Merge branch main into dev/wasi_threads

.devcontainer/Dockerfile

@@ -12,7 +12,7 @@ ENV TZ=Asian/Shanghai
 # hadolint ignore=DL3008
 RUN apt-get update \
   && apt-get install -y apt-transport-https apt-utils build-essential \
-  ca-certificates curl g++-multilib git gnupg \
+  ca-certificates ccache curl g++-multilib git gnupg \
   libgcc-9-dev lib32gcc-9-dev lsb-release \
   ninja-build ocaml ocamlbuild python2.7 \
   software-properties-common tree tzdata \
@@ -20,6 +20,15 @@ RUN apt-get update \
   && apt-get clean -y \
   && rm -rf /var/lib/apt/lists/*
 
+#
+# binaryen
+ARG BINARYEN_VER=111
+WORKDIR /opt
+RUN wget -c --progress=dot:giga https://github.com/WebAssembly/binaryen/releases/download/version_${BINARYEN_VER}/binaryen-version_${BINARYEN_VER}-x86_64-linux.tar.gz \
+  && tar xf binaryen-version_${BINARYEN_VER}-x86_64-linux.tar.gz \
+  && ln -sf /opt/binaryen-version_111 /opt/binaryen \
+  && rm binaryen-version_${BINARYEN_VER}-x86_64-linux.tar.gz
+
 #
 # CMAKE (https://apt.kitware.com/)
 SHELL ["/bin/bash", "-o", "pipefail", "-c"]
@@ -38,18 +47,19 @@ RUN wget --progress=dot:giga -O - https://apt.kitware.com/keys/kitware-archive-l
 WORKDIR /opt
 RUN git clone https://github.com/emscripten-core/emsdk.git
 
+ARG EMSDK_VER=3.0.0
 WORKDIR /opt/emsdk
 RUN  git pull \
-  && ./emsdk install 2.0.26 \
+  && ./emsdk install ${EMSDK_VER} \
-  && ./emsdk activate 2.0.26 \
+  && ./emsdk activate ${EMSDK_VER} \
   && echo "source /opt/emsdk/emsdk_env.sh" >> /root/.bashrc
 
 #
 # install wasi-sdk
-ARG WASI_SDK_VER=16
+ARG WASI_SDK_VER=19
 RUN wget -c --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-${WASI_SDK_VER}/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz -P /opt \
   && tar xf /opt/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz -C /opt \
-  && ln -fs /opt/wasi-sdk-${WASI_SDK_VER}.0 /opt/wasi-sdk \
+  && ln -sf /opt/wasi-sdk-${WASI_SDK_VER}.0 /opt/wasi-sdk \
   && rm /opt/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz
 
 #
@@ -57,7 +67,7 @@ RUN wget -c --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases
 ARG WABT_VER=1.0.29
 RUN wget -c --progress=dot:giga https://github.com/WebAssembly/wabt/releases/download/${WABT_VER}/wabt-${WABT_VER}-ubuntu.tar.gz -P /opt \
   && tar xf /opt/wabt-${WABT_VER}-ubuntu.tar.gz -C /opt \
-  && ln -fs /opt/wabt-${WABT_VER} /opt/wabt \
+  && ln -sf /opt/wabt-${WABT_VER} /opt/wabt \
   && rm /opt/wabt-${WABT_VER}-ubuntu.tar.gz
 
 #
@@ -70,6 +80,8 @@ RUN mkdir /opt/bazelisk \
 
 #
 # install clang+llvm
+ARG LLVM_VER=14
+RUN apt-get purge -y clang-10 llvm-10 && apt autoremove -y
 WORKDIR /etc/apt/apt.conf.d
 RUN touch 99verfiy-peer.conf \
   && echo "Acquire { https::Verify-Peer false }" > 99verfiy-peer.conf
@@ -77,9 +89,7 @@ RUN touch 99verfiy-peer.conf \
 WORKDIR /tmp
 RUN wget --progress=dot:giga https://apt.llvm.org/llvm.sh \
   && chmod a+x ./llvm.sh \
-  && /tmp/llvm.sh 12 all \ 
+  && ./llvm.sh ${LLVM_VER} all
-  && ln -sf /usr/bin/clang-format-12 /usr/bin/clang-format \
-  && rm -rf /tmp/*
 
 #
 # [Optional]
@@ -96,17 +106,28 @@ RUN apt-get update \
 # Install required python packages
 # hadolint ignore=DL3013
 RUN python3 -m pip install --no-cache-dir --upgrade pip \
-  && pip3 install --no-cache-dir --user black nose pycparser pylint
+  && pip3 install --no-cache-dir black nose pycparser pylint
 
-# set path, PS and clean up
+#
-ENV PATH "/opt/bazelisk:/opt/clang-llvm/bin:${PATH}"
+# Install github-cli. It doens't work as a feature of devcontainer.json
-RUN echo "export PATH=/opt/bazelisk:/opt/clang-llvm/bin:${PATH}" >> /root/.bashrc \
+RUN cd /tmp \
-  && printf "%s\n" "PS1='\n[ \u@wamr-dev-docker \W ]\n$ '" >> /root/.bashrc \
+  && wget https://github.com/cli/cli/releases/download/v2.20.2/gh_2.20.2_linux_amd64.deb \
+  && dpkg -i gh_2.20.2_linux_amd64.deb
+
+#
+# Install NodeJS
+RUN curl -fsSL https://deb.nodesource.com/setup_19.x | bash -
+RUN apt-get install -y nodejs
+
+# set path
+ENV PATH="/opt/bazelisk:/usr/lib/llvm-${LLVM_VER}/bin:${PATH}"
+ENV CC=/usr/lib/llvm-${LLVM_VER}/bin/clang CXX=/usr/lib/llvm-${LLVM_VER}/bin/clang++
+RUN printf "%s\n" "PS1='\n[ \u@wamr-dev-docker \W ]\n$ '" >> /root/.bashrc \
   && apt-get autoremove -y \
   && apt-get clean -y \
   && rm -rf /var/lib/apt/lists/* \
   && rm -rf /tmp/*
 
 # set workdir when container run
-VOLUME /workspace
+VOLUME /workspaces
-WORKDIR /workspace
+WORKDIR /workspaces

.devcontainer/devcontainer.json

@@ -1,6 +1,5 @@
 // Copyright (C) 2019 Intel Corporation.  All rights reserved.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-
 // For format details, see https://aka.ms/vscode-remote/devcontainer.json or this file's README at:
 // https://github.com/microsoft/vscode-dev-containers/tree/v0.195.0/containers/cpp
 {
@@ -10,7 +9,12 @@
     // Update 'VARIANT' to pick an Debian / Ubuntu OS version: debian-11, debian-10, debian-9, ubuntu-21.04, ubuntu-20.04, ubuntu-18.04
     // Use Debian 11, Debian 9, Ubuntu 18.04 or Ubuntu 21.04 on local arm64/Apple Silicon
     "args": {
-      "VARIANT": "ubuntu-20.04"
+      "BINARYEN_VER": "111",
+      "EMSDK_VER": "3.0.0",
+      "LLVM_VER": "15",
+      "VARIANT": "ubuntu-20.04",
+      "WASI_SDK_VER": "19",
+      "WABT_VER": "1.0.31"
     }
   },
   "runArgs": [
@@ -27,12 +31,10 @@
       // Add the IDs of extensions you want installed when the container is created.
       "extensions": [
         "dtsvet.vscode-wasm",
-        "esbenp.prettier-vscode",
+        "llvm-vs-code-extensions.vscode-clangd",
         "ms-python.python",
         "ms-python.vscode-pylance",
         "ms-vscode.cmake-tools",
-        "ms-vscode.cpptools",
-        "twxs.cmake"
       ]
     }
   },

.github/workflows/build_iwasm_release.yml

@@ -36,7 +36,7 @@ jobs:
       - name: generate iwasm binary release
         run: |
           cmake -S . -B build \
-            -DWAMR_BUILD_AOT=1 -DWAMR_BUILD_INTERP=1 -DWAMR_BUILD_JIT=0 \
+            -DWAMR_BUILD_AOT=1 -DWAMR_BUILD_INTERP=1 -DWAMR_BUILD_FAST_JIT=1 -DWAMR_BUILD_JIT=1 \
             -DWAMR_BUILD_CUSTOM_NAME_SECTION=0 \
             -DWAMR_BUILD_DEBUG_INTERP=0 \
             -DWAMR_BUILD_DEBUG_AOT=0 \

.github/workflows/build_llvm_libraries.yml

@@ -5,23 +5,46 @@ name: Reusable workflow-build_llvm_libraries
 on:
   workflow_call:
     inputs:
-      runs-on:
+      os:
         required: true
         type: string
+      arch:
+        required: true
+        type: string
+    outputs:
+      cache_key:
+        description: "A cached key of LLVM libraries"
+        value: ${{ jobs.build_llvm_libraries.outputs.key}}
 
 jobs:
   build_llvm_libraries:
-    runs-on: ${{ matrix.os }}
+    runs-on: ${{ inputs.os }}
-    strategy:
+    outputs:
-      matrix:
+      key: ${{ steps.create_lib_cache_key.outputs.key}}
-        os: ${{ fromJson(inputs.runs-on) }}
 
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
+      - name: install dependencies
+        run: /usr/bin/env python3 -m pip install -r requirements.txt
+        working-directory: build-scripts
+
+      - name: retrive the last commit ID
+        id: get_last_commit
+        run: echo "last_commit=$(GH_TOKEN=${{ secrets.GITHUB_TOKEN }} /usr/bin/env python3 ./build_llvm.py --llvm-ver)" >> $GITHUB_OUTPUT
+        working-directory: build-scripts
+
+      # Bump the prefix number to evict all previous caches and
+      # enforce a clean build, in the unlikely case that some
+      # weird build error occur and llvm/build becomes a potential
+      # suspect.
+      - name: form the cache key of libraries
+        id: create_lib_cache_key
+        run: echo "key=0-llvm-libraries-${{ inputs.os }}-${{ inputs.arch }}-${{ steps.get_last_commit.outputs.last_commit }}" >> $GITHUB_OUTPUT
+
       - name: Cache LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -30,10 +53,39 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-build-llvm_libraries_ex
+          key: ${{ steps.create_lib_cache_key.outputs.key}}
 
-      - name: Build llvm
+      - uses: actions/cache@v3
-        id: build_llvm
+        with:
-        if: ${{ steps.cache_llvm.outputs.cache-hit != 'true' }} 
+          path: ~/.ccache
-        run: /usr/bin/env python3 ./build_llvm.py --arch X86 WebAssembly
+          key: 0-ccache-${{ inputs.os }}-${{ steps.get_last_commit.outputs.last_commit }}
+          restore-keys: |
+            0-ccache-${{ inputs.os }}
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true' && inputs.os == 'ubuntu-20.04'
+
+      - uses: actions/cache@v3
+        with:
+          path: ~/.cache/ccache
+          key: 0-ccache-${{ inputs.os }}-${{ steps.get_last_commit.outputs.last_commit }}
+          restore-keys: |
+            0-ccache-${{ inputs.os }}
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true' && inputs.os == 'ubuntu-22.04'
+
+      - run: sudo apt install -y ccache ninja-build
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true' && startsWith(inputs.os, 'ubuntu')
+
+      - uses: actions/cache@v3
+        with:
+          path: ~/Library/Caches/ccache
+          key: 0-ccache-${{ inputs.os }}-${{ steps.get_last_commit.outputs.last_commit }}
+          restore-keys: |
+            0-ccache-${{ inputs.os }}
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true' && startsWith(inputs.os, 'macos')
+
+      - run: brew install ccache ninja
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true' && startsWith(inputs.os, 'macos')
+
+      - name: Build LLVM libraries
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
+        run: /usr/bin/env python3 ./build_llvm.py --arch ${{ inputs.arch }}
         working-directory: build-scripts

.github/workflows/coding_guidelines.yml

@@ -15,7 +15,7 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  complinace_job:
+  compliance_job:
     runs-on: ubuntu-latest
     steps:
       - name: checkout

.github/workflows/compilation_on_android_ubuntu.yml

@@ -53,8 +53,7 @@ env:
   FAST_JIT_BUILD_OPTIONS: "      -DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_FAST_JIT=1 -DWAMR_BUILD_JIT=0 -DWAMR_BUILD_LAZY_JIT=0"
   LLVM_LAZY_JIT_BUILD_OPTIONS: " -DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_FAST_JIT=0 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=1"
   LLVM_EAGER_JIT_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_FAST_JIT=0 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=0"
-  # LLVM
+  MULTI_TIER_JIT_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=1 -DWAMR_BUILD_FAST_JIT=1 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=1"
-  LLVM_CACHE_SUFFIX: "build-llvm_libraries_ex"
   # For Spec Test
   DEFAULT_TEST_OPTIONS: "-s spec -b -P"
   MULTI_MODULES_TEST_OPTIONS: "-s spec -b -M -P"
@@ -64,23 +63,37 @@ env:
   WASI_TEST_OPTIONS: "-s wasi_certification"
 
 jobs:
-  build_llvm_libraries:
+  build_llvm_libraries_on_ubuntu_2004:
     uses: ./.github/workflows/build_llvm_libraries.yml
     with:
-      runs-on: "['ubuntu-20.04', 'ubuntu-22.04']"
+      os: "ubuntu-20.04"
+      arch: "X86"
+
+  build_llvm_libraries_on_ubuntu_2204:
+    uses: ./.github/workflows/build_llvm_libraries.yml
+    with:
+      os: "ubuntu-22.04"
+      arch: "X86"
 
   build_wamrc:
-    needs: [build_llvm_libraries]
+    needs:
+      [build_llvm_libraries_on_ubuntu_2004, build_llvm_libraries_on_ubuntu_2204]
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-20.04, ubuntu-22.04]
+        include:
+          - os: ubuntu-20.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2004.outputs.cache_key }}
+          - os: ubuntu-22.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2204.outputs.cache_key }}
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
+      # since jobs.id can't contain the dot character
+      # it is hard to use `format` to assemble the cache key
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -89,10 +102,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: steps.cache_llvm.outputs.cache-hit != 'true'
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: Build wamrc
@@ -103,7 +116,8 @@ jobs:
         working-directory: wamr-compiler
 
