biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc][Easy] Annotate unused vars in the csrc files (#14798)

Browse Source 
Signed-off-by: Lu Fang <lufang@fb.com>
This commit is contained in:
Lu Fang
2025-03-14 21:40:09 -07:00
committed by GitHub
parent 97ac781c62
commit 8c0d15d5c5
4 changed files with 14 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
csrc/rocm/attention.cu
View File

@@ -308,8 +308,8 @@ __launch_bounds__(NUM_THREADS, 5) void paged_attention_ll4mi_QKV_mfma16_kernel(
constexpr int GQA_RATIO4 = DIVIDE_ROUND_UP(GQA_RATIO, 4);
__shared__ float shared_qk_max[NWARPS][16 + 1];
__shared__ float shared_exp_sum[NWARPS][16 + 1];
[[maybe_unused]] __shared__ float shared_qk_max[NWARPS][16 + 1];
[[maybe_unused]] __shared__ float shared_exp_sum[NWARPS][16 + 1];
// shared_logits is used for multiple purposes
__shared__ _B16x4 shared_logits[NWARPS][4][16][4];
@@ -426,7 +426,8 @@ __launch_bounds__(NUM_THREADS, 5) void paged_attention_ll4mi_QKV_mfma16_kernel(
const cache_t* k_ptr2 = k_ptr + kblock_number * kv_block_stride;
const int klocal_token_idx =
TOKENS_PER_WARP * warpid + token_depth * 16 + lane16id;
const int kglobal_token_idx = partition_start_token_idx + klocal_token_idx;
[[maybe_unused]] const int kglobal_token_idx =
partition_start_token_idx + klocal_token_idx;
const int kphysical_block_offset = klocal_token_idx % BLOCK_SIZE;
const cache_t* k_ptr3 = k_ptr2 + kphysical_block_offset * KX;