[EVM] Run some opt passes at the beginning of codegen pipeline

This includes early-cse, gvn, dse, simplifycfg and passes that
optimize GEP instructions.

Author: PavelKopyl

llvm/lib/Analysis/MemoryLocation.cpp

@@ -190,6 +190,10 @@ MemoryLocation MemoryLocation::getForArgument(const CallBase *Call,
       case Intrinsic::memcpy_inline:
       case Intrinsic::memmove:
       case Intrinsic::memset:
+      case Intrinsic::evm_memmoveas1as1:
+      case Intrinsic::evm_memcpyas1as2:
+      case Intrinsic::evm_memcpyas1as3:
+      case Intrinsic::evm_memcpyas1as4:
         if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2)))
           if (LenCI->getValue().getActiveBits() > 64)
             return MemoryLocation::getAfter(Arg, AATags);

llvm/lib/Analysis/VMAliasAnalysis.cpp

@@ -95,6 +95,10 @@ static std::optional<APInt> getConstStartLoc(const MemoryLocation &Loc,
 AliasResult VMAAResult::alias(const MemoryLocation &LocA,
                               const MemoryLocation &LocB, AAQueryInfo &AAQI,
                               const Instruction *I) {
+  if (!isa<PointerType>(LocA.Ptr->getType()) ||
+      !isa<PointerType>(LocB.Ptr->getType()))
+    return AAResultBase::alias(LocA, LocB, AAQI, I);
+
   const unsigned ASA = LocA.Ptr->getType()->getPointerAddressSpace();
   const unsigned ASB = LocB.Ptr->getType()->getPointerAddressSpace();
 

llvm/lib/Target/EVM/EVMTargetMachine.cpp

@@ -25,8 +25,16 @@
 #include "llvm/InitializePasses.h"
 #include "llvm/MC/TargetRegistry.h"
 #include "llvm/Passes/PassBuilder.h"
+#include "llvm/Transforms/IPO.h"
 #include "llvm/Transforms/IPO/GlobalDCE.h"
+#include "llvm/Transforms/InstCombine/InstCombine.h"
+#include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Scalar/DeadStoreElimination.h"
+#include "llvm/Transforms/Scalar/EarlyCSE.h"
+#include "llvm/Transforms/Scalar/GVN.h"
 #include "llvm/Transforms/Scalar/MergeIdenticalBB.h"
+#include "llvm/Transforms/Scalar/NewGVN.h"
+#include "llvm/Transforms/Scalar/SimplifyCFG.h"
 #include "llvm/Transforms/Utils.h"
 
 using namespace llvm;
@@ -141,6 +149,14 @@ void EVMTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
     FAM.registerPass([] { return EVMAA(); });
   });
 
+  PB.registerOptimizerLastEPCallback(
+      [](ModulePassManager &PM, OptimizationLevel Level) {
+        if (Level != OptimizationLevel::O0)
+          // Earlier transformations may expose new opportunities for DSE,
+          // so we run it again.
+          PM.addPass(createModuleToFunctionPassAdaptor(DSEPass()));
+      });
+
   PB.registerPipelineParsingCallback(
       [](StringRef PassName, ModulePassManager &PM,
          ArrayRef<PassBuilder::PipelineElement>) {
@@ -211,6 +227,29 @@ class EVMPassConfig final : public TargetPassConfig {
 void EVMPassConfig::addIRPasses() {
   addPass(createEVMLowerIntrinsicsPass());
   if (TM->getOptLevel() != CodeGenOptLevel::None) {
+    // Call SeparateConstOffsetFromGEP pass to extract constants within
+    // indices and lower a GEP with multiple indices to either arithmetic
+    // operations or multiple GEPs with single index.
+    addPass(createSeparateConstOffsetFromGEPPass(true));
+    // ReassociateGEPs exposes more opportunites for SLSR.
+    addPass(createStraightLineStrengthReducePass());
+    // SeparateConstOffsetFromGEP and SLSR creates common expressions which
+    // GVN or EarlyCSE can reuse.
+    addPass(createNewGVNPass());
+    addPass(createGVNHoistPass());
+    // Run NaryReassociate after EarlyCSE/GVN to be more effective.
+    addPass(createNaryReassociatePass());
+    // Call EarlyCSE pass to find and remove subexpressions in the lowered
+    // result.
+    addPass(createEarlyCSEPass(true));
+    // The DSE pass we run at the end of the optimization pipeline may open
+    // up new opportunities for further CFG simplification.
+    addPass(createCFGSimplificationPass(SimplifyCFGOptions()
+                                            .convertSwitchRangeToICmp(true)
+                                            .hoistCommonInsts(true)
+                                            .sinkCommonInsts(true)));
+    addPass(createLICMPass());
+
     addPass(createEVMAAWrapperPass());
     addPass(createEVMExternalAAWrapperPass());
   }

