!38597 fix error links for r1.6

Merge pull request !38597 from lvmingfu/code_docs_r1.666

.jenkins/check/config/filter_cppcheck.txt

@@ -24,6 +24,7 @@
 "mindspore/mindspore/ccsrc/runtime/hccl_adapter/hccl_adapter.cc"                                      "useStlAlgorithm"
 "mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/cast_gpu_kernel.cc"                               "unknownMacro"
 "mindspore/mindspore/ccsrc/runtime/device/ascend/ascend_memory_manager.cc"                            "nullPointerArithmeticRedundantCheck"
+"mindspore/mindspore/ccsrc/pipeline/jit/static_analysis/auto_monad.cc"                                "containerOutOfBounds"
 
 # MindData
 "mindspore/mindspore/ccsrc/minddata/dataset/engine/dataset_iterator.cc"                               "useStlAlgorithm"

.jenkins/check/config/whitelizard.txt

@@ -95,6 +95,7 @@ mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/deconv_winograd
 mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/deconv_winograd_fp32.c:DeConvWgMerge
 mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/intrinsics/avx/TiledC8MatMulFp32.c:TiledC8MatmulFp32
 mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp16/quant_dtype_cast_fp16.c:Fp16ToInt8_arm64
+mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_fp32.c:MatMul4x1Kernel
 mindspore/mindspore/ccsrc/backend/session/gpu_session.cc:mindspore::session::gpu::GPUSession::LoadInputData
 mindspore/mindspore/ccsrc/debug/dump_proto.cc:mindspore::ProtoExporter::SetNodeOutputType
 mindspore/mindspore/ccsrc/debug/dump_proto.cc:mindspore::ProtoExporter::SetValueToProto
@@ -164,11 +165,30 @@ mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_
 mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_4x32_kernel_nhwc_fp32.c:nnacl_gemm_avx512_4x32_kernel_nhwc_fp32
 mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_5x64_kernel_nhwc_fp32.c:nnacl_gemm_avx512_5x64_kernel_nhwc_fp32
 mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_5x32_kernel_nhwc_fp32.c:nnacl_gemm_avx512_5x32_kernel_nhwc_fp32
-mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_avx512_fp32.c:nnacl_gemm_avx512_2x64_kernel_nhwc_fp32
-mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_avx512_fp32.c:nnacl_gemm_avx512_3x64_kernel_nhwc_fp32
-mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_avx512_fp32.c:nnacl_gemm_avx512_4x64_kernel_nhwc_fp32
-mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_avx512_fp32.c:nnacl_gemm_avx512_5x64_kernel_nhwc_fp32
-mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_avx512_fp32.c:nnacl_gemm_avx512_6x64_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_11x32_kernel_nhwc_fp32.c:nnacl_gemm_avx512_11x32_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_10x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_10x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_7x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_7x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_9x32_kernel_nhwc_fp32.c:nnacl_gemm_avx512_9x32_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_9x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_9x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_5x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_5x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_12x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_12x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_5x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_5x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_6x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_6x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_1x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_1x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_8x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_8x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_4x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_4x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_7x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_7x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_6x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_6x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_3x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_3x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_1x32_kernel_nhwc_fp32.c:nnacl_gemm_avx512_1x32_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_2x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_2x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_11x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_11x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_1x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_1x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_10x32_kernel_nhwc_fp32.c:nnacl_gemm_avx512_10x32_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_4x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_4x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_8x16_kernel_nhwc_fp32.c:nnacl_gemm_avx512_8x16_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_3x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_3x48_kernel_nhwc_fp32
+mindspore/mindspore/lite/experiment/HPC-generator/gemm_avx512/nnacl_gemm_avx512_2x48_kernel_nhwc_fp32.c:nnacl_gemm_avx512_2x48_kernel_nhwc_fp32
 mindspore/mindspore/lite/src/runtime/kernel/arm/fp32/matmul_fp32_base.cc:mindspore::kernel::MatmulFp32BaseCPUKernel::Run
 mindspore/mindspore/ccsrc/frontend/parallel/auto_parallel/rec_core/rec_partition.cc:mindspore::parallel::GetWeights
 mindspore/mindspore/ccsrc/frontend/parallel/auto_parallel/rec_core/rec_partition.cc:mindspore::parallel::PartitionNode

CMakeLists.txt

@@ -19,51 +19,13 @@ if(NOT CMAKE_SYSTEM_NAME MATCHES "Windows")
 endif()
 
 if(${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
-    # find appropriate macosx SDK and set SDKROOT
-    if(NOT DEFINED ENV{SDKROOT})
-        # arm64: macosx11.x
-        # x86_64: macosx10.x, macosx11.x
-        if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "arm64")
-            set(MACOSX_SDK_REGEX "MacOSX11(\\.\\d+)?")
-        else()
-            set(MACOSX_SDK_REGEX "MacOSX1[01](\\.\\d+)?")
-        endif()
-        exec_program(xcrun ARGS --show-sdk-path OUTPUT_VARIABLE MACOSX_SDK_PATH)
-        get_filename_component(MACOSX_SDK_PATH ${MACOSX_SDK_PATH} DIRECTORY)
-        file(GLOB ALL_SDK_NAME RELATIVE ${MACOSX_SDK_PATH} ${MACOSX_SDK_PATH}/*)
-        # get highest SDK version meets the requirements
-        execute_process(
-            COMMAND bash -c "echo '${ALL_SDK_NAME}' | grep -Eo '${MACOSX_SDK_REGEX}' | sort -n | tail -1 | tr -d '\\n'"
-            OUTPUT_VARIABLE MACOSX_FIND_SDK_NAME
-        )
-        if(NOT MACOSX_FIND_SDK_NAME)
-            message(FATAL_ERROR "can not find appropriate macosx SDK, you may need upgrade xcode")
-        endif()
-        set(ENV{SDKROOT} "${MACOSX_SDK_PATH}/${MACOSX_FIND_SDK_NAME}.sdk")
-    endif()
-    message("macosx sdkroot: $ENV{SDKROOT}")
-    # set macosx deployment target based on SDK
-    if(NOT DEFINED ENV{MACOSX_DEPLOYMENT_TARGET})
-        execute_process(
-            COMMAND bash -c "cat $ENV{SDKROOT}/SDKSettings.json | \
-                grep -Eo 'MACOSX_DEPLOYMENT_TARGET\\\":\\\"\\d{2}\\.\\d+' | cut -d '\"' -f 3 | tr -d '\\n'"
-            OUTPUT_VARIABLE MACOSX_FIND_SDK_VERSION
-        )
-        if(NOT MACOSX_FIND_SDK_VERSION)
-            message(FATAL_ERROR "can not find MACOSX_DEPLOYMENT_TARGET in SDKROOT, \
-                please check whether it's a valid SDK path")
-        endif()
-        set(CMAKE_OSX_DEPLOYMENT_TARGET ${MACOSX_FIND_SDK_VERSION} CACHE STRING
-            "minimum macosx deployment target version" FORCE)
-        if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "arm64")
-            set(CMAKE_OSX_DEPLOYMENT_TARGET "11.0")
-        endif()
-    endif()
-    message("macosx deployment target version: ${CMAKE_OSX_DEPLOYMENT_TARGET}")
     set(CMAKE_CXX_FLAGS_RELEASE "$ENV{CXXFLAGS} -O2 -Winconsistent-missing-override -Wno-user-defined-warnings \
         -Wno-return-std-move -Wno-unused-private-field -Wno-unused-lambda-capture -Wno-sign-compare \
         -Wno-overloaded-virtual -Wno-unneeded-internal-declaration -Wno-unused-variable -Wno-pessimizing-move \
         -Wno-inconsistent-missing-override -DHALF_ENABLE_CPP11_USER_LITERALS=0 -D_FORTIFY_SOURCE=2")
+    if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang" AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 13.1)
+        set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -Wno-unused-but-set-variable")
+    endif()
 elseif(ENABLE_SYM_FILE)
     set(CMAKE_CXX_FLAGS_RELEASE "$ENV{CXXFLAGS} -O2 -g -ggdb -Wl,--allow-shlib-undefined \
         -DHALF_ENABLE_CPP11_USER_LITERALS=0 -D_FORTIFY_SOURCE=2")

README.md

@@ -44,7 +44,7 @@ enrichment of the AI software/hardware application ecosystem.
 
 <img src="https://gitee.com/mindspore/mindspore/raw/master/docs/MindSpore-architecture.png" alt="MindSpore Architecture"/>
 
-For more details please check out our [Architecture Guide](https://www.mindspore.cn/docs/programming_guide/en/master/architecture.html).
+For more details please check out our [Architecture Guide](https://www.mindspore.cn/docs/programming_guide/en/r1.6/architecture.html).
 
 ### Automatic Differentiation
 
@@ -240,7 +240,7 @@ please check out [docker](https://gitee.com/mindspore/mindspore/blob/master/scri
 
 ## Quickstart
 
-See the [Quick Start](https://www.mindspore.cn/tutorials/en/master/quick_start.html)
+See the [Quick Start](https://www.mindspore.cn/tutorials/en/r1.6/quick_start.html)
 to implement the image classification.
 
 ## Docs
@@ -282,6 +282,7 @@ Project stable branches will be in one of the following states:
 
 | **Branch** | **Status**   | **Initial Release Date** | **Next Phase**                         | **EOL Date**|
 |------------|--------------|--------------------------|----------------------------------------|-------------|
+| **r1.6**   | Maintained   | 2022-01-29               | Unmaintained <br> 2023-01-29 estimated |             |
 | **r1.5**   | Maintained   | 2021-10-15               | Unmaintained <br> 2022-10-15 estimated |             |
 | **r1.4**   | Maintained   | 2021-08-15               | Unmaintained <br> 2022-08-15 estimated |             |
 | **r1.3**   | Maintained   | 2021-07-15               | Unmaintained <br> 2022-07-15 estimated |             |

README_CN.md

@@ -41,7 +41,7 @@ MindSpore提供了友好的设计和高效的执行，旨在提升数据科学
 
 <img src="https://gitee.com/mindspore/mindspore/raw/master/docs/MindSpore-architecture-zh.png" alt="MindSpore Architecture"/>
 
-欲了解更多详情，请查看我们的[总体架构](https://www.mindspore.cn/docs/programming_guide/zh-CN/master/architecture.html)。
+欲了解更多详情，请查看我们的[总体架构](https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/architecture.html)。
 
 ### 自动微分
 
@@ -236,7 +236,7 @@ MindSpore的Docker镜像托管在[Docker Hub](https://hub.docker.com/r/mindspore
 
 ## 快速入门
 
-参考[快速入门](https://www.mindspore.cn/tutorials/zh-CN/master/quick_start.html)实现图片分类。
+参考[快速入门](https://www.mindspore.cn/tutorials/zh-CN/r1.6/quick_start.html)实现图片分类。
 
 ## 文档
 
@@ -275,6 +275,7 @@ MindSpore的版本分支有以下几种维护阶段：
 
 | **分支名** | **当前状态**  | **上线时间**          | **后续状态**                           | **EOL 日期**|
 |------------|--------------|----------------------|----------------------------------------|------------|
+| **r1.6**   | Maintained   | 2022-01-29           | Unmaintained <br> 2023-01-29 estimated |            |
 | **r1.5**   | Maintained   | 2021-10-15           | Unmaintained <br> 2022-10-15 estimated |            |
 | **r1.4**   | Maintained   | 2021-08-15           | Unmaintained <br> 2022-08-15 estimated |            |
 | **r1.3**   | Maintained   | 2021-07-15           | Unmaintained <br> 2022-07-15 estimated |            |

RELEASE.md

@@ -1,3 +1,197 @@
+# MindSpore 1.6.2
+
+## MindSpore 1.6.2 Release Notes
+
+### Bug fixes
+
+- Fix the problem that ASR dynamic shape memory leak.
+- Support ModelArts trains while reading.
+- Dropout operator Ascend recalculation adaptation.
+- Provide MindIR to ONNX model conversion interface.
+
+### Contributors
+
+Thanks goes to these wonderful people:
+
+Adel, AGroupofProbiotocs, anthonyaje, anzhengqi, askmiao, baihuawei, baiyangfan, bai-yangfan, bingyaweng, BowenK, buxue, caifubi, CaoJian, caojian05, caozhou, Cathy, changzherui, chenbo116, chenfei, chengxianbin, chenhaozhe, chenjianping, chenzomi, chenzupeng, chujinjin, cj, cjh9368, Corleone, damon0626, danish, Danish, davidmc, dayschan, doitH, dong-li001, eric, Eric, fary86, fuzhiye, Gaoxiong, GAO_HYP_XYJ, gengdongjie, Gogery, gongdaguo, gray0v0, gukecai, guoqi, gzhcv, hangq, hanhuifeng2020, Harshvardhan, He, heleiwang, hexia, Hoai, HuangBingjian, huangdongrun, huanghui, huangxinjing, huqi, huzhifeng, hwjiaorui, Islam Amin, Jesse, , Jiabin Liu, jianghui58, jiangzhiwen, Jiaqi, jin-xiulang, jinyaohui, jjfeing, John, Jonathan, jonyguo, JulyAi, jzg, kai00, kingfo, kingxian, kpy, kswang, laiyongqiang, leonwanghui, Li, liangchenghui, liangzelang, lichen_101010, lichenever, lihongkang, lilei, limingqi107, ling, linqingke, Lin Xh, liubuyu, liuwenhao4, liuxiao78, liuxiao93, liuyang_655, liuzhongkai, Lixia, lixian, liyanliu, liyong, lizhenyu, luopengting, luoyang, lvchangquan, lvliang, lz, mahdi, Mahdi, maning202007, Margaret_wangrui, mayang, mengyuanli, Ming_blue, nhussain, ougongchang, panfengfeng, panyifeng, Payne, Peilin, peixu_ren, Pengyongrong, qianlong, qianjiahong, r1chardf1d0, riemann_penn, rmdyh, Sheng, shenwei41, simson, Simson, Su, sunsuodong, tao_yunhao, tinazhang, VectorSL, , Wan, wandongdong, wangdongxu, wangmin, wangnan39@huawei.com, wangyue01, wangzhe, wanyiming, Wei, wenchunjiang, wilfChen, WilliamLian, wsc, wudenggang, wukesong, wuweikang, wuxuejian, Xiao Tianci, Xiaoda, xiefangqi, xinyunfan, xuanyue, xulei2020, Xun, xuyongfei, yanghaitao, yanghaitao1, yanghaoran, YangLuo, yangruoqi713, yankai, yanzhenxiang2020, yao_yf, yepei6, yeyunpeng, Yi, yoni, yoonlee666, yuchaojie, yujianfeng, yuximiao, zengzitao, Zhang, zhanghaibo5@huawei.com, zhanghuiyao, zhanghui_china, zhangxinfeng3, zhangyihui, zhangz0911gm, zhanke, zhanyuan, zhaodezan, zhaojichen, zhaoting, zhaozhenlong, zhengjun10, Zhenglong Li, zhiqwang, zhoufeng, zhousiyi, zhouyaqiang, zhouyifengCode, Zichun, Zirui, Ziyan, zjun, ZPaC, wangfengwfwf, zymaa, gerayking.
+
+Contributions of any kind are welcome!
+
+# MindSpore 1.6.1
+
+## MindSpore 1.6.1 Release Notes
+
+### Bug fixes
+
+- Fix the problem that the accuracy of the transformer network decreases.
+- Fix the problem that the accuracy of the warpctc network decreases.
+
+### Contributors
+
+Thanks goes to these wonderful people:
+
+Adel, AGroupofProbiotocs, anthonyaje, anzhengqi, askmiao, baihuawei, baiyangfan, bai-yangfan, bingyaweng, BowenK, buxue, caifubi, CaoJian, caojian05, caozhou, Cathy, changzherui, chenbo116, chenfei, chengxianbin, chenhaozhe, chenjianping, chenzomi, chenzupeng, chujinjin, cj, cjh9368, Corleone, damon0626, danish, Danish, davidmc, dayschan, doitH, dong-li001, eric, Eric, fary86, fuzhiye, Gaoxiong, GAO_HYP_XYJ, gengdongjie, Gogery, gongdaguo, gray0v0, gukecai, guoqi, gzhcv, hangq, hanhuifeng2020, Harshvardhan, He, heleiwang, hexia, Hoai, HuangBingjian, huangdongrun, huanghui, huangxinjing, huqi, huzhifeng, hwjiaorui, Islam Amin, Jesse, , Jiabin Liu, jianghui58, jiangzhiwen, Jiaqi, jin-xiulang, jinyaohui, jjfeing, John, Jonathan, jonyguo, JulyAi, jzg, kai00, kingfo, kingxian, kpy, kswang, laiyongqiang, leonwanghui, Li, liangchenghui, liangzelang, lichen_101010, lichenever, lihongkang, lilei, limingqi107, ling, linqingke, Lin Xh, liubuyu, liuwenhao4, liuxiao78, liuxiao93, liuyang_655, liuzhongkai, Lixia, lixian, liyanliu, liyong, lizhenyu, luopengting, luoyang, lvchangquan, lvliang, lz, mahdi, Mahdi, maning202007, Margaret_wangrui, mayang, mengyuanli, Ming_blue, nhussain, ougongchang, panfengfeng, panyifeng, Payne, Peilin, peixu_ren, Pengyongrong, qianlong, qianjiahong, r1chardf1d0, riemann_penn, rmdyh, Sheng, shenwei41, simson, Simson, Su, sunsuodong, tao_yunhao, tinazhang, VectorSL, , Wan, wandongdong, wangdongxu, wangmin, wangnan39@huawei.com, wangyue01, wangzhe, wanyiming, Wei, wenchunjiang, wilfChen, WilliamLian, wsc, wudenggang, wukesong, wuweikang, wuxuejian, Xiao Tianci, Xiaoda, xiefangqi, xinyunfan, xuanyue, xulei2020, Xun, xuyongfei, yanghaitao, yanghaitao1, yanghaoran, YangLuo, yangruoqi713, yankai, yanzhenxiang2020, yao_yf, yepei6, yeyunpeng, Yi, yoni, yoonlee666, yuchaojie, yujianfeng, yuximiao, zengzitao, Zhang, zhanghaibo5@huawei.com, zhanghuiyao, zhanghui_china, zhangxinfeng3, zhangyihui, zhangz0911gm, zhanke, zhanyuan, zhaodezan, zhaojichen, zhaoting, zhaozhenlong, zhengjun10, Zhenglong Li, zhiqwang, zhoufeng, zhousiyi, zhouyaqiang, zhouyifengCode, Zichun, Zirui, Ziyan, zjun, ZPaC, wangfengwfwf, zymaa, gerayking.
+
+Contributions of any kind are welcome!
+
+# MindSpore 1.6.0
+
+## MindSpore 1.6.0 Release Notes
+
+### Major Features and Improvements
+
+#### OS
+
+- [STABLE] Support macOS with CPU(X86)
+- [BETA] Supoport macOS with CPU(M1)
+
+#### FrontEnd
+
+- [STABLE] Support JIT Fallback feature in Graph mode.
+- [STABLE] Support compile cache feature in Graph mode.
+- [STABLE] Add new optimizers, including ASGD and Rprop.
+- [STABLE] Add new initializers, including Identity, Orthogonal, Dirac, Sparse and VarianceScaling.
+- [STABLE] Support resuming training when an exception occurs in the process.
+- [STABLE] Change `mindspore.nn.LSTMCell` from single-layer LSTM to single-cell LSTM.
+- [BETA] Introduce `mindspore.ops.Custom` to customize your own operators for Ascend(AICore, AICPU), GPU, CPU backends, and the custom type can be one of TBE, AKG, pure Python function or prebuild binary(called aot operator).
+
+#### PyNative
+
+- [STABLE] Support heterogeneous feature in PyNative mode.
+- [STABLE] Optimize memory allocation in PyNative mode.
+
+#### Auto Parallel
+
+- [STABLE] Support configuring the output shard strategy of the MatMul distributed operator.
+- [STABLE] Support multi-instances parallel.
+- [STABLE] Support activation slice communication and calculation overlap in Transformer.
+- [STABLE] Support heterogeneous parallel tensor swap.
+- [STABLE] Add implementations of distributed operator of ResizeNearestNeighbor.
+- [STABLE] Add a communication operator named NeighborExchangeV2 that supports data exchange between adjacent 8 rank ids.
+- [STABLE] Pipeline parallel support GPU platform.
+- [STABLE] Add cell-level data parallel interface.
+- [STABLE] Support gradient AllReduce fusion according to the amount of data.
+- [STABLE] Support a sharding strategy search algorithm called sharding propagation.
+
+#### Executor
+
+- [STABLE] Support multigraph sink and subgraph sink of MindRT.
+- [STABLE] Support memory swap to break the device memory size limit on Ascend platform.
+- [STABLE] Support dynamic deployment of distributed training cluster(GPU).
+- [BETA] Support automatic failover of parameter server.
+
+#### DataSet
+
+- [STABLE] Support overwrite feature in MindRecord.
+- [STABLE] Log improvement and more friendly to users.
+- [BETA] Support new feature [Dataset Offload](https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/enable_dataset_offload.html) to speed up data processing by heterogeneous computing.
+- [BETA] Support new feature [Dataset Autotune](https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/enable_dataset_autotune.html) to adjust parallelism of dataset pipeline automatically.
+
+#### GraphKernel Fusion
+
+- [STABLE] Support kernel fusion and generation for CPU backend.
+
+#### Federated Learning
+
+- [STABLE] FL-Client framework and model decoupling.
+- [BETA] Support Cross-silo federated learning framework.
+
+#### Debug
+
+- [STABLE] Support dump in cell level(Ascend).
+- [STABLE] Support dump Tensor statistics(Ascend/GPU).
+- [STABLE] Support displaying corresponding code lines for fusion nodes.
+- [STABLE] Support passing dump flag in Ascend backend in order to dump correct operators after fusion transformation.
+
+### API Change
+
+#### Backwards Incompatible Change
+
+##### Python API
+
+###### `mindspore.dataset.MindDataset` interface changes input parameter dataset_file([!27542](https://gitee.com/mindspore/mindspore/pulls/27542))
+
+`MindDataset` contains the input parameter `dataset_file`, which is in the singular format. It can receive a single file path or a list that stores multiple file paths. Thus It is preferred to change the input parameter `dataset_file` into plural format. In addition, the input parameters of most dataset API, such as `TFRecordDataset`, are in plural formart (`dataset_files`). To ensure consistency, the input parameter `dataset_file` of MindDataset is changed to plural formart as `dataset_files`,  we can see the updated version in api of [mindspore.dataset.MindDataset](https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/dataset/mindspore.dataset.MindDataset.html#mindspore.dataset.MindDataset).
+
+###### Deprecated usage: `import mindspore.dataset.engine.datasets as ds`. Use `import mindspore.dataset as ds` instead as recommended
+
+We're reconstructed `mindspore/dataset/engine/datasets.py` to `datasets.py`, `datasets_version.py`, `datasets_text.py`, `datasets_audio.py`, `datasets_standard_format.py` and `datasets_user_defined.py`. This is more convenient for subsequent maintenance, the specific dataset loading classes will be scattered in various files, which will cause that the corresponding class cannot be found from the file name, so it is recommended to use a unified usage `import mindspore.dataset as ds`.
+
+###### Delete `mindspore.Tensor`'s property `virtual_flag`([!26989](https://gitee.com/mindspore/mindspore/pulls/26989))
+
+###### Delete `mindspore.Parameter`'s property `is_init`([!26989](https://gitee.com/mindspore/mindspore/pulls/26989))
+
+###### Delete `mindspore.nn.ROC`'s interface `roc`([!25713](https://gitee.com/mindspore/mindspore/pulls/25713))
+
+###### The `shard()` interface of primitive is changed from `shard(strategy)` to `shard(in_strategy=None, out_strategy=None)`
+
+###### The `set_auto_parallel_context()` interface of context is changed from
+
+###### `set_auto_parallel_context(parallel_mode=AUTO_PARALLEL, auto_parallel_search_mode="dynamic_programming")` to ` set_auto_parallel_context(parallel_mode=AUTO_PARALLEL, search_mode="dynamic_programming")`
+
+#### Collect Data and Create Landscape
+
+##### Python API
+
+###### `mindspore.train.callback.SummaryCollector` interface's parameter `collect_specified_data` add new operations `collect_landscape` ([!26229](https://gitee.com/mindspore/mindspore/pulls/26229))
+
+`collect_landscape` can collect the parameters needed to create the loss landscape. we can see the updated version in api of [mindspore.train.callback.SummaryCollector](https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.train.html#mindspore.train.callback.SummaryCollector).
+
+###### `mindspore.train.callback` add new interface `SummaryLandscape` ([!26229](https://gitee.com/mindspore/mindspore/pulls/26229))
+
+`SummaryLandscape` can help you to collect loss landscape information. It can create landscape in PCA direction or random direction by calculating loss. We can see the updated version in api of [mindspore.train.callback.SummaryLandscape](https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.train.html#mindspore.train.callback.SummaryLandscape).
+
+### Bug fixes
+
+#### Executor
+
+- Fix process hanging while calling MPI_comm_create in asymmetric pipeline split scenario. ([!28707](https://gitee.com/mindspore/mindspore/pulls/28707))
+- Fix the execution error when the weights are shared between graph mode and PyNative mode.([!26635](https://gitee.com/mindspore/mindspore/pulls/26635))
+- Fixed the probability coredump when free memory under PyNative mode.([!25472](https://gitee.com/mindspore/mindspore/pulls/25472))
+
+#### Dataset
+
+- Fix memory increase abnormally when running dataset for a long time. ([!26237](https://gitee.com/mindspore/mindspore/pulls/26237))
+- Fix saving MindRecord files with Chinese path on Windows. ([!28378](https://gitee.com/mindspore/mindspore/pulls/28378))
+
+## MindSpore Lite
+
+### Major Features and Improvements
+
+#### Converter and runtime
+
+- [STABLE] Add more fusion patterns in the converter tool to improve runtime performance.
+- [STABLE] Support take OpenGL texture as input and output of inference.
+- [STABLE] Refactor the JAVA API.
+- [BETA] Support inference on Ascend310.
+
+#### x86 backend optimization
+
+- [STABLE] Optimize kernels for x86 using Advanced Vector Extensions(AVX512).
+
+#### ARM backend optimization
+
+- [STABLE] Support heterogeneous parallel inference, including splitting operators, constructing heterogeneous subgraphs, and heterogeneous parallel scheduling between CPUs and GPUs.
+- [STABLE] Add more FP16 operators.
+
+#### Post quantization
+
+- [STABLE] Post quantization supports debugging.
+- [STABLE] Full quantization supports choosing non-quantized nodes.
+- [STABLE] Mixed bit quantization supports auto-tune.
+
+#### Training on Device
+
+- [STABLE] Support user-defined algorithm models to access the federated learning framework.
+
+### Contributors
+
+Thanks goes to these wonderful people:
+
+AGroupofProbiotocs, anzhengqi, askmiao, baihuawei, baiyangfan, bai-yangfan, bingyaweng, BowenK, buxue, caifubi, CaoJian, caojian05, caozhou, Cathy, changzherui, chenbo116, chenfei, chengxianbin, chenhaozhe, chenjianping, chenzomi, chenzupeng, chujinjin, cj, cjh9368, Corleone, damon0626, danish, Danish, davidmc, dayschan, doitH, dong-li001, fary86, fuzhiye, Gaoxiong, GAO_HYP_XYJ, gengdongjie, Gogery, gongdaguo, gray0v0, gukecai, guoqi, gzhcv, hangq, hanhuifeng2020, Harshvardhan, He, heleiwang, hesham, hexia, Hoai, HuangBingjian, huangdongrun, huanghui, huangxinjing, huqi, huzhifeng, hwjiaorui, Jiabin Liu, jianghui58, Jiaqi, jin-xiulang, jinyaohui, jjfeing, John, jonyguo, JulyAi, jzg, kai00, kingfo, kingxian, kpy, kswang, liuyongqi, laiyongqiang, leonwanghui, liangchenghui, liangzelang, lichen_101010, lichenever, lihongkang, lilei, limingqi107, ling, linqingke, Lin Xh, liubuyu, liuwenhao4, liuxiao78, liuxiao93, liuyang_655, liuzhongkai, Lixia, lixian, liyanliu, liyong, lizhenyu, luopengting, lvchangquan, lvliang, lz, maning202007, Margaret_wangrui, mengyuanli, Ming_blue, ms_yan, ougongchang, panfengfeng, panyifeng, Payne, Peilin, peixu_ren, Pengyongrong, qianlong, qianjiahong, r1chardf1d0, riemann_penn, rmdyh, Sheng, shenwei41, simson, Simson, Su, sunsuodong, tao_yunhao, tinazhang, VectorSL, , Wan, wandongdong, wangdongxu, wangmin, [wangnan39@huawei.com](mailto:wangnan39@huawei.com), wangyue01, wangzhe, wanyiming, Wei, wenchunjiang, wilfChen, WilliamLian, wsc, wudenggang, wukesong, wuweikang, wuxuejian, Xiao Tianci, Xiaoda, xiefangqi, xinyunfan, xuanyue, xuyongfei, yanghaitao, yanghaitao1, yanghaoran, YangLuo, yangruoqi713, yankai, yanzhenxiang2020, yao_yf, yepei6, yeyunpeng, Yi, yoni, yoonlee666, yuchaojie, yujianfeng, yuximiao, zengzitao, Zhang, [zhanghaibo5@huawei.com](mailto:zhanghaibo5@huawei.com), zhanghuiyao, zhanghui_china, zhangxinfeng3, zhangyihui, zhangz0911gm, zhanke, zhanyuan, zhaodezan, zhaojichen, zhaoting, zhaozhenlong, zhengjun10, zhiqwang, zhoufeng, zhousiyi, zhouyaqiang, zhouyifengCode, Zichun, Ziyan, zjun, ZPaC, wangfengwfwf, zymaa, gerayking.
+
+Contributions of any kind are welcome!
+
 # MindSpore 1.5.2
 
