[clang] [llvm] Reland "[AArch64] Decouple feature dependency expansion. (#94279)" (PR #95519)
Alexandros Lamprineas via llvm-commits
llvm-commits at lists.llvm.org
Tue Jun 18 07:28:14 PDT 2024
https://github.com/labrinea updated https://github.com/llvm/llvm-project/pull/95519
>From 66b4dcd6d074c700346b405c48c7ff608122ae8d Mon Sep 17 00:00:00 2001
From: Alexandros Lamprineas <alexandros.lamprineas at arm.com>
Date: Fri, 14 Jun 2024 08:46:12 +0100
Subject: [PATCH 1/5] Reland "[AArch64] Decouple feature dependency expansion.
(#94279)"
This is the second attempt. We should be inserting the Driver
features in front of the features of a parsed target attribute
to avoid errors like the following:
```
% cat neon.c
__attribute__((target("arch=armv8-a")))
uint64x2_t foo(uint64x2_t a, uint64x2_t b) { return veorq_u64(a, b); }
% clang --target=aarch64-linux-gnu -c neon.c
error: always_inline function 'veorq_u64' requires target feature
'outline-atomics', but would be inlined into function 'foo'
that is compiled without support for 'outline-atomics'
```
---
clang/include/clang/AST/ASTContext.h | 3 -
clang/lib/AST/ASTContext.cpp | 59 +++++-----
clang/lib/AST/CMakeLists.txt | 2 +
clang/lib/Basic/Targets/AArch64.cpp | 105 ++++++-----------
clang/lib/Basic/Targets/AArch64.h | 4 -
.../CodeGen/aarch64-cpu-supports-target.c | 4 +-
.../aarch64-sme-attrs.cpp | 2 +-
clang/test/CodeGen/aarch64-targetattr.c | 48 ++++----
clang/test/CodeGen/attr-target-version.c | 46 ++++----
clang/test/Sema/aarch64-neon-target.c | 4 +-
.../llvm/TargetParser/AArch64TargetParser.h | 107 +++++++++++-------
llvm/lib/TargetParser/AArch64TargetParser.cpp | 51 ++++++---
12 files changed, 213 insertions(+), 222 deletions(-)
diff --git a/clang/include/clang/AST/ASTContext.h b/clang/include/clang/AST/ASTContext.h
index 53ece996769a8..f1f20fca477a4 100644
--- a/clang/include/clang/AST/ASTContext.h
+++ b/clang/include/clang/AST/ASTContext.h
@@ -3210,9 +3210,6 @@ class ASTContext : public RefCountedBase<ASTContext> {
/// valid feature names.
ParsedTargetAttr filterFunctionTargetAttrs(const TargetAttr *TD) const;
- std::vector<std::string>
- filterFunctionTargetVersionAttrs(const TargetVersionAttr *TV) const;
-
void getFunctionFeatureMap(llvm::StringMap<bool> &FeatureMap,
const FunctionDecl *) const;
void getFunctionFeatureMap(llvm::StringMap<bool> &FeatureMap,
diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
index 34aa399fda2f8..aa22825602a40 100644
--- a/clang/lib/AST/ASTContext.cpp
+++ b/clang/lib/AST/ASTContext.cpp
@@ -87,6 +87,7 @@
#include "llvm/Support/MD5.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
+#include "llvm/TargetParser/AArch64TargetParser.h"
#include "llvm/TargetParser/Triple.h"
#include <algorithm>
#include <cassert>
@@ -13676,17 +13677,20 @@ QualType ASTContext::getCorrespondingSignedFixedPointType(QualType Ty) const {
}
}
-std::vector<std::string> ASTContext::filterFunctionTargetVersionAttrs(
- const TargetVersionAttr *TV) const {
- assert(TV != nullptr);
- llvm::SmallVector<StringRef, 8> Feats;
- std::vector<std::string> ResFeats;
- TV->getFeatures(Feats);
- for (auto &Feature : Feats)
- if (Target->validateCpuSupports(Feature.str()))
- // Use '?' to mark features that came from TargetVersion.
- ResFeats.push_back("?" + Feature.str());
- return ResFeats;
+// Given a list of FMV features, return a concatenated list of the
+// corresponding backend features (which may contain duplicates).
+static std::vector<std::string> getFMVBackendFeaturesFor(
+ const llvm::SmallVectorImpl<StringRef> &FMVFeatStrings) {
+ std::vector<std::string> BackendFeats;
+ for (StringRef F : FMVFeatStrings) {
+ if (auto FMVExt = llvm::AArch64::parseArchExtension(F)) {
+ SmallVector<StringRef, 8> Feats;
+ FMVExt->DependentFeatures.split(Feats, ',', -1, false);
+ for (StringRef F : Feats)
+ BackendFeats.push_back(F.str());
+ }
+ }
+ return BackendFeats;
}
ParsedTargetAttr
@@ -13721,10 +13725,12 @@ void ASTContext::getFunctionFeatureMap(llvm::StringMap<bool> &FeatureMap,
// Make a copy of the features as passed on the command line into the
// beginning of the additional features from the function to override.
- ParsedAttr.Features.insert(
- ParsedAttr.Features.begin(),
- Target->getTargetOpts().FeaturesAsWritten.begin(),
- Target->getTargetOpts().FeaturesAsWritten.end());
+ // AArch64 handles command line option features in parseTargetAttr().
+ if (!Target->getTriple().isAArch64())
+ ParsedAttr.Features.insert(
+ ParsedAttr.Features.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.end());
if (ParsedAttr.CPU != "" && Target->isValidCPUName(ParsedAttr.CPU))
TargetCPU = ParsedAttr.CPU;
@@ -13745,32 +13751,31 @@ void ASTContext::getFunctionFeatureMap(llvm::StringMap<bool> &FeatureMap,
Target->getTargetOpts().FeaturesAsWritten.end());
Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Features);
} else if (const auto *TC = FD->getAttr<TargetClonesAttr>()) {
- std::vector<std::string> Features;
if (Target->getTriple().isAArch64()) {
- // TargetClones for AArch64
llvm::SmallVector<StringRef, 8> Feats;
TC->getFeatures(Feats, GD.getMultiVersionIndex());
- for (StringRef Feat : Feats)
- if (Target->validateCpuSupports(Feat.str()))
- // Use '?' to mark features that came from AArch64 TargetClones.
