[memprof] Speed up llvm-profdata (#117446)

CallStackRadixTreeBuilder::build takes the parameter
MemProfFrameIndexes by value, involving copies:

  std::optional<const llvm::DenseMap<FrameIdTy, LinearFrameId>>
    MemProfFrameIndexes

Then "build" makes another copy of MemProfFrameIndexe and passes it to
encodeCallStack for every call stack, which is painfully slow.

This patch changes the type to a pointer so that we don't have to make
a copy every time we pass the argument.

Without this patch, it takes 553 seconds to run "llvm-profdata merge"
on a large MemProf raw profile.  This patch shortenes that down to 67
seconds.

Author: kazutakahirata

llvm/include/llvm/ProfileData/MemProf.h

@@ -1117,21 +1117,20 @@ template <typename FrameIdTy> class CallStackRadixTreeBuilder {
 
   // Encode a call stack into RadixArray.  Return the starting index within
   // RadixArray.
-  LinearCallStackId
-  encodeCallStack(const llvm::SmallVector<FrameIdTy> *CallStack,
-                  const llvm::SmallVector<FrameIdTy> *Prev,
-                  std::optional<const llvm::DenseMap<FrameIdTy, LinearFrameId>>
-                      MemProfFrameIndexes);
+  LinearCallStackId encodeCallStack(
+      const llvm::SmallVector<FrameIdTy> *CallStack,
+      const llvm::SmallVector<FrameIdTy> *Prev,
+      const llvm::DenseMap<FrameIdTy, LinearFrameId> *MemProfFrameIndexes);
 
 public:
   CallStackRadixTreeBuilder() = default;
 
   // Build a radix tree array.
-  void build(llvm::MapVector<CallStackId, llvm::SmallVector<FrameIdTy>>
-                 &&MemProfCallStackData,
-             std::optional<const llvm::DenseMap<FrameIdTy, LinearFrameId>>
-                 MemProfFrameIndexes,
-             llvm::DenseMap<FrameIdTy, FrameStat> &FrameHistogram);
+  void
+  build(llvm::MapVector<CallStackId, llvm::SmallVector<FrameIdTy>>
+            &&MemProfCallStackData,
+        const llvm::DenseMap<FrameIdTy, LinearFrameId> *MemProfFrameIndexes,
+        llvm::DenseMap<FrameIdTy, FrameStat> &FrameHistogram);
 
   ArrayRef<LinearFrameId> getRadixArray() const { return RadixArray; }
 

llvm/lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -4235,7 +4235,7 @@ static DenseMap<CallStackId, LinearCallStackId> writeMemoryProfileRadixTree(
   CallStackRadixTreeBuilder<LinearFrameId> Builder;
   // We don't need a MemProfFrameIndexes map as we have already converted the
   // full stack id hash to a linear offset into the StackIds array.
-  Builder.build(std::move(CallStacks), /*MemProfFrameIndexes=*/std::nullopt,
+  Builder.build(std::move(CallStacks), /*MemProfFrameIndexes=*/nullptr,
                 FrameHistogram);
   Stream.EmitRecord(bitc::FS_CONTEXT_RADIX_TREE_ARRAY, Builder.getRadixArray(),
                     RadixAbbrev);

llvm/lib/ProfileData/InstrProfWriter.cpp

@@ -641,7 +641,7 @@ writeMemProfCallStackArray(
       MemProfCallStackIndexes;
 
   memprof::CallStackRadixTreeBuilder<memprof::FrameId> Builder;
-  Builder.build(std::move(MemProfCallStackData), MemProfFrameIndexes,
+  Builder.build(std::move(MemProfCallStackData), &MemProfFrameIndexes,
                 FrameHistogram);
   for (auto I : Builder.getRadixArray())
     OS.write32(I);

