mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-03-25 09:37:42 +00:00
Fix the bug
This commit is contained in:
@@ -258,14 +258,14 @@ struct BlockGemmARegBRegCRegV1
|
||||
// read A warp tensor from A Block window
|
||||
AWarpTensor a_warp_tensor;
|
||||
a_warp_tensor.get_thread_buffer() = a_block_tensor.get_y_sliced_thread_data(
|
||||
merge_sequences(sequence<mIter, kIter>{}, a_warp_y_index_zeros),
|
||||
merge_sequences(sequence<kIter, mIter>{}, a_warp_y_index_zeros),
|
||||
merge_sequences(sequence<1, 1>{}, a_warp_y_lengths));
|
||||
|
||||
static_for<0, NIterPerWarp, 1>{}([&](auto nIter) {
|
||||
// read B warp tensor from B block tensor
|
||||
BWarpTensor b_warp_tensor;
|
||||
b_warp_tensor.get_thread_buffer() = b_block_tensor.get_y_sliced_thread_data(
|
||||
merge_sequences(sequence<nIter, kIter>{}, b_warp_y_index_zeros),
|
||||
merge_sequences(sequence<kIter, nIter>{}, b_warp_y_index_zeros),
|
||||
merge_sequences(sequence<1, 1>{}, b_warp_y_lengths));
|
||||
|
||||
// read C warp tensor from C block tensor
|
||||
|
||||
Reference in New Issue
Block a user