llc: Change behavior of -mcpu with existing attribute

[arsenm]

Summary

Don't overwrite existing target-cpu attributes.

I've often found the replacement behavior annoying, and this is
inconsistent with how the fast math command line flags interact with
the function attributes.

Does not yet change target-features, since I think that should behave
as a concatenation.

[craig.topper]

X86 changes look good to me. I've also found the previous behavior strange.

[arsenm]

Rebase

[craig.topper]

LGTM

[arsenm]

f26ed6e47cb8b080c236d11c4942a12265180084

llvm/include/llvm/CodeGen/CommandFlags.inc

 /// Set function attributes of function \p F based on CPU, Features, and command
 /// line flags.
 LLVM_ATTRIBUTE_UNUSED static void
 setFunctionAttributes(StringRef CPU, StringRef Features, Function &F) {
   auto &Ctx = F.getContext();
   AttributeList Attrs = F.getAttributes();
   AttrBuilder NewAttrs;
 
-  if (!CPU.empty())
+  if (!CPU.empty() && !F.hasFnAttribute("target-cpu"))
     NewAttrs.addAttribute("target-cpu", CPU);
   if (!Features.empty())
     NewAttrs.addAttribute("target-features", Features);
   if (FramePointerUsage.getNumOccurrences() > 0) {
     if (FramePointerUsage == llvm::FramePointer::All)
       NewAttrs.addAttribute("frame-pointer", "all");
     else if (FramePointerUsage == llvm::FramePointer::NonLeaf)
       NewAttrs.addAttribute("frame-pointer", "non-leaf");

llvm/test/CodeGen/AMDGPU/memory-legalizer-atomic-insert-end.mir

-# RUN: llc -march=amdgcn -mcpu=gfx803 -run-pass si-memory-legalizer  %s -o - | FileCheck %s
+# RUN: llc -march=amdgcn -run-pass=si-memory-legalizer  %s -o - | FileCheck %s
 
 --- |
   declare i32 @llvm.amdgcn.workitem.id.x() #0
 
   define amdgpu_kernel void @atomic_max_i32_noret(
       i32 addrspace(1)* %out,
       i32 addrspace(1)* addrspace(1)* %in,
       i32 addrspace(1)* %x,
exit:                                             ; preds = %atomic, %0
     call void @llvm.amdgcn.end.cf(i64 %3)
     ret void
   }
 
   declare { i1, i64 } @llvm.amdgcn.if(i1)
 
   declare void @llvm.amdgcn.end.cf(i64)
 
-  ; Function Attrs: nounwind
-  declare void @llvm.stackprotector(i8*, i8**) #3
-
-  attributes #0 = { nounwind readnone "target-cpu"="tahiti" }
-  attributes #1 = { nounwind "target-cpu"="tahiti" }
-  attributes #2 = { readnone }
-  attributes #3 = { nounwind }
+  attributes #0 = { nounwind readnone }
+  attributes #1 = { nounwind "target-cpu"="gfx803" }
 
 ...
 ---
 
 # CHECK-LABEL: name: atomic_max_i32_noret
 
 # CHECK-LABEL: bb.1.atomic:
 # CHECK:       BUFFER_ATOMIC_SMAX_ADDR64

llvm/test/CodeGen/MIR/AMDGPU/llc-target-cpu-attr-from-cmdline-ir.mir

 # RUN: llc -march=amdgcn -mcpu=hawaii -run-pass=none -o - %s | FileCheck -check-prefix=MCPU %s
 # RUN: llc -march=amdgcn -mattr=+unaligned-buffer-access -run-pass=none -o - %s | FileCheck -check-prefix=MATTR %s
 
 # FIXME: This overrides attributes that already are present. It should probably
 # only touch functions without an existing attribute.
 
