[clang][FMV][AArch64] Diagnose/ignore unreachable functions versions. #171496
+120
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The commit llvm#150267 allows the user to override version priority. As a result it is now possible to define an unreachable function version if a higher priority version contains a subset of its FMV features. For example: target_clones("sve;priority=2", "sve2;priority=1") the sve2 version is unreachable, since if you don't have sve we can't have sve2 either. The patch emits a warning about such cases and ignores those versions when generating the resolver. Also removes their definitions.
llvmbot
added
clang
Member
|
@llvm/pr-subscribers-clang-codegen @llvm/pr-subscribers-clang Author: Alexandros Lamprineas (labrinea) ChangesThe commit #150267 allows the user to override version priority. As a result it is now possible to define an unreachable function version if a higher priority version contains a subset of its FMV features. For example: target_clones("sve;priority=2", "sve2;priority=1") the sve2 version is unreachable, since if you don't have sve we can't have sve2 either. The patch emits a warning about such cases and ignores those versions when generating the resolver. Also removes their definitions. Full diff: https://github.com/llvm/llvm-project/pull/171496.diff 4 Files Affected:
diff --git a/clang/include/clang/Basic/DiagnosticFrontendKinds.td b/clang/include/clang/Basic/DiagnosticFrontendKinds.td
index 9e344160ff934..a2d4564e23285 100644
--- a/clang/include/clang/Basic/DiagnosticFrontendKinds.td
+++ b/clang/include/clang/Basic/DiagnosticFrontendKinds.td
@@ -399,6 +399,10 @@ def err_invalid_llvm_ir : Error<"invalid LLVM IR input: %0">;
def err_os_unsupport_riscv_fmv : Error<
"function multiversioning is currently only supported on Linux">;
+def warn_unreachable_version
+ : Warning<"function version '%0' is unreachable; ignoring version">,
+ InGroup<FunctionMultiVersioning>;
+
def warn_hlsl_langstd_minimal :
Warning<"support for HLSL language version %0 is incomplete, "
"recommend using %1 instead">,
diff --git a/clang/lib/CodeGen/CodeGenFunction.cpp b/clang/lib/CodeGen/CodeGenFunction.cpp
index ac25bd95f0463..2bf2fd68b860e 100644
--- a/clang/lib/CodeGen/CodeGenFunction.cpp
+++ b/clang/lib/CodeGen/CodeGenFunction.cpp
@@ -3143,6 +3143,10 @@ void CodeGenFunction::EmitAArch64MultiVersionResolver(
Builder.SetInsertPoint(CurBlock);
}
+ // Skip unreachable versions.
+ if (RO.Function == nullptr)
+ continue;
+
llvm::BasicBlock *RetBlock = createBasicBlock("resolver_return", Resolver);
CGBuilderTy RetBuilder(*this, RetBlock);
CreateMultiVersionResolverReturn(CGM, Resolver, RetBuilder, RO.Function,
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 1dcf94fc35e07..9864c4cc03274 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -68,6 +68,7 @@
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/Hash.h"
#include "llvm/Support/TimeProfiler.h"
+#include "llvm/TargetParser/AArch64TargetParser.h"
#include "llvm/TargetParser/RISCVISAInfo.h"
#include "llvm/TargetParser/Triple.h"
#include "llvm/TargetParser/X86TargetParser.h"
@@ -4674,6 +4675,7 @@ void CodeGenModule::emitMultiVersionFunctions() {
// in this TU. For other architectures it is always emitted.
