fix xnnpack build

cmake/deps.txt

@@ -45,9 +45,9 @@ protoc_linux_x86;https://github.com/protocolbuffers/protobuf/releases/download/v
 protoc_linux_aarch64;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-linux-aarch_64.zip;df9d45470b0b8cf939dd2f0ec6b88e9cafc4d617
 protoc_mac_universal;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-osx-universal_binary.zip;23710c3d1c2036d8d65a6a22234372fa2d7af9ef
 psimd;https://github.com/Maratyszcza/psimd/archive/072586a71b55b7f8c584153d223e95687148a900.zip;1f5454b01f06f9656b77e4a5e2e31d7422487013
-pthreadpool;https://github.com/google/pthreadpool/archive/4e80ca24521aa0fb3a746f9ea9c3eaa20e9afbb0.zip;bd4ea65c8292801e9555b527a0ecbb2e0092c917
+pthreadpool;https://github.com/google/pthreadpool/archive/dcc9f28589066af0dbd4555579281230abbf74dd.zip;533a77943203ef15ca608bcd9dbe2c94da7451d2
 pybind11;https://github.com/pybind/pybind11/archive/refs/tags/v2.13.6.zip;f780292da9db273c8ef06ccf5fd4b623624143e9
-pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/8a1772a0c5c447df2d18edf33ec4603a8c9c04a6.zip;85bf8a60dae026b99b6ccd78606c85ed83bfb2cd
+pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/de0ce7c7251372892e53ce9bc891750d2c9a4fd8.zip;c45b8d3619b9bccbd26dc5f657959aee38b18b7a
 re2;https://github.com/google/re2/archive/refs/tags/2024-07-02.zip;646e1728269cde7fcef990bf4a8e87b047882e88
 safeint;https://github.com/dcleblanc/SafeInt/archive/refs/tags/3.0.28.zip;23f252040ff6cb9f1fd18575b32fa8fb5928daac
 tensorboard;https://github.com/tensorflow/tensorboard/archive/373eb09e4c5d2b3cc2493f0949dc4be6b6a45e81.zip;67b833913605a4f3f499894ab11528a702c2b381

cmake/external/onnxruntime_external_deps.cmake

@@ -570,7 +570,7 @@ if (onnxruntime_USE_XNNPACK)
      ENDIF()
      ADD_LIBRARY(xnnpack STATIC IMPORTED)
      find_library(xnnpack_LIBRARY NAMES XNNPACK)
-     find_library(microkernels_prod_LIBRARY NAMES microkernels-prod)
+     find_library(microkernels_prod_LIBRARY NAMES xnnpack-microkernels-prod)
      find_package(unofficial-pthreadpool CONFIG REQUIRED)
 
      target_include_directories(xnnpack INTERFACE "${XNNPACK_HDR}")

cmake/external/xnnpack.cmake

@@ -90,7 +90,7 @@ onnxruntime_fetchcontent_makeavailable(googlexnnpack)
 set(XNNPACK_DIR ${googlexnnpack_SOURCE_DIR})
 set(XNNPACK_INCLUDE_DIR ${XNNPACK_DIR}/include)
 
-set(onnxruntime_EXTERNAL_LIBRARIES_XNNPACK XNNPACK microkernels-prod pthreadpool)
+set(onnxruntime_EXTERNAL_LIBRARIES_XNNPACK XNNPACK xnnpack-microkernels-prod pthreadpool)
 if(ORT_TARGET_PROCESSOR MATCHES "^arm64.*" AND NOT CMAKE_C_COMPILER_ID STREQUAL "MSVC")
   list(APPEND onnxruntime_EXTERNAL_LIBRARIES_XNNPACK kleidiai)
 endif()

