[clang][codegen] Remove the insertion of `correctly-rounded-divide-sqrt-fp-math` fn-attr.

[hliao]

Summary

• -cl-fp32-correctly-rounded-divide-sqrt is already handled in a per-instruction manner to annotate the accuracy required. There's no need to add that fn-attr. So far, there's no backend handling that attr and that option is an OpenCL-specific one, which should not be inserted everywhere.

[hliao]

Remove the irrelevant change on .clang-format.

[bader]

I'm okay with this change in general. One note though.

There are two aspects of this change:

1. I agree that cl-fp32-correctly-rounded-divide-sqrt is OpenCL specific and nothing should not be added in other modes.
2. I have concerns WRT removing the attribute in OpenCL mode. Most of the OpenCL back-ends are out-of-tree and might still rely on a function attribute. In addition to that https://reviews.llvm.org/D22940 handles division part of the requirement, but I don't know if sqrt computation will be handled correctly.

Can we split the patch into two: first to add attribute only in OpenCL language mode and second one removing function attribute in OpenCL mode as well?
In case we need to rollback the patch, I expect no issues with the first part.

[hliao]

Split the original into 2. This is the first part, which add
correctly-rounded-device-sqrt-fp-math for OpenCL only. The second part will
remove that attribute annotating completely.

[bader]

Thanks!

[hliao]

In D88303#2298259, @bader wrote:

Thanks!

Part 2 is submitted @ https://reviews.llvm.org/D88424

clang/lib/CodeGen/CGCall.cpp

if (AttrOnCallSite) {
     FuncAttrs.addAttribute("unsafe-fp-math",
                            llvm::toStringRef(LangOpts.UnsafeFPMath));
     FuncAttrs.addAttribute("use-soft-float",
                            llvm::toStringRef(CodeGenOpts.SoftFloat));
     FuncAttrs.addAttribute("stack-protector-buffer-size",
                            llvm::utostr(CodeGenOpts.SSPBufferSize));
     FuncAttrs.addAttribute("no-signed-zeros-fp-math",
                            llvm::toStringRef(LangOpts.NoSignedZero));
-    FuncAttrs.addAttribute(
-        "correctly-rounded-divide-sqrt-fp-math",
-        llvm::toStringRef(CodeGenOpts.CorrectlyRoundedDivSqrt));
 
     // TODO: Reciprocal estimate codegen options should apply to instructions?
     const std::vector<std::string> &Recips = CodeGenOpts.Reciprocals;
     if (!Recips.empty())
       FuncAttrs.addAttribute("reciprocal-estimates",
                              llvm::join(Recips, ","));
 
     if (!CodeGenOpts.PreferVectorWidth.empty() &&

clang/test/CodeGen/complex-builtins.c

 // NO__ERRNO: declare { double, double } @ctanh(double, double) [[READNONE]]
 // NO__ERRNO: declare <2 x float> @ctanhf(<2 x float>) [[READNONE]]
 // NO__ERRNO: declare { x86_fp80, x86_fp80 } @ctanhl({ x86_fp80, x86_fp80 }* byval({ x86_fp80, x86_fp80 }) align 16) [[NOT_READNONE]]
 // HAS_ERRNO: declare { double, double } @ctanh(double, double) [[NOT_READNONE]]
 // HAS_ERRNO: declare <2 x float> @ctanhf(<2 x float>) [[NOT_READNONE]]
 // HAS_ERRNO: declare { x86_fp80, x86_fp80 } @ctanhl({ x86_fp80, x86_fp80 }* byval({ x86_fp80, x86_fp80 }) align 16) [[NOT_READNONE]]
 };
 
-
 // NO__ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
-// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 
-// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 // HAS_ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
-

clang/test/CodeGen/complex-libcalls.c

 // NO__ERRNO: declare { double, double } @ctanh(double, double) [[READNONE]]
 // NO__ERRNO: declare <2 x float> @ctanhf(<2 x float>) [[READNONE]]
 // NO__ERRNO: declare { x86_fp80, x86_fp80 } @ctanhl({ x86_fp80, x86_fp80 }* byval({ x86_fp80, x86_fp80 }) align 16) [[NOT_READNONE]]
 // HAS_ERRNO: declare { double, double } @ctanh(double, double) [[NOT_READNONE]]
 // HAS_ERRNO: declare <2 x float> @ctanhf(<2 x float>) [[NOT_READNONE]]
 // HAS_ERRNO: declare { x86_fp80, x86_fp80 } @ctanhl({ x86_fp80, x86_fp80 }* byval({ x86_fp80, x86_fp80 }) align 16) [[NOT_READNONE]]
 };
 
