[Perf][DSV4] Optimize sparse top-k metadata kernels for higher prefill throughput (#52084)

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
This commit is contained in:
Chauncey
2026-08-16 21:15:47 +08:00
committed by GitHub
parent 4d2a68d64d
commit 836aac92ff
@@ -580,7 +580,7 @@ def combine_topk_swa_indices(
return combined_indices, combined_lens
_COMBINE_TOPK_SWA_NUM_WORKERS = 128
_COMBINE_TOPK_SWA_NUM_WORKERS = 256
# Representative pointer alignment variants for Triton pointer specialization.