cmake/patches/xnnpack/AddEmscriptenAndIosSupport.patch

@@ -1,8 +1,8 @@
 diff --git a/CMakeLists.txt b/CMakeLists.txt
-index f0b3410ae..1e3cb8178 100644
+index 94bcad92e3..be7dfe95fd 100644
 --- a/CMakeLists.txt
 +++ b/CMakeLists.txt
-@@ -337,7 +337,7 @@ ENDIF()
+@@ -360,7 +360,7 @@ ENDIF()
  # ---[ Build flags
  IF(NOT CMAKE_SYSTEM_NAME)
    MESSAGE(FATAL_ERROR "CMAKE_SYSTEM_NAME not defined")
@@ -11,21 +11,30 @@ index f0b3410ae..1e3cb8178 100644
    MESSAGE(FATAL_ERROR "Unrecognized CMAKE_SYSTEM_NAME value \"${CMAKE_SYSTEM_NAME}\"")
  ENDIF()
  IF(CMAKE_SYSTEM_NAME MATCHES "Windows")
-@@ -848,7 +848,12 @@ IF(XNNPACK_BUILD_LIBRARY)
-   TARGET_LINK_LIBRARIES(operator-utils PRIVATE xnnpack-base logging)
-   TARGET_LINK_LIBRARIES(reference-ukernels PRIVATE xnnpack-base)
-   TARGET_LINK_LIBRARIES(subgraph PRIVATE xnnpack-base allocator logging memory mutex operators operator-run datatype)
--  TARGET_LINK_LIBRARIES(XNNPACK PRIVATE xnnpack-base allocator cache hardware-config indirection memory microkernel-utils microparams-init mutex normalization operators operator-run operator-utils packing microkernels-prod subgraph datatype reference-ukernels)
+@@ -903,10 +903,18 @@ IF(XNNPACK_BUILD_LIBRARY)
+   TARGET_LINK_LIBRARIES(xnnpack-operator-utils PRIVATE xnnpack-base xnnpack-logging)
+   TARGET_LINK_LIBRARIES(xnnpack-reference-ukernels PRIVATE xnnpack-base xnnpack-datatype)
+   TARGET_LINK_LIBRARIES(xnnpack-subgraph PRIVATE xnnpack-base xnnpack-allocator xnnpack-logging xnnpack-memory xnnpack-mutex xnnpack-operators xnnpack-operator-run xnnpack-datatype)
+-  TARGET_LINK_LIBRARIES(XNNPACK PRIVATE xnnpack-base xnnpack-allocator xnnpack-cache
+-      xnnpack-hardware-config xnnpack-indirection xnnpack-memory xnnpack-microkernel-utils xnnpack-microparams-init
+-      xnnpack-mutex xnnpack-normalization xnnpack-operators xnnpack-operator-run xnnpack-operator-utils xnnpack-pack-lh xnnpack-packing
+-      xnnpack-microkernels-prod xnnpack-subgraph xnnpack-datatype xnnpack-reference-ukernels)
 +  IF(CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
 +    # omit microkernels-prod as the list is manually created by ORT in cmake/external/xnnpack.cmake
-+    TARGET_LINK_LIBRARIES(XNNPACK PRIVATE xnnpack-base allocator cache hardware-config indirection memory microkernel-utils microparams-init mutex normalization operators operator-run operator-utils packing subgraph datatype reference-ukernels)
++    TARGET_LINK_LIBRARIES(XNNPACK PRIVATE xnnpack-base xnnpack-allocator xnnpack-cache
++        xnnpack-hardware-config xnnpack-indirection xnnpack-memory xnnpack-microkernel-utils xnnpack-microparams-init
++        xnnpack-mutex xnnpack-normalization xnnpack-operators xnnpack-operator-run xnnpack-operator-utils xnnpack-pack-lh xnnpack-packing
++        xnnpack-subgraph xnnpack-datatype xnnpack-reference-ukernels)
 +  ELSE()
-+    TARGET_LINK_LIBRARIES(XNNPACK PRIVATE xnnpack-base allocator cache hardware-config indirection memory microkernel-utils microparams-init mutex normalization operators operator-run operator-utils packing microkernels-prod subgraph datatype reference-ukernels)
++    TARGET_LINK_LIBRARIES(XNNPACK PRIVATE xnnpack-base xnnpack-allocator xnnpack-cache
++        xnnpack-hardware-config xnnpack-indirection xnnpack-memory xnnpack-microkernel-utils xnnpack-microparams-init
++        xnnpack-mutex xnnpack-normalization xnnpack-operators xnnpack-operator-run xnnpack-operator-utils xnnpack-pack-lh xnnpack-packing
++        xnnpack-microkernels-prod xnnpack-subgraph xnnpack-datatype xnnpack-reference-ukernels)
 +  ENDIF()
-   TARGET_LINK_LIBRARIES(XNNPACK PUBLIC pthreadpool logging)
+   TARGET_LINK_LIBRARIES(XNNPACK PUBLIC pthreadpool xnnpack-logging)
    SET_TARGET_PROPERTIES(XNNPACK PROPERTIES C_EXTENSIONS YES)
  ENDIF()
-@@ -857,7 +862,8 @@ IF(NOT MSVC)
+@@ -915,7 +923,8 @@ IF(NOT MSVC)
  ENDIF()
  IF(XNNPACK_TARGET_PROCESSOR STREQUAL "arm")
    SET_PROPERTY(SOURCE ${ALL_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -marm ")
@@ -35,3 +44,6 @@ index f0b3410ae..1e3cb8178 100644
    SET_PROPERTY(SOURCE ${ALL_NEON_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv7-a -mfpu=neon ")
    SET_PROPERTY(SOURCE ${ALL_NEONFP16_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv7-a -mfpu=neon-fp16 ")
    # GCC requires -mfp16-format=ieee to define __fp16 type, but Clang doesn't support this option at all.
+-- 
+2.46.1
+

