SHL: version 2.0

CMakeLists.txt

@@ -3,28 +3,26 @@ project(CSI-NN2)
 
 enable_language(ASM)
 
-option(USE_CSI_NN2_DEBUG "option for debug" ON)
+option(USE_SHL_DEBUG "option for debug" ON)
 
 option(BUILD_X86 "build x86" OFF)
 option(BUILD_RISCV "build riscv" OFF)
 option(BUILD_RISCV_ELF "build riscv elf" OFF)
 option(BUILD_CSKY "build csky" OFF)
 option(BUILD_CSKY_ELF "build csky elf" OFF)
 
+if (NOT USE_COMPILER_PATH)
+
 # riscv linux compiler
 if (BUILD_RISCV)
-    if(IS_DIRECTORY $ENV{RISCV_GNU_GCC_PATH})
-        set(RISCV_GNU_GCC $ENV{RISCV_GNU_GCC_PATH})
-    else()
-        set(RISCV_GNU_GCC "${PROJECT_SOURCE_DIR}/tools/gcc-toolchain/bin")
-    endif()
-
-    set(CMAKE_C_COMPILER ${RISCV_GNU_GCC}/riscv64-unknown-linux-gnu-gcc)
-    set(CMAKE_CXX_COMPILER ${RISCV_GNU_GCC}/riscv64-unknown-linux-gnu-g++)
+    set(CMAKE_C_COMPILER riscv64-unknown-linux-gnu-gcc)
+    set(CMAKE_CXX_COMPILER riscv64-unknown-linux-gnu-g++)
+    set(CMAKE_ASM_COMPILER riscv64-unknown-linux-gnu-gcc)
 endif()
 
 # riscv elf compiler
 if (BUILD_RISCV_ELF)
+    set(CMAKE_ASM_COMPILER riscv64-unknown-elf-gcc)
     set(CMAKE_C_COMPILER riscv64-unknown-elf-gcc)
 endif()
 
@@ -40,9 +38,11 @@ if (BUILD_CSKY_ELF)
     set(CMAKE_ASM_COMPILER csky-abiv2-elf-gcc)
 endif()
 
-# CSI-NN2 debug module
-if(USE_CSI_NN2_DEBUG)
-    add_definitions(-D CSI_DEBUG)
+endif()
+
+# SHL debug module
+if(USE_SHL_DEBUG)
+    add_definitions(-D SHL_DEBUG)
 endif()
 
 # reduce elf size
@@ -55,78 +55,133 @@ file(GLOB_RECURSE NN2_SRCS source/nn2/*.c source/utils/*.c)
 file(GLOB_RECURSE REF_SRCS source/reference/*.c)
 file(GLOB_RECURSE GREF_SRCS source/graph_ref/*.c)
 file(GLOB_RECURSE OPENVX_SRCS source/openvx/*.c)
-file(GLOB_RECURSE C906_SRCS source/c906_opt/*.c)
-file(GLOB_RECURSE C908_SRCS source/c908/*.c)
+file(GLOB_RECURSE PNNA_SRCS source/pnna/*.c source/pnna/*.cpp)
+file(GLOB_RECURSE C906_SRCS source/c906_opt/*.c source/c906_opt/*.S)
+file(GLOB_RECURSE C908_SRCS source/c908_opt/*.c source/c908_opt/gemm_kernel/*.S)
 file(GLOB_RECURSE THEAD_RVV_SRCS source/thead_rvv/*.c)
 file(GLOB_RECURSE C860_SRCS source/c860_opt/*.S)
 file(GLOB_RECURSE I805_REF_SRCS source/i805_ref/*.c)
 file(GLOB_RECURSE I805_SRCS source/i805_opt/*.c source/i805_opt/*.S)
 file(GLOB_RECURSE E804_SRCS source/e804_opt/*.c source/e804_opt/*.S)
+file(GLOB_RECURSE ASP_SRCS source/asp/*.c)
 
 include_directories(include)
 
-option(CSINN_LAYER_BENCHMARK "Layer information and performance" OFF)
-if(CSINN_LAYER_BENCHMARK)
-    add_definitions(-DCSINN_LAYER_BENCHMARK)
+option(SHL_LAYER_BENCHMARK "Layer information and performance" OFF)
+if(SHL_LAYER_BENCHMARK)
+    add_definitions(-DSHL_LAYER_BENCHMARK)
     message(STATUS "Print the execution time of each layer - ON")
 endif()
 
-set(CMAKE_INSTALL_PREFIX "${PROJECT_SOURCE_DIR}/install")
-
 if(BUILD_X86)
     # build x86_ref so
     LIST(APPEND X86_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS})
-    add_library(x86_share SHARED ${X86_LST})
-    SET_TARGET_PROPERTIES(x86_share PROPERTIES OUTPUT_NAME "csi_nn2_ref_x86")
-    set(X86_BUILD_FLAGS -DCSI_AVX_OPT -DCSI_BUILD_REF -DCSI_BUILD_GREF -mavx -mfma -fopenmp)
-    target_compile_options(x86_share PRIVATE ${X86_BUILD_FLAGS})
-
-    install(TARGETS x86_share DESTINATION lib)
-
+    add_library(x86_static STATIC ${X86_LST})
+    SET_TARGET_PROPERTIES(x86_static PROPERTIES OUTPUT_NAME "shl_ref_x86")
+    set(X86_BUILD_FLAGS -DSHL_AVX_OPT -DSHL_BUILD_REF -DSHL_BUILD_GREF -fPIC -mavx -mfma -fopenmp)
+    target_compile_options(x86_static PRIVATE ${X86_BUILD_FLAGS})
+
+    install(TARGETS x86_static DESTINATION lib)
+
+    # build pnna x86 simulate so
+    LIST(APPEND PNNA_LST ${NN2_SRCS} ${REF_SRCS} ${PNNA_SRCS})
+    add_library(pnna_share SHARED ${PNNA_LST})
+    SET_TARGET_PROPERTIES(pnna_share PROPERTIES OUTPUT_NAME "shl_pnna_x86")
+    set(PNNA_BUILD_FLAGS -DSHL_BUILD_PNNA)
+    target_compile_options(pnna_share PRIVATE ${PNNA_BUILD_FLAGS})
+    target_include_directories(pnna_share PRIVATE module/nna_ddk_install/include/)
+    set(PNNA_LINK_DIR ${CMAKE_CURRENT_SOURCE_DIR}/module/nna_ddk_install/x86/)
+    target_link_libraries(pnna_share PRIVATE -L${PNNA_LINK_DIR} -limgdnn_csim -lnnasession_csim)
+
+    install(TARGETS pnna_share DESTINATION lib)
+
+    # build heterogeneous pnna x86 simulate so
+    LIST(APPEND HLIGHT_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${PNNA_SRCS})
+    add_library(hlight_share SHARED ${HLIGHT_LST})
+    SET_TARGET_PROPERTIES(hlight_share PROPERTIES OUTPUT_NAME "shl_hlight_x86")
+    set(HLIGHT_BUILD_FLAGS -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_BUILD_PNNA)
+    target_compile_options(hlight_share PRIVATE ${HLIGHT_BUILD_FLAGS})
+    target_include_directories(hlight_share PRIVATE module/nna_ddk_install/include/)
+    set(PNNA_LINK_DIR ${CMAKE_CURRENT_SOURCE_DIR}/module/nna_ddk_install/x86/)
+    target_link_libraries(hlight_share PRIVATE -L${PNNA_LINK_DIR} -limgdnn_csim -lnnasession_csim)
+
+    install(TARGETS hlight_share DESTINATION lib)
 endif()
 
 if(BUILD_RISCV)
     # build rvv a
     LIST(APPEND RVV_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${THEAD_RVV_SRCS})
     add_library(rvv_static STATIC ${RVV_LST})
-    SET_TARGET_PROPERTIES(rvv_static PROPERTIES OUTPUT_NAME "csi_nn2_rvv")
-    set(RVV_BUILD_FLAGS -march=rv64gcv0p7_zfh_xtheadc -mabi=lp64d -DCSI_BUILD_RVV -DCSI_BUILD_REF -DCSI_BUILD_GREF)
+    SET_TARGET_PROPERTIES(rvv_static PROPERTIES OUTPUT_NAME "shl_rvv")
+    set(RVV_BUILD_FLAGS -march=rv64gcv0p7_zfh_xtheadc -mabi=lp64d -DSHL_BUILD_RVV -DSHL_BUILD_REF -DSHL_BUILD_GREF)
     target_compile_options(rvv_static PRIVATE ${RVV_BUILD_FLAGS})
 
     install(TARGETS rvv_static DESTINATION lib)
 
     # build c906 a
     LIST(APPEND C906_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${THEAD_RVV_SRCS} ${C906_SRCS})
     add_library(c906_static STATIC ${C906_LST})
-    SET_TARGET_PROPERTIES(c906_static PROPERTIES OUTPUT_NAME "csi_nn2_c906")
-    set(C906_BUILD_FLAGS -march=rv64gcv0p7_zfh_xtheadc -mabi=lp64d -DCSI_BUILD_C906 -DCSI_BUILD_REF -DCSI_BUILD_GREF)
+    SET_TARGET_PROPERTIES(c906_static PROPERTIES OUTPUT_NAME "shl_c906")
+    set(C906_BUILD_FLAGS -march=rv64gcv0p7_zfh_xtheadc -mabi=lp64d -DSHL_BUILD_C906 -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_BUILD_RVV)
     target_compile_options(c906_static PRIVATE ${C906_BUILD_FLAGS})
 
     install(TARGETS c906_static DESTINATION lib)
 
     add_library(c906_share SHARED ${C906_LST})
-    SET_TARGET_PROPERTIES(c906_share PROPERTIES OUTPUT_NAME "csi_nn2_c906")
+    SET_TARGET_PROPERTIES(c906_share PROPERTIES OUTPUT_NAME "shl_c906")
     target_compile_options(c906_share PRIVATE ${C906_BUILD_FLAGS})
 
     install(TARGETS c906_share DESTINATION lib)
 
+    # build c908 a
+    LIST(APPEND C908_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${THEAD_RVV_SRCS} ${C908_SRCS})
+    add_library(c908_static STATIC ${C908_LST})
+    SET_TARGET_PROPERTIES(c908_static PROPERTIES OUTPUT_NAME "shl_c908")
+    set(C908_BUILD_FLAGS -march=rv64gcv_zfh_xtheadc_xtheadv -mabi=lp64d -DSHL_BUILD_C908 -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_BUILD_RVV)
+    target_compile_options(c908_static PRIVATE ${C908_BUILD_FLAGS})
+
+    install(TARGETS c908_static DESTINATION lib)
+
+    # build pnna so
+    LIST(APPEND PNNA_LST ${NN2_SRCS} ${REF_SRCS} ${PNNA_SRCS})
+    add_library(pnna_share SHARED ${PNNA_LST})
+    SET_TARGET_PROPERTIES(pnna_share PROPERTIES OUTPUT_NAME "shl_pnna")
+    set(PNNA_BUILD_FLAGS -DSHL_BUILD_PNNA)
+    target_compile_options(pnna_share PRIVATE ${PNNA_BUILD_FLAGS})
+    target_include_directories(pnna_share PRIVATE module/nna_ddk_install/include/)
+    set(PNNA_LINK_DIR ${CMAKE_CURRENT_SOURCE_DIR}/module/nna_ddk_install/light/)
+    target_link_libraries(pnna_share PRIVATE -L${PNNA_LINK_DIR} -limgdnn -lnnasession)
+
+    install(TARGETS pnna_share DESTINATION lib)
+
+    # build heterogeneous pnna so
+    LIST(APPEND HLIGHT_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${PNNA_SRCS})
+    add_library(hlight_share SHARED ${HLIGHT_LST})
+    SET_TARGET_PROPERTIES(hlight_share PROPERTIES OUTPUT_NAME "shl_hlight")
+    set(HLIGHT_BUILD_FLAGS -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_BUILD_PNNA)
+    target_compile_options(hlight_share PRIVATE ${HLIGHT_BUILD_FLAGS})
+    target_include_directories(hlight_share PRIVATE module/nna_ddk_install/include/)
+    set(PNNA_LINK_DIR ${CMAKE_CURRENT_SOURCE_DIR}/module/nna_ddk_install/light/)
+    target_link_libraries(hlight_share PRIVATE -L${PNNA_LINK_DIR} -limgdnn -lnnasession)
+
+    install(TARGETS hlight_share DESTINATION lib)
 endif()
 
 if(BUILD_RISCV_ELF)
     # build c906 elf a
     LIST(APPEND C906_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${THEAD_RVV_SRCS} ${C906_SRCS})
     add_library(c906_elf_static STATIC ${C906_LST})
-    SET_TARGET_PROPERTIES(c906_elf_static PROPERTIES OUTPUT_NAME "csi_nn2_c906_rtos")
-    set(C906_BUILD_FLAGS -march=rv64gcv0p7_zfh_xtheadc -mabi=lp64d -DCSI_BUILD_C906 -DCSI_BUILD_REF -DCSI_BUILD_GREF -DCSI_BUILD_RTOS)
+    SET_TARGET_PROPERTIES(c906_elf_static PROPERTIES OUTPUT_NAME "shl_c906_rtos")
+    set(C906_BUILD_FLAGS -march=rv64gcv0p7_zfh_xtheadc -mabi=lp64d -mcmodel=medany -DSHL_BUILD_C906 -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_BUILD_RTOS)
     target_compile_options(c906_elf_static PRIVATE ${C906_BUILD_FLAGS})
 
     install(TARGETS c906_elf_static DESTINATION lib)
 
     # build ASP elf a
     LIST(APPEND ASP_LST ${NN2_SRCS} ${REF_SRCS} ${GREF_SRCS} ${ASP_SRCS})
     add_library(asp_elf_static STATIC ${ASP_LST})
-    SET_TARGET_PROPERTIES(asp_elf_static PROPERTIES OUTPUT_NAME "csi_nn2_asp")
-    set(ASP_BUILD_FLAGS -march=rv32imafdcp -mabi=ilp32d -DCSI_BUILD_ASP -DCSI_BUILD_REF -DCSI_BUILD_GREF -DCSI_BUILD_RTOS)
+    SET_TARGET_PROPERTIES(asp_elf_static PROPERTIES OUTPUT_NAME "shl_asp")
+    set(ASP_BUILD_FLAGS -march=rv32imafdcp -mabi=ilp32d -DSHL_BUILD_ASP -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_USE_ATAT_MALLOC -DSHL_BUILD_RTOS)
     target_compile_options(asp_elf_static PRIVATE ${ASP_BUILD_FLAGS})
 
     install(TARGETS asp_elf_static DESTINATION lib)
@@ -136,8 +191,8 @@ if(BUILD_CSKY)
     # build openvx so
     LIST(APPEND OPENVX_LST ${NN2_SRCS} ${OPENVX_SRCS})
     add_library(openvx_share SHARED ${OPENVX_LST})
-    SET_TARGET_PROPERTIES(openvx_share PROPERTIES OUTPUT_NAME "csi_nn2_openvx")
-    set(OPENVX_BUILD_FLAGS -mcpu=c860v -fPIC -DCSI_BUILD_OPENVX -mhard-float)
+    SET_TARGET_PROPERTIES(openvx_share PROPERTIES OUTPUT_NAME "shl_openvx")
+    set(OPENVX_BUILD_FLAGS -mcpu=c860v -fPIC -DSHL_BUILD_OPENVX -mhard-float)
     target_compile_options(openvx_share PRIVATE ${OPENVX_BUILD_FLAGS})
     set(OPENVX_LINK_DIR ${CMAKE_CURRENT_SOURCE_DIR}/module/acuity-driver/driver/build/sdk/drivers)
     target_link_libraries(openvx_share PRIVATE -mcpu=c860v -fPIC -mhard-float -L${OPENVX_LINK_DIR} -lArchModelSw -lNNArchPerf -lOpenVX -lOpenVXU -lCLC -lVSC -lGAL -lNNGPUBinary -lovxlib -lOvx12VXCBinary)
@@ -149,8 +204,8 @@ if(BUILD_CSKY)
     # build c860 a
     LIST(APPEND C860_LST ${NN2_SRCS} ${REF_SRCS} ${C860_SRCS})
     add_library(c860_static STATIC ${C860_LST})
-    SET_TARGET_PROPERTIES(c860_static PROPERTIES OUTPUT_NAME "csi_nn2_c860")
-    set(C860_BUILD_FLAGS -mcpu=c860v -DCSI_BUILD_REF)
+    SET_TARGET_PROPERTIES(c860_static PROPERTIES OUTPUT_NAME "shl_c860")
+    set(C860_BUILD_FLAGS -mcpu=c860v -DSHL_BUILD_REF)
     target_compile_options(c860_static PRIVATE ${C860_BUILD_FLAGS})
 
     install(TARGETS c860_static DESTINATION lib)
@@ -160,32 +215,41 @@ if(BUILD_CSKY_ELF)
     # build i805 ref a
     LIST(APPEND I805_REF_LST ${NN2_SRCS} ${REF_SRCS} ${I805_REF_SRCS})
     add_library(i805_ref_static STATIC ${I805_REF_LST})
-    SET_TARGET_PROPERTIES(i805_ref_static PROPERTIES OUTPUT_NAME "csi_nn2_ref_i805")
-    set(I805_REF_BUILD_FLAGS -DCSI_BUILD_REF_I805 -DCSI_MATH_DSP -DCSI_BUILD_RTOS -mcpu=i805)
+    SET_TARGET_PROPERTIES(i805_ref_static PROPERTIES OUTPUT_NAME "shl_ref_i805")
+    set(I805_REF_BUILD_FLAGS -DSHL_BUILD_REF_I805 -DSHL_BUILD_RTOS -mcpu=i805)
     target_compile_options(i805_ref_static PRIVATE ${I805_REF_BUILD_FLAGS})
-    target_include_directories(i805_ref_static PRIVATE include/include_xt800)
+    target_include_directories(i805_ref_static PRIVATE source/i805_ref)
 
     install(TARGETS i805_ref_static DESTINATION lib)
 
     # build i805 a
     LIST(APPEND I805_LST ${NN2_SRCS} ${REF_SRCS} ${I805_SRCS})
     add_library(i805_static STATIC ${I805_LST})
-    SET_TARGET_PROPERTIES(i805_static PROPERTIES OUTPUT_NAME "csi_nn2_i805")
-    set(I805_BUILD_FLAGS -DCSI_BUILD_I805 -DCSI_MATH_DSP -DCSI_BUILD_RTOS -mcpu=ck805ef -mhard-float)
+    SET_TARGET_PROPERTIES(i805_static PROPERTIES OUTPUT_NAME "shl_i805")
+    set(I805_BUILD_FLAGS -DSHL_BUILD_I805 -DSHL_BUILD_RTOS -mcpu=ck805ef -mhard-float)
     target_compile_options(i805_static PRIVATE ${I805_BUILD_FLAGS})
-    target_include_directories(i805_static PRIVATE include/include_xt800)
+    target_include_directories(i805_static PRIVATE source/i805_opt)
 
     install(TARGETS i805_static DESTINATION lib)
 
     # build e804 a
     LIST(APPEND E804_LST ${NN2_SRCS} ${REF_SRCS} ${E804_SRCS})
     add_library(e804_static STATIC ${E804_LST})
-    SET_TARGET_PROPERTIES(e804_static PROPERTIES OUTPUT_NAME "csi_nn2_e804")
-    set(E804_BUILD_FLAGS -DCSI_BUILD_E804 -mcpu=e804d -DCSI_BUILD_RTOS -mno-required-attr-fpu-abi)
+    SET_TARGET_PROPERTIES(e804_static PROPERTIES OUTPUT_NAME "shl_e804")
+    set(E804_BUILD_FLAGS -DSHL_BUILD_E804 -mcpu=e804d -DSHL_BUILD_RTOS -mno-required-attr-fpu-abi)
     target_compile_options(e804_static PRIVATE ${E804_BUILD_FLAGS})
-    target_include_directories(e804_static PRIVATE include/include_xt800)
+    target_include_directories(e804_static PRIVATE source/e804_opt)
 
     install(TARGETS e804_static DESTINATION lib)
 endif()
 
+# coverage options
+OPTION(ENABLE_GCOV "Enable gcov" OFF)
+if(ENABLE_GCOV)
+  SET(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fprofile-arcs -ftest-coverage")
+  SET(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} -fprofile-arcs -ftest-coverage")
+  SET(CMAKE_EXE_LINKER_FLAGS_DEBUG "${CMAKE_EXE_LINKER_FLAGS_DEBUG} -fprofile-arcs -ftest-coverage -lgcov")
+endif()
+
+
 install(DIRECTORY "include/." DESTINATION "include" FILES_MATCHING PATTERN "*.h")

Makefile

@@ -15,19 +15,37 @@ nn2_c906_so:
 nn2_c906_elf:
 	mkdir -p riscv_elf_build; cd riscv_elf_build; cmake ../ -DBUILD_RISCV_ELF=ON -DCMAKE_BUILD_TYPE=Release; make c906_elf_static -j8; cd -
 
+nn2_asp_elf:
+	mkdir -p riscv_elf_build; cd riscv_elf_build; cmake ../ -DBUILD_RISCV_ELF=ON -DCMAKE_BUILD_TYPE=Release; make asp_elf_static -j8; cd -
+
 nn2_c908:
 	mkdir -p riscv_build; cd riscv_build; cmake ../ -DBUILD_RISCV=ON -DCMAKE_BUILD_TYPE=Release; make c908_static -j8; cd -
 
 nn2_ref_x86:
-	mkdir -p x86_build; cd x86_build; cmake ../ -DBUILD_X86=ON -DCMAKE_BUILD_TYPE=Release; make x86_share -j8; cd -
+	mkdir -p x86_build; cd x86_build; cmake ../ -DBUILD_X86=ON -DCMAKE_BUILD_TYPE=Release; make x86_static -j8; cd -
+
+nn2_openvx:
+	mkdir -p csky_build; cd csky_build; cmake ../ -DBUILD_CSKY=ON -DCMAKE_BUILD_TYPE=Release; make openvx_share -j8; cd -
+
+nn2_pnna:
+	mkdir -p riscv_build; cd riscv_build; cmake ../ -DBUILD_RISCV=ON -DCMAKE_BUILD_TYPE=Release; make pnna_share -j8; cd -
+
+nn2_pnna_x86:
+	mkdir -p x86_build; cd x86_build; cmake ../ -DBUILD_X86=ON -DCMAKE_BUILD_TYPE=Release; make pnna_share -j8; cd -
+
+nn2_hlight_x86:
+	mkdir -p x86_build; cd x86_build; cmake ../ -DBUILD_X86=ON -DCMAKE_BUILD_TYPE=Release; make hlight_share -j8; cd -
+
+nn2_hlight:
+	mkdir -p riscv_build; cd riscv_build; cmake ../ -DBUILD_RISCV=ON -DCMAKE_BUILD_TYPE=Release; make hlight_share -j8; cd -
 
 .PHONY: install_nn2
 install_nn2: include
 	mkdir -p install_nn2/lib
 	cp include install_nn2 -r
-	-cp riscv_build/libcsi_nn2_* install_nn2/lib -rf
-	-cp csky_build/libcsi_nn2_* install_nn2/lib -rf
-	-cp x86_build/libcsi_nn2_* install_nn2/lib -rf
+	-cp riscv_build/libshl_* install_nn2/lib -rf
+	-cp csky_build/libshl_* install_nn2/lib -rf
+	-cp x86_build/libshl_* install_nn2/lib -rf
 	cp version install_nn2/ -rf
 
 clint:

README.md

@@ -1,27 +1,27 @@
 ## 简介
 
-CSI-NN2 是 T-HEAD 提供的一组针对无剑 SoC 平台的神经网络库 API。抽象了各种常用的网络层的接口，并且提供一系列已优化的二进制库。
+SHL(曾用名CSI-NN2) 是 T-HEAD 提供的一组针对玄铁 CPU 平台的神经网络库 API。抽象了各种常用的网络层的接口，并且提供一系列已优化的二进制库。
 
-CSI-NN2 的特性：
+SHL 的特性：
 
 - C 代码版本的参考实现。
 - 提供玄铁系列 CPU 的汇编优化实现。
 - 支持对称量化和非对称量化。
 - 支持8位定点，16位定点和16位浮点等数据类型。
 - 兼容 NCHW 和 NHWC 格式。
-- 搭配 [HHB](https://occ.t-head.cn/development/series/index?spm=a2cl5.14300690.0.0.4aca475a4yHCxV&id=3865005559921381376&type=kind) 实现代码自动调用。
-- 覆盖 CPU，NPU 架构。
-- 附加一些辅助接口，参考使用。
+- 搭配 [HHB](https://www.yuque.com/za4k4z/oxlbxl) 实现代码自动调用。
+- 覆盖 CPU，NPU 等不同体系结构。
+- 附加异构参考实现。
 
-CSI-NN2 提供了完成的接口声明和接口的参考实现，各个设备提供商可以依此针对性的完成各个接口的优化工作。
+SHL 提供了完成的接口声明和接口的参考实现，各个设备提供商可以依此针对性的完成各个接口的优化工作。
 
 ## 文档说明
 
 - [中文文档](https://www.yuque.com/za4k4z/isgz8o)
 
 ## 致谢
 
-CSI-NN2 参考、借鉴了下列项目：
+SHL 参考、借鉴了下列项目：
 - [Caffe](https://github.com/BVLC/caffe)
 - [Tensorflow](https://github.com/tensorflow/tensorflow)
 - [ncnn](https://github.com/Tencent/ncnn)