 ## MindSpore 1.5.2 Release Notes
@@ -417,7 +611,7 @@ thor(net, learning_rate, damping, momentum, weight_decay=0.0, loss_scale=1.0, ba
 
 ##### Dump Config
 
-Previously, we could only dump tensor data for one or all steps. To make the dump feature easier to use, we changed the dump configuration format and dump structure. View the [New Dump Tutorial](https://www.mindspore.cn/docs/programming_guide/en/master/dump_in_graph_mode.html#dump).
+Previously, we could only dump tensor data for one or all steps. To make the dump feature easier to use, we changed the dump configuration format and dump structure. View the [New Dump Tutorial](https://www.mindspore.cn/docs/programming_guide/en/r1.6/dump_in_graph_mode.html#dump).
 
 | 1.2.1                                                  | 1.3.0                                                                                       |
 | ------------------------------------------------------ | ------------------------------------------------------------------------------------------- |
@@ -801,7 +995,7 @@ However, currently MindSpore Parser cannot parse numpy.ndarray in JIT-graph. To
 
 ###### mindspore.numpy interfaces remove support for keyword arguments `out` and `where`([!12726](https://gitee.com/mindspore/mindspore/pulls/12726))
 
-Previously, we have incomplete support for keyword arguments `out` and `where` in mindspore.numpy interfaces, however, the `out` argument is only functional when `where` argument is also provided, and `out` cannot be used to pass reference to numpy functions. Therefore, we have removed these two arguments to avoid any confusion users may have. Their original functionality can be found in [np.where](https://www.mindspore.cn/docs/api/en/master/api_python/numpy/mindspore.numpy.where.html#mindspore.numpy.where)
+Previously, we have incomplete support for keyword arguments `out` and `where` in mindspore.numpy interfaces, however, the `out` argument is only functional when `where` argument is also provided, and `out` cannot be used to pass reference to numpy functions. Therefore, we have removed these two arguments to avoid any confusion users may have. Their original functionality can be found in [np.where](https://www.mindspore.cn/docs/api/en/r1.6/api_python/numpy/mindspore.numpy.where.html#mindspore.numpy.where)
 
 <table>
 <tr>
@@ -1100,7 +1294,7 @@ MSTensor::DestroyTensorPtr(tensor);
 
 ###### `nn.MatMul` is now deprecated in favor of `ops.matmul` ([!12817](https://gitee.com/mindspore/mindspore/pulls/12817))
 
-[ops.matmul](https://www.mindspore.cn/docs/api/en/master/api_python/ops/mindspore.ops.matmul.html#mindspore.ops.matmul) follows the API of [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html) as closely as possible. As a function interface, [ops.matmul](https://www.mindspore.cn/docs/api/en/master/api_python/ops/mindspore.ops.matmul.html#mindspore.ops.matmul) is applied without instantiation, as opposed to `nn.MatMul`, which should only be used as a class instance.
+[ops.matmul](https://www.mindspore.cn/docs/api/en/r1.6/api_python/ops/mindspore.ops.matmul.html#mindspore.ops.matmul) follows the API of [numpy.matmul](https://numpy.org/doc/stable/reference/generated/numpy.matmul.html) as closely as possible. As a function interface, [ops.matmul](https://www.mindspore.cn/docs/api/en/r1.6/api_python/ops/mindspore.ops.matmul.html#mindspore.ops.matmul) is applied without instantiation, as opposed to `nn.MatMul`, which should only be used as a class instance.
 
 <table>
 <tr>

akg

@@ -1 +1 @@
-Subproject commit 288658ca72e8de65b17c3109170d1ef0d511eb28
+Subproject commit e53ed0355b0cb1422ee406755196c964ebd4c4ad

cmake/check_requirements.cmake

@@ -75,3 +75,60 @@ if(NOT CMAKE_SYSTEM_NAME MATCHES "Windows")
         find_required_package(FLEX)
     endif()
 endif()
+
+# for macos, find appropriate macosx SDK then set SDKROOT and MACOSX_DEPLOYMENT_TARGET
+if(${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
+    if(NOT DEFINED ENV{SDKROOT})
+        # arm64: macosx11.x
+        # x86_64: macosx10.x, macosx11.x
+        if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "arm64")
+            set(MACOSX_SDK_REGEX "MacOSX11(\\.\\d+)?")
+        else()
+            set(MACOSX_SDK_REGEX "MacOSX1[01](\\.\\d+)?")
+        endif()
+        set(MACOSX_XCODE_SDK_PATH "/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs")
+        set(MACOSX_CLT_SDK_PATH "/Library/Developer/CommandLineTools/SDKs")
+        set(MACOSX_SDK_SEARCH_PATHS "${MACOSX_XCODE_SDK_PATH}/*" "${MACOSX_CLT_SDK_PATH}/*")
+        file(GLOB ALL_SDK_NAME ${MACOSX_SDK_SEARCH_PATHS})
+        # get highest SDK version meets the requirements
+        execute_process(
+            COMMAND bash -c "echo '${ALL_SDK_NAME}' | grep -Eo '${MACOSX_SDK_REGEX}' | sort -n | tail -1 | tr -d '\\n'"
+            OUTPUT_VARIABLE MACOSX_FIND_SDK_NAME
+        )
+        if(NOT MACOSX_FIND_SDK_NAME)
+            message(FATAL_ERROR
+                "can not find appropriate macosx SDK, find in ${ALL_SDK_NAME}, you may set SDKROOT manually"
+            )
+        endif()
+        if(IS_DIRECTORY "${MACOSX_XCODE_SDK_PATH}/${MACOSX_FIND_SDK_NAME}.sdk")
+            set(CMAKE_OSX_SYSROOT "${MACOSX_XCODE_SDK_PATH}/${MACOSX_FIND_SDK_NAME}.sdk")
+        else()
+            set(CMAKE_OSX_SYSROOT "${MACOSX_CLT_SDK_PATH}/${MACOSX_FIND_SDK_NAME}.sdk")
+        endif()
+        set(ENV{SDKROOT} ${CMAKE_OSX_SYSROOT})
+    endif()
+    message("macosx sdkroot: $ENV{SDKROOT}")
+    # set macosx deployment target based on SDK
+    if(NOT DEFINED ENV{MACOSX_DEPLOYMENT_TARGET})
+        execute_process(
+            COMMAND bash -c "cat $ENV{SDKROOT}/SDKSettings.json | \
+                grep -Eo 'MACOSX_DEPLOYMENT_TARGET\\\":\\\"\\d{2}\\.\\d+' | cut -d '\"' -f 3 | tr -d '\\n'"
+            OUTPUT_VARIABLE MACOSX_FIND_SDK_VERSION
+        )
+        if(NOT MACOSX_FIND_SDK_VERSION)
+            message(FATAL_ERROR "can not find MACOSX_DEPLOYMENT_TARGET in SDKROOT, \
+                please check whether it's a valid SDK path")
+        endif()
+
+        if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "arm64")
+            set(CMAKE_OSX_DEPLOYMENT_TARGET "11.0")
+        elseif(${MACOSX_FIND_SDK_VERSION} VERSION_LESS "10.15")
+            set(CMAKE_OSX_DEPLOYMENT_TARGET ${MACOSX_FIND_SDK_VERSION} CACHE STRING
+                "minimum macosx deployment target version" FORCE)
+        else()
+            set(CMAKE_OSX_DEPLOYMENT_TARGET "10.15")
+        endif()
+        set(ENV{MACOSX_DEPLOYMENT_TARGET} ${CMAKE_OSX_DEPLOYMENT_TARGET})
+    endif()
+    message("macosx deployment target version: $ENV{MACOSX_DEPLOYMENT_TARGET}")
+endif()

cmake/external_libs/openssl.cmake

@@ -21,6 +21,7 @@ if(${CMAKE_SYSTEM_NAME} MATCHES "Linux" OR APPLE)
     CONFIGURE_COMMAND ./config no-zlib no-shared
     PATCHES ${OPENSSL_PATCH_ROOT}/CVE-2021-3711.patch
     PATCHES ${OPENSSL_PATCH_ROOT}/CVE-2021-3712.patch
+    PATCHES ${OPENSSL_PATCH_ROOT}/CVE-2022-0778.patch
   )
   include_directories(${openssl_INC})
   add_library(mindspore::ssl ALIAS openssl::ssl)

cmake/external_libs/protobuf.cmake

@@ -44,6 +44,12 @@ else()
     set(MD5 "1a6274bc4a65b55a6fa70e264d796490")
 endif()
 
+if(BUILD_LITE)
+  set(PROTOBUF_PATCH_ROOT ${TOP_DIR}/third_party/patch/protobuf)
+else()
+  set(PROTOBUF_PATCH_ROOT ${CMAKE_SOURCE_DIR}/third_party/patch/protobuf)
+endif()
+
 mindspore_add_pkg(protobuf
         VER 3.13.0
         LIBS protobuf
@@ -51,7 +57,8 @@ mindspore_add_pkg(protobuf
         URL ${REQ_URL}
         MD5 ${MD5}
         CMAKE_PATH cmake/
-        CMAKE_OPTION -Dprotobuf_BUILD_TESTS=OFF -Dprotobuf_BUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE=Release)
+        CMAKE_OPTION -Dprotobuf_BUILD_TESTS=OFF -Dprotobuf_BUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE=Release
+        PATCHES ${PROTOBUF_PATCH_ROOT}/CVE-2021-22570.patch)
 
 include_directories(${protobuf_INC})
 add_library(mindspore::protobuf ALIAS protobuf::protobuf)

