yaxunl created this revision.
yaxunl added reviewers: Anastasia, bader, pxli168.
yaxunl added subscribers: cfe-commits, tstellarAMD.
This patch attempts to use target-dependent function attribute to represent
reqd_work_group_size, work_group_size_hint and vector_type_hint kernel
attributes and use function metadata to represent kernel argument info.
This is to elicit further discussion about a better way to represent kernel
attributes.
The cfe-dev discussion is at
http://lists.llvm.org/pipermail/cfe-dev/2016-June/049223.html
http://reviews.llvm.org/D20979
Files:
lib/CodeGen/CodeGenFunction.cpp
test/CodeGenOpenCL/kernel-arg-info.cl
test/CodeGenOpenCL/kernel-attributes.cl
test/CodeGenOpenCL/kernel-metadata.cl
Index: test/CodeGenOpenCL/kernel-metadata.cl
===================================================================
--- test/CodeGenOpenCL/kernel-metadata.cl
+++ test/CodeGenOpenCL/kernel-metadata.cl
@@ -6,10 +6,5 @@
__kernel void kernel_function() {
}
-// CHECK: !opencl.kernels = !{!0}
-// CHECK: !0 = !{void ()* @kernel_function, !1, !2, !3, !4, !5}
-// CHECK: !1 = !{!"kernel_arg_addr_space"}
-// CHECK: !2 = !{!"kernel_arg_access_qual"}
-// CHECK: !3 = !{!"kernel_arg_type"}
-// CHECK: !4 = !{!"kernel_arg_base_type"}
-// CHECK: !5 = !{!"kernel_arg_type_qual"}
+// CHECK: define void @kernel_function() #{{[0-9]+}} !kernel_arg_addr_space ![[MD:[0-9]+]] !kernel_arg_access_qual ![[MD]] !kernel_arg_type ![[MD]] !kernel_arg_base_type ![[MD]] !kernel_arg_type_qual ![[MD]] {
+// CHECK: ![[MD]] = !{}
Index: test/CodeGenOpenCL/kernel-attributes.cl
===================================================================
--- test/CodeGenOpenCL/kernel-attributes.cl
+++ test/CodeGenOpenCL/kernel-attributes.cl
@@ -3,14 +3,14 @@
typedef unsigned int uint4 __attribute__((ext_vector_type(4)));
kernel __attribute__((vec_type_hint(int))) __attribute__((reqd_work_group_size(1,2,4))) void kernel1(int a) {}
+// CHECK: define void @kernel1(i32 %a) #[[AT1:[0-9]+]]
kernel __attribute__((vec_type_hint(uint4))) __attribute__((work_group_size_hint(8,16,32))) void kernel2(int a) {}
+// CHECK: define void @kernel2(i32 %a) #[[AT2:[0-9]+]]
-// CHECK: opencl.kernels = !{[[MDNODE0:![0-9]+]], [[MDNODE3:![0-9]+]]}
-
-// CHECK: [[MDNODE0]] = !{void (i32)* @kernel1, {{.*}} [[MDNODE1:![0-9]+]], [[MDNODE2:![0-9]+]]}
-// CHECK: [[MDNODE1]] = !{!"vec_type_hint", i32 undef, i32 1}
-// CHECK: [[MDNODE2]] = !{!"reqd_work_group_size", i32 1, i32 2, i32 4}
-// CHECK: [[MDNODE3]] = !{void (i32)* @kernel2, {{.*}} [[MDNODE4:![0-9]+]], [[MDNODE5:![0-9]+]]}
-// CHECK: [[MDNODE4]] = !{!"vec_type_hint", <4 x i32> undef, i32 0}
-// CHECK: [[MDNODE5]] = !{!"work_group_size_hint", i32 8, i32 16, i32 32}
+// CHECK: attributes #[[AT1]] = {
+// CHECK-DAG: "reqd_work_group_size"="1 2 4"
+// CHECK-DAG: "vec_type_hint"="int"
+// CHECK: attributes #[[AT2]] = {
+// CHECK-DAG: "vec_type_hint"="uint4"
+// CHECK-DAG: "work_group_size_hint"="8 16 32"
Index: test/CodeGenOpenCL/kernel-arg-info.cl
===================================================================
--- test/CodeGenOpenCL/kernel-arg-info.cl
+++ test/CodeGenOpenCL/kernel-arg-info.cl
@@ -1,55 +1,88 @@
-// RUN: %clang_cc1 %s -cl-kernel-arg-info -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s -check-prefix ARGINFO
-// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s -check-prefix NO-ARGINFO
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown -cl-kernel-arg-info | FileCheck %s -check-prefix ARGINFO
kernel void foo(__global int * restrict X, const int Y,
volatile int anotherArg, __constant float * restrict Z) {
*X = Y + anotherArg;
}
-
