AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[HIPIFY] Return to wrapping with HIP_KERNEL_NAME(...) macro of a template instantiation kernel launch

Browse Source 
[REASON]

1. hip-clang is fine with the templated kernel launch, brackets are unneeded: HIP_KERNEL_NAME(...) __VA_ARGS__
2. HCC is not, thus: HIP_KERNEL_NAME(...) (__VA_ARGS__)

[TODO] Clean-up entirely kernel name wrapping when HCC is finally obsolete.

+ Update perl generation, hipify-perl, and affected tests accordingly.
This commit is contained in:
Evgeny Mankov
2019-10-02 16:01:07 +03:00
parent 8549cc7e1c
commit fee2f0fa78
7 changed files with 28 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hipamd/tests/hipify-clang/unit_tests/device/atomics.cu
+1 -1
View File
@@ -257,7 +257,7 @@ void runTest() {
// CHECK: hipMemcpy(dOData, hOData, memSize, hipMemcpyHostToDevice);
cudaMemcpy(dOData, hOData, memSize, cudaMemcpyHostToDevice);
// Execute the kernel
// CHECK: hipLaunchKernelGGL((testKernel), dim3(numBlocks), dim3(numThreads), 0, 0, dOData);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(testKernel), dim3(numBlocks), dim3(numThreads), 0, 0, dOData);
testKernel<<<numBlocks, numThreads>>>(dOData);
// Copy result from device to host
// CHECK: hipMemcpy(hOData, dOData, memSize, hipMemcpyDeviceToHost);
hipamd/tests/hipify-clang/unit_tests/libraries/cuRAND/benchmark_curand_kernel.cpp
+5 -5
View File
@@ -121,7 +121,7 @@ struct runner
const size_t states_size = blocks * threads;
// CHECK: CUDA_CALL(hipMalloc((void **)&states, states_size * sizeof(GeneratorState)));
CUDA_CALL(cudaMalloc((void **)&states, states_size * sizeof(GeneratorState)));
// CHECK: hipLaunchKernelGGL((init_kernel), dim3(blocks), dim3(threads), 0, 0, states, seed, offset);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(init_kernel), dim3(blocks), dim3(threads), 0, 0, states, seed, offset);
init_kernel<<<blocks, threads>>>(states, seed, offset);
// CHECK: CUDA_CALL(hipPeekAtLastError());
// CHECK: CUDA_CALL(hipDeviceSynchronize());
@@ -142,7 +142,7 @@ struct runner
const GenerateFunc& generate_func,
const Extra extra)
{
// CHECK: hipLaunchKernelGGL((generate_kernel), dim3(blocks), dim3(threads), 0, 0, states, data, size, generate_func, extra);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(generate_kernel), dim3(blocks), dim3(threads), 0, 0, states, data, size, generate_func, extra);
generate_kernel<<<blocks, threads>>>(states, data, size, generate_func, extra);
}
};
@@ -223,7 +223,7 @@ struct runner<curandStateMtgp32_t>
const GenerateFunc& generate_func,
const Extra extra)
{
// CHECK: hipLaunchKernelGGL((generate_kernel), dim3(std::min((size_t)200, blocks)), dim3(256), 0, 0, states, data, size, generate_func, extra);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(generate_kernel), dim3(std::min((size_t)200, blocks)), dim3(256), 0, 0, states, data, size, generate_func, extra);
generate_kernel<<<std::min((size_t)200, blocks), 256>>>(states, data, size, generate_func, extra);
}
};
@@ -304,7 +304,7 @@ struct runner<curandStateSobol32_t>
CUDA_CALL(cudaMemcpy(directions, h_directions, size, cudaMemcpyHostToDevice));
const size_t blocks_x = next_power2((blocks + dimensions - 1) / dimensions);
// CHECK: hipLaunchKernelGGL((init_kernel), dim3(blocks_x, dimensions), dim3(threads), 0, 0, states, directions, offset);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(init_kernel), dim3(blocks_x, dimensions), dim3(threads), 0, 0, states, directions, offset);
init_kernel<<<dim3(blocks_x, dimensions), threads>>>(states, directions, offset);
// CHECK: CUDA_CALL(hipPeekAtLastError());
// CHECK: CUDA_CALL(hipDeviceSynchronize());
@@ -329,7 +329,7 @@ struct runner<curandStateSobol32_t>
const Extra extra)
{
const size_t blocks_x = next_power2((blocks + dimensions - 1) / dimensions);
// CHECK: hipLaunchKernelGGL((generate_kernel), dim3(blocks_x, dimensions), dim3(threads), 0, 0, states, data, size / dimensions, generate_func, extra);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(generate_kernel), dim3(blocks_x, dimensions), dim3(threads), 0, 0, states, data, size / dimensions, generate_func, extra);
generate_kernel<<<dim3(blocks_x, dimensions), threads>>>(states, data, size / dimensions, generate_func, extra);
}
};
hipamd/tests/hipify-clang/unit_tests/samples/axpy.cu
+7 -7
View File
@@ -10,7 +10,7 @@
#define KERNEL_CALL_AS_MACRO axpy<float><<<1, kDataLen>>>
#define KERNEL_NAME_MACRO axpy<float>
// CHECK: #define COMPLETE_LAUNCH hipLaunchKernelGGL((axpy), dim3(1), dim3(kDataLen), 0, 0, a, device_x, device_y)
// CHECK: #define COMPLETE_LAUNCH hipLaunchKernelGGL(HIP_KERNEL_NAME(axpy), dim3(1), dim3(kDataLen), 0, 0, a, device_x, device_y)
#define COMPLETE_LAUNCH axpy<<<1, kDataLen>>>(a, device_x, device_y)
@@ -48,22 +48,22 @@ int main(int argc, char* argv[]) {
cudaMemcpy(device_x, host_x, kDataLen * sizeof(float), cudaMemcpyHostToDevice);
// Launch the kernel in numerous different strange ways to exercise the prerocessor.
// CHECK: hipLaunchKernelGGL((axpy), dim3(1), dim3(kDataLen), 0, 0, a, device_x, device_y);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(axpy), dim3(1), dim3(kDataLen), 0, 0, a, device_x, device_y);
axpy<<<1, kDataLen>>>(a, device_x, device_y);
// CHECK: hipLaunchKernelGGL((axpy<float>), dim3(1), dim3(kDataLen), 0, 0, a, device_x, device_y);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(axpy<float>), dim3(1), dim3(kDataLen), 0, 0, a, device_x, device_y);
axpy<float><<<1, kDataLen>>>(a, device_x, device_y);
// CHECK: hipLaunchKernelGGL((axpy<float>), dim3(1), dim3(kDataLen), 0, 0, a, TOKEN_PASTE(device, _x), device_y);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(axpy<float>), dim3(1), dim3(kDataLen), 0, 0, a, TOKEN_PASTE(device, _x), device_y);
axpy<float><<<1, kDataLen>>>(a, TOKEN_PASTE(device, _x), device_y);
// CHECK: hipLaunchKernelGGL((axpy<float>), dim3(1), dim3(kDataLen), 0, 0, ARG_LIST_AS_MACRO);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(axpy<float>), dim3(1), dim3(kDataLen), 0, 0, ARG_LIST_AS_MACRO);
axpy<float><<<1, kDataLen>>>(ARG_LIST_AS_MACRO);
// CHECK: hipLaunchKernelGGL((KERNEL_NAME_MACRO), dim3(1), dim3(kDataLen), 0, 0, ARG_LIST_AS_MACRO);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(KERNEL_NAME_MACRO), dim3(1), dim3(kDataLen), 0, 0, ARG_LIST_AS_MACRO);
KERNEL_NAME_MACRO<<<1, kDataLen>>>(ARG_LIST_AS_MACRO);
// CHECK: hipLaunchKernelGGL((axpy<float>), dim3(1), dim3(kDataLen), 0, 0, ARG_LIST_AS_MACRO);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(axpy<float>), dim3(1), dim3(kDataLen), 0, 0, ARG_LIST_AS_MACRO);
KERNEL_CALL_AS_MACRO(ARG_LIST_AS_MACRO);
// CHECK: hipLaunchKernelGGL(empty, dim3(1), dim3(kDataLen), 0, 0);
hipamd/tests/hipify-clang/unit_tests/samples/square.cu
+1 -1
View File
@@ -94,7 +94,7 @@ int main(int argc, char *argv[])
const unsigned threadsPerBlock = 256;
printf ("info: launch 'vector_square' kernel\n");
// CHECK: hipLaunchKernelGGL((vector_square), dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
// CHECK: hipLaunchKernelGGL(HIP_KERNEL_NAME(vector_square), dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
vector_square <<<blocks, threadsPerBlock>>> (C_d, A_d, N);
printf ("info: copy Device2Host\n");