cmake/package_lite.cmake

@@ -239,6 +239,10 @@ if(PLATFORM_ARM64)
     if(NOT TARGET_MIX210)
         __install_micro_wrapper()
     endif()
+    if(MSLITE_ENABLE_RUNTIME_GLOG)
+        install(FILES ${glog_LIBPATH}/libglog.so.0.4.0 DESTINATION ${GLOG_DIR} RENAME libglog.so.0
+                COMPONENT ${RUNTIME_COMPONENT_NAME})
+    endif()
     if(MSLITE_ENABLE_TOOLS)
         if(NOT MSLITE_COMPILE_TWICE)
             install(TARGETS ${BENCHMARK_NAME} RUNTIME DESTINATION ${BENCHMARK_ROOT_DIR}
@@ -265,6 +269,142 @@ if(PLATFORM_ARM64)
             install(TARGETS ${BENCHMARK_TRAIN_NAME} RUNTIME DESTINATION ${BENCHMARK_TRAIN_ROOT_DIR} COMPONENT
                     ${RUNTIME_COMPONENT_NAME})
         endif()
+        if(MSLITE_ENABLE_CONVERTER)
+            install(DIRECTORY ${TOP_DIR}/mindspore/lite/include/ DESTINATION ${CONVERTER_ROOT_DIR}/include
+                    COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h"
+                    PATTERN "train*" EXCLUDE PATTERN "delegate.h" EXCLUDE PATTERN "lite_session.h" EXCLUDE)
+            install(FILES ${API_HEADER}  DESTINATION ${CONVERTER_ROOT_DIR}/include/api
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${MINDAPI_BASE_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/mindapi/base
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${MINDAPI_IR_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/mindapi/ir
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${ABSTRACT_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/abstract
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${API_IR_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/api/ir
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${BASE_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/base
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${IR_DTYPE_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/ir/dtype
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${IR_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/ir
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(DIRECTORY ${TOP_DIR}/mindspore/core/ops/ DESTINATION ${CONVERTER_ROOT_DIR}/include/core/ops
+                    COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+            install(FILES ${UTILS_HEADER} DESTINATION ${CONVERTER_ROOT_DIR}/include/core/utils
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(DIRECTORY ${TOP_DIR}/mindspore/lite/build/schema/
+                    DESTINATION ${CONVERTER_ROOT_DIR}/include/schema
+                    COMPONENT ${RUNTIME_COMPONENT_NAME}
+                    FILES_MATCHING PATTERN "*.h" PATTERN "schema_generated.h" EXCLUDE)
+            install(DIRECTORY ${flatbuffers_INC}/ DESTINATION ${CONVERTER_ROOT_DIR}/include/third_party
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(DIRECTORY ${glog_LIBPATH}/../include/glog/
+                    DESTINATION ${CONVERTER_ROOT_DIR}/include/third_party/glog
+                    COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+            install(DIRECTORY ${TOP_DIR}/third_party/securec/include/
+                    DESTINATION ${CONVERTER_ROOT_DIR}/include/third_party/securec
+                    COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+            install(TARGETS converter_lite RUNTIME DESTINATION ${CONVERTER_ROOT_DIR}/converter
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${TOP_DIR}/mindspore/lite/build/tools/converter/registry/libmslite_converter_plugin.so
+                    DESTINATION ${CONVERTER_ROOT_DIR}/lib COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${glog_LIBPATH}/libglog.so.0.4.0 DESTINATION ${CONVERTER_ROOT_DIR}/lib RENAME libglog.so.0
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+
+            install(FILES ${opencv_LIBPATH}/libopencv_core.so.4.5.2
+                    DESTINATION ${CONVERTER_ROOT_DIR}/lib RENAME libopencv_core.so.4.5
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${opencv_LIBPATH}/libopencv_imgcodecs.so.4.5.2
+                    DESTINATION ${CONVERTER_ROOT_DIR}/lib RENAME libopencv_imgcodecs.so.4.5
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+            install(FILES ${opencv_LIBPATH}/libopencv_imgproc.so.4.5.2
+                    DESTINATION ${CONVERTER_ROOT_DIR}/lib RENAME libopencv_imgproc.so.4.5
+                    COMPONENT ${RUNTIME_COMPONENT_NAME})
+
+            if(MSLITE_ENABLE_ACL)
+                set(LITE_ACL_DIR ${TOP_DIR}/mindspore/lite/build/tools/converter/adapter/acl)
+                install(FILES ${LITE_ACL_DIR}/mindspore_shared_lib/libmindspore_shared_lib.so
+                        DESTINATION ${CONVERTER_ROOT_DIR}/lib COMPONENT ${RUNTIME_COMPONENT_NAME})
+                if(MSLITE_ENABLE_RUNTIME_CONVERT)
+                    install(FILES ${LITE_ACL_DIR}/mindspore_shared_lib/libmindspore_shared_lib.so
+                            DESTINATION ${RUNTIME_LIB_DIR} COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${glog_LIBPATH}/libglog.so.0.4.0 DESTINATION ${RUNTIME_LIB_DIR} RENAME libglog.so.0
+                            COMPONENT ${RUNTIME_COMPONENT_NAME})
+                endif()
+                if(MSLITE_MINDDATA_IMPLEMENT STREQUAL "cloud" AND MSLITE_ENABLE_RUNTIME_CONVERT)
+                    file(GLOB DATA_ENGINE_LIB_LIST ${LITE_ACL_DIR}/_c_dataengine/*.so)
+                    file(GLOB DATA_RECORD_LIB_LIST ${LITE_ACL_DIR}/_c_mindrecord/*.so)
+                    install(FILES ${DATA_ENGINE_LIB_LIST}
+                            DESTINATION ${RUNTIME_LIB_DIR} COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${DATA_RECORD_LIB_LIST}
+                            DESTINATION ${RUNTIME_LIB_DIR} COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${jpeg_turbo_LIBPATH}/libjpeg.so.62.3.0
+                            DESTINATION ${RUNTIME_LIB_DIR} RENAME libjpeg.so.62 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${jpeg_turbo_LIBPATH}/libturbojpeg.so.0.2.0
+                            DESTINATION ${RUNTIME_LIB_DIR} RENAME libturbojpeg.so.0 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${tinyxml2_LIBPATH}/libtinyxml2.so.8.0.0
+                            DESTINATION ${RUNTIME_LIB_DIR} RENAME libtinyxml2.so.8 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${icu4c_LIBPATH}/libicuuc.so.67.1
+                            DESTINATION ${RUNTIME_LIB_DIR} RENAME libicuuc.so.67 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${icu4c_LIBPATH}/libicudata.so.67.1
+                            DESTINATION ${RUNTIME_LIB_DIR} RENAME libicudata.so.67 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${icu4c_LIBPATH}/libicui18n.so.67.1
+                            DESTINATION ${RUNTIME_LIB_DIR} RENAME libicui18n.so.67 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${grpc_LIBPATH}/libmindspore_grpc++.so.1.36.1 DESTINATION ${RUNTIME_LIB_DIR}
+                            RENAME libmindspore_grpc++.so.1 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${grpc_LIBPATH}/libmindspore_grpc.so.15.0.0 DESTINATION
+                            ${RUNTIME_LIB_DIR} RENAME libmindspore_grpc.so.15 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${grpc_LIBPATH}/libmindspore_gpr.so.15.0.0 DESTINATION
+                            ${RUNTIME_LIB_DIR} RENAME libmindspore_gpr.so.15 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${grpc_LIBPATH}/libmindspore_upb.so.15.0.0 DESTINATION
+                            ${RUNTIME_LIB_DIR} RENAME libmindspore_upb.so.15 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    install(FILES ${grpc_LIBPATH}/libmindspore_address_sorting.so.15.0.0 DESTINATION ${RUNTIME_LIB_DIR}
+                            RENAME libmindspore_address_sorting.so.15 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                    ## Public header files for minddata
+                    install(
+                            FILES ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/config.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/constants.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/execute.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/text.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/transforms.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/vision.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/vision_lite.h
+                            ${TOP_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/vision_ascend.h
+                            DESTINATION ${RUNTIME_INC_DIR}/dataset COMPONENT ${RUNTIME_COMPONENT_NAME})
+                endif()
+            endif()
+
+            if(MSLITE_ENABLE_DPICO_ATC_ADAPTER)
+                install(FILES ${TOP_DIR}/mindspore/lite/build/tools/converter/adapter/dpico/libdpico_atc_adapter.so
+                        DESTINATION ${CONVERTER_ROOT_DIR}/providers/SD3403 COMPONENT ${RUNTIME_COMPONENT_NAME})
+                if(MSLITE_ENABLE_TOOLS)
+                    install(TARGETS ${BECHCHMARK_NAME} RUNTIME DESTINATION ${BENCHMARK_ROOT_DIR}
+                            COMPONENT ${RUNTIME_COMPONENT_NAME})
+                endif()
+            endif()
+
+            if(MSLITE_ENABLE_RUNTIME_GLOG)
+                install(DIRECTORY ${glog_LIBPATH}/../include/glog/ DESTINATION ${RUNTIME_INC_DIR}/third_party/glog
+                        COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+                install(FILES ${glog_LIBPATH}/libglog.so.0.4.0 DESTINATION ${GLOG_DIR} RENAME libglog.so.0
+                        COMPONENT ${RUNTIME_COMPONENT_NAME})
+            endif()
+            if(MSLITE_ENABLE_RUNTIME_CONVERT)
+                install(FILES ${TOP_DIR}/mindspore/lite/build/tools/converter/registry/libmslite_converter_plugin.so
+                        DESTINATION ${RUNTIME_LIB_DIR} COMPONENT ${RUNTIME_COMPONENT_NAME})
+
+                install(FILES ${opencv_LIBPATH}/libopencv_core.so.4.5.2
+                        DESTINATION ${RUNTIME_LIB_DIR} RENAME libopencv_core.so.4.5
+                        COMPONENT ${RUNTIME_COMPONENT_NAME})
+                install(FILES ${opencv_LIBPATH}/libopencv_imgcodecs.so.4.5.2
+                        DESTINATION ${RUNTIME_LIB_DIR} RENAME libopencv_imgcodecs.so.4.5
+                        COMPONENT ${RUNTIME_COMPONENT_NAME})
+                install(FILES ${opencv_LIBPATH}/libopencv_imgproc.so.4.5.2
+                        DESTINATION ${RUNTIME_LIB_DIR} RENAME libopencv_imgproc.so.4.5
+                        COMPONENT ${RUNTIME_COMPONENT_NAME})
+            endif()
+        endif()
     endif()
     if(MSLITE_ENABLE_TESTCASES)
         install(FILES ${TOP_DIR}/mindspore/lite/build/test/lite-test DESTINATION ${TEST_CASE_DIR}
@@ -495,6 +635,10 @@ else()
         install(FILES ${TOP_DIR}/mindspore/lite/tools/obfuscator/lib/linux-x64/libmsdeobfuscator-lite.so
                 DESTINATION ${RUNTIME_LIB_DIR} COMPONENT ${RUNTIME_COMPONENT_NAME})
     endif()
+    if(MSLITE_ENABLE_RUNTIME_GLOG)
+        install(FILES ${glog_LIBPATH}/libglog.so.0.4.0 DESTINATION ${GLOG_DIR} RENAME libglog.so.0
+                COMPONENT ${RUNTIME_COMPONENT_NAME})
+    endif()
     if(MSLITE_ENABLE_CONVERTER)
         install(DIRECTORY ${TOP_DIR}/mindspore/lite/include/ DESTINATION ${CONVERTER_ROOT_DIR}/include
                 COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h"

cmake/package_tar.cmake

@@ -141,6 +141,14 @@ install(
         COMPONENT mindspore
 )
 
+## Public header files for mindapi
+install(
+        DIRECTORY ${CMAKE_SOURCE_DIR}/mindspore/core/mindapi/base
+        ${CMAKE_SOURCE_DIR}/mindspore/core/mindapi/ir
+        DESTINATION ${INSTALL_BASE_DIR}/include/mindapi
+        COMPONENT mindspore
+)
+
 ## Public header files for minddata
 install(
         FILES ${CMAKE_SOURCE_DIR}/mindspore/ccsrc/minddata/dataset/include/dataset/config.h

cmake/utils.cmake

@@ -114,6 +114,7 @@ function(__find_pkg_then_add_target pkg_name pkg_exe lib_path)
     message("_FIND:${${pkg_name}_BASE_DIR}")
 
     if(pkg_exe)
+        unset(${pkg_exe}_EXE CACHE)
         find_program(${pkg_exe}_EXE ${pkg_exe} PATHS ${${pkg_name}_BASE_DIR}/bin NO_DEFAULT_PATH)
         if(NOT ${pkg_exe}_EXE)
             return()
@@ -206,7 +207,6 @@ endfunction()
 set(MS_FIND_NO_DEFAULT_PATH NO_CMAKE_PATH NO_CMAKE_ENVIRONMENT_PATH NO_SYSTEM_ENVIRONMENT_PATH
                             NO_CMAKE_BUILDS_PATH NO_CMAKE_PACKAGE_REGISTRY NO_CMAKE_SYSTEM_PATH
                             NO_CMAKE_SYSTEM_PACKAGE_REGISTRY)
-set(MS_FIND_NO_DEFAULT_PATH ${MS_FIND_NO_DEFAULT_PATH} PARENT_SCOPE)
 function(mindspore_add_pkg pkg_name)
 
     set(options)
@@ -239,6 +239,9 @@ function(mindspore_add_pkg pkg_name)
             "${CMAKE_CXX_COMPILER_VERSION}-${CMAKE_C_COMPILER_VERSION}
             ${ARGN} - ${${pkg_name}_USE_STATIC_LIBS}- ${${pkg_name}_PATCHES_HASH}
             ${${pkg_name}_CXXFLAGS}--${${pkg_name}_CFLAGS}--${${pkg_name}_LDFLAGS}")
+    if(${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
+        set(${pkg_name}_CONFIG_TXT "${${pkg_name}_CONFIG_TXT}--${CMAKE_OSX_DEPLOYMENT_TARGET}")
+    endif()
     string(REPLACE ";" "-" ${pkg_name}_CONFIG_TXT ${${pkg_name}_CONFIG_TXT})
     string(MD5 ${pkg_name}_CONFIG_HASH ${${pkg_name}_CONFIG_TXT})
 
@@ -268,7 +271,7 @@ function(mindspore_add_pkg pkg_name)
             return()
         endif()
     elseif(NOT PKG_HEAD_ONLY)
-        find_package(${__FIND_PKG_NAME} ${PKG_VER} ${MS_FIND_NO_DEFAULT_PATH})
+        find_package(${__FIND_PKG_NAME} ${PKG_VER} PATHS ${${pkg_name}_BASE_DIR} ${MS_FIND_NO_DEFAULT_PATH})
         if(${__FIND_PKG_NAME}_FOUND)
             set(${pkg_name}_INC ${${pkg_name}_BASE_DIR}/include PARENT_SCOPE)
             message("Found pkg: ${__FIND_PKG_NAME}")

config/op_info.config

@@ -619,7 +619,7 @@
 {"op_name": "Mod", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""], ["uint8", ""]], [["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "mod.so", "compute_cost": 10, "kernel_name": "mod", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "MaxPoolGradGrad", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "grad", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_mode", "param_type": "required", "type": "str", "value": "all"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "max_pool_grad_grad.so", "compute_cost": 10, "kernel_name": "max_pool_grad_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "MaxPoolGradGradWithArgmax", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "argmax", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_mode", "param_type": "required", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["uint16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["int64", "NC1HWC0"], ["float16", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "max_pool_grad_grad_with_argmax.so", "compute_cost": 10, "kernel_name": "max_pool_grad_grad_with_argmax", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "TensorMove", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]], [["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "tensor_move.so", "compute_cost": 10, "kernel_name": "tensor_move", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
+{"op_name": "TensorMove", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["bool", ""], ["bool", ""]], [["int8", ""], ["int8", ""]], [["int16", ""], ["int16", ""]], [["int32", ""], ["int32", ""]], [["int64", ""], ["int64", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]], [["float64", ""], ["float64", ""]], [["uint8", ""], ["uint8", ""]], [["uint16", ""], ["uint16", ""]], [["uint32", ""], ["uint32", ""]], [["uint64", ""], ["uint64", ""]], [["complex64", "DefaultFormat"], ["complex64", "DefaultFormat"]], [["complex128", "DefaultFormat"], ["complex128", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "tensor_move.so", "compute_cost": 10, "kernel_name": "tensor_move", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "PopulationCount", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int16", "NC1HWC0"], ["uint8", "NC1HWC0"]], [["int16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "NC1HWC0"], ["uint8", "NC1HWC0"]], [["uint16", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "population_count.so", "compute_cost": 10, "kernel_name": "population_count", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ParallelConcat", "inputs": [{"index": 0, "name": "values", "need_compile": false, "param_type": "dynamic", "shape": "all"}], "outputs": [{"index": 0, "name": "output_data", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "shape", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "N", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["bool", "NC1HWC0"], ["bool", "NC1HWC0"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int8", "NC1HWC0"], ["int8", "NC1HWC0"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint8", "NC1HWC0"], ["uint8", "NC1HWC0"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int16", "NC1HWC0"], ["int16", "NC1HWC0"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint16", "NC1HWC0"], ["uint16", "NC1HWC0"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "NC1HWC0"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint32", "NC1HWC0"], ["uint32", "NC1HWC0"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int64", "NC1HWC0"], ["int64", "NC1HWC0"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint64", "NC1HWC0"], ["uint64", "NC1HWC0"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["bool", "NHWC"], ["bool", "NHWC"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "NHWC"], ["int8", "NHWC"]], [["uint8", "NHWC"], ["uint8", "NHWC"]], [["int16", "NHWC"], ["int16", "NHWC"]], [["uint16", "NHWC"], ["uint16", "NHWC"]], [["int32", "NHWC"], ["int32", "NHWC"]], [["uint32", "NHWC"], ["uint32", "NHWC"]], [["int64", "NHWC"], ["int64", "NHWC"]], [["uint64", "NHWC"], ["uint64", "NHWC"]], [["float16", "NHWC"], ["float16", "NHWC"]], [["float32", "NHWC"], ["float32", "NHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "parallel_concat.so", "compute_cost": 10, "kernel_name": "parallel_concat", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "AdamApplyOneAssign", "inputs": [{"index": 0, "name": "input0", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "input2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "input3", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "input4", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "mul0_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "mul1_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 7, "name": "mul2_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 8, "name": "mul3_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 9, "name": "add2_y", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output0", "need_compile": true, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output1", "need_compile": true, "param_type": "required", "shape": "all"}, {"index": 2, "name": "output2", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "adam_apply_one_assign.so", "compute_cost": 10, "kernel_name": "adam_apply_one_assign", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "dynamic_compile_static": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}

docs/api/api_python/dataset/mindspore.dataset.CLUEDataset.rst

@@ -23,7 +23,7 @@ mindspore.dataset.CLUEDataset
 
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后, `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     根据给定的 `task` 参数配置，数据集会生成不同的输出列：
 

docs/api/api_python/dataset/mindspore.dataset.CSVDataset.rst

@@ -22,7 +22,7 @@
 
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后, `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.CelebADataset.rst

@@ -19,7 +19,7 @@ mindspore.dataset.CelebADataset
     - **num_samples** (int, 可选) - 指定从数据集中读取的样本数，可以小于数据集总数。默认值：None，读取全部样本图片。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.Cifar100Dataset.rst

@@ -18,7 +18,7 @@ mindspore.dataset.Cifar100Dataset
     - **sampler** (Sampler, 可选) - 指定从数据集中选取样本的采样器，默认值：None，下表中会展示不同配置的预期行为。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.Cifar10Dataset.rst

@@ -18,7 +18,7 @@ mindspore.dataset.Cifar10Dataset
     - **sampler** (Sampler, 可选) - 指定从数据集中选取样本的采样器，默认值：None，下表中会展示不同配置的预期行为。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.CocoDataset.rst

@@ -17,7 +17,7 @@
     - **sampler** (Sampler, 可选) - 指定从数据集中选取样本的采样器，默认值：None，下表中会展示不同配置的预期行为。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
     - **extra_metadata** (bool, 可选) - 用于指定是否额外输出一个数据列用于表示图片元信息。如果为True，则将额外输出一个名为 `[_meta-filename, dtype=string]` 的数据列，默认值：False。
 
     根据不同 `task` 参数设置，生成数据集具有不同的输出列：

docs/api/api_python/dataset/mindspore.dataset.DatasetCache.rst

@@ -5,8 +5,8 @@ mindspore.dataset.DatasetCache
 
     创建数据缓存客户端实例。
 
-    关于单节点数据缓存的使用，请参阅 `单节点数据缓存教程 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/enable_cache.html>`_ 、
-    `单节点数据缓存编程指南 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_。
+    关于单节点数据缓存的使用，请参阅 `单节点数据缓存教程 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/enable_cache.html>`_ 、
+    `单节点数据缓存编程指南 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_。
 
     **参数：**
 

docs/api/api_python/dataset/mindspore.dataset.ImageFolderDataset.rst

@@ -19,7 +19,7 @@ mindspore.dataset.ImageFolderDataset
     - **decode** (bool, 可选) - 是否对读取的图片进行解码操作，默认值：False，不解码。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.ManifestDataset.rst

@@ -19,7 +19,7 @@
     - **decode** (bool, 可选) - 是否对读取的图片进行解码操作，默认值：False，不解码。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.MindDataset.rst

@@ -24,7 +24,7 @@
     - **padded_sample** (dict, 可选): 指定额外添加到数据集的样本，可用于在分布式训练时补齐分片数据，注意字典的键名需要与 `column_list` 指定的列名相同。默认值：None，不添加样本。需要与 `num_padded` 参数同时使用。
     - **num_padded** (int, 可选) - 指定额外添加的数据集样本的数量。在分布式训练时可用于为数据集补齐样本，使得总样本数量可被 `num_shards` 整除。默认值：None，不添加样本。需要与 `padded_sample` 参数同时使用。
     - **num_samples** (int, 可选) - 指定从数据集中读取的样本数。默认值：None，读取所有样本。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.MnistDataset.rst

@@ -18,7 +18,7 @@ mindspore.dataset.MnistDataset
     - **sampler** (Sampler, 可选) - 指定从数据集中选取样本的采样器，默认值：None，下表中会展示不同配置的预期行为。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.TFRecordDataset.rst

@@ -27,7 +27,7 @@ mindspore.dataset.TFRecordDataset
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后, `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
     - **shard_equal_rows** (bool, 可选) - 分布式训练时，为所有分片获取等量的数据行数。默认值：False。如果 `shard_equal_rows` 为False，则可能会使得每个分片的数据条目不相等，从而导致分布式训练失败。因此当每个TFRecord文件的数据数量不相等时，建议将此参数设置为True。注意，只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.TextFileDataset.rst

@@ -19,7 +19,7 @@
 
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后, `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读 `单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
 
     **异常：**
 

docs/api/api_python/dataset/mindspore.dataset.VOCDataset.rst

@@ -22,7 +22,7 @@ mindspore.dataset.VOCDataset
     - **sampler** (Sampler, 可选) - 指定从数据集中选取样本的采样器，默认值：None，下表中会展示不同配置的预期行为。
     - **num_shards** (int, 可选) - 指定分布式训练时将数据集进行划分的分片数，默认值：None。指定此参数后， `num_samples` 表示每个分片的最大样本数。
     - **shard_id** (int, 可选) - 指定分布式训练时使用的分片ID号，默认值：None。只有当指定了 `num_shards` 时才能指定此参数。
-    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_ 。默认值：None，不使用缓存。
+    - **cache** (DatasetCache, 可选) - 单节点数据缓存服务，用于加快数据集处理，详情请阅读`单节点数据缓存 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_ 。默认值：None，不使用缓存。
     - **extra_metadata** (bool, 可选) - 用于指定是否额外输出一个数据列用于表示图片元信息。如果为True，则将额外输出一个名为 `[_meta-filename, dtype=string]` 的数据列，默认值：False。
 
     根据给定的`task`配置，生成数据集具有不同的输出列：

docs/api/api_python/dataset/mindspore.dataset.WaitedDSCallback.rst

@@ -3,14 +3,14 @@ mindspore.dataset.WaitedDSCallback
 
 .. py:class:: mindspore.dataset.WaitedDSCallback(step_size=1)
 
-    数据集自定义回调类的抽象基类，用于与训练回调类(`mindspore.callback <https://mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.train.html#mindspore.train.callback.Callback>`_)的同步。
+    数据集自定义回调类的抽象基类，用于与训练回调类 `mindspore.train.callback <https://mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.train.html#mindspore.train.callback.Callback>`_ 的同步。
 
     可用于在每个step或epoch开始前执行自定义的回调方法，注意，第二个step或epoch开始时才会触发该调用。
     例如在自动数据增强中根据上一个epoch的loss值来更新增强算子参数配置。
 
