caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[llvm-cov][gcov] Support multi-files coverage in one basic block (#144504)

Browse Source 
In the current gcov implementation, all lines within a basic block are
attributed to the source file of the block's containing function. This
is inaccurate when a block contains lines from other files (e.g., via
#include "foo.inc").

Commit
[406e81b](406e81b79d)
attempted to address this by filtering lines based on debug info types,
but this approach has two limitations:

* **Over-filtering**: Some valid lines belonging to the function are
incorrectly excluded.
* **Under-counting**: Lines not belonging to the function are filtered
out and omitted from coverage statistics.

**GCC Reference Behavior**
GCC's gcov implementation handles this case correctly.This change aligns
the LLVM behavior with GCC.

**Proposed Solution**
1. **GCNO Generation**:

* **Current**: Each block stores a single GCOVLines record (filename +
lines).

* **New**: Dynamically create new GCOVLines records whenever consecutive
lines in a block originate from different source files. Group subsequent
lines from the same file under one record.

2. **GCNO Parsing**:

* **Current**: Lines are directly attributed to the function's source
file.

* **New**: Introduce a GCOVLocation type to track filename/line mappings
within blocks. Statistics will reflect the actual source file for each
line.
This commit is contained in:
int-zjt
2025-06-20 16:24:19 +08:00
committed by GitHub
parent bc14e5e5e9
commit b6b8fa3b15
5 changed files with 80 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
compiler-rt/test/profile/Posix/gcov-file-change-line.cpp Normal file
View File

@@ -0,0 +1,15 @@
// RUN: rm -rf %t && split-file %s %t && cd %t
// RUN: %clangxx --coverage main.cpp -o t
// RUN: %run ./t
// RUN: llvm-cov gcov -t t-main. | FileCheck %s
//--- main.cpp
#include <stdio.h>
int main(int argc, char *argv[]) { // CHECK: 2: [[#]]:int main
puts(""); // CHECK-NEXT: 2: [[#]]:
#line 3
puts(""); // line 3
return 0; // line 4
}
// CHECK-NOT: {{^ +[0-9]+:}}

12
compiler-rt/test/profile/Posix/gcov-file-change.cpp
View File

@@ -16,8 +16,8 @@ inline auto *const inl_var_main = // CHECK: 1: [[#]]:inline auto
void foo(int x) { // CHECK-NEXT: 1: [[#]]:
if (x) { // CHECK-NEXT: 1: [[#]]:
#include "a.inc"
}
}
} // CHECK: 1: [[#]]:
} // CHECK-NEXT: 1: [[#]]:
// CHECK-NOT: {{^ +[0-9]+:}}
int main(int argc, char *argv[]) { // CHECK: 1: [[#]]:int main
@@ -32,10 +32,8 @@ int main(int argc, char *argv[]) { // CHECK: 1: [[#]]:int main
//--- a.h
/// Apple targets doesn't enable -mconstructor-aliases by default and the count may be 4.
struct A { A() { } }; // CHECK: {{[24]}}: [[#]]:struct A
inline auto *const inl_var_a =
new A;
/// TODO a.inc:1 should have line execution.
// CHECK-NOT: {{^ +[0-9]+:}}
inline auto *const inl_var_a = // CHECK-NEXT: 1: [[#]]:
new A; // CHECK-NEXT: 1: [[#]]:
//--- a.inc
puts("");
puts(""); // CHECK: 1: [[#]]:puts

20
llvm/include/llvm/ProfileData/GCOV.h
View File

@@ -271,6 +271,14 @@ public:
DenseSet<const GCOVBlock *> visited;
};
/// Represent file of lines same with block_location_info in gcc.
struct GCOVBlockLocation {
GCOVBlockLocation(unsigned idx) : srcIdx(idx) {}
unsigned srcIdx;
SmallVector<uint32_t, 4> lines;
};
/// GCOVBlock - Collects block information.
class GCOVBlock {
public:
@@ -281,8 +289,13 @@ public:
GCOVBlock(uint32_t N) : number(N) {}
void addLine(uint32_t N) { lines.push_back(N); }
uint32_t getLastLine() const { return lines.back(); }
void addLine(uint32_t N) {
locations.back().lines.push_back(N);
lastLine = N;
}
void addFile(unsigned fileIdx) { locations.emplace_back(fileIdx); }
uint32_t getLastLine() const { return lastLine; }
uint64_t getCount() const { return count; }
void addSrcEdge(GCOVArc *Edge) { pred.push_back(Edge); }
@@ -311,7 +324,8 @@ public:
uint64_t count = 0;
SmallVector<GCOVArc *, 2> pred;
SmallVector<GCOVArc *, 2> succ;
SmallVector<uint32_t, 4> lines;
SmallVector<GCOVBlockLocation> locations;
uint32_t lastLine = 0;
bool traversable = false;
GCOVArc *incoming = nullptr;
};

