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/0_Intro/bit_extract/bit_extract.cpp
+63 -63
View File
@@ -26,85 +26,85 @@ THE SOFTWARE.
#include "hip_helper.h"
__global__ void bit_extract_kernel(uint32_t* C_d, const uint32_t* A_d, size_t N) {
size_t offset = (blockIdx.x * blockDim.x + threadIdx.x);
size_t stride = blockDim.x * gridDim.x;
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) {
for (size_t i = offset; i < N; i += stride) {
#ifdef __HIP_PLATFORM_AMD__
C_d[i] = __bitextract_u32(A_d[i], 8, 4);
C_d[i] = __bitextract_u32(A_d[i], 8, 4);
#else /* defined __HIP_PLATFORM_NVIDIA__ 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[]) {
uint32_t *A_d, *C_d;
uint32_t *A_h, *C_h;
size_t N = 1000000;
size_t Nbytes = N * sizeof(uint32_t);
uint32_t *A_d, *C_d;
uint32_t *A_h, *C_h;
size_t N = 1000000;
size_t Nbytes = N * sizeof(uint32_t);
#ifdef __HIP_ENABLE_PCH
// Verify hip_pch.o
const char* pch = nullptr;
unsigned int size = 0;
__hipGetPCH(&pch, &size);
printf("pch size: %u\n", size);
if (size == 0) {
printf("__hipGetPCH failed!\n");
return -1;
} else {
printf("__hipGetPCH succeeded!\n");
}
// Verify hip_pch.o
const char* pch = nullptr;
unsigned int size = 0;
__hipGetPCH(&pch, &size);
printf("pch size: %u\n", size);
if (size == 0) {
printf("__hipGetPCH failed!\n");
return -1;
} else {
printf("__hipGetPCH succeeded!\n");
}
#endif
int deviceId;
checkHipErrors(hipGetDevice(&deviceId));
hipDeviceProp_t props;
checkHipErrors(hipGetDeviceProperties(&props, deviceId));
printf("info: running on device #%d %s\n", deviceId, props.name);
int deviceId;
checkHipErrors(hipGetDevice(&deviceId));
hipDeviceProp_t props;
checkHipErrors(hipGetDeviceProperties(&props, deviceId));
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);
A_h = (uint32_t*)malloc(Nbytes);
checkHipErrors(A_h == 0 ? hipErrorOutOfMemory : hipSuccess);
C_h = (uint32_t*)malloc(Nbytes);
checkHipErrors(C_h == 0 ? hipErrorOutOfMemory : hipSuccess);
printf("info: allocate host mem (%6.2f MB)\n", 2 * Nbytes / 1024.0 / 1024.0);
A_h = (uint32_t*)malloc(Nbytes);
checkHipErrors(A_h == 0 ? hipErrorOutOfMemory : hipSuccess);
C_h = (uint32_t*)malloc(Nbytes);
checkHipErrors(C_h == 0 ? hipErrorOutOfMemory : hipSuccess);
for (size_t i = 0; i < N; i++) {
A_h[i] = 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);
checkHipErrors(hipMalloc(&A_d, Nbytes));
checkHipErrors(hipMalloc(&C_d, Nbytes));
printf("info: copy Host2Device\n");
checkHipErrors(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
printf("info: launch 'bit_extract_kernel' \n");
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
hipLaunchKernelGGL(bit_extract_kernel, dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
printf("info: copy Device2Host\n");
checkHipErrors(hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
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]);
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));
printf("info: launch 'bit_extract_kernel' \n");
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
hipLaunchKernelGGL(bit_extract_kernel, dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
printf("info: copy Device2Host\n");
checkHipErrors(hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
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]);
checkHipErrors(hipErrorUnknown);
}
}
checkHipErrors(hipFree(A_d));
checkHipErrors(hipFree(C_d));
free(A_h);
free(C_h);
printf("PASSED!\n");
printf("PASSED!\n");
}
projects/hip-tests/samples/0_Intro/generic_target/saxpy.cpp
+41 -39
View File
