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

SWDEV-546485 Port and clean up for all tests in catch/perftests/memory folder. (#558)

Browse Source 
* SWDEV-546485 Port and clean up for hipPerfBufferCopyRectSpeed

* SWDEV-546485 Port and clean up for hipPerfDevMemReadSpeed

* SWDEV-546485 Port and clean up for hipPerfDevMemWriteSpeed

* SWDEV-546485 Port and clean up for hipPerfHostNumaAlloc

* SWDEV-546485 Port and clean up for hipPerfMemcpy

* SWDEV-546485 Port and clean up for hipPerfMemMallocCpyFree

* SWDEV-546485 Port and clean up for hipPerfMemset

* SWDEV-546485 Port and clean up for hipPerfSampleRate

* SWDEV-546485 Port and clean up for hipPerfSharedMemReadSpeed

* SWDEV-546485 Ported and fixed up segfault for hipPerfMemFill

* SWDEV-545485 Returning to unedited stage
This commit is contained in:
Naeisseh, Hadi
2025-08-15 16:09:19 -04:00
committed by GitHub
parent 9fdc9a98b7
commit 04469c0cde
20 changed files with 603 additions and 3195 deletions
Download Patch File Download Diff File Expand all files Collapse all files
catch/perftests/memory/hipPerfMemMallocCpyFree.cc
+41 -40
View File
@@ -18,13 +18,13 @@ THE SOFTWARE.
*/
/**
* @addtogroup hipMemcpy hipMemcpy
* @{
* @ingroup perfMemoryTest
* `hipMemcpy(void* dst, const void* src, size_t count, hipMemcpyKind kind)` -
* Copies data between host and device.
*/
* @addtogroup hipMemcpy hipMemcpy
* @{
* @ingroup perfMemoryTest
* `hipMemcpy(void* dst, const void* src, size_t count, hipMemcpyKind kind)` -
* Copies data between host and device.
*/
// #define ENABLE_DEBUG 1
#include <time.h>
#include <hip_test_common.hh>
@@ -38,7 +38,7 @@ void valSet(int* A, int val, size_t size) {
}
}
void setup(size_t *size, int *num, int **pA, const size_t totalGlobalMem) {
void setup(size_t* size, int* num, int** pA, const size_t totalGlobalMem) {
for (int i = 0; i < *num; i++) {
size[i] = 1 << (i + 6);
if ((NUM_ITER + 1) * size[i] > totalGlobalMem) {
@@ -50,39 +50,39 @@ void setup(size_t *size, int *num, int **pA, const size_t totalGlobalMem) {
valSet(*pA, 1, size[*num - 1]);
}
void testInit(size_t size, int *A) {
int *Ad;
void testInit(size_t size, int* A) {
int* Ad;
clock_t start = clock();
HIP_CHECK(hipMalloc(&Ad, size)); // hip::init() will be called
HIP_CHECK(hipMalloc(&Ad, size)); // hip::init() will be called
clock_t end = clock();
double uS = (end - start) * 1000000. / CLOCKS_PER_SEC;
INFO("Initial: hipMalloc(" << size << ") cost " << uS << "us" << "\n");
CONSOLE_PRINT("Initial: hipMalloc(%zu) cost %.2fus\n", size, uS);
start = clock();
HIP_CHECK(hipMemcpy(Ad, A, size, hipMemcpyHostToDevice));
HIP_CHECK(hipDeviceSynchronize());
end = clock();
uS = (end - start) * 1000000. / CLOCKS_PER_SEC;
INFO("hipMemcpy(" << size << ") cost " << uS << "us" << "\n");
CONSOLE_PRINT("hipMemcpy(%zu) cost %.2fus\n", size, uS);
start = clock();
HIP_CHECK(hipFree(Ad));
end = clock();