-    用户可通过 `train_run_context` 获取模型相关信息。如 `network` 、 `train_network` 、 `epoch_num` 、 `batch_num` 、 `loss_fn` 、 `optimizer` 、 `parallel_mode` 、 `device_number` 、 `list_callback` 、 `cur_epoch_num` 、 `cur_step_num` 、 `dataset_sink_mode` 、 `net_outputs` 等，详见 `mindspore.callback <https://mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.train.html#mindspore.train.callback.Callback>`_ 。
+    用户可通过 `train_run_context` 获取网络训练相关信息，如 `network` 、 `train_network` 、 `epoch_num` 、 `batch_num` 、 `loss_fn` 、 `optimizer` 、 `parallel_mode` 、 `device_number` 、 `list_callback` 、 `cur_epoch_num` 、 `cur_step_num` 、 `dataset_sink_mode` 、 `net_outputs` 等，详见 `mindspore.train.callback <https://mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.train.html#mindspore.train.callback.Callback>`_ 。
 
-    用户可通过 `ds_run_context` 获取数据处理管道相关信息。包括 `cur_epoch_num` (当前epoch数)、 `cur_step_num_in_epoch` (当前epoch的step数)、 `cur_step_num` (当前step数)。
+    用户可通过 `ds_run_context` 获取数据处理管道相关信息，包括 `cur_epoch_num` (当前epoch数)、 `cur_step_num_in_epoch` (当前epoch的step数)、 `cur_step_num` (当前step数)。
 
     **参数：**
 

docs/api/api_python/mindspore.context.rst

@@ -106,7 +106,7 @@ MindSpore上下文，用于配置当前执行环境，包括执行模式、执
 
     - **pynative_synchronize** (bool) - 表示是否在PyNative模式下启动设备同步执行。默认值：False。设置为False时，将在设备上异步执行算子。当算子执行出错时，将无法定位特定错误脚本代码的位置。当设置为True时，将在设备上同步执行算子。这将降低程序的执行性能。此时，当算子执行出错时，可以根据错误的调用栈来定位错误脚本代码的位置。
     - **mode** (int) - 表示在GRAPH_MODE(0)或PYNATIVE_MODE(1)模式中的运行。默认值：GRAPH_MODE(0)。GRAPH_MODE或PYNATIVE_MODE可以通过 `mode` 属性设置，两种模式都支持所有后端。默认模式为GRAPH_MODE。
-    - **enable_graph_kernel** (bool) - 表示是否启用图算融合去优化网络执行性能。默认值：False。如果 `enable_graph_kernel` 设置为True，则可以启用加速。有关图算融合的详细信息，请查看 `使能图算融合 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/enable_graph_kernel_fusion.html>`_ 。
+    - **enable_graph_kernel** (bool) - 表示是否启用图算融合去优化网络执行性能。默认值：False。如果 `enable_graph_kernel` 设置为True，则可以启用加速。有关图算融合的详细信息，请查看 `使能图算融合 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/enable_graph_kernel_fusion.html>`_ 。
     - **graph_kernel_flags** (str) - 图算融合的优化选项，当与enable_graph_kernel冲突时，它的优先级更高。其仅适用于有经验的用户。例如，context.set_context(graph_kernel_flags="--opt_level=2 --dump_as_text")。一些常用选项：
 
       - **opt_level**：设置优化级别。默认值：2。当opt_level的值大于0时，启动图算融合。可选值包括：
@@ -128,11 +128,11 @@ MindSpore上下文，用于配置当前执行环境，包括执行模式、执
       - RL，GA：当RL和GA优化同时打开时，工具会根据网络模型中的不同算子类型自动选择RL或GA。RL和GA的顺序没有区别。（自动选择）。
 
 
-      有关启用算子调优工具设置的更多信息，请查看 `使能算子调优工具 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/enable_auto_tune.html>`_。
+      有关启用算子调优工具设置的更多信息，请查看 `使能算子调优工具 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/enable_auto_tune.html>`_。
 
     - **check_bprop** (bool) - 表示是否检查反向传播节点，以确保反向传播节点输出的形状(shape)和数据类型与输入参数相同。默认值：False。
     - **max_call_depth** (int) - 指定函数调用的最大深度。其值必须为正整数。默认值：1000。当嵌套Cell太深或子图数量太多时，需要设置 `max_call_depth` 参数。如果 `max_call_depth` 的值比以前的大，则应把系统最大堆栈深度设得更大，否则可能会因为系统堆栈溢出而引发 `core dumped` 异常。
-    - **enable_sparse** (bool) - 表示是否启用稀疏特征。默认值：False。有关稀疏特征和稀疏张量的详细信息，请查看 `稀疏张量 <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/tensor.html#sparse-tensor>`_。
+    - **enable_sparse** (bool) - 表示是否启用稀疏特征。默认值：False。有关稀疏特征和稀疏张量的详细信息，请查看 `稀疏张量 <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/tensor.html#sparse-tensor>`_。
     - **grad_for_scalar** (bool)：  表示是否获取标量梯度。默认值：False。当 `grad_for_scalar` 设置为True时，则衍生函数的标量输入。默认值为False。由于后端目前不支持伸缩操作，所以该接口只支持在前端可推演的简单操作。
     - **enable_compile_cache** (bool) - 表示是否加载或者保存前端编译的图。当 `enable_compile_cache` 被设置为True时，在第一次执行的过程中，一个硬件无关的编译缓存会被生成并且导出为一个MINDIR文件。当该网络被再次执行时，如果 `enable_compile_cache` 仍然为True并且网络脚本没有被更改，那么这个编译缓存会被加载。注意目前只支持有限的Python脚本更改的自动检测，这意味着可能有正确性风险。默认值：False。这是一个实验原型，可能会被更改或者删除。
     - **compile_cache_path** (str) - 保存前端图编译缓存的路径。默认值："."。如果目录不存在，系统会自动创建这个目录。缓存会被保存到如下目录：`compile_cache_path/rank_${rank_id}/` 。 `rank_id` 是集群上当前设备的ID。

docs/api/api_python/mindspore.dataset.rst

@@ -8,7 +8,7 @@ mindspore.dataset
 
 大多数数据集可以通过指定参数 `cache` 启用缓存服务，以提升整体数据处理效率。
 请注意Windows平台上还不支持缓存服务，因此在Windows上加载和处理数据时，请勿使用。更多介绍和限制，
-请参考 `Single-Node Tensor Cache <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/cache.html>`_。
+请参考 `Single-Node Tensor Cache <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/cache.html>`_。
 
 
 在API示例中，常用的模块导入方法如下：

docs/api/api_python/mindspore.train/mindspore.train.SummaryCollector.rst

@@ -3,7 +3,7 @@
     SummaryCollector可以收集一些常用信息。
 
     它可以帮助收集loss、学习率、计算图等。
-    SummaryCollector还可以允许通过 `summary算子 <https://www.mindspore.cn/mindinsight/docs/zh-CN/master/summary_record.html#summarysummarycollector>`_ 将数据收集到summary文件中。
+    SummaryCollector还可以允许通过 `summary算子 <https://www.mindspore.cn/mindinsight/docs/zh-CN/r1.6/summary_record.html#summarysummarycollector>`_ 将数据收集到summary文件中。
 
     .. note:: 
         - 不允许在回调列表中存在多个SummaryCollector实例。
@@ -20,7 +20,7 @@
 
       - **collect_metric** (bool) - 表示是否收集训练metrics，目前只收集loss。把第一个输出视为loss，并且算出其平均数。默认值：True。
       - **collect_graph** (bool) - 表示是否收集计算图。目前只收集训练计算图。默认值：True。
-      - **collect_train_lineage** (bool) - 表示是否收集训练阶段的lineage数据，该字段将显示在MindInsight的 `lineage页面 <https://www.mindspore.cn/mindinsight/docs/zh-CN/master/lineage_and_scalars_comparison.html>`_ 上。默认值：True。
+      - **collect_train_lineage** (bool) - 表示是否收集训练阶段的lineage数据，该字段将显示在MindInsight的 `lineage页面 <https://www.mindspore.cn/mindinsight/docs/zh-CN/r1.6/lineage_and_scalars_comparison.html>`_ 上。默认值：True。
       - **collect_eval_lineage** (bool) - 表示是否收集评估阶段的lineage数据，该字段将显示在MindInsight的lineage页面上。默认值：True。
       - **collect_input_data** (bool) - 表示是否为每次训练收集数据集。目前仅支持图像数据。如果数据集中有多列数据，则第一列应为图像数据。默认值：True。
       - **collect_dataset_graph** (bool) - 表示是否收集训练阶段的数据集图。默认值：True。

docs/api/api_python/mindspore.train/mindspore.train.summary.SummaryRecord.rst

@@ -4,7 +4,7 @@
 
     该方法将在一个指定的目录中创建summary文件和lineage文件，并将数据写入文件。
 
-    它通过执行 `record` 方法将数据写入文件。除了通过 `summary算子 <https://www.mindspore.cn/mindinsight/docs/zh-CN/master/summary_record.html#summarysummarycollector>`_ 记录网络的数据外，SummaryRecord还支持通过 `自定义回调函数和自定义训练循环 <https://www.mindspore.cn/mindinsight/docs/zh-CN/master/summary_record.html#callback>`_ 记录数据。
+    它通过执行 `record` 方法将数据写入文件。除了通过 `summary算子 <https://www.mindspore.cn/mindinsight/docs/zh-CN/r1.6/summary_record.html#summarysummarycollector>`_ 记录网络的数据外，SummaryRecord还支持通过 `自定义回调函数和自定义训练循环 <https://www.mindspore.cn/mindinsight/docs/zh-CN/r1.6/summary_record.html#callback>`_ 记录数据。
 
     .. note::
         - 确保在最后关闭SummaryRecord，否则进程不会退出。请参阅下面的示例部分，了解如何用两种方式正确关闭SummaryRecord。
@@ -144,7 +144,7 @@
 
         **异常：**
 
-        - **TypeError：** `step` 不为整型，或 `train_network` 的类型不为`mindspore.nn.Cell <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/nn/mindspore.nn.Cell.html?highlight=MindSpore.nn.cell#mindspore-nn-cell>`_ 。
+        - **TypeError：** `step` 不为整型，或 `train_network` 的类型不为`mindspore.nn.Cell <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/nn/mindspore.nn.Cell.html?highlight=MindSpore.nn.cell#mindspore-nn-cell>`_ 。
 
         **样例：**
 
@@ -157,7 +157,7 @@
 
     .. py:method:: set_mode(mode)
 
-        设置模型运行状态。不同的状态会影响记录数据的内容。
+        设置模型运行阶段。不同的阶段会影响记录数据的内容。
 
         **参数：**
 

docs/api/api_python/mindspore/mindspore.build_searched_strategy.rst

@@ -3,7 +3,7 @@ mindspore.build_searched_strategy
 
 .. py:class:: mindspore.build_searched_strategy(strategy_filename)
 
-    构建网络中每个参数的策略，用于分布式推理。关于它的使用细节，请参考： `保存和加载模型（HyBrid Parallel模式） <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/save_load_model_hybrid_parallel.html>`_。
+    构建网络中每个参数的策略，用于分布式推理。关于它的使用细节，请参考： `保存和加载模型（HyBrid Parallel模式） <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/save_load_model_hybrid_parallel.html>`_。
 
     **参数：**
 

docs/api/api_python/mindspore/mindspore.load_distributed_checkpoint.rst

@@ -3,7 +3,7 @@ mindspore.load_distributed_checkpoint
 
 .. py:method:: mindspore.load_distributed_checkpoint(network, checkpoint_filenames, predict_strategy=None, train_strategy_filename=None, strict_load=False, dec_key=None, dec_mode='AES-GCM')
 
-    给分布式预测加载checkpoint文件到网络，用于分布式推理。关于分布式推理的细节，请参考： https://www.mindspore.cn/docs/programming_guide/zh-CN/master/distributed_inference.html 。
+    给分布式预测加载checkpoint文件到网络，用于分布式推理。关于分布式推理的细节，请参考： https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/distributed_inference.html 。
 
     **参数：**
 

docs/api/api_python/mindspore/mindspore.merge_sliced_parameter.rst

@@ -3,7 +3,7 @@ mindspore.merge_sliced_parameter
 
 .. py:method:: mindspore.merge_sliced_parameter(sliced_parameters, strategy=None)
 
-    将参数切片合并为一个完整的参数，用于分布式推理。关于它的细节，请参考：`保存和加载模型（HyBrid Parallel模式） <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/save_load_model_hybrid_parallel.html>`_。
+    将参数切片合并为一个完整的参数，用于分布式推理。关于它的细节，请参考：`保存和加载模型（HyBrid Parallel模式） <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/save_load_model_hybrid_parallel.html>`_。
 
     **参数：**
 

docs/api/api_python/nn/mindspore.nn.Flatten.rst

@@ -7,7 +7,7 @@ mindspore.nn.Flatten
 
     **输入：**
 
-    - **x** (Tensor) - 要展平的输入Tensor。shape为 :math:`(N, *)`，其中 :math:`*` 表示任意的附加维度。数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_。
+    - **x** (Tensor) - 要展平的输入Tensor。shape为 :math:`(N, *)`，其中 :math:`*` 表示任意的附加维度。数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_。
 
     **输出：**
 

docs/api/api_python/nn/mindspore.nn.ReLU.rst

@@ -15,7 +15,7 @@ mindspore.nn.ReLU
 
     **输入：**
 
-    - **x** (Tensor) - 用于计算ReLU的任意维度的Tensor。数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_。
+    - **x** (Tensor) - 用于计算ReLU的任意维度的Tensor。数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_。
 
     **输出：**
 

docs/api/api_python/nn/mindspore.nn.Tril.rst

@@ -11,7 +11,7 @@ mindspore.nn.Tril
 
     **输入：**
 
-    - **x** (Tensor)：输入Tensor。数据类型为`number <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_。
+    - **x** (Tensor)：输入Tensor。数据类型为`number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_。
     - **k** (int)：对角线的索引。默认值：0。假设输入的矩阵的维度分别为d1，d2，则k的范围应在[-min(d1, d2)+1, min(d1, d2)-1]，超出该范围时输出值与输入 `x` 一致。
 
     **输出：**

docs/api/api_python/nn/mindspore.nn.optim_note_loss_scale.rst

@@ -1 +1 @@
-优化器和混合精度之间通常没有联系。但是，当使用 `FixedLossScaleManager` 且 `FixedLossScaleManager` 中的 `drop_overflow_update` 设置为False时，优化器需要设置'loss_scale'。由于此优化器没有 `loss_scale` 的参数，因此需要通过其他方式处理 `loss_scale` ，如何正确处理 `loss_scale` 详见 `LossScale <https://www.mindspore.cn/docs/programming_guide/zh-CN/master/lossscale.html>`_。
+优化器和混合精度之间通常没有联系。但是，当使用 `FixedLossScaleManager` 且 `FixedLossScaleManager` 中的 `drop_overflow_update` 设置为False时，优化器需要设置'loss_scale'。由于此优化器没有 `loss_scale` 的参数，因此需要通过其他方式处理 `loss_scale` ，如何正确处理 `loss_scale` 详见 `LossScale <https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/lossscale.html>`_。

docs/api/api_python/nn_probability/mindspore.nn.probability.bijector.GumbelCDF.rst

@@ -39,7 +39,7 @@ mindspore.nn.probability.bijector.GumbelCDF
     >>>
     >>> # 初始化GumbelCDF Bijector，loc设置为1.0和scale设置为2.0。
     >>> gumbel_cdf = msb.GumbelCDF(1.0, 2.0)
-    >>> # 在网络中使用ScalarAffinebijector。
+    >>> # 在网络中使用GumbelCDF bijector。
     >>> x = Tensor([1, 2, 3], dtype=mindspore.float32)
     >>> y = Tensor([0.1, 0.2, 0.3], dtype=mindspore.float32)
     >>> ans1 = gumbel_cdf.forward(x)

docs/api/api_python/nn_probability/mindspore.nn.probability.bijector.PowerTransform.rst

@@ -3,7 +3,7 @@ mindspore.nn.probability.bijector.PowerTransform
 
 .. py:class:: mindspore.nn.probability.bijector.PowerTransform(power=0., name='PowerTransform')
 
-    乘方Bijector（Power Bijector）。
+    乘方Bijector（PowerTransform Bijector）。
     此Bijector对应的映射函数为：
 
     .. math::
@@ -11,7 +11,7 @@ mindspore.nn.probability.bijector.PowerTransform
 
     其中幂c >= 0。
 
-    Power Bijector将输入从 `[-1/c, inf]` 映射到 `[0, inf]` 。
+    PowerTransform Bijector将输入从 `[-1/c, inf]` 映射到 `[0, inf]` 。
 
     当 `c=0` 时，此Bijector等于 :class:`mindspore.nn.probability.bijector.Exp` Bijector。
 

docs/api/api_python/nn_probability/mindspore.nn.probability.distribution.Gumbel.rst

@@ -15,7 +15,7 @@ mindspore.nn.probability.distribution.Gumbel
 
     - **loc** (int, float, list, numpy.ndarray, Tensor) - Gumbel分布的位置。
     - **scale** (int, float, list, numpy.ndarray, Tensor) - Gumbel分布的尺度。
-    - **seed** (int) - 采样时使用的种子。如果为None，则使用全局种子。默认值：None。
+    - **seed** (int) - 采样时使用的种子。如果为None，则使用全局种子。默认值：0。
     - **dtype** (mindspore.dtype) - 分布类型。默认值：mindspore.float32。
     - **name** (str) - 分布的名称。默认值：'Gumbel'。
 

docs/api/api_python/nn_probability/mindspore.nn.probability.distribution.LogNormal.rst

@@ -16,7 +16,7 @@ mindspore.nn.probability.distribution.LogNormal
 
     - **loc** (int, float, list, numpy.ndarray, Tensor) - 基础正态分布的平均值。默认值：None。
     - **scale** (int, float, list, numpy.ndarray, Tensor) - 基础正态分布的标准差。默认值：None。
-    - **seed** (int) - 采样时使用的种子。如果为None，则使用全局种子。默认值：None。
+    - **seed** (int) - 采样时使用的种子。如果为None，则使用全局种子。默认值：0。
     - **dtype** (mindspore.dtype) - 分布类型。默认值：mindspore.float32。
     - **name** (str) - 分布的名称。默认值：'LogNormal'。
 

docs/api/api_python/ops/mindspore.ops.Add.rst

@@ -1,31 +1,32 @@
 mindspore.ops.Add
 =================
 
-.. py:class:: mindspore.ops.Add(*args, **kwargs)
+.. py:class:: mindspore.ops.Add()
 
-    两个输入Tensor按元素相加。
-
-    输入 `x` 和 `y` 遵循隐式类型转换规则，使数据类型保持一致。
-    输入必须是两个Tensor，或一个Tensor和一个Scalar。
-    当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。
-    当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+    两个输入Tensor逐元素相加。
 
     .. math::
 
         out_{i} = x_{i} + y_{i}
 
+    .. note::
+        - 输入 `x` 和 `y` 遵循 `隐式类型转换规则 <https://www.mindspore.cn/docs/note/zh-CN/r1.6/operator_list_implicit.html>`_ ，使数据类型保持一致。
+        - 输入必须是两个Tensor，或一个Tensor和一个Scalar。
+        - 当输入是两个Tensor时，它们的数据类型不能同时是bool，并保证其shape可以广播。
+        - 当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+
     **输入：**
 
-    - **x** (Union[Tensor, Number, bool]) - 第一个输入，是一个Number、bool值或数据类型为Number或bool的Tensor。
-    - **y** (Union[Tensor, Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个Number或bool值，或数据类型为Number或bool的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 的Tensor。
+    - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**
 
-    Tensor，shape与广播后的shape相同，数据类型为两个输入中精度较高的类型。
+    Tensor，shape与输入 `x`，`y` 广播后的shape相同，数据类型为两个输入中精度较高的类型。
 
     **异常：**
 
-    - **TypeError** - `x` 和 `y` 不是Tensor、Number或bool。
+    - **TypeError** - `x` 和 `y` 不是Tensor、number.Number或bool。
 
     **支持平台：**
 

docs/api/api_python/ops/mindspore.ops.AddN.rst

@@ -1,15 +1,15 @@
 mindspore.ops.AddN
 ===================
 
-.. py:class:: mindspore.ops.AddN(*args, **kwargs)
+.. py:class:: mindspore.ops.AddN()
 
-    按元素将所有输入的Tensor相加。
+    逐元素将所有输入的Tensor相加。
 
     所有输入Tensor必须具有相同的shape。
 
     **输入：**
 
-    - **x** (Union(tuple[Tensor], list[Tensor])) - 输入tuple或list由多个Tensor组成，其数据类型为Number或bool，用于相加。
+    - **x** (Union(tuple[Tensor], list[Tensor])) - Tensor组成的tuble或list，类型为 `bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 。
 
     **输出：**
 
@@ -18,6 +18,7 @@ mindspore.ops.AddN
     **异常：**
 
     - **TypeError** - `x` 既不是tuple，也不是list。
+    - **ValueError** - `x` 中存在shape不同的Tensor。
 
     **支持平台：**
 

docs/api/api_python/ops/mindspore.ops.Div.rst

@@ -1,28 +1,33 @@
 mindspore.ops.Div
 =================
 
-.. py:class:: mindspore.ops.Div(*args, **kwargs)
+.. py:class:: mindspore.ops.Div()
 
-    按元素计算第一输入Tensor除以第二输入Tensor的商。
-
-    输入 `x` 和 `y` 遵循隐式类型转换规则，使数据类型保持一致。输入必须是两个Tensor，或一个Tensor和一个Scalar。当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+    逐元素计算第一输入Tensor除以第二输入Tensor的商。
 
