[PATCH] D104159: [not for review][lld-macho] Simple cstring literal implementation

[Jez Ng via Phabricator via llvm-commits]

int3 created this revision.
Herald added a reviewer: gkm.
Herald added a project: lld-macho.
Herald added a reviewer: lld-macho.
int3 requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

This is just an experiment to see if not creating heavyweight
InputSections for literals is worth it. The answer seems to be "yes".
This diff only creates the additional InputSections w/o deduplicating
them, and it's a 3.6% regression when linking chromium_framework. In
contrast, the StringPiece-based implementation in D102964 <https://reviews.llvm.org/D102964> (which
includes deduplication) is only a 2.2% regression.

      N           Min           Max        Median           Avg        Stddev
  x  20          3.98          4.09         4.025         4.032   0.035629674
  +  20          4.14          4.22         4.175         4.177   0.019493589
  Difference at 95.0% confidence
          0.145 +/- 0.018381
          3.59623% +/- 0.455877%
          (Student's t, pooled s = 0.0287182)


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D104159

Files:
  lld/MachO/InputFiles.cpp


Index: lld/MachO/InputFiles.cpp
===================================================================
--- lld/MachO/InputFiles.cpp
+++ lld/MachO/InputFiles.cpp
@@ -589,6 +589,47 @@
     // We populate subsecMap by repeatedly splitting the last (highest address)
     // subsection.
     SubsectionEntry subsecEntry = subsecMap.back();
+
+    if (sectionType(sectionHeaders[i].flags) == S_CSTRING_LITERALS) {
+      size_t off = 0;
+      StringRef s = toStringRef(subsecEntry.isec->data);
+      size_t nextSymIdxIdx = 0;
+      while (!s.empty()) {
+        InputSection *isec = subsecEntry.isec;
+        size_t end = s.find(0);
+        if (end == StringRef::npos)
+          fatal(toString(this) + ": string is not null terminated");
+        size_t size = end + 1;
+        isec->data = ArrayRef<uint8_t>(
+            reinterpret_cast<const uint8_t *>(s.substr(0, size).data()), size);
+        s = s.substr(size);
+
+        while (nextSymIdxIdx < symbolIndices.size()) {
+          uint32_t symIndex = symbolIndices[nextSymIdxIdx];
+          const NList &sym = nList[symIndex];
+          uint64_t subsecAddr = sectionAddr + off;
+          if (sym.n_value >= subsecAddr + size)
+            break;
+          nextSymIdxIdx++;
+          StringRef name = strtab + sym.n_strx;
+          uint64_t symbolOffset = sym.n_value - subsecAddr;
+          symbols[symIndex] =
+              createDefined(sym, name, isec, symbolOffset, size - symbolOffset);
+        }
+
+        if (s.empty())
+          break;
+
+        auto *nextIsec = make<InputSection>(*isec);
+        nextIsec->numRefs = 0;
+        nextIsec->wasCoalesced = false;
+        off += size;
+        subsecMap.push_back({off, nextIsec});
+        subsecEntry = subsecMap.back();
+      }
+      continue;
+    }
+
     for (size_t j = 0; j < symbolIndices.size(); ++j) {
       uint32_t symIndex = symbolIndices[j];
       const NList &sym = nList[symIndex];


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D104159.351578.patch
Type: text/x-patch
Size: 1930 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210611/e2e54837/attachment.bin>