- Features.push_back("?" + Feat.str());
+ std::vector<std::string> Features = getFMVBackendFeaturesFor(Feats);
Features.insert(Features.begin(),
Target->getTargetOpts().FeaturesAsWritten.begin(),
Target->getTargetOpts().FeaturesAsWritten.end());
+ Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Features);
} else {
+ std::vector<std::string> Features;
StringRef VersionStr = TC->getFeatureStr(GD.getMultiVersionIndex());
if (VersionStr.starts_with("arch="))
TargetCPU = VersionStr.drop_front(sizeof("arch=") - 1);
else if (VersionStr != "default")
Features.push_back((StringRef{"+"} + VersionStr).str());
+ Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Features);
}
- Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Features);
} else if (const auto *TV = FD->getAttr<TargetVersionAttr>()) {
- std::vector<std::string> Feats = filterFunctionTargetVersionAttrs(TV);
- Feats.insert(Feats.begin(),
- Target->getTargetOpts().FeaturesAsWritten.begin(),
- Target->getTargetOpts().FeaturesAsWritten.end());
- Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Feats);
+ llvm::SmallVector<StringRef, 8> Feats;
+ TV->getFeatures(Feats);
+ std::vector<std::string> Features = getFMVBackendFeaturesFor(Feats);
+ Features.insert(Features.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.end());
+ Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Features);
} else {
FeatureMap = Target->getTargetOpts().FeatureMap;
}
diff --git a/clang/lib/AST/CMakeLists.txt b/clang/lib/AST/CMakeLists.txt
index a5d3dacfc1a84..0328666d59b1f 100644
--- a/clang/lib/AST/CMakeLists.txt
+++ b/clang/lib/AST/CMakeLists.txt
@@ -139,4 +139,6 @@ add_clang_library(clangAST
omp_gen
ClangDriverOptions
intrinsics_gen
+ # These generated headers are included transitively.
+ AArch64TargetParserTableGen
)
diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp
index 08d13c41a4857..6fba5fff7bcc1 100644
--- a/clang/lib/Basic/Targets/AArch64.cpp
+++ b/clang/lib/Basic/Targets/AArch64.cpp
@@ -1052,57 +1052,18 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
return true;
}
-bool AArch64TargetInfo::initFeatureMap(
- llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags, StringRef CPU,
- const std::vector<std::string> &FeaturesVec) const {
- std::vector<std::string> UpdatedFeaturesVec;
- // Parse the CPU and add any implied features.
- std::optional<llvm::AArch64::CpuInfo> CpuInfo = llvm::AArch64::parseCpu(CPU);
- if (CpuInfo) {
- auto Exts = CpuInfo->getImpliedExtensions();
- std::vector<StringRef> CPUFeats;
- llvm::AArch64::getExtensionFeatures(Exts, CPUFeats);
- for (auto F : CPUFeats) {
- assert((F[0] == '+' || F[0] == '-') && "Expected +/- in target feature!");
- UpdatedFeaturesVec.push_back(F.str());
- }
- }
-
- // Process target and dependent features. This is done in two loops collecting
- // them into UpdatedFeaturesVec: first to add dependent '+'features, second to
- // add target '+/-'features that can later disable some of features added on
- // the first loop. Function Multi Versioning features begin with '?'.
- for (const auto &Feature : FeaturesVec)
- if (((Feature[0] == '?' || Feature[0] == '+')) &&
- AArch64TargetInfo::doesFeatureAffectCodeGen(Feature.substr(1))) {
- StringRef DepFeatures =
- AArch64TargetInfo::getFeatureDependencies(Feature.substr(1));
- SmallVector<StringRef, 1> AttrFeatures;
- DepFeatures.split(AttrFeatures, ",");
- for (auto F : AttrFeatures)
- UpdatedFeaturesVec.push_back(F.str());
- }
- for (const auto &Feature : FeaturesVec)
- if (Feature[0] != '?') {
- std::string UpdatedFeature = Feature;
- if (Feature[0] == '+') {
- std::optional<llvm::AArch64::ExtensionInfo> Extension =
- llvm::AArch64::parseArchExtension(Feature.substr(1));
- if (Extension)
- UpdatedFeature = Extension->Feature.str();
- }
- UpdatedFeaturesVec.push_back(UpdatedFeature);
- }
-
- return TargetInfo::initFeatureMap(Features, Diags, CPU, UpdatedFeaturesVec);
-}
-
// Parse AArch64 Target attributes, which are a comma separated list of:
// "arch=<arch>" - parsed to features as per -march=..
// "cpu=<cpu>" - parsed to features as per -mcpu=.., with CPU set to <cpu>
// "tune=<cpu>" - TuneCPU set to <cpu>
// "feature", "no-feature" - Add (or remove) feature.
// "+feature", "+nofeature" - Add (or remove) feature.
+//
+// A feature may correspond to an Extension (anything with a corresponding
+// AEK_), in which case an ExtensionSet is used to parse it and expand its
+// dependencies. Otherwise the feature is passed through (e.g. +v8.1a,
+// +outline-atomics, -fmv, etc). Features coming from the command line are
+// already parsed, therefore their dependencies do not need expansion.
ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
ParsedTargetAttr Ret;
if (Features == "default")
@@ -1112,23 +1073,26 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
bool FoundArch = false;
auto SplitAndAddFeatures = [](StringRef FeatString,
- std::vector<std::string> &Features) {
+ std::vector<std::string> &Features,
+ llvm::AArch64::ExtensionSet &FeatureBits) {
SmallVector<StringRef, 8> SplitFeatures;
FeatString.split(SplitFeatures, StringRef("+"), -1, false);
for (StringRef Feature : SplitFeatures) {
- StringRef FeatureName = llvm::AArch64::getArchExtFeature(Feature);
- if (!FeatureName.empty())
- Features.push_back(FeatureName.str());
+ if (FeatureBits.parseModifier(Feature, /* AllowNoDashForm = */ true))
+ continue;
+ // Pass through features that are not extensions, e.g. +v8.1a,
+ // +outline-atomics, -fmv, etc.
+ if (Feature.starts_with("no"))
+ Features.push_back("-" + Feature.drop_front(2).str());
else
- // Pushing the original feature string to give a sema error later on
- // when they get checked.
- if (Feature.starts_with("no"))
- Features.push_back("-" + Feature.drop_front(2).str());
- else
- Features.push_back("+" + Feature.str());
+ Features.push_back("+" + Feature.str());
}
};
+ llvm::AArch64::ExtensionSet FeatureBits;
+ // Reconstruct the bitset from the command line option features.
+ FeatureBits.reconstructFromParsedFeatures(getTargetOpts().FeaturesAsWritten);
+
for (auto &Feature : AttrFeatures) {
Feature = Feature.trim();
if (Feature.starts_with("fpmath="))
@@ -1151,9 +1115,9 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
// Ret.Features.
if (!AI)
continue;
- Ret.Features.push_back(AI->ArchFeature.str());
+ FeatureBits.addArchDefaults(*AI);
// Add any extra features, after the +
- SplitAndAddFeatures(Split.second, Ret.Features);
+ SplitAndAddFeatures(Split.second, Ret.Features, FeatureBits);
} else if (Feature.starts_with("cpu=")) {
if (!Ret.CPU.empty())
Ret.Duplicate = "cpu=";
@@ -1163,7 +1127,10 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
std::pair<StringRef, StringRef> Split =
Feature.split("=").second.trim().split("+");
Ret.CPU = Split.first;
- SplitAndAddFeatures(Split.second, Ret.Features);
+ if (auto CpuInfo = llvm::AArch64::parseCpu(Ret.CPU)) {
+ FeatureBits.addCPUDefaults(*CpuInfo);
+ SplitAndAddFeatures(Split.second, Ret.Features, FeatureBits);
+ }
}
} else if (Feature.starts_with("tune=")) {
if (!Ret.Tune.empty())
@@ -1171,25 +1138,19 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
else
Ret.Tune = Feature.split("=").second.trim();
} else if (Feature.starts_with("+")) {
- SplitAndAddFeatures(Feature, Ret.Features);
- } else if (Feature.starts_with("no-")) {
- StringRef FeatureName =
- llvm::AArch64::getArchExtFeature(Feature.split("-").second);
- if (!FeatureName.empty())
- Ret.Features.push_back("-" + FeatureName.drop_front(1).str());
- else
- Ret.Features.push_back("-" + Feature.split("-").second.str());
+ SplitAndAddFeatures(Feature, Ret.Features, FeatureBits);
} else {
- // Try parsing the string to the internal target feature name. If it is
- // invalid, add the original string (which could already be an internal
- // name). These should be checked later by isValidFeatureName.
- StringRef FeatureName = llvm::AArch64::getArchExtFeature(Feature);
- if (!FeatureName.empty())
- Ret.Features.push_back(FeatureName.str());
+ if (FeatureBits.parseModifier(Feature, /* AllowNoDashForm = */ true))
+ continue;
+ // Pass through features that are not extensions, e.g. +v8.1a,
+ // +outline-atomics, -fmv, etc.
+ if (Feature.starts_with("no-"))
+ Ret.Features.push_back("-" + Feature.drop_front(3).str());
else
Ret.Features.push_back("+" + Feature.str());
}
}
+ FeatureBits.toLLVMFeatureList(Ret.Features);
return Ret;
}
diff --git a/clang/lib/Basic/Targets/AArch64.h b/clang/lib/Basic/Targets/AArch64.h
index 12fb50286f751..696553ef8038a 100644
--- a/clang/lib/Basic/Targets/AArch64.h
+++ b/clang/lib/Basic/Targets/AArch64.h
@@ -107,10 +107,6 @@ class LLVM_LIBRARY_VISIBILITY AArch64TargetInfo : public TargetInfo {
unsigned multiVersionSortPriority(StringRef Name) const override;
unsigned multiVersionFeatureCost() const override;
- bool
- initFeatureMap(llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags,
- StringRef CPU,
- const std::vector<std::string> &FeaturesVec) const override;
bool useFP16ConversionIntrinsics() const override {
return false;
}
diff --git a/clang/test/CodeGen/aarch64-cpu-supports-target.c b/clang/test/CodeGen/aarch64-cpu-supports-target.c
index e023944b24e53..28187bcf74533 100644
--- a/clang/test/CodeGen/aarch64-cpu-supports-target.c
+++ b/clang/test/CodeGen/aarch64-cpu-supports-target.c
@@ -48,5 +48,5 @@ int test_versions() {
return code();
}
// CHECK: attributes #0 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-// CHECK: attributes #1 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon" }
-// CHECK: attributes #2 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve" }
+// CHECK: attributes #1 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon" }
+// CHECK: attributes #2 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+sve" }
diff --git a/clang/test/CodeGen/aarch64-sme-intrinsics/aarch64-sme-attrs.cpp b/clang/test/CodeGen/aarch64-sme-intrinsics/aarch64-sme-attrs.cpp
index af8933d93d6cb..9885ac45e6a0e 100644
--- a/clang/test/CodeGen/aarch64-sme-intrinsics/aarch64-sme-attrs.cpp
+++ b/clang/test/CodeGen/aarch64-sme-intrinsics/aarch64-sme-attrs.cpp
@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme \
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +bf16 \
// RUN: -disable-O0-optnone -Werror -emit-llvm -o - %s \
// RUN: | opt -S -passes=mem2reg \
// RUN: | opt -S -passes=inline \
diff --git a/clang/test/CodeGen/aarch64-targetattr.c b/clang/test/CodeGen/aarch64-targetattr.c
index 3e7a209245607..644e6a692c3be 100644
--- a/clang/test/CodeGen/aarch64-targetattr.c
+++ b/clang/test/CodeGen/aarch64-targetattr.c
@@ -58,58 +58,50 @@ void v1msve() {}
// CHECK-LABEL: @plussve() #12
__attribute__((target("+sve")))
void plussve() {}
-// CHECK-LABEL: @plussveplussve2() #13
+// CHECK-LABEL: @plussveplussve2() #12
__attribute__((target("+sve+nosve2")))
void plussveplussve2() {}
-// CHECK-LABEL: @plussveminusnosve2() #13
+// CHECK-LABEL: @plussveminusnosve2() #12
__attribute__((target("sve,no-sve2")))
void plussveminusnosve2() {}
-// CHECK-LABEL: @plusfp16() #14
+// CHECK-LABEL: @plusfp16() #13
__attribute__((target("+fp16")))
void plusfp16() {}
-// CHECK-LABEL: @all() #15
+// CHECK-LABEL: @all() #14
__attribute__((target("cpu=neoverse-n1,tune=cortex-a710,arch=armv8.6-a+sve2")))
void all() {}
-// CHECK-LABEL: @allplusbranchprotection() #16
+// CHECK-LABEL: @allplusbranchprotection() #15
__attribute__((target("cpu=neoverse-n1,tune=cortex-a710,arch=armv8.6-a+sve2,branch-protection=standard")))
void allplusbranchprotection() {}
-// These tests check that the user facing and internal llvm name are both accepted.
-// CHECK-LABEL: @plusnoneon() #17
-__attribute__((target("+noneon")))
-void plusnoneon() {}
-// CHECK-LABEL: @plusnosimd() #17
+// CHECK-LABEL: @plusnosimd() #16
__attribute__((target("+nosimd")))
void plusnosimd() {}
-// CHECK-LABEL: @noneon() #17
-__attribute__((target("no-neon")))
-void noneon() {}
-// CHECK-LABEL: @nosimd() #17
+// CHECK-LABEL: @nosimd() #16
__attribute__((target("no-simd")))
void nosimd() {}
// This isn't part of the standard interface, but test that -arch features should not apply anything else.