-# MCPU: attributes #0 = { "target-cpu"="hawaii" }
-# MCPU-NOT: attributes #1
+# MCPU:  @with_cpu_attr() #0 {
+# MCPU: @no_cpu_attr() #1 {
+
+# MCPU: attributes #0 = { "target-cpu"="fiji" }
+# MCPU: attributes #1 = { "target-cpu"="hawaii" }
 
 # MATTR:  attributes #0 = { "target-cpu"="fiji" "target-features"="+unaligned-buffer-access" }
 # MATTR: attributes #1 = { "target-features"="+unaligned-buffer-access" }
 
 --- |
   define amdgpu_kernel void @with_cpu_attr() #0 {
     ret void
   }

llvm/test/CodeGen/X86/avoid-sfb-overlaps.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-linux -verify-machineinstrs | FileCheck %s -check-prefix=CHECK
-; RUN: llc < %s -mtriple=x86_64-linux --x86-disable-avoid-SFB -verify-machineinstrs | FileCheck %s --check-prefix=DISABLED
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=x86-64 -verify-machineinstrs | FileCheck %s -check-prefix=CHECK
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=x86-64 --x86-disable-avoid-SFB -verify-machineinstrs | FileCheck %s --check-prefix=DISABLED
 ; RUN: llc < %s -mtriple=x86_64-linux -mcpu=core-avx2 -verify-machineinstrs | FileCheck %s -check-prefix=CHECK-AVX2
 ; RUN: llc < %s -mtriple=x86_64-linux -mcpu=skx -verify-machineinstrs | FileCheck %s -check-prefix=CHECK-AVX512
 
 ; ModuleID = '../testSFB/testOverlapBlocks.c'
 source_filename = "../testSFB/testOverlapBlocks.c"
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
entry:
   store i8 %conv2, i8* %add.ptr3, align 1
   %add.ptr4 = getelementptr inbounds i8, i8* %A, i64 -11
   store i8 0, i8* %add.ptr4, align 1
   %add.ptr5 = getelementptr inbounds i8, i8* %A, i64 16
   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 4 %add.ptr5, i8* nonnull align 4 %add.ptr, i64 16, i1 false)
   ret void
 }
 
-attributes #0 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
-attributes #1 = { argmemonly nounwind }
-
+attributes #0 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
 
+attributes #1 = { argmemonly nounwind }

llvm/test/CodeGen/X86/avoid-sfb.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-linux -verify-machineinstrs | FileCheck %s -check-prefix=CHECK
-; RUN: llc < %s -mtriple=x86_64-linux --x86-disable-avoid-SFB -verify-machineinstrs | FileCheck %s --check-prefix=DISABLED
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=x86-64 -verify-machineinstrs | FileCheck %s -check-prefix=CHECK
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=x86-64 --x86-disable-avoid-SFB -verify-machineinstrs | FileCheck %s --check-prefix=DISABLED
 ; RUN: llc < %s -mtriple=x86_64-linux -mcpu=core-avx2 -verify-machineinstrs | FileCheck %s -check-prefix=CHECK-AVX2
 ; RUN: llc < %s -mtriple=x86_64-linux -mcpu=skx -verify-machineinstrs | FileCheck %s -check-prefix=CHECK-AVX512
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 %struct.S = type { i32, i32, i32, i32 }
 
if.end:                                           ; preds = %if.then, %entry
   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 4, i1 false)
   ret void
 }
 
 
 declare void @bar(%struct.S*) local_unnamed_addr #1
 
 
-; Function Attrs: argmemonly nounwind
-declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1) #1
-
-attributes #0 = { nounwind uwtable "target-cpu"="x86-64" }
-
 %struct.S7 = type { float, float, float , float, float, float, float, float }
 
 ; Function Attrs: nounwind uwtable
 define void @test_conditional_block_float(%struct.S7* nocapture noalias %s1, %struct.S7* nocapture %s2, i32 %x, %struct.S7* nocapture %s3, %struct.S7* nocapture readonly %s4, float %y) local_unnamed_addr #0 {
 ; CHECK-LABEL: test_conditional_block_float:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    cmpl $18, %edx
 ; CHECK-NEXT:    jl .LBB11_2

 entry:
   %a = bitcast i8* %A to i32*
   store i32 %x, i32* %a, align 4
   %add.ptr = getelementptr inbounds i8, i8* %A, i64 20
   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 4 %add.ptr, i8* align 4 %A, i64 16, i32 4, i1 false)
   ret void
 }
 
+; Function Attrs: argmemonly nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1) #1
 
-
+attributes #0 = { nounwind uwtable }

llvm/test/CodeGen/X86/llc-override-mcpu-mattr.ll

 ; RUN: llc < %s -mtriple=x86_64-- -mcpu=broadwell | FileCheck %s
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s
 
-; Check that llc can overide function attributes target-cpu and target-features
+; Check that llc can set function attributes target-cpu and target-features
 ; using command line options -mcpu and -mattr.
 
 ; CHECK: vpsadbw (%r{{si|dx}}), %ymm{{[0-9]+}}, %ymm{{[0-9]+}}
 
-define <4 x i64> @foo1(<4 x i64>* %s1, <4 x i64>* %s2) #0 {
+define <4 x i64> @foo1(<4 x i64>* %s1, <4 x i64>* %s2) {
 entry:
   %ps1 = load <4 x i64>, <4 x i64>* %s1
   %ps2 = load <4 x i64>, <4 x i64>* %s2
   %0 = bitcast <4 x i64> %ps1 to <32 x i8>
   %1 = bitcast <4 x i64> %ps2 to <32 x i8>
   %2 = tail call <4 x i64> @llvm.x86.avx2.psad.bw(<32 x i8> %0, <32 x i8> %1)
   ret <4 x i64> %2
 }
 
 declare <4 x i64> @llvm.x86.avx2.psad.bw(<32 x i8>, <32 x i8>)
-
-attributes #0 = { "target-cpu"="core2" "target-features"="+ssse3,+cx16,+sse4.2,+sse4.1,+sse,+sse2,+sse3,+avx,+popcnt" }

llvm/test/DebugInfo/COFF/inlining-files.ll

entry:
   %inc3.i = add nsw i32 %4, 1, !dbg !27
   store volatile i32 %inc3.i, i32* @x, align 4, !dbg !27, !tbaa !16
   %5 = load volatile i32, i32* @x, align 4, !dbg !29, !tbaa !16
   %inc1 = add nsw i32 %5, 1, !dbg !29
   store volatile i32 %inc1, i32* @x, align 4, !dbg !29, !tbaa !16
   ret void, !dbg !30
 }
 
-attributes #0 = { norecurse nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #0 = { norecurse nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-features"="+fxsr,+mmx,+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" }
 
 !llvm.dbg.cu = !{!2}
 !llvm.module.flags = !{!8, !9, !10}
 !llvm.ident = !{!11}
 
 !0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
 !1 = !DIGlobalVariable(name: "x", scope: !2, file: !3, line: 1, type: !6, isLocal: false, isDefinition: true)
 !2 = distinct !DICompileUnit(language: DW_LANG_C99, file: !3, producer: "clang version 3.9.0 ", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !4, globals: !5)

llvm/test/DebugInfo/COFF/inlining-header.ll

entry:
   store volatile i32 %add1.i, i32* @"\01?x@@3HC", align 4, !dbg !29, !tbaa !16
   %4 = load volatile i32, i32* @"\01?x@@3HC", align 4, !dbg !30, !tbaa !16
   %add1 = add nsw i32 %4, 1, !dbg !30
   store volatile i32 %add1, i32* @"\01?x@@3HC", align 4, !dbg !30, !tbaa !16
   %5 = load volatile i32, i32* @"\01?x@@3HC", align 4, !dbg !31, !tbaa !16
   ret i32 %5, !dbg !32
 }
 
-attributes #0 = { norecurse nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #0 = { norecurse nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "stack-protector-buffer-size"="8" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
 
 !llvm.dbg.cu = !{!2}
 !llvm.module.flags = !{!8, !9, !10}
 !llvm.ident = !{!11}
 
 !0 = distinct !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
 !1 = !DIGlobalVariable(name: "x", linkageName: "\01?x@@3HC", scope: !2, file: !3, line: 1, type: !6, isLocal: false, isDefinition: true)
 !2 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !3, producer: "clang version 3.9.0 (trunk 275430) (llvm/trunk 275433)", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !4, globals: !5)

llvm/test/Other/opt-override-mcpu-mattr.ll

 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -mcpu=broadwell -mattr=+avx2 -S | FileCheck %s
 
-; Check that opt can rewrite function attributes target-cpu and target-features
-; using command line options -mcpu and -mattr.
+; Check that opt can add but not rewrite function attributes
+; target-cpu and target-features using command line options -mcpu and
+; -mattr.
 