     .. math::
 
         out_{i} = \frac{x_i}{y_i}
 
+    .. note::
+        - 输入 `x` 和 `y` 遵循 `隐式类型转换规则 <https://www.mindspore.cn/docs/note/zh-CN/r1.6/operator_list_implicit.html>`_ ，使数据类型保持一致。
+        - 输入必须是两个Tensor，或一个Tensor和一个Scalar。
+        - 当输入是两个Tensor时，它们的数据类型不能同时是bool，并保证其shape可以广播。
+        - 当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+
     **输入：**
 
-    - **x** (Union[Tensor, Number, bool]) - 第一个输入，是一个Number、bool值或数据类型为Number或bool的Tensor。
-    - **y** (Union[Tensor, Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个Number或bool值，或数据类型为Number或bool的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 的Tensor。
+    - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**
 
-    Tensor，shape与广播后的shape相同，数据类型为两个输入中精度较高的类型。
+    Tensor，shape与输入 `x`，`y` 广播后的shape相同，数据类型为两个输入中精度较高的类型。
 
     **异常：**
 
     - **TypeError** - `x` 和 `y` 都不是Tensor。
+    - **TypeError** - `x` 和 `y` 数据类型都是bool_的Tensor。
 
     **支持平台：**
 
@@ -38,8 +43,8 @@ mindspore.ops.Div
     >>> print(output)
     [-1.3333334  2.5        2.        ]
     >>> # 用例2：两个输入的数据类型和shape不同
-    >>> x = Tensor(np.array([-4.0, 5.0, 6.0]), mindspore.int32)
-    >>> y = Tensor(2, mindspore.float32)
+    >>> x = Tensor(np.array([-4.0, 5.0, 6.0]), mindspore.float32)
+    >>> y = Tensor(2, mindspore.int32)
     >>> output = div(x, y)
     >>> print(output)
     [-2.  2.5  3.]

docs/api/api_python/ops/mindspore.ops.Eps.rst

@@ -1,18 +1,23 @@
 mindspore.ops.Eps
 =================
 
-.. py:class:: mindspore.ops.Eps(*args, **kwargs)
+.. py:class:: mindspore.ops.Eps()
 
-    创建一个填充 `x` 数据类型最小值的Tensor。
+    创建一个与输入数据类型和shape都相同的Tensor，元素值为对应数据类型能表达的最小值。
 
     **输入：**
 
-    - **x** (Tensor) - 用于获取其数据类型最小值的Tensor。数据类型必须为float16或float32。shape为 :math:`(N,*)` ，其中 :math:`*` 表示任意的附加维度数。
+    - **x** (Tensor) - 用于获取其数据类型能表达的最小值的任意维度的Tensor。数据类型必须为float16或float32。
 
     **输出：**
 
     Tensor，具有与 `x` 相同的数据类型和shape，填充了 `x` 数据类型的最小值。
 
+    **异常：**
+
+    - **TypeError** - `x` 不是Tensor。
+    - **TypeError** - `x` 的数据类型不是float16或者float32。
+
     **支持平台：**
 
     ``Ascend`` ``GPU`` ``CPU``

docs/api/api_python/ops/mindspore.ops.Erf.rst

@@ -1,9 +1,9 @@
 mindspore.ops.Erf
 =================
 
-.. py:class:: mindspore.ops.Erf(*args, **kwargs)
+.. py:class:: mindspore.ops.Erf()
 
-    按元素计算 `x` 的高斯误差函数。
+    逐元素计算 `x` 的高斯误差函数。
 
     .. math::
 
@@ -11,7 +11,7 @@ mindspore.ops.Erf
 
     **输入：**
 
-    - **x** (Tensor) - 用于计算高斯误差函数的Tensor。数据类型必须为float16或float32。shape为 :math:`(N,*)` ，其中 :math:`*` 表示任意的附加维度数，其秩应小于8。
+    - **x** (Tensor) - 高斯误差函数的输入Tensor。数据类型必须为float16或float32。任意维度小于8的Tensor。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Eye.rst

@@ -9,7 +9,7 @@ mindspore.ops.Eye
 
     - **n** (int) - 指定返回Tensor的行数。仅支持常量值。
     - **m** (int) - 指定返回Tensor的列数。仅支持常量值。
-    - **t** (mindspore.dtype) - 指定返回Tensor的数据类型。数据类型必须是`bool_ <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_或`number <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_。
+    - **t** (mindspore.dtype) - 指定返回Tensor的数据类型。数据类型必须是`bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_或`number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Fill.rst

@@ -7,7 +7,7 @@ mindspore.ops.Fill
 
     **输入：**
 
-    - **type** (mindspore.dtype) - 指定输出Tensor的数据类型。数据类型只支持`bool_ <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_和`number <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_。
+    - **type** (mindspore.dtype) - 指定输出Tensor的数据类型。数据类型只支持`bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_和`number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_。
     - **shape** (tuple[int]) - 指定输出Tensor的shape。
     - **value** (Union(number.Number, bool)) - 用来填充输出Tensor的值。
 

docs/api/api_python/ops/mindspore.ops.Gamma.rst

@@ -3,7 +3,7 @@ mindspore.ops.Gamma
 
 .. py:class:: mindspore.ops.Gamma(seed=0, seed2=0)
 
-    根据概率密度函数分布生成随机正浮点数x。
+    根据概率密度函数分布生成随机正值浮点数x。
 
     .. math::
 
@@ -14,6 +14,13 @@ mindspore.ops.Gamma
     - **seed** (int) - 算子层的随机种子，用于生成随机数。必须是非负的。默认值：0。
     - **seed2** (int)：全局的随机种子，和算子层的随机种子共同决定最终生成的随机数。必须是非负的。默认值：0。
 
+    .. note::
+        - 随机种子：通过一些复杂的数学算法，可以得到一组有规律的随机数，而随机种子就是这个随机数的初始值。随机种子相同，得到的随机数就不会改变。
+        - 全局的随机种子和算子层的随机种子都没设置：使用默认值当做随机种子。
+        - 全局的随机种子设置了，算子层的随机种子未设置：随机生成一个种子和全局的随机种子拼接。
+        - 全局的随机种子未设置，算子层的随机种子设置了：使用默认的全局的随机种子，和算子层的随机种子拼接。
+        - 全局的随机种子和算子层的随机种子都设置了：全局的随机种子和算子层的随机种子拼接。
+
     **输入：**
 
     - **shape** (tuple) - 待生成的随机Tensor的shape。只支持常量值。
@@ -22,12 +29,13 @@ mindspore.ops.Gamma
 
     **输出：**
 
-    Tensor。shape是输入 `shape` 以及alpha、beta广播后的shape。数据类型为float32。
+    Tensor。shape是输入 `shape`， `alpha`， `beta` 广播后的shape。数据类型为float32。
 
     **异常：**
 
-    - **TypeError** - `seed` 和 `seed2` 都不是int。
-    - **TypeError** - `alpha` 和 `beta` 都不是Tensor。
+    - **TypeError** - `seed` 或 `seed2` 的数据类型不是int。
+    - **TypeError** - `alpha` 或 `beta` 不是Tensor。
+    - **TypeError** - `alpha` 或 `beta` 的数据类型不是float32。
     - **ValueError** - `shape` 不是常量值。
 
     **支持平台：**

docs/api/api_python/ops/mindspore.ops.GeLU.rst

@@ -1,7 +1,7 @@
 mindspore.ops.GeLU
 ==================
 
-.. py:class:: mindspore.ops.GeLU(*args, **kwargs)
+.. py:class:: mindspore.ops.GeLU()
 
     高斯误差线性单元激活函数（Gaussian Error Linear Units activation function）。
 
@@ -11,13 +11,13 @@ mindspore.ops.GeLU
     GeLU函数定义如下：
 
     .. math::
-        \text{output} = 0.5 * x * (1 + tanh(x / \sqrt{2})),
+        GELU(x_i) = x_i*P(X < x_i)
 
-    其中 :math:`tanh` 是双曲正切函数。
+    其中  :math:`P` 是标准高斯分布的累积分布函数， :math:`x_i` 是输入的元素。
 
     **输入：**
 
-    - **x** (Tensor) - 用于计算GeLU函数的Tensor，数据类型为float16或float32。
+    - **x** (Tensor) - 激活函数GeLU的输入，数据类型为float16或float32。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Greater.rst

@@ -21,7 +21,7 @@ mindspore.ops.Greater
 
     **输入：**
 
-    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为`number <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_或`bool_ <https://www.mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.html#mindspore.dtype>`_的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为`number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_或`bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_的Tensor。
     - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**

docs/api/api_python/ops/mindspore.ops.L2Loss.rst

@@ -1,14 +1,14 @@
 mindspore.ops.L2Loss
 ====================
 
-.. py:class:: mindspore.ops.L2Loss(*args, **kwargs)
+.. py:class:: mindspore.ops.L2Loss()
 
-    计算Tensor的L2范数的一半，不对结果进行开方。
+    用于计算L2范数，但不对结果进行开方操作。
 
-    把 `input_x` 设为x，输出设为loss。
+    把输入设为x，输出设为loss。
 
     .. math::
-        loss = sum(x ** 2) / 2
+        loss = \frac{\sum x ^ 2}{2}
 
     **输入：**
 
@@ -16,7 +16,7 @@ mindspore.ops.L2Loss
 
     **输出：**
 
-    Tensor，具有与 `input_x` 相同的数据类型。输出Tensor是loss的值，是一个scalar Tensor。
+    Tensor，具有与 `input_x` 相同的数据类型的Scalar Tensor。
 
     **异常：**
 

docs/api/api_python/ops/mindspore.ops.LessEqual.rst

@@ -1,14 +1,9 @@
 mindspore.ops.LessEqual
 ========================
 
-.. py:class:: mindspore.ops.LessEqual(*args, **kwargs)
+.. py:class:: mindspore.ops.LessEqual()
 
-    按元素计算 :math:`x <= y` 的bool值。
-
-    输入 `x` 和 `y` 遵循隐式类型转换规则，使数据类型保持一致。
-    输入必须是两个Tensor，或一个Tensor和一个Scalar。
-    当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。
-    当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+    逐元素计算 :math:`x <= y` 的bool值。
 
     .. math::
 
@@ -17,10 +12,16 @@ mindspore.ops.LessEqual
             & \text{False,   if } x_{i}>y_{i}
             \end{cases}
 
+    .. note::
+        - 输入 `x` 和 `y` 遵循 `隐式类型转换规则 <https://www.mindspore.cn/docs/note/zh-CN/r1.6/operator_list_implicit.html>`_ ，使数据类型保持一致。
+        - 输入必须是两个Tensor，或一个Tensor和一个Scalar。
+        - 当输入是两个Tensor时，它们的数据类型不能同时是bool，并保证其shape可以广播。
+        - 当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+
     **输入：**
 
-    - **x** (Union[Tensor, Number, bool]) - 第一个输入，是一个Number、bool值或数据类型为Number或bool的Tensor。
-    - **y** (Union[Tensor, Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个Number或bool值，或数据类型为Number或bool的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 的Tensor。
+    - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Log.rst

@@ -1,20 +1,23 @@
 mindspore.ops.Log
 =================
 
-.. py:class:: mindspore.ops.Log(*args, **kwargs)
+.. py:class:: mindspore.ops.Log()
 
-    按元素返回Tensor的自然对数。
+    逐元素返回Tensor的自然对数。
 
     .. math::
         y_i = log_e(x_i)
 
     .. warning::
 
-        如果算子Log的输入值在(0，0.01]或[0.95，1.05]范围内，则输出精度可能会发生变化。
+        如果算子Log的输入值在(0，0.01]或[0.95，1.05]范围内，则输出精度可能会存在误差。
+
+    .. note::
+        Ascend上输入Tensor的维度要小于等于8，CPU上输入Tensor的维度要小于8。
 
     **输入：**
 
-    - **x** (Tensor) - 输入Tensor。该值必须大于0。shape为 :math:`(N,*)`，其中 :math:`*` 表示任意的附加维度数，它的秩应小于8。
+    - **x** (Tensor) - 任意维度的输入Tensor。该值必须大于0。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.MatMul.rst

@@ -28,7 +28,7 @@ mindspore.ops.matmul
     **异常：**
 
     - **TypeError** - `transpose_a` 或 `transpose_b` 不是bool。
-    - **ValueError** - 矩阵 `a` 的维度的列不等于矩阵 `b` 的维度的行。
+    - **ValueError** - 矩阵 `a` 的列不等于矩阵 `b` 的行。
     - **ValueError** - `a` 或 `b` 的维度不等于2。
 
     **支持平台：**

docs/api/api_python/ops/mindspore.ops.Mul.rst

@@ -1,23 +1,24 @@
 mindspore.ops.Mul
 =================
 
-.. py:class:: mindspore.ops.Mul(*args, **kwargs)
+.. py:class:: mindspore.ops.Mul()
 
-    两个Tensor按元素相乘。
-
-    输入 `x` 和 `y` 遵循隐式类型转换规则，使数据类型保持一致。
-    输入必须是两个Tensor，或一个Tensor和一个Scalar。
-    当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。
-    当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+    两个Tensor逐元素相乘。
 
     .. math::
 
         out_{i} = x_{i} * y_{i}
 
+    .. note::
+        - 输入 `x` 和 `y` 遵循 `隐式类型转换规则 <https://www.mindspore.cn/docs/note/zh-CN/r1.6/operator_list_implicit.html>`_ ，使数据类型保持一致。
+        - 输入必须是两个Tensor，或一个Tensor和一个Scalar。
+        - 当输入是两个Tensor时，它们的数据类型不能同时是bool，并保证其shape可以广播。
+        - 当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+
     **输入：**
 
-    - **x** (Union[Tensor, Number, bool]) - 第一个输入，是一个Number、bool值或数据类型为Number或bool的Tensor。
-    - **y** (Union[Tensor, Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个Number或bool值，或数据类型为Number或bool的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 的Tensor。
+    - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**
 
@@ -25,7 +26,7 @@ mindspore.ops.Mul
 
     **异常：**
 
-    - **TypeError** - `x` 和 `y` 不是Tensor、Number或bool。
+    - **TypeError** - `x` 和 `y` 不是Tensor、number.Number或bool。
     - **ValueError** - `x` 和 `y` 的shape不相同。
 
     **支持平台：**

docs/api/api_python/ops/mindspore.ops.OnesLike.rst

@@ -1,15 +1,13 @@
 mindspore.ops.OnesLike
 ======================
 
-.. py:class:: mindspore.ops.OnesLike(*args, **kwargs)
+.. py:class:: mindspore.ops.OnesLike()
 
-    创建新Tensor。所有元素的值都为1。
-
-    返回填充了Scalar值为1的具有与输入相同shape和数据类型的Tensor。
+    返回值为1的Tensor，shape和数据类型与输入相同。
 
     **输入：**
 
-    - **input_x** (Tensor) - 输入Tensor。shape为 :math:`(N,*)` ，其中 :math:`*` 表示任意的附加维度数。
+    - **input_x** (Tensor) - 任意维度的Tensor。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.PReLU.rst

@@ -1,7 +1,7 @@
 mindspore.ops.PReLU
 ===================
 
-.. py:class:: mindspore.ops.PReLU(*args, **kwargs)
+.. py:class:: mindspore.ops.PReLU()
 
     带参数的线性修正单元激活函数（Parametric Rectified Linear Unit activation function）。
 
@@ -14,18 +14,18 @@ mindspore.ops.PReLU
 
     .. note::
 
-        Ascend不支持0-D或1-D的x。
+        Ascend不支持标量和1维向量的输入x。
 
     **输入：**
 
-    - **x** (Tensor) - 用于计算激活函数的Tensor。数据类型为float16或float32。shape为 :math:`(N, C, *)` ，其中 :math:`*` 表示任意的附加维度数。
-    - **weight** (Tensor) - 权重Tensor。数据类型为float16或float32。只有两种shape是合法的，1或 `input_x` 的通道数。通道维度是输入的第二维。当输入为0-D或1-D Tensor时，通道数为1。
+    - **x** (Tensor) - 激活函数的输入Tensor。数据类型为float16或float32。shape为 :math:`(N, C, *)` ，其中 :math:`*` 表示任意的附加维度。
+    - **weight** (Tensor) - 权重Tensor。数据类型为float16或float32。weight只可以是向量，长度与输入x的通道数C相同。在GPU设备上，当输入为标量时，shape为1。
 
     **输出：**
 
     Tensor，数据类型与 `x` 的相同。
 
-    有关详细信息，请参考:class:`nn.PReLU`。
+    有关详细信息，请参考 :class:`mindspore.nn.PReLU` 。
 
     **异常：**
 

docs/api/api_python/ops/mindspore.ops.Pow.rst

@@ -1,23 +1,24 @@
 mindspore.ops.Pow
 ==================
 
-.. py:class:: mindspore.ops.Pow(*args, **kwargs)
+.. py:class:: mindspore.ops.Pow()
 
-    计算 `x` 中每个元素的 `y` 的幂次。
-
-    输入 `x` 和 `y` 遵循隐式类型转换规则，使数据类型保持一致。
-    输入必须是两个Tensor，或一个Tensor和一个Scalar。
-    当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。
-    当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+    计算 `x` 中每个元素的 `y` 次幂。
 
     .. math::
 
         out_{i} = x_{i} ^{ y_{i}}
 
+    .. note::
+        - 输入 `x` 和 `y` 遵循 `隐式类型转换规则 <https://www.mindspore.cn/docs/note/zh-CN/r1.6/operator_list_implicit.html>`_ ，使数据类型保持一致。
+        - 输入必须是两个Tensor，或一个Tensor和一个Scalar。
+        - 当输入是两个Tensor时，它们的数据类型不能同时是bool，并保证其shape可以广播。
+        - 当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+
     **输入：**
 
-    - **x** (Union[Tensor, Number, bool]) - 第一个输入，是一个Number、bool值或数据类型为Number或bool的Tensor。
-    - **y** (Union[Tensor, Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个Number或bool值，或数据类型为Number或bool的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为 `number <https://www.mindspore.cn/docs/api/en/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `bool_ <https://www.mindspore.cn/docs/api/en/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 的Tensor。
+    - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**
     
@@ -25,8 +26,8 @@ mindspore.ops.Pow
 
     **异常：**
 
-    - **TypeError** - `x` 和 `y` 不是Tensor、Number或bool。
-    - **ValueError** - `x` 和 `y` 的shape不相同。
+    - **TypeError** - `x` 和 `y` 不是Tensor、number.Number或bool。
+    - **ValueError** - 当`x` 和 `y` 都为Tensor时，它们的shape不相同。
 
     **支持平台：**
 

docs/api/api_python/ops/mindspore.ops.ReLUV2.rst

@@ -1,7 +1,7 @@
 mindspore.ops.ReLUV2
 ====================
 
-.. py:class:: mindspore.ops.ReLUV2(*args, **kwargs)
+.. py:class:: mindspore.ops.ReLUV2()
 
     线性修正单元激活函数（Rectified Linear Unit activation function）。
 
@@ -11,10 +11,6 @@ mindspore.ops.ReLUV2
 
         \text{ReLU}(x) = (x)^+ = \max(0, x)，
 
-    .. note::
-
-        与 `ReLu` 的区别在于该算子多输出一个mask，且算子的kernel与 `ReLu` 的不同。
-
     **输入：**
 
     - **input_x** (Tensor) - 输入Tensor必须是4-D Tensor。
@@ -22,7 +18,7 @@ mindspore.ops.ReLUV2
     **输出：**
 
     - **output** (Tensor) - 数据类型和shape与 `input_x` 的相同。
-    - **mask** (Tensor) - 数据类型必须为uint8的Tensor。
+    - **mask** (Tensor) - 保留输出，无实际意义。
 
     **异常：**
 
@@ -37,14 +33,9 @@ mindspore.ops.ReLUV2
 
     >>> input_x = Tensor(np.array([[[[1, -2], [-3, 4]], [[-5, 6], [7, -8]]]]), mindspore.float32)
     >>> relu_v2 = ops.ReLUV2()
-    >>> output, mask= relu_v2(input_x)
+    >>> output, _= relu_v2(input_x)
     >>> print(output)
     [[[[1. 0.]
        [0. 4.]]
        [[0. 6.]
-       [7. 0.]]]]
-    >>> print(mask)
-    [[[[[1 0]
-        [2 0]]
-        [[2 0]
-        [1 0]]]]]
+       [7. 0.]]]]

docs/api/api_python/ops/mindspore.ops.Reshape.rst

@@ -1,9 +1,9 @@
 mindspore.ops.Reshape
 ======================
 
-.. py:class:: mindspore.ops.Reshape(*args, **kwargs)
+.. py:class:: mindspore.ops.Reshape()
 
-    基于给定的shape，使用相同的值对输入Tensor进行reshape操作。
+    基于给定的shape，对输入Tensor进行重新排列。
 
     `input_shape` 最多只能有一个-1，在这种情况下，它可以从剩余的维度和输入的元素个数中推断出来。
 
@@ -18,7 +18,7 @@ mindspore.ops.Reshape
 
     **异常：**
 
-    - **ValueError** - 给定的 `input_shape`，如果它有几个-1，或者除-1（若存在）之外的元素的乘积小于或等于0，或者无法被输入Tensor的shape的乘积相除，或者与输入的数组大小不匹配。
+    - **ValueError** - 给定的 `input_shape`，如果它有多个-1，或者除-1（若存在）之外的元素的乘积小于或等于0，或者无法被输入Tensor的shape的乘积整除，或者与输入的数组大小不匹配。
 