-
 // NO__ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
-// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 
-// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 // HAS_ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
-

clang/test/CodeGen/math-builtins.c

 // NO__ERRNO: declare double @llvm.trunc.f64(double) [[READNONE_INTRINSIC]]
 // NO__ERRNO: declare float @llvm.trunc.f32(float) [[READNONE_INTRINSIC]]
 // NO__ERRNO: declare x86_fp80 @llvm.trunc.f80(x86_fp80) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @llvm.trunc.f64(double) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare float @llvm.trunc.f32(float) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare x86_fp80 @llvm.trunc.f80(x86_fp80) [[READNONE_INTRINSIC]]
 };
 
-
 // NO__ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
 // NO__ERRNO: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
-// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 // NO__ERRNO: attributes [[PURE]] = { {{.*}}readonly{{.*}} }
 
-// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 // HAS_ERRNO: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
 // HAS_ERRNO: attributes [[PURE]] = { {{.*}}readonly{{.*}} }
 // HAS_ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
 
 // HAS_ERRNO_GNU: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
 // HAS_ERRNO_ANDROID: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
 // HAS_ERRNO_WIN: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
 

clang/test/CodeGen/math-libcalls.c

 // NO__ERRNO: declare double @llvm.trunc.f64(double) [[READNONE_INTRINSIC]]
 // NO__ERRNO: declare float @llvm.trunc.f32(float) [[READNONE_INTRINSIC]]
 // NO__ERRNO: declare x86_fp80 @llvm.trunc.f80(x86_fp80) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @llvm.trunc.f64(double) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare float @llvm.trunc.f32(float) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare x86_fp80 @llvm.trunc.f80(x86_fp80) [[READNONE_INTRINSIC]]
 };
 
-
 // NO__ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
 // NO__ERRNO: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
-// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// NO__ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 // NO__ERRNO: attributes [[READONLY]] = { {{.*}}readonly{{.*}} }
 
-// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind "correctly{{.*}} }
+// HAS_ERRNO: attributes [[NOT_READNONE]] = { nounwind {{.*}} }
 // HAS_ERRNO: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
 // HAS_ERRNO: attributes [[READONLY]] = { {{.*}}readonly{{.*}} }
 // HAS_ERRNO: attributes [[READNONE]] = { {{.*}}readnone{{.*}} }
 
 // HAS_ERRNO_GNU: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }
 // HAS_ERRNO_WIN: attributes [[READNONE_INTRINSIC]] = { {{.*}}readnone{{.*}} }

clang/test/CodeGenOpenCL/amdgpu-attrs.cl

 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64_WAVES_PER_EU_2_NUM_SGPR_32]] = {{.*}} "amdgpu-flat-work-group-size"="32,64" "amdgpu-implicitarg-num-bytes"="56" "amdgpu-num-sgpr"="32" "amdgpu-waves-per-eu"="2"
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64_WAVES_PER_EU_2_NUM_VGPR_64]] = {{.*}} "amdgpu-flat-work-group-size"="32,64" "amdgpu-implicitarg-num-bytes"="56" "amdgpu-num-vgpr"="64" "amdgpu-waves-per-eu"="2"
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64_WAVES_PER_EU_2_4_NUM_SGPR_32]] = {{.*}} "amdgpu-flat-work-group-size"="32,64" "amdgpu-implicitarg-num-bytes"="56" "amdgpu-num-sgpr"="32" "amdgpu-waves-per-eu"="2,4"
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64_WAVES_PER_EU_2_4_NUM_VGPR_64]] = {{.*}} "amdgpu-flat-work-group-size"="32,64" "amdgpu-implicitarg-num-bytes"="56" "amdgpu-num-vgpr"="64" "amdgpu-waves-per-eu"="2,4"
 
