[llvm] [BOLT] Run EliminateUnreachableBlocks in parallel (PR #71299)

[Vladislav Khmelevsky via llvm-commits]

https://github.com/yota9 created https://github.com/llvm/llvm-project/pull/71299

The wall time for this pass decreased on my laptop from ~80 sec to 5
sec.


>From 3c3d5e3c1b6fe1bdeca5352bbb88adb296b6f996 Mon Sep 17 00:00:00 2001
From: Vladislav Khmelevsky <och95 at yandex.ru>
Date: Sun, 5 Nov 2023 01:57:13 +0400
Subject: [PATCH] [BOLT] Run EliminateUnreachableBlocks in parallel

The wall time for this pass decreased on my laptop from ~80 sec to 5
sec.
---
 bolt/include/bolt/Core/BinaryFunction.h |  2 +-
 bolt/lib/Core/BinaryFunction.cpp        |  2 +-
 bolt/lib/Passes/BinaryPasses.cpp        | 48 ++++++++++++++-----------
 3 files changed, 30 insertions(+), 22 deletions(-)

diff --git a/bolt/include/bolt/Core/BinaryFunction.h b/bolt/include/bolt/Core/BinaryFunction.h
index c67ddccbf4892a7..2abaff663624874 100644
--- a/bolt/include/bolt/Core/BinaryFunction.h
+++ b/bolt/include/bolt/Core/BinaryFunction.h
@@ -1445,7 +1445,7 @@ class BinaryFunction {
 
   /// Rebuilds BBs layout, ignoring dead BBs. Returns the number of removed
   /// BBs and the removed number of bytes of code.
-  std::pair<unsigned, uint64_t> eraseInvalidBBs();
+  std::pair<unsigned, uint64_t> eraseInvalidBBs(const MCCodeEmitter *Emitter = nullptr);
 
   /// Get the relative order between two basic blocks in the original
   /// layout.  The result is > 0 if B occurs before A and < 0 if B
diff --git a/bolt/lib/Core/BinaryFunction.cpp b/bolt/lib/Core/BinaryFunction.cpp
index 2e1fa739e53f20b..11aa9b364220a87 100644
--- a/bolt/lib/Core/BinaryFunction.cpp
+++ b/bolt/lib/Core/BinaryFunction.cpp
@@ -322,7 +322,7 @@ void BinaryFunction::markUnreachableBlocks() {
 
 // Any unnecessary fallthrough jumps revealed after calling eraseInvalidBBs
 // will be cleaned up by fixBranches().
-std::pair<unsigned, uint64_t> BinaryFunction::eraseInvalidBBs() {
+std::pair<unsigned, uint64_t> BinaryFunction::eraseInvalidBBs(const MCCodeEmitter *Emitter) {
   DenseSet<const BinaryBasicBlock *> InvalidBBs;
   unsigned Count = 0;
   uint64_t Bytes = 0;
diff --git a/bolt/lib/Passes/BinaryPasses.cpp b/bolt/lib/Passes/BinaryPasses.cpp
index e50fa9dea602bc7..870ea6497b63aca 100644
--- a/bolt/lib/Passes/BinaryPasses.cpp
+++ b/bolt/lib/Passes/BinaryPasses.cpp
@@ -317,11 +317,11 @@ void NormalizeCFG::runOnFunctions(BinaryContext &BC) {
 }
 
 void EliminateUnreachableBlocks::runOnFunction(BinaryFunction &Function) {
-  if (!Function.getLayout().block_empty()) {
-    unsigned Count;
-    uint64_t Bytes;
-    Function.markUnreachableBlocks();
-    LLVM_DEBUG({
+  BinaryContext &BC = Function.getBinaryContext();
+  unsigned Count;
+  uint64_t Bytes;
+  Function.markUnreachableBlocks();
+  LLVM_DEBUG({
       for (BinaryBasicBlock &BB : Function) {
         if (!BB.isValid()) {
           dbgs() << "BOLT-INFO: UCE found unreachable block " << BB.getName()
@@ -330,25 +330,33 @@ void EliminateUnreachableBlocks::runOnFunction(BinaryFunction &Function) {
         }
       }
     });
-    std::tie(Count, Bytes) = Function.eraseInvalidBBs();
-    DeletedBlocks += Count;
-    DeletedBytes += Bytes;
-    if (Count) {
-      Modified.insert(&Function);
-      if (opts::Verbosity > 0)
-        outs() << "BOLT-INFO: removed " << Count
-               << " dead basic block(s) accounting for " << Bytes
-               << " bytes in function " << Function << '\n';
-    }
+  BinaryContext::IndependentCodeEmitter Emitter =
+    BC.createIndependentMCCodeEmitter();
+  std::tie(Count, Bytes) = Function.eraseInvalidBBs(Emitter.MCE.get());
+  DeletedBlocks += Count;
+  DeletedBytes += Bytes;
+  if (Count) {
+    auto L = BC.scopeLock();
+    Modified.insert(&Function);
+    if (opts::Verbosity > 0)
+      outs() << "BOLT-INFO: removed " << Count
+             << " dead basic block(s) accounting for " << Bytes
+             << " bytes in function " << Function << '\n';
   }
 }
 
 void EliminateUnreachableBlocks::runOnFunctions(BinaryContext &BC) {
-  for (auto &It : BC.getBinaryFunctions()) {
-    BinaryFunction &Function = It.second;
-    if (shouldOptimize(Function))
-      runOnFunction(Function);
-  }
+  ParallelUtilities::WorkFuncTy WorkFun = [&](BinaryFunction &BF) {
+    runOnFunction(BF);
+  };
+
+  ParallelUtilities::PredicateTy SkipPredicate = [&](const BinaryFunction &BF) {
+    return !shouldOptimize(BF) || BF.getLayout().block_empty();
+  };
+
+  ParallelUtilities::runOnEachFunction(
+    BC, ParallelUtilities::SchedulingPolicy::SP_CONSTANT, WorkFun,
+    SkipPredicate, "elimininate-unreachable");
 
   if (DeletedBlocks)
     outs() << "BOLT-INFO: UCE removed " << DeletedBlocks << " blocks and "