-// CHECK-LABEL: @minusarch() #18
+// CHECK-LABEL: @minusarch() #17
__attribute__((target("no-v9.3a")))
void minusarch() {}
// CHECK: attributes #0 = { {{.*}} "target-features"="+crc,+fp-armv8,+lse,+neon,+ras,+rdm,+v8.1a,+v8.2a,+v8a" }
// CHECK: attributes #1 = { {{.*}} "target-features"="+crc,+fp-armv8,+fullfp16,+lse,+neon,+ras,+rdm,+sve,+v8.1a,+v8.2a,+v8a" }
// CHECK: attributes #2 = { {{.*}} "target-features"="+crc,+fp-armv8,+fullfp16,+lse,+neon,+ras,+rdm,+sve,+sve2,+v8.1a,+v8.2a,+v8a" }
-// CHECK: attributes #3 = { {{.*}} "target-features"="+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8.6a,+v8a" }
-// CHECK: attributes #4 = { {{.*}} "target-cpu"="cortex-a710" "target-features"="+bf16,+complxnum,+crc,+dotprod,+flagm,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+mte,+neon,+pauth,+ras,+rcpc,+rdm,+sb,+sve,+sve2,+sve2-bitperm" }
+// CHECK: attributes #3 = { {{.*}} "target-features"="+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8.6a,+v8a" }
+// CHECK: attributes #4 = { {{.*}} "target-cpu"="cortex-a710" "target-features"="+bf16,+complxnum,+crc,+dotprod,+flagm,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+mte,+neon,+pauth,+ras,+rcpc,+rdm,+sb,+sve,+sve2,+sve2-bitperm,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8a,+v9a" }
// CHECK: attributes #5 = { {{.*}} "tune-cpu"="cortex-a710" }
// CHECK: attributes #6 = { {{.*}} "target-cpu"="generic" }
// CHECK: attributes #7 = { {{.*}} "tune-cpu"="generic" }
-// CHECK: attributes #8 = { {{.*}} "target-cpu"="neoverse-n1" "target-features"="+aes,+crc,+dotprod,+fp-armv8,+fullfp16,+lse,+neon,+ras,+rcpc,+rdm,+sha2,+spe,+ssbs" "tune-cpu"="cortex-a710" }
-// CHECK: attributes #9 = { {{.*}} "target-features"="+fp-armv8,+fullfp16,+neon,+sve" "tune-cpu"="cortex-a710" }
-// CHECK: attributes #10 = { {{.*}} "target-cpu"="neoverse-v1" "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+rand,+ras,+rcpc,+rdm,+sha2,+sha3,+sm4,+spe,+ssbs,+sve,+sve2" }
-// CHECK: attributes #11 = { {{.*}} "target-cpu"="neoverse-v1" "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+rand,+ras,+rcpc,+rdm,+sha2,+sha3,+sm4,+spe,+ssbs,-sve" }
-// CHECK: attributes #12 = { {{.*}} "target-features"="+fp-armv8,+fullfp16,+neon,+sve" }
-// CHECK: attributes #13 = { {{.*}} "target-features"="+fp-armv8,+fullfp16,+neon,+sve,-sve2" }
-// CHECK: attributes #14 = { {{.*}} "target-features"="+fullfp16" }
-// CHECK: attributes #15 = { {{.*}} "target-cpu"="neoverse-n1" "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sha2,+spe,+ssbs,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8.6a,+v8a" "tune-cpu"="cortex-a710" }
-// CHECK: attributes #16 = { {{.*}} "branch-target-enforcement"="true" "guarded-control-stack"="true" {{.*}} "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sha2,+spe,+ssbs,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8.6a,+v8a" "tune-cpu"="cortex-a710" }
-// CHECK: attributes #17 = { {{.*}} "target-features"="-neon" }
-// CHECK: attributes #18 = { {{.*}} "target-features"="-v9.3a" }
+// CHECK: attributes #8 = { {{.*}} "target-cpu"="neoverse-n1" "target-features"="+aes,+crc,+dotprod,+fp-armv8,+fullfp16,+lse,+neon,+ras,+rcpc,+rdm,+sha2,+spe,+ssbs,+v8.1a,+v8.2a,+v8a" "tune-cpu"="cortex-a710" }
+// CHECK: attributes #9 = { {{.*}} "target-features"="+fp-armv8,+fullfp16,+sve" "tune-cpu"="cortex-a710" }
+// CHECK: attributes #10 = { {{.*}} "target-cpu"="neoverse-v1" "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+rand,+ras,+rcpc,+rdm,+sha2,+sha3,+sm4,+spe,+ssbs,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8a" }
+// CHECK: attributes #11 = { {{.*}} "target-cpu"="neoverse-v1" "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+rand,+ras,+rcpc,+rdm,+sha2,+sha3,+sm4,+spe,+ssbs,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8a,-sve" }
+// CHECK: attributes #12 = { {{.*}} "target-features"="+fp-armv8,+fullfp16,+sve" }
+// CHECK: attributes #13 = { {{.*}} "target-features"="+fp-armv8,+fullfp16" }
+// CHECK: attributes #14 = { {{.*}} "target-cpu"="neoverse-n1" "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sha2,+spe,+ssbs,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8.6a,+v8a" "tune-cpu"="cortex-a710" }
+// CHECK: attributes #15 = { {{.*}} "branch-target-enforcement"="true" "guarded-control-stack"="true" {{.*}} "target-features"="+aes,+bf16,+complxnum,+crc,+dotprod,+fp-armv8,+fullfp16,+i8mm,+jsconv,+lse,+neon,+pauth,+ras,+rcpc,+rdm,+sha2,+spe,+ssbs,+sve,+sve2,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8.6a,+v8a" "tune-cpu"="cortex-a710" }
+// CHECK-NOT: attributes #16 = {{.*}} "target-features"
+// CHECK: attributes #17 = { {{.*}} "target-features"="-v9.3a" }
diff --git a/clang/test/CodeGen/attr-target-version.c b/clang/test/CodeGen/attr-target-version.c
index 3597711333d34..75f8734e5aaf3 100644
--- a/clang/test/CodeGen/attr-target-version.c
+++ b/clang/test/CodeGen/attr-target-version.c
@@ -1129,42 +1129,42 @@ int caller(void) { return used_def_without_default_decl() + used_decl_without_de
// CHECK-NOFMV-NEXT: ret i32 0
//
//.
