[X86] Honor command line features along with cpu_specific attribute

[craig.topper]

Summary

If the feature is on the command line we should honor it for all
functions. I don't think we could reliably target a single function
for a less capable processor than what the rest of the program is
compiled for.

Fixes PR52407.

[erichkeane]

Is there a lit-test that could be added to make sure this happens? We put this in an llvm-attribute, so it should be checkable.

[craig.topper]

In D113647#3124363, @erichkeane wrote:

Is there a lit-test that could be added to make sure this happens? We put this in an llvm-attribute, so it should be checkable.

Forgot to git add the file I wrote.

[craig.topper]

Add test file

clang/lib/AST/ASTContext.cpp

if (const auto *TD = FD->getAttr<TargetAttr>()) {
     // the attribute.
     Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU,
                            ParsedAttr.Features);
   } else if (const auto *SD = FD->getAttr<CPUSpecificAttr>()) {
     llvm::SmallVector<StringRef, 32> FeaturesTmp;
     Target->getCPUSpecificCPUDispatchFeatures(
         SD->getCPUName(GD.getMultiVersionIndex())->getName(), FeaturesTmp);
     std::vector<std::string> Features(FeaturesTmp.begin(), FeaturesTmp.end());
+    Features.insert(Features.begin(),
+                    Target->getTargetOpts().FeaturesAsWritten.begin(),
+                    Target->getTargetOpts().FeaturesAsWritten.end());
     Target->initFeatureMap(FeatureMap, getDiagnostics(), TargetCPU, Features);
   } else {
     FeatureMap = Target->getTargetOpts().FeatureMap;
   }
 }
 
 OMPTraitInfo &ASTContext::getNewOMPTraitInfo() {
   OMPTraitInfoVector.emplace_back(new OMPTraitInfo());

clang/test/CodeGen/attr-cpuspecific-avx-abi.c

+// RUN: %clang_cc1 -triple x86_64-linux-gnu -target-feature +avx -emit-llvm -o - %s | FileCheck %s --check-prefixes=CHECK
+
+// Make sure the features from the command line are honored regardless of what
+// CPU is specified in the cpu_specific attribute.
+// In this test, if the 'avx' feature isn't honored, we'll generate an error for
+// the return type having a different ABI without 'avx' being enabled.
+
+typedef double __m256d __attribute__((vector_size(32)));
+
+extern __m256d bar_avx1(void);
+extern __m256d bar_avx2(void);
+
+// AVX1/AVX2 dispatcher
+__attribute__((cpu_dispatch(generic, core_4th_gen_avx)))
+__m256d foo_pd64x4(void);
+
+__attribute__((cpu_specific(generic)))
+__m256d foo(void) { return bar_avx1(); }
+// CHECK: define{{.*}} @foo.A() #[[A:[0-9]+]]
+
+__attribute__((cpu_specific(core_4th_gen_avx)))
+__m256d foo(void) { return bar_avx2(); }
+// CHECK: define{{.*}} @foo.V() #[[V:[0-9]+]]
+
+// CHECK: attributes #[[A]] = {{.*}}"target-features"="+avx,+crc32,+cx8,+mmx,+popcnt,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave"
+// CHECK: attributes #[[V]] = {{.*}}"target-features"="+avx,+avx2,+bmi,+cmov,+crc32,+cx8,+f16c,+fma,+lzcnt,+mmx,+movbe,+popcnt,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave"