mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-19 22:39:03 +00:00
fix the flatmm (#2349)
This commit is contained in:
@@ -447,6 +447,7 @@ struct FlatmmKernel
|
||||
// Run GEMM cooperatively by whole workgroup.
|
||||
const auto& a_block_window = gemm_tile_windows.at(I0);
|
||||
const auto& b_flat_block_window = gemm_tile_windows.at(I1);
|
||||
const auto& d_block_window = gemm_tile_windows.at(I2);
|
||||
const auto& c_block_tile = FlatmmPipeline{}.template operator()(
|
||||
a_block_window, b_flat_block_window, num_loop, smem_ptr);
|
||||
|
||||
@@ -454,7 +455,7 @@ struct FlatmmKernel
|
||||
auto& c_block_window = gemm_tile_windows.at(I2);
|
||||
|
||||
EpiloguePipeline{}.template operator()<decltype(c_block_window), decltype(c_block_tile)>(
|
||||
c_block_window, c_block_tile, smem_ptr);
|
||||
c_block_window, c_block_tile, d_block_window, smem_ptr);
|
||||
}
|
||||
|
||||
CK_TILE_DEVICE void operator()(FlatmmKernelArgs kargs) const
|
||||
|
||||
Reference in New Issue
Block a user