MachineTom
5f76cb916d
1. Create a set of mini numa interface. In Linux, the interface is based on system call rather than libnuma. In Windows, the interface can also work, but the policy class is dummy. Different from Linux, Windows doesn't provide numactl tool or numa lib to setup numa policy, thus the default policy is followed in Windows, that is, using the closest host numa node to allocate pinned host memory in hipHostMalloc(). To get the closest host numa node of a GPU device, you need query the new attribute hipDeviceAttributeHostNumaId. Then you can create a thread with CPU affinity on the numa node. For example, reference the test in hip-tests/catch/perftests/memory/hipPerfHostNumaAllocWin.cc. 2. Remove pfnSetThreadGroupAffinity and pfnGetNumaNodeProcessorMaskEx as the functions have been exposed since Win7 and Win server 2008. 3. Other minor fixes.
254 línte
7.8 KiB
C++
254 línte
7.8 KiB
C++
/*
|
||
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
|
||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||
of this software and associated documentation files (the "Software"), to deal
|
||
in the Software without restriction, including without limitation the rights
|
||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||
copies of the Software, and to permit persons to whom the Software is
|
||
furnished to do so, subject to the following conditions:
|
||
The above copyright notice and this permission notice shall be included in
|
||
all copies or substantial portions of the Software.
|
||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
||
THE SOFTWARE.
|
||
*/
|
||
|
||
/**
|
||
* @addtogroup hipMemcpy hipMemcpy
|
||
* @{
|
||
* @ingroup perfMemoryTest
|
||
* `hipMemcpy(void* dst, const void* src, size_t count, hipMemcpyKind kind)` -
|
||
* Copies data between host and device.
|
||
*/
|
||
|
||
#include <hip_test_common.hh>
|
||
// #define ENABLE_DEBUG 1
|
||
#define NUM_SIZE 14
|
||
#define NUM_ITER 1000
|
||
// max BW number for DevicetoDeviceNoCU
|
||
#define NOCU_MAX_BW 128
|
||
|
||
class hipPerfMemcpy {
|
||
private:
|
||
size_t totalSizes_[NUM_SIZE];
|
||
void setHostBuffer(int* A, int val, size_t size);
|
||
|
||
public:
|
||
hipPerfMemcpy();
|
||
~hipPerfMemcpy() {}
|
||
void TestResult(unsigned int numTests, std::chrono::duration<double, std::micro> diff,
|
||
hipMemcpyKind type);
|
||
bool run_h2d(unsigned int numTests);
|
||
bool run_d2h(unsigned int numTests);
|
||
bool run_d2d(unsigned int numTests);
|
||
bool run_d2d_nocu(unsigned int numTests);
|
||
};
|
||
|
||
hipPerfMemcpy::hipPerfMemcpy() {
|
||
for (int i = 0; i < NUM_SIZE; i++) {
|
||
totalSizes_[i] = 1 << (i + 9);
|
||
}
|
||
}
|
||
|
||
void hipPerfMemcpy::setHostBuffer(int* A, int val, size_t size) {
|
||
size_t len = size / sizeof(int);
|
||
for (int i = 0; i < len; i++) {
|
||
A[i] = val;
|
||
}
|
||
}
|
||
|
||
void hipPerfMemcpy::TestResult(unsigned int numTests,
|
||
std::chrono::duration<double, std::micro> diff, hipMemcpyKind type) {
|
||
// BW in GB/s
|
||
double perf =
