AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

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

Telusuri Sumber 
[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
orang tua 108992428d
melakukan 6bb9913e8a
7 mengubah file dengan 28 tambahan dan 27 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
tests/hipify-clang/unit_tests/samples/axpy.cu
+7 -7
Melihat 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);
tests/hipify-clang/unit_tests/samples/square.cu
+1 -1
Melihat 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");