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

add support for GPUs using wavefront size of 32 (#285)

Browse Source 
* 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
This commit is contained in:
Edgar Gabriel
2025-10-22 16:04:58 -05:00
committed by GitHub
parent 955c22aeed
commit d0c2845031
19 changed files with 192 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/unit_tests/free_list_gtest.cpp
+5 -5
View File
@@ -108,7 +108,7 @@ TYPED_TEST(FreeListTestFixture, push_host_pop_device) {
CHECK_HIP(hipMemset(results, 0, size_bytes));
is_empty = reinterpret_cast<bool*>(results + h_input.size());
const auto block_size = WF_SIZE;
const auto block_size = this->wf_size;
rocshmem::pop_all<<<1, block_size>>>(free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
@@ -140,7 +140,7 @@ TYPED_TEST(FreeListTestFixture, push_host_concurrent_pop_device) {
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 = WF_SIZE;
const auto block_size = this->wf_size;
rocshmem::pop_all<<<num_blocks, block_size>>>(
free_list, results, h_input.size());
CHECK_HIP(hipDeviceSynchronize());
@@ -184,7 +184,7 @@ TYPED_TEST(FreeListTestFixture, push_host_pop_push_device) {
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 = WF_SIZE;
const auto block_size = this->wf_size;
CHECK_HIP(hipMemcpy(d_input, h_input.data(), sizeof(T) * h_input.size(),
hipMemcpyHostToDevice));
@@ -223,7 +223,7 @@ TYPED_TEST(FreeListTestFixture, push_host_pop_concurrent_push_device) {
CHECK_HIP(hipMemset(results, 0, size_bytes));
d_input = reinterpret_cast<T*>(results + h_input.size());
const auto block_size = WF_SIZE;
const auto block_size = this->wf_size;
CHECK_HIP(hipMemcpy(d_input, h_input.data(), sizeof(T) * h_input.size(),
hipMemcpyHostToDevice));
@@ -277,7 +277,7 @@ TYPED_TEST(FreeListTestFixture, push_host_concurrent_pop_push_device) {
CHECK_HIP(hipMemcpy(d_input, h_input.data(), sizeof(T) * h_input.size(),
hipMemcpyHostToDevice));
const auto block_size = WF_SIZE;
const auto block_size = this->wf_size;
rocshmem::pop_all<FreeListType, T><<<1, block_size>>>(
free_list, nullptr, h_input.size());
CHECK_HIP(hipDeviceSynchronize());