Add smalltypeof to inactive

EnzymeAD · Nov 4, 2023 · 4cf8160 · 4cf8160
1 parent 1d69c3e
commit 4cf8160
Show file tree

Hide file tree

Showing 8 changed files with 25 additions and 16 deletions.
diff --git a/enzyme/Enzyme/ActivityAnalysis.cpp b/enzyme/Enzyme/ActivityAnalysis.cpp
@@ -109,6 +109,7 @@ const char *KnownInactiveFunctionsContains[] = {
  "__enzyme_pointer"};
 
 const StringSet<> InactiveGlobals = {
+ "small_typeof",
  "ompi_request_null",
  "ompi_mpi_double",
  "ompi_mpi_comm_world",

diff --git a/enzyme/Enzyme/GradientUtils.cpp b/enzyme/Enzyme/GradientUtils.cpp
@@ -6910,7 +6910,8 @@ Value *GradientUtils::lookupM(Value *val, IRBuilder<> &BuilderM,
 
  ValueToValueMapTy available;
  for (const auto &pair : originalToNewFn) {
- available[pair.first] = pair.second;
+ if (pair.first->getType() == pair.second->getType())
+ available[pair.first] = pair.second;
  }
 
  // Sort so that later instructions do not dominate earlier
@@ -6924,6 +6925,7 @@ Value *GradientUtils::lookupM(Value *val, IRBuilder<> &BuilderM,
  auto uw = cast<Instruction>(
  unwrapM(a, v, available, UnwrapMode::AttemptSingleUnwrap,
  /*scope*/ nullptr, /*cache*/ false));
+ assert(uw->getType() == a->getType());
  for (size_t i = 0; i < uw->getNumOperands(); i++) {
  auto op = uw->getOperand(i);
  if (auto arg = dyn_cast<Argument>(op))
@@ -6947,6 +6949,7 @@ Value *GradientUtils::lookupM(Value *val, IRBuilder<> &BuilderM,
  available.clear();
  for (auto I : llvm::reverse(InsertedInstructions)) {
  assert(I->getNumUses() == 0);
+ OrigSE.forgetValue(I);
  I->eraseFromParent();
  }
 #endif

diff --git a/enzyme/Enzyme/Utils.cpp b/enzyme/Enzyme/Utils.cpp
@@ -1914,7 +1914,8 @@ bool overwritesToMemoryReadBy(llvm::AAResults &AA, llvm::TargetLibraryInfo &TLI,
 
  if (auto LI = dyn_cast<LoadInst>(maybeReader)) {
  LoadBegin = SE.getSCEV(LI->getPointerOperand());
- if (LoadBegin != SE.getCouldNotCompute()) {
+ if (LoadBegin != SE.getCouldNotCompute() &&
+ !LoadBegin->getType()->isIntegerTy()) {
  auto &DL = maybeWriter->getModule()->getDataLayout();
  auto width = cast<IntegerType>(DL.getIndexType(LoadBegin->getType()))
  ->getBitWidth();
@@ -1930,7 +1931,8 @@ bool overwritesToMemoryReadBy(llvm::AAResults &AA, llvm::TargetLibraryInfo &TLI,
  }
  if (auto SI = dyn_cast<StoreInst>(maybeWriter)) {
  StoreBegin = SE.getSCEV(SI->getPointerOperand());
- if (StoreBegin != SE.getCouldNotCompute()) {
+ if (StoreBegin != SE.getCouldNotCompute() &&
+ !StoreBegin->getType()->isIntegerTy()) {
  auto &DL = maybeWriter->getModule()->getDataLayout();
  auto width = cast<IntegerType>(DL.getIndexType(StoreBegin->getType()))
  ->getBitWidth();
@@ -1948,7 +1950,8 @@ bool overwritesToMemoryReadBy(llvm::AAResults &AA, llvm::TargetLibraryInfo &TLI,
  }
  if (auto MS = dyn_cast<MemSetInst>(maybeWriter)) {
  StoreBegin = SE.getSCEV(MS->getArgOperand(0));
- if (StoreBegin != SE.getCouldNotCompute()) {
+ if (StoreBegin != SE.getCouldNotCompute() &&
+ !StoreBegin->getType()->isIntegerTy()) {
  if (auto Len = dyn_cast<ConstantInt>(MS->getArgOperand(2))) {
  auto &DL = MS->getModule()->getDataLayout();
  auto width = cast<IntegerType>(DL.getIndexType(StoreBegin->getType()))
@@ -1961,7 +1964,8 @@ bool overwritesToMemoryReadBy(llvm::AAResults &AA, llvm::TargetLibraryInfo &TLI,
  }
  if (auto MS = dyn_cast<MemTransferInst>(maybeWriter)) {
  StoreBegin = SE.getSCEV(MS->getArgOperand(0));
- if (StoreBegin != SE.getCouldNotCompute()) {
+ if (StoreBegin != SE.getCouldNotCompute() &&
+ !StoreBegin->getType()->isIntegerTy()) {
  if (auto Len = dyn_cast<ConstantInt>(MS->getArgOperand(2))) {
  auto &DL = MS->getModule()->getDataLayout();
  auto width = cast<IntegerType>(DL.getIndexType(StoreBegin->getType()))
@@ -1974,7 +1978,8 @@ bool overwritesToMemoryReadBy(llvm::AAResults &AA, llvm::TargetLibraryInfo &TLI,
  }
  if (auto MS = dyn_cast<MemTransferInst>(maybeReader)) {
  LoadBegin = SE.getSCEV(MS->getArgOperand(1));
- if (LoadBegin != SE.getCouldNotCompute()) {
+ if (LoadBegin != SE.getCouldNotCompute() &&
+ !LoadBegin->getType()->isIntegerTy()) {
  if (auto Len = dyn_cast<ConstantInt>(MS->getArgOperand(2))) {
  auto &DL = MS->getModule()->getDataLayout();
  auto width = cast<IntegerType>(DL.getIndexType(LoadBegin->getType()))