@@ -36,7 +36,7 @@ static constexpr auto NUM_BLOCKS{32};
using namespace std;
static constexpr auto saxpy{
R"(
R"(
#include "test_header.h"
#include "test_header1.h"
extern "C"
@@ -50,8 +50,7 @@ void saxpy(real a, realptr x, realptr y, realptr out, size_t n)
}
)"};
int main()
{
int main() {
hipDeviceProp_t props;
int device = 0;
checkHipErrors(hipSetDevice(device));
@@ -62,11 +61,11 @@ int main()
auto pos = agentTarget.find(':');
if (pos != std::string::npos) {
postFix = agentTarget.substr(pos); // Features
postFix = agentTarget.substr(pos); // Features
agentTarget.resize(pos);
}
if (!getGenericTarget(agentTarget, genericTarget)) {
cout << props.gcnArchName <<" has no generic target support. Skipped!" << endl;
cout << props.gcnArchName << " has no generic target support. Skipped!" << endl;
return 0;
}
if (agentTarget.find("gfx906") != std::string::npos) {
@@ -86,16 +85,20 @@ int main()
vector<const char*> header_sources;
header_names.push_back("test_header.h");
header_names.push_back("test_header1.h");
header_sources.push_back("#ifndef HIPRTC_TEST_HEADER_H\n#define HIPRTC_TEST_HEADER_H\ntypedef float real;\n#endif //HIPRTC_TEST_HEADER_H\n");
header_sources.push_back("#ifndef HIPRTC_TEST_HEADER1_H\n#define HIPRTC_TEST_HEADER1_H\ntypedef float* realptr;\n#endif //HIPRTC_TEST_HEADER1_H\n");
hiprtcCreateProgram(&prog, // prog
saxpy, // buffer
"saxpy.cu", // name
num_headers, // numHeaders
&header_sources[0], // headers
&header_names[0]); // includeNames
header_sources.push_back(
"#ifndef HIPRTC_TEST_HEADER_H\n#define HIPRTC_TEST_HEADER_H\ntypedef float real;\n#endif "
"//HIPRTC_TEST_HEADER_H\n");
header_sources.push_back(
"#ifndef HIPRTC_TEST_HEADER1_H\n#define HIPRTC_TEST_HEADER1_H\ntypedef float* "
"realptr;\n#endif //HIPRTC_TEST_HEADER1_H\n");
hiprtcCreateProgram(&prog, // prog
saxpy, // buffer
"saxpy.cu", // name
num_headers, // numHeaders
&header_sources[0], // headers
&header_names[0]); // includeNames
string offload {"--offload-arch="};
string offload{"--offload-arch="};
offload += genericTarget.c_str();
/*
* offload must be one of following:
@@ -108,17 +111,17 @@ int main()
*
* */
const char* options[] = {offload.c_str(), "-mcode-object-version=6", "-w"};
hiprtcResult compileResult {
hiprtcCompileProgram(prog, sizeof(options) / sizeof(options[0]), options) };
hiprtcResult compileResult{
hiprtcCompileProgram(prog, sizeof(options) / sizeof(options[0]), options)};
size_t logSize;
hiprtcGetProgramLogSize(prog, &logSize);
if (logSize) {
string log(logSize, '\0');
hiprtcGetProgramLog(prog, &log[0]);
string log(logSize, '\0');
hiprtcGetProgramLog(prog, &log[0]);
cout << log << '\n';
cout << log << '\n';
}
if (compileResult != HIPRTC_SUCCESS) {
@@ -148,43 +151,42 @@ int main()
unique_ptr<float[]> hOut{new float[n]};
for (size_t i = 0; i < n; ++i) {
hX[i] = static_cast<float>(i);
hY[i] = static_cast<float>(i * 2);
hX[i] = static_cast<float>(i);
hY[i] = static_cast<float>(i * 2);
}
hipDeviceptr_t dX, dY, dOut;
checkHipErrors(hipMalloc((void **)&dX, bufferSize));
checkHipErrors(hipMalloc((void **)&dY, bufferSize));
checkHipErrors(hipMalloc((void **)&dOut, bufferSize));
checkHipErrors(hipMalloc((void**)&dX, bufferSize));
checkHipErrors(hipMalloc((void**)&dY, bufferSize));
checkHipErrors(hipMalloc((void**)&dOut, bufferSize));
checkHipErrors(hipMemcpyHtoD(dX, hX.get(), bufferSize));
checkHipErrors(hipMemcpyHtoD(dY, hY.get(), bufferSize));
struct {
float a_;
hipDeviceptr_t b_;
hipDeviceptr_t c_;
hipDeviceptr_t d_;
size_t e_;
float a_;
hipDeviceptr_t b_;
