[CK TILE] Grouped Conv Explicit Gemm (#3289)

* [CK TILE] Grouped Conv Explicit Gemm

* fixes

* apply builder fixes
This commit is contained in:
Bartłomiej Kocot
2025-11-25 23:28:35 +01:00
committed by GitHub
parent 37ea160088
commit 00dfa2f2ce
13 changed files with 386 additions and 269 deletions

View File

@@ -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);
}
}
};