adasdadsd's picture
v2.0: Full float32 precision pipeline, headDim=64 support, 448/448 production tests passed
2145fce verified
Raw
History Blame Contribute Delete
827 Bytes
add_ops_compile_options(
OP_NAME SmallqFlashAttention
OPTIONS --cce-auto-sync=on
-Wno-deprecated-declarations
-mllvm -cce-aicore-hoist-movemask=false
--op_relocatable_kernel_binary=true
)
set(smallq_flash_attention_depends transformer/attention/smallq_flash_attention PARENT_SCOPE)
target_sources(op_host_aclnn PRIVATE
op_host/smallq_flash_attention_def.cpp
)
target_sources(optiling PRIVATE
op_host/smallq_flash_attention_tiling.cpp
)
if (NOT BUILD_OPEN_PROJECT)
target_sources(opmaster_ct PRIVATE
op_host/smallq_flash_attention_tiling.cpp
)
endif ()
target_include_directories(optiling PRIVATE
${CMAKE_CURRENT_SOURCE_DIR}/op_host
)
target_sources(opsproto PRIVATE
op_host/smallq_flash_attention_proto.cpp
)