Merge commit '30727c48fcdf2178f013cbb843db563abd77d09c' into develop

This commit is contained in:
assistant-librarian[bot]
2025-11-27 23:12:24 +00:00
parent d0b319035a
commit 6032baee56
15 changed files with 2530 additions and 19 deletions

View File

@@ -86,7 +86,7 @@ invoke_gemm(ck_tile::DeviceMem& a_m_k_dev_buf,
std::tuple<float, ck_tile::index_t> ave_time_and_batch;
if(args.reduction_strategy == ck_tile::StreamKReductionStrategy::Atomic)
if(reduction_strategy == ck_tile::StreamKReductionStrategy::Atomic)
{
ave_time_and_batch = gemm<GemmConfig,
ADataType,

View File

@@ -105,13 +105,13 @@ std::tuple<float, ck_tile::index_t> gemm(const ck_tile::StreamKHostArgs& args,
}
auto reset_data_buffers = [&]() {
if(ReductionStrategy == ck_tile::StreamKReductionStrategy::Atomic)
if constexpr(ReductionStrategy == ck_tile::StreamKReductionStrategy::Atomic)
{
// Clear the output C tensor results after each repetition of the kernel
hipGetErrorString(hipMemsetAsync(
args.e_ptr, 0, args.M * args.N * sizeof(CDataType), s.stream_id_));
}
else if(ReductionStrategy == ck_tile::StreamKReductionStrategy::Reduction)
else if constexpr(ReductionStrategy == ck_tile::StreamKReductionStrategy::Reduction)
{
// Reset sk flags to zero before each repetition of the kernel
workspace_data.SetZero();