bool ShouldEmitResolver = !getTarget().getTriple().isAArch64();
SmallVector<CodeGenFunction::FMVResolverOption, 10> Options;
+ llvm::DenseMap<llvm::Function *, const FunctionDecl *> DeclMap;
getContext().forEachMultiversionedFunctionVersion(
FD, [&](const FunctionDecl *CurFD) {
@@ -4684,11 +4686,13 @@ void CodeGenModule::emitMultiVersionFunctions() {
assert(getTarget().getTriple().isX86() && "Unsupported target");
TA->getX86AddedFeatures(Feats);
llvm::Function *Func = createFunction(CurFD);
+ DeclMap.insert({Func, CurFD});
Options.emplace_back(Func, Feats, TA->getX86Architecture());
} else if (const auto *TVA = CurFD->getAttr<TargetVersionAttr>()) {
if (TVA->isDefaultVersion() && IsDefined)
ShouldEmitResolver = true;
llvm::Function *Func = createFunction(CurFD);
+ DeclMap.insert({Func, CurFD});
char Delim = getTarget().getTriple().isAArch64() ? '+' : ',';
TVA->getFeatures(Feats, Delim);
Options.emplace_back(Func, Feats);
@@ -4699,6 +4703,7 @@ void CodeGenModule::emitMultiVersionFunctions() {
if (TC->isDefaultVersion(I) && IsDefined)
ShouldEmitResolver = true;
llvm::Function *Func = createFunction(CurFD, I);
+ DeclMap.insert({Func, CurFD});
Feats.clear();
if (getTarget().getTriple().isX86()) {
TC->getX86Feature(Feats, I);
@@ -4744,6 +4749,24 @@ void CodeGenModule::emitMultiVersionFunctions() {
const CodeGenFunction::FMVResolverOption &RHS) {
return getFMVPriority(TI, LHS).ugt(getFMVPriority(TI, RHS));
});
+
+ // Diagnose unreachable function versions.
+ if (getTarget().getTriple().isAArch64()) {
+ for (auto I = Options.begin() + 1, E = Options.end(); I != E; ++I) {
+ llvm::APInt RHS = llvm::AArch64::getCpuSupportsMask(I->Features);
+ if (std::any_of(Options.begin(), I, [RHS](auto RO) {
+ llvm::APInt LHS = llvm::AArch64::getCpuSupportsMask(RO.Features);
+ return LHS.isSubsetOf(RHS);
+ })) {
+ Diags.Report(DeclMap[I->Function]->getLocation(),
+ diag::warn_unreachable_version)
+ << I->Function->getName();
+ assert(I->Function->user_empty() && "unexpected users");
+ I->Function->eraseFromParent();
+ I->Function = nullptr;
+ }
+ }
+ }
CodeGenFunction CGF(*this);
CGF.EmitMultiVersionResolver(ResolverFunc, Options);
diff --git a/clang/test/CodeGen/AArch64/fmv-unreachable-version.c b/clang/test/CodeGen/AArch64/fmv-unreachable-version.c
new file mode 100644
index 0000000000000..c9626c5ef34f3
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/fmv-unreachable-version.c
@@ -0,0 +1,89 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --check-attributes --check-globals --include-generated-funcs
+// RUN: %clang_cc1 -triple aarch64-linux-gnu -verify -emit-llvm -o - %s | FileCheck %s
+
+__attribute__((target_version("sve;priority=5"))) int unreachable_versions(void) { return 5; }
+// expected-warning@+1 {{function version 'unreachable_versions._MlseMmops' is unreachable; ignoring version}}
+__attribute__((target_version("mops+lse;priority=1"))) int unreachable_versions(void) { return 1; }
+int foo() { return unreachable_versions(); }
+// expected-warning@+1 {{function version 'unreachable_versions._Msve2' is unreachable; ignoring version}}
+__attribute__((target_clones("sve2;priority=4", "aes+sve2;priority=3", "lse;priority=2", "default"))) int unreachable_versions(void) { return 0; }
+// expected-warning@-1 {{function version 'unreachable_versions._MaesMsve2' is unreachable; ignoring version}}
+
+//.
+// CHECK: @__aarch64_cpu_features = external dso_local global { i64 }
+// CHECK: @unreachable_versions = weak_odr ifunc i32 (), ptr @unreachable_versions.resolver
+//.