     **支持平台：**
 

docs/api/api_python/ops/mindspore.ops.SeLU.rst

@@ -1,9 +1,9 @@
 mindspore.ops.SeLU
 ==================
 
-.. py:class:: mindspore.ops.SeLU(*args, **kwargs)
+.. py:class:: mindspore.ops.SeLU()
 
-    按元素计算输入Tensor的SeLU（scaled exponential Linear Unit）函数。
+    激活函数SeLU（Scaled exponential Linear Unit）。
 
     该激活函数定义为：
 
@@ -21,7 +21,7 @@ mindspore.ops.SeLU
 
     **输入：**
 
-    - **input_x** (Tensor) - shape为 :math:`(N, *)` 的Tensor，其中， :math:`*` 表示任意的附加维度数，数据类型为float16或float32。
+    - **input_x** (Tensor) - 任意维度的Tensor，数据类型为float16或float32。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Sigmoid.rst

@@ -1,11 +1,11 @@
 mindspore.ops.Sigmoid
 =====================
 
-.. py:class:: mindspore.ops.Sigmoid(*args, **kwargs)
+.. py:class:: mindspore.ops.Sigmoid()
 
     Sigmoid激活函数。
 
-    按元素计算输入的Sigmoid函数。Sigmoid函数定义为：
+    逐元素计算Sgmoid激活函数。Sigmoid函数定义为：
 
     .. math::
 
@@ -15,7 +15,7 @@ mindspore.ops.Sigmoid
 
     **输入：**
 
-    - **input_x** (Tensor) - shape为 :math:`(N, *)` 的tensor，其中， :math:`*` 表示任意的附加维度数，数据类型为float16或float32。
+    - **input_x** (Tensor) - 任意维度的Tensor，数据类型为float16或float32。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Size.rst

@@ -1,15 +1,13 @@
 mindspore.ops.Size
 ==================
 
-.. py:class:: mindspore.ops.Size(*args, **kwargs)
+.. py:class:: mindspore.ops.Size()
 
-    返回Tensor的大小。
-
-    返回一个整数Scalar，表示输入的元素大小，即Tensor中元素的总数。
+    返回一个Scalar，类型为整数，表示输入Tensor的大小，即Tensor中元素的总数。
 
     **输入：**
 
-    - **input_x** (Tensor) - Tensor的shape为 :math:`(x_1, x_2, ..., x_R)` 。数据类型为Number。
+    - **input_x** (Tensor) - 输入参数，shape为 :math:`(x_1, x_2, ..., x_R)` 。数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 。
 
     **输出：**
 

docs/api/api_python/ops/mindspore.ops.Sub.rst

@@ -1,23 +1,24 @@
 mindspore.ops.Sub
 =================
 
-.. py:class:: mindspore.ops.Sub(*args, **kwargs)
+.. py:class:: mindspore.ops.Sub()
 
-    按元素用第一个输入Tensor减去第二个输入Tensor。
-
-    输入 `x` 和 `y` 遵循隐式类型转换规则，使数据类型保持一致。
-    输入必须是两个Tensor，或一个Tensor和一个Scalar。
-    当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。
-    当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+    逐元素用第一个输入Tensor减去第二个输入Tensor。
 
     .. math::
 
         out_{i} = x_{i} - y_{i}
 
+     .. note::
+        - 输入 `x` 和 `y` 遵循 `隐式类型转换规则 <https://www.mindspore.cn/docs/note/zh-CN/r1.6/operator_list_implicit.html>`_ ，使数据类型保持一致。
+        - 输入必须是两个Tensor，或一个Tensor和一个Scalar。
+        - 当输入是两个Tensor时，它们的数据类型不能同时是bool，它们的shape可以广播。
+        - 当输入是一个Tensor和一个Scalar时，Scalar只能是一个常数。
+
     **输入：**
 
-    - **x** (Union[Tensor, Number, bool]) - 第一个输入，是一个Number、bool值或数据类型为Number或bool的Tensor。
-    - **y** (Union[Tensor, Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个Number或bool值，或数据类型为Number或bool的Tensor。
+    - **x** (Union[Tensor, number.Number, bool]) - 第一个输入，是一个number.Number、bool值或数据类型为 `number <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 或 `bool_ <https://www.mindspore.cn/docs/api/zh-CN/r1.6/api_python/mindspore.html#mindspore.dtype>`_ 的Tensor。
+    - **y** (Union[Tensor, number.Number, bool]) - 第二个输入，当第一个输入是Tensor时，第二个输入应该是一个number.Number或bool值，或数据类型为number或bool_的Tensor。当第一个输入是Scalar时，第二个输入必须是数据类型为number或bool_的Tensor。
 
     **输出：**
 
@@ -25,7 +26,7 @@ mindspore.ops.Sub
 
     **异常：**
 
-    - **TypeError** - `x` 和 `y` 不是Tensor、Number或bool。
+    - **TypeError** - `x` 和 `y` 不是Tensor、number.Number或bool。
 
     **支持平台：**
 

docs/api/api_python/ops/mindspore.ops.Tile.rst

@@ -1,9 +1,9 @@
 mindspore.ops.Tile
 ===================
 
-.. py:class:: mindspore.ops.Tile(*args, **kwargs)
+.. py:class:: mindspore.ops.Tile()
 
-    按照给定的次数复制Tensor。
+    按照给定的次数复制输入Tensor。
 
     通过复制 `multiples` 次 `input_x` 来创建新的Tensor。输出Tensor的第i维度有 `input_x.shape[i] * multiples[i]` 个元素，并且 `input_x` 的值沿第i维度被复制 `multiples[i]` 次。
 
@@ -13,12 +13,12 @@ mindspore.ops.Tile
 
     **输入：**
 
-    - **input_x** (Tensor) - 1-D或更高的Tensor。将输入Tensor的shape设置为 :math:`(x_1, x_2, ..., x_S)` 。
-    - **multiples** (tuple[int]) - 输入tuple由多个整数构成，如 :math:`(y_1, y_2, ..., y_S)` 。`multiples` 的长度不能小于 `input_x` 的维度。只支持常量值。
+    - **input_x** (Tensor) - 1-D或更高维的Tensor。
+    - **multiples** (tuple[int]) - 指定复制次数的参数，参数类型为tuple，数据类型为整数。如 :math:`(y_1, y_2, ..., y_S)` 。`multiples` 的长度不能小于 `input_x` 的维度。只支持常量值。
 
     **输出：**
 
-    Tensor，具有与 `input_x` 相同的数据类型。假设 `multiples` 的长度为 `d` ，`input_x` 的维度为 `input_x.dim`。
+    Tensor，具有与 `input_x` 相同的数据类型。假设 `multiples` 的长度为 `d` ，`input_x` 的维度为 `input_x.dim`，`input_x`的shape为 :math:`(x_1, x_2, ..., x_S)` 。
 
     - 如果 `input_x.dim = d`， 将其相应位置的shape相乘，输出的shape为 :math:`(x_1*y_1, x_2*y_2, ..., x_S*y_S)` 。
     - 如果 `input_x.dim < d`， 在 `input_x` 的shape的前面填充1，直到它们的长度一致。例如将 `input_x` 的shape设置为 :math:`(1, ..., x_1, ..., x_R, x_S)` ，然后可以将其相应位置的shape相乘，输出的shape为 :math:`(1*y_1, ..., x_R*y_R, x_S*y_S)` 。

docs/api/api_python/ops/mindspore.ops.UniformReal.rst

@@ -3,24 +3,31 @@ mindspore.ops.UniformReal
 
 .. py:class:: mindspore.ops.UniformReal(seed=0, seed2=0)
 
-    产生随机的浮点数i，均匀分布在[0，1)范围内。
+    产生随机的浮点数，均匀分布在[0，1)范围内。
 
     **参数：**
 
     - **seed** (int) - 算子层的随机种子，用于生成随机数。必须是非负的。默认值：0。
     - **seed2** (int)：全局的随机种子，和算子层的随机种子共同决定最终生成的随机数。必须是非负的。默认值：0。
 
+    .. note::
+        - 随机种子：通过一些复杂的数学算法，可以得到一组有规律的随机数，而随机种子就是这个随机数的初始值。随机种子相同，得到的随机数就不会改变。
+        - 全局的随机种子和算子层的随机种子都没设置：使用默认值当做随机种子。
+        - 全局的随机种子设置了，算子层的随机种子未设置：随机生成一个种子和全局的随机种子拼接。
+        - 全局的随机种子未设置，算子层的随机种子设置了：使用默认的全局的随机种子，和算子层的随机种子拼接。
+        - 全局的随机种子和算子层的随机种子都设置了：全局的随机种子和算子层的随机种子拼接。
+
     **输入：**
 
-    - **shape** (tuple) - 待生成的随机Tensor的shape。只支持常量值。
+    - **shape** (tuple) - 待生成的Tensor的shape。只支持常量值。
 
     **输出：**
 
-    Tensor。它的shape为输入 `shape` 表示的值。数据类型为float32。
+    Tensor。它的shape为输入 `shape`。数据类型为float32。
 
     **异常：**
 
-    - **TypeError** - `seed` 和 `seed2` 都不是int。
+    - **TypeError** - `seed` 或 `seed2` 不是int。
     - **TypeError** - `shape` 不是tuple。
     - **ValueError** - `shape` 不是常量值。
 

docs/api/api_python/ops/mindspore.ops.ZerosLike.rst

@@ -1,15 +1,13 @@
 mindspore.ops.ZerosLike
 =======================
 
-.. py:class:: mindspore.ops.ZerosLike(*args, **kwargs)
+.. py:class:: mindspore.ops.ZerosLike()
 
-    创建新的Tensor。它的所有元素的值都为0。
-
-    返回具有与输入Tensor相同shape和数据类型的值为0的Tensor。
+    返回值为0的Tensor，其shape和数据类型与输入Tensor相同。
 
     **输入：**
 
-    - **input_x** (Tensor) - 输入Tensor。数据类型为int32、int64、float16或float32。shape为 :math:`(N,*)` ，其中 :math:`*` 表示任意的附加维度数。
+    - **input_x** (Tensor) - 任意维度的输入Tensor。数据类型为int32、int64、float16或float32。
 
     **输出：**
 

graphengine

@@ -1 +1 @@
-Subproject commit 2158c0a9b8c53d89c5be65046b41c9ac058b916d
+Subproject commit 008fafbb15d6e1126b424dbb8689a4795578b5b6

include/api/context.h

@@ -99,6 +99,16 @@ class MS_API Context {
   /// \return Pointer to the custom delegate.
   std::shared_ptr<Delegate> GetDelegate() const;
 
+  /// \brief Set quant model to run as float model in multi device.
+  ///
+  /// \param[in] float_mode: true, run as float model; false, not run as float model.
+  void SetMultiModalHW(bool float_mode);
+
+  /// \brief Get the mode of the model run.
+  ///
+  /// \return Bool value that indicates whether run as float model
+  bool GetMultiModalHW() const;
+
   /// \brief Get a mutable reference of DeviceInfoContext vector in this context. Only MindSpore Lite supports
   /// heterogeneous scenarios with multiple members in the vector.
   ///

include/api/kernel.h

@@ -139,7 +139,6 @@ class MS_API Kernel {
   /// \param[in] value define the kernel's attribute value.
   void SetAttr(const std::string &key, const std::string &value) { attrs_[key] = value; }
 
- protected:
   std::string name_;
   const mindspore::Context *context_ = nullptr;
   std::vector<mindspore::MSTensor> inputs_;

include/api/model_parallel_runner.h

@@ -0,0 +1,71 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_INCLUDE_API_MODEL_PARALLEL_RUNNER_H
+#define MINDSPORE_INCLUDE_API_MODEL_PARALLEL_RUNNER_H
+#include <vector>
+#include <memory>
+#include <utility>
+#include <string>
+#include "include/api/status.h"
+#include "include/api/context.h"
+namespace mindspore {
+struct RunnerConfig {
+  std::shared_ptr<Context> context = nullptr;
+  int workers_num = 0;
+};
+class ModelPool;
+
+/// \brief The ModelParallelRunner class is used to define a MindSpore ModelParallelRunner, facilitating Model
+/// management.
+class MS_API ModelParallelRunner {
+ public:
+  ModelParallelRunner() = default;
+  ~ModelParallelRunner() = default;
+
+  /// \brief build a model parallel runner from model path so that it can run on a device. Only valid for Lite.
+  ///
+  /// \param[in] model_path Define the model path.
+  /// \param[in] runner_config Define the config used to store options during model pool init.
+  ///
+  /// \return Status.
+  Status Init(const std::string &model_path, const std::shared_ptr<RunnerConfig> &runner_config = nullptr);
+
+  /// \brief Obtains all input tensors information of the model.
+  ///
+  /// \return The vector that includes all input tensors.
+  std::vector<MSTensor> GetInputs();
+
+  /// \brief Obtains all output tensors information of the model.
+  ///
+  /// \return The vector that includes all output tensors.
+  std::vector<MSTensor> GetOutputs();
+
+  /// \brief Inference ModelParallelRunner.
+  ///
+  /// \param[in] inputs A vector where model inputs are arranged in sequence.
+  /// \param[out] outputs Which is a pointer to a vector. The model outputs are filled in the container in sequence.
+  /// \param[in] before CallBack before predict.
+  /// \param[in] after CallBack after predict.
+  ///
+  /// \return Status.
+  Status Predict(const std::vector<MSTensor> &inputs, std::vector<MSTensor> *outputs,
+                 const MSKernelCallBack &before = nullptr, const MSKernelCallBack &after = nullptr);
+
+ private:
+  std::shared_ptr<ModelPool> model_pool_ = nullptr;
+};
+}  // namespace mindspore
+#endif  // MINDSPORE_INCLUDE_API_MODEL_PARALLEL_RUNNER_H

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_kernel_build.cc

@@ -108,7 +108,7 @@ void ParseAttrValue(const std::string &type, const std::string &attr_name, const
   MS_EXCEPTION_IF_NULL(node_attr);
   MS_EXCEPTION_IF_NULL(value);
   if (type == "int") {
-    auto attr_value = static_cast<int>(GetValue<int64_t>(value));
+    auto attr_value = value->isa<Int32Imm>() ? GetValue<int>(value) : GetValue<int64_t>(value);
     (*node_attr)[attr_name].set_i(attr_value);
   } else if (type == "str") {
     auto attr_value = GetValue<std::string>(value);
@@ -186,6 +186,12 @@ void SetNodeInputs(const std::shared_ptr<AnfNode> &anf_node, mindspore::NodeDef
     return;
   }
 
+  std::vector<size_t> input_size_list;
+  if (!SetIOIputSize(anf_node, input_num, &input_size_list)) {
+    MS_LOG(ERROR) << "Node [" << AnfAlgo::GetCNodeName(anf_node) << "] get input size list failed.";
+    return;
+  }
+
   for (size_t input_index = 0; input_index < input_num; input_index++) {
     ::mindspore::Tensor *node_inputs = proto->add_inputs();
     MS_EXCEPTION_IF_NULL(node_inputs);
@@ -215,6 +221,7 @@ void SetNodeInputs(const std::shared_ptr<AnfNode> &anf_node, mindspore::NodeDef
     }
     node_inputs->set_tensor_type(input_data_type);
     node_inputs->set_mem_device("HBM");
+    node_inputs->set_data_size(input_size_list[input_index]);
   }
 }
 
@@ -243,8 +250,17 @@ void SetNodeOutputs(const std::shared_ptr<AnfNode> &anf_node, mindspore::NodeDef
     }
     TypeId output_type = AnfAlgo::GetOutputDeviceDataType(anf_node, output_index);
     int32_t output_data_type = AicpuOpUtil::MsTypeToProtoType(output_type);
+
+    int64_t data_size = 1;
+    if (!GetShapeSize(output_shape, TypeIdToType(output_type), &data_size)) {
+      MS_LOG(ERROR) << "Node [" << AnfAlgo::GetCNodeName(anf_node) << "] get output size failed for output "
+                    << output_index;
+      return;
+    }
+
     node_outputs->set_tensor_type(output_data_type);
     node_outputs->set_mem_device("HBM");
+    node_outputs->set_data_size(LongToSize(data_size));
   }
 }
 

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/CMakeLists.txt

@@ -24,13 +24,18 @@ if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "aarch64" AND EXISTS ${CMAKE_C_COMPILE
         ${CMAKE_CURRENT_SOURCE_DIR}/aicpu_sharder/aicpu_pulse.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/aicpu_sharder/aicpu_sharder.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/random_choice_with_mask_kernels.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/environ/aicpu_environ_manager.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/environ/environ_create.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/environ/environ_set.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/environ/environ_get.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/environ/environ_destroy_all.cc
     )
 
-    add_library(aicpu_kernels SHARED
+    add_library(mindspore_aicpu_kernels SHARED
         ${AICPU_SRC}
     )
 
-    target_compile_options(aicpu_kernels PRIVATE
+    target_compile_options(mindspore_aicpu_kernels PRIVATE
         -march=armv8-a
         -O2
         -fvisibility-inlines-hidden
@@ -39,7 +44,7 @@ if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "aarch64" AND EXISTS ${CMAKE_C_COMPILE
         -fno-common
     )
 
-    target_link_libraries(aicpu_kernels PRIVATE
+    target_link_libraries(mindspore_aicpu_kernels PRIVATE
         -ldl
         -shared
         PUBLIC
@@ -53,8 +58,8 @@ if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "aarch64" AND EXISTS ${CMAKE_C_COMPILE
     )
 
     set(INSTALL_LIBRARY_DIR lib)
-    install(TARGETS aicpu_kernels OPTIONAL
-        EXPORT aicpu_kernels-targets
+    install(TARGETS mindspore_aicpu_kernels OPTIONAL
+        EXPORT mindspore_aicpu_kernels-targets
         LIBRARY DESTINATION ${INSTALL_LIBRARY_DIR}
     )
 

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/common/kernel_log.h

@@ -1,77 +1,95 @@
-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-#ifndef AICPU_OPS_AICPU_COMMON_KERNEL_LOG_H_
-#define AICPU_OPS_AICPU_COMMON_KERNEL_LOG_H_
-
-#include <unistd.h>
-#include <sys/syscall.h>
-#include <iostream>
-#include <utility>
-#include "common/kernel_errcode.h"
-
-inline int GetTid(void) {
-  thread_local static int tid = syscall(__NR_gettid);
-  return tid;
-}
-static const int LOG_COUNT = 0;
-
-namespace aicpu {
-#define AICPU_LOG_DEBUG 0
-#define AICPU_LOG_INFO 1
-#define AICPU_LOG_WARN 2
-#define AICPU_LOG_ERROR 3
-#define AICPU_LOG_EVENT 0x10
-
-inline void PrintLog(const int level) { std::cerr << level << std::endl; }
-
-template <typename T, typename... Args>
-inline void PrintLog(const int level, T &&head, Args &&... tail) {
-  std::cerr << std::forward<T>(head) << " ";
-  PrintLog(level, std::forward<Args>(tail)...);
-}
-
-int LogSetLevel(int level);
-
-int LogGetLevel(void);
-
-bool CheckLogLevel(int log_level_check);
-
-#define AICPU_LOGD(fmt, ...) \
-  AICPU_LOG(AICPU_LOG_DEBUG, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
-#define AICPU_LOGI(fmt, ...) \
-  AICPU_LOG(AICPU_LOG_INFO, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
-#define AICPU_LOGW(fmt, ...) \
-  AICPU_LOG(AICPU_LOG_WARN, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
-#define AICPU_LOGE(fmt, ...) \
-  AICPU_LOG(AICPU_LOG_ERROR, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
-#define AICPU_LOGEVENT(fmt, ...) \
-  AICPU_LOG(AICPU_LOG_EVENT, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
-#define AICPU_LOG(level, fmt, ...)                                              \
-  do {                                                                          \
-    if (aicpu::CheckLogLevel(level)) {                                          \
-      aicpu::PrintLog(level, "[%s:%d]" fmt, __FILE__, __LINE__, ##__VA_ARGS__); \
-    }                                                                           \
-  } while (LOG_COUNT != 0)
-
-#define AICPU_CHK_STATUS_RET(expr...)           \
-  do {                                          \
-    const uint32_t status = (expr);             \
-    if (status != AICPU_KERNEL_STATE_SUCCESS) { \
-      return status;                            \
-    }                                           \
-  } while (0);
-}  // namespace aicpu
-#endif  // AICPU_OPS_AICPU_COMMON_KERNEL_LOG_H_
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef AICPU_OPS_AICPU_COMMON_KERNEL_LOG_H_
+#define AICPU_OPS_AICPU_COMMON_KERNEL_LOG_H_
+
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <iostream>
+#include <utility>
+#include "common/kernel_errcode.h"
+
+inline int GetTid(void) {
+  thread_local static int tid = syscall(__NR_gettid);
+  return tid;
+}
+static const int LOG_COUNT = 0;
+
+namespace aicpu {
+#define AICPU_LOG_DEBUG 0
+#define AICPU_LOG_INFO 1
+#define AICPU_LOG_WARN 2
+#define AICPU_LOG_ERROR 3
+#define AICPU_LOG_EVENT 0x10
+
+inline void PrintLog(const int level) { std::cerr << level << std::endl; }
+
+template <typename T, typename... Args>
+inline void PrintLog(const int level, T &&head, Args &&... tail) {
+  std::cerr << std::forward<T>(head) << " ";
+  PrintLog(level, std::forward<Args>(tail)...);
+}
+
+int LogSetLevel(int level);
+
+int LogGetLevel(void);
+
+bool CheckLogLevel(int log_level_check);
+
+#define AICPU_LOGD(fmt, ...) \
+  AICPU_LOG(AICPU_LOG_DEBUG, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
+#define AICPU_LOGI(fmt, ...) \
+  AICPU_LOG(AICPU_LOG_INFO, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
+#define AICPU_LOGW(fmt, ...) \
+  AICPU_LOG(AICPU_LOG_WARN, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
+#define AICPU_LOGE(fmt, ...) \
+  AICPU_LOG(AICPU_LOG_ERROR, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
+#define AICPU_LOGEVENT(fmt, ...) \
+  AICPU_LOG(AICPU_LOG_EVENT, "%s:%s:%d[tid:%lu]:" #fmt, __FUNCTION__, __FILE__, __LINE__, GetTid(), ##__VA_ARGS__);
+#define AICPU_LOG(level, fmt, ...)                                              \
+  do {                                                                          \
+    if (aicpu::CheckLogLevel(level)) {                                          \
+      aicpu::PrintLog(level, "[%s:%d]" fmt, __FILE__, __LINE__, ##__VA_ARGS__); \
+    }                                                                           \
+  } while (LOG_COUNT != 0)
+
+#define AICPU_CHK_STATUS_RET(expr...)           \
+  do {                                          \
+    const uint32_t status = (expr);             \
+    if (status != AICPU_KERNEL_STATE_SUCCESS) { \
+      return status;                            \
+    }                                           \
+  } while (0);
+
+#define AICPU_CHECK_NULLPTR_VOID(value, logText...) \
+  if (value == nullptr) {                           \
+    AICPU_LOGE(logText);                            \
+    return;                                         \
+  }
+
+#define AICPU_CHECK_FALSE(condition, errorCode, logText...) \
+  if (!(condition)) {                                       \
+    AICPU_LOGE(logText);                                    \
+    return errorCode;                                       \
+  }
+
+#define AICPU_CHECK_NULLPTR(value, errorCode, logText...) \
+  if (value == nullptr) {                                 \
+    AICPU_LOGE(logText);                                  \
+    return errorCode;                                     \
+  }
+}  // namespace aicpu
+#endif  // AICPU_OPS_AICPU_COMMON_KERNEL_LOG_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/aicpu_environ.h