hipDeviceptr_t c_;
hipDeviceptr_t d_;
size_t e_;
} args{a, dX, dY, dOut, n};
auto size = sizeof(args);
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args,
HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
checkHipErrors(hipModuleLaunchKernel(kernel, NUM_BLOCKS, 1, 1, NUM_THREADS, 1, 1,
0, nullptr, nullptr, config));
checkHipErrors(hipModuleLaunchKernel(kernel, NUM_BLOCKS, 1, 1, NUM_THREADS, 1, 1, 0, nullptr,
nullptr, config));
checkHipErrors(hipMemcpyDtoH(hOut.get(), dOut, bufferSize));
for (size_t i = 0; i < n; ++i) {
if (fabs(a * hX[i] + hY[i] - hOut[i]) > fabs(hOut[i])* 1e-6) {
cout << "Validation failed." << endl;
}
if (fabs(a * hX[i] + hY[i] - hOut[i]) > fabs(hOut[i]) * 1e-6) {
cout << "Validation failed." << endl;
}
}
checkHipErrors(hipFree((void *)dX));
checkHipErrors(hipFree((void *)dY));
checkHipErrors(hipFree((void *)dOut));
checkHipErrors(hipFree((void*)dX));
checkHipErrors(hipFree((void*)dY));
checkHipErrors(hipFree((void*)dOut));
checkHipErrors(hipModuleUnload(module));
projects/hip-tests/samples/0_Intro/generic_target/square.cpp
+1 -2
View File
@@ -28,8 +28,7 @@ THE SOFTWARE.
/*
* Square each element in the array A and write to array C.
*/
template <typename T>
__global__ void vector_square(T* C_d, const T* A_d, size_t N) {
template <typename T> __global__ void vector_square(T* C_d, const T* A_d, size_t N) {
size_t offset = (blockIdx.x * blockDim.x + threadIdx.x);
size_t stride = blockDim.x * gridDim.x;
projects/hip-tests/samples/0_Intro/module_api/defaultDriver.cpp
+48 -48
View File
@@ -33,56 +33,56 @@ THE SOFTWARE.
#define kernel_name "hello_world"
int main() {
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
checkHipErrors(hipDeviceGet(&device, 0));
checkHipErrors(hipCtxCreate(&context, 0, device));
checkHipErrors(hipMalloc((void**)&Ad, SIZE));
checkHipErrors(hipMalloc((void**)&Bd, SIZE));
checkHipErrors(hipMemcpyHtoD(Ad, A, SIZE));
checkHipErrors(hipMemcpyHtoD(Bd, B, SIZE));
hipModule_t Module;
hipFunction_t Function;
checkHipErrors(hipModuleLoad(&Module, fileName));
checkHipErrors(hipModuleGetFunction(&Function, Module, kernel_name));
void* args[2] = {&Ad, &Bd};
checkHipErrors(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, args, nullptr));
checkHipErrors(hipMemcpyDtoH(B, Bd, SIZE));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
}
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
checkHipErrors(hipDeviceGet(&device, 0));
checkHipErrors(hipCtxCreate(&context, 0, device));
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
checkHipErrors(hipMalloc((void**)&Ad, SIZE));
checkHipErrors(hipMalloc((void**)&Bd, SIZE));
checkHipErrors(hipMemcpyHtoD(Ad, A, SIZE));
checkHipErrors(hipMemcpyHtoD(Bd, B, SIZE));
hipModule_t Module;
hipFunction_t Function;
checkHipErrors(hipModuleLoad(&Module, fileName));
checkHipErrors(hipModuleGetFunction(&Function, Module, kernel_name));
void* args[2] = {&Ad, &Bd};
checkHipErrors(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, args, nullptr));
checkHipErrors(hipMemcpyDtoH(B, Bd, SIZE));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
}
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
checkHipErrors(hipFree(Ad));
checkHipErrors(hipFree(Bd));
delete[] A;
delete[] B;
checkHipErrors(hipCtxDestroy(context));
return 0;
checkHipErrors(hipFree(Ad));
checkHipErrors(hipFree(Bd));
delete[] A;
delete[] B;
checkHipErrors(hipCtxDestroy(context));
return 0;
}
projects/hip-tests/samples/0_Intro/module_api/launchKernelHcc.cpp
+61 -61
View File
@@ -38,69 +38,69 @@ THE SOFTWARE.
