AI/rocm-systems
2
0
Форкнуть 0
Код Задачи Запросы на слияние Действия Пакеты Проекты Релизы Вики Активность

Apply .clangformat to all repo source files

Просмотр исходного кода 
Change-Id: I7e79c6058f0303f9a98911e3b7dd2e8596079344
Этот коммит содержится в:
Maneesh Gupta
2018-03-12 11:29:03 +05:30
родитель 18e70b1e6b
Коммит 1ba06f63c4
293 изменённых файлов: 43980 добавлений и 45830 удалений
Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы
samples/0_Intro/bit_extract/bit_extract.cpp
+32 -35
Просмотреть файл
@@ -28,79 +28,76 @@ THE SOFTWARE.
#endif
#define CHECK(cmd) \
{\
hipError_t error = cmd;\
if (error != hipSuccess) { \
fprintf(stderr, "error: '%s'(%d) at %s:%d\n", hipGetErrorString(error), error,__FILE__, __LINE__); \
exit(EXIT_FAILURE);\
}\
}
#define CHECK(cmd) \
{ \
hipError_t error = cmd; \
if (error != hipSuccess) { \
fprintf(stderr, "error: '%s'(%d) at %s:%d\n", hipGetErrorString(error), error, \
__FILE__, __LINE__); \
exit(EXIT_FAILURE); \
} \
}
__global__ void
bit_extract_kernel(hipLaunchParm lp, uint32_t *C_d, const uint32_t *A_d, size_t N)
{
__global__ void bit_extract_kernel(hipLaunchParm lp, uint32_t* C_d, const uint32_t* A_d, size_t N) {
size_t offset = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
size_t stride = hipBlockDim_x * hipGridDim_x ;
size_t stride = hipBlockDim_x * hipGridDim_x;
for (size_t i=offset; i<N; i+=stride) {
for (size_t i = offset; i < N; i += stride) {
#ifdef __HIP_PLATFORM_HCC__
C_d[i] = hc::__bitextract_u32(A_d[i], 8, 4);
#else /* defined __HIP_PLATFORM_NVCC__ or other path */
C_d[i] = ((A_d[i] & 0xf00) >> 8);
C_d[i] = ((A_d[i] & 0xf00) >> 8);
#endif
}
}
int main(int argc, char *argv[])
{
int main(int argc, char* argv[]) {
uint32_t *A_d, *C_d;
uint32_t *A_h, *C_h;
size_t N = 1000000;
size_t Nbytes = N * sizeof(uint32_t);
int deviceId;
CHECK (hipGetDevice(&deviceId));
CHECK(hipGetDevice(&deviceId));
hipDeviceProp_t props;
CHECK(hipGetDeviceProperties(&props, deviceId));
printf ("info: running on device #%d %s\n", deviceId, props.name);
printf("info: running on device #%d %s\n", deviceId, props.name);
printf ("info: allocate host mem (%6.2f MB)\n", 2*Nbytes/1024.0/1024.0);
printf("info: allocate host mem (%6.2f MB)\n", 2 * Nbytes / 1024.0 / 1024.0);
A_h = (uint32_t*)malloc(Nbytes);
CHECK(A_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
CHECK(A_h == 0 ? hipErrorMemoryAllocation : hipSuccess);
C_h = (uint32_t*)malloc(Nbytes);
CHECK(C_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
CHECK(C_h == 0 ? hipErrorMemoryAllocation : hipSuccess);
for (size_t i=0; i<N; i++)
{
for (size_t i = 0; i < N; i++) {
A_h[i] = i;
}
printf ("info: allocate device mem (%6.2f MB)\n", 2*Nbytes/1024.0/1024.0);
printf("info: allocate device mem (%6.2f MB)\n", 2 * Nbytes / 1024.0 / 1024.0);
CHECK(hipMalloc(&A_d, Nbytes));
CHECK(hipMalloc(&C_d, Nbytes));
printf ("info: copy Host2Device\n");
CHECK ( hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
printf("info: copy Host2Device\n");
CHECK(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
printf ("info: launch 'bit_extract_kernel' \n");
printf("info: launch 'bit_extract_kernel' \n");
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
hipLaunchKernel(bit_extract_kernel, dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
hipLaunchKernel(bit_extract_kernel, dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
printf ("info: copy Device2Host\n");
CHECK ( hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
printf("info: copy Device2Host\n");
CHECK(hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
printf ("info: check result\n");
for (size_t i=0; i<N; i++) {
printf("info: check result\n");
for (size_t i = 0; i < N; i++) {
unsigned Agold = ((A_h[i] & 0xf00) >> 8);
if (C_h[i] != Agold) {
fprintf (stderr, "mismatch detected.\n");
printf ("%zu: %08x =? %08x (Ain=%08x)\n", i, C_h[i], Agold, A_h[i]);
fprintf(stderr, "mismatch detected.\n");
printf("%zu: %08x =? %08x (Ain=%08x)\n", i, C_h[i], Agold, A_h[i]);
CHECK(hipErrorUnknown);
}
}
printf ("PASSED!\n");
printf("PASSED!\n");
}