uS = (end - start) * 1000000. / CLOCKS_PER_SEC;
INFO("hipFree(" << size << ") cost " << uS << "us" << "\n");
CONSOLE_PRINT("hipFree(%zu) cost %.2fus\n", size, uS);
}
static bool hipPerfMemMallocCpyFree_test() {
double uS;
clock_t start, end;
size_t size[NUM_SIZE] = { 0 };
int *Ad[NUM_ITER] = { nullptr };
int *A;
size_t size[NUM_SIZE] = {0};
int* Ad[NUM_ITER] = {nullptr};
int* A;
hipDeviceProp_t props;
memset(&props, 0, sizeof(props));
HIP_CHECK(hipGetDeviceProperties(&props, 0));
INFO("totalGlobalMem: " << props.totalGlobalMem << "\n");
CONSOLE_PRINT("totalGlobalMem: %zu\n", props.totalGlobalMem);
int num = NUM_SIZE;
setup(size, &num, &A, props.totalGlobalMem);
@@ -91,59 +91,60 @@ static bool hipPerfMemMallocCpyFree_test() {
for (int i = 0; i < num; i++) {
start = clock();
for (int j = 0; j < NUM_ITER; j++) {
HIP_CHECK(hipMalloc(&Ad[j], size[i]));
HIP_CHECK(hipMalloc(&Ad[j], size[i]));
}
end = clock();
uS = (end - start) * 1000000. / (NUM_ITER * CLOCKS_PER_SEC);
INFO("hipMalloc(" << size[i] << ") cost " << uS << "us" << "\n");
CONSOLE_PRINT("hipMalloc(%zu) cost %.2fus\n", size[i], uS);
start = clock();
for (int j = 0; j < NUM_ITER; j++) {
HIP_CHECK(hipMemcpy(Ad[j], A, size[i], hipMemcpyHostToDevice));
HIP_CHECK(hipMemcpy(Ad[j], A, size[i], hipMemcpyHostToDevice));
}
HIP_CHECK(hipDeviceSynchronize());
end = clock();
uS = (end - start) * 1000000. / (NUM_ITER * CLOCKS_PER_SEC);
INFO("hipMemcpy(" << size[i] << ") cost " << uS << "us" << "\n");
CONSOLE_PRINT("hipMemcpy(%zu) cost %.2fus\n", size[i], uS);
start = clock();
for (int j = 0; j < NUM_ITER; j++) {
HIP_CHECK(hipFree(Ad[j]));
Ad[j] = nullptr;
HIP_CHECK(hipFree(Ad[j]));
Ad[j] = nullptr;
}
end = clock();
double uS = (end - start) * 1000000. / (NUM_ITER * CLOCKS_PER_SEC);
INFO("hipFree(" << size[i] << ") cost " << uS << "us" << "\n");
CONSOLE_PRINT("hipFree(%zu) cost %.2fus\n", size[i], uS);
}
free(A);
return true;
}
/**
* Test Description
* ------------------------
*  - Verify hipPerfMemMallocCpyFree status.
* Test source
* ------------------------
*  - perftests/memory/hipPerfMemMallocCpyFree.cc
* Test requirements
* ------------------------
*  - HIP_VERSION >= 5.6
*/
* Test Description
* ------------------------
*  - Verify hipPerfMemMallocCpyFree status.
* Test source
* ------------------------
*  - perftests/memory/hipPerfMemMallocCpyFree.cc
* Test requirements
* ------------------------
*  - HIP_VERSION >= 5.6
*/
TEST_CASE("Perf_hipPerfMemMallocCpyFree_test") {
int numDevices = 0;
HIP_CHECK(hipGetDeviceCount(&numDevices));
if (numDevices <= 0) {
SUCCEED("Skipped testcase hipPerfDevMemReadSpeed as"
"there is no device to test.");
SUCCEED(
"Skipped testcase hipPerfDevMemReadSpeed as"
"there is no device to test.");
} else {
REQUIRE(true == hipPerfMemMallocCpyFree_test());
}
}
/**
* End doxygen group perfMemoryTest.
* @}
*/
* End doxygen group perfMemoryTest.
* @}
*/