llvm/test/CodeGen/EVM/O3-pipeline.ll

@@ -9,25 +9,56 @@ target triple = "evm"
 ; CHECK-NEXT: Target Pass Configuration
 ; CHECK-NEXT: Machine Module Information
 ; CHECK-NEXT: Target Transform Information
+; CHECK-NEXT: Assumption Cache Tracker
 ; CHECK-NEXT: EVM Address space based Alias Analysis
 ; CHECK-NEXT: EVM Address space based Alias Analysis Wrapper
 ; CHECK-NEXT: Type-Based Alias Analysis
 ; CHECK-NEXT: Scoped NoAlias Alias Analysis
-; CHECK-NEXT: Assumption Cache Tracker
 ; CHECK-NEXT: Profile summary info
 ; CHECK-NEXT: Create Garbage Collector Module Metadata
 ; CHECK-NEXT: Machine Branch Probability Analysis
 ; CHECK-NEXT:   ModulePass Manager
 ; CHECK-NEXT:     Pre-ISel Intrinsic Lowering
 ; CHECK-NEXT:     EVM Lower Intrinsics
 ; CHECK-NEXT:     FunctionPass Manager
-; CHECK-NEXT:       Module Verifier
-; CHECK-NEXT:       Dominator Tree Construction
-; CHECK-NEXT:       Basic Alias Analysis (stateless AA impl)
-; CHECK-NEXT:       Natural Loop Information
-; CHECK-NEXT:       Canonicalize natural loops
-; CHECK-NEXT:       Scalar Evolution Analysis
-; CHECK-NEXT:       Loop Pass Manager
+; CHECK-NEXT:     Dominator Tree Construction
+; CHECK-NEXT:      Natural Loop Information
+; CHECK-NEXT:      Split GEPs to a variadic base and a constant offset for better CSE
+; CHECK-NEXT:      Scalar Evolution Analysis
+; CHECK-NEXT:      Straight line strength reduction
+; CHECK-NEXT:      Basic Alias Analysis (stateless AA impl)
+; CHECK-NEXT:      Function Alias Analysis Results
+; CHECK-NEXT:      Memory SSA
+; CHECK-NEXT:      Global Value Numbering
+; CHECK-NEXT:      Post-Dominator Tree Construction
+; CHECK-NEXT:      Basic Alias Analysis (stateless AA impl)
+; CHECK-NEXT:      Function Alias Analysis Results
+; CHECK-NEXT:      Memory Dependence Analysis
+; CHECK-NEXT:      Memory SSA
+; CHECK-NEXT:      Early GVN Hoisting of Expressions
+; CHECK-NEXT:      Natural Loop Information
+; CHECK-NEXT:      Scalar Evolution Analysis
+; CHECK-NEXT:      Nary reassociation
+; CHECK-NEXT:      Basic Alias Analysis (stateless AA impl)
+; CHECK-NEXT:      Function Alias Analysis Results
+; CHECK-NEXT:      Memory SSA
+; CHECK-NEXT:      Early CSE w/ MemorySSA
+; CHECK-NEXT:      Simplify the CFG
+; CHECK-NEXT:      Dominator Tree Construction
+; CHECK-NEXT:      Basic Alias Analysis (stateless AA impl)
+; CHECK-NEXT:      Function Alias Analysis Results
+; CHECK-NEXT:      Memory SSA
+; CHECK-NEXT:      Natural Loop Information
+; CHECK-NEXT:      Canonicalize natural loops
+; CHECK-NEXT:      LCSSA Verifier
+; CHECK-NEXT:      Loop-Closed SSA Form Pass
+; CHECK-NEXT:      Scalar Evolution Analysis
+; CHECK-NEXT:      Lazy Branch Probability Analysis
+; CHECK-NEXT:      Lazy Block Frequency Analysis
+; CHECK-NEXT:      Loop Pass Manager
+; CHECK-NEXT:        Loop Invariant Code Motion
+; CHECK-NEXT:      Module Verifier
+; CHECK-NEXT:      Loop Pass Manager
 ; CHECK-NEXT:         Canonicalize Freeze Instructions in Loops
 ; CHECK-NEXT:         Induction Variable Users
 ; CHECK-NEXT:         Loop Strength Reduction

