Skip to content

LLM inference in C/C++

C++ ggml

sycl(build): parallelize ocloc invocations (#25903)

Titaniumtown8e8681e0e208
diff --git a/ggml/src/ggml-sycl/CMakeLists.txt b/ggml/src/ggml-sycl/CMakeLists.txt
index 1c17d20df1..a8d9c0d804 100644
--- a/ggml/src/ggml-sycl/CMakeLists.txt
+++ b/ggml/src/ggml-sycl/CMakeLists.txt
@@ -199,9 +199,20 @@ if (GGML_SYCL_DEVICE_ARCH)
-fsycl-targets=spir64_gen
"SHELL:-Xsycl-target-backend=spir64_gen \"-device ${GGML_SYCL_DEVICE_ARCH}\""
)
+
+ # Pass through parallel job (process) count for parallelising the
+ # `llvm-foreach -- ocloc` invocation for compiling AOT device images.
+ include(ProcessorCount)
+ ProcessorCount(_ggml_sycl_nproc)
+ if (_ggml_sycl_nproc LESS 1)
+ set(_ggml_sycl_nproc 1)
+ endif()
+ set(GGML_SYCL_MAX_PARALLEL_LINK_JOBS ${_ggml_sycl_nproc} CACHE STRING
+ "Parallel ocloc jobs for spir64_gen AOT device-image lowering")
target_link_options(
ggml-sycl PRIVATE
-fsycl-targets=spir64_gen
"SHELL:-Xsycl-target-backend=spir64_gen \"-device ${GGML_SYCL_DEVICE_ARCH}\""
+ -fsycl-max-parallel-link-jobs=${GGML_SYCL_MAX_PARALLEL_LINK_JOBS}
)
endif()