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 0, i32 0, i32 2}
-// CHECK: !{!"kernel_arg_access_qual", !"none", !"none", !"none", !"none"}
-// CHECK: !{!"kernel_arg_type", !"int*", !"int", !"int", !"float*"}
-// CHECK: !{!"kernel_arg_base_type", !"int*", !"int", !"int", !"float*"}
-// CHECK: !{!"kernel_arg_type_qual", !"restrict", !"const", !"volatile", !"restrict const"}
-// ARGINFO: !{!"kernel_arg_name", !"X", !"Y", !"anotherArg", !"Z"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"X", !"Y", !"anotherArg", !"Z"}
+// CHECK: define spir_kernel void @foo{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD11:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD12:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD13:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD13]]
+// CHECK: !kernel_arg_type_qual ![[MD14:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD15:[0-9]+]]
kernel void foo2(read_only image1d_t img1, image2d_t img2, write_only image2d_array_t img3) {
}
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 1, i32 1}
-// CHECK: !{!"kernel_arg_access_qual", !"read_only", !"read_only", !"write_only"}
-// CHECK: !{!"kernel_arg_type", !"image1d_t", !"image2d_t", !"image2d_array_t"}
-// CHECK: !{!"kernel_arg_base_type", !"image1d_t", !"image2d_t", !"image2d_array_t"}
-// CHECK: !{!"kernel_arg_type_qual", !"", !"", !""}
-// ARGINFO: !{!"kernel_arg_name", !"img1", !"img2", !"img3"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"img1", !"img2", !"img3"}
+// CHECK: define spir_kernel void @foo2{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD21:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD22:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD23:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD23]]
+// CHECK: !kernel_arg_type_qual ![[MD24:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD25:[0-9]+]]
kernel void foo3(__global half * X) {
}
-// CHECK: !{!"kernel_arg_addr_space", i32 1}
-// CHECK: !{!"kernel_arg_access_qual", !"none"}
-// CHECK: !{!"kernel_arg_type", !"half*"}
-// CHECK: !{!"kernel_arg_base_type", !"half*"}
-// CHECK: !{!"kernel_arg_type_qual", !""}
-// ARGINFO: !{!"kernel_arg_name", !"X"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"X"}
+// CHECK: define spir_kernel void @foo3{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD31:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD32:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD33:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD33]]
+// CHECK: !kernel_arg_type_qual ![[MD34:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD35:[0-9]+]]
typedef unsigned int myunsignedint;
kernel void foo4(__global unsigned int * X, __global myunsignedint * Y) {
}
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 1}
-// CHECK: !{!"kernel_arg_access_qual", !"none", !"none"}
-// CHECK: !{!"kernel_arg_type", !"uint*", !"myunsignedint*"}
-// CHECK: !{!"kernel_arg_base_type", !"uint*", !"uint*"}
-// CHECK: !{!"kernel_arg_type_qual", !"", !""}
-// ARGINFO: !{!"kernel_arg_name", !"X", !"Y"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"X", !"Y"}
+// CHECK: define spir_kernel void @foo4{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD41:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD42:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD43:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD44:[0-9]+]]
+// CHECK: !kernel_arg_type_qual ![[MD45:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD46:[0-9]+]]
typedef image1d_t myImage;
kernel void foo5(read_only myImage img1, write_only image1d_t img2) {
