MatchInfo &MatchInfo) {
+bool AMDGPURegBankCombinerImpl::matchFPMinMaxToMed3(
+    MachineInstr &MI, Med3MatchInfo &MatchInfo) const {
   Register Dst = MI.getOperand(0).getReg();
   LLT Ty = MRI.getType(Dst);
 
   // med3 for f16 is only available on gfx9+, and not available for v2f16.
-  if ((Ty != LLT::scalar(16) || !Subtarget.hasMed3_16()) &&
-      Ty != LLT::scalar(32))
+  if ((Ty != LLT::scalar(16) || !STI.hasMed3_16()) && Ty != LLT::scalar(32))
     return false;
 
   auto OpcodeTriple = getMinMaxPair(MI.getOpcode());
 
   Register Val;
   std::optional<FPValueAndVReg> K0, K1;
   // Match min(max(Val, K0), K1) or max(min(Val, K1), K0). Then see if K0 <= K1.
   if (!matchMed<GFCstAndRegMatch>(MI, MRI, OpcodeTriple, Val, K0, K1))
if ((!MRI.hasOneNonDBGUse(K0->VReg) || TII.isInlineConstant(K0->Value)) &&
       MatchInfo = {OpcodeTriple.Med, Val, K0->VReg, K1->VReg};
       return true;
     }
   }
 
   return false;
 }
 
-bool AMDGPURegBankCombinerHelper::matchFPMinMaxToClamp(MachineInstr &MI,
-                                                       Register &Reg) {
+bool AMDGPURegBankCombinerImpl::matchFPMinMaxToClamp(MachineInstr &MI,
+                                                     Register &Reg) const {
   // Clamp is available on all types after regbankselect (f16, f32, f64, v2f16).
   auto OpcodeTriple = getMinMaxPair(MI.getOpcode());
   Register Val;
   std::optional<FPValueAndVReg> K0, K1;
   // Match min(max(Val, K0), K1) or max(min(Val, K1), K0).
   if (!matchMed<GFCstOrSplatGFCstMatch>(MI, MRI, OpcodeTriple, Val, K0, K1))
     return false;
 

 // Val = SNaN only for ieee = true. It is important which operand is NaN.
 // min(min(SNaN, 0.0), 1.0) = min(QNaN, 1.0) = 1.0
 // min(min(SNaN, 1.0), 0.0) = min(QNaN, 0.0) = 0.0
 // min(min(0.0, 1.0), SNaN) = min(0.0, SNaN) = QNaN
 // Val = NaN,ieee = false or Val = QNaN,ieee = true
 // min(min(NaN, 0.0), 1.0) = min(0.0, 1.0) = 0.0
 // min(min(NaN, 1.0), 0.0) = min(1.0, 0.0) = 0.0
 // min(min(0.0, 1.0), NaN) = min(0.0, NaN) = 0.0
-bool AMDGPURegBankCombinerHelper::matchFPMed3ToClamp(MachineInstr &MI,
-                                                     Register &Reg) {
+bool AMDGPURegBankCombinerImpl::matchFPMed3ToClamp(MachineInstr &MI,
+                                                   Register &Reg) const {
   // In llvm-ir, clamp is often represented as an intrinsic call to
   // @llvm.amdgcn.fmed3.f32(%Val, 0.0, 1.0). Check for other operand orders.
   MachineInstr *Src0 = getDefIgnoringCopies(MI.getOperand(1).getReg(), MRI);
   MachineInstr *Src1 = getDefIgnoringCopies(MI.getOperand(2).getReg(), MRI);
   MachineInstr *Src2 = getDefIgnoringCopies(MI.getOperand(3).getReg(), MRI);
 
   if (isFCst(Src0) && !isFCst(Src1))
     std::swap(Src0, Src1);
if (isKnownNeverNaN(MI.getOperand(0).getReg(), MRI) ||
        (isKnownNeverSNaN(Val, MRI) || isOp3Zero()))) {
     Reg = Val;
     return true;
   }
 
   return false;
 }
 
-void AMDGPURegBankCombinerHelper::applyClamp(MachineInstr &MI, Register &Reg) {
+void AMDGPURegBankCombinerImpl::applyClamp(MachineInstr &MI,
+                                           Register &Reg) const {
   B.setInstrAndDebugLoc(MI);
   B.buildInstr(AMDGPU::G_AMDGPU_CLAMP, {MI.getOperand(0)}, {Reg},
                MI.getFlags());
   MI.eraseFromParent();
 }
 