@@ -0,0 +1,97 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_AICPU_ENVIRON_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_AICPU_ENVIRON_H_
+
+#include <vector>
+#include <string>
+#include <memory>
+#include <map>
+#include "common/kernel_log.h"
+
+namespace aicpu {
+constexpr int64_t kObjectTypeTensorType = 17;
+constexpr int64_t kObjectTypeEnvType = 26;
+// index of input or output
+enum Index : size_t {
+  kIndex0 = 0,
+  kIndex1,
+  kIndex2,
+  kIndex3,
+  kIndex4,
+  kIndex5,
+  kIndex6,
+  kIndex7,
+  kIndex8,
+  kIndex9,
+  kIndex10,
+  kIndex11,
+  kIndex12,
+  kIndex13,
+  kIndex14,
+  kIndex15,
+  kIndex16,
+};
+
+struct EnvironValue {
+  EnvironValue() : addr_(nullptr), size_(0), value_type_(kObjectTypeTensorType) {}
+  EnvironValue(void *address_addr, size_t address_size, int32_t value_type)
+      : addr_(address_addr), size_(address_size), value_type_(value_type) {}
+
+  void *addr_;
+  size_t size_;
+  int32_t value_type_;
+};
+
+using EnvironValuePtr = std::shared_ptr<EnvironValue>;
+
+// Environ is the meaning expression of map.
+class Environ {
+ public:
+  explicit Environ(int64_t handle) : handle_(handle) {}
+  virtual ~Environ() = default;
+
+  void Set(int64_t key, const EnvironValuePtr &value) { values_[key] = value; }
+
+  EnvironValuePtr Get(int64_t key) {
+    if (values_.count(key) > 0) {
+      return values_[key];
+    }
+    return nullptr;
+  }
+
+  void Clear() {
+    // Foreach values to free the value addr.
+    for (auto &value : values_) {
+      AICPU_CHECK_NULLPTR_VOID(value.second, "value.second is null.");
+      free(value.second->addr_);
+    }
+    values_.clear();
+    handle_ = 0;
+  }
+
+ private:
+  // The handle is unique for each env.
+  int64_t handle_ = 0;
+
+  // Store the tensors in map, as <key, tensor>.
+  std::map<int64_t, EnvironValuePtr> values_;
+};
+using EnvironPtr = std::shared_ptr<Environ>;
+}  // namespace aicpu
+
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_AICPU_ENVIRON_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/aicpu_environ_manager.cc

@@ -0,0 +1,100 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "environ/aicpu_environ_manager.h"
+#include <string>
+
+namespace aicpu {
+constexpr auto kScalarTensorShapeDim = 1;
+constexpr auto kScalarTensorShapeSize = 1;
+constexpr auto kEnvValueTypeAttr = "value_type";
+
+int64_t EnvironMgr::Create() {
+  std::unique_lock<std::mutex> lock(mutex);
+  if (env_handles_count_ >= INT64_MAX) {
+    AICPU_LOGE(" The handles number:%d is out of range: ", env_handles_count_);
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+  int64_t ret_handle = ++env_handles_count_;
+  auto env = std::make_shared<Environ>(ret_handle);
+  AICPU_CHECK_NULLPTR(env, AICPU_KERNEL_STATE_PARAM_INVALID, "env is null.");
+  envs_[ret_handle] = env;
+
+  return ret_handle;
+}
+
+EnvironPtr EnvironMgr::Get(int64_t handle) {
+  std::unique_lock<std::mutex> lock(mutex);
+  const auto &envIter = envs_.find(handle);
+  if (envIter != envs_.end()) {
+    auto &result = envIter->second;
+    return result;
+  }
+  return nullptr;
+}
+
+void EnvironMgr::Clear() {
+  std::unique_lock<std::mutex> lock(mutex);
+  for (auto &env : envs_) {
+    AICPU_CHECK_NULLPTR_VOID(env.second, "env is null.")
+    env.second->Clear();
+  }
+  envs_.clear();
+}
+
+bool EnvironMgr::IsScalarTensor(const aicpuops::Tensor &tensor) {
+  aicpuops::TensorShape shape = tensor.tensor_shape();
+  if (shape.dim_size() == 0) {
+    AICPU_LOGD("The shape is empty.");
+    return true;
+  }
+
+  if ((shape.dim_size() == kScalarTensorShapeDim) && (shape.dim(aicpu::kIndex0).size() == kScalarTensorShapeSize)) {
+    AICPU_LOGD("The tensor is scalar.");
+    return true;
+  }
+  return false;
+}
+
+bool EnvironMgr::CheckEnvInput(const aicpuops::NodeDef &node_def) {
+  ::google::protobuf::Map<::std::string, ::aicpuops::AttrValue> nodedef_map = node_def.attrs();
+  auto value_type_attr = nodedef_map[kEnvValueTypeAttr].i();
+  if ((value_type_attr != kObjectTypeTensorType) && (value_type_attr != kObjectTypeEnvType)) {
+    AICPU_LOGE("The value type is not supported: [%d]", value_type_attr);
+    return false;
+  }
+
+  // Check the input handle.
+  if (!IsScalarTensor(node_def.inputs(aicpu::kIndex0))) {
+    AICPU_LOGE("The input handle checks invalid.");
+    return false;
+  }
+
+  // Check the input key
+  if (!IsScalarTensor(node_def.inputs(aicpu::kIndex1))) {
+    AICPU_LOGE("The input key checks invalid.");
+    return false;
+  }
+
+  // Check the input value
+  if ((value_type_attr == kObjectTypeEnvType) && (!IsScalarTensor(node_def.inputs(aicpu::kIndex2)))) {
+    AICPU_LOGE("The input value checks invalid.");
+    return false;
+  }
+
+  return true;
+}
+}  // namespace aicpu

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/aicpu_environ_manager.h

@@ -0,0 +1,69 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_AICPU_ENVIRON_MANAGER_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_AICPU_ENVIRON_MANAGER_H_
+
+#include <utility>
+#include <map>
+#include <memory>
+#include <vector>
+#include <mutex>
+#include "environ/aicpu_environ.h"
+#include "aicpu_sharder/aicpu_sharder.h"
+#include "proto/aicpu_tensor.pb.h"
+#include "common/distinct_uniform_int_distribution.h"
+#include "common/tensor.h"
+
+namespace aicpu {
+class EnvironMgr {
+ public:
+  static EnvironMgr &GetInstance() noexcept {
+    static EnvironMgr instance;
+    return instance;
+  }
+
+  EnvironMgr(const EnvironMgr &) = delete;
+  EnvironMgr(EnvironMgr &&) = delete;
+  EnvironMgr &operator=(const EnvironMgr &) = delete;
+  EnvironMgr &operator=(EnvironMgr &&) = delete;
+
+  // Create the env object and return the unique env handle.
+  int64_t Create();
+
+  EnvironPtr Get(int64_t handle);
+
+  void Clear();
+
+  // Check whether the inputs of EnvironGet kernel or EnvironSet kernel are valid.
+  bool CheckEnvInput(const aicpuops::NodeDef &node_def);
+  // Check whether is scalar tensor. Environ handle and env key only support scalar tensor currently.
+  bool IsScalarTensor(const aicpuops::Tensor &tensor);
+
+ private:
+  EnvironMgr() = default;
+  ~EnvironMgr() = default;
+
+  // Store the envs in map, as <handle, env>.
+  std::map<int64_t, EnvironPtr> envs_;
+
+  int64_t env_handles_count_{0};
+
+  std::mutex mutex;
+};
+}  // namespace aicpu
+
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_AICPU_ENVIRON_MANAGER_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_create.cc

@@ -0,0 +1,46 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "environ/environ_create.h"
+#include "environ/aicpu_environ_manager.h"
+
+namespace aicpu {
+uint32_t EnvironCreateKernel::DoCompute() {
+  // Generate an unique handle.
+  int64_t env_handle = EnvironMgr::GetInstance().Create();
+  AICPU_LOGD("Create env handle:%d", env_handle);
+  auto *output_data = reinterpret_cast<int64_t *>(io_addrs_[aicpu::kIndex0]);
+  output_data[0] = env_handle;
+
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+
+uint32_t EnvironCreateKernel::ParseKernelParam() {
+  AICPU_LOGD("Enter ParseKernelParam.");
+  if (!EnvironMgr::GetInstance().IsScalarTensor(node_def_.outputs(aicpu::kIndex0))) {
+    AICPU_LOGE("The output is not scalar tensor.");
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+}  // namespace aicpu
+
+extern "C" {
+__attribute__((visibility("default"))) uint32_t EnvironCreate(void *param) {
+  aicpu::EnvironCreateKernel environCreateKernel;
+  return environCreateKernel.Compute(param);
+}
+}

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_create.h

@@ -0,0 +1,33 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_CREATE_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_CREATE_H_
+
+#include "common/kernel_base.h"
+
+namespace aicpu {
+class EnvironCreateKernel : public KernelBase {
+ public:
+  EnvironCreateKernel() : KernelBase("EnvironCreate") {}
+  ~EnvironCreateKernel() = default;
+
+ protected:
+  uint32_t DoCompute() override;
+  uint32_t ParseKernelParam() override;
+};
+}  // namespace aicpu
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_CREATE_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_destroy_all.cc

@@ -0,0 +1,42 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "environ/environ_destroy_all.h"
+#include "environ/aicpu_environ_manager.h"
+
+namespace aicpu {
+uint32_t EnvironDestroyAllKernel::DoCompute() {
+  AICPU_LOGD("Destroy all env handle");
+  EnvironMgr::GetInstance().Clear();
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+
+uint32_t EnvironDestroyAllKernel::ParseKernelParam() {
+  AICPU_LOGD("Enter ParseKernelParam.");
+  if (!EnvironMgr::GetInstance().IsScalarTensor(node_def_.outputs(aicpu::kIndex0))) {
+    AICPU_LOGE("The output is not scalar tensor.");
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+}  // namespace aicpu
+
+extern "C" {
+__attribute__((visibility("default"))) uint32_t EnvironDestroyAll(void *param) {
+  aicpu::EnvironDestroyAllKernel environDestroyAllKernel;
+  return environDestroyAllKernel.Compute(param);
+}
+}

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_destroy_all.h

@@ -0,0 +1,33 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_DESTORY_ALL_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_DESTORY_ALL_H_
+
+#include "common/kernel_base.h"
+
+namespace aicpu {
+class EnvironDestroyAllKernel : public KernelBase {
+ public:
+  EnvironDestroyAllKernel() : KernelBase("EnvironDestroyAll") {}
+  ~EnvironDestroyAllKernel() = default;
+
+ protected:
+  uint32_t DoCompute() override;
+  uint32_t ParseKernelParam() override;
+};
+}  // namespace aicpu
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_DESTORY_ALL_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_get.cc

@@ -0,0 +1,107 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "environ/environ_get.h"
+#include <random>
+#include <climits>
+#include <vector>
+#include <algorithm>
+#include <string>
+#include "mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/aicpu_sharder/aicpu_sharder.h"
+#include "mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/common/tensor.h"
+#include "environ/aicpu_environ_manager.h"
+
+namespace aicpu {
+constexpr auto kEnvValueTypeAttr = "value_type";
+
+uint32_t EnvironGetKernel::DoCompute() {
+  AICPU_LOGD("Enter DoCompute.");
+  auto &env_mgr = EnvironMgr::GetInstance();
+
+  auto *input_handle_ptr = reinterpret_cast<int64_t *>((io_addrs_[aicpu::kIndex0]));
+  auto *input_key_ptr = reinterpret_cast<int64_t *>((io_addrs_[aicpu::kIndex1]));
+  auto *default_value_ptr = reinterpret_cast<void *>((io_addrs_[aicpu::kIndex2]));
+  auto *output_ptr = reinterpret_cast<void *>((io_addrs_[aicpu::kIndex3]));
+
+  // Get handle and key
+  int64_t handle = input_handle_ptr[0];
+  int64_t key = input_key_ptr[0];
+
+  // Get env and value by handle and key
+  const auto &env = env_mgr.Get(handle);
+  AICPU_CHECK_NULLPTR(env, AICPU_KERNEL_STATE_PARAM_INVALID, "Get env [%d] failed", handle)
+  const auto &env_value = env->Get(key);
+
+  AICPU_LOGD("EnvironGetKernel: hindle[%d], key[%d], value[%d]", handle, key, (void *)&env_value);
+  // Default value
+  auto *output_value_ptr = default_value_ptr;
+  auto output_value_size = default_value_size_;
+  auto output_value_type = attr_value_type_;
+  if (env_value != nullptr) {
+    output_value_ptr = env_value->addr_;
+    output_value_size = env_value->size_;
+    output_value_type = env_value->value_type_;
+  } else {
+    AICPU_LOGE("Get key[%d] value checks failed.", key);
+  }
+
+  if ((output_value_size_ < output_value_size) || (output_value_type != attr_value_type_)) {
+    AICPU_LOGE("The env value checks invalid, value_size: %d vs %d, value_type:%d vs %d", output_value_size_,
+               output_value_size, output_value_type, attr_value_type_);
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+
+  auto ret = memcpy_s(output_ptr, output_value_size_, output_value_ptr, output_value_size_);
+  AICPU_CHECK_FALSE((ret == EOK), AICPU_KERNEL_STATE_PARAM_INVALID,
+                    "Memcpy size[%zu] from env map to output[0] failed.", output_value_size_);
+
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+
+uint32_t EnvironGetKernel::ParseKernelParam() {
+  AICPU_LOGD("Enter ParseKernelParam.");
+  auto &env_mgr = EnvironMgr::GetInstance();
+  if (!env_mgr.CheckEnvInput(node_def_)) {
+    AICPU_LOGE("The input checks invalid. ");
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+
+  // Get value type attr
+  ::google::protobuf::Map<::std::string, ::aicpuops::AttrValue> nodedef_map = node_def_.attrs();
+  attr_value_type_ = nodedef_map[kEnvValueTypeAttr].i();
+
+  // check output value
+  auto default_value_tensor = node_def_.inputs(aicpu::kIndex2);
+  auto output_value_ptr_tensor = node_def_.outputs(aicpu::kIndex0);
+  if ((output_value_ptr_tensor.tensor_shape().dim_size() != default_value_tensor.tensor_shape().dim_size()) ||
+      (output_value_ptr_tensor.tensor_type() != default_value_tensor.tensor_type())) {
+    AICPU_LOGE("The env value checks invalid.");
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+
+  // Get value size.
+  default_value_size_ = default_value_tensor.data_size();
+  output_value_size_ = output_value_ptr_tensor.data_size();
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+}  // namespace aicpu
+
+extern "C" {
+__attribute__((visibility("default"))) uint32_t EnvironGet(void *param) {
+  aicpu::EnvironGetKernel environGetKernel;
+  return environGetKernel.Compute(param);
+}
+}

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_get.h

@@ -0,0 +1,39 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_GET_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_GET_H_
+
+#include <vector>
+#include "mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/common/kernel_base.h"
+
+namespace aicpu {
+class EnvironGetKernel : public KernelBase {
+ public:
+  EnvironGetKernel() : KernelBase("EnvironGet") {}
+  ~EnvironGetKernel() = default;
+
+ protected:
+  uint32_t DoCompute() override;
+  uint32_t ParseKernelParam() override;
+
+ private:
+  int32_t attr_value_type_{0};
+  size_t default_value_size_{0};
+  size_t output_value_size_{0};
+};
+}  // namespace aicpu
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_GET_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_set.cc

@@ -0,0 +1,85 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "environ/environ_set.h"
+#include <string>
+#include <memory>
+#include "mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/aicpu_sharder/aicpu_sharder.h"
+#include "mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/common/tensor.h"
+#include "environ/aicpu_environ_manager.h"
+
+namespace aicpu {
+constexpr auto kEnvValueTypeAttr = "value_type";
+
+uint32_t EnvironSetKernel::DoCompute() {
+  AICPU_LOGD("Enter DoCompute.");
+  auto &env_mgr = EnvironMgr::GetInstance();
+
+  auto *input_handle_ptr = reinterpret_cast<int64_t *>(io_addrs_[aicpu::kIndex0]);
+  auto *input_key_ptr = reinterpret_cast<int64_t *>(io_addrs_[aicpu::kIndex1]);
+  auto *input_value_ptr = reinterpret_cast<void *>(io_addrs_[aicpu::kIndex2]);
+  auto *output_handle_ptr = reinterpret_cast<int64_t *>(io_addrs_[aicpu::kIndex3]);
+
+  auto *value_ptr = malloc(value_size_);
+  AICPU_CHECK_NULLPTR(value_ptr, AICPU_KERNEL_STATE_PARAM_INVALID, "Malloc failed.")
+  auto ret = memcpy_s(value_ptr, value_size_, input_value_ptr, value_size_);
+  AICPU_CHECK_FALSE((ret == EOK), AICPU_KERNEL_STATE_PARAM_INVALID, "Memcpy size from input[2] to environ failed.",
+                    value_size_);
+
+  // Set env member.
+  const auto &env = env_mgr.Get(input_handle_ptr[0]);
+  AICPU_CHECK_NULLPTR(env, AICPU_KERNEL_STATE_PARAM_INVALID, "Get handle[%d] failed.", input_handle_ptr[0]);
+
+  auto env_value = std::make_shared<EnvironValue>(value_ptr, value_size_, attr_value_type_);
+  env->Set(input_key_ptr[0], env_value);
+  AICPU_LOGD("EnvironSetKernel: handle[%d], key[%d], value[%d]", input_handle_ptr[0], input_key_ptr[0],
+             (void *)&env_value);
+
+  // Set output handle
+  output_handle_ptr[0] = input_handle_ptr[0];
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+
+uint32_t EnvironSetKernel::ParseKernelParam() {
+  AICPU_LOGD("Enter ParseKernelParam.");
+  auto &env_mgr = EnvironMgr::GetInstance();
+  if (!env_mgr.CheckEnvInput(node_def_)) {
+    AICPU_LOGE("The input checks invalid. ");
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+
+  if (!env_mgr.IsScalarTensor(node_def_.outputs(aicpu::kIndex0))) {
+    AICPU_LOGE("The output handle is not equal of input handle.");
+    return AICPU_KERNEL_STATE_PARAM_INVALID;
+  }
+
+  // Get value type.
+  ::google::protobuf::Map<::std::string, ::aicpuops::AttrValue> nodedef_map = node_def_.attrs();
+  attr_value_type_ = nodedef_map[kEnvValueTypeAttr].i();
+
+  // Get value size.
+  aicpuops::Tensor value_tensor = node_def_.inputs(aicpu::kIndex2);
+  value_size_ = value_tensor.data_size();
+  return AICPU_KERNEL_STATE_SUCCESS;
+}
+}  // namespace aicpu
+
+extern "C" {
+__attribute__((visibility("default"))) uint32_t EnvironSet(void *param) {
+  aicpu::EnvironSetKernel environSetKernel;
+  return environSetKernel.Compute(param);
+}
+}