   build_iwasm:
-    needs: [build_llvm_libraries]
+    needs:
+      [build_llvm_libraries_on_ubuntu_2004, build_llvm_libraries_on_ubuntu_2204]
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
@@ -115,6 +129,7 @@ jobs:
             $FAST_JIT_BUILD_OPTIONS,
             $LLVM_LAZY_JIT_BUILD_OPTIONS,
             $LLVM_EAGER_JIT_BUILD_OPTIONS,
+            $MULTI_TIER_JIT_BUILD_OPTIONS,
           ]
         make_options_feature: [
             # Features
@@ -147,6 +162,8 @@ jobs:
             make_options_feature: "-DWAMR_BUILD_MULTI_MODULE=1"
           - make_options_run_mode: $LLVM_EAGER_JIT_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_MULTI_MODULE=1"
+          - make_options_run_mode: $MULTI_TIER_JIT_BUILD_OPTIONS
+            make_options_feature: "-DWAMR_BUILD_MULTI_MODULE=1"
           # SIMD only on JIT/AOT mode
           - make_options_run_mode: $CLASSIC_INTERP_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_SIMD=1"
@@ -163,6 +180,8 @@ jobs:
             make_options_feature: "-DWAMR_BUILD_DEBUG_INTERP=1"
           - make_options_run_mode: $LLVM_EAGER_JIT_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_DEBUG_INTERP=1"
+          - make_options_run_mode: $MULTI_TIER_JIT_BUILD_OPTIONS
+            make_options_feature: "-DWAMR_BUILD_DEBUG_INTERP=1"
           # DEBUG_AOT only on JIT/AOT mode
           - make_options_run_mode: $CLASSIC_INTERP_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_DEBUG_AOT=1"
@@ -175,6 +194,8 @@ jobs:
             make_options_feature: "-DWAMR_BUILD_DEBUG_AOT=1"
           - make_options_run_mode: $LLVM_EAGER_JIT_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_DEBUG_AOT=1"
+          - make_options_run_mode: $MULTI_TIER_JIT_BUILD_OPTIONS
+            make_options_feature: "-DWAMR_BUILD_DEBUG_AOT=1"
           # MINI_LOADER only on INTERP mode
           - make_options_run_mode: $AOT_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_MINI_LOADER=1"
@@ -184,16 +205,28 @@ jobs:
             make_options_feature: "-DWAMR_BUILD_MINI_LOADER=1"
           - make_options_run_mode: $LLVM_EAGER_JIT_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_MINI_LOADER=1"
-          # Fast-JIT mode doesn't support android(X86-32)
+          - make_options_run_mode: $MULTI_TIER_JIT_BUILD_OPTIONS
+            make_options_feature: "-DWAMR_BUILD_MINI_LOADER=1"
+          # Fast-JIT and Multi-Tier-JIT mode don't support android(X86-32)
           - make_options_run_mode: $FAST_JIT_BUILD_OPTIONS
             platform: android
+          - make_options_run_mode: $MULTI_TIER_JIT_BUILD_OPTIONS
+            platform: android
+          # only test andorid on ubuntu latest
+          - os: ubuntu-20.04
+            platform: android
+        include:
+          - os: ubuntu-20.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2004.outputs.cache_key }}
+          - os: ubuntu-22.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2204.outputs.cache_key }}
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
       # only download llvm cache when needed
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         if: endsWith(matrix.make_options_run_mode, '_JIT_BUILD_OPTIONS')
         uses: actions/cache@v3
         with:
@@ -203,10 +236,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: endsWith(matrix.make_options_run_mode, '_JIT_BUILD_OPTIONS') && (steps.cache_llvm.outputs.cache-hit != 'true')
+        if: endsWith(matrix.make_options_run_mode, '_JIT_BUILD_OPTIONS') && (steps.retrieve_llvm_libs.outputs.cache-hit != 'true')
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: Build iwasm
@@ -217,7 +250,13 @@ jobs:
         working-directory: product-mini/platforms/${{ matrix.platform }}
 
   build_samples_wasm_c_api:
-    needs: [build_iwasm, build_llvm_libraries, build_wamrc]
+    needs:
+      [
+        build_iwasm,
+        build_llvm_libraries_on_ubuntu_2004,
+        build_llvm_libraries_on_ubuntu_2204,
+        build_wamrc,
+      ]
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
@@ -229,22 +268,28 @@ jobs:
             $FAST_JIT_BUILD_OPTIONS,
             $LLVM_LAZY_JIT_BUILD_OPTIONS,
             $LLVM_EAGER_JIT_BUILD_OPTIONS,
+            $MULTI_TIER_JIT_BUILD_OPTIONS,
           ]
         os: [ubuntu-20.04, ubuntu-22.04]
         wasi_sdk_release:
           [
-            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-16/wasi-sdk-16.0-linux.tar.gz",
+            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz",
           ]
         wabt_release:
           [
-            "https://github.com/WebAssembly/wabt/releases/download/1.0.24/wabt-1.0.24-ubuntu.tar.gz",
+            "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-ubuntu.tar.gz",
           ]
+        include:
+          - os: ubuntu-20.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2004.outputs.cache_key }}
+          - os: ubuntu-22.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2204.outputs.cache_key }}
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         if: (!endsWith(matrix.make_options, '_INTERP_BUILD_OPTIONS'))
         uses: actions/cache@v3
         with:
@@ -254,18 +299,18 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: (!endsWith(matrix.make_options, '_INTERP_BUILD_OPTIONS')) && (steps.cache_llvm.outputs.cache-hit != 'true')
+        if: (!endsWith(matrix.make_options, '_INTERP_BUILD_OPTIONS')) && (steps.retrieve_llvm_libs.outputs.cache-hit != 'true')
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: download and install wabt
         run: |
           cd /opt
           sudo wget ${{ matrix.wabt_release }}
-          sudo tar -xzf wabt-1.0.24-*.tar.gz
+          sudo tar -xzf wabt-1.0.31-*.tar.gz
-          sudo mv wabt-1.0.24 wabt
+          sudo mv wabt-1.0.31 wabt
 
       - name: Build wamrc
         if: (!endsWith(matrix.make_options, '_INTERP_BUILD_OPTIONS'))
@@ -277,19 +322,9 @@ jobs:
 
       - name: Build Sample [wasm-c-api]
         run: |
-          mkdir build && cd build
+          cmake -S . -B build ${{ matrix.make_options }}
-          cmake .. ${{ matrix.make_options }}
+          cmake --build build --config Release --parallel 4
-          cmake --build . --config Release --parallel 4
+          ctest --test-dir build
-          ./callback
-          ./callback_chain
-          ./empty_imports
-          ./global
-          ./hello
-          ./hostref
-          ./memory
-          ./reflect
-          ./table
-          ./trap
         working-directory: samples/wasm-c-api
 
   build_samples_others:
@@ -298,14 +333,13 @@ jobs:
     strategy:
       matrix:
         os: [ubuntu-20.04, ubuntu-22.04]
-        wasi_sdk_release:
+        include:
-          [
+          - os: ubuntu-20.04
-            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-16/wasi-sdk-16.0-linux.tar.gz",
+            wasi_sdk_release: "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz"
-          ]
+            wabt_release: "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-ubuntu.tar.gz"
-        wabt_release:
+          - os: ubuntu-22.04
-          [
+            wasi_sdk_release: "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz"
-            "https://github.com/WebAssembly/wabt/releases/download/1.0.24/wabt-1.0.24-ubuntu.tar.gz",
+            wabt_release: "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-ubuntu.tar.gz"
-          ]
     steps:
       - name: checkout
         uses: actions/checkout@v3
@@ -314,15 +348,15 @@ jobs:
         run: |
           cd /opt
           sudo wget ${{ matrix.wasi_sdk_release }}
-          sudo tar -xzf wasi-sdk-16.0-*.tar.gz
+          sudo tar -xzf wasi-sdk-*.tar.gz
-          sudo mv wasi-sdk-16.0 wasi-sdk
+          sudo mv wasi-sdk-19.0 wasi-sdk
 
       - name: download and install wabt
         run: |
           cd /opt
           sudo wget ${{ matrix.wabt_release }}
-          sudo tar -xzf wabt-1.0.24-*.tar.gz
+          sudo tar -xzf wabt-1.0.31-*.tar.gz
-          sudo mv wabt-1.0.24 wabt
+          sudo mv wabt-1.0.31 wabt
 
       - name: build wasi-libc (needed for wasi-threads)
         run: |
@@ -402,12 +436,19 @@ jobs:
           ./iwasm wasm-apps/no_pthread.wasm
 
   test:
-    needs: [build_iwasm, build_llvm_libraries, build_wamrc]
+    needs: [build_iwasm, build_llvm_libraries_on_ubuntu_2004, build_wamrc]
     runs-on: ubuntu-20.04
     strategy:
       matrix:
         running_mode:
-          ["classic-interp", "fast-interp", "jit", "aot", "fast-jit"]
+          [
+            "classic-interp",
+            "fast-interp",
+            "jit",
+            "aot",
+            "fast-jit",
+            "multi-tier-jit",
+          ]
         test_option:
           [
             $DEFAULT_TEST_OPTIONS,
@@ -416,6 +457,8 @@ jobs:
             $THREADS_TEST_OPTIONS,
             $WASI_TEST_OPTIONS,
           ]
+        llvm_cache_key:
+          ["${{ needs.build_llvm_libraries_on_ubuntu_2004.outputs.cache_key }}"]
         exclude:
           # uncompatiable modes and features
           # classic-interp and fast-interp don't support simd
@@ -431,32 +474,39 @@ jobs:
             test_option: $WASI_TEST_OPTIONS
           - running_mode: "jit"
             test_option: $MULTI_MODULES_TEST_OPTIONS
-          # fast-jit is only tested on default mode, exclude other three
+          # fast-jit doesn't support multi module, simd, and threads
           - running_mode: "fast-jit"
             test_option: $MULTI_MODULES_TEST_OPTIONS
           - running_mode: "fast-jit"
             test_option: $SIMD_TEST_OPTIONS
           - running_mode: "fast-jit"
             test_option: $THREADS_TEST_OPTIONS
+          # multi-tier-jit doesn't support multi module, simd, and threads
+          - running_mode: "multi-tier-jit"
+            test_option: $MULTI_MODULES_TEST_OPTIONS
+          - running_mode: "multi-tier-jit"
+            test_option: $SIMD_TEST_OPTIONS
+          - running_mode: "multi-tier-jit"
+            test_option: $THREADS_TEST_OPTIONS
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
       - name: set env variable(if llvm are used)
-        if: matrix.running_mode == 'aot' || matrix.running_mode == 'jit'
+        if: matrix.running_mode == 'aot' || matrix.running_mode == 'jit' || matrix.running_mode == 'multi-tier-jit'
         run: echo "USE_LLVM=true" >> $GITHUB_ENV
 
       - name: set env variable(if x86_32 test needed)
         if: >
           (matrix.test_option == '$DEFAULT_TEST_OPTIONS' || matrix.test_option == '$THREADS_TEST_OPTIONS'
            || matrix.test_option == '$WASI_TEST_OPTIONS')
-          && matrix.running_mode != 'fast-jit' && matrix.running_mode != 'jit'
+          && matrix.running_mode != 'fast-jit' && matrix.running_mode != 'jit' && matrix.running_mode != 'multi-tier-jit'
         run: echo "TEST_ON_X86_32=true" >> $GITHUB_ENV
 
       #only download llvm libraries in jit and aot mode
       - name: Get LLVM libraries
         if: env.USE_LLVM == 'true'
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -465,10 +515,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ubuntu-20.04-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: env.USE_LLVM == 'true' && steps.cache_llvm.outputs.cache-hit != 'true'
+        if: env.USE_LLVM == 'true' && steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: run tests

.github/workflows/compilation_on_macos.yml

@@ -51,26 +51,28 @@ env:
   FAST_INTERP_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=0 -DWAMR_BUILD_FAST_INTERP=1 -DWAMR_BUILD_INTERP=1 -DWAMR_BUILD_JIT=0 -DWAMR_BUILD_LAZY_JIT=0"
   LLVM_LAZY_JIT_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=1"
   LLVM_EAGER_JIT_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=0"
-  LLVM_CACHE_SUFFIX: "build-llvm_libraries_ex"
 
 jobs:
   build_llvm_libraries:
     uses: ./.github/workflows/build_llvm_libraries.yml
     with:
-      runs-on: "['macos-latest']"
+      os: "macos-latest"
+      arch: "X86"
 
   build_wamrc:
     needs: [build_llvm_libraries]
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [macos-latest]
+        include:
+          - os: macos-latest
+            llvm_cache_key: ${{ needs.build_llvm_libraries.outputs.cache_key }}
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -79,10 +81,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: steps.cache_llvm.outputs.cache-hit != 'true'
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: Build wamrc
@@ -166,13 +168,16 @@ jobs:
             make_options_feature: "-DWAMR_BUILD_MINI_LOADER=1"
           - make_options_run_mode: $LLVM_EAGER_JIT_BUILD_OPTIONS
             make_options_feature: "-DWAMR_BUILD_MINI_LOADER=1"
+        include:
+          - os: macos-latest
+            llvm_cache_key: ${{ needs.build_llvm_libraries.outputs.cache_key }}
     steps:
       - name: checkout
         uses: actions/checkout@v3
 
       # only download llvm cache when needed
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         if: endsWith(matrix.make_options_run_mode, '_JIT_BUILD_OPTIONS')
         uses: actions/cache@v3
         with:
@@ -182,10 +187,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: endsWith(matrix.make_options_run_mode, '_JIT_BUILD_OPTIONS') && (steps.cache_llvm.outputs.cache-hit != 'true')
+        if: endsWith(matrix.make_options_run_mode, '_JIT_BUILD_OPTIONS') && (steps.retrieve_llvm_libs.outputs.cache-hit != 'true')
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: Build iwasm
@@ -210,8 +215,14 @@ jobs:
             #$AOT_BUILD_OPTIONS,
           ]
         os: [macos-latest]
-        wasi_sdk_release: ["https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-16/wasi-sdk-16.0-macos.tar.gz"]
+        wasi_sdk_release:
-        wabt_release: ["https://github.com/WebAssembly/wabt/releases/download/1.0.24/wabt-1.0.24-macos.tar.gz"]
+          [
+            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-macos.tar.gz",
+          ]
+        wabt_release:
+          [
+            "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-macos-12.tar.gz",
+          ]
     steps:
       - name: checkout
         uses: actions/checkout@v3
@@ -220,24 +231,14 @@ jobs:
         run: |
           cd /opt
           sudo wget ${{ matrix.wabt_release }}
-          sudo tar -xzf wabt-1.0.24-*.tar.gz
+          sudo tar -xzf wabt-1.0.31-*.tar.gz
-          sudo mv wabt-1.0.24 wabt
+          sudo mv wabt-1.0.31 wabt
 
       - name: Build Sample [wasm-c-api]
         run: |
-          mkdir build && cd build
+          cmake -S . -B build ${{ matrix.make_options }}
-          cmake .. ${{ matrix.make_options }}
+          cmake --build build --config Release --parallel 4
-          cmake --build . --config Release --parallel 4
+          ctest --test-dir build
-          ./callback
-          ./callback_chain
-          ./empty_imports
-          ./global
-          ./hello
-          ./hostref
-          ./memory
-          ./reflect
-          ./table
-          ./trap
         working-directory: samples/wasm-c-api
 
   build_samples_others:
@@ -246,8 +247,14 @@ jobs:
     strategy:
       matrix:
         os: [macos-latest]
-        wasi_sdk_release: ["https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-16/wasi-sdk-16.0-macos.tar.gz"]
+        wasi_sdk_release:
-        wabt_release: ["https://github.com/WebAssembly/wabt/releases/download/1.0.24/wabt-1.0.24-macos.tar.gz"]
+          [
+            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-macos.tar.gz",
+          ]
+        wabt_release:
+          [
+            "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-macos-12.tar.gz",
+          ]
     steps:
       - name: checkout
         uses: actions/checkout@v3
@@ -256,15 +263,15 @@ jobs:
         run: |
           cd /opt
           sudo wget ${{ matrix.wasi_sdk_release }}
-          sudo tar -xzf wasi-sdk-16.0-*.tar.gz
+          sudo tar -xzf wasi-sdk-*.tar.gz
-          sudo mv wasi-sdk-16.0 wasi-sdk
+          sudo mv wasi-sdk-19.0 wasi-sdk
 
       - name: download and install wabt
         run: |
           cd /opt
           sudo wget ${{ matrix.wabt_release }}
-          sudo tar -xzf wabt-1.0.24-*.tar.gz
+          sudo tar -xzf wabt-1.0.31-*.tar.gz
-          sudo mv wabt-1.0.24 wabt
+          sudo mv wabt-1.0.31 wabt
 
       - name: build wasi-libc (needed for wasi-threads)
         run: |

.github/workflows/compilation_on_nuttx.yml

@@ -65,9 +65,12 @@ jobs:
           "boards/risc-v/k210/maix-bit/configs/nsh",
         ]
         wamr_config_option: [
-          "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_FAST=y\\nCONFIG_INTERPRETERS_WAMR_LIBC_BUILTIN=y\\n",
           "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_FAST=y\\n",
+          "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_FAST=y\\nCONFIG_INTERPRETERS_WAMR_LIBC_WASI=y\\n",
+          "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_FAST=y\\nCONFIG_INTERPRETERS_WAMR_LIBC_BUILTIN=y\\n",
           "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_CLASSIC=y\\n",
+          "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_CLASSIC=y\\nCONFIG_INTERPRETERS_WAMR_LIBC_WASI=y\\n",
+          "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_CLASSIC=y\\nCONFIG_INTERPRETERS_WAMR_LIBC_BUILTIN=y\\n",
           "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\nCONFIG_INTERPRETERS_WAMR_LIBC_BUILTIN=y\\n",
           "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_AOT=y\\n",
           "CONFIG_INTERPRETERS_WAMR=y\\nCONFIG_INTERPRETERS_WAMR_FAST=y\\n",
@@ -112,7 +115,7 @@ jobs:
 
       - name: Enable WAMR for NuttX
         run: |
-          find nuttx/boards -name defconfig | xargs sed -i '$a\CONFIG_EOL_IS_LF=y\n${{ matrix.wamr_config_option }}'
+          find nuttx/boards -name defconfig | xargs sed -i '$a\CONFIG_EOL_IS_LF=y\nCONFIG_PSEUDOFS_SOFTLINKS=y\n${{ matrix.wamr_config_option }}'
           find nuttx/boards/sim -name defconfig | xargs sed -i '$a\CONFIG_LIBM=y\n'
 
       - name: Build

.github/workflows/compilation_on_sgx.yml

@@ -51,13 +51,13 @@ env:
   FAST_INTERP_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=0 -DWAMR_BUILD_FAST_INTERP=1 -DWAMR_BUILD_INTERP=1 -DWAMR_BUILD_JIT=0 -DWAMR_BUILD_LAZY_JIT=0"
   LLVM_LAZY_JIT_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=1"
   LLVM_EAGER_JIT_BUILD_OPTIONS: "-DWAMR_BUILD_AOT=1 -DWAMR_BUILD_FAST_INTERP=0 -DWAMR_BUILD_INTERP=0 -DWAMR_BUILD_JIT=1 -DWAMR_BUILD_LAZY_JIT=0"
-  LLVM_CACHE_SUFFIX: "build-llvm_libraries_ex"
 
 jobs:
   build_llvm_libraries:
     uses: ./.github/workflows/build_llvm_libraries.yml
     with:
-      runs-on: "['ubuntu-20.04']"
+      os: "ubuntu-20.04"
+      arch: "X86"
 
   build_iwasm:
     runs-on: ${{ matrix.os }}
@@ -131,7 +131,9 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-20.04]
+        include:
+          - os: ubuntu-20.04
+            llvm_cache_key: ${{ needs.build_llvm_libraries.outputs.cache_key }}
     steps:
       - name: install SGX SDK and necessary libraries
         run: |
@@ -150,7 +152,7 @@ jobs:
         uses: actions/checkout@v3
 
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -159,10 +161,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: steps.cache_llvm.outputs.cache-hit != 'true'
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: Build wamrc
@@ -189,11 +191,11 @@ jobs:
         os: [ubuntu-20.04]
         wasi_sdk_release:
           [
-            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-12/wasi-sdk-12.0-linux.tar.gz",
+            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz",
           ]
         wabt_release:
           [
-            "https://github.com/WebAssembly/wabt/releases/download/1.0.24/wabt-1.0.24-ubuntu.tar.gz",
+            "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-ubuntu.tar.gz",
           ]
     steps:
       - name: checkout
@@ -203,8 +205,8 @@ jobs:
         run: |
           cd /opt
           sudo wget ${{ matrix.wabt_release }}
-          sudo tar -xzf wabt-1.0.24-*.tar.gz
+          sudo tar -xzf wabt-1.0.31-*.tar.gz
-          sudo mv wabt-1.0.24 wabt
+          sudo mv wabt-1.0.31 wabt
 
       - name: install SGX SDK and necessary libraries
         run: |
@@ -221,19 +223,9 @@ jobs:
 
       - name: Build Sample [wasm-c-api]
         run: |
-          mkdir build && cd build
+          cmake -S . -B build ${{ matrix.make_options }}
-          cmake .. ${{ matrix.make_options }}
+          cmake --build build --config Release --parallel 4
-          cmake --build . --config Release --parallel 4
+          ctest --test-dir build
-          ./callback
-          ./callback_chain
-          ./empty_imports
-          ./global
-          ./hello
-          ./hostref
-          ./memory
-          ./reflect
-          ./table
-          ./trap
         working-directory: samples/wasm-c-api
 
   build_samples_others:
@@ -244,11 +236,11 @@ jobs:
         os: [ubuntu-20.04]
         wasi_sdk_release:
           [
-            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-16/wasi-sdk-16.0-linux.tar.gz",
+            "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz",
           ]
         wabt_release:
           [
-            "https://github.com/WebAssembly/wabt/releases/download/1.0.24/wabt-1.0.24-ubuntu.tar.gz",
+            "https://github.com/WebAssembly/wabt/releases/download/1.0.31/wabt-1.0.31-ubuntu.tar.gz",
           ]
     steps:
       - name: checkout
@@ -258,15 +250,15 @@ jobs:
         run: |
           cd /opt
           sudo wget ${{ matrix.wasi_sdk_release }}
-          sudo tar -xzf wasi-sdk-16.0-*.tar.gz
+          sudo tar -xzf wasi-sdk-*.tar.gz
-          sudo mv wasi-sdk-16.0 wasi-sdk
+          sudo mv wasi-sdk-19.0 wasi-sdk
 
       - name: download and install wabt
         run: |
           cd /opt
           sudo wget ${{ matrix.wabt_release }}
-          sudo tar -xzf wabt-1.0.24-*.tar.gz
+          sudo tar -xzf wabt-1.0.31-*.tar.gz
-          sudo mv wabt-1.0.24 wabt
+          sudo mv wabt-1.0.31 wabt
 
       - name: build wasi-libc (needed for wasi-threads)
         run: |
@@ -358,6 +350,7 @@ jobs:
       matrix:
         running_mode: ["classic-interp", "fast-interp", "aot"]
         test_option: ["-x -p -s spec -b -P", "-x -p -s spec -S -b -P"]
+        llvm_cache_key: ["${{ needs.build_llvm_libraries.outputs.cache_key }}"]
         # classic-interp and fast-interp don't support simd
         exclude:
           - running_mode: "classic-interp"
@@ -371,7 +364,7 @@ jobs:
 
       - name: Get LLVM libraries
         if: matrix.running_mode == 'aot'
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -380,10 +373,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key: ubuntu-20.04-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: matrix.running_mode == 'aot' && steps.cache_llvm.outputs.cache-hit != 'true'
+        if: matrix.running_mode == 'aot' && steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: install SGX SDK and necessary libraries

.github/workflows/release_process.yml

@@ -123,7 +123,7 @@ jobs:
       runner: ubuntu-20.04
       upload_url: ${{ needs.create_release.outputs.upload_url }}
       ver_num: ${{ needs.create_tag.outputs.new_ver}}
-      wasi_sdk_url: https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-12/wasi-sdk-12.0-linux.tar.gz
+      wasi_sdk_url: https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz
 
   release_wamr_sdk_on_ubuntu_2204:
     needs: [create_tag, create_release]
@@ -133,7 +133,7 @@ jobs:
       runner: ubuntu-22.04
       upload_url: ${{ needs.create_release.outputs.upload_url }}
       ver_num: ${{ needs.create_tag.outputs.new_ver}}
-      wasi_sdk_url: https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-12/wasi-sdk-12.0-linux.tar.gz
+      wasi_sdk_url: https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-linux.tar.gz
 
   release_wamr_sdk_on_macos:
     needs: [create_tag, create_release]
@@ -143,7 +143,7 @@ jobs:
       runner: macos-latest
       upload_url: ${{ needs.create_release.outputs.upload_url }}
       ver_num: ${{ needs.create_tag.outputs.new_ver}}
-      wasi_sdk_url: https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-12/wasi-sdk-12.0-macos.tar.gz
+      wasi_sdk_url: https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-19/wasi-sdk-19.0-macos.tar.gz
 