cmake/vcpkg-ports/cpuinfo/portfile.cmake

@@ -6,8 +6,8 @@ endif()
 vcpkg_from_github(
     OUT_SOURCE_PATH SOURCE_PATH
     REPO pytorch/cpuinfo
-    REF 8a1772a0c5c447df2d18edf33ec4603a8c9c04a6
-    SHA512 b94ccbfa886221d6bb16513d074675af0a72928a9dd9485dcacdc1124a8a60aacbbe91913a1579e766dfb024f0be1d52eeead40342004ff0238a8b94a095ed08
+    REF de0ce7c7251372892e53ce9bc891750d2c9a4fd8
+    SHA512 0fde9210b700d2648d37c8deeb0d5c0d007d8ca5689578dd3bce4c460886b20d7649f0194d2ea06b02238fe9d4f06193599ec3ab5cafb19f1f860b00404264fa
     HEAD_REF master
 )
 

cmake/vcpkg-ports/pthreadpool/fix-cmakelists.patch

@@ -1,8 +1,17 @@
+From 3621e82fb4b46cfe1960cf13fdea92e8760a305d Mon Sep 17 00:00:00 2001
+From: Fanchen Kong <[email protected]>
+Date: Wed, 2 Jul 2025 09:58:02 +0800
+Subject: [PATCH] fix-cmakelists
+
+---
+ CMakeLists.txt | 29 ++++++++++++-----------------
+ 1 file changed, 12 insertions(+), 17 deletions(-)
+
 diff --git a/CMakeLists.txt b/CMakeLists.txt
-index f06aada..3c6c6e2 100644
+index efff8cc..1a0f7e9 100644
 --- a/CMakeLists.txt
 +++ b/CMakeLists.txt
-@@ -31,8 +31,6 @@ IF(CCACHE_BINARY)
+@@ -41,8 +41,6 @@ IF(CMAKE_C_COMPILER_ID STREQUAL "MSVC")
  ENDIF()
 
  # ---[ Options.
@@ -11,7 +20,7 @@ index f06aada..3c6c6e2 100644
  OPTION(PTHREADPOOL_ALLOW_DEPRECATED_API "Enable deprecated API functions" ON)
  SET(PTHREADPOOL_SYNC_PRIMITIVE "default" CACHE STRING "Synchronization primitive (condvar, futex, gcd, event, or default) for worker threads")
  SET_PROPERTY(CACHE PTHREADPOOL_SYNC_PRIMITIVE PROPERTY STRINGS default condvar futex gcd event)
-@@ -41,7 +39,7 @@ IF(CMAKE_SYSTEM_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?)$")
+@@ -51,7 +49,7 @@ IF(CMAKE_SYSTEM_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?)$")
  ELSE()
    OPTION(PTHREADPOOL_ENABLE_FASTPATH "Enable fast path using atomic decrement instead of atomic compare-and-swap" OFF)
  ENDIF()
@@ -20,8 +29,8 @@ index f06aada..3c6c6e2 100644
    OPTION(PTHREADPOOL_BUILD_TESTS "Build pthreadpool unit tests" ON)
    OPTION(PTHREADPOOL_BUILD_BENCHMARKS "Build pthreadpool micro-benchmarks" ON)
  ELSE()
-@@ -67,7 +65,8 @@ MACRO(PTHREADPOOL_TARGET_ENABLE_CXX11 target)
- ENDMACRO()
+@@ -71,7 +69,8 @@ IF(PTHREADPOOL_BUILD_TESTS)
+ ENDIF()
 
  # ---[ Download deps
 -IF(NOT DEFINED FXDIV_SOURCE_DIR)
@@ -30,7 +39,7 @@ index f06aada..3c6c6e2 100644
    MESSAGE(STATUS "Downloading FXdiv to ${CMAKE_BINARY_DIR}/FXdiv-source (define FXDIV_SOURCE_DIR to avoid it)")
    CONFIGURE_FILE(cmake/DownloadFXdiv.cmake "${CMAKE_BINARY_DIR}/FXdiv-download/CMakeLists.txt")
    EXECUTE_PROCESS(COMMAND "${CMAKE_COMMAND}" -G "${CMAKE_GENERATOR}" .
-@@ -118,21 +117,13 @@ ELSE()
+@@ -122,21 +121,13 @@ ELSE()
  ENDIF()
 
