AI/rocm-systems
2
0
Fork 0
Koodi Ongelmat Pull-pyynnöt Actions Paketit Projektit Julkaisut Wiki Toiminta
Files
bbad1d8539f442beee893ba0755caabcded71e0e
rocm-systems/tests/unit_tests/free_list_gtest.cpp
T
Edgar Gabriel d0c2845031 add support for GPUs using wavefront size of 32 (#285) 
* add gfx1100 support

Add support for Radeon 7900 GPUs (RX and PRO), and 7800 PRO.

I was contemplating to add gfx1101 and gfx1102 GPUs as well, but those are the lower end models that are more unlikely to be used for compute intensive jobs. In addition, I do not have access to them to test the support.

* update WF_SIZe for different options

Radeon systems use a WarpSize of 32, unlike current Instinct systems,
which use a warp size of 64. For the device side, a gfx specific ifdef
is sufficient. For the host side, we need to query the device
properties.

* adjust functional tests to wf_size of 32

* update unit tests to handle wf_size of 32

* address reviewer comments
2025-10-22 16:04:58 -05:00

310 rivejä
11 KiB
C++
Raaka Selitys Historia

/******************************************************************************
* Copyright (c) Advanced Micro Devices, Inc. All rights reserved.
*
* SPDX-License-Identifier: MIT
*
* 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.
*****************************************************************************/
#include "free_list_gtest.hpp"
#include "../src/util.hpp"
using namespace rocshmem;
/*****************************************************************************
******************************* Fixture Tests *******************************
*****************************************************************************/
namespace rocshmem {
template <typename List, typename Value>
__global__ void pop_all(List* list, Value* values, const std::size_t count) {
const auto stride = blockDim.x * gridDim.x;
const auto thread_index = blockIdx.x * blockDim.x + threadIdx.x;
// One push per block. block size is always WF_SIZE
for (std::size_t i = thread_index; i < count * WF_SIZE; i += stride) {
if (is_thread_zero_in_wave()) {
auto last = list->pop_front();
if (values != nullptr) {
values[i / WF_SIZE] = last.value;
}
}
}
}
template <typename List, typename Value>
__global__ void push_all(List* list, const Value* values,
const std::size_t count) {
const auto stride = blockDim.x * gridDim.x;
const auto thread_index = blockIdx.x * blockDim.x + threadIdx.x;
// One push per block. block size is always WF_SIZE
for (std::size_t i = thread_index; i < count * WF_SIZE; i += stride) {
if (is_thread_zero_in_wave()) {
list->push_back(values[i / WF_SIZE]);
}
}
}
template <typename List>
__global__ void pop_empty(List* list, bool* empty) {
auto pop_result = list->pop_front();
*empty = !pop_result.success;
}
} // namespace rocshmem
TYPED_TEST(FreeListTestFixture, pop_empty_device) {
using Allocator = typename TestFixture::Allocator;
using T = typename TestFixture::T;
auto& h_input = this->h_input;
auto& free_list = this->free_list;
auto& hip_allocator_ = this->hip_allocator_;
bool *is_empty {nullptr};
hip_allocator_.allocate(reinterpret_cast<void**>(&is_empty),
sizeof(bool));
CHECK_HIP(hipMemset(is_empty, 0, sizeof(bool)));
FreeListProxy<Allocator, T> empty_list_proxy{};
FreeList<T, Allocator>* empty_free_list{empty_list_proxy.get()};
rocshmem::pop_empty<<<1, 1>>>(empty_free_list, is_empty);
CHECK_HIP(hipDeviceSynchronize());
EXPECT_TRUE(is_empty[0]);
hip_allocator_.deallocate(is_empty);
}
TYPED_TEST(FreeListTestFixture, push_host_pop_device) {
using Allocator = typename TestFixture::Allocator;
using T = typename TestFixture::T;
auto& h_input = this->h_input;
auto& free_list = this->free_list;
auto& hip_allocator_ = this->hip_allocator_;
T *results {nullptr};
bool *is_empty {nullptr};
size_t size_bytes = sizeof(T) * h_input.size() + sizeof(bool);
hip_allocator_.allocate(reinterpret_cast<void**>(&results),
size_bytes);
CHECK_HIP(hipMemset(results, 0, size_bytes));
is_empty = reinterpret_cast<bool*>(results + h_input.size());
const auto block_size = this->wf_size;
rocshmem::pop_all<<<1, block_size>>>(free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
for (std::size_t i = 0; i < h_input.size(); i++) {
EXPECT_EQ(results[i], h_input[i]);
}
rocshmem::pop_empty<<<1, 1>>>(free_list, is_empty);
CHECK_HIP(hipDeviceSynchronize());
EXPECT_TRUE(is_empty[0]);
hip_allocator_.deallocate(results);
}
TYPED_TEST(FreeListTestFixture, push_host_concurrent_pop_device) {
using Allocator = typename TestFixture::Allocator;
using T = typename TestFixture::T;
auto& h_input = this->h_input;
auto& free_list = this->free_list;
auto& hip_allocator_ = this->hip_allocator_;
T *results {nullptr};
bool *is_empty {nullptr};
size_t size_bytes = sizeof(T) * h_input.size() + sizeof(bool);
hip_allocator_.allocate(reinterpret_cast<void**>(&results),
size_bytes);
CHECK_HIP(hipMemset(results, 0, size_bytes));
is_empty = reinterpret_cast<bool*>(results + h_input.size());
const auto num_blocks = h_input.size();
const auto block_size = this->wf_size;
rocshmem::pop_all<<<num_blocks, block_size>>>(