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_ops/environ/environ_set.h

@@ -0,0 +1,36 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_SET_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_SET_H_
+
+#include "common/kernel_base.h"
+namespace aicpu {
+class EnvironSetKernel : public KernelBase {
+ public:
+  EnvironSetKernel() : KernelBase("EnvironSet") {}
+  ~EnvironSetKernel() = default;
+
+ protected:
+  uint32_t DoCompute() override;
+  uint32_t ParseKernelParam() override;
+
+ private:
+  int32_t attr_value_type_{0};
+  size_t value_size_{0};
+};
+}  // namespace aicpu
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_AICPU_AICPU_OPS_ENVIRON_ENVIRON_SET_H_

mindspore/ccsrc/backend/kernel_compiler/aicpu/aicpu_util.h

@@ -71,11 +71,16 @@ constexpr auto kSearchSorted = "SearchSorted";
 constexpr auto kResizeBilinear = "ResizeBilinear";
 constexpr auto kResizeBilinearGrad = "ResizeBilinearGrad";
 constexpr auto kScatterElements = "ScatterElements";
+constexpr auto kEnvironCreate = "EnvironCreate";
+constexpr auto kEnvironSet = "EnvironSet";
+constexpr auto kEnvironGet = "EnvironGet";
+constexpr auto kEnvironDestroyAll = "EnvironDestroyAll";
 const std::set<std::string> kCpuKernelOps{kIdentity,     kMaskedSelect,   kMaskedSelectGrad,   kDynamicStitch,
                                           kSearchSorted, kResizeBilinear, kResizeBilinearGrad, kScatterElements};
 const std::set<std::string> kCacheKernelOps{kUpdateCache, kCacheSwapTable, kSubAndFilter,       kPadAndShift,
                                             kDropout3D,   kDropout2D,      kNonMaxSuppressionV3};
-const std::set<std::string> kCpuKernelBaseOps{kGetNext, kInitData, kRandomChoiceWithMask};
+const std::set<std::string> kCpuKernelBaseOps{kGetNext,    kInitData,   kRandomChoiceWithMask, kEnvironCreate,
+                                              kEnvironSet, kEnvironGet, kEnvironDestroyAll};
 const std::set<std::string> kDynamicInputOps{
   kPrint, kPack, kMeshgrid, kStackInitOpName, kStackDestroyOpName, kStackPushOpName, kStackPopOpName, kDynamicStitch};
 struct AicpuParamHead {

mindspore/ccsrc/backend/kernel_compiler/akg/akg_kernel_build.cc

@@ -22,6 +22,7 @@
 #include <fcntl.h>
 #include <unistd.h>
 
+#include <chrono>
 #include <algorithm>
 #include <map>
 #include <memory>
@@ -44,6 +45,8 @@ constexpr int32_t MAX_ERROR_LEN = 1024;
 constexpr int32_t PROCESS_NUM = 16;
 constexpr int32_t TIME_OUT = 300;
 
+#define ACQUIRE_LOCK LockMng lock(fd_, __func__, __LINE__)
+
 inline std::string GetErrorInfo() {
   char buf[MAX_ERROR_LEN + 1] = {0};
   auto ret = strerror_r(errno, buf, MAX_ERROR_LEN);
@@ -59,8 +62,9 @@ inline std::string GetErrorInfo() {
 }
 
 bool AkgKernelPool::LockMng::TryLock() const {
-  // Try to lock 100 times. Return errno if lock unsuccessfully
-  uint32_t trial = 100;
+  // Try to lock trial times. Return errno if lock unsuccessfully
+  uint32_t trial = 2000;
+  const uint32_t sleep_time_us = 5000;
 
   int32_t ret = -1;
   while (trial > 0) {
@@ -70,14 +74,15 @@ bool AkgKernelPool::LockMng::TryLock() const {
     }
 
     trial--;
-    (void)usleep(5000);
+    (void)usleep(sleep_time_us);
   }
 
   if (ret == -1) {
-    MS_LOG(ERROR) << "Failed to acquire the lock, error msg:" << GetErrorInfo() << ".";
+    MS_LOG(ERROR) << "Failed to acquire the lock, error msg:" << GetErrorInfo() << ", left trying times: " << trial;
     return false;
   }
 
+  MS_LOG(INFO) << "AkgKernelBuild successfully acquire lock called at " << calling_position_;
   return true;
 }
 
@@ -86,6 +91,7 @@ void AkgKernelPool::LockMng::Unlock() const {
   if (ret == -1) {
     MS_LOG(ERROR) << "Failed to release the lock, error msg:" << GetErrorInfo();
   }
+  MS_LOG(INFO) << "AkgKernelBuild successfully release lock called at " << calling_position_;
 }
 
 std::string AkgKernelPool::GetCurrentPath() const {
@@ -114,7 +120,7 @@ void *AkgKernelPool::CreateSharedMem(const std::string &path) {
   auto mem_size = sizeof(size_t) * kListNum_ * (kMaxKernelNum_ + 1) + 512;
 
   {
-    LockMng lock(fd_);
+    ACQUIRE_LOCK;
     if (!lock.locked_) {
       MS_LOG(ERROR) << "Failed to acquire lock.";
       return nullptr;
@@ -140,7 +146,7 @@ void *AkgKernelPool::CreateSharedMem(const std::string &path) {
     }
   }
 
-  LockMng lock(fd_);
+  ACQUIRE_LOCK;
   if (!lock.locked_) {
     MS_LOG(ERROR) << "Failed to acquire lock.";
     return nullptr;
@@ -203,7 +209,7 @@ int32_t AkgKernelPool::Init(const std::vector<JsonNodePair> &build_args) {
 
 int32_t AkgKernelPool::Release() const {
   {
-    LockMng lock(fd_);
+    ACQUIRE_LOCK;
     if (!lock.locked_) {
       MS_LOG(ERROR) << "Failed to acquire lock.";
       return -1;
@@ -244,7 +250,7 @@ int32_t AkgKernelPool::Release() const {
 }
 
 int32_t AkgKernelPool::AddKernels(const std::vector<JsonNodePair> &build_args) {
-  LockMng lock(fd_);
+  ACQUIRE_LOCK;
   if (!lock.locked_) {
     MS_LOG(ERROR) << "Failed to acquire lock.";
     return -1;
@@ -293,7 +299,7 @@ int32_t AkgKernelPool::AddKernels(const std::vector<JsonNodePair> &build_args) {
 }
 
 int32_t AkgKernelPool::FetchKernels(std::set<size_t> *out) {
-  LockMng lock(fd_);
+  ACQUIRE_LOCK;
   if (!lock.locked_) {
     MS_LOG(ERROR) << "Failed to acquire lock.";
     return -1;
@@ -301,7 +307,7 @@ int32_t AkgKernelPool::FetchKernels(std::set<size_t> *out) {
 
   std::set<size_t> left_in_todo_list;
 
-  // filter out kernels which belongs to other processes
+  // filter out kernels which does not belongs to this process
   auto FilterBySelfList = [&left_in_todo_list, &out, this](size_t id) {
     if (this->self_kernel_ids_.count(id) != 0) {
       (void)out->emplace(id);
@@ -323,7 +329,7 @@ int32_t AkgKernelPool::FetchKernels(std::set<size_t> *out) {
 
 int32_t AkgKernelPool::UpdateAndWait(const std::set<size_t> &ids) {
   if (!ids.empty()) {
-    LockMng lock(fd_);
+    ACQUIRE_LOCK;
     if (!lock.locked_) {
       MS_LOG(ERROR) << "Failed to acquire lock.";
       return -1;
@@ -355,10 +361,11 @@ int32_t AkgKernelPool::UpdateAndWait(const std::set<size_t> &ids) {
 int32_t AkgKernelPool::Wait() const {
   // wait until all the kernels which belong to this process finish compiling
   uint32_t trials = 1000;
+  const uint32_t sleep_time_us = 1000000;
 
   while (trials > 0) {
     {
-      LockMng lock(fd_);
+      ACQUIRE_LOCK;
       if (!lock.locked_) {
         MS_LOG(ERROR) << "Failed to acquire lock.";
         return -1;
@@ -372,7 +379,7 @@ int32_t AkgKernelPool::Wait() const {
       }
     }
 
-    (void)usleep(1000000);
+    (void)usleep(sleep_time_us);
     trials--;
   }
 
@@ -565,6 +572,7 @@ void AkgKernelBuilder::LoadCache() {
     (void)bin_map->Insert(kernel_name, kernel_dir + kernel_json);
   }
   has_load = true;
+  (void)closedir(dir);
   return;
 }
 

mindspore/ccsrc/backend/kernel_compiler/akg/akg_kernel_build.h

@@ -61,8 +61,9 @@ class AkgKernelPool {
  public:
   class LockMng {
    public:
-    explicit LockMng(int32_t fd) {
+    explicit LockMng(const int32_t fd, const char *function, const uint32_t line) {
       fd_ = fd;
+      calling_position_ = std::string(function) + ":" + std::to_string(line);
       locked_ = TryLock();
     }
 
@@ -79,6 +80,7 @@ class AkgKernelPool {
     void Unlock() const;
 
     int32_t fd_{-1};
+    std::string calling_position_;
   };
 
  public:

mindspore/ccsrc/backend/kernel_compiler/akg/akg_kernel_json_generator.cc

@@ -90,7 +90,8 @@ std::vector<std::pair<AnfNodePtr, std::pair<size_t, size_t>>> GetInputIndex(cons
           accum_idx += LongToInt(dyn_input_sizes[dyn_i]);
           if (used_as_idx < accum_idx) {
             input_index.push_back(std::make_pair(
-              anf_node, std::make_pair(dyn_i, IntToSize(used_as_idx - (accum_idx - dyn_input_sizes[dyn_i])))));
+              anf_node,
+              std::make_pair(dyn_i, IntToSize(used_as_idx - (accum_idx - LongToInt(dyn_input_sizes[dyn_i]))))));
             found = true;
             break;
           }
@@ -335,7 +336,7 @@ bool AkgKernelJsonGenerator::CreateInputDescJson(const AnfNodePtr &anf_node, con
       input_list.emplace_back(input_desc_json);
       real_input_index++;
     }
-    inputs_json->emplace_back(input_list);
+    (void)inputs_json->emplace_back(input_list);
   }
   return true;
 }
@@ -877,7 +878,7 @@ nlohmann::json AkgKernelJsonGenerator::CreateInputsJson(const std::vector<AnfNod
       input_shape.push_back(1);
     }
     input_desc_json[kJsonKeyShape] = input_shape;
-    inputs_json.emplace_back(std::vector<nlohmann::json>{input_desc_json});
+    (void)inputs_json.emplace_back(std::vector<nlohmann::json>{input_desc_json});
   }
   return inputs_json;
 }

mindspore/ccsrc/backend/kernel_compiler/akg/ascend/akg_ascend_kernel_mod.cc

@@ -73,12 +73,12 @@ bool AkgKernelMod::Launch(const std::vector<AddressPtr> &inputs, const std::vect
   // pack all addresses into a vector.
   std::vector<void *> runtime_args;
   (void)std::transform(std::begin(inputs), std::end(inputs), std::back_inserter(runtime_args),
-                       [](const AddressPtr &input) -> void * { return input->addr; });
+                       [](const AddressPtr &input) { return input->addr; });
   (void)std::transform(std::begin(outputs), std::end(outputs), std::back_inserter(runtime_args),
-                       [](const AddressPtr &output) -> void * { return output->addr; });
+                       [](const AddressPtr &output) { return output->addr; });
   if (!workspace.empty()) {
     (void)std::transform(std::begin(workspace), std::end(workspace), std::back_inserter(runtime_args),
-                         [](const AddressPtr &addr) -> void * { return addr->addr; });
+                         [](const AddressPtr &addr) { return addr->addr; });
   }
 
   rtL2Ctrl_t *l2ctrl = nullptr;
@@ -111,12 +111,12 @@ std::vector<TaskInfoPtr> AkgKernelMod::GenTask(const std::vector<AddressPtr> &in
 
   // pack all addresses into a vector.
   (void)std::transform(std::begin(inputs), std::end(inputs), std::back_inserter(input_data_addrs),
-                       [](const AddressPtr &input) -> void * { return input->addr; });
+                       [](const AddressPtr &input) { return input->addr; });
   (void)std::transform(std::begin(outputs), std::end(outputs), std::back_inserter(output_data_addrs),
-                       [](const AddressPtr &output) -> void * { return output->addr; });
+                       [](const AddressPtr &output) { return output->addr; });
   if (!workspace.empty()) {
     (void)std::transform(std::begin(workspace), std::end(workspace), std::back_inserter(workspace_addrs),
-                         [](const AddressPtr &workspace) -> void * { return workspace->addr; });
+                         [](const AddressPtr &workspace) { return workspace->addr; });
   }
 
   uint32_t block_dim = DEFAULT_BLOCK_DIM;  // default blockdim equal to 1.

mindspore/ccsrc/backend/kernel_compiler/akg/cpu/akg_cpu_kernel_mod.cc

@@ -33,7 +33,7 @@ namespace kernel {
 class AkgParallelLaunch {
  public:
   using AkgParallelLambda = int (*)(int task_id, int num_task, void *cdata);
-  static int AkgLaunchFunc(AkgParallelLambda flambda, void *cdata, int num_task) {
+  static int AkgLaunchFunc(AkgParallelLambda flambda, void *cdata, int) {
     auto nthreads = omp_get_max_threads();
 #pragma omp parallel num_threads(nthreads)
     { flambda(omp_get_thread_num(), nthreads, cdata); }
@@ -41,10 +41,11 @@ class AkgParallelLaunch {
   }
 };
 
-struct AkgCallBack {
-  void *parallel_launch_func;
-  void *(*malloc_func)(size_t);
-  void (*free_func)(void *);
+class AkgCallBack {
+ public:
+  void *parallel_launch_func = nullptr;
+  void *(*malloc_func)(size_t) = nullptr;
+  void (*free_func)(void *) = nullptr;
 
   AkgCallBack() {
     parallel_launch_func = reinterpret_cast<void *>(&AkgParallelLaunch::AkgLaunchFunc);
@@ -121,16 +122,16 @@ CpuKernelMod::CpuKernelMod(const KernelPackPtr &kp) {
 }
 
 bool CpuKernelMod::Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
-                          const std::vector<AddressPtr> &outputs, void *stream_ptr) {
+                          const std::vector<AddressPtr> &outputs, void *) {
   if (launch_func_ == nullptr) {
     MS_LOG(ERROR) << "GetFunction failed. kernel: " << kernel_name_;
     return false;
   }
   std::vector<void *> runtimeargs;
   (void)std::transform(std::begin(inputs), std::end(inputs), std::back_inserter(runtimeargs),
-                       [](const AddressPtr &input) -> void * { return input->addr; });
+                       [](const AddressPtr &input) { return input->addr; });
   (void)std::transform(std::begin(outputs), std::end(outputs), std::back_inserter(runtimeargs),
-                       [](const AddressPtr &output) -> void * { return output->addr; });
+                       [](const AddressPtr &output) { return output->addr; });
   static AkgCallBack akg_callback;
   (void)runtimeargs.emplace_back(reinterpret_cast<void *>(&akg_callback));
   using AkgCpuKernelFunction = void (*)(void *);

mindspore/ccsrc/backend/kernel_compiler/akg/cpu/akg_cpu_kernel_mod.h

@@ -55,7 +55,7 @@ class CpuKernelMod : public KernelMod {
   const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
   const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
   bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
-              const std::vector<AddressPtr> &outputs, void *stream_ptr) override;
+              const std::vector<AddressPtr> &outputs, void *) override;
 
   static CpuKernelManagerPtr kernelmanager_;
 

mindspore/ccsrc/backend/kernel_compiler/akg/gpu/akg_gpu_kernel_mod.cc

@@ -16,6 +16,7 @@
 
 #include "backend/kernel_compiler/akg/gpu/akg_gpu_kernel_mod.h"
 
+#include <algorithm>
 #include "nlohmann/json.hpp"
 #include "utils/ms_utils.h"
 
@@ -29,6 +30,12 @@ const int MAX_REGISTER_PER_THREAD_BLOCK = 65536;
 const int REGISTER_UNIT_IN_WARP = 256;
 const int WARP_SIZE = 32;
 const int WARP_ALLOC_GRAN = 4;
+const int BLOCKIDX_X_INDEX = 0;
+const int BLOCKIDX_Y_INDEX = 1;
+const int BLOCKIDX_Z_INDEX = 2;
+const int THREADIDX_X_INDEX = 3;
+const int THREADIDX_Y_INDEX = 4;
+const int THREADIDX_Z_INDEX = 5;
 
 GpuKernelManagerPtr GpuKernelMod::kernelmanager_ = std::make_shared<GpuKernelManager>();
 GpuKernelManager::GpuKernelManager() {}
@@ -120,15 +127,16 @@ bool GpuKernelMod::Launch(const std::vector<AddressPtr> &inputs, const std::vect
   }
   std::vector<void *> runtimeargs;
   (void)std::transform(std::begin(inputs), std::end(inputs), std::back_inserter(runtimeargs),
-                       [](const AddressPtr &input) -> void * { return reinterpret_cast<void *>(&(input->addr)); });
+                       [](const AddressPtr &input) { return reinterpret_cast<void *>(&(input->addr)); });
   (void)std::transform(std::begin(outputs), std::end(outputs), std::back_inserter(runtimeargs),
-                       [](const AddressPtr &output) -> void * { return reinterpret_cast<void *>(&(output->addr)); });
+                       [](const AddressPtr &output) { return reinterpret_cast<void *>(&(output->addr)); });
   if (!workspace.empty()) {
     (void)std::transform(std::begin(workspace), std::end(workspace), std::back_inserter(runtimeargs),
-                         [](const AddressPtr &addr) -> void * { return reinterpret_cast<void *>(&(addr->addr)); });
+                         [](const AddressPtr &addr) { return reinterpret_cast<void *>(&(addr->addr)); });
   }
-  result = cuLaunchKernel(kernel_addr, thread_info[0], thread_info[1], thread_info[2], thread_info[3], thread_info[4],
-                          thread_info[5], 0, reinterpret_cast<CUstream>(stream_ptr),
+  result = cuLaunchKernel(kernel_addr, thread_info[BLOCKIDX_X_INDEX], thread_info[BLOCKIDX_Y_INDEX],
+                          thread_info[BLOCKIDX_Z_INDEX], thread_info[THREADIDX_X_INDEX], thread_info[THREADIDX_Y_INDEX],
+                          thread_info[THREADIDX_Z_INDEX], 0, reinterpret_cast<CUstream>(stream_ptr),
                           reinterpret_cast<void **>(&runtimeargs[0]), 0);
   if (result != CUDA_SUCCESS) {
     const char *msg = nullptr;

mindspore/ccsrc/backend/kernel_compiler/common_utils.cc

@@ -215,8 +215,7 @@ KernelPackPtr SearchCache(const std::string &kernel_name, const std::string &pro
 KernelPackPtr InsertCache(const std::string &kernel_name, const std::string &processor) {
   MS_LOG(INFO) << "Insert cache for kernel:" << kernel_name << ", processr:" << processor;
   KernelMeta *bin_map = KernelMeta::GetInstance();
-  std::string kernel_json;
-  kernel_json = bin_map->kernel_meta_path();
+  std::string kernel_json = bin_map->kernel_meta_path();
   (void)kernel_json.append(kernel_name).append(kJsonSuffix);
   KernelPackPtr kernel_pack = std::make_shared<KernelPack>();
   if (!kernel_pack->ReadFromJsonFile(kernel_json, processor)) {
@@ -984,9 +983,9 @@ size_t UnitSizeInBytes(const mindspore::TypeId &t) {
     case kNumberTypeFloat64:
       bytes = sizeof(int64_t);
       break;
+    case kNumberTypeInt4:
     default:
       MS_LOG(EXCEPTION) << "Invalid types " << t;
-      break;
   }
 
   return bytes;

mindspore/ccsrc/backend/kernel_compiler/cpu/arithmetic_cpu_kernel.cc

@@ -459,12 +459,13 @@ bool ArithmeticCPUKernel<T>::Launch(const std::vector<AddressPtr> &inputs, const
                                     const std::vector<AddressPtr> &outputs) {
   CHECK_KERNEL_INPUTS_NUM(inputs.size(), kInputsNum, kernel_name_);
   CHECK_KERNEL_OUTPUTS_NUM(outputs.size(), kOutputsNum, kernel_name_);
-  if (output_size_ == 0) {
-    return true;
-  }
   auto *input1 = reinterpret_cast<T *>(inputs[0]->addr);
   const auto *input2 = reinterpret_cast<T *>(inputs[1]->addr);
   auto *output = reinterpret_cast<T *>(outputs[0]->addr);
+  if (output_size_ == 0) {
+    MS_LOG(WARNING) << kernel_name_ << " output shape contain 0, output_shape: " << output_shape_;
+    return true;
+  }
   if (kernel_name_ == prim::kPrimAssignAdd->name()) {
     AssignAdd(input1, input2, output);
   } else {

mindspore/ccsrc/backend/kernel_compiler/cpu/broadcast_to_cpu_kernel.cc

@@ -64,10 +64,8 @@ void BroadcastToCPUKernel<T>::CheckArgs() {
     if (input_shape_[i] != output_shape_[i + offset] && input_shape_[i] != 1) {
       MS_LOG(EXCEPTION)
         << "For '" << kernel_name_ << "', when the " << i
-        << "'th dimension of input tensor 'input_x' "
-           "and target shape 'shape' not equal, the dimension length of input tensor 'input_x' should be "
-           "1. But got the dimension of input tensor 'input_x': "
-        << Vector2Str(input_shape_) << ", and the dimension of target shape 'shape': " << Vector2Str(output_shape_);
+        << "'th, the shape of input should be 1 and equal to the shape of output, but got the shape of input: "
+        << Vector2Str(input_shape_) << ", and the shape of output: " << Vector2Str(output_shape_);
     }
   }
 }