-void AMDGPURegBankCombinerHelper::applyMed3(MachineInstr &MI,
-                                            Med3MatchInfo &MatchInfo) {
+void AMDGPURegBankCombinerImpl::applyMed3(MachineInstr &MI,
+                                          Med3MatchInfo &MatchInfo) const {
   B.setInstrAndDebugLoc(MI);
   B.buildInstr(MatchInfo.Opc, {MI.getOperand(0)},
                {getAsVgpr(MatchInfo.Val0), getAsVgpr(MatchInfo.Val1),
                 getAsVgpr(MatchInfo.Val2)},
                MI.getFlags());
   MI.eraseFromParent();
 }
 
-SIModeRegisterDefaults AMDGPURegBankCombinerHelper::getMode() {
+SIModeRegisterDefaults AMDGPURegBankCombinerImpl::getMode() const {
   return MF.getInfo<SIMachineFunctionInfo>()->getMode();
 }
 
-bool AMDGPURegBankCombinerHelper::getIEEE() { return getMode().IEEE; }
+bool AMDGPURegBankCombinerImpl::getIEEE() const { return getMode().IEEE; }
 
-bool AMDGPURegBankCombinerHelper::getDX10Clamp() { return getMode().DX10Clamp; }
+bool AMDGPURegBankCombinerImpl::getDX10Clamp() const {
+  return getMode().DX10Clamp;
+}
 
-bool AMDGPURegBankCombinerHelper::isFminnumIeee(const MachineInstr &MI) {
+bool AMDGPURegBankCombinerImpl::isFminnumIeee(const MachineInstr &MI) const {
   return MI.getOpcode() == AMDGPU::G_FMINNUM_IEEE;
 }
 
-bool AMDGPURegBankCombinerHelper::isFCst(MachineInstr *MI) {
+bool AMDGPURegBankCombinerImpl::isFCst(MachineInstr *MI) const {
   return MI->getOpcode() == AMDGPU::G_FCONSTANT;
 }
 
-bool AMDGPURegBankCombinerHelper::isClampZeroToOne(MachineInstr *K0,
-                                                   MachineInstr *K1) {
+bool AMDGPURegBankCombinerImpl::isClampZeroToOne(MachineInstr *K0,
+                                                 MachineInstr *K1) const {
   if (isFCst(K0) && isFCst(K1)) {
     const ConstantFP *KO_FPImm = K0->getOperand(1).getFPImm();
     const ConstantFP *K1_FPImm = K1->getOperand(1).getFPImm();
     return (KO_FPImm->isExactlyValue(0.0) && K1_FPImm->isExactlyValue(1.0)) ||
            (KO_FPImm->isExactlyValue(1.0) && K1_FPImm->isExactlyValue(0.0));
   }
   return false;
 }
 
-class AMDGPURegBankCombinerHelperState {
-protected:
-  CombinerHelper &Helper;
-  AMDGPURegBankCombinerHelper &RegBankHelper;
-
-public:
-  AMDGPURegBankCombinerHelperState(CombinerHelper &Helper,
-                                   AMDGPURegBankCombinerHelper &RegBankHelper)
-      : Helper(Helper), RegBankHelper(RegBankHelper) {}
-};
-
-#define AMDGPUREGBANKCOMBINERHELPER_GENCOMBINERHELPER_DEPS
-#include "AMDGPUGenRegBankGICombiner.inc"
-#undef AMDGPUREGBANKCOMBINERHELPER_GENCOMBINERHELPER_DEPS
-
-namespace {
-#define AMDGPUREGBANKCOMBINERHELPER_GENCOMBINERHELPER_H
-#include "AMDGPUGenRegBankGICombiner.inc"
-#undef AMDGPUREGBANKCOMBINERHELPER_GENCOMBINERHELPER_H
-
 class AMDGPURegBankCombinerInfo final : public CombinerInfo {
   GISelKnownBits *KB;
   MachineDominatorTree *MDT;
+  AMDGPURegBankCombinerImplRuleConfig RuleConfig;
 
