[PATCH] D45808: [OpenCL] Add 'denorms-are-zero' function attribute

[Kristina Bessonova via Phabricator via cfe-commits]

krisb created this revision.
krisb added reviewers: Anastasia, yaxunl.
Herald added a subscriber: cfe-commits.

Generate attribute 'denorms-are-zero'='true' if '-cl-denorms-are-zero'
compile option was specified and 'denorms-are-zero'='false' otherwise.


Repository:
  rC Clang

https://reviews.llvm.org/D45808

Files:
  lib/CodeGen/CGCall.cpp
  test/CodeGenOpenCL/denorms-are-zero.cl


Index: test/CodeGenOpenCL/denorms-are-zero.cl
===================================================================
--- test/CodeGenOpenCL/denorms-are-zero.cl
+++ test/CodeGenOpenCL/denorms-are-zero.cl
@@ -1,19 +1,20 @@
-// RUN: %clang_cc1 -S -cl-denorms-are-zero -o - %s 2>&1
-// RUN: %clang_cc1 -emit-llvm -cl-denorms-are-zero -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck %s
-// RUN: %clang_cc1 -emit-llvm -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck %s --check-prefix=CHECK-DENORM
-// RUN: %clang_cc1 -emit-llvm -target-feature +fp32-denormals -target-feature -fp64-fp16-denormals -cl-denorms-are-zero -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck --check-prefix=CHECK-FEATURE %s
+// RUN: %clang_cc1 -emit-llvm -cl-denorms-are-zero -o - %s | FileCheck %s
+// RUN: %clang_cc1 -emit-llvm -cl-denorms-are-zero -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck %s --check-prefix=AMDGCN
+// RUN: %clang_cc1 -emit-llvm -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck %s --check-prefix=AMDGCN-DENORM
+// RUN: %clang_cc1 -emit-llvm -target-feature +fp32-denormals -target-feature -fp64-fp16-denormals -cl-denorms-are-zero -o - -triple amdgcn--amdhsa -target-cpu fiji %s | FileCheck --check-prefix=AMDGCN-FEATURE %s
 
-// For non-amdgcn targets, this test just checks that the -cl-denorms-are-zero argument is accepted
-// by clang.  This option is currently a no-op, which is allowed by the
-// OpenCL specification.
+// For all targets 'denorms-are-zero' attribute is set to 'true'
+
+// CHECK-LABEL: define void @f()
+// CHECK: attributes #{{[0-9]*}} = {{{[^}]*}} "denorms-are-zero"="true"
 
 // For amdgcn target cpu fiji, fp32 should be flushed since fiji does not support fp32 denormals, unless +fp32-denormals is
 // explicitly set. amdgcn target always do not flush fp64 denormals. The control for fp64 and fp16 denormals is the same.
 
-// CHECK-DENORM-LABEL: define void @f()
-// CHECK-DENORM: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp64-fp16-denormals,{{[^"]*}}-fp32-denormals{{[^"]*}}"
-// CHECK-LABEL: define void @f()
-// CHECK: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp64-fp16-denormals,{{[^"]*}}-fp32-denormals{{[^"]*}}"
-// CHECK-FEATURE-LABEL: define void @f()
-// CHECK-FEATURE: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp32-denormals,{{[^"]*}}-fp64-fp16-denormals{{[^"]*}}"
+// AMDGCN-LABEL: define void @f()
+// AMDGCN: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp64-fp16-denormals,{{[^"]*}}-fp32-denormals{{[^"]*}}"
+// AMDGCN-DENORM-LABEL: define void @f()
+// AMDGCN-DENORM: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp64-fp16-denormals,{{[^"]*}}-fp32-denormals{{[^"]*}}"
+// AMDGCN-FEATURE-LABEL: define void @f()
+// AMDGCN-FEATURE: attributes #{{[0-9]*}} = {{{[^}]*}} "target-features"="{{[^"]*}}+fp32-denormals,{{[^"]*}}-fp64-fp16-denormals{{[^"]*}}"
 void f() {}
Index: lib/CodeGen/CGCall.cpp
===================================================================
--- lib/CodeGen/CGCall.cpp
+++ lib/CodeGen/CGCall.cpp
@@ -1745,6 +1745,10 @@
         "correctly-rounded-divide-sqrt-fp-math",
         llvm::toStringRef(CodeGenOpts.CorrectlyRoundedDivSqrt));
 
+    if (getLangOpts().OpenCL)
+      FuncAttrs.addAttribute("denorms-are-zero",
+                             llvm::toStringRef(CodeGenOpts.FlushDenorm));
+
     // TODO: Reciprocal estimate codegen options should apply to instructions?
     const std::vector<std::string> &Recips = CodeGenOpts.Reciprocals;
     if (!Recips.empty())


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D45808.143065.patch
Type: text/x-patch
Size: 3600 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20180419/2b68004b/attachment-0001.bin>