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

Fix ROCm 6.4 warnings (#47)

Browse Source 
* Removed __AMDGCN_WAVEFRONT_SIZE

* Added unit test to validate WF_SIZE
This commit is contained in:
Yiltan
2025-02-24 13:34:13 -05:00
committed by GitHub
parent 57d60aa727
commit 487e5b7d0f
8 changed files with 100 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/reverse_offload/context_ro_device.cpp
+3 -3
View File
@@ -470,9 +470,9 @@ __device__ uint64_t active_logical_lane_id() {
}
__device__ uint64_t broadcast_lds(bool lowest_active, uint64_t value) {
constexpr size_t SIZE = 1024 / __AMDGCN_WAVEFRONT_SIZE;
constexpr size_t SIZE = 1024 / WF_SIZE;
__shared__ uint64_t value_per_warp[SIZE];
auto wavefront_number {get_flat_block_id() / __AMDGCN_WAVEFRONT_SIZE};
auto wavefront_number {get_flat_block_id() / WF_SIZE};
if (lowest_active) {
value_per_warp[wavefront_number] = value;
__threadfence_block();
@@ -481,7 +481,7 @@ __device__ uint64_t broadcast_lds(bool lowest_active, uint64_t value) {
}
__device__ uint64_t broadcast_shfl_up(uint64_t value) {
for (unsigned i{0}; i < __AMDGCN_WAVEFRONT_SIZE; i++) {
for (unsigned i{0}; i < WF_SIZE; i++) {
uint64_t temp{__shfl_up(value, i)};
if (temp) {
value = temp;