[clang] [FMV][AArch64] Do not optimize away runtime checks for implied features (PR #99522)

Alexandros Lamprineas via cfe-commits cfe-commits at lists.llvm.org
Thu Jul 18 09:28:43 PDT 2024


https://github.com/labrinea created https://github.com/llvm/llvm-project/pull/99522

When generating the body of the ifunc resolver, clang skips runtime checks for features that are implied from the command line. We bend this rule for certain features (memtag, bti, dgh), but this happens quite arbitrarily in my opinion. The reasoning is that some features are in the HINT instruction space, meaning they operate as NOPs if the hardware does not support them. Still the user wants to detect their presence with runtime checks. See #90928 for details.

I think we should always perform runtime checks regardless of the feature and then try to statically resolve calls whenever a function is compiled with a sufficiently high set of architecture features (so including target/target_version/target_clones attributes, and command line options). This is what GCC does. We have an open PR in LLVM GlobalOpt since it was suggested not to perform such codegen optimizations in clang anyway. See #87939.

>From 5ecefdf788cf9759092d3f349f845353a7023ec9 Mon Sep 17 00:00:00 2001
From: Alexandros Lamprineas <alexandros.lamprineas at arm.com>
Date: Thu, 18 Jul 2024 17:02:13 +0100
Subject: [PATCH] [FMV][AArch64] Do not optimize away runtime checks for
 implied features

When generating the body of the ifunc resolver, clang skips runtime
checks for features that are implied from the command line. We bend
this rule for certain features (memtag, bti, dgh), but this happens
quite arbitrarily in my opinion. The reasoning is that some features
are in the HINT instruction space, meaning they operate as NOPs if
the hardware does not support them. Still the user wants to detect
their presence with runtime checks. See #90928 for details.

I think we should always perform runtime checks regardless of the
feature and then try to statically resolve calls whenever a function
is compiled with a sufficiently high set of architecture features
(so including target/target_version/target_clones attributes, and
command line options). This is what GCC does. We have an open PR in
LLVM GlobalOpt since it was suggested not to perform such codegen
optimizations in clang anyway. See #87939.
---
 clang/lib/CodeGen/CodeGenFunction.cpp         | 12 +--
 .../test/CodeGen/attr-target-clones-aarch64.c | 16 ++++
 clang/test/CodeGen/attr-target-version.c      | 74 +++++++++----------
 3 files changed, 55 insertions(+), 47 deletions(-)