-// CHECK: attributes #[[ATTR0]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+flagm,+fp16fml,+fullfp16,+neon,+rand,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR0]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+flagm,+fp-armv8,+fp16fml,+fullfp16,+neon,+rand,-v9.5a" }
// CHECK: attributes #[[ATTR1]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+altnzcv,+bf16,+flagm,+sme,+sme-i16i64,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR2]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse,+neon,+sha2,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR3]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+ls64,+neon,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR4]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp16fml,+fullfp16,+neon,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR5]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR2]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+lse,+neon,+sha2,-v9.5a" }
+// CHECK: attributes #[[ATTR3]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+fp-armv8,+ls64,+neon,-v9.5a" }
+// CHECK: attributes #[[ATTR4]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fp16fml,+fullfp16,+neon,-v9.5a" }
+// CHECK: attributes #[[ATTR5]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,-v9.5a" }
// CHECK: attributes #[[ATTR6]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+crc,-fp-armv8,-v9.5a" }
// CHECK: attributes #[[ATTR7]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bti,-fp-armv8,-v9.5a" }
// CHECK: attributes #[[ATTR8]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme,+sme2,-fp-armv8,-v9.5a" }
// CHECK: attributes #[[ATTR9:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fp-armv8,-v9.5a" }
// CHECK: attributes #[[ATTR10]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccpp,-fp-armv8,-v9.5a" }
// CHECK: attributes #[[ATTR11]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR12]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR12]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,-v9.5a" }
// CHECK: attributes #[[ATTR13]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+sb,-fp-armv8,-v9.5a" }
// CHECK: attributes #[[ATTR14]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mops,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR15]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+neon,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR16]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,+sve,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR15]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+fp-armv8,+neon,-v9.5a" }
+// CHECK: attributes #[[ATTR16]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,-v9.5a" }
// CHECK: attributes #[[ATTR17]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR18]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,+rdm,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR19:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+jsconv,+neon,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR20:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,+rdm,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR21]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+jsconv,+neon,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR22]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+aes,+f64mm,+fullfp16,+neon,+sve,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR23]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+complxnum,+fullfp16,+neon,+rdm,+sme,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR24]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+f32mm,+fullfp16,+i8mm,+neon,+sha2,+sha3,+sve,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR25]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+dit,+fullfp16,+neon,+sve,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR18]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rdm,-v9.5a" }
+// CHECK: attributes #[[ATTR19:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+jsconv,+neon,-v9.5a" }
+// CHECK: attributes #[[ATTR20:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rdm,-v9.5a" }
+// CHECK: attributes #[[ATTR21]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+jsconv,+neon,-v9.5a" }
+// CHECK: attributes #[[ATTR22]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+aes,+f64mm,+fp-armv8,+fullfp16,+neon,+sve,-v9.5a" }
+// CHECK: attributes #[[ATTR23]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+complxnum,+fp-armv8,+fullfp16,+neon,+rdm,+sme,-v9.5a" }
+// CHECK: attributes #[[ATTR24]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+f32mm,+fp-armv8,+fullfp16,+i8mm,+neon,+sha2,+sha3,+sve,-v9.5a" }
+// CHECK: attributes #[[ATTR25]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+dit,+fp-armv8,+fullfp16,+neon,+sve,-v9.5a" }
// CHECK: attributes #[[ATTR26]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccpp,+rcpc,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR27]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccdp,+ccpp,+jsconv,+neon,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR27]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccdp,+ccpp,+fp-armv8,+jsconv,+neon,-v9.5a" }
// CHECK: attributes #[[ATTR28]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fptoint,+rcpc,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR29]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fullfp16,+neon,+sve,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR30]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,+sve,+sve2,+sve2-aes,+sve2-sha3,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR31]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,+sve,+sve2,+sve2-aes,+sve2-bitperm,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR32]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+mte,+neon,+sve,+sve2,+sve2-sm4,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR29]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sve,-v9.5a" }
+// CHECK: attributes #[[ATTR30]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,+sve2,+sve2-aes,+sve2-sha3,-v9.5a" }
+// CHECK: attributes #[[ATTR31]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,+sve2,+sve2-aes,+sve2-bitperm,-v9.5a" }
+// CHECK: attributes #[[ATTR32]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+mte,+neon,+sve,+sve2,+sve2-sm4,-v9.5a" }
// CHECK: attributes #[[ATTR33]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mops,+mte,+rcpc,+rcpc3,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR34]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,+sm4,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR35]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse,+neon,+rdm,-fp-armv8,-v9.5a" }
+// CHECK: attributes #[[ATTR34]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+sm4,-v9.5a" }
+// CHECK: attributes #[[ATTR35]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+lse,+neon,+rdm,-v9.5a" }
//.
// CHECK-NOFMV: attributes #[[ATTR0]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fmv" }
// CHECK-NOFMV: attributes #[[ATTR1:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fmv" }
diff --git a/clang/test/Sema/aarch64-neon-target.c b/clang/test/Sema/aarch64-neon-target.c
index fa45fff1d183d..642afddd88c15 100644
--- a/clang/test/Sema/aarch64-neon-target.c
+++ b/clang/test/Sema/aarch64-neon-target.c
@@ -69,8 +69,8 @@ void undefined(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t
vrnd_f16(v4f16); // expected-error {{always_inline function 'vrnd_f16' requires target feature 'fullfp16'}}
vmaxnm_f16(v4f16, v4f16); // expected-error {{always_inline function 'vmaxnm_f16' requires target feature 'fullfp16'}}
vrndi_f16(v4f16); // expected-error {{always_inline function 'vrndi_f16' requires target feature 'fullfp16'}}
- // fp16fml
- vfmlal_low_f16(v2f32, v4f16, v4f16); // expected-error {{always_inline function 'vfmlal_low_f16' requires target feature 'fp16fml'}}
+ // fp16fml depends on fp-armv8
+ vfmlal_low_f16(v2f32, v4f16, v4f16); // expected-error {{always_inline function 'vfmlal_low_f16' requires target feature 'fp-armv8'}}
// i8mm
vmmlaq_s32(v4i32, v8i16, v8i16); // expected-error {{always_inline function 'vmmlaq_s32' requires target feature 'i8mm'}}
vusdot_laneq_s32(v2i32, v8i8, v8i16, 0); // expected-error {{always_inline function 'vusdot_s32' requires target feature 'i8mm'}}
diff --git a/llvm/include/llvm/TargetParser/AArch64TargetParser.h b/llvm/include/llvm/TargetParser/AArch64TargetParser.h
index df8e685eb6667..9da4bf4471c9b 100644
--- a/llvm/include/llvm/TargetParser/AArch64TargetParser.h
+++ b/llvm/include/llvm/TargetParser/AArch64TargetParser.h
@@ -132,48 +132,6 @@ struct ExtensionInfo {
#define EMIT_EXTENSIONS
#include "llvm/TargetParser/AArch64TargetParserDef.inc"
-struct ExtensionSet {
- // Set of extensions which are currently enabled.
- ExtensionBitset Enabled;
- // Set of extensions which have been enabled or disabled at any point. Used
- // to avoid cluttering the cc1 command-line with lots of unneeded features.
- ExtensionBitset Touched;
- // Base architecture version, which we need to know because some feature
- // dependencies change depending on this.
- const ArchInfo *BaseArch;
-
- ExtensionSet() : Enabled(), Touched(), BaseArch(nullptr) {}
-
- // Enable the given architecture extension, and any other extensions it
- // depends on. Does not change the base architecture, or follow dependencies
- // between features which are only related by required arcitecture versions.
- void enable(ArchExtKind E);
-
- // Disable the given architecture extension, and any other extensions which
- // depend on it. Does not change the base architecture, or follow
- // dependencies between features which are only related by required
- // arcitecture versions.
- void disable(ArchExtKind E);
-
- // Add default extensions for the given CPU. Records the base architecture,
- // to later resolve dependencies which depend on it.
- void addCPUDefaults(const CpuInfo &CPU);
-
- // Add default extensions for the given architecture version. Records the
- // base architecture, to later resolve dependencies which depend on it.