  ADD_LIBRARY(pthreadpool_interface INTERFACE)
@@ -54,7 +63,7 @@ index f06aada..3c6c6e2 100644
 
  IF(PTHREADPOOL_SYNC_PRIMITIVE STREQUAL "condvar")
    TARGET_COMPILE_DEFINITIONS(pthreadpool PRIVATE PTHREADPOOL_USE_FUTEX=0)
-@@ -181,18 +172,22 @@ IF(CMAKE_SYSTEM_NAME STREQUAL "Linux")
+@@ -182,18 +173,22 @@ IF(CMAKE_SYSTEM_NAME STREQUAL "Linux")
  ENDIF()
 
  # ---[ Configure FXdiv
@@ -80,3 +89,6 @@ index f06aada..3c6c6e2 100644
 
  IF(PTHREADPOOL_BUILD_TESTS)
    # ---[ Build google test
+-- 
+2.46.1
+

cmake/vcpkg-ports/pthreadpool/portfile.cmake

@@ -5,8 +5,8 @@ endif()
 vcpkg_from_github(
     OUT_SOURCE_PATH SOURCE_PATH
     REPO google/pthreadpool
-    REF 4e80ca24521aa0fb3a746f9ea9c3eaa20e9afbb0
-    SHA512 776017cc5d2aa94337292f2f4fbd54d099ef29abf736ab8147f07f98f12b7654cbd2fe38d34646a479a519c261ac253bbaf19c6dcbb0ec4cc0859de70f7e6472
+    REF dcc9f28589066af0dbd4555579281230abbf74dd 
+    SHA512 61853fa8f6c3297d8760be3af1df3f2a00583c1e0e58bdd03cd9cb915e8660a4f2817b22e6463cf53f10de902a1c6204ec6054fcbeada72eeee9e44baeb97178
     PATCHES
         fix-cmakelists.patch
 )

cmake/vcpkg-ports/xnnpack/fix-build.patch

@@ -1,21 +1,26 @@
+From 715b9463289feef7d6bf736502a166bd00d70785 Mon Sep 17 00:00:00 2001
+From: Fanchen Kong <[email protected]>
+Date: Tue, 1 Jul 2025 13:42:35 +0800
+Subject: [PATCH] fix-build
+
+---
+ CMakeLists.txt | 29 ++++++++++-------------------
+ 1 file changed, 10 insertions(+), 19 deletions(-)
+
 diff --git a/CMakeLists.txt b/CMakeLists.txt
-index f0b3410ae..ba54c3bfe 100644
+index 9f6fb5e256..4387298e59 100644
 --- a/CMakeLists.txt
 +++ b/CMakeLists.txt
-@@ -1047,9 +1047,11 @@ ENDIF()
- IF(XNNPACK_BUILD_ALL_MICROKERNELS)
-   TARGET_INCLUDE_DIRECTORIES(microkernels-all PRIVATE include src)
+@@ -1125,7 +1125,7 @@ ELSE()
  ENDIF()
-+
- TARGET_INCLUDE_DIRECTORIES(datatype PRIVATE include src)
- TARGET_INCLUDE_DIRECTORIES(microkernels-prod PRIVATE include src)
--TARGET_INCLUDE_DIRECTORIES(hardware-config PRIVATE include src ${CPUINFO_SOURCE_DIR}/include)
-+TARGET_INCLUDE_DIRECTORIES(hardware-config PRIVATE include src)
-+
- TARGET_INCLUDE_DIRECTORIES(indirection PRIVATE include src)
- TARGET_INCLUDE_DIRECTORIES(microparams-init PRIVATE include src)
- TARGET_INCLUDE_DIRECTORIES(normalization PRIVATE include src)
-@@ -1104,14 +1106,9 @@ IF(NOT TARGET cpuinfo)
+
+ INCLUDE_DIRECTORIES(.)
+-TARGET_INCLUDE_DIRECTORIES(xnnpack-hardware-config PRIVATE include src ${CPUINFO_SOURCE_DIR}/include)
++TARGET_INCLUDE_DIRECTORIES(xnnpack-hardware-config PRIVATE include src)
+ IF(XNNPACK_BUILD_LIBRARY)
+   TARGET_INCLUDE_DIRECTORIES(XNNPACK PUBLIC include)
+   IF(WIN32)
+@@ -1164,14 +1164,9 @@ IF(NOT TARGET cpuinfo)
        "${CPUINFO_SOURCE_DIR}"
        "${CMAKE_BINARY_DIR}/cpuinfo")
    ELSE()
@@ -33,7 +38,7 @@ index f0b3410ae..ba54c3bfe 100644
    ENDIF()
  ENDIF()
  IF(XNNPACK_BUILD_LIBRARY)
-@@ -1129,16 +1126,12 @@ IF(NOT TARGET pthreadpool)
+@@ -1189,16 +1184,12 @@ IF(NOT TARGET pthreadpool)
        "${PTHREADPOOL_SOURCE_DIR}"
        "${CMAKE_BINARY_DIR}/pthreadpool")
    ELSE()
@@ -53,7 +58,7 @@ index f0b3410ae..ba54c3bfe 100644
    ENDIF()
  ENDIF()
  TARGET_LINK_LIBRARIES(xnnpack-base INTERFACE pthreadpool)
-@@ -1152,12 +1145,12 @@ IF(NOT TARGET fxdiv)
+@@ -1212,12 +1203,12 @@ IF(NOT TARGET fxdiv)
        "${FXDIV_SOURCE_DIR}"
        "${CMAKE_BINARY_DIR}/FXdiv")
    ELSE()
@@ -69,3 +74,6 @@ index f0b3410ae..ba54c3bfe 100644
      SET_PROPERTY(TARGET fxdiv PROPERTY LINKER_LANGUAGE C)
    ENDIF()
  ENDIF()
+-- 
+2.46.1
+

