add Enzyme timings to -ftime-trace reports (#1892)

* Move preservenvvm to modulepass

* sprinkle some scoped timers around

* two more timers

* remove test trace in NVVM, respond to comments

* clang-format

* fix fmt

* Update ActivityAnalysis.cpp

---------

Co-authored-by: William S. Moses <gh@wsmoses.com>
Co-authored-by: Manuel Drehwald <git@manuel.drehwald.info>

enzyme/Enzyme/ActivityAnalysis.cpp

@@ -46,6 +46,7 @@
 
 #include "llvm/IR/InstIterator.h"
 
+#include "llvm/Support/TimeProfiler.h"
 #include "llvm/Support/raw_ostream.h"
 
 #include "llvm/IR/InlineAsm.h"
@@ -730,6 +731,10 @@ bool isPossibleFloat(const TypeResults &TR, Value *I, const DataLayout &DL) {
 /// do not propagate adjoints themselves
 bool ActivityAnalyzer::isConstantInstruction(TypeResults const &TR,
                                              Instruction *I) {
+
+  TimeTraceScope timeScope("isConstantInstruction",
+                           I->getParent()->getParent()->getName());
+
   // This analysis may only be called by instructions corresponding to
   // the function analyzed by TypeInfo
   assert(I);
@@ -1044,6 +1049,8 @@ bool ActivityAnalyzer::isConstantValue(TypeResults const &TR, Value *Val) {
   // This analysis may only be called by instructions corresponding to
   // the function analyzed by TypeInfo -- however if the Value
   // was created outside a function (e.g. global, constant), that is allowed
+  TimeTraceScope timeScope("isConstantValue");
+
   assert(Val);
   if (auto I = dyn_cast<Instruction>(Val)) {
     if (TR.getFunction() != I->getParent()->getParent()) {

enzyme/Enzyme/Enzyme.cpp

@@ -3198,6 +3198,8 @@ class EnzymeBase {
     Logic.clear();
 
     if (changed && Logic.PostOpt) {
+      TimeTraceScope timeScope("Enzyme PostOpt", M.getName());
+
       PassBuilder PB;
       LoopAnalysisManager LAM;
       FunctionAnalysisManager FAM;

enzyme/Enzyme/EnzymeLogic.cpp

@@ -78,6 +78,7 @@
 #include "llvm/Analysis/GlobalsModRef.h"
 
 #include "llvm/Support/AMDGPUMetadata.h"
+#include "llvm/Support/TimeProfiler.h"
 
 #include "llvm/ADT/StringSet.h"
 
@@ -1976,6 +1977,9 @@ const AugmentedReturn &EnzymeLogic::CreateAugmentedPrimal(
     bool shadowReturnUsed, const FnTypeInfo &oldTypeInfo_,
     const std::vector<bool> _overwritten_args, bool forceAnonymousTape,
     unsigned width, bool AtomicAdd, bool omp) {
+
+  TimeTraceScope timeScope("CreateAugmentedPrimal", todiff->getName());
+
   if (returnUsed)
     assert(!todiff->getReturnType()->isEmptyTy() &&
            !todiff->getReturnType()->isVoidTy());
@@ -3677,6 +3681,8 @@ Function *EnzymeLogic::CreatePrimalAndGradient(
     RequestContext context, const ReverseCacheKey &&key, TypeAnalysis &TA,
     const AugmentedReturn *augmenteddata, bool omp) {
 
+  TimeTraceScope timeScope("CreatePrimalAndGradient", key.todiff->getName());
+
   assert(key.mode == DerivativeMode::ReverseModeCombined ||
          key.mode == DerivativeMode::ReverseModeGradient);
 
@@ -4557,6 +4563,9 @@ Function *EnzymeLogic::CreateForwardDiff(
     llvm::Type *additionalArg, const FnTypeInfo &oldTypeInfo_,
     const std::vector<bool> _overwritten_args,
     const AugmentedReturn *augmenteddata, bool omp) {
+
+  TimeTraceScope timeScope("CreateForwardDiff", todiff->getName());
+
   assert(retType != DIFFE_TYPE::OUT_DIFF);
 
   assert(mode == DerivativeMode::ForwardMode ||

enzyme/Enzyme/FunctionUtils.cpp

@@ -69,6 +69,8 @@
 #include "llvm/Analysis/ScopedNoAliasAA.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 
+#include "llvm/Support/TimeProfiler.h"
+
 #include "llvm/Transforms/IPO/FunctionAttrs.h"
 #include "llvm/Transforms/Utils/Mem2Reg.h"
 
@@ -1416,6 +1418,8 @@ void SplitPHIs(llvm::Function &F) {
 Function *PreProcessCache::preprocessForClone(Function *F,
                                               DerivativeMode mode) {
 
+  TimeTraceScope timeScope("preprocessForClone", F->getName());
+
   if (mode == DerivativeMode::ReverseModeGradient)
     mode = DerivativeMode::ReverseModePrimal;
   if (mode == DerivativeMode::ForwardModeSplit)

enzyme/Enzyme/GradientUtils.cpp

@@ -8242,6 +8242,9 @@ void GradientUtils::eraseFictiousPHIs() {
 }
 
 void GradientUtils::forceActiveDetection() {
+
+  TimeTraceScope timeScope("Activity Analysis", oldFunc->getName());
+
   for (auto &Arg : oldFunc->args()) {
     ATA->isConstantValue(TR, &Arg);
   }

enzyme/Enzyme/PreserveNVVM.cpp

@@ -37,6 +37,8 @@
 #include "llvm/IR/Module.h"
 #include "llvm/Support/raw_ostream.h"
 
+#include "llvm/Support/TimeProfiler.h"
+
 #include "llvm/Pass.h"
 
 #include "llvm/Transforms/Utils.h"

enzyme/Enzyme/TypeAnalysis/TypeAnalysis.cpp

@@ -42,6 +42,7 @@
 #include "llvm/IR/InstIterator.h"
 
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/TimeProfiler.h"
 #include "llvm/Support/raw_ostream.h"
 
 #include "llvm/ADT/SmallSet.h"
@@ -1538,6 +1539,9 @@ void TypeAnalyzer::runPHIHypotheses() {
 }
 
 void TypeAnalyzer::run() {
+
+  TimeTraceScope timeScope("Type Analysis", fntypeinfo.Function->getName());
+
   // This function runs a full round of type analysis.
   // This works by doing two stages of analysis,
   // with a "deduced integer types for unused" values