}
-// CHECK: !{!"kernel_arg_access_qual", !"read_only", !"write_only"}
-// CHECK: !{!"kernel_arg_type", !"myImage", !"image1d_t"}
-// CHECK: !{!"kernel_arg_base_type", !"image1d_t", !"image1d_t"}
-// ARGINFO: !{!"kernel_arg_name", !"img1", !"img2"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"img1", !"img2"}
+// CHECK: define spir_kernel void @foo5{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD41:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD51:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD52:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD53:[0-9]+]]
+// CHECK: !kernel_arg_type_qual ![[MD45]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD54:[0-9]+]]
+
+// CHECK: ![[MD11]] = !{i32 1, i32 0, i32 0, i32 2}
+// CHECK: ![[MD12]] = !{!"none", !"none", !"none", !"none"}
+// CHECK: ![[MD13]] = !{!"int*", !"int", !"int", !"float*"}
+// CHECK: ![[MD14]] = !{!"restrict", !"const", !"volatile", !"restrict const"}
+// ARGINFO: ![[MD15]] = !{!"X", !"Y", !"anotherArg", !"Z"}
+// CHECK: ![[MD21]] = !{i32 1, i32 1, i32 1}
+// CHECK: ![[MD22]] = !{!"read_only", !"read_only", !"write_only"}
+// CHECK: ![[MD23]] = !{!"__read_only image1d_t", !"__read_only image2d_t", !"__write_only image2d_array_t"}
+// CHECK: ![[MD24]] = !{!"", !"", !""}
+// ARGINFO: ![[MD25]] = !{!"img1", !"img2", !"img3"}
+// CHECK: ![[MD31]] = !{i32 1}
+// CHECK: ![[MD32]] = !{!"none"}
+// CHECK: ![[MD33]] = !{!"half*"}
+// CHECK: ![[MD34]] = !{!""}
+// ARGINFO: ![[MD35]] = !{!"X"}
+// CHECK: ![[MD41]] = !{i32 1, i32 1}
+// CHECK: ![[MD42]] = !{!"none", !"none"}
+// CHECK: ![[MD43]] = !{!"uint*", !"myunsignedint*"}
+// CHECK: ![[MD44]] = !{!"uint*", !"uint*"}
+// CHECK: ![[MD45]] = !{!"", !""}
+// ARGINFO: ![[MD46]] = !{!"X", !"Y"}
+// CHECK: ![[MD51]] = !{!"read_only", !"write_only"}
+// CHECK: ![[MD52]] = !{!"myImage", !"__write_only image1d_t"}
+// CHECK: ![[MD53]] = !{!"__read_only image1d_t", !"__write_only image1d_t"}
+// ARGINFO: ![[MD54]] = !{!"img1", !"img2"}
+
Index: lib/CodeGen/CodeGenFunction.cpp
===================================================================
--- lib/CodeGen/CodeGenFunction.cpp
+++ lib/CodeGen/CodeGenFunction.cpp
@@ -435,7 +435,6 @@
// includes the argument name, its type, the address and access qualifiers used.
static void GenOpenCLArgMetadata(const FunctionDecl *FD, llvm::Function *Fn,
CodeGenModule &CGM, llvm::LLVMContext &Context,
- SmallVector<llvm::Metadata *, 5> &kernelMDArgs,
CGBuilderTy &Builder, ASTContext &ASTCtx) {
// Create MDNodes that represent the kernel arg metadata.
// Each MDNode is a list in the form of "key", N number of values which is
@@ -445,28 +444,21 @@
// MDNode for the kernel argument address space qualifiers.
SmallVector<llvm::Metadata *, 8> addressQuals;
- addressQuals.push_back(llvm::MDString::get(Context, "kernel_arg_addr_space"));
// MDNode for the kernel argument access qualifiers (images only).
SmallVector<llvm::Metadata *, 8> accessQuals;
- accessQuals.push_back(llvm::MDString::get(Context, "kernel_arg_access_qual"));
// MDNode for the kernel argument type names.
SmallVector<llvm::Metadata *, 8> argTypeNames;
- argTypeNames.push_back(llvm::MDString::get(Context, "kernel_arg_type"));
// MDNode for the kernel argument base type names.
SmallVector<llvm::Metadata *, 8> argBaseTypeNames;
- argBaseTypeNames.push_back(
- llvm::MDString::get(Context, "kernel_arg_base_type"));
// MDNode for the kernel argument type qualifiers.
SmallVector<llvm::Metadata *, 8> argTypeQuals;
- argTypeQuals.push_back(llvm::MDString::get(Context, "kernel_arg_type_qual"));
// MDNode for the kernel argument names.
