impl gemm2 swizzle

This commit is contained in:
coderfeli
2025-02-17 02:33:52 +00:00
parent aecd6a38e4
commit fcc2c867af
2 changed files with 21 additions and 6 deletions

View File

@@ -236,7 +236,7 @@ int main(int argc, char* argv[])
// const ck::index_t experts = 8;
Tensor<ck::index_t> expert_ids(HostTensorDescriptor({experts}, {1}));
Tensor<ck::index_t> expert_ids(HostTensorDescriptor({sorted_tile_num}, {1}));
Tensor<ck::index_t> sorted_token_ids(HostTensorDescriptor({sorted_size}, {1}));
Tensor<ck::index_t> max_token_id(HostTensorDescriptor({1}));
max_token_id.mData[0] = valid_size;