kernel: support slightly faster merge_state_v2 cuda kernel (#5381)

This commit is contained in:
DefTruth
2025-04-15 12:28:23 +08:00
committed by GitHub
parent 11421a3f44
commit 388e15c0db
7 changed files with 638 additions and 4 deletions

View File

@@ -170,6 +170,7 @@ string(REPLACE "-D__CUDA_NO_HALF2_OPERATORS__" "" CMAKE_CUDA_FLAGS "${CMAKE
set(SOURCES
"csrc/allreduce/custom_all_reduce.cu"
"csrc/attention/cascade.cu"
"csrc/attention/merge_attn_states.cu"
"csrc/attention/cutlass_mla_kernel.cu"
"csrc/attention/lightning_attention_decode_kernel.cu"
"csrc/elementwise/activation.cu"