- void addArchDefaults(const ArchInfo &Arch);
-
- // Add or remove a feature based on a modifier string. The string must be of
- // the form "<name>" to enable a feature or "no<name>" to disable it. This
- // will also enable or disable any features as required by the dependencies
- // between them.
- bool parseModifier(StringRef Modifier);
-
- // Convert the set of enabled extension to an LLVM feature list, appending
- // them to Features.
- void toLLVMFeatureList(std::vector<StringRef> &Features) const;
-};
-
// Represents a dependency between two architecture extensions. Later is the
// feature which was added to the architecture after Earlier, and expands the
// functionality provided by it. If Later is enabled, then Earlier will also be
@@ -596,6 +554,65 @@ inline constexpr CpuInfo CpuInfos[] = {
AArch64::AEK_PROFILE}))},
};
+struct ExtensionSet {
+ // Set of extensions which are currently enabled.
+ ExtensionBitset Enabled;
+ // Set of extensions which have been enabled or disabled at any point. Used
+ // to avoid cluttering the cc1 command-line with lots of unneeded features.
+ ExtensionBitset Touched;
+ // Base architecture version, which we need to know because some feature
+ // dependencies change depending on this.
+ const ArchInfo *BaseArch;
+
+ ExtensionSet() : Enabled(), Touched(), BaseArch(nullptr) {}
+
+ // Enable the given architecture extension, and any other extensions it
+ // depends on. Does not change the base architecture, or follow dependencies
+ // between features which are only related by required arcitecture versions.
+ void enable(ArchExtKind E);
+
+ // Disable the given architecture extension, and any other extensions which
+ // depend on it. Does not change the base architecture, or follow
+ // dependencies between features which are only related by required
+ // arcitecture versions.
+ void disable(ArchExtKind E);
+
+ // Add default extensions for the given CPU. Records the base architecture,
+ // to later resolve dependencies which depend on it.
+ void addCPUDefaults(const CpuInfo &CPU);
+
+ // Add default extensions for the given architecture version. Records the
+ // base architecture, to later resolve dependencies which depend on it.
+ void addArchDefaults(const ArchInfo &Arch);
+
+ // Add or remove a feature based on a modifier string. The string must be of
+ // the form "<name>" to enable a feature or "no<name>" to disable it. This
+ // will also enable or disable any features as required by the dependencies
+ // between them.
+ bool parseModifier(StringRef Modifier, const bool AllowNoDashForm = false);
+
+ // Constructs a new ExtensionSet by toggling the corresponding bits for every
+ // feature in the \p Features list without expanding their dependencies. Used
+ // for reconstructing an ExtensionSet from the output of toLLVMFeatures().
+ void reconstructFromParsedFeatures(const std::vector<std::string> &Features);
+
+ // Convert the set of enabled extension to an LLVM feature list, appending
+ // them to Features.
+ template <typename T> void toLLVMFeatureList(std::vector<T> &Features) const {
+ if (BaseArch && !BaseArch->ArchFeature.empty())
+ Features.emplace_back(T(BaseArch->ArchFeature));
+
+ for (const auto &E : Extensions) {
+ if (E.Feature.empty() || !Touched.test(E.ID))
+ continue;
+ if (Enabled.test(E.ID))
+ Features.emplace_back(T(E.Feature));
+ else
+ Features.emplace_back(T(E.NegFeature));
+ }
+ }
+};
+
// Name alias.
struct Alias {
StringRef AltName;
@@ -619,7 +636,13 @@ const ArchInfo *getArchForCpu(StringRef CPU);
// Parser
const ArchInfo *parseArch(StringRef Arch);
+
+// Return the extension which has the given -target-feature name.
+std::optional<ExtensionInfo> targetFeatureToExtension(StringRef TargetFeature);
+
+// Parse a name as defined by the Extension class in tablegen.
std::optional<ExtensionInfo> parseArchExtension(StringRef Extension);
+
// Given the name of a CPU or alias, return the correponding CpuInfo.
std::optional<CpuInfo> parseCpu(StringRef Name);
// Used by target parser tests
diff --git a/llvm/lib/TargetParser/AArch64TargetParser.cpp b/llvm/lib/TargetParser/AArch64TargetParser.cpp
index ca356ec82bf1f..d1cc306790522 100644
--- a/llvm/lib/TargetParser/AArch64TargetParser.cpp
+++ b/llvm/lib/TargetParser/AArch64TargetParser.cpp
@@ -122,6 +122,14 @@ AArch64::parseArchExtension(StringRef ArchExt) {
return {};
}
+std::optional<AArch64::ExtensionInfo>
+AArch64::targetFeatureToExtension(StringRef TargetFeature) {
+ for (const auto &E : Extensions)
+ if (TargetFeature == E.Feature)
+ return E;
+ return {};
+}
+
std::optional<AArch64::CpuInfo> AArch64::parseCpu(StringRef Name) {
// Resolve aliases first.
Name = resolveCPUAlias(Name);
@@ -213,21 +221,6 @@ void AArch64::ExtensionSet::disable(ArchExtKind E) {
disable(Dep.Later);
}
-void AArch64::ExtensionSet::toLLVMFeatureList(
- std::vector<StringRef> &Features) const {
- if (BaseArch && !BaseArch->ArchFeature.empty())
- Features.push_back(BaseArch->ArchFeature);
-
- for (const auto &E : Extensions) {
- if (E.Feature.empty() || !Touched.test(E.ID))
- continue;
- if (Enabled.test(E.ID))
- Features.push_back(E.Feature);
- else
- Features.push_back(E.NegFeature);
- }
-}
-
void AArch64::ExtensionSet::addCPUDefaults(const CpuInfo &CPU) {
LLVM_DEBUG(llvm::dbgs() << "addCPUDefaults(" << CPU.Name << ")\n");
BaseArch = &CPU.Arch;
@@ -247,11 +240,18 @@ void AArch64::ExtensionSet::addArchDefaults(const ArchInfo &Arch) {
enable(E.ID);
}
-bool AArch64::ExtensionSet::parseModifier(StringRef Modifier) {
+bool AArch64::ExtensionSet::parseModifier(StringRef Modifier,
+ const bool AllowNoDashForm) {
LLVM_DEBUG(llvm::dbgs() << "parseModifier(" << Modifier << ")\n");
- bool IsNegated = Modifier.starts_with("no");
- StringRef ArchExt = IsNegated ? Modifier.drop_front(2) : Modifier;
+ size_t NChars = 0;
+ // The "no-feat" form is allowed in the target attribute but nowhere else.