diff --git a/enzyme/test/Enzyme/ReverseMode/addrbug.ll b/enzyme/test/Enzyme/ReverseMode/addrbug.ll
@@ -1,5 +1,5 @@
-; RUN: if [ %llvmver -lt 16 ]; then %opt < %s %loadEnzyme -enzyme -enzyme-preopt=false -mem2reg -instsimplify -loop-deletion -simplifycfg -correlated-propagation -adce -instsimplify -S | FileCheck %s; fi
-; RUN: %opt < %s %newLoadEnzyme -passes="enzyme,function(mem2reg,instsimplify,loop(loop-deletion),%simplifycfg,correlated-propagation,adce,instsimplify)" -enzyme-preopt=false -S | FileCheck %s
+; RUN: if [ %llvmver -lt 16 ] && [ %llvmver -gt 11 ]; then %opt < %s %loadEnzyme -enzyme -enzyme-preopt=false -mem2reg -instsimplify -loop-deletion -simplifycfg -correlated-propagation -adce -instsimplify -S | FileCheck %s; fi
+; RUN: if [ %llvmver -gt 11 ]; then %opt < %s %newLoadEnzyme -passes="enzyme,function(mem2reg,instsimplify,loop(loop-deletion),%simplifycfg,correlated-propagation,adce,instsimplify)" -enzyme-preopt=false -S | FileCheck %s; fi
 
 ; Function Attrs: nounwind
 declare void @__enzyme_autodiff(i8*, ...)

diff --git a/enzyme/test/Enzyme/ReverseMode/makememcpy1.ll b/enzyme/test/Enzyme/ReverseMode/makememcpy1.ll
@@ -1,5 +1,5 @@
-; RUN: if [ %llvmver -lt 16 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -loop-deletion -correlated-propagation -adce -simplifycfg -S | FileCheck %s; fi
-; RUN: %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,loop(loop-deletion),correlated-propagation,adce,%simplifycfg)" -S | FileCheck %s
+; RUN: if [ %llvmver -lt 16 ] && [ %llvmver -gt 11 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -loop-deletion -correlated-propagation -adce -simplifycfg -S | FileCheck %s; fi
+; RUN: if [ %llvmver -gt 11 ]; then %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,loop(loop-deletion),correlated-propagation,adce,%simplifycfg)" -S | FileCheck %s; fi
 
 ; This requires the additional optimization to create memcpy's
 

diff --git a/enzyme/test/Enzyme/ReverseMode/metacachelicm.ll b/enzyme/test/Enzyme/ReverseMode/metacachelicm.ll
@@ -1,5 +1,5 @@
-; RUN: if [ %llvmver -lt 16 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s; fi
-; RUN: %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,%simplifycfg)" -S | FileCheck %s
+; RUN: if [ %llvmver -lt 16 ] && [ %llvmver -gt 11 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s; fi
+; RUN: if [ %llvmver -gt 11 ]; then %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,%simplifycfg)" -S | FileCheck %s; fi
 
 ; Function Attrs: nounwind uwtable
 define dso_local void @compute(double* noalias nocapture %data, i64* noalias nocapture readnone %array, double* noalias nocapture %out) #0 {

diff --git a/enzyme/test/Enzyme/ReverseMode/metacachelicm2.ll b/enzyme/test/Enzyme/ReverseMode/metacachelicm2.ll
@@ -1,5 +1,5 @@
-; RUN: if [ %llvmver -lt 16 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s; fi
-; RUN: %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,%simplifycfg)" -S | FileCheck %s
+; RUN: if [ %llvmver -lt 16 ] && [ %llvmver -gt 11 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s; fi
+; RUN: if [ %llvmver -gt 11 ]; then %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,%simplifycfg)" -S | FileCheck %s; fi
 
 ; Function Attrs: nounwind uwtable
 define dso_local void @compute(double* noalias nocapture %data, i64* noalias nocapture readonly %array, double* noalias nocapture %out) #0 {

diff --git a/enzyme/test/Enzyme/ReverseMode/sploop2.ll b/enzyme/test/Enzyme/ReverseMode/sploop2.ll
@@ -1,5 +1,5 @@
-; RUN: if [ %llvmver -lt 16 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -adce -loop-deletion -correlated-propagation -simplifycfg -S | FileCheck %s; fi
-; RUN: %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,adce,loop(loop-deletion),correlated-propagation,%simplifycfg)" -S | FileCheck %s
+; RUN: if [ %llvmver -lt 16 ] && [ %llvmver -gt 11 ]; then %opt < %s %loadEnzyme -enzyme-preopt=false -enzyme -mem2reg -instsimplify -adce -loop-deletion -correlated-propagation -simplifycfg -S | FileCheck %s; fi
+; RUN: if [ %llvmver -gt 11 ]; then %opt < %s %newLoadEnzyme -enzyme-preopt=false -passes="enzyme,function(mem2reg,instsimplify,adce,loop(loop-deletion),correlated-propagation,%simplifycfg)" -S | FileCheck %s; fi
 
 ; This requires the memcpy optimization to run