[SYCL] supprt Flash Attention for fp32/fp16/Q4/Q5/Q8 (#20190)

* support flash-attention for fp32/fp16/Q4/Q5/Q8

* rm warining

* update for JIT
This commit is contained in:
Neo Zhang
2026-03-08 12:00:07 +08:00
committed by GitHub
parent c5a778891b
commit 213c4a0b81
65 changed files with 20091 additions and 8593 deletions
+6
View File
@@ -25,6 +25,11 @@ ggml_add_backend_library(ggml-sycl
file(GLOB GGML_HEADERS_SYCL "*.hpp")
file(GLOB GGML_SOURCES_SYCL "*.cpp")
file(GLOB SRCS "template-instances/fattn-tile*.cpp")
list(APPEND GGML_SOURCES_SYCL ${SRCS})
file(GLOB SRCS "template-instances/fattn-vec*.cpp")
list(APPEND GGML_SOURCES_SYCL ${SRCS})
target_sources(ggml-sycl PRIVATE ${GGML_HEADERS_SYCL} ${GGML_SOURCES_SYCL})
if (WIN32)
@@ -145,6 +150,7 @@ else()
endif()
if (GGML_SYCL_GRAPH)
message(STATUS "find GGML_SYCL_GRAPH")
target_compile_definitions(ggml-sycl PRIVATE GGML_SYCL_GRAPH)
endif()