llvm/lib/ProfileData/MemProf.cpp

@@ -335,8 +335,7 @@ template <typename FrameIdTy>
 LinearCallStackId CallStackRadixTreeBuilder<FrameIdTy>::encodeCallStack(
     const llvm::SmallVector<FrameIdTy> *CallStack,
     const llvm::SmallVector<FrameIdTy> *Prev,
-    std::optional<const llvm::DenseMap<FrameIdTy, LinearFrameId>>
-        MemProfFrameIndexes) {
+    const llvm::DenseMap<FrameIdTy, LinearFrameId> *MemProfFrameIndexes) {
   // Compute the length of the common root prefix between Prev and CallStack.
   uint32_t CommonLen = 0;
   if (Prev) {
@@ -381,8 +380,7 @@ template <typename FrameIdTy>
 void CallStackRadixTreeBuilder<FrameIdTy>::build(
     llvm::MapVector<CallStackId, llvm::SmallVector<FrameIdTy>>
         &&MemProfCallStackData,
-    std::optional<const llvm::DenseMap<FrameIdTy, LinearFrameId>>
-        MemProfFrameIndexes,
+    const llvm::DenseMap<FrameIdTy, LinearFrameId> *MemProfFrameIndexes,
     llvm::DenseMap<FrameIdTy, FrameStat> &FrameHistogram) {
   // Take the vector portion of MemProfCallStackData.  The vector is exactly
   // what we need to sort.  Also, we no longer need its lookup capability.

llvm/unittests/ProfileData/MemProfTest.cpp

@@ -628,7 +628,7 @@ TEST(MemProf, RadixTreeBuilderEmpty) {
       FrameHistogram =
           llvm::memprof::computeFrameHistogram<FrameId>(MemProfCallStackData);
   llvm::memprof::CallStackRadixTreeBuilder<FrameId> Builder;
-  Builder.build(std::move(MemProfCallStackData), MemProfFrameIndexes,
+  Builder.build(std::move(MemProfCallStackData), &MemProfFrameIndexes,
                 FrameHistogram);
   ASSERT_THAT(Builder.getRadixArray(), testing::IsEmpty());
   const auto Mappings = Builder.takeCallStackPos();
@@ -646,7 +646,7 @@ TEST(MemProf, RadixTreeBuilderOne) {
       FrameHistogram =
           llvm::memprof::computeFrameHistogram<FrameId>(MemProfCallStackData);
   llvm::memprof::CallStackRadixTreeBuilder<FrameId> Builder;
-  Builder.build(std::move(MemProfCallStackData), MemProfFrameIndexes,
+  Builder.build(std::move(MemProfCallStackData), &MemProfFrameIndexes,
                 FrameHistogram);
   EXPECT_THAT(Builder.getRadixArray(), testing::ElementsAreArray({
                                            3U, // Size of CS1,
@@ -673,7 +673,7 @@ TEST(MemProf, RadixTreeBuilderTwo) {
       FrameHistogram =
           llvm::memprof::computeFrameHistogram<FrameId>(MemProfCallStackData);
   llvm::memprof::CallStackRadixTreeBuilder<FrameId> Builder;
-  Builder.build(std::move(MemProfCallStackData), MemProfFrameIndexes,
+  Builder.build(std::move(MemProfCallStackData), &MemProfFrameIndexes,
                 FrameHistogram);
   EXPECT_THAT(Builder.getRadixArray(),
               testing::ElementsAreArray({
@@ -711,7 +711,7 @@ TEST(MemProf, RadixTreeBuilderSuccessiveJumps) {
       FrameHistogram =
           llvm::memprof::computeFrameHistogram<FrameId>(MemProfCallStackData);
   llvm::memprof::CallStackRadixTreeBuilder<FrameId> Builder;
-  Builder.build(std::move(MemProfCallStackData), MemProfFrameIndexes,
+  Builder.build(std::move(MemProfCallStackData), &MemProfFrameIndexes,
                 FrameHistogram);
   EXPECT_THAT(Builder.getRadixArray(),
               testing::ElementsAreArray({