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

Apply .clangformat to all repo source files

Browse Source 
Change-Id: I7e79c6058f0303f9a98911e3b7dd2e8596079344


[ROCm/clr commit: 9e47fccc89]
This commit is contained in:
Maneesh Gupta
2018-03-12 11:29:03 +05:30
parent ecbb701440
commit 46ddefedee
293 changed files with 43980 additions and 45830 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/hipamd/tests/src/deviceLib/hip_ffs.cpp
+107 -115
View File
@@ -34,153 +34,145 @@ THE SOFTWARE.
#include <hip/hip_runtime.h>
#include <hip/device_functions.h>
#define HIP_ASSERT(x) (assert((x)==hipSuccess))
#define HIP_ASSERT(x) (assert((x) == hipSuccess))
#define WIDTH 8
#define HEIGHT 8
#define WIDTH 8
#define HEIGHT 8
#define NUM (WIDTH*HEIGHT)
#define NUM (WIDTH * HEIGHT)
#define THREADS_PER_BLOCK_X 8
#define THREADS_PER_BLOCK_Y 8
#define THREADS_PER_BLOCK_Z 1
#define THREADS_PER_BLOCK_X 8
#define THREADS_PER_BLOCK_Y 8
#define THREADS_PER_BLOCK_Z 1
template<typename T>
int lastbit( T a)
{
if (a == 0)
#if defined (__HIP_PLATFORM_HCC__) && !defined ( NVCC_COMPAT )
return -1;
template <typename T>
int lastbit(T a) {
if (a == 0)
#if defined(__HIP_PLATFORM_HCC__) && !defined(NVCC_COMPAT)
return -1;
#else
return 0;
return 0;
#endif
int pos = 1;
while ((a&1) != 1) {
a >>= 1; pos++;
}
#if defined (__HIP_PLATFORM_HCC__) && !defined ( NVCC_COMPAT )
return pos-1;
int pos = 1;
while ((a & 1) != 1) {
a >>= 1;
pos++;
}
#if defined(__HIP_PLATFORM_HCC__) && !defined(NVCC_COMPAT)
return pos - 1;
#else
return pos;
return pos;
#endif
}
__global__ void
HIP_kernel(hipLaunchParm lp,
unsigned int* a, unsigned int* b, unsigned int* c, unsigned long long int* d,
int width, int height)
{
__global__ void HIP_kernel(hipLaunchParm lp, unsigned int* a, unsigned int* b, unsigned int* c,
unsigned long long int* d, int width, int height) {
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
int i = y * width + x;
if ( i < (width * height)) {
int i = y * width + x;
if (i < (width * height)) {
a[i] = __ffs(b[i]);
c[i] = __ffsll(d[i]);
}
}
c[i] = __ffsll(d[i]);
}
}
using namespace std;
int main() {
unsigned int* hostA;
unsigned int* hostB;
unsigned int* hostC;
unsigned long long int* hostD;
unsigned int* hostA;
unsigned int* hostB;
unsigned int* hostC;
unsigned long long int* hostD;
unsigned int* deviceA;
unsigned int* deviceB;
unsigned int* deviceC;
unsigned long long int* deviceD;
unsigned int* deviceA;
unsigned int* deviceB;
unsigned int* deviceC;
unsigned long long int* deviceD;
hipDeviceProp_t devProp;
hipGetDeviceProperties(&devProp, 0);
cout << " System minor " << devProp.minor << endl;
cout << " System major " << devProp.major << endl;
cout << " agent prop name " << devProp.name << endl;
hipDeviceProp_t devProp;
hipGetDeviceProperties(&devProp, 0);
cout << " System minor " << devProp.minor << endl;
cout << " System major " << devProp.major << endl;
cout << " agent prop name " << devProp.name << endl;
cout << "hip Device prop succeeded " << endl ;
cout << "hip Device prop succeeded " << endl;
int i;
int errors;
int i;
int errors;
hostA = (unsigned int*)malloc(NUM * sizeof(unsigned int));
hostB = (unsigned int*)malloc(NUM * sizeof(unsigned int));
hostC = (unsigned int*)malloc(NUM * sizeof(unsigned int));
hostD = (unsigned long long int*)malloc(NUM * sizeof(unsigned long long int));
hostA = (unsigned int*)malloc(NUM * sizeof(unsigned int));
hostB = (unsigned int*)malloc(NUM * sizeof(unsigned int));
hostC = (unsigned int*)malloc(NUM * sizeof(unsigned int));
