[Support] Add parallelForEach with fixed TaskSize and use it in lld/ELF's Writer

parallelForEach.patch

From 7a28e9d1cc99e997ca8e2d463776aaa7e8c1d6da Mon Sep 17 00:00:00 2001
From: Fangrui Song <>
Date: Mon, 17 Jan 2022 11:25:51 -0800
Subject: [PATCH] [Support] Add parallelForEach with fixed TaskSize and use it
 in lld/ELF's Writer

---
 lld/ELF/OutputSections.cpp           |  2 +-
 llvm/include/llvm/Support/Parallel.h | 21 +++++++++++++++++++++
 2 files changed, 22 insertions(+), 1 deletion(-)

diff --git a/lld/ELF/OutputSections.cpp b/lld/ELF/OutputSections.cpp
index 07ee7d84a2cd..57916a08ca2d 100644
--- a/lld/ELF/OutputSections.cpp
+++ b/lld/ELF/OutputSections.cpp
@@ -353,7 +353,7 @@ template <class ELFT> void OutputSection::writeTo(uint8_t *buf) {
   if (nonZeroFiller)
     fill(buf, sections.empty() ? size : sections[0]->outSecOff, filler);
 
-  parallelForEachN(0, sections.size(), [&](size_t i) {
+  parallelForEachN(0, sections.size(), 128, [&](size_t i) {
     InputSection *isec = sections[i];
     isec->writeTo<ELFT>(buf + isec->outSecOff);
 
diff --git a/llvm/include/llvm/Support/Parallel.h b/llvm/include/llvm/Support/Parallel.h
index 5c3b26d5754c..122b68018d36 100644
--- a/llvm/include/llvm/Support/Parallel.h
+++ b/llvm/include/llvm/Support/Parallel.h
@@ -274,6 +274,27 @@ void parallelForEachN(size_t Begin, size_t End, FuncTy Fn) {
     Fn(I);
 }
 
+template <class IndexTy, class FuncTy>
+void parallelForEach(IndexTy Begin, IndexTy End, IndexTy TaskSize, FuncTy Fn) {
+  IndexTy I = Begin;
+#if LLVM_ENABLE_THREADS
+  parallel::detail::TaskGroup TG;
+  for (; I + TaskSize < End; I += TaskSize) {
+    TG.spawn([=, &Fn] {
+      for (IndexTy J = I, E = I + TaskSize; J != E; ++J)
+        Fn(J);
+    });
+  }
+#endif
+  for (; I != End; ++I)
+    Fn(I);
+}
+
+template <class FuncTy>
+void parallelForEachN(size_t Begin, size_t End, size_t TaskSize, FuncTy Fn) {
+  parallelForEach<size_t>(Begin, End, TaskSize, Fn);
+}
+
 template <class IterTy, class ResultTy, class ReduceFuncTy,
           class TransformFuncTy>
 ResultTy parallelTransformReduce(IterTy Begin, IterTy End, ResultTy Init,
-- 
2.34

2% speedup when linking -DCMAKE_BUILD_TYPE=Release clang.

No significant difference when linking default chrome.