[SYCL] Fix the sub group size of Intel (ggerganov#8106)

* use warp_size macro for all sycl kernels * fix mask of permute_sub_group_by_xor * fix rms_norm with correct warp number * fix rms_norm_f32/group_norm_f32 * move norm to norm.cpp file * fix quantize bug * fix mmvq's batch size
arthw · Jul 2, 2024 · d08c20e · d08c20e
1 parent 5fac350
commit d08c20e
Show file tree

Hide file tree

Showing 9 changed files with 587 additions and 509 deletions.
diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt
@@ -486,9 +486,11 @@ if (GGML_SYCL)
     add_compile_options(-I./) #include DPCT
 
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-narrowing")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3")
     if (GGML_SYCL_TARGET STREQUAL "NVIDIA")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl-targets=nvptx64-nvidia-cuda")
+        add_compile_definitions(GGML_SYCL_WARP_SIZE=32)
+    else()
+        add_compile_definitions(GGML_SYCL_WARP_SIZE=16)
     endif()
 
     file(GLOB   GGML_HEADERS_SYCL "ggml-sycl/*.hpp")