+ if (AllowNoDashForm && Modifier.starts_with("no-"))
+ NChars = 3;
+ else if (Modifier.starts_with("no"))
+ NChars = 2;
+ bool IsNegated = NChars != 0;
+ StringRef ArchExt = Modifier.drop_front(NChars);
if (auto AE = parseArchExtension(ArchExt)) {
if (AE->Feature.empty() || AE->NegFeature.empty())
@@ -265,6 +265,21 @@ bool AArch64::ExtensionSet::parseModifier(StringRef Modifier) {
return false;
}
+void AArch64::ExtensionSet::reconstructFromParsedFeatures(
+ const std::vector<std::string> &Features) {
+ assert(Touched.none() && "Bitset already initialized");
+ for (auto &F : Features) {
+ bool IsNegated = F[0] == '-';
+ if (auto AE = targetFeatureToExtension(F)) {
+ Touched.set(AE->ID);
+ if (IsNegated)
+ Enabled.reset(AE->ID);
+ else
+ Enabled.set(AE->ID);
+ }
+ }
+}
+
const AArch64::ExtensionInfo &
AArch64::getExtensionByID(AArch64::ArchExtKind ExtID) {
return lookupExtensionByID(ExtID);
>From 442e4eef2001fbf65a95236d8afb547e82bf92d0 Mon Sep 17 00:00:00 2001
From: Alexandros Lamprineas <alexandros.lamprineas at arm.com>
Date: Fri, 14 Jun 2024 09:10:23 +0100
Subject: [PATCH 2/5] Changes from last revision:
* We should be inserting the Driver features in front of the features
of a parsed target attribute.
* Added a sema test.
---
clang/lib/AST/ASTContext.cpp | 10 ++++------
clang/test/Sema/aarch64-neon-target.c | 6 ++++++
2 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
index aa22825602a40..7fb5966c0bf7c 100644
--- a/clang/lib/AST/ASTContext.cpp
+++ b/clang/lib/AST/ASTContext.cpp
@@ -13725,12 +13725,10 @@ void ASTContext::getFunctionFeatureMap(llvm::StringMap<bool> &FeatureMap,
// Make a copy of the features as passed on the command line into the
// beginning of the additional features from the function to override.
- // AArch64 handles command line option features in parseTargetAttr().
- if (!Target->getTriple().isAArch64())
- ParsedAttr.Features.insert(
- ParsedAttr.Features.begin(),
- Target->getTargetOpts().FeaturesAsWritten.begin(),
- Target->getTargetOpts().FeaturesAsWritten.end());
+ ParsedAttr.Features.insert(
+ ParsedAttr.Features.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.end());
if (ParsedAttr.CPU != "" && Target->isValidCPUName(ParsedAttr.CPU))
TargetCPU = ParsedAttr.CPU;
diff --git a/clang/test/Sema/aarch64-neon-target.c b/clang/test/Sema/aarch64-neon-target.c
index 642afddd88c15..fd84b7f2eb00a 100644
--- a/clang/test/Sema/aarch64-neon-target.c
+++ b/clang/test/Sema/aarch64-neon-target.c
@@ -1,4 +1,5 @@
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +neon -verify -emit-llvm -o - %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +neon -target-feature +outline-atomics -verify -emit-llvm -o - %s
// REQUIRES: aarch64-registered-target
// Test that functions with the correct target attributes can use the correct NEON intrinsics.
@@ -41,6 +42,11 @@ void bf16(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t v16i8
vcvt_bf16_f32(v4f32);
}
+__attribute__((target("arch=armv8-a")))
+uint64x2_t test_v8(uint64x2_t a, uint64x2_t b) {
+ return veorq_u64(a, b);
+}
+
__attribute__((target("arch=armv8.1-a")))
void test_v81(int32x2_t d, int32x4_t v, int s) {
vqrdmlahq_s32(v, v, v);
>From cf4dcdb9972d6bc0dd1ab879f211c4aaf174b8b5 Mon Sep 17 00:00:00 2001
From: Alexandros Lamprineas <alexandros.lamprineas at arm.com>
Date: Fri, 14 Jun 2024 12:27:33 +0100
Subject: [PATCH 3/5] Changes from last revision:
* Revert inserting the Driver features in front of the features
of a parsed target attribute.
* Instead make reconstructFromParsedFeatures() add unrecognized
features to the features of a parsed target attribute.
* Do not allow 'no-' form when parsing '+' separated features.
* Updated misleading comments and added FIXMEs.
---
clang/lib/AST/ASTContext.cpp | 10 ++++---
clang/lib/Basic/Targets/AArch64.cpp | 26 ++++++++++++-------
.../llvm/TargetParser/AArch64TargetParser.h | 4 ++-
llvm/lib/TargetParser/AArch64TargetParser.cpp | 5 +++-
4 files changed, 30 insertions(+), 15 deletions(-)
diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
index 7fb5966c0bf7c..aa22825602a40 100644
--- a/clang/lib/AST/ASTContext.cpp
+++ b/clang/lib/AST/ASTContext.cpp
@@ -13725,10 +13725,12 @@ void ASTContext::getFunctionFeatureMap(llvm::StringMap<bool> &FeatureMap,
// Make a copy of the features as passed on the command line into the
// beginning of the additional features from the function to override.
- ParsedAttr.Features.insert(
- ParsedAttr.Features.begin(),
- Target->getTargetOpts().FeaturesAsWritten.begin(),
- Target->getTargetOpts().FeaturesAsWritten.end());
+ // AArch64 handles command line option features in parseTargetAttr().
+ if (!Target->getTriple().isAArch64())
+ ParsedAttr.Features.insert(
+ ParsedAttr.Features.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.begin(),
+ Target->getTargetOpts().FeaturesAsWritten.end());
if (ParsedAttr.CPU != "" && Target->isValidCPUName(ParsedAttr.CPU))
TargetCPU = ParsedAttr.CPU;
diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp
index 6fba5fff7bcc1..a43937de819cf 100644
--- a/clang/lib/Basic/Targets/AArch64.cpp
+++ b/clang/lib/Basic/Targets/AArch64.cpp
@@ -1061,9 +1061,8 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
//
// A feature may correspond to an Extension (anything with a corresponding
// AEK_), in which case an ExtensionSet is used to parse it and expand its
-// dependencies. Otherwise the feature is passed through (e.g. +v8.1a,
-// +outline-atomics, -fmv, etc). Features coming from the command line are
-// already parsed, therefore their dependencies do not need expansion.
+// dependencies. If the feature does not yield a successful parse then it
+// is be passed through.
ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
ParsedTargetAttr Ret;
if (Features == "default")
@@ -1078,10 +1077,14 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
SmallVector<StringRef, 8> SplitFeatures;
FeatString.split(SplitFeatures, StringRef("+"), -1, false);
for (StringRef Feature : SplitFeatures) {
- if (FeatureBits.parseModifier(Feature, /* AllowNoDashForm = */ true))
+ if (FeatureBits.parseModifier(Feature))
continue;
- // Pass through features that are not extensions, e.g. +v8.1a,
- // +outline-atomics, -fmv, etc.
+ // Pass through anything that failed to parse so that we can emit
+ // diagnostics, as well as valid internal feature names.
+ //
+ // FIXME: We should consider rejecting internal feature names like
+ // neon, v8a, etc.