llvm/test/CodeGen/EVM/br.ll

@@ -6,21 +6,21 @@ target triple = "evm"
 
 define i256 @diamond(i256 %rs1, i256 %rs2) nounwind {
 ; CHECK-LABEL: diamond:
-; CHECK:       ; %bb.0:
+; CHECK:       ; %bb.0: ; %end_bb
 ; CHECK-NEXT:    JUMPDEST
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    DUP3
+; CHECK-NEXT:    DUP2
+; CHECK-NEXT:    DUP2
 ; CHECK-NEXT:    SUB
 ; CHECK-NEXT:    PUSH4 @.BB0_2
 ; CHECK-NEXT:    JUMPI
-; CHECK-NEXT:  ; %bb.1: ; %true_bb
+; CHECK-NEXT:  ; %bb.1:
 ; CHECK-NEXT:    DUP1
 ; CHECK-NEXT:    SWAP2
 ; CHECK-NEXT:    POP
 ; CHECK-NEXT:    MUL
 ; CHECK-NEXT:    PUSH4 @.BB0_3
 ; CHECK-NEXT:    JUMP
-; CHECK-NEXT:  .BB0_2: ; %false_bb
+; CHECK-NEXT:  .BB0_2: ; %end_bb
 ; CHECK-NEXT:    JUMPDEST
 ; CHECK-NEXT:    ADD
 ; CHECK-NEXT:  .BB0_3: ; %end_bb

llvm/test/CodeGen/EVM/branch-folder-after-stackification.ll

@@ -8,31 +8,22 @@ define i256 @test(i256 %arg) {
 ; CHECK-LABEL: test:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    JUMPDEST
+; CHECK-NEXT:    PUSH1 0xA
 ; CHECK-NEXT:    PUSH0
-; CHECK-NEXT:    DUP2
+; CHECK-NEXT:    DUP3
 ; CHECK-NEXT:    SLT
-; CHECK-NEXT:    PUSH4 @.BB0_2
-; CHECK-NEXT:    JUMPI
-; CHECK-NEXT:  ; %bb.1:
-; CHECK-NEXT:    PUSH1 0x1
+; CHECK-NEXT:    MUL
 ; CHECK-NEXT:    PUSH1 0x14
-; CHECK-NEXT:    PUSH4 @.BB0_3
-; CHECK-NEXT:    JUMP
-; CHECK-NEXT:  .BB0_2: ; %bb1
-; CHECK-NEXT:    JUMPDEST
-; CHECK-NEXT:    PUSH1 0x1
-; CHECK-NEXT:    PUSH1 0xA
-; CHECK-NEXT:  .BB0_3: ; %bb2
-; CHECK-NEXT:    JUMPDEST
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    SLT
-; CHECK-NEXT:    PUSH4 @.BB0_5
-; CHECK-NEXT:    JUMPI
-; CHECK-NEXT:  ; %bb.4: ; %bb3
-; CHECK-NEXT:    POP
+; CHECK-NEXT:    SUB
+; CHECK-NEXT:    SWAP1
+; CHECK-NEXT:    PUSH0
+; CHECK-NEXT:    DUP3
 ; CHECK-NEXT:    PUSH1 0x5
-; CHECK-NEXT:  .BB0_5: ; %bb4
-; CHECK-NEXT:    JUMPDEST
+; CHECK-NEXT:    SUB
+; CHECK-NEXT:    SWAP2
+; CHECK-NEXT:    SGT
+; CHECK-NEXT:    MUL
+; CHECK-NEXT:    ADD
 ; CHECK-NEXT:    SWAP1
 ; CHECK-NEXT:    JUMP
 entry: