[runtime] To upgrade libtorch CPU runtime with IPEX version

runtime/README.md

@@ -26,7 +26,7 @@ Here is a brief summary of all platforms and OSs. please note the corresponding
 | ios             | ios                 | libtorch             | ios demo, [link](TODO)                                                                           |
 | kunlun          | linux               | xpu runtime          | Kunlun XPU runtime                                                                               |
 | libtorch        | linux, windows, mac | libtorch             | c++ build with libtorch                                                                          |
-| ipex            | linux               | libtorch + ipex      | c++ build with libtorch with ipex optimization                                                                         |
+| ipex            | linux               | libtorch + ipex      | c++ build with libtorch and ipex optimization                                                                         |
 | onnxrutnime     | linux, windows, mac | onnxruntime          | c++ build with onnxruntime                                                                       |
 | openvino        | linux, windows, mac | openvino             | c++ build with openvino                                                                          |
 | raspberrypi     | linux               | onnxruntime          | c++ build on raspberrypi with onnxruntime                                                        |

runtime/core/bin/CMakeLists.txt

@@ -1,5 +1,8 @@
 add_executable(decoder_main decoder_main.cc)
 target_link_libraries(decoder_main PUBLIC decoder)
+if(IPEX)
+  target_link_libraries(decoder_main PUBLIC "-Wl,--no-as-needed ${IPEX_CPU_CORE_LIBRARY}")
+endif()
 
 add_executable(label_checker_main label_checker_main.cc)
 target_link_libraries(label_checker_main PUBLIC decoder)

runtime/core/cmake/ipex.cmake

@@ -5,11 +5,11 @@ if(NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
 endif()
 
 if(CXX11_ABI)
-  set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-1.13.0%2Bcpu.zip")
-  set(URL_HASH "SHA256=d52f63577a07adb0bfd6d77c90f7da21896e94f71eb7dcd55ed7835ccb3b2b59")
+  set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-2.0.1%2Bcpu.zip")
+  set(URL_HASH "SHA256=137a842d1cf1e9196b419390133a1623ef92f8f84dc7a072f95ada684f394afd")
 else()
-  set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cpu/libtorch-shared-with-deps-1.13.0%2Bcpu.zip")
-  set(URL_HASH "SHA256=bee1b7be308792aa60fc95a4f5274d9658cb7248002d0e333d49eb81ec88430c")
+  set(LIBTORCH_URL "https://download.pytorch.org/libtorch/cpu/libtorch-shared-with-deps-2.0.1%2Bcpu.zip")
+  set(URL_HASH "SHA256=90d50350fd24ce5cf9dfbf47888d0cfd9f943eb677f481b86fe1b8e90f7fda5d")
 endif()
 FetchContent_Declare(libtorch
 URL      ${LIBTORCH_URL}
@@ -19,13 +19,13 @@ FetchContent_MakeAvailable(libtorch)
 find_package(Torch REQUIRED PATHS ${libtorch_SOURCE_DIR} NO_DEFAULT_PATH)
 
 if(CXX11_ABI)
-  set(LIBIPEX_URL "https://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-cxx11-abi-1.13.100%2Bcpu.run")
-  set(URL_HASH "SHA256=e26778d68a5b76e5d4e540812433ade4f3e917bafc7a252a1fc1c0d1e0ed9c73")
-  set(LIBIPEX_SCRIPT_NAME "libintel-ext-pt-cxx11-abi-1.13.100%2Bcpu.run")
+  set(LIBIPEX_URL "https://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-cxx11-abi-2.0.100%2Bcpu.run")
+  set(URL_HASH "SHA256=f172d9ebc2ca0c39cc93bb395721194f79767e1bc3f82b13e1edc07d1530a600")
+  set(LIBIPEX_SCRIPT_NAME "libintel-ext-pt-cxx11-abi-2.0.100%2Bcpu.run")
 else()
-  set(LIBIPEX_URL "https://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-1.13.100%2Bcpu.run")
-  set(URL_HASH "SHA256=0e0e09384106a1f00f4dbd9b543ff42643b769f654c693ff9ecdaba04c607987")
-  set(LIBIPEX_SCRIPT_NAME "libintel-ext-pt-1.13.100%2Bcpu.run")
+  set(LIBIPEX_URL "https://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-2.0.100%2Bcpu.run")
+  set(URL_HASH "SHA256=8392f965dd9b8f6c0712acbb805c7e560e4965a0ade279b47a5f5a8363888268")
+  set(LIBIPEX_SCRIPT_NAME "libintel-ext-pt-2.0.100%2Bcpu.run")
 endif()
 FetchContent_Declare(intel_ext_pt
 URL                  ${LIBIPEX_URL}
@@ -35,6 +35,6 @@ DOWNLOAD_NO_EXTRACT  TRUE
 PATCH_COMMAND        bash ${FETCHCONTENT_BASE_DIR}/${LIBIPEX_SCRIPT_NAME} install ${libtorch_SOURCE_DIR}
 )
 FetchContent_MakeAvailable(intel_ext_pt)
-find_package(intel_ext_pt_cpu REQUIRED PATHS ${libtorch_SOURCE_DIR} NO_DEFAULT_PATH)
+find_package(IPEX REQUIRED PATHS ${libtorch_SOURCE_DIR} NO_DEFAULT_PATH)
 
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS} -DC10_USE_GLOG")

runtime/ipex/CMakeLists.txt

@@ -10,6 +10,7 @@ option(GRPC "whether to build with gRPC" OFF)
 option(WEBSOCKET "whether to build with websocket" OFF)
 option(HTTP "whether to build with http" OFF)
 option(TORCH "whether to build with Torch" ON)
+option(IPEX "whether to build with Torch+IPEX" ON)
 
 set(CMAKE_VERBOSE_MAKEFILE OFF)
 
@@ -63,4 +64,4 @@ add_subdirectory(bin)
 if(BUILD_TESTING)
   include(gtest)
   add_subdirectory(test)
-endif()
+endif()

runtime/ipex/README.md

@@ -16,7 +16,7 @@ pip install -r requirements.txt
 ```
 Installation of IPEX
 ``` sh
-pip install intel_extension_for_pytorch==1.13.100 -f https://developer.intel.com/ipex-whl-stable-cpu
+pip install intel_extension_for_pytorch==2.0.100 -f https://developer.intel.com/ipex-whl-stable-cpu
 ```
 
 Installation of related tools: Intel® OpenMP and TCMalloc
@@ -36,14 +36,16 @@ based on the package manager of your system.
 For exporting FP32 runtime model
 ``` sh
 source examples/aishell/s0/path.sh
+export OMP_NUM_THREADS=1
 python wenet/bin/export_ipex.py \
     --config <model_config_yaml_filename> \
     --checkpoint <model_ckpt_filename> \
     --output_file <runtime_model_output_filename>
 ```
-If you have an Intel® 4th Generation Xeon (Sapphire Rapids) server, you can export a BF16 runtime model and get better performance by virtue of AMX instructions
+If you have an Intel® 4th Generation Xeon (Sapphire Rapids) server, you can export a BF16 runtime model and get better performance by virtue of [AMX instructions](https://en.wikipedia.org/wiki/Advanced_Matrix_Extensions)
 ``` sh
 source examples/aishell/s0/path.sh
+export OMP_NUM_THREADS=1
 python wenet/bin/export_ipex.py \
     --config <model_config_yaml_filename> \
     --checkpoint <model_ckpt_filename> \
@@ -53,6 +55,7 @@ python wenet/bin/export_ipex.py \
 And for exporting int8 quantized runtime model
 ``` sh
 source examples/aishell/s0/path.sh
+export OMP_NUM_THREADS=1
 python wenet/bin/export_ipex.py \
     --config <model_config_yaml_filename> \
     --checkpoint <model_ckpt_filename> \
@@ -73,13 +76,13 @@ export GLOG_logtostderr=1
 export GLOG_v=2
 wav_path=your_test_wav_path
 model_dir=your_model_dir
-ipexrun --no_python \
+ipexrun --no-python \
     ./build/bin/decoder_main \
         --chunk_size -1 \
         --wav_path $wav_path \
         --model_path $model_dir/final.zip \
         --unit_path $model_dir/units.txt 2>&1 | tee log.txt
 ```
-NOTE: Please refer [IPEX Launch Script Usage Guide](https://intel.github.io/intel-extension-for-pytorch/cpu/1.13.100+cpu/tutorials/performance_tuning/launch_script.html) for usage of advanced features.
+NOTE: Please refer [IPEX Launch Script Usage Guide](https://intel.github.io/intel-extension-for-pytorch/cpu/2.0.100+cpu/tutorials/performance_tuning/launch_script.html) for usage of advanced features.
 
 For advanced usage of WeNet, such as building Web/RPC/HTTP services, please refer [LibTorch Tutorial](../libtorch#advanced-usage). The difference is that the executables should be invoked via IPEX launch script `ipexrun`.