#define kernel_name "hello_world"
int main() {
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
checkHipErrors(hipDeviceGet(&device, 0));
checkHipErrors(hipCtxCreate(&context, 0, device));
checkHipErrors(hipMalloc((void**)&Ad, SIZE));
checkHipErrors(hipMalloc((void**)&Bd, SIZE));
checkHipErrors(hipMemcpyHtoD(Ad, A, SIZE));
checkHipErrors(hipMemcpyHtoD(Bd, B, SIZE));
hipModule_t Module;
hipFunction_t Function;
checkHipErrors(hipModuleLoad(&Module, fileName));
checkHipErrors(hipModuleGetFunction(&Function, Module, kernel_name));
struct {
void* _Ad;
void* _Bd;
} args;
args._Ad = Ad;
args._Bd = Bd;
size_t size = sizeof(args);
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
checkHipErrors(
hipExtModuleLaunchKernel(Function, LEN, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config, 0));
checkHipErrors(hipMemcpyDtoH(B, Bd, SIZE));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
}
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
checkHipErrors(hipDeviceGet(&device, 0));
checkHipErrors(hipCtxCreate(&context, 0, device));
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
checkHipErrors(hipMalloc((void**)&Ad, SIZE));
checkHipErrors(hipMalloc((void**)&Bd, SIZE));
checkHipErrors(hipMemcpyHtoD(Ad, A, SIZE));
checkHipErrors(hipMemcpyHtoD(Bd, B, SIZE));
hipModule_t Module;
hipFunction_t Function;
checkHipErrors(hipModuleLoad(&Module, fileName));
checkHipErrors(hipModuleGetFunction(&Function, Module, kernel_name));
struct {
void* _Ad;
void* _Bd;
} args;
args._Ad = Ad;
args._Bd = Bd;
size_t size = sizeof(args);
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
checkHipErrors(
hipExtModuleLaunchKernel(Function, LEN, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config, 0));
checkHipErrors(hipMemcpyDtoH(B, Bd, SIZE));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
}
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
checkHipErrors(hipFree(Ad));
checkHipErrors(hipFree(Bd));
delete[] A;
delete[] B;
checkHipErrors(hipCtxDestroy(context));
return 0;
checkHipErrors(hipFree(Ad));
checkHipErrors(hipFree(Bd));
delete[] A;
delete[] B;
checkHipErrors(hipCtxDestroy(context));
return 0;
}
projects/hip-tests/samples/0_Intro/module_api/runKernel.cpp
+59 -59
View File
@@ -35,67 +35,67 @@ THE SOFTWARE.
#define kernel_name "hello_world"
int main() {
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
checkHipErrors(hipDeviceGet(&device, 0));
checkHipErrors(hipCtxCreate(&context, 0, device));
checkHipErrors(hipMalloc((void**)&Ad, SIZE));
checkHipErrors(hipMalloc((void**)&Bd, SIZE));
checkHipErrors(hipMemcpyHtoD(Ad, A, SIZE));
checkHipErrors(hipMemcpyHtoD(Bd, B, SIZE));
hipModule_t Module;
hipFunction_t Function;
checkHipErrors(hipModuleLoad(&Module, fileName));
checkHipErrors(hipModuleGetFunction(&Function, Module, kernel_name));
struct {
void* _Ad;
void* _Bd;
} args;
args._Ad = (void*)Ad;
args._Bd = (void*)Bd;
size_t size = sizeof(args);
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
checkHipErrors(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
checkHipErrors(hipMemcpyDtoH(B, Bd, SIZE));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
}
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
checkHipErrors(hipDeviceGet(&device, 0));
checkHipErrors(hipCtxCreate(&context, 0, device));
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
checkHipErrors(hipMalloc((void**)&Ad, SIZE));
checkHipErrors(hipMalloc((void**)&Bd, SIZE));
checkHipErrors(hipMemcpyHtoD(Ad, A, SIZE));
checkHipErrors(hipMemcpyHtoD(Bd, B, SIZE));
hipModule_t Module;
hipFunction_t Function;
checkHipErrors(hipModuleLoad(&Module, fileName));
checkHipErrors(hipModuleGetFunction(&Function, Module, kernel_name));
struct {
void* _Ad;
void* _Bd;
} args;
args._Ad = (void*) Ad;
args._Bd = (void*) Bd;
size_t size = sizeof(args);
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
checkHipErrors(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
checkHipErrors(hipMemcpyDtoH(B, Bd, SIZE));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
}
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
checkHipErrors(hipFree(Ad));
hipFree(Bd);
delete[] A;
delete[] B;
checkHipErrors(hipCtxDestroy(context));
return 0;
checkHipErrors(hipFree(Ad));
hipFree(Bd);
delete[] A;
delete[] B;
checkHipErrors(hipCtxDestroy(context));
return 0;
}
projects/hip-tests/samples/0_Intro/module_api/vcpy_kernel.cpp
+2 -2
View File
@@ -23,6 +23,6 @@ THE SOFTWARE.