onnxruntime/core/providers/xnnpack/math/gemm.cc

@@ -139,7 +139,7 @@ Status Gemm::PrePack(const Tensor& tensor, int input_idx, AllocatorPtr,
 
   // flags - 1 - for no transpose - 0 for transpose
   uint32_t flags = trans_B_ == CblasTrans ? 0 : XNN_FLAG_TRANSPOSE_WEIGHTS;
-  auto code_cache = GetCodeCache();
+  // auto code_cache = GetCodeCache();
   auto weights_cache = GetWeightsCache();
   xnn_status status = xnn_status::xnn_status_uninitialized;
   struct xnn_operator* p = nullptr;
@@ -159,7 +159,7 @@ Status Gemm::PrePack(const Tensor& tensor, int input_idx, AllocatorPtr,
         bias_data,                                                   // const float* bias,
         foutput_min, foutput_max,
         flags,
-        code_cache, weights_cache,
+        /*code_cache, */weights_cache,
         &p);
   } else if (op_compute_type_ == OpComputeType::op_compute_type_fp16) {
     const MLFloat16* bias_data = nullptr;
@@ -175,7 +175,7 @@ Status Gemm::PrePack(const Tensor& tensor, int input_idx, AllocatorPtr,
         bias_data,                                                   // const float* bias,
         foutput_min, foutput_max,
         flags,
-        code_cache, weights_cache,
+        /*code_cache, */weights_cache,
         &p);
   }
 

onnxruntime/core/providers/xnnpack/math/matmul.cc

@@ -102,10 +102,10 @@ Status MatMul::PrePack(const Tensor& tensor, int input_idx, AllocatorPtr alloc,
   }
 
 #ifdef XNN_CACHE_ENABLE
-  xnn_code_cache_t code_cache = GetCodeCache();
+  // xnn_code_cache_t code_cache = GetCodeCache();
   xnn_weights_cache_t weight_cache = GetWeightsCache();
 #else
-  xnn_code_cache_t code_cache = nullptr;
+  // xnn_code_cache_t code_cache = nullptr;
   xnn_weights_cache_t weight_cache = nullptr;
 #endif
 
@@ -122,7 +122,7 @@ Status MatMul::PrePack(const Tensor& tensor, int input_idx, AllocatorPtr alloc,
         foutput_min,
         foutput_max,
         flags,
-        code_cache,
+        // code_cache,
         weight_cache,
         &p);
   } else if (op_type_ == OpComputeType::op_compute_type_fp16) {
@@ -136,7 +136,7 @@ Status MatMul::PrePack(const Tensor& tensor, int input_idx, AllocatorPtr alloc,
         foutput_min,
         foutput_max,
         flags,
-        code_cache,
+        // code_cache,
         weight_cache,
         &p);
   }