[PATCH] D110018: [lld-macho] Speed up markLive()

[Jez Ng via Phabricator via llvm-commits]

int3 added a comment.

I'm a bit ambivalent about this change. The typical way that parallel mark-sweep is implemented is via a work-stealing queue. That way we have one lock per thread, instead of all threads contending on one lock. Granted this diff is already an improvement, but perhaps it would be best to implement the optimal solution right off the bat...



================
Comment at: lld/MachO/InputFiles.cpp:278
         // Copying requires less memory than constructing a fresh InputSection.
-        auto *copy = make<ConcatInputSection>(*isec);
+        auto copy = make<ConcatInputSection>(
+            segname, name, this, data.slice(0, recordSize), align, flags);
----------------
why this change?


================
Comment at: lld/MachO/InputSection.h:120
+  void markLive(uint64_t off) override {
+    const std::lock_guard<std::mutex> l(liveNessMutex);
+    live = true;
----------------
I think this lock isn't necessary... multiple concurrent calls to this method will still end up setting `live` to true


================
Comment at: lld/MachO/InputSection.h:259-260
+  void markLive(uint64_t off) override {
+    const std::lock_guard<std::mutex> l(liveNessMutex);
+    live[off >> power2LiteralSize] = 1;
+  }
----------------
atomics might perform better here, assuming low contention


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110018/new/

https://reviews.llvm.org/D110018