kleidiai: add and integrate SVE 256-bit vector-length kernel (#18458)

* kleidiai: add and integrate SVE 256-bit vector-length kernel

* updated for review comments
This commit is contained in:
Charles Xu
2025-12-30 14:04:53 +02:00
committed by GitHub
parent d77d7c5c06
commit 2d6c00a9b8
4 changed files with 115 additions and 37 deletions
+12 -2
View File
@@ -561,9 +561,9 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
# Fetch KleidiAI sources:
include(FetchContent)
set(KLEIDIAI_COMMIT_TAG "v1.14.0")
set(KLEIDIAI_COMMIT_TAG "v1.16.0")
set(KLEIDIAI_DOWNLOAD_URL "https://github.com/ARM-software/kleidiai/archive/refs/tags/${KLEIDIAI_COMMIT_TAG}.tar.gz")
set(KLEIDIAI_ARCHIVE_MD5 "45e110675d93f99f82c23a1afcca76bc")
set(KLEIDIAI_ARCHIVE_MD5 "0a9e9008adb6031f9e8cf70dff4a3321")
if (POLICY CMP0135)
cmake_policy(SET CMP0135 NEW)
@@ -615,6 +615,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
string(FIND "${ARCH_FLAGS_TEMP}" "+dotprod" DOTPROD_ENABLED)
string(FIND "${ARCH_FLAGS_TEMP}" "+i8mm" I8MM_ENABLED)
string(FIND "${ARCH_FLAGS_TEMP}" "+sme" SME_ENABLED)
string(FIND "${ARCH_FLAGS_TEMP}" "+sve" SVE_ENABLED)
set(PRIVATE_ARCH_FLAGS ${ARCH_FLAGS_TEMP})
@@ -659,6 +660,15 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
set(PRIVATE_ARCH_FLAGS "-fno-tree-vectorize;${PRIVATE_ARCH_FLAGS}+sve+sve2")
endif()
if (NOT SVE_ENABLED MATCHES -1)
list(APPEND GGML_KLEIDIAI_SOURCES
${KLEIDIAI_SRC}/kai/kai_common_sve_asm.S
${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x8_qsi4c32p8x8_1x8_sve_dotprod_asm.S
${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x8_qsi4c32p8x8_1x8_sve_dotprod.c
${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p4x8_qsi4c32p8x8_16x8_sve_i8mm_asm.S
${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p4x8_qsi4c32p8x8_16x8_sve_i8mm.c)
endif()
set_source_files_properties(${GGML_KLEIDIAI_SOURCES} PROPERTIES COMPILE_OPTIONS "${PRIVATE_ARCH_FLAGS}")
list(APPEND GGML_CPU_SOURCES ${GGML_KLEIDIAI_SOURCES})
endif()