[PATCH] D97061: [llvm-cov] Compare path only to find the same file

[Choongwoo Han via Phabricator via llvm-commits]

tunz created this revision.
tunz added reviewers: vsk, phosek, zequanwu.
tunz requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Currently, getSourceFile accesses file system to
check if two paths are the same file with a thread
lock, which is a huge performance bottleneck in some
cases.

Thus, if -ignore-symlinks option is given, let it
check only path strings after path normalization
so that we can avoid file system accesses.

When I tested it with two binaries and 16 cpu cores,
it saved over 70% of time.

Binary 1: 56 secs -> 3 secs
Binary 2: 17 hours -> 4 hours


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D97061

Files:
  llvm/tools/llvm-cov/CodeCoverage.cpp


Index: llvm/tools/llvm-cov/CodeCoverage.cpp
===================================================================
--- llvm/tools/llvm-cov/CodeCoverage.cpp
+++ llvm/tools/llvm-cov/CodeCoverage.cpp
@@ -135,6 +135,7 @@
   CoverageViewOptions ViewOpts;
   CoverageFiltersMatchAll Filters;
   CoverageFilters IgnoreFilenameFilters;
+  bool ComparePathStringOnly = false;
 
   /// True if InputSourceFiles are provided.
   bool HadSourceFiles = false;
@@ -248,9 +249,22 @@
     if (Loc != RemappedFilenames.end())
       SourceFile = Loc->second;
   }
-  for (const auto &Files : LoadedSourceFiles)
-    if (sys::fs::equivalent(SourceFile, Files.first))
-      return *Files.second;
+
+  SmallString<256> NormalizedPath;
+  sys::path::native(SourceFile, NormalizedPath);
+  sys::path::remove_dots(NormalizedPath, true);
+  SourceFile = NormalizedPath;
+
+  if (ComparePathStringOnly) {
+    for (const auto &Files : LoadedSourceFiles)
+      if (SourceFile == Files.first)
+        return *Files.second;
+  } else {
+    for (const auto &Files : LoadedSourceFiles)
+      if (sys::fs::equivalent(SourceFile, Files.first))
+        return *Files.second;
+  }
+
   auto Buffer = MemoryBuffer::getFile(SourceFile);
   if (auto EC = Buffer.getError()) {
     error(EC.message(), SourceFile);
@@ -654,6 +668,12 @@
                "regular expression"),
       cl::ZeroOrMore, cl::cat(FilteringCategory));
 
+  cl::opt<bool> IgnoreSymlinks(
+      "ignore-symlinks", cl::Optional,
+      cl::desc("Do not follow symlinks to check file equivalence. "
+               "Use only normalized path strings to find the same file."),
+      cl::init(false));
+
   cl::opt<double> RegionCoverageLtFilter(
       "region-coverage-lt", cl::Optional,
       cl::desc("Show code coverage only for functions with region coverage "
@@ -838,6 +858,7 @@
       ::exit(0);
     }
 
+    ComparePathStringOnly = IgnoreSymlinks;
     ViewOpts.ShowBranchSummary = BranchSummary;
     ViewOpts.ShowRegionSummary = RegionSummary;
     ViewOpts.ShowInstantiationSummary = InstantiationSummary;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D97061.325009.patch
Type: text/x-patch
Size: 2056 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210219/5d094bcf/attachment.bin>