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64_WAVES_PER_EU_2_NUM_SGPR_32_NUM_VGPR_64]] = {{.*}} "amdgpu-flat-work-group-size"="32,64" "amdgpu-implicitarg-num-bytes"="56" "amdgpu-num-sgpr"="32" "amdgpu-num-vgpr"="64" "amdgpu-waves-per-eu"="2"
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64_WAVES_PER_EU_2_4_NUM_SGPR_32_NUM_VGPR_64]] = {{.*}} "amdgpu-flat-work-group-size"="32,64" "amdgpu-implicitarg-num-bytes"="56" "amdgpu-num-sgpr"="32" "amdgpu-num-vgpr"="64" "amdgpu-waves-per-eu"="2,4"
 
-// CHECK-DAG: attributes [[A_FUNCTION]] = {{.*}} "correctly-rounded-divide-sqrt-fp-math"="false"
+// CHECK-DAG: attributes [[A_FUNCTION]] = {{.*}}
 // CHECK-DAG: attributes [[DEFAULT_KERNEL_ATTRS]] = {{.*}} "amdgpu-flat-work-group-size"="1,256" "amdgpu-implicitarg-num-bytes"="56"

clang/test/CodeGenOpenCL/fpmath.cl

 // RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck --check-prefix=CHECK --check-prefix=NODIVOPT %s
 // RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown -cl-fp32-correctly-rounded-divide-sqrt | FileCheck --check-prefix=CHECK --check-prefix=DIVOPT %s
 // RUN: %clang_cc1 %s -emit-llvm -o - -DNOFP64 -cl-std=CL1.2 -triple r600-unknown-unknown -target-cpu r600 -pedantic | FileCheck --check-prefix=CHECK-FLT %s
 // RUN: %clang_cc1 %s -emit-llvm -o - -DFP64 -cl-std=CL1.2 -triple spir-unknown-unknown -pedantic | FileCheck --check-prefix=CHECK-DBL %s
 
 typedef __attribute__(( ext_vector_type(4) )) float float4;
 
 float spscalardiv(float a, float b) {
   // CHECK: @spscalardiv
-  // CHECK: #[[ATTR:[0-9]+]]
   // CHECK: fdiv{{.*}},
   // NODIVOPT: !fpmath ![[MD:[0-9]+]]
   // DIVOPT-NOT: !fpmath ![[MD:[0-9]+]]
   return a / b;
 }
 
 float4 spvectordiv(float4 a, float4 b) {
   // CHECK: @spvectordiv
-  // CHECK: #[[ATTR2:[0-9]+]]
   // CHECK: fdiv{{.*}},
   // NODIVOPT: !fpmath ![[MD]]
   // DIVOPT-NOT: !fpmath ![[MD]]
   return a / b;
 }
 
 #if __OPENCL_C_VERSION__ >=120
 void printf(constant char* fmt, ...);
 
 void testdbllit(long *val) {
   // CHECK-FLT: float 2.000000e+01
   // CHECK-DBL: double 2.000000e+01
   printf("%f", 20.0);
 }
 
 #endif
 
 #ifndef NOFP64
 #pragma OPENCL EXTENSION cl_khr_fp64 : enable
 double dpscalardiv(double a, double b) {
   // CHECK: @dpscalardiv
-  // CHECK: #[[ATTR]]
   // CHECK-NOT: !fpmath
   return a / b;
 }
 #endif
 
-// CHECK: attributes #[[ATTR]] = {
-// NODIVOPT-SAME: "correctly-rounded-divide-sqrt-fp-math"="false"
-// DIVOPT-SAME: "correctly-rounded-divide-sqrt-fp-math"="true"
-// CHECK-SAME: }
-// CHECK: attributes #[[ATTR2]] = {
-// NODIVOPT-SAME: "correctly-rounded-divide-sqrt-fp-math"="false"
-// DIVOPT-SAME: "correctly-rounded-divide-sqrt-fp-math"="true"
-// CHECK-SAME: }
 // NODIVOPT: ![[MD]] = !{float 2.500000e+00}