[GCov] Emit memset instead of stores in __llvm_gcov_reset

[aeubanks]

Summary

For a very large module, llvm_gcov_reset can become very large.
llvm_gcov_reset previously emitted stores to a bunch of globals in one
huge basic block. MemCpyOpt would turn many of these stores into
memsets, and updating MemorySSA would be extremely slow.

Verified that this makes the compile time of certain files go down
drastically (20min -> 5min).

[MaskRay]

The test does not belong to module-flags.ll :)

D78477 did not add an IR test. Perhaps you need to add a new test file. Add two functions so that memset will be emitted more than once.

[aeubanks]

create new test

[rnk]

There are pre-commit test suite failures related to gcov, which seem related.

[MaskRay]

(When making a gcov change, ninja check-llvm-transforms-gcovprofiling check-llvm-tools-llvm-cov check-clang-codegen check-profile. These directories have tests.)

[aeubanks]

fix size calculation
(not sure if there's a better way to do this or not)

[MaskRay]

Looks great!

llvm/test/Transforms/GCOVProfiling/reset.ll

35

It is more common to place a synthesized function before metadata nodes.

[aeubanks]

move test function

[asbirlea]

Thanks for this fix! Timings look great!

llvm/lib/Transforms/Instrumentation/GCOVProfiling.cpp

Function *GCOVProfiler::insertReset(
   FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), false);
   Function *ResetF = M->getFunction("__llvm_gcov_reset");
   if (!ResetF)
     ResetF = createInternalFunction(FTy, "__llvm_gcov_reset");
   ResetF->addFnAttr(Attribute::NoInline);
 
   BasicBlock *Entry = BasicBlock::Create(*Ctx, "entry", ResetF);
   IRBuilder<> Builder(Entry);
+  LLVMContext &C = Entry->getContext();
 
   // Zero out the counters.
   for (const auto &I : CountersBySP) {
     GlobalVariable *GV = I.first;
-    Constant *Null = Constant::getNullValue(GV->getValueType());
-    Builder.CreateStore(Null, GV);
+    auto *GVTy = cast<ArrayType>(GV->getValueType());
+    Builder.CreateMemSet(GV, Constant::getNullValue(Type::getInt8Ty(C)),
+                         GVTy->getNumElements() *
+                             GVTy->getElementType()->getScalarSizeInBits() / 8,
+                         GV->getAlign());
   }
 
   Type *RetTy = ResetF->getReturnType();
   if (RetTy->isVoidTy())
     Builder.CreateRetVoid();
   else if (RetTy->isIntegerTy())
     // Used if __llvm_gcov_reset was implicitly declared.
     Builder.CreateRet(ConstantInt::get(RetTy, 0));
   else
     report_fatal_error("invalid return type for __llvm_gcov_reset");
 
   return ResetF;
 }

llvm/test/Transforms/GCOVProfiling/reset.ll

+; RUN: mkdir -p %t && cd %t
+; RUN: opt < %s -S -passes=insert-gcov-profiling | FileCheck %s
+
+target triple = "x86_64-unknown-linux-gnu"
+
+define void @foo() !dbg !8 {
+entry:
+  ret void, !dbg !12
+}
+
+define void @bar() !dbg !13 {
+entry:
+  ret void, !dbg !14
+}
+
+; CHECK: define internal void @__llvm_gcov_reset()
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* bitcast ([1 x i64]* @__llvm_gcov_ctr to i8*), i8 0, i64 8, i1 false)
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* bitcast ([1 x i64]* @__llvm_gcov_ctr.1 to i8*), i8 0, i64 8, i1 false)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!3, !4, !5, !6}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 14.0.0", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "/tmp/a.c", directory: "/tmp")
+!2 = !{}
+!3 = !{i32 7, !"Dwarf Version", i32 4}
+!4 = !{i32 2, !"Debug Info Version", i32 3}
+!5 = !{i32 1, !"wchar_size", i32 4}
+!6 = !{i32 7, !"uwtable", i32 1}
+!8 = distinct !DISubprogram(name: "foo", scope: !9, file: !9, line: 1, type: !10, scopeLine: 1, flags: DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+!9 = !DIFile(filename: "/tmp/a.c", directory: "")
+!10 = !DISubroutineType(types: !11)
+!11 = !{null}
+!12 = !DILocation(line: 1, column: 13, scope: !8)
+!13 = distinct !DISubprogram(name: "bar", scope: !9, file: !9, line: 2, type: !10, scopeLine: 2, flags: DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+!14 = !DILocation(line: 2, column: 13, scope: !13)