+ // FIXME: We should consider emitting diagnostics here.
if (Feature.starts_with("no"))
Features.push_back("-" + Feature.drop_front(2).str());
else
@@ -1091,7 +1094,8 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
llvm::AArch64::ExtensionSet FeatureBits;
// Reconstruct the bitset from the command line option features.
- FeatureBits.reconstructFromParsedFeatures(getTargetOpts().FeaturesAsWritten);
+ FeatureBits.reconstructFromParsedFeatures(getTargetOpts().FeaturesAsWritten,
+ Ret.Features);
for (auto &Feature : AttrFeatures) {
Feature = Feature.trim();
@@ -1142,8 +1146,12 @@ ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
} else {
if (FeatureBits.parseModifier(Feature, /* AllowNoDashForm = */ true))
continue;
- // Pass through features that are not extensions, e.g. +v8.1a,
- // +outline-atomics, -fmv, etc.
+ // Pass through anything that failed to parse so that we can emit
+ // diagnostics, as well as valid internal feature names.
+ //
+ // FIXME: We should consider rejecting internal feature names like
+ // neon, v8a, etc.
+ // FIXME: We should consider emitting diagnostics here.
if (Feature.starts_with("no-"))
Ret.Features.push_back("-" + Feature.drop_front(3).str());
else
diff --git a/llvm/include/llvm/TargetParser/AArch64TargetParser.h b/llvm/include/llvm/TargetParser/AArch64TargetParser.h
index 9da4bf4471c9b..7b5d59f82d9bd 100644
--- a/llvm/include/llvm/TargetParser/AArch64TargetParser.h
+++ b/llvm/include/llvm/TargetParser/AArch64TargetParser.h
@@ -594,7 +594,9 @@ struct ExtensionSet {
// Constructs a new ExtensionSet by toggling the corresponding bits for every
// feature in the \p Features list without expanding their dependencies. Used
// for reconstructing an ExtensionSet from the output of toLLVMFeatures().
- void reconstructFromParsedFeatures(const std::vector<std::string> &Features);
+ // Features that are not recognized are pushed back to \p NonExtensions.
+ void reconstructFromParsedFeatures(const std::vector<std::string> &Features,
+ std::vector<std::string> &NonExtensions);
// Convert the set of enabled extension to an LLVM feature list, appending
// them to Features.
diff --git a/llvm/lib/TargetParser/AArch64TargetParser.cpp b/llvm/lib/TargetParser/AArch64TargetParser.cpp
index d1cc306790522..0f6a97144e04d 100644
--- a/llvm/lib/TargetParser/AArch64TargetParser.cpp
+++ b/llvm/lib/TargetParser/AArch64TargetParser.cpp
@@ -266,7 +266,8 @@ bool AArch64::ExtensionSet::parseModifier(StringRef Modifier,
}
void AArch64::ExtensionSet::reconstructFromParsedFeatures(
- const std::vector<std::string> &Features) {
+ const std::vector<std::string> &Features,
+ std::vector<std::string> &NonExtensions) {
assert(Touched.none() && "Bitset already initialized");
for (auto &F : Features) {
bool IsNegated = F[0] == '-';
@@ -276,7 +277,9 @@ void AArch64::ExtensionSet::reconstructFromParsedFeatures(
Enabled.reset(AE->ID);
else
Enabled.set(AE->ID);
+ continue;
}
+ NonExtensions.push_back(F);
}
}
>From a7d83638992762fbde60623b2bbe0fa9943ec992 Mon Sep 17 00:00:00 2001
From: Alexandros Lamprineas <alexandros.lamprineas at arm.com>
Date: Fri, 14 Jun 2024 12:45:13 +0100
Subject: [PATCH 4/5] Fix typo
---
clang/lib/Basic/Targets/AArch64.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp
index a43937de819cf..418f124c4df56 100644
--- a/clang/lib/Basic/Targets/AArch64.cpp
+++ b/clang/lib/Basic/Targets/AArch64.cpp
@@ -1062,7 +1062,7 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
// A feature may correspond to an Extension (anything with a corresponding
// AEK_), in which case an ExtensionSet is used to parse it and expand its
// dependencies. If the feature does not yield a successful parse then it
-// is be passed through.
+// is passed through.
ParsedTargetAttr AArch64TargetInfo::parseTargetAttr(StringRef Features) const {
ParsedTargetAttr Ret;
if (Features == "default")
>From 2ef457d985400062ba868a6b15e5787e28dcc7d5 Mon Sep 17 00:00:00 2001
From: Alexandros Lamprineas <alexandros.lamprineas at arm.com>
Date: Tue, 18 Jun 2024 15:22:54 +0100
Subject: [PATCH 5/5] Changes from last revision:
Fixes three tests which fail after the rebase by adding -target-feature +sve.
---
.../acle_sve2_aes_bitperm_sha3_sm4.cpp | 4 ++--
.../test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_lane.cpp | 4 ++--
.../Sema/aarch64-sve2-intrinsics/acle_sve2_imm_rotation.cpp | 4 ++--
3 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_aes_bitperm_sha3_sm4.cpp b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_aes_bitperm_sha3_sm4.cpp
index 898f18efa9386..81cbb2dce921b 100644
--- a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_aes_bitperm_sha3_sm4.cpp
+++ b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_aes_bitperm_sha3_sm4.cpp
@@ -1,7 +1,7 @@
// REQUIRES: aarch64-registered-target
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2 -verify -verify-ignore-unexpected=error,note -emit-llvm -o - %s
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -verify=overload -verify-ignore-unexpected=error,note -emit-llvm -o - %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -verify -verify-ignore-unexpected=error,note -emit-llvm -o - %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -verify=overload -verify-ignore-unexpected=error,note -emit-llvm -o - %s
#ifdef SVE_OVERLOADED_FORMS
// A simple used,unused... macro, long enough to represent any SVE builtin.
diff --git a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_lane.cpp b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_lane.cpp
index 520238ba3c6b0..bca063385420a 100644
--- a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_lane.cpp
+++ b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_lane.cpp
@@ -1,7 +1,7 @@
// REQUIRES: aarch64-registered-target
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2 -fsyntax-only -verify %s
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -fsyntax-only -verify %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -fsyntax-only -verify %s
#ifdef SVE_OVERLOADED_FORMS
// A simple used,unused... macro, long enough to represent any SVE builtin.
diff --git a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_rotation.cpp b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_rotation.cpp
index 0ccc35fb44d49..343d2a0a97462 100644
--- a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_rotation.cpp
+++ b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_imm_rotation.cpp
@@ -1,7 +1,7 @@
// REQUIRES: aarch64-registered-target
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2 -fsyntax-only -verify %s
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -fsyntax-only -verify %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -fsyntax-only -verify %s
#ifdef SVE_OVERLOADED_FORMS
// A simple used,unused... macro, long enough to represent any SVE builtin.
More information about the llvm-commits
mailing list