diff --git a/clang/lib/CodeGen/CodeGenFunction.cpp b/clang/lib/CodeGen/CodeGenFunction.cpp
index 551db09165dbe..1e98bea8c8ce3 100644
--- a/clang/lib/CodeGen/CodeGenFunction.cpp
+++ b/clang/lib/CodeGen/CodeGenFunction.cpp
@@ -2797,16 +2797,8 @@ void CodeGenFunction::EmitKCFIOperandBundle(
 llvm::Value *CodeGenFunction::FormAArch64ResolverCondition(
     const MultiVersionResolverOption &RO) {
   llvm::SmallVector<StringRef, 8> CondFeatures;
-  for (const StringRef &Feature : RO.Conditions.Features) {
-    // Optimize the Function Multi Versioning resolver by creating conditions
-    // only for features that are not enabled in the target. The exception is
-    // for features whose extension instructions are executed as NOP on targets
-    // without extension support.
-    if (!getContext().getTargetInfo().hasFeature(Feature) || Feature == "bti" ||
-        Feature == "memtag" || Feature == "memtag2" || Feature == "memtag3" ||
-        Feature == "dgh")
-      CondFeatures.push_back(Feature);
-  }
+  for (const StringRef &Feature : RO.Conditions.Features)
+    CondFeatures.push_back(Feature);
   if (!CondFeatures.empty()) {
     return EmitAArch64CpuSupports(CondFeatures);
   }
diff --git a/clang/test/CodeGen/attr-target-clones-aarch64.c b/clang/test/CodeGen/attr-target-clones-aarch64.c
index 846b08298cc72..274e05de594b8 100644
--- a/clang/test/CodeGen/attr-target-clones-aarch64.c
+++ b/clang/test/CodeGen/attr-target-clones-aarch64.c
@@ -157,7 +157,15 @@ inline int __attribute__((target_clones("fp16", "sve2-bitperm+fcma", "default"))
 // CHECK:       resolver_return:
 // CHECK-NEXT:    ret ptr @ftc_dup2._McrcMdotprod
 // CHECK:       resolver_else:
+// CHECK-NEXT:    [[TMP4:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT:    [[TMP5:%.*]] = and i64 [[TMP4]], 256
+// CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[TMP5]], 256
+// CHECK-NEXT:    [[TMP7:%.*]] = and i1 true, [[TMP6]]
+// CHECK-NEXT:    br i1 [[TMP7]], label [[RESOLVER_RETURN1:%.*]], label [[RESOLVER_ELSE2:%.*]]
+// CHECK:       resolver_return1:
 // CHECK-NEXT:    ret ptr @ftc_dup2._Mfp
+// CHECK:       resolver_else2:
+// CHECK-NEXT:    ret ptr @ftc_dup2.default
 //
 //
 // CHECK: Function Attrs: noinline nounwind optnone
@@ -571,7 +579,15 @@ inline int __attribute__((target_clones("fp16", "sve2-bitperm+fcma", "default"))
 // CHECK-MTE-BTI:       resolver_return:
 // CHECK-MTE-BTI-NEXT:    ret ptr @ftc_dup2._McrcMdotprod
 // CHECK-MTE-BTI:       resolver_else:
+// CHECK-MTE-BTI-NEXT:    [[TMP4:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-MTE-BTI-NEXT:    [[TMP5:%.*]] = and i64 [[TMP4]], 256
+// CHECK-MTE-BTI-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[TMP5]], 256
+// CHECK-MTE-BTI-NEXT:    [[TMP7:%.*]] = and i1 true, [[TMP6]]
+// CHECK-MTE-BTI-NEXT:    br i1 [[TMP7]], label [[RESOLVER_RETURN1:%.*]], label [[RESOLVER_ELSE2:%.*]]
+// CHECK-MTE-BTI:       resolver_return1:
 // CHECK-MTE-BTI-NEXT:    ret ptr @ftc_dup2._Mfp
+// CHECK-MTE-BTI:       resolver_else2:
+// CHECK-MTE-BTI-NEXT:    ret ptr @ftc_dup2.default
 //
 //
 // CHECK-MTE-BTI: Function Attrs: noinline nounwind optnone
diff --git a/clang/test/CodeGen/attr-target-version.c b/clang/test/CodeGen/attr-target-version.c
index fbe34a51b40b6..b058f84f78baa 100644
--- a/clang/test/CodeGen/attr-target-version.c
+++ b/clang/test/CodeGen/attr-target-version.c
@@ -1,5 +1,5 @@
 // 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-none-linux-gnu -target-feature -v9.5a -target-feature -fp-armv8 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -emit-llvm -o - %s | FileCheck %s
 // RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature -fmv -emit-llvm -o - %s | FileCheck %s -check-prefix=CHECK-NOFMV
 
