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

SWDEV-470698 - fix formatting, add format check workflow (#657)

Browse Source
This commit is contained in:
Danylo Lytovchenko
2025-08-20 16:28:06 +02:00
committed by GitHub
parent 5840940caa
commit f7338717ae
1574 changed files with 162972 additions and 199346 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/hip-tests/samples/2_Cookbook/16_assembly_to_executable/square.cpp
+51 -56
View File
@@ -26,69 +26,64 @@ THE SOFTWARE.
*
* Square each element in the array A and write to array C.
*/
template <typename T>
__global__ void
vector_square(T *C_d, T *A_d, size_t N)
{
size_t offset = (blockIdx.x * blockDim.x + threadIdx.x);
size_t stride = blockDim.x * gridDim.x ;
template <typename T> __global__ void vector_square(T* C_d, T* A_d, size_t N) {
size_t offset = (blockIdx.x * blockDim.x + threadIdx.x);
size_t stride = blockDim.x * gridDim.x;
for (size_t i=offset; i<N; i+=stride) {
C_d[i] = A_d[i] * A_d[i];
}
for (size_t i = offset; i < N; i += stride) {
C_d[i] = A_d[i] * A_d[i];
}
}
int main(int argc, char *argv[])
{
float *A_d, *C_d;
float *A_h, *C_h;
size_t N = 1000000;
size_t Nbytes = N * sizeof(float);
int main(int argc, char* argv[]) {
float *A_d, *C_d;
float *A_h, *C_h;
size_t N = 1000000;
size_t Nbytes = N * sizeof(float);
hipDeviceProp_t props;
checkHipErrors(hipGetDeviceProperties(&props, 0/*deviceID*/));
printf ("info: running on device %s\n", props.name);
hipDeviceProp_t props;
checkHipErrors(hipGetDeviceProperties(&props, 0 /*deviceID*/));
printf("info: running on device %s\n", props.name);
printf ("info: allocate host mem (%6.2f MB)\n", 2*Nbytes/1024.0/1024.0);
A_h = (float*)malloc(Nbytes);
checkHipErrors(A_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
C_h = (float*)malloc(Nbytes);
checkHipErrors(C_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
// Fill with Phi + i
for (size_t i=0; i<N; i++)
{
A_h[i] = 1.618f + i;
printf("info: allocate host mem (%6.2f MB)\n", 2 * Nbytes / 1024.0 / 1024.0);
A_h = (float*)malloc(Nbytes);
checkHipErrors(A_h == 0 ? hipErrorMemoryAllocation : hipSuccess);
C_h = (float*)malloc(Nbytes);
checkHipErrors(C_h == 0 ? hipErrorMemoryAllocation : hipSuccess);
// Fill with Phi + i
for (size_t i = 0; i < N; i++) {
A_h[i] = 1.618f + i;
}
printf("info: allocate device mem (%6.2f MB)\n", 2 * Nbytes / 1024.0 / 1024.0);
checkHipErrors(hipMalloc(&A_d, Nbytes));
checkHipErrors(hipMalloc(&C_d, Nbytes));
printf("info: copy Host2Device\n");
checkHipErrors(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
printf("info: launch 'vector_square' kernel\n");
vector_square<<<blocks, threadsPerBlock>>>(C_d, A_d, N);
printf("info: copy Device2Host\n");
checkHipErrors(hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
printf("info: checkHipErrors result\n");
for (size_t i = 0; i < N; i++) {
if (C_h[i] != A_h[i] * A_h[i]) {
checkHipErrors(hipErrorUnknown);
}
}
printf ("info: allocate device mem (%6.2f MB)\n", 2*Nbytes/1024.0/1024.0);
checkHipErrors(hipMalloc(&A_d, Nbytes));
checkHipErrors(hipMalloc(&C_d, Nbytes));
checkHipErrors(hipFree(A_d));
checkHipErrors(hipFree(C_d));
free(A_h);
free(C_h);
printf ("info: copy Host2Device\n");
checkHipErrors ( hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
printf ("info: launch 'vector_square' kernel\n");
vector_square <<<blocks, threadsPerBlock>>> (C_d, A_d, N);
printf ("info: copy Device2Host\n");
checkHipErrors ( hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
printf ("info: checkHipErrors result\n");
for (size_t i=0; i<N; i++) {
if (C_h[i] != A_h[i] * A_h[i]) {
checkHipErrors(hipErrorUnknown);
}
}
checkHipErrors(hipFree(A_d));
checkHipErrors(hipFree(C_d));
free(A_h);
free(C_h);
printf ("PASSED!\n");
printf("PASSED!\n");
}