+// CHECK: Function Attrs: noinline nounwind optnone vscale_range(1,16)
+// CHECK-LABEL: define {{[^@]+}}@unreachable_versions._Msve
+// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: ret i32 5
+//
+//
+// CHECK: Function Attrs: noinline nounwind optnone
+// CHECK-LABEL: define {{[^@]+}}@foo
+// CHECK-SAME: () #[[ATTR1:[0-9]+]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[CALL:%.*]] = call i32 @unreachable_versions()
+// CHECK-NEXT: ret i32 [[CALL]]
+//
+//
+// CHECK: Function Attrs: noinline nounwind optnone vscale_range(1,16)
+// CHECK-LABEL: define {{[^@]+}}@unreachable_versions._Mlse
+// CHECK-SAME: () #[[ATTR2:[0-9]+]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: ret i32 0
+//
+//
+// CHECK: Function Attrs: noinline nounwind optnone vscale_range(1,16)
+// CHECK-LABEL: define {{[^@]+}}@unreachable_versions.default
+// CHECK-SAME: () #[[ATTR3:[0-9]+]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: ret i32 0
+//
+//
+// CHECK: Function Attrs: disable_sanitizer_instrumentation
+// CHECK-LABEL: define {{[^@]+}}@unreachable_versions.resolver
+// CHECK-SAME: () #[[ATTR4:[0-9]+]] comdat {
+// CHECK-NEXT: resolver_entry:
+// CHECK-NEXT: call void @__init_cpu_features_resolver()
+// CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT: [[TMP1:%.*]] = and i64 [[TMP0]], 1073807616
+// CHECK-NEXT: [[TMP2:%.*]] = icmp eq i64 [[TMP1]], 1073807616
+// CHECK-NEXT: [[TMP3:%.*]] = and i1 true, [[TMP2]]
+// CHECK-NEXT: br i1 [[TMP3]], label [[RESOLVER_RETURN:%.*]], label [[RESOLVER_ELSE:%.*]]
+// CHECK: resolver_return:
+// CHECK-NEXT: ret ptr @unreachable_versions._Msve
+// CHECK: resolver_else:
+// CHECK-NEXT: [[TMP4:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT: [[TMP5:%.*]] = and i64 [[TMP4]], 69793284352
+// CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[TMP5]], 69793284352
+// CHECK-NEXT: [[TMP7:%.*]] = and i1 true, [[TMP6]]
+// CHECK-NEXT: [[TMP8:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT: [[TMP9:%.*]] = and i64 [[TMP8]], 69793317632
+// CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[TMP9]], 69793317632
+// CHECK-NEXT: [[TMP11:%.*]] = and i1 true, [[TMP10]]
+// CHECK-NEXT: [[TMP12:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT: [[TMP13:%.*]] = and i64 [[TMP12]], 128
+// CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[TMP13]], 128
+// CHECK-NEXT: [[TMP15:%.*]] = and i1 true, [[TMP14]]
+// CHECK-NEXT: br i1 [[TMP15]], label [[RESOLVER_RETURN1:%.*]], label [[RESOLVER_ELSE2:%.*]]
+// CHECK: resolver_return1:
+// CHECK-NEXT: ret ptr @unreachable_versions._Mlse
+// CHECK: resolver_else2:
+// CHECK-NEXT: [[TMP16:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT: [[TMP17:%.*]] = and i64 [[TMP16]], 576460752303423616
+// CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[TMP17]], 576460752303423616
+// CHECK-NEXT: [[TMP19:%.*]] = and i1 true, [[TMP18]]
+// CHECK-NEXT: ret ptr @unreachable_versions.default
+//
+//.
+// CHECK: attributes #[[ATTR0]] = { noinline nounwind optnone vscale_range(1,16) "fmv-features"="P0,P2,sve" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+sve" }
+// CHECK: attributes #[[ATTR1]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #[[ATTR2]] = { noinline nounwind optnone vscale_range(1,16) "fmv-features"="P1,lse" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse" }
+// CHECK: attributes #[[ATTR3]] = { noinline nounwind optnone vscale_range(1,16) "fmv-features" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #[[ATTR4]] = { disable_sanitizer_instrumentation }
+//.
+// CHECK: [[META0:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: [[META1:![0-9]+]] = !{!"{{.*}}clang version {{.*}}"}
+//.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The commit #150267 allows the user to override version priority. As a result it is now possible to define an unreachable function version if a higher priority version contains a subset of its FMV features. For example:
target_clones("sve;priority=2", "sve2;priority=1")
the sve2 version is unreachable, since if you don't have sve we can't have sve2 either.
The patch emits a warning about such cases and ignores those versions when generating the resolver. Also removes their definitions.