45
llvm/lib/ProfileData/GCOV.cpp
View File

@@ -191,7 +191,7 @@ bool GCOVFile::readGCNO(GCOVBuffer &buf) {
buf.readString(filename);
if (filename.empty())
break;
// TODO Unhandled
Block.addFile(addNormalizedPathToMap(filename));
}
}
}
@@ -456,11 +456,13 @@ void GCOVBlock::print(raw_ostream &OS) const {
}
OS << "\n";
}
if (!lines.empty()) {
OS << "\tLines : ";
for (uint32_t N : lines)
OS << (N) << ",";
OS << "\n";
if (!locations.empty()) {
for (const GCOVBlockLocation &loc : locations) {
OS << "\tFile: " << loc.srcIdx << ": ";
for (uint32_t N : loc.lines)
OS << (N) << ",";
OS << "\n";
}
}
}
@@ -701,20 +703,25 @@ void Context::collectFunction(GCOVFunction &f, Summary &summary) {
SmallSet<uint32_t, 16> lines;
SmallSet<uint32_t, 16> linesExec;
for (const GCOVBlock &b : f.blocksRange()) {
if (b.lines.empty())
if (b.locations.empty())
continue;
uint32_t maxLineNum = *llvm::max_element(b.lines);
if (maxLineNum >= si.lines.size())
si.lines.resize(maxLineNum + 1);
for (uint32_t lineNum : b.lines) {
LineInfo &line = si.lines[lineNum];
if (lines.insert(lineNum).second)
++summary.lines;
if (b.count && linesExec.insert(lineNum).second)
++summary.linesExec;
line.exists = true;
line.count += b.count;
line.blocks.push_back(&b);
for (const GCOVBlockLocation &loc : b.locations) {
SourceInfo &locSource = sources[loc.srcIdx];
uint32_t maxLineNum = *llvm::max_element(loc.lines);
if (maxLineNum >= locSource.lines.size())
locSource.lines.resize(maxLineNum + 1);
for (uint32_t lineNum : loc.lines) {
LineInfo &line = locSource.lines[lineNum];
line.exists = true;
line.count += b.count;
line.blocks.push_back(&b);
if (f.srcIdx == loc.srcIdx) {
if (lines.insert(lineNum).second)
++summary.lines;
if (b.count && linesExec.insert(lineNum).second)
++summary.linesExec;
}
}
}
}
}

37
llvm/lib/Transforms/Instrumentation/GCOVProfiling.cpp
View File

@@ -210,12 +210,12 @@ static StringRef getFunctionName(const DISubprogram *SP) {
return SP->getName();
}
/// Extract a filename for a DISubprogram.
/// Extract a filename for a DIScope.
///
/// Prefer relative paths in the coverage notes. Clang also may split
/// up absolute paths into a directory and filename component. When
/// the relative path doesn't exist, reconstruct the absolute path.
static SmallString<128> getFilename(const DISubprogram *SP) {
static SmallString<128> getFilename(const DIScope *SP) {
SmallString<128> Path;
StringRef RelPath = SP->getFilename();
if (sys::fs::exists(RelPath))
@@ -244,7 +244,9 @@ namespace {
// list of line numbers and a single filename, representing lines that belong
// to the block.
class GCOVLines : public GCOVRecord {
public:
public:
const StringRef getFilename() { return Filename; }
void addLine(uint32_t Line) {
assert(Line != 0 && "Line zero is not a valid real line number.");
Lines.push_back(Line);
@@ -276,7 +278,9 @@ namespace {
class GCOVBlock : public GCOVRecord {
public:
GCOVLines &getFile(StringRef Filename) {
return LinesByFile.try_emplace(Filename, P, Filename).first->second;
if (Lines.empty() || Lines.back().getFilename() != Filename)
Lines.emplace_back(P, Filename);
return Lines.back();
}
void addEdge(GCOVBlock &Successor, uint32_t Flags) {
@@ -285,22 +289,16 @@ namespace {
void writeOut() {
uint32_t Len = 3;
SmallVector<StringMapEntry<GCOVLines> *, 32> SortedLinesByFile;
for (auto &I : LinesByFile) {
Len += I.second.length();
SortedLinesByFile.push_back(&I);
}
for (auto &L : Lines)
Len += L.length();
write(GCOV_TAG_LINES);
write(Len);
write(Number);
llvm::sort(SortedLinesByFile, [](StringMapEntry<GCOVLines> *LHS,
StringMapEntry<GCOVLines> *RHS) {
return LHS->getKey() < RHS->getKey();
});
for (auto &I : SortedLinesByFile)
I->getValue().writeOut();
for (auto &L : Lines)
L.writeOut();
write(0);
write(0);
}
@@ -309,7 +307,7 @@ namespace {
// Only allow copy before edges and lines have been added. After that,
// there are inter-block pointers (eg: edges) that won't take kindly to
// blocks being copied or moved around.
assert(LinesByFile.empty());
assert(Lines.empty());
assert(OutEdges.empty());
}
@@ -322,7 +320,7 @@ namespace {
GCOVBlock(GCOVProfiler *P, uint32_t Number)
: GCOVRecord(P), Number(Number) {}
StringMap<GCOVLines> LinesByFile;
SmallVector<GCOVLines> Lines;
};
// A function has a unique identifier, a checksum (we leave as zero) and a
@@ -881,11 +879,10 @@ bool GCOVProfiler::emitProfileNotes(
if (Line == Loc.getLine()) continue;
Line = Loc.getLine();
MDNode *Scope = Loc.getScope();
// TODO: Handle blocks from another file due to #line, #include, etc.
if (isa<DILexicalBlockFile>(Scope) || SP != getDISubprogram(Scope))
if (SP != getDISubprogram(Scope))
continue;
GCOVLines &Lines = Block.getFile(Filename);
GCOVLines &Lines = Block.getFile(getFilename(Loc->getScope()));
Lines.addLine(Loc.getLine());
}
Line = 0;