#include "hip/hip_runtime.h"
extern "C" __global__ void hello_world(float* a, float* b) {
int tx = threadIdx.x;
b[tx] = a[tx];
int tx = threadIdx.x;
b[tx] = a[tx];
}
projects/hip-tests/samples/0_Intro/module_api_global/runKernel.cpp
+107 -102
View File
@@ -31,131 +31,136 @@ THE SOFTWARE.
#define SIZE LEN * sizeof(float)
#define fileName "vcpy_kernel.code"
#define checkHipErrors(cmd) \
{ \
hipError_t status = cmd; \
if (status != hipSuccess) { \
std::cout << "error: #" << status << " (" << hipGetErrorString(status) \
<< ") at line:" << __LINE__ << ": " << #cmd << std::endl; \
abort(); \
} \
}
#define checkHipErrors(cmd) \
{ \
hipError_t status = cmd; \
if (status != hipSuccess) { \
std::cout << "error: #" << status << " (" << hipGetErrorString(status) \
<< ") at line:" << __LINE__ << ": " << #cmd << std::endl; \
abort(); \
} \
}
int main() {
float *A, *B;
float *Ad, *Bd;
A = new float[LEN];
B = new float[LEN];
float *A, *B;
float *Ad, *Bd;
A = new float[LEN];
B = new float[LEN];
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
}
for (uint32_t i = 0; i < LEN; i++) {
A[i] = i * 1.0f;
B[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
hipDeviceGet(&device, 0);
hipCtxCreate(&context, 0, device);
hipInit(0);
hipDevice_t device;
hipCtx_t context;
hipDeviceGet(&device, 0);
hipCtxCreate(&context, 0, device);
hipMalloc((void**)&Ad, SIZE);
hipMalloc((void**)&Bd, SIZE);
hipMalloc((void**)&Ad, SIZE);
hipMalloc((void**)&Bd, SIZE);
hipMemcpyHtoD(hipDeviceptr_t(Ad), A, SIZE);
hipMemcpyHtoD((hipDeviceptr_t)(Bd), B, SIZE);
hipModule_t Module;
checkHipErrors(hipModuleLoad(&Module, fileName));
hipMemcpyHtoD(hipDeviceptr_t(Ad), A, SIZE);
hipMemcpyHtoD((hipDeviceptr_t)(Bd), B, SIZE);
hipModule_t Module;
checkHipErrors(hipModuleLoad(&Module, fileName));
float myDeviceGlobal_h = 42.0;
float* deviceGlobal;
size_t deviceGlobalSize;
checkHipErrors(hipModuleGetGlobal((void**)&deviceGlobal, &deviceGlobalSize, Module, "myDeviceGlobal"));
checkHipErrors(hipMemcpyHtoD(hipDeviceptr_t(deviceGlobal), &myDeviceGlobal_h, deviceGlobalSize));
float myDeviceGlobal_h = 42.0;
float* deviceGlobal;
size_t deviceGlobalSize;
checkHipErrors(
hipModuleGetGlobal((void**)&deviceGlobal, &deviceGlobalSize, Module, "myDeviceGlobal"));
checkHipErrors(hipMemcpyHtoD(hipDeviceptr_t(deviceGlobal), &myDeviceGlobal_h, deviceGlobalSize));
#define ARRAY_SIZE 16
float myDeviceGlobalArray_h[ARRAY_SIZE];
float *myDeviceGlobalArray;
size_t myDeviceGlobalArraySize;
checkHipErrors(hipModuleGetGlobal((void**)&myDeviceGlobalArray, &myDeviceGlobalArraySize, Module, "myDeviceGlobalArray"));
for (int i = 0; i < ARRAY_SIZE; i++) {
myDeviceGlobalArray_h[i] = i * 1000.0f;
}
checkHipErrors(hipMemcpyHtoD(hipDeviceptr_t(myDeviceGlobalArray), &myDeviceGlobalArray_h, myDeviceGlobalArraySize));
float myDeviceGlobalArray_h[ARRAY_SIZE];
float* myDeviceGlobalArray;
size_t myDeviceGlobalArraySize;
checkHipErrors(hipModuleGetGlobal((void**)&myDeviceGlobalArray, &myDeviceGlobalArraySize, Module,
"myDeviceGlobalArray"));
for (int i = 0; i < ARRAY_SIZE; i++) {
myDeviceGlobalArray_h[i] = i * 1000.0f;
}
checkHipErrors(hipMemcpyHtoD(hipDeviceptr_t(myDeviceGlobalArray), &myDeviceGlobalArray_h,
myDeviceGlobalArraySize));
struct {
void* _Ad;
void* _Bd;
} args;
struct {
void* _Ad;
void* _Bd;
} args;
args._Ad = (void*) Ad;
args._Bd = (void*) Bd;
args._Ad = (void*)Ad;
args._Bd = (void*)Bd;
size_t size = sizeof(args);
