mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-20 06:49:15 +00:00
[CK TILE] Grouped Conv Explicit Gemm (#3289)
* [CK TILE] Grouped Conv Explicit Gemm * fixes * apply builder fixes
This commit is contained in:
@@ -228,10 +228,34 @@ struct BatchedGemmKernel
|
||||
CDataType* c_ptr = static_cast<CDataType*>(kargs.e_ptr) + batch_offset_C;
|
||||
|
||||
// allocate LDS
|
||||
__shared__ char smem_ptr[GetSmemSize()];
|
||||
__shared__ char smem_ptr0[GetSmemSize()];
|
||||
|
||||
UniversalGemmKernel::RunGemm(
|
||||
{a_ptr}, {b_ptr}, {/*ds_ptr*/}, c_ptr, smem_ptr, kargs, splitk_batch_offset, i_m, i_n);
|
||||
if constexpr(GemmPipeline::DoubleSmemBuffer == true)
|
||||
{
|
||||
__shared__ char smem_ptr1[GetSmemSize()];
|
||||
UniversalGemmKernel::RunGemm2LDS({a_ptr},
|
||||
{b_ptr},
|
||||
{/*ds_ptr*/},
|
||||
c_ptr,
|
||||
smem_ptr0,
|
||||
smem_ptr1,
|
||||
kargs,
|
||||
splitk_batch_offset,
|
||||
i_m,
|
||||
i_n);
|
||||
}
|
||||
else
|
||||
{
|
||||
UniversalGemmKernel::RunGemm({a_ptr},
|
||||
{b_ptr},
|
||||
{/*ds_ptr*/},
|
||||
c_ptr,
|
||||
smem_ptr0,
|
||||
kargs,
|
||||
splitk_batch_offset,
|
||||
i_m,
|
||||
i_n);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
Reference in New Issue
Block a user