|
||
(static_cast<double>(totalSizes_[numTests] * NUM_ITER) * static_cast<double>(1e-03)) /
|
||
diff.count();
|
||
|
||
const char* typestr = NULL;
|
||
|
||
if (type == hipMemcpyHostToDevice) {
|
||
typestr = "Host to Device";
|
||
} else if (type == hipMemcpyDeviceToHost) {
|
||
typestr = "Device to Host";
|
||
} else if (type == hipMemcpyDeviceToDevice) {
|
||
typestr = "Device to Device";
|
||
perf *= 2.0;
|
||
} else if (type == hipMemcpyDeviceToDeviceNoCU) {
|
||
typestr = "Device to Device No CU";
|
||
perf *= 2.0;
|
||
}
|
||
|
||
CONSOLE_PRINT("hipPerfMemcpy[%d] %s copy BW %.2f GB/s for memory size of %zu Bytes.\n", numTests,
|
||
typestr, perf, totalSizes_[numTests]);
|
||
}
|
||
|
||
bool hipPerfMemcpy::run_h2d(unsigned int numTests) {
|
||
int *A, *Ad;
|
||
A = new int[totalSizes_[numTests]];
|
||
HIP_CHECK(hipHostRegister(A, totalSizes_[numTests], hipHostRegisterDefault));
|
||
setHostBuffer(A, 1, totalSizes_[numTests]);
|
||
HIP_CHECK(hipMalloc(&Ad, totalSizes_[numTests]));
|
||
|
||
// measure performance based on host time
|
||
auto all_start = std::chrono::steady_clock::now();
|
||
|
||
for (int j = 0; j < NUM_ITER; j++) {
|
||
HIP_CHECK(hipMemcpyAsync(Ad, A, totalSizes_[numTests], hipMemcpyHostToDevice, nullptr));
|
||
}
|
||
|
||
HIP_CHECK(hipDeviceSynchronize());
|
||
|
||
auto all_end = std::chrono::steady_clock::now();
|
||
std::chrono::duration<double, std::micro> diff = all_end - all_start;
|
||
|
||
TestResult(numTests, diff, hipMemcpyHostToDevice);
|
||
|
||
HIP_CHECK(hipHostUnregister(A));
|
||
delete[] A;
|
||
HIP_CHECK(hipFree(Ad));
|
||
|
||
return true;
|
||
}
|
||
|
||
bool hipPerfMemcpy::run_d2h(unsigned int numTests) {
|
||
int *A, *Ad;
|
||
A = new int[totalSizes_[numTests]];
|
||
HIP_CHECK(hipHostRegister(A, totalSizes_[numTests], hipHostRegisterDefault));
|
||
HIP_CHECK(hipMalloc(&Ad, totalSizes_[numTests]));
|
||
HIP_CHECK(hipMemset(Ad, 0x1, totalSizes_[numTests]));
|
||
|
||
// measure performance based on host time
|
||
auto all_start = std::chrono::steady_clock::now();
|
||
|
||
for (int j = 0; j < NUM_ITER; j++) {
|
||
HIP_CHECK(hipMemcpyAsync(A, Ad, totalSizes_[numTests], hipMemcpyDeviceToHost, nullptr));
|
||
}
|
||
|
||
HIP_CHECK(hipDeviceSynchronize());
|
||
|
||
auto all_end = std::chrono::steady_clock::now();
|
||
std::chrono::duration<double, std::micro> diff = all_end - all_start;
|
||
|
||
TestResult(numTests, diff, hipMemcpyDeviceToHost);
|
||
|
||
HIP_CHECK(hipHostUnregister(A));
|
||
delete[] A;
|
||
HIP_CHECK(hipFree(Ad));
|
||
|
||
return true;
|
||
}
|
||
|
||
bool hipPerfMemcpy::run_d2d(unsigned int numTests) {
|
||
int *Ad1, *Ad2;
|
||
HIP_CHECK(hipMalloc(&Ad1, totalSizes_[numTests]));
|
||
HIP_CHECK(hipMalloc(&Ad2, totalSizes_[numTests]));
|
||
HIP_CHECK(hipMemset(Ad2, 0x1, totalSizes_[numTests]));
|
||
|
||
|
||
// measure performance based on host time
|
||
auto all_start = std::chrono::steady_clock::now();
|
||
|
||
for (int j = 0; j < NUM_ITER; j++) {
|
||
HIP_CHECK(hipMemcpyAsync(Ad1, Ad2, totalSizes_[numTests], hipMemcpyDeviceToDevice, nullptr));
