xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-11 07:14:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Refactor] Remove unused variables in moe_permute_unpermute_kernel.inl (#19573)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-06-13 09:12:15 -04:00 committed by GitHub
parent a24cb91600
commit ce9dc02c93
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 0 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
csrc/moe/permute_unpermute_kernels/moe_permute_unpermute_kernel.inl
View File

@ -20,7 +20,6 @@ __global__ void expandInputRowsKernel(
int expert_id = sorted_experts[expanded_dest_row]; int expert_id = sorted_experts[expanded_dest_row];
extern __shared__ int64_t smem_expert_first_token_offset[]; extern __shared__ int64_t smem_expert_first_token_offset[];
int64_t align_expanded_row_accumulate = 0;
if constexpr (ALIGN_BLOCK_SIZE) { if constexpr (ALIGN_BLOCK_SIZE) {
// load g2s // load g2s
for (int idx = threadIdx.x; idx < num_local_experts + 1; for (int idx = threadIdx.x; idx < num_local_experts + 1;
@ -63,7 +62,6 @@ __global__ void expandInputRowsKernel(
using DataElem = cutlass::Array<T, ELEM_PER_THREAD>; using DataElem = cutlass::Array<T, ELEM_PER_THREAD>;
// Duplicate and permute rows // Duplicate and permute rows
int64_t const source_k_rank = expanded_source_row / num_rows;
int64_t const source_row = expanded_source_row % num_rows; int64_t const source_row = expanded_source_row % num_rows;
auto const* source_row_ptr = auto const* source_row_ptr =
@ -160,7 +158,6 @@ __global__ void finalizeMoeRoutingKernel(
elem_index += stride) { elem_index += stride) {
ComputeElem thread_output; ComputeElem thread_output;
thread_output.fill(0); thread_output.fill(0);
float row_rescale{0.f};
for (int k_idx = 0; k_idx < k; ++k_idx) { for (int k_idx = 0; k_idx < k; ++k_idx) {
int64_t const expanded_original_row = original_row + k_idx * num_rows; int64_t const expanded_original_row = original_row + k_idx * num_rows;
int64_t const expanded_permuted_row = int64_t const expanded_permuted_row =
@ -177,8 +174,6 @@ __global__ void finalizeMoeRoutingKernel(
auto const* expanded_permuted_rows_row_ptr = auto const* expanded_permuted_rows_row_ptr =
expanded_permuted_rows_v + expanded_permuted_row * num_elems_in_col; expanded_permuted_rows_v + expanded_permuted_row * num_elems_in_col;
int64_t const expert_idx = expert_for_source_row[k_offset];
ComputeElem expert_result = arrayConvert<InputElem, ComputeElem>( ComputeElem expert_result = arrayConvert<InputElem, ComputeElem>(
expanded_permuted_rows_row_ptr[elem_index]); expanded_permuted_rows_row_ptr[elem_index]);
thread_output = thread_output + row_scale * (expert_result); thread_output = thread_output + row_scale * (expert_result);