Update the KleidiAI version to r0.5.0.

source/backend/cpu/CMakeLists.txt

@@ -49,11 +49,3 @@ IF(MNN_ARM82)
         list(APPEND MNN_OBJECTS_TO_LINK $<TARGET_OBJECTS:MNN_Arm82>)
     ENDIF()
 ENDIF()
-
-# Kleidi AI
-IF(MNN_KLEIDIAI)
-  add_definitions(-DMNN_KLEIDIAI_ENABLED=1)
-  include(${CMAKE_CURRENT_LIST_DIR}/arm/kleidiAI/CMakeLists.txt)
-  list(APPEND MNN_TARGETS MNN_KleidiAI)
-  list(APPEND MNN_OBJECTS_TO_LINK $<TARGET_OBJECTS:MNN_KleidiAI>)
-ENDIF()

source/backend/cpu/CPUBackend.hpp

@@ -18,7 +18,7 @@
 #include "MNN_generated.h"
 
 #ifdef MNN_KLEIDIAI_ENABLED
-#include "arm/kleidiAI/mnn_kleidiai.h"
+#include "arm/mnn_kleidiai.h"
 #endif
 
 namespace MNN {

source/backend/cpu/arm/CMakeLists.txt

@@ -19,6 +19,55 @@ if (MNN_CPU_WEIGHT_DEQUANT_GEMM)
     FILE(GLOB MNN_AArch64_SRC ${MNN_AArch64_SRC} ${CMAKE_CURRENT_LIST_DIR}/arm64/normal_memory/*.[sS])
 endif()
 
+if (MNN_KLEIDIAI)
+    add_definitions(-DMNN_KLEIDIAI_ENABLED=1)
+    # Disable the KleidiAI tests
+    set(KLEIDIAI_BUILD_TESTS  OFF)
+    # Fetch KleidiAI sources:
+    include(FetchContent)
+    set(KLEIDIAI_COMMIT_SHA "v0.5.0")
+    set(KLEIDIAI_DOWNLOAD_URL "${CMAKE_CURRENT_LIST_DIR}/kleidiai-v0.5.0.tar.gz")
+    set(KLEIDIAI_ARCHIVE_MD5  "b75f2d3c1bd2a5457091cc3976e2fdde")
+
+    if (POLICY CMP0135)
+        cmake_policy(SET CMP0135 NEW)
+    endif()
+    FetchContent_Declare(KleidiAI_Download
+        URL ${KLEIDIAI_DOWNLOAD_URL}
+        DOWNLOAD_EXTRACT_TIMESTAMP NEW
+        URL_HASH MD5=${KLEIDIAI_ARCHIVE_MD5})
+
+    FetchContent_MakeAvailable(KleidiAI_Download)
+    FetchContent_GetProperties(KleidiAI_Download
+        SOURCE_DIR  KLEIDIAI_SRC
+        POPULATED   KLEIDIAI_POPULATED)
+
+    if (NOT KLEIDIAI_POPULATED)
+        message(FATAL_ERROR "KleidiAI source downloaded failed.")
+    endif()
+
+    list(APPEND MNN_SOURCES_KLEIDIAI ${CMAKE_CURRENT_LIST_DIR}/mnn_kleidiai.cpp)
+    list(APPEND MNN_HEADERS_KLEIDIAI ${CMAKE_CURRENT_LIST_DIR}/mnn_kleidiai.h)
+
+    # KleidiAI
+    include_directories(
+        ${KLEIDIAI_SRC}/
+        ${KLEIDIAI_SRC}/kai/
+        ${KLEIDIAI_SRC}/kai/ukernels/
+        ${KLEIDIAI_SRC}/kai/ukernels/matmul/
+        ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qai8dxp_qsi4cxp/
+        ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/)
+
+    list(APPEND MNN_SOURCES_KLEIDIAI ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qai8dxp_f32.c)
+    list(APPEND MNN_SOURCES_KLEIDIAI ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4cxp_qs4cxs1s0.c)
+    list(APPEND MNN_SOURCES_KLEIDIAI ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qai8dxp_qsi4cxp/kai_matmul_clamp_f32_qai8dxp1x8_qsi4cxp4x8_1x4x32_neon_dotprod.c)
+    list(APPEND MNN_SOURCES_KLEIDIAI ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qai8dxp_qsi4cxp/kai_matmul_clamp_f32_qai8dxp4x8_qsi4cxp4x8_8x4x32_neon_i8mm.c)
+    list(APPEND MNN_CDEF_PUBLIC MNN_USE_KLEIDIAI)
+
+    set_source_files_properties(${MNN_SOURCES_KLEIDIAI} PROPERTIES COMPILE_OPTIONS -march=armv8.2-a+i8mm+dotprod+sve+sve2+fp16)
+
+endif()
+
 if(CMAKE_SYSTEM_PROCESSOR MATCHES "^armv7" OR ARCHS MATCHES "^armv7(;armv7s)?")
     message(STATUS "Enabling AArch32 Assemblies")
     add_library(MNNARM32 OBJECT ${MNN_AArch32_SRC} ${MNN_NEON_SRC})
@@ -32,8 +81,9 @@ if(CMAKE_SYSTEM_PROCESSOR MATCHES "^armv7" OR ARCHS MATCHES "^armv7(;armv7s)?")
     endif()
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^aarch64" OR ARCHS STREQUAL "arm64" OR ARCHS STREQUAL "ARM64")
     message(STATUS "Enabling AArch64 Assemblies")
-    add_library(MNNARM64 OBJECT ${MNN_AArch64_SRC} ${MNN_NEON_SRC})
+    add_library(MNNARM64 OBJECT ${MNN_AArch64_SRC} ${MNN_NEON_SRC} ${MNN_SOURCES_KLEIDIAI})
     target_include_directories(MNNARM64 PRIVATE ${CMAKE_CURRENT_LIST_DIR}/)
+    target_include_directories(MNNARM64 PRIVATE ${MNN_HEADERS_KLEIDIAI})
     list(APPEND MNN_OBJECTS_TO_LINK $<TARGET_OBJECTS:MNNARM64>)
     list(APPEND MNN_TARGETS MNNARM64)
     add_definitions(-DMNN_USE_NEON)