   #
   # vscode extension cross-platform

.github/workflows/spec_test_on_nuttx.yml

@@ -16,7 +16,8 @@ jobs:
   build_llvm_libraries:
     uses: ./.github/workflows/build_llvm_libraries.yml
     with:
-      runs-on: "['ubuntu-22.04']"
+      os: "ubuntu-22.04"
+      arch: "ARM RISCV AArch64"
 
   spec_test_on_qemu:
     runs-on: ${{ matrix.os }}
@@ -37,6 +38,7 @@ jobs:
           "-t aot",
           "-t aot -X"
         ]
+        llvm_cache_key: [ "${{ needs.build_llvm_libraries.outputs.cache_key }}" ]
     steps:
       - name: Install Utilities
         run: |
@@ -72,7 +74,7 @@ jobs:
           path: apps/interpreters/wamr/wamr
 
       - name: Get LLVM libraries
-        id: cache_llvm
+        id: retrieve_llvm_libs
         uses: actions/cache@v3
         with:
           path: |
@@ -81,10 +83,10 @@ jobs:
             ./core/deps/llvm/build/lib
             ./core/deps/llvm/build/libexec
             ./core/deps/llvm/build/share
-          key:  ${{ matrix.os }}-${{ env.LLVM_CACHE_SUFFIX }}
+          key: ${{ matrix.llvm_cache_key }}
 
       - name: Quit if cache miss
-        if: steps.cache_llvm.outputs.cache-hit != 'true'
+        if: steps.retrieve_llvm_libs.outputs.cache-hit != 'true'
         run: echo "::error::can not get prebuilt llvm libraries" && exit 1
 
       - name: Copy LLVM

.gitignore

@@ -32,3 +32,6 @@ samples/socket-api/wasm-src/inc/pthread.h
 **/__pycache__
 
 tests/benchmarks/coremark/coremark*
+
+samples/workload/include/**
+!samples/workload/include/.gitkeep

build-scripts/build_llvm.py

@@ -7,6 +7,7 @@
 import argparse
 import os
 import pathlib
+import requests
 import shlex
 import shutil
 import subprocess
@@ -21,28 +22,43 @@ def clone_llvm(dst_dir, llvm_repo, llvm_branch):
     llvm_dir = dst_dir.joinpath("llvm").resolve()
 
     if not llvm_dir.exists():
-        print(f"Clone llvm to {llvm_dir} ...")
         GIT_CLONE_CMD = f"git clone --depth 1 --branch {llvm_branch} {llvm_repo} llvm"
         subprocess.check_output(shlex.split(GIT_CLONE_CMD), cwd=dst_dir)
-    else:
-        print(f"There is an LLVM local repo in {llvm_dir}, clean and keep using it")
 
     return llvm_dir
 
 
-def build_llvm(llvm_dir, platform, backends, projects):
+def query_llvm_version(llvm_info):
+    github_token = os.environ['GH_TOKEN']
+    owner_project = llvm_info['repo'].replace("https://github.com/", "").replace(".git", "")
+    url = f"https://api.github.com/repos/{owner_project}/commits/{llvm_info['branch']}"
+    headers = {
+        'Authorization': f"Bearer {github_token}"
+    }
+
+    try:
+        response = requests.request("GET", url, headers=headers, data={})
+        response.raise_for_status()
+    except requests.exceptions.HTTPError as error:
+        print (error) # for debugging purpose
+        return None
+
+    response = response.json()
+    return response['sha']
+
+
+def build_llvm(llvm_dir, platform, backends, projects, use_clang=False):
     LLVM_COMPILE_OPTIONS = [
         '-DCMAKE_BUILD_TYPE:STRING="Release"',
         "-DCMAKE_EXPORT_COMPILE_COMMANDS=ON",
         "-DLLVM_APPEND_VC_REV:BOOL=ON",
-        "-DLLVM_BUILD_BENCHMARKS:BOOL=OFF",
-        "-DLLVM_BUILD_DOCS:BOOL=OFF",
         "-DLLVM_BUILD_EXAMPLES:BOOL=OFF",
         "-DLLVM_BUILD_LLVM_DYLIB:BOOL=OFF",
         "-DLLVM_BUILD_TESTS:BOOL=OFF",
-        "-DLLVM_CCACHE_BUILD:BOOL=OFF",
+        "-DLLVM_CCACHE_BUILD:BOOL=ON",
         "-DLLVM_ENABLE_BINDINGS:BOOL=OFF",
         "-DLLVM_ENABLE_IDE:BOOL=OFF",
+        "-DLLVM_ENABLE_LIBEDIT=OFF",
         "-DLLVM_ENABLE_TERMINFO:BOOL=OFF",
         "-DLLVM_ENABLE_ZLIB:BOOL=OFF",
         "-DLLVM_INCLUDE_BENCHMARKS:BOOL=OFF",
@@ -54,6 +70,18 @@ def build_llvm(llvm_dir, platform, backends, projects):
         "-DLLVM_OPTIMIZED_TABLEGEN:BOOL=ON",
     ]
 
+    # use clang/clang++/lld. but macos doesn't support lld
+    if not sys.platform.startswith("darwin") and use_clang:
+        if shutil.which("clang") and shutil.which("clang++") and shutil.which("lld"):
+            os.environ["CC"] = "clang"
+            os.environ["CXX"] = "clang++"
+            LLVM_COMPILE_OPTIONS.append('-DLLVM_USE_LINKER:STRING="lld"')
+            print("Use the clang toolchain")
+        else:
+            print("Can not find clang, clang++ and lld, keep using the gcc toolchain")
+    else:
+        print("Use the gcc toolchain")
+
     LLVM_EXTRA_COMPILE_OPTIONS = {
         "arc": [
             '-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD:STRING="ARC"',
@@ -99,8 +127,10 @@ def build_llvm(llvm_dir, platform, backends, projects):
 
     lib_llvm_core_library = build_dir.joinpath("lib/libLLVMCore.a").resolve()
     if lib_llvm_core_library.exists():
-        print(f"Please remove {build_dir} manually and try again")
+        print(
-        return build_dir
+            f"It has already been fully compiled. If want to a re-build, please remove {build_dir} manually and try again"
+        )
+        return None
 
     compile_options = " ".join(
         LLVM_COMPILE_OPTIONS
@@ -119,10 +149,11 @@ def build_llvm(llvm_dir, platform, backends, projects):
             CONFIG_CMD += " -G'Unix Makefiles'"
         else:
             CONFIG_CMD += " -A x64"
-    print(f"{CONFIG_CMD}")
+    else:
+        CONFIG_CMD += " -G'Ninja'"
     subprocess.check_call(shlex.split(CONFIG_CMD), cwd=build_dir)
 
-    BUILD_CMD = f"cmake --build . --target package --parallel {os.cpu_count()}" + (
+    BUILD_CMD = "cmake --build . --target package" + (
         " --config Release" if "windows" == platform else ""
     )
     subprocess.check_call(shlex.split(BUILD_CMD), cwd=build_dir)
@@ -133,23 +164,25 @@ def build_llvm(llvm_dir, platform, backends, projects):
 def repackage_llvm(llvm_dir):
     build_dir = llvm_dir.joinpath("./build").resolve()
 
-    packs = [f for f in build_dir.glob("LLVM-13*.tar.gz")]
+    packs = [f for f in build_dir.glob("LLVM-*.tar.gz")]
     if len(packs) > 1:
-        raise Exception("Find more than one LLVM-13*.tar.gz")
+        raise Exception("Find more than one LLVM-*.tar.gz")
 
     if not packs:
         return
 
     llvm_package = packs[0].name
-    # mv build/LLVM-13.0.0*.gz .
+    # mv build/LLVM-*.gz .
     shutil.move(str(build_dir.joinpath(llvm_package).resolve()), str(llvm_dir))
     # rm -r build
     shutil.rmtree(str(build_dir))
     # mkdir build
     build_dir.mkdir()
-    # tar xf ./LLVM-13.0.0-*.tar.gz --strip-components=1 --directory=build
+    # tar xf ./LLVM-*.tar.gz --strip-components=1 --directory=build
     CMD = f"tar xf {llvm_dir.joinpath(llvm_package).resolve()} --strip-components=1 --directory={build_dir}"
     subprocess.check_call(shlex.split(CMD), cwd=llvm_dir)
+    # rm ./LLVM-1*.gz
+    os.remove(llvm_dir.joinpath(llvm_package).resolve())
 
 
 def main():
@@ -184,8 +217,17 @@ def main():
         choices=["clang", "lldb"],
         help="identify extra LLVM projects, separate by space, like '--project clang lldb'",
     )
+    parser.add_argument(
+        "--llvm-ver",
+        action="store_true",
+        help="return the version info of generated llvm libraries",
+    )
+    parser.add_argument(
+        "--use-clang",
+        action="store_true",
+        help="use clang instead of gcc",
+    )
     options = parser.parse_args()
-    print(f"options={options}")
 