size_t size = sizeof(args);
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
void* config[] = {HIP_LAUNCH_PARAM_BUFFER_POINTER, &args, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END};
{
hipFunction_t Function;
checkHipErrors(hipModuleGetFunction(&Function, Module, "hello_world"));
checkHipErrors(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
{
hipFunction_t Function;
checkHipErrors(hipModuleGetFunction(&Function, Module, "hello_world"));
checkHipErrors(
hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
hipMemcpyDtoH(B, Bd, SIZE);
hipMemcpyDtoH(B, Bd, SIZE);
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
if (mismatchCount >= 10) {
break;
}
}
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
if (A[i] != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << A[i] << " != " << B[i] << std::endl;
if (mismatchCount >= 10) {
break;
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
}
}
{
hipFunction_t Function;
checkHipErrors(hipModuleGetFunction(&Function, Module, "test_globals"));
int val =-1;
checkHipErrors(hipFuncGetAttribute(&val, HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES,Function));
printf("Shared Size Bytes = %d\n",val);
checkHipErrors(hipFuncGetAttribute(&val, HIP_FUNC_ATTRIBUTE_NUM_REGS, Function));
printf("Num Regs = %d\n",val);
checkHipErrors(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
}
hipMemcpyDtoH(B, Bd, SIZE);
{
hipFunction_t Function;
checkHipErrors(hipModuleGetFunction(&Function, Module, "test_globals"));
int val = -1;
checkHipErrors(hipFuncGetAttribute(&val, HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES, Function));
printf("Shared Size Bytes = %d\n", val);
checkHipErrors(hipFuncGetAttribute(&val, HIP_FUNC_ATTRIBUTE_NUM_REGS, Function));
printf("Num Regs = %d\n", val);
checkHipErrors(
hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
float expected = A[i] + myDeviceGlobal_h + myDeviceGlobalArray_h[i % 16];
if (expected != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << expected << " != " << B[i] << std::endl;
if (mismatchCount >= 10) {
break;
}
}
hipMemcpyDtoH(B, Bd, SIZE);
int mismatchCount = 0;
for (uint32_t i = 0; i < LEN; i++) {
float expected = A[i] + myDeviceGlobal_h + myDeviceGlobalArray_h[i % 16];
if (expected != B[i]) {
mismatchCount++;
std::cout << "error: mismatch " << expected << " != " << B[i] << std::endl;
if (mismatchCount >= 10) {
break;
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
}
}
hipFree(Ad);
hipFree(Bd);
delete[] A;
delete[] B;
hipCtxDestroy(context);
return 0;
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
}
hipFree(Ad);
hipFree(Bd);
delete[] A;
delete[] B;
hipCtxDestroy(context);
return 0;
}
projects/hip-tests/samples/0_Intro/module_api_global/vcpy_kernel.cpp
+4 -4
View File
@@ -28,11 +28,11 @@ __device__ float myDeviceGlobal;
__device__ float myDeviceGlobalArray[16];
extern "C" __global__ void hello_world(const float* a, float* b) {
int tx = threadIdx.x;
b[tx] = a[tx];
int tx = threadIdx.x;
b[tx] = a[tx];
}
extern "C" __global__ void test_globals(const float* a, float* b) {
int tx = threadIdx.x;
b[tx] = a[tx] + myDeviceGlobal + myDeviceGlobalArray[tx % ARRAY_SIZE];
int tx = threadIdx.x;
b[tx] = a[tx] + myDeviceGlobal + myDeviceGlobalArray[tx % ARRAY_SIZE];
}
projects/hip-tests/samples/0_Intro/square/square.hipref.cpp
+60 -61
View File
@@ -24,79 +24,78 @@ THE SOFTWARE.