SmallVector<llvm::Metadata *, 8> argNames;
- argNames.push_back(llvm::MDString::get(Context, "kernel_arg_name"));
for (unsigned i = 0, e = FD->getNumParams(); i != e; ++i) {
const ParmVarDecl *parm = FD->getParamDecl(i);
@@ -577,13 +569,25 @@
argNames.push_back(llvm::MDString::get(Context, parm->getName()));
}
- kernelMDArgs.push_back(llvm::MDNode::get(Context, addressQuals));
- kernelMDArgs.push_back(llvm::MDNode::get(Context, accessQuals));
- kernelMDArgs.push_back(llvm::MDNode::get(Context, argTypeNames));
- kernelMDArgs.push_back(llvm::MDNode::get(Context, argBaseTypeNames));
- kernelMDArgs.push_back(llvm::MDNode::get(Context, argTypeQuals));
+ Fn->setMetadata("kernel_arg_addr_space",
+ llvm::MDNode::get(Context, addressQuals));
+ Fn->setMetadata("kernel_arg_access_qual",
+ llvm::MDNode::get(Context, accessQuals));
+ Fn->setMetadata("kernel_arg_type",
+ llvm::MDNode::get(Context, argTypeNames));
+ Fn->setMetadata("kernel_arg_base_type",
+ llvm::MDNode::get(Context, argBaseTypeNames));
+ Fn->setMetadata("kernel_arg_type_qual",
+ llvm::MDNode::get(Context, argTypeQuals));
if (CGM.getCodeGenOpts().EmitOpenCLArgMetadata)
- kernelMDArgs.push_back(llvm::MDNode::get(Context, argNames));
+ Fn->setMetadata("kernel_arg_name",
+ llvm::MDNode::get(Context, argNames));
+}
+
+static void SetFunctionAttribute(llvm::Function &F, llvm::StringRef Name,
+ llvm::StringRef Value) {
+ F.setAttributes(F.getAttributes().addAttribute(
+ F.getContext(), llvm::AttributeSet::FunctionIndex, Name, Value));
}
void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD,
@@ -594,50 +598,24 @@
llvm::LLVMContext &Context = getLLVMContext();
- SmallVector<llvm::Metadata *, 5> kernelMDArgs;
- kernelMDArgs.push_back(llvm::ConstantAsMetadata::get(Fn));
-
- GenOpenCLArgMetadata(FD, Fn, CGM, Context, kernelMDArgs, Builder,
- getContext());
-
- if (const VecTypeHintAttr *A = FD->getAttr<VecTypeHintAttr>()) {
- QualType hintQTy = A->getTypeHint();
- const ExtVectorType *hintEltQTy = hintQTy->getAs<ExtVectorType>();
- bool isSignedInteger =
- hintQTy->isSignedIntegerType() ||
- (hintEltQTy && hintEltQTy->getElementType()->isSignedIntegerType());
- llvm::Metadata *attrMDArgs[] = {
- llvm::MDString::get(Context, "vec_type_hint"),
- llvm::ConstantAsMetadata::get(llvm::UndefValue::get(
- CGM.getTypes().ConvertType(A->getTypeHint()))),
- llvm::ConstantAsMetadata::get(llvm::ConstantInt::get(
- llvm::IntegerType::get(Context, 32),
- llvm::APInt(32, (uint64_t)(isSignedInteger ? 1 : 0))))};
- kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
- }
+ GenOpenCLArgMetadata(FD, Fn, CGM, Context, Builder, getContext());
+
+ if (const VecTypeHintAttr *A = FD->getAttr<VecTypeHintAttr>())
+ SetFunctionAttribute(*Fn, "vec_type_hint", A->getTypeHint().getAsString());
if (const WorkGroupSizeHintAttr *A = FD->getAttr<WorkGroupSizeHintAttr>()) {
- llvm::Metadata *attrMDArgs[] = {
- llvm::MDString::get(Context, "work_group_size_hint"),
- llvm::ConstantAsMetadata::get(Builder.getInt32(A->getXDim())),
- llvm::ConstantAsMetadata::get(Builder.getInt32(A->getYDim())),
- llvm::ConstantAsMetadata::get(Builder.getInt32(A->getZDim()))};
- kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
+ std::string S;
+ llvm::raw_string_ostream SS(S);
+ SS << A->getXDim() << " " << A->getYDim() << " " << A->getZDim();
+ SetFunctionAttribute(*Fn, "work_group_size_hint", SS.str());
}
if (const ReqdWorkGroupSizeAttr *A = FD->getAttr<ReqdWorkGroupSizeAttr>()) {
- llvm::Metadata *attrMDArgs[] = {
- llvm::MDString::get(Context, "reqd_work_group_size"),
- llvm::ConstantAsMetadata::get(Builder.getInt32(A->getXDim())),
- llvm::ConstantAsMetadata::get(Builder.getInt32(A->getYDim())),
- llvm::ConstantAsMetadata::get(Builder.getInt32(A->getZDim()))};
- kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
+ std::string S;
+ llvm::raw_string_ostream SS(S);
+ SS << A->getXDim() << " " << A->getYDim() << " " << A->getZDim();
+ SetFunctionAttribute(*Fn, "reqd_work_group_size", SS.str());
}
-
- llvm::MDNode *kernelMDNode = llvm::MDNode::get(Context, kernelMDArgs);
- llvm::NamedMDNode *OpenCLKernelMetadata =
- CGM.getModule().getOrInsertNamedMetadata("opencl.kernels");
- OpenCLKernelMetadata->addOperand(kernelMDNode);
}
/// Determine whether the function F ends with a return stmt.
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