 int __attribute__((target_version("rng+flagm+fp16fml"))) fmv(void) { return 1; }
@@ -1112,42 +1112,42 @@ int caller(void) { return used_def_without_default_decl() + used_decl_without_de
 // CHECK-NOFMV-NEXT:    ret i32 1
 //
 //.
-// 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"="+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]] = { noinline nounwind optnone "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,+fullfp16,+neon,-v9.5a" }
-// CHECK: attributes #[[ATTR12:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR13]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mops,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR14]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+fp-armv8,+neon,-v9.5a" }
-// CHECK: attributes #[[ATTR15]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,-v9.5a" }
-// CHECK: attributes #[[ATTR16]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR17]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rdm,-v9.5a" }
-// CHECK: attributes #[[ATTR18:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+jsconv,+neon,-v9.5a" }
-// CHECK: attributes #[[ATTR19]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+jsconv,+neon,-v9.5a" }
-// CHECK: attributes #[[ATTR20]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+sb,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR21]] = { 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 #[[ATTR22]] = { 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 #[[ATTR23]] = { 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 #[[ATTR24]] = { 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 #[[ATTR25]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccpp,+rcpc,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR26]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccdp,+ccpp,+fp-armv8,+jsconv,+neon,-v9.5a" }
-// CHECK: attributes #[[ATTR27]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fptoint,+rcpc,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR28]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sve,-v9.5a" }
-// CHECK: attributes #[[ATTR29]] = { 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 #[[ATTR30]] = { 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 #[[ATTR31]] = { 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 #[[ATTR32]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mops,+mte,+rcpc,+rcpc3,-fp-armv8,-v9.5a" }
-// CHECK: attributes #[[ATTR33]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+sm4,-v9.5a" }
-// CHECK: attributes #[[ATTR34]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+lse,+neon,+rdm,-v9.5a" }
-// CHECK: attributes #[[ATTR35:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rdm,-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" }
+// CHECK: attributes #[[ATTR1]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+altnzcv,+bf16,+flagm,+sme,+sme-i16i64" }
+// CHECK: attributes #[[ATTR2]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+lse,+neon,+sha2" }
+// CHECK: attributes #[[ATTR3]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+fp-armv8,+ls64,+neon" }
+// CHECK: attributes #[[ATTR4]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fp16fml,+fullfp16,+neon" }
+// CHECK: attributes #[[ATTR5]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon" }
+// CHECK: attributes #[[ATTR6]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+crc" }
+// CHECK: attributes #[[ATTR7]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bti" }
+// CHECK: attributes #[[ATTR8]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme,+sme2" }
+// CHECK: attributes #[[ATTR9]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #[[ATTR10]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccpp" }
+// CHECK: attributes #[[ATTR11]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon" }
+// CHECK: attributes #[[ATTR12:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #[[ATTR13]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mops" }
+// CHECK: attributes #[[ATTR14]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+dotprod,+fp-armv8,+neon" }
+// CHECK: attributes #[[ATTR15]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve" }
+// CHECK: attributes #[[ATTR16]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse" }
+// CHECK: attributes #[[ATTR17]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rdm" }
+// CHECK: attributes #[[ATTR18:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+jsconv,+neon" }
+// CHECK: attributes #[[ATTR19]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+jsconv,+neon" }
+// CHECK: attributes #[[ATTR20]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+sb" }
+// CHECK: attributes #[[ATTR21]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+aes,+f64mm,+fp-armv8,+fullfp16,+neon,+sve" }
+// CHECK: attributes #[[ATTR22]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+complxnum,+fp-armv8,+fullfp16,+neon,+rdm,+sme" }
+// CHECK: attributes #[[ATTR23]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+f32mm,+fp-armv8,+fullfp16,+i8mm,+neon,+sha2,+sha3,+sve" }
+// CHECK: attributes #[[ATTR24]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+dit,+fp-armv8,+fullfp16,+neon,+sve" }
+// CHECK: attributes #[[ATTR25]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccpp,+rcpc" }
+// CHECK: attributes #[[ATTR26]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ccdp,+ccpp,+fp-armv8,+jsconv,+neon" }
+// CHECK: attributes #[[ATTR27]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fptoint,+rcpc" }
+// CHECK: attributes #[[ATTR28]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sve" }
+// CHECK: attributes #[[ATTR29]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,+sve2,+sve2-aes,+sve2-sha3" }
+// 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-bitperm" }
+// CHECK: attributes #[[ATTR31]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+mte,+neon,+sve,+sve2,+sve2-sm4" }
+// CHECK: attributes #[[ATTR32]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mops,+mte,+rcpc,+rcpc3" }
+// CHECK: attributes #[[ATTR33]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+sm4" }
+// CHECK: attributes #[[ATTR34]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+lse,+neon,+rdm" }
+// CHECK: attributes #[[ATTR35:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rdm" }
 //.
 // 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" }



More information about the cfe-commits mailing list