 ; CHECK: attributes #0 = { nounwind readnone ssp uwtable "target-cpu"="broadwell" "target-features"="+avx2" "use-soft-float"="false" }
+; CHECK: attributes #1 = { nounwind readnone ssp uwtable "target-cpu"="core2" "target-features"="+avx2" "use-soft-float"="false" }
 
-define i32 @foo1() #0 {
+define i32 @no_target_cpu() #0 {
 entry:
   ret i32 0
 }
 
-attributes #0 = { nounwind readnone ssp uwtable "target-cpu"="core2" "target-features"="+ssse3,+cx16,+sse,+sse2,+sse3" "use-soft-float"="false" }
+define i32 @has_targe_cpu() #1 {
+entry:
+  ret i32 0
+}
+
+attributes #0 = { nounwind readnone ssp uwtable "target-features"="+ssse3,+cx16,+sse,+sse2,+sse3" "use-soft-float"="false" }
+attributes #1 = { nounwind readnone ssp uwtable "target-cpu"="core2" "target-features"="+ssse3,+cx16,+sse,+sse2,+sse3" "use-soft-float"="false" }

llvm/test/Transforms/LoopVectorize/X86/tail_folding_and_assume_safety.ll

 
 ;CHECK-LABEL: @fold_tail
 ;CHECK: vector.body:
 ;CHECK: call <8 x i32> @llvm.masked.load
 ;CHECK: call <8 x i32> @llvm.masked.load
 ;CHECK: call void @llvm.masked.store
 
 ; Function Attrs: nofree norecurse nounwind uwtable
-define dso_local void @fold_tail(i32* noalias nocapture %p, i32* noalias nocapture readonly %q1, i32* noalias nocapture readonly %q2, 
+define dso_local void @fold_tail(i32* noalias nocapture %p, i32* noalias nocapture readonly %q1, i32* noalias nocapture readonly %q2,
 i32 %guard) local_unnamed_addr #0 {
 entry:
   %0 = sext i32 %guard to i64
   br label %for.body
 
 for.cond.cleanup:
   ret void
 

 
 ;CHECK-LABEL: @fold_tail_and_assume_safety
 ;CHECK: vector.body:
 ;CHECK: call <8 x i32> @llvm.masked.load
 ;CHECK: call <8 x i32> @llvm.masked.load
 ;CHECK: call void @llvm.masked.store
 
 ; Function Attrs: nofree norecurse nounwind uwtable
-define dso_local void @fold_tail_and_assume_safety(i32* noalias nocapture %p, i32* noalias nocapture readonly %q1, i32* noalias nocapture readonly %q2, 
+define dso_local void @fold_tail_and_assume_safety(i32* noalias nocapture %p, i32* noalias nocapture readonly %q1, i32* noalias nocapture readonly %q2,
 i32 %guard) local_unnamed_addr #0 {
 entry:
   %0 = sext i32 %guard to i64
   br label %for.body
 
 for.cond.cleanup:
   ret void
 
if.then:
   br label %for.inc
 
 for.inc:
   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
   %exitcond = icmp eq i64 %indvars.iv.next, 1021
   br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !11
 }
 
-attributes #0 = { norecurse nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #0 = { norecurse nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
 
 !llvm.module.flags = !{!0}
 !llvm.ident = !{!1}
 
 !0 = !{i32 1, !"wchar_size", i32 4}
 !1 = !{!"clang version 6.0.0-1ubuntu2 (tags/RELEASE_600/final)"}
 !2 = !{!3, !3, i64 0}
 !3 = !{!"int", !4, i64 0}

llvm/test/tools/llvm-objdump/AMDGPU/source-lines.ll

entry:
   %3 = load i32 addrspace(1)*, i32 addrspace(1)* addrspace(5)* %Out.addr, align 4, !dbg !29
   store i32 %2, i32 addrspace(1)* %3, align 4, !dbg !30
   ret void, !dbg !31
 }
 
 ; Function Attrs: nounwind readnone
 declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
 
-attributes #0 = { noinline nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="gfx700" "target-features"="+fp64-fp16-denormals,-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #0 = { noinline nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-features"="+fp64-fp16-denormals,-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
 attributes #1 = { nounwind readnone }
 
 !llvm.dbg.cu = !{!0}
 !opencl.ocl.version = !{!3}
 !llvm.module.flags = !{!4, !5}
 !llvm.ident = !{!6}
 
 !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 5.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2)