     # if the "platform" is not identified in the command line option,
     # detect it
@@ -199,12 +241,10 @@ def main():
     else:
         platform = options.platform
 
-    print(f"========== Build LLVM for {platform} ==========\n")
-
     llvm_repo_and_branch = {
         "arc": {
             "repo": "https://github.com/llvm/llvm-project.git",
-            "branch": "release/13.x",
+            "branch": "release/15.x",
         },
         "xtensa": {
             "repo": "https://github.com/espressif/llvm-project.git",
@@ -212,7 +252,7 @@ def main():
         },
         "default": {
             "repo": "https://github.com/llvm/llvm-project.git",
-            "branch": "release/13.x",
+            "branch": "release/15.x",
         },
     }
 
@@ -225,19 +265,22 @@ def main():
     deps_dir = current_dir.joinpath("../core/deps").resolve()
 
     try:
-        print(f"==================== CLONE LLVM ====================")
         llvm_info = llvm_repo_and_branch.get(platform, llvm_repo_and_branch["default"])
+
+        if options.llvm_ver:
+            commit_hash = query_llvm_version(llvm_info)
+            print(commit_hash)
+            return commit_hash is not None
+        
         llvm_dir = clone_llvm(deps_dir, llvm_info["repo"], llvm_info["branch"])
+        if (
+            build_llvm(
+                llvm_dir, platform, options.arch, options.project, options.use_clang
+            )
+            is not None
+        ):
+            repackage_llvm(llvm_dir)
 
-        print()
-        print(f"==================== BUILD LLVM ====================")
-        build_llvm(llvm_dir, platform, options.arch, options.project)
-
-        print()
-        print(f"==================== PACKAGE LLVM ====================")
-        repackage_llvm(llvm_dir)
-
-        print()
         return True
     except subprocess.CalledProcessError:
         return False

build-scripts/requirements.txt

@@ -0,0 +1 @@
+requests==2.28.2

build-scripts/runtime_lib.cmake

@@ -19,6 +19,11 @@ endif ()
 if (NOT DEFINED DEPS_DIR)
     set (DEPS_DIR ${WAMR_ROOT_DIR}/core/deps)
 endif ()
+if (NOT DEFINED SHARED_PLATFORM_CONFIG)
+    # CMake file for platform configuration. The PLATFORM_SHARED_SOURCE varable
+    # should point to a list of platform-specfic source files to compile.
+    set (SHARED_PLATFORM_CONFIG ${SHARED_DIR}/platform/${WAMR_BUILD_PLATFORM}/shared_platform.cmake)
+endif ()
 
 if (DEFINED EXTRA_SDK_INCLUDE_PATH)
     message(STATUS, "EXTRA_SDK_INCLUDE_PATH = ${EXTRA_SDK_INCLUDE_PATH} ")
@@ -96,9 +101,13 @@ if (WAMR_BUILD_LIB_PTHREAD_SEMAPHORE EQUAL 1)
 endif ()
 
 if (WAMR_BUILD_WASI_NN EQUAL 1)
-    execute_process(COMMAND ${WAMR_ROOT_DIR}/core/deps/install_tensorflow.sh
+    if (NOT EXISTS "${WAMR_ROOT_DIR}/core/deps/tensorflow-src")
-                    RESULT_VARIABLE TENSORFLOW_RESULT
+        execute_process(COMMAND ${WAMR_ROOT_DIR}/core/deps/install_tensorflow.sh
-    )
+                        RESULT_VARIABLE TENSORFLOW_RESULT
+        )
+    else ()
+        message("Tensorflow is already downloaded.")
+    endif()
     set(TENSORFLOW_SOURCE_DIR "${WAMR_ROOT_DIR}/core/deps/tensorflow-src")
     include_directories (${CMAKE_CURRENT_BINARY_DIR}/flatbuffers/include)
     include_directories (${TENSORFLOW_SOURCE_DIR})
@@ -169,7 +178,7 @@ LIST (APPEND RUNTIME_LIB_HEADER_LIST ${header})
 
 enable_language (ASM)
 
-include (${SHARED_DIR}/platform/${WAMR_BUILD_PLATFORM}/shared_platform.cmake)
+include (${SHARED_PLATFORM_CONFIG})
 include (${SHARED_DIR}/mem-alloc/mem_alloc.cmake)
 include (${IWASM_DIR}/common/iwasm_common.cmake)
 include (${SHARED_DIR}/utils/shared_utils.cmake)

core/iwasm/aot/aot_runtime.c

@@ -30,6 +30,8 @@ bh_static_assert(offsetof(WASMExecEnv, aux_stack_boundary)
 bh_static_assert(offsetof(WASMExecEnv, aux_stack_bottom)
                  == 7 * sizeof(uintptr_t));
 bh_static_assert(offsetof(WASMExecEnv, native_symbol) == 8 * sizeof(uintptr_t));
+bh_static_assert(offsetof(WASMExecEnv, native_stack_top_min)
+                 == 9 * sizeof(uintptr_t));
 
 bh_static_assert(offsetof(AOTModuleInstance, memories) == 1 * sizeof(uint64));
 bh_static_assert(offsetof(AOTModuleInstance, func_ptrs) == 5 * sizeof(uint64));
@@ -1083,6 +1085,17 @@ aot_instantiate(AOTModule *module, bool is_sub_inst, uint32 stack_size,
     }
 #endif
 
+#if WASM_ENABLE_WASI_NN != 0
+    if (!is_sub_inst) {
+        if (!(((AOTModuleInstanceExtra *)module_inst->e)->wasi_nn_ctx =
+                  wasi_nn_initialize())) {
+            set_error_buf(error_buf, error_buf_size,
+                          "wasi nn initialization failed");
+            goto fail;
+        }
+    }
+#endif
+
     /* Initialize the thread related data */
     if (stack_size == 0)
         stack_size = DEFAULT_WASM_STACK_SIZE;
@@ -1194,6 +1207,15 @@ aot_deinstantiate(AOTModuleInstance *module_inst, bool is_sub_inst)
         wasm_runtime_free(
             ((AOTModuleInstanceExtra *)module_inst->e)->c_api_func_imports);
 
+#if WASM_ENABLE_WASI_NN != 0
+    if (!is_sub_inst) {
+        WASINNContext *wasi_nn_ctx =
+            ((AOTModuleInstanceExtra *)module_inst->e)->wasi_nn_ctx;
+        if (wasi_nn_ctx)
+            wasi_nn_destroy(wasi_nn_ctx);
+    }
+#endif
+
     wasm_runtime_free(module_inst);
 }
 
@@ -1237,6 +1259,7 @@ invoke_native_with_hw_bound_check(WASMExecEnv *exec_env, void *func_ptr,
     /* Check native stack overflow firstly to ensure we have enough
        native stack to run the following codes before actually calling
        the aot function in invokeNative function. */
+    RECORD_STACK_USAGE(exec_env, (uint8 *)&module_inst);
     if ((uint8 *)&module_inst < exec_env->native_stack_boundary
                                     + page_size * (guard_page_count + 1)) {
         aot_set_exception_with_id(module_inst, EXCE_NATIVE_STACK_OVERFLOW);
@@ -1836,6 +1859,7 @@ aot_call_indirect(WASMExecEnv *exec_env, uint32 tbl_idx, uint32 table_elem_idx,
        exec_env->native_stack_boundary must have been set, we don't set
        it again */
 