hostD = (unsigned long long int*)malloc(NUM * sizeof(unsigned long long int));
// initialize the input data
for (i = 0; i < NUM; i++) {
hostB[i] = i;
hostD[i] = 1099511627776+i;
}
HIP_ASSERT(hipMalloc((void**)&deviceA, NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)&deviceB, NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)&deviceC, NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)&deviceD, NUM * sizeof(unsigned long long int)));
HIP_ASSERT(hipMemcpy(deviceB, hostB, NUM*sizeof(unsigned int), hipMemcpyHostToDevice));
HIP_ASSERT(hipMemcpy(deviceD, hostD, NUM*sizeof(unsigned long long int), hipMemcpyHostToDevice));
hipLaunchKernel(HIP_kernel,
dim3(WIDTH/THREADS_PER_BLOCK_X, HEIGHT/THREADS_PER_BLOCK_Y),
dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y),
0, 0,
deviceA ,deviceB, deviceC,deviceD, WIDTH ,HEIGHT);
HIP_ASSERT(hipMemcpy(hostA, deviceA, NUM*sizeof(unsigned int), hipMemcpyDeviceToHost));
HIP_ASSERT(hipMemcpy(hostC, deviceC, NUM*sizeof(unsigned int), hipMemcpyDeviceToHost));
// verify the results
errors = 0;
for (i = 0; i < NUM; i++) {
printf("gpu_ffs =%d, cpu_ffs =%d \n",hostA[i],lastbit(hostB[i]));
if (hostA[i] != lastbit(hostB[i])) {
errors++;
// initialize the input data
for (i = 0; i < NUM; i++) {
hostB[i] = i;
hostD[i] = 1099511627776 + i;
}
}
if (errors!=0) {
cout << "FAILED: ffs" << endl;
return -1;
} else {
cout << "__ffs() for unsigned checked!" << endl;
}
errors = 0;
for (i = 0; i < NUM; i++) {
printf("gpu_ffsll =%d, cpu_ffsll =%d \n",hostC[i],lastbit(hostD[i]));
if (hostC[i] != lastbit(hostD[i])) {
errors++;
HIP_ASSERT(hipMalloc((void**)&deviceA, NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)&deviceB, NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)&deviceC, NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)&deviceD, NUM * sizeof(unsigned long long int)));
HIP_ASSERT(hipMemcpy(deviceB, hostB, NUM * sizeof(unsigned int), hipMemcpyHostToDevice));
HIP_ASSERT(
hipMemcpy(deviceD, hostD, NUM * sizeof(unsigned long long int), hipMemcpyHostToDevice));
hipLaunchKernel(HIP_kernel, dim3(WIDTH / THREADS_PER_BLOCK_X, HEIGHT / THREADS_PER_BLOCK_Y),
dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y), 0, 0, deviceA, deviceB, deviceC,
deviceD, WIDTH, HEIGHT);
HIP_ASSERT(hipMemcpy(hostA, deviceA, NUM * sizeof(unsigned int), hipMemcpyDeviceToHost));
HIP_ASSERT(hipMemcpy(hostC, deviceC, NUM * sizeof(unsigned int), hipMemcpyDeviceToHost));
// verify the results
errors = 0;
for (i = 0; i < NUM; i++) {
printf("gpu_ffs =%d, cpu_ffs =%d \n", hostA[i], lastbit(hostB[i]));
if (hostA[i] != lastbit(hostB[i])) {
errors++;
}
}
if (errors != 0) {
cout << "FAILED: ffs" << endl;
return -1;
} else {
cout << "__ffs() for unsigned checked!" << endl;
}
errors = 0;
for (i = 0; i < NUM; i++) {
printf("gpu_ffsll =%d, cpu_ffsll =%d \n", hostC[i], lastbit(hostD[i]));
if (hostC[i] != lastbit(hostD[i])) {
errors++;
}
}
if (errors != 0) {
cout << "FAILED: ffs" << endl;
return -1;
} else {
cout << "__ffsll() for unsigned checked!" << endl;
}
}
if (errors!=0) {
cout << "FAILED: ffs" << endl;
return -1;
} else {
cout << "__ffsll() for unsigned checked!" << endl;
}
cout << "ffs test PASSED!" << endl;
HIP_ASSERT(hipFree(deviceA));
HIP_ASSERT(hipFree(deviceB));
HIP_ASSERT(hipFree(deviceC));
HIP_ASSERT(hipFree(deviceD));
HIP_ASSERT(hipFree(deviceA));
HIP_ASSERT(hipFree(deviceB));
HIP_ASSERT(hipFree(deviceC));
HIP_ASSERT(hipFree(deviceD));
free(hostA);
free(hostB);
free(hostC);
free(hostD);
free(hostA);
free(hostB);
free(hostC);
free(hostD);
return errors;
return errors;
}