|
||
}
|
||
|
||
HIP_CHECK(hipDeviceSynchronize());
|
||
|
||
auto all_end = std::chrono::steady_clock::now();
|
||
std::chrono::duration<double, std::micro> diff = all_end - all_start;
|
||
|
||
TestResult(numTests, diff, hipMemcpyDeviceToDevice);
|
||
|
||
HIP_CHECK(hipFree(Ad1));
|
||
HIP_CHECK(hipFree(Ad2));
|
||
|
||
return true;
|
||
}
|
||
|
||
bool hipPerfMemcpy::run_d2d_nocu(unsigned int numTests) {
|
||
int *Ad1, *Ad2;
|
||
HIP_CHECK(hipMalloc(&Ad1, totalSizes_[numTests]));
|
||
HIP_CHECK(hipMalloc(&Ad2, totalSizes_[numTests]));
|
||
HIP_CHECK(hipMemset(Ad2, 0x1, totalSizes_[numTests]));
|
||
|
||
// measure performance based on host time
|
||
auto all_start = std::chrono::steady_clock::now();
|
||
|
||
for (int j = 0; j < NUM_ITER; j++) {
|
||
HIP_CHECK(
|
||
hipMemcpyAsync(Ad1, Ad2, totalSizes_[numTests], hipMemcpyDeviceToDeviceNoCU, nullptr));
|
||
}
|
||
|
||
HIP_CHECK(hipDeviceSynchronize());
|
||
|
||
auto all_end = std::chrono::steady_clock::now();
|
||
std::chrono::duration<double, std::micro> diff = all_end - all_start;
|
||
|
||
TestResult(numTests, diff, hipMemcpyDeviceToDeviceNoCU);
|
||
|
||
HIP_CHECK(hipFree(Ad1));
|
||
HIP_CHECK(hipFree(Ad2));
|
||
|
||
return true;
|
||
}
|
||
|
||
/**
|
||
* Test Description
|
||
* ------------------------
|
||
* - Verify hipPerfMemcpy status.
|
||
* Test source
|
||
* ------------------------
|
||
* - perftests/memory/hipPerfMemcpy.cc
|
||
* Test requirements
|
||
* ------------------------
|
||
* - HIP_VERSION >= 5.6
|
||
*/
|
||
|
||
TEST_CASE("Perf_hipPerfMemcpy_test") {
|
||
int numDevices = 0;
|
||
HIP_CHECK(hipGetDeviceCount(&numDevices));
|
||
|
||
if (numDevices <= 0) {
|
||
SUCCEED("Skipped testcase hipPerfMemcpy as there is no device to test.");
|
||
} else {
|
||
int deviceId = 0;
|
||
HIP_CHECK(hipSetDevice(deviceId));
|
||
hipDeviceProp_t props;
|
||
HIP_CHECK(hipGetDeviceProperties(&props, deviceId));
|
||
|
||
CONSOLE_PRINT("info: running on bus 0x%x %s with %d CUs and device id: %d\n", props.pciBusID,
|
||
props.name, props.multiProcessorCount, deviceId);
|
||
|
||
hipPerfMemcpy hipPerfMemcpy;
|
||
SECTION("Perf test Host Memory to Device Memory") {
|
||
for (auto testCase = 0; testCase < NUM_SIZE; testCase++) {
|
||
REQUIRE(true == hipPerfMemcpy.run_h2d(testCase));
|
||
}
|
||
}
|
||
SECTION("Perf test Device Memory to Host Memory") {
|
||
for (auto testCase = 0; testCase < NUM_SIZE; testCase++) {
|
||
REQUIRE(true == hipPerfMemcpy.run_d2h(testCase));
|
||
}
|
||
}
|
||
SECTION("Perf test Device Memory to Device Memory") {
|
||
for (auto testCase = 0; testCase < NUM_SIZE; testCase++) {
|
||
REQUIRE(true == hipPerfMemcpy.run_d2d(testCase));
|
||
}
|
||
}
|
||
SECTION("Perf test Device Memory to Device Memory No CU") {
|
||
for (auto testCase = 0; testCase < NUM_SIZE; testCase++) {
|
||
REQUIRE(true == hipPerfMemcpy.run_d2d_nocu(testCase));
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
/**
|
||
* End doxygen group perfMemoryTest.
|
||
* @}
|
||
*/
|