[PATCH] D155677: [lld-macho] Switch to xxh3_64bits

[Fangrui Song via Phabricator via llvm-commits]

MaskRay created this revision.
MaskRay added a reviewer: lld-macho.
Herald added projects: lld-macho, All.
MaskRay requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

xxh3 is substantially faster than xxh64.
For lld/ELF, there is substantial speedup in `.debug_str` duplicate
elimination (D154813 <https://reviews.llvm.org/D154813>). Use xxh3 for lld-macho as well.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D155677

Files:
  lld/MachO/ICF.cpp
  lld/MachO/InputSection.cpp
  lld/MachO/SyntheticSections.cpp
  lld/MachO/Writer.cpp


Index: lld/MachO/Writer.cpp
===================================================================
--- lld/MachO/Writer.cpp
+++ lld/MachO/Writer.cpp
@@ -1190,14 +1190,14 @@
   threadFutures.reserve(chunks.size());
   for (size_t i = 0; i < chunks.size(); ++i)
     threadFutures.emplace_back(threadPool.async(
-        [&](size_t j) { hashes[j] = xxHash64(chunks[j]); }, i));
+        [&](size_t j) { hashes[j] = xxh3_64bits(chunks[j]); }, i));
   for (std::shared_future<void> &future : threadFutures)
     future.wait();
   // Append the output filename so that identical binaries with different names
   // don't get the same UUID.
-  hashes[chunks.size()] = xxHash64(sys::path::filename(config->finalOutput));
-  uint64_t digest = xxHash64({reinterpret_cast<uint8_t *>(hashes.data()),
-                              hashes.size() * sizeof(uint64_t)});
+  hashes[chunks.size()] = xxh3_64bits(sys::path::filename(config->finalOutput));
+  uint64_t digest = xxh3_64bits({reinterpret_cast<uint8_t *>(hashes.data()),
+                                 hashes.size() * sizeof(uint64_t)});
   uuidCommand->writeUuid(digest);
 }
 
Index: lld/MachO/SyntheticSections.cpp
===================================================================
--- lld/MachO/SyntheticSections.cpp
+++ lld/MachO/SyntheticSections.cpp
@@ -1677,7 +1677,7 @@
 DeduplicatedCStringSection::StringOffset
 DeduplicatedCStringSection::getStringOffset(StringRef str) const {
   // StringPiece uses 31 bits to store the hashes, so we replicate that
-  uint32_t hash = xxHash64(str) & 0x7fffffff;
+  uint32_t hash = xxh3_64bits(str) & 0x7fffffff;
   auto offset = stringOffsetMap.find(CachedHashStringRef(str, hash));
   assert(offset != stringOffsetMap.end() &&
          "Looked-up strings should always exist in section");
Index: lld/MachO/InputSection.cpp
===================================================================
--- lld/MachO/InputSection.cpp
+++ lld/MachO/InputSection.cpp
@@ -246,7 +246,7 @@
     size_t end = s.find(0);
     if (end == StringRef::npos)
       fatal(getLocation(off) + ": string is not null terminated");
-    uint32_t hash = deduplicateLiterals ? xxHash64(s.take_front(end)) : 0;
+    uint32_t hash = deduplicateLiterals ? xxh3_64bits(s.take_front(end)) : 0;
     pieces.emplace_back(off, hash);
     size_t size = end + 1; // include null terminator
     s = s.substr(size);
Index: lld/MachO/ICF.cpp
===================================================================
--- lld/MachO/ICF.cpp
+++ lld/MachO/ICF.cpp
@@ -457,7 +457,7 @@
     assert(isec->icfEqClass[0] == 0); // don't overwrite a unique ID!
     // Turn-on the top bit to guarantee that valid hashes have no collisions
     // with the small-integer unique IDs for ICF-ineligible sections
-    isec->icfEqClass[0] = xxHash64(isec->data) | (1ull << 31);
+    isec->icfEqClass[0] = xxh3_64bits(isec->data) | (1ull << 31);
   });
   // Now that every input section is either hashed or marked as unique, run the
   // segregation algorithm to detect foldable subsections.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D155677.541861.patch
Type: text/x-patch
Size: 3024 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230719/43c46530/attachment.bin>