 public:
-  AMDGPUGenRegBankCombinerHelperRuleConfig GeneratedRuleCfg;
-
   AMDGPURegBankCombinerInfo(bool EnableOpt, bool OptSize, bool MinSize,
-                                  const AMDGPULegalizerInfo *LI,
-                                  GISelKnownBits *KB, MachineDominatorTree *MDT)
+                            const AMDGPULegalizerInfo *LI, GISelKnownBits *KB,
+                            MachineDominatorTree *MDT)
       : CombinerInfo(/*AllowIllegalOps*/ false, /*ShouldLegalizeIllegal*/ true,
                      /*LegalizerInfo*/ LI, EnableOpt, OptSize, MinSize),
         KB(KB), MDT(MDT) {
-    if (!GeneratedRuleCfg.parseCommandLineOption())
+    if (!RuleConfig.parseCommandLineOption())
       report_fatal_error("Invalid rule identifier");
   }
 
   bool combine(GISelChangeObserver &Observer, MachineInstr &MI,
                MachineIRBuilder &B) const override;
 };
 
 bool AMDGPURegBankCombinerInfo::combine(GISelChangeObserver &Observer,
-                                              MachineInstr &MI,
-                                              MachineIRBuilder &B) const {
   CombinerHelper Helper(Observer, B, /* IsPreLegalize*/ false, KB, MDT);
-  AMDGPURegBankCombinerHelper RegBankHelper(B, Helper);
-  AMDGPUGenRegBankCombinerHelper Generated(GeneratedRuleCfg, Helper,
-                                           RegBankHelper);
-
-  if (Generated.tryCombineAll(Observer, MI, B))
-    return true;
-
-  return false;
+  // TODO: Do not re-create the Impl on every inst, it should be per function.
+  AMDGPURegBankCombinerImpl Impl(RuleConfig, B, Helper, Observer);
+  Impl.setupMF(*MI.getMF(), KB);
+  return Impl.tryCombineAll(MI);
 }
 
-#define AMDGPUREGBANKCOMBINERHELPER_GENCOMBINERHELPER_CPP
-#include "AMDGPUGenRegBankGICombiner.inc"
-#undef AMDGPUREGBANKCOMBINERHELPER_GENCOMBINERHELPER_CPP
-
 // Pass boilerplate
 // ================
 
 class AMDGPURegBankCombiner : public MachineFunctionPass {
 public:
   static char ID;
 
   AMDGPURegBankCombiner(bool IsOptNone = false);
 
-  StringRef getPassName() const override {
-    return "AMDGPURegBankCombiner";
-  }
+  StringRef getPassName() const override { return "AMDGPURegBankCombiner"; }
 
   bool runOnMachineFunction(MachineFunction &MF) override;
 
   void getAnalysisUsage(AnalysisUsage &AU) const override;
 private:
   bool IsOptNone;
 };
 } // end anonymous namespace
 
 void AMDGPURegBankCombiner::getAnalysisUsage(AnalysisUsage &AU) const {
   AU.addRequired<TargetPassConfig>();
   AU.setPreservesCFG();
   getSelectionDAGFallbackAnalysisUsage(AU);
   AU.addRequired<GISelKnownBitsAnalysis>();
   AU.addPreserved<GISelKnownBitsAnalysis>();
   if (!IsOptNone) {
     AU.addRequired<MachineDominatorTree>();
     AU.addPreserved<MachineDominatorTree>();
   }
   MachineFunctionPass::getAnalysisUsage(AU);
 }
 
 AMDGPURegBankCombiner::AMDGPURegBankCombiner(bool IsOptNone)
     : MachineFunctionPass(ID), IsOptNone(IsOptNone) {
   initializeAMDGPURegBankCombinerPass(*PassRegistry::getPassRegistry());
 }
 
 bool AMDGPURegBankCombiner::runOnMachineFunction(MachineFunction &MF) {
   if (MF.getProperties().hasProperty(
           MachineFunctionProperties::Property::FailedISel))
     return false;
   auto *TPC = &getAnalysis<TargetPassConfig>();
   const Function &F = MF.getFunction();
   bool EnableOpt =
       MF.getTarget().getOptLevel() != CodeGenOpt::None && !skipFunction(F);
 
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
-  const AMDGPULegalizerInfo *LI
-    = static_cast<const AMDGPULegalizerInfo *>(ST.getLegalizerInfo());
+  const AMDGPULegalizerInfo *LI =
+      static_cast<const AMDGPULegalizerInfo *>(ST.getLegalizerInfo());
 
