[Bugfix][Kernel] Fix perf regression caused by PR #12405 (#12434)

Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>

Author: LucasWilkinson

CMakeLists.txt

@@ -576,7 +576,7 @@ else()
   FetchContent_Declare(
           vllm-flash-attn
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
-          GIT_TAG 9732b0ce005d1e6216864788502d5570004678f5
+          GIT_TAG d4e09037abf588af1ec47d0e966b237ee376876c
           GIT_PROGRESS TRUE
           # Don't share the vllm-flash-attn build between build types
           BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn