This is an archive of the discontinued LLVM Phabricator instance.

Differential D19918

AMDGPU/SI: Use amdgpu_kernel calling convention for OpenCL kernels.
ClosedPublic

Authored by nhaustov on May 4 2016, 6:53 AM.

Details

Reviewers
• tstellarAMD
arsenm
Commits
rG4961ea85d7c1: AMDGPU/SI: Use amdgpu_kernel calling convention for OpenCL kernels.
rC268718: AMDGPU/SI: Use amdgpu_kernel calling convention for OpenCL kernels.
rL268718: AMDGPU/SI: Use amdgpu_kernel calling convention for OpenCL kernels.

Diff Detail

Event Timeline

nhaustov updated this revision to Diff 56141.May 4 2016, 6:53 AM
nhaustov retitled this revision from to AMDGPU/SI: Use amdgpu_kernel calling convention for OpenCL kernels..
nhaustov updated this object.
nhaustov added reviewers: • tstellarAMD, arsenm.
nhaustov added a subscriber: cfe-commits.
nhaustov added a parent revision: D19917: AMDGPU/SI: Add amdgpu_kernel calling convention..May 4 2016, 6:54 AM
• tstellarAMD accepted this revision.May 5 2016, 10:15 AM
• tstellarAMD edited edge metadata.

LGTM.

This revision is now accepted and ready to land.May 5 2016, 10:15 AM
Closed by commit rL268718: AMDGPU/SI: Use amdgpu_kernel calling convention for OpenCL kernels. (authored by nhaustov). · Explain WhyMay 6 2016, 2:21 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
CodeGen/
8 lines
test/
CodeGenOpenCL/
12 lines
8 lines

Diff 56141

lib/CodeGen/TargetInfo.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 6,820 Lines • ▼ Show 20 Lines
void AMDGPUTargetCodeGenInfo::setTargetAttributes( void AMDGPUTargetCodeGenInfo::setTargetAttributes(
const Decl *D, const Decl *D,
llvm::GlobalValue *GV, llvm::GlobalValue *GV,
CodeGen::CodeGenModule &M) const { CodeGen::CodeGenModule &M) const {
const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D); const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D);
if (!FD) if (!FD)
return; return;
if (M.getLangOpts().OpenCL) {
if (FD->hasAttr<OpenCLKernelAttr>()) {
// Set amdgpu_kernel calling convention for OpenCL kernels.
llvm::Function *Fn = cast<llvm::Function>(GV);
Fn->setCallingConv(llvm::CallingConv::AMDGPU_KERNEL);
}
}
if (const auto Attr = FD->getAttr<AMDGPUNumVGPRAttr>()) { if (const auto Attr = FD->getAttr<AMDGPUNumVGPRAttr>()) {
llvm::Function *F = cast<llvm::Function>(GV); llvm::Function *F = cast<llvm::Function>(GV);
uint32_t NumVGPR = Attr->getNumVGPR(); uint32_t NumVGPR = Attr->getNumVGPR();
if (NumVGPR != 0) if (NumVGPR != 0)
F->addFnAttr("amdgpu_num_vgpr", llvm::utostr(NumVGPR)); F->addFnAttr("amdgpu_num_vgpr", llvm::utostr(NumVGPR));
} }
if (const auto Attr = FD->getAttr<AMDGPUNumSGPRAttr>()) { if (const auto Attr = FD->getAttr<AMDGPUNumSGPRAttr>()) {
▲ Show 20 Lines • Show All 1,139 Lines • Show Last 20 Lines

test/CodeGenOpenCL/amdgpu-calling-conv.cl

  • This file was added.
// REQUIRES: amdgpu-registered-target
// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -S -emit-llvm -o - %s | FileCheck %s
// CHECK: define amdgpu_kernel void @calling_conv_amdgpu_kernel()
kernel void calling_conv_amdgpu_kernel()
{
}
// CHECK: define void @calling_conv_none()
void calling_conv_none()
{
}

test/CodeGenOpenCL/amdgpu-num-gpr-attr.cl

// RUN: %clang_cc1 -triple amdgcn-- -target-cpu tahiti -O0 -emit-llvm -o - %s | FileCheck %s // RUN: %clang_cc1 -triple amdgcn-- -target-cpu tahiti -O0 -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -O0 -emit-llvm -verify -o - %s | FileCheck -check-prefix=X86 %s // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -O0 -emit-llvm -verify -o - %s | FileCheck -check-prefix=X86 %s
// Make sure this is silently accepted on other targets. // Make sure this is silently accepted on other targets.
__attribute__((amdgpu_num_vgpr(64))) // expected-no-diagnostics __attribute__((amdgpu_num_vgpr(64))) // expected-no-diagnostics
kernel void test_num_vgpr64() { kernel void test_num_vgpr64() {
// CHECK: define void @test_num_vgpr64() [[ATTR_VGPR64:#[0-9]+]] // CHECK: define amdgpu_kernel void @test_num_vgpr64() [[ATTR_VGPR64:#[0-9]+]]
} }
__attribute__((amdgpu_num_sgpr(32))) // expected-no-diagnostics __attribute__((amdgpu_num_sgpr(32))) // expected-no-diagnostics
kernel void test_num_sgpr32() { kernel void test_num_sgpr32() {
// CHECK: define void @test_num_sgpr32() [[ATTR_SGPR32:#[0-9]+]] // CHECK: define amdgpu_kernel void @test_num_sgpr32() [[ATTR_SGPR32:#[0-9]+]]
} }
__attribute__((amdgpu_num_vgpr(64), amdgpu_num_sgpr(32))) // expected-no-diagnostics __attribute__((amdgpu_num_vgpr(64), amdgpu_num_sgpr(32))) // expected-no-diagnostics
kernel void test_num_vgpr64_sgpr32() { kernel void test_num_vgpr64_sgpr32() {
// CHECK: define void @test_num_vgpr64_sgpr32() [[ATTR_VGPR64_SGPR32:#[0-9]+]] // CHECK: define amdgpu_kernel void @test_num_vgpr64_sgpr32() [[ATTR_VGPR64_SGPR32:#[0-9]+]]
} }
__attribute__((amdgpu_num_sgpr(20), amdgpu_num_vgpr(40))) // expected-no-diagnostics __attribute__((amdgpu_num_sgpr(20), amdgpu_num_vgpr(40))) // expected-no-diagnostics
kernel void test_num_sgpr20_vgpr40() { kernel void test_num_sgpr20_vgpr40() {
// CHECK: define void @test_num_sgpr20_vgpr40() [[ATTR_SGPR20_VGPR40:#[0-9]+]] // CHECK: define amdgpu_kernel void @test_num_sgpr20_vgpr40() [[ATTR_SGPR20_VGPR40:#[0-9]+]]
} }
__attribute__((amdgpu_num_vgpr(0))) // expected-no-diagnostics __attribute__((amdgpu_num_vgpr(0))) // expected-no-diagnostics
kernel void test_num_vgpr0() { kernel void test_num_vgpr0() {
} }
__attribute__((amdgpu_num_sgpr(0))) // expected-no-diagnostics __attribute__((amdgpu_num_sgpr(0))) // expected-no-diagnostics
kernel void test_num_sgpr0() { kernel void test_num_sgpr0() {
Show All 16 Lines