   GISelKnownBits *KB = &getAnalysis<GISelKnownBitsAnalysis>().get(MF);
   MachineDominatorTree *MDT =
       IsOptNone ? nullptr : &getAnalysis<MachineDominatorTree>();
-  AMDGPURegBankCombinerInfo PCInfo(EnableOpt, F.hasOptSize(),
-                                         F.hasMinSize(), LI, KB, MDT);
+  AMDGPURegBankCombinerInfo PCInfo(EnableOpt, F.hasOptSize(), F.hasMinSize(),
+                                   LI, KB, MDT);
   Combiner C(PCInfo, TPC);
   return C.combineMachineInstrs(MF, /*CSEInfo*/ nullptr);
 }
 
 char AMDGPURegBankCombiner::ID = 0;
 INITIALIZE_PASS_BEGIN(AMDGPURegBankCombiner, DEBUG_TYPE,
                       "Combine AMDGPU machine instrs after regbankselect",
                       false, false)

llvm/lib/Target/AMDGPU/CMakeLists.txt

 tablegen(LLVM AMDGPUGenMCPseudoLowering.inc -gen-pseudo-lowering)
 tablegen(LLVM AMDGPUGenRegisterBank.inc -gen-register-bank)
 tablegen(LLVM AMDGPUGenRegisterInfo.inc -gen-register-info)
 tablegen(LLVM AMDGPUGenSearchableTables.inc -gen-searchable-tables)
 tablegen(LLVM AMDGPUGenSubtargetInfo.inc -gen-subtarget)
 
 set(LLVM_TARGET_DEFINITIONS AMDGPUGISel.td)
 tablegen(LLVM AMDGPUGenGlobalISel.inc -gen-global-isel)
-tablegen(LLVM AMDGPUGenPreLegalizeGICombiner.inc -gen-global-isel-combiner
-              -combiners="AMDGPUPreLegalizerCombinerHelper")
-tablegen(LLVM AMDGPUGenPostLegalizeGICombiner.inc -gen-global-isel-combiner
-              -combiners="AMDGPUPostLegalizerCombinerHelper")
-tablegen(LLVM AMDGPUGenRegBankGICombiner.inc -gen-global-isel-combiner
-              -combiners="AMDGPURegBankCombinerHelper")
+tablegen(LLVM AMDGPUGenPreLegalizeGICombiner.inc -gen-global-isel-combiner-matchtable
+              -combiners="AMDGPUPreLegalizerCombiner")
+tablegen(LLVM AMDGPUGenPostLegalizeGICombiner.inc -gen-global-isel-combiner-matchtable
+              -combiners="AMDGPUPostLegalizerCombiner")
+tablegen(LLVM AMDGPUGenRegBankGICombiner.inc -gen-global-isel-combiner-matchtable
+              -combiners="AMDGPURegBankCombiner")
 
 set(LLVM_TARGET_DEFINITIONS R600.td)
 tablegen(LLVM R600GenAsmWriter.inc -gen-asm-writer)
 tablegen(LLVM R600GenCallingConv.inc -gen-callingconv)
 tablegen(LLVM R600GenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM R600GenDFAPacketizer.inc -gen-dfa-packetizer)
 tablegen(LLVM R600GenInstrInfo.inc -gen-instr-info)
 tablegen(LLVM R600GenMCCodeEmitter.inc -gen-emitter)

llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp

class SPIRVInstructionSelector : public InstructionSelector {
   SPIRVGlobalRegistry &GR;
   MachineRegisterInfo *MRI;
 
 public:
   SPIRVInstructionSelector(const SPIRVTargetMachine &TM,
                            const SPIRVSubtarget &ST,
                            const RegisterBankInfo &RBI);
   void setupMF(MachineFunction &MF, GISelKnownBits *KB,
-               CodeGenCoverage &CoverageInfo, ProfileSummaryInfo *PSI,
+               CodeGenCoverage *CoverageInfo, ProfileSummaryInfo *PSI,
                BlockFrequencyInfo *BFI) override;
   // Common selection code. Instruction-specific selection occurs in spvSelect.
   bool select(MachineInstr &I) override;
   static const char *getName() { return DEBUG_TYPE; }
 
 #define GET_GLOBALISEL_PREDICATES_DECL
 #include "SPIRVGenGlobalISel.inc"
 #undef GET_GLOBALISEL_PREDICATES_DECL