free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
std::vector<T> h_results(h_input.size());
CHECK_HIP(hipMemcpy(h_results.data(), results, sizeof(T) * h_input.size(),
hipMemcpyDeviceToHost));
// sort to guarantee that the ordering is correct
std::sort(h_input.begin(), h_input.end());
std::sort(h_results.begin(), h_results.end());
for (std::size_t i = 0; i < h_results.size(); i++) {
EXPECT_EQ(h_results[i], h_input[i]);
}
rocshmem::pop_empty<<<1, 1>>>(free_list, is_empty);
CHECK_HIP(hipDeviceSynchronize());
EXPECT_TRUE(is_empty[0]);
hip_allocator_.deallocate(results);
}
TYPED_TEST(FreeListTestFixture, push_host_pop_push_device) {
using Allocator = typename TestFixture::Allocator;
using T = typename TestFixture::T;
using FreeListType = FreeList<T, Allocator>;
auto& h_input = this->h_input;
auto& free_list = this->free_list;
auto& hip_allocator_ = this->hip_allocator_;
T *results {nullptr};
T *d_input {nullptr};
bool *is_empty {nullptr};
size_t size_bytes = 2 * sizeof(T) * h_input.size() + sizeof(bool);
hip_allocator_.allocate(reinterpret_cast<void**>(&results),
size_bytes);
CHECK_HIP(hipMemset(results, 0, size_bytes));
d_input = reinterpret_cast<T*>(results + h_input.size());
is_empty = reinterpret_cast<bool*>(d_input + h_input.size());
const auto block_size = this->wf_size;
CHECK_HIP(hipMemcpy(d_input, h_input.data(), sizeof(T) * h_input.size(),
hipMemcpyHostToDevice));
rocshmem::pop_all<FreeListType, T><<<1, block_size>>>(
free_list, nullptr, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
rocshmem::push_all<<<1, block_size>>>(free_list, d_input, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
rocshmem::pop_all<<<1, block_size>>>(free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
for (std::size_t i = 0; i < h_input.size(); i++) {
EXPECT_EQ(results[i], h_input[i]);
}
hip_allocator_.deallocate(results);
}
TYPED_TEST(FreeListTestFixture, push_host_pop_concurrent_push_device) {
using Allocator = typename TestFixture::Allocator;
using T = typename TestFixture::T;
using FreeListType = FreeList<T, Allocator>;
auto& h_input = this->h_input;
auto& free_list = this->free_list;
auto& hip_allocator_ = this->hip_allocator_;
T *results {nullptr};
T *d_input {nullptr};
size_t size_bytes = 2 * sizeof(T) * h_input.size();
hip_allocator_.allocate(reinterpret_cast<void**>(&results),
size_bytes);
CHECK_HIP(hipMemset(results, 0, size_bytes));
d_input = reinterpret_cast<T*>(results + h_input.size());
const auto block_size = this->wf_size;
CHECK_HIP(hipMemcpy(d_input, h_input.data(), sizeof(T) * h_input.size(),
hipMemcpyHostToDevice));
rocshmem::pop_all<FreeListType, T><<<1, block_size>>>(
free_list, nullptr,h_input.size());
CHECK_HIP(hipDeviceSynchronize());
// Concurrently push all values
const auto num_blocks = h_input.size();
rocshmem::push_all<<<num_blocks, block_size>>>(
free_list, d_input, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
rocshmem::pop_all<<<1, block_size>>>(free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
std::vector<T> h_results(h_input.size());
CHECK_HIP(hipMemcpy(h_results.data(), results, sizeof(T) * h_input.size(),
hipMemcpyDeviceToHost));
// sort to guarantee that the ordering is correct
std::sort(h_input.begin(), h_input.end());
std::sort(h_results.begin(), h_results.end());
for (std::size_t i = 0; i < h_results.size(); i++) {
EXPECT_EQ(h_results[i], h_input[i]);
}
hip_allocator_.deallocate(results);
}
TYPED_TEST(FreeListTestFixture, push_host_concurrent_pop_push_device) {
using Allocator = typename TestFixture::Allocator;
using T = typename TestFixture::T;
using FreeListType = FreeList<T, Allocator>;
auto& h_input = this->h_input;
auto& free_list = this->free_list;
auto& hip_allocator_ = this->hip_allocator_;
T *results {nullptr};
T *d_input {nullptr};
size_t size_bytes = 2 * sizeof(T) * h_input.size();
hip_allocator_.allocate(reinterpret_cast<void**>(&results),
size_bytes);
CHECK_HIP(hipMemset(results, 0, size_bytes));
d_input = reinterpret_cast<T*>(results + h_input.size());
CHECK_HIP(hipMemcpy(d_input, h_input.data(), sizeof(T) * h_input.size(),
hipMemcpyHostToDevice));
const auto block_size = this->wf_size;
rocshmem::pop_all<FreeListType, T><<<1, block_size>>>(
free_list, nullptr, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
// Concurrently push all values
const auto num_blocks = h_input.size();
rocshmem::push_all<<<num_blocks, block_size>>>(
free_list, d_input, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
// Concurrently pop all values
rocshmem::pop_all<<<num_blocks, block_size>>>(
free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
std::vector<T> h_results(h_input.size());
CHECK_HIP(hipMemcpy(h_results.data(), results, sizeof(T) * h_input.size(),
hipMemcpyDeviceToHost));
// sort to guarantee that the ordering is correct
std::sort(h_input.begin(), h_input.end());
std::sort(h_results.begin(), h_results.end());
for (std::size_t i = 0; i < h_results.size(); i++) {
EXPECT_EQ(h_results[i], h_input[i]);
}
hip_allocator_.deallocate(results);
}
Viittaa uudesa ongelmassa View Git Blame Copy Permalink