#include "hip/hip_runtime.h"
#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); \
} \
}
{ \
hipError_t error = cmd; \
if (error != hipSuccess) { \
fprintf(stderr, "error: '%s'(%d) at %s:%d\n", hipGetErrorString(error), error, __FILE__, \
__LINE__); \
exit(EXIT_FAILURE); \
} \
}
/*
* Square each element in the array A and write to array C.
*/
template <typename T>
__global__ void vector_square(T* C_d, const 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, const 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);
static int device = 0;
CHECK(hipSetDevice(device));
hipDeviceProp_t props;
CHECK(hipGetDeviceProperties(&props, device /*deviceID*/));
printf("info: running on device %s\n", props.name);
float *A_d, *C_d;
float *A_h, *C_h;
size_t N = 1000000;
size_t Nbytes = N * sizeof(float);
static int device = 0;
CHECK(hipSetDevice(device));
hipDeviceProp_t props;
CHECK(hipGetDeviceProperties(&props, device /*deviceID*/));
printf("info: running on device %s\n", props.name);
#ifdef __HIP_PLATFORM_AMD__
printf("info: architecture on AMD GPU device is: %s\n", props.gcnArchName);
printf("info: architecture on AMD GPU device is: %s\n", props.gcnArchName);
#endif
printf("info: allocate host mem (%6.2f MB)\n", 2 * Nbytes / 1024.0 / 1024.0);
A_h = (float*)malloc(Nbytes);
CHECK(A_h == 0 ? hipErrorOutOfMemory : hipSuccess);
C_h = (float*)malloc(Nbytes);
CHECK(C_h == 0 ? hipErrorOutOfMemory : 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);
CHECK(A_h == 0 ? hipErrorOutOfMemory : hipSuccess);
C_h = (float*)malloc(Nbytes);
CHECK(C_h == 0 ? hipErrorOutOfMemory : 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);
CHECK(hipMalloc(&A_d, Nbytes));
CHECK(hipMalloc(&C_d, Nbytes));
printf("info: copy Host2Device\n");
CHECK(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
printf("info: launch 'vector_square' kernel\n");
hipLaunchKernelGGL(vector_square, 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: check result\n");
for (size_t i = 0; i < N; i++) {
if (C_h[i] != A_h[i] * A_h[i]) {
CHECK(hipErrorUnknown);
}
}
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));
CHECK(hipFree(A_d));
CHECK(hipFree(C_d));
free(A_h);
free(C_h);
printf("info: copy Host2Device\n");
CHECK(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
const unsigned blocks = 512;
const unsigned threadsPerBlock = 256;
printf("info: launch 'vector_square' kernel\n");
hipLaunchKernelGGL(vector_square, 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: check result\n");
for (size_t i = 0; i < N; i++) {
if (C_h[i] != A_h[i] * A_h[i]) {
CHECK(hipErrorUnknown);
}
}
CHECK(hipFree(A_d));
CHECK(hipFree(C_d));
free(A_h);
free(C_h);
printf("PASSED!\n");
printf("PASSED!\n");
}