+    RECORD_STACK_USAGE(exec_env, (uint8 *)&module_inst);
     if ((uint8 *)&module_inst < exec_env->native_stack_boundary) {
         aot_set_exception_with_id(module_inst, EXCE_NATIVE_STACK_OVERFLOW);
         goto fail;

core/iwasm/aot/aot_runtime.h

@@ -11,6 +11,10 @@
 #include "../interpreter/wasm_runtime.h"
 #include "../compilation/aot.h"
 
+#if WASM_ENABLE_WASI_NN != 0
+#include "../libraries/wasi-nn/src/wasi_nn_private.h"
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -75,6 +79,9 @@ typedef struct AOTFunctionInstance {
 
 typedef struct AOTModuleInstanceExtra {
     CApiFuncImport *c_api_func_imports;
+#if WASM_ENABLE_WASI_NN != 0
+    WASINNContext *wasi_nn_ctx;
+#endif
 } AOTModuleInstanceExtra;
 
 #if defined(OS_ENABLE_HW_BOUND_CHECK) && defined(BH_PLATFORM_WINDOWS)

core/iwasm/common/arch/invokeNative_ia32.s

@@ -16,9 +16,14 @@ _invokeNative:
     push    %ebp
     movl    %esp, %ebp
     movl    16(%ebp), %ecx          /* ecx = argc */
-    movl    12(%ebp), %edx          /* edx = argv */
+    leal    2(%ecx), %edx           /* edx = ecx + 2 (count return address and saved ebp) */
+    andl    $3, %edx                /* edx = edx % 4 */
+    jz   stack_aligned              /* if edx == 0, stack is already 16 bytes aligned */
+    leal    -16(%esp, %edx, 4), %esp /* esp = esp - 16 + edx * 4 */
+stack_aligned:
     test    %ecx, %ecx
     jz      skip_push_args          /* if ecx == 0, skip pushing arguments */
+    movl    12(%ebp), %edx          /* edx = argv */
     leal    -4(%edx,%ecx,4), %edx   /* edx = edx + ecx * 4 - 4 */
     subl    %esp, %edx              /* edx = edx - esp */
 1:

core/iwasm/common/wasm_c_api.c

@@ -276,7 +276,7 @@ WASM_DEFINE_VEC_OWN(store, wasm_store_delete)
 WASM_DEFINE_VEC_OWN(valtype, wasm_valtype_delete)
 
 #ifndef NDEBUG
-#if WAMR_BUILD_MEMORY_PROFILING != 0
+#if WASM_ENABLE_MEMORY_PROFILING != 0
 #define WASM_C_DUMP_PROC_MEM() LOG_PROC_MEM()
 #else
 #define WASM_C_DUMP_PROC_MEM() (void)0
@@ -398,7 +398,7 @@ wasm_engine_new_internal(mem_alloc_type_t type, const MemAllocOption *opts)
 }
 
 /* global engine instance */
-static wasm_engine_t *singleton_engine = NULL;
+static wasm_engine_t *singleton_engine;
 #ifdef os_thread_local_attribute
 /* categorize wasm_store_t as threads*/
 static os_thread_local_attribute unsigned thread_local_stores_num = 0;
@@ -1458,6 +1458,30 @@ wasm_importtype_type(const wasm_importtype_t *import_type)
     return import_type->extern_type;
 }
 
+bool
+wasm_importtype_is_linked(const wasm_importtype_t *import_type)
+{
+    if (!import_type)
+        return false;
+
+    const wasm_name_t *module_name = wasm_importtype_module(import_type);
+    const wasm_name_t *field_name = wasm_importtype_name(import_type);
+
+    switch (wasm_externtype_kind(wasm_importtype_type(import_type))) {
+        case WASM_EXTERN_FUNC:
+            return wasm_runtime_is_import_func_linked(module_name->data,
+                                                      field_name->data);
+        case WASM_EXTERN_GLOBAL:
+            return wasm_runtime_is_import_global_linked(module_name->data,
+                                                        field_name->data);
+        case WASM_EXTERN_MEMORY:
+        case WASM_EXTERN_TABLE:
+        default:
+            break;
+    }
+    return false;
+}
+
 own wasm_exporttype_t *
 wasm_exporttype_new(own wasm_byte_vec_t *name,
                     own wasm_externtype_t *extern_type)
@@ -2537,12 +2561,12 @@ wasm_module_imports(const wasm_module_t *module, own wasm_importtype_vec_t *out)
 
         bh_assert(extern_type);
 
-        wasm_name_new_from_string(&module_name, module_name_rt);
+        wasm_name_new_from_string_nt(&module_name, module_name_rt);
         if (strlen(module_name_rt) && !module_name.data) {
             goto failed;
         }
 
-        wasm_name_new_from_string(&name, field_name_rt);
+        wasm_name_new_from_string_nt(&name, field_name_rt);
         if (strlen(field_name_rt) && !name.data) {
             goto failed;
         }
@@ -2622,7 +2646,7 @@ wasm_module_exports(const wasm_module_t *module, wasm_exporttype_vec_t *out)
         }
 
         /* byte* -> wasm_byte_vec_t */
-        wasm_name_new_from_string(&name, export->name);
+        wasm_name_new_from_string_nt(&name, export->name);
         if (strlen(export->name) && !name.data) {
             goto failed;
         }
@@ -3008,6 +3032,20 @@ failed:
     return NULL;
 }
 
+static wasm_func_t *
+wasm_func_new_empty(wasm_store_t *store)
+{
+    wasm_func_t *func = NULL;
+
+    if (!(func = malloc_internal(sizeof(wasm_func_t))))
+        goto failed;
+
+    func->store = store;
+    func->kind = WASM_EXTERN_FUNC;
+
+    RETURN_OBJ(func, wasm_func_delete)
+}
+
 void
 wasm_func_delete(wasm_func_t *func)
 {
@@ -3211,7 +3249,8 @@ wasm_func_call(const wasm_func_t *func, const wasm_val_vec_t *params,
         wasm_name_t message = { 0 };
         wasm_trap_t *trap;
 
-        wasm_name_new_from_string(&message, "failed to call unlinked function");
+        wasm_name_new_from_string_nt(&message,
+                                     "failed to call unlinked function");
         trap = wasm_trap_new(func->store, &message);
         wasm_byte_vec_delete(&message);
 
@@ -3371,6 +3410,25 @@ failed:
     return NULL;
 }
 
+static wasm_global_t *
+wasm_global_new_empty(wasm_store_t *store)
+{
+    wasm_global_t *global = NULL;
+
+    global = malloc_internal(sizeof(wasm_global_t));
+    if (!global)
+        goto failed;
+
+    global->store = store;
+    global->kind = WASM_EXTERN_GLOBAL;
+
+    return global;
+failed:
+    LOG_DEBUG("%s failed", __FUNCTION__);
+    wasm_global_delete(global);
+    return NULL;
+}
+
 /* almost same with wasm_global_new */
 wasm_global_t *
 wasm_global_copy(const wasm_global_t *src)
@@ -4205,7 +4263,8 @@ wasm_memory_data_size(const wasm_memory_t *memory)
             (WASMModuleInstance *)module_inst_comm;
         WASMMemoryInstance *memory_inst =
             module_inst->memories[memory->memory_idx_rt];
-        return memory_inst->cur_page_count * memory_inst->num_bytes_per_page;
+        return (size_t)memory_inst->cur_page_count
+               * memory_inst->num_bytes_per_page;
     }
 #endif
 
@@ -4215,7 +4274,8 @@ wasm_memory_data_size(const wasm_memory_t *memory)
         AOTMemoryInstance *memory_inst =
             ((AOTMemoryInstance **)
                  module_inst->memories)[memory->memory_idx_rt];
-        return memory_inst->cur_page_count * memory_inst->num_bytes_per_page;
+        return (size_t)memory_inst->cur_page_count
+               * memory_inst->num_bytes_per_page;
     }
 #endif
 
@@ -4286,6 +4346,11 @@ interp_link_func(const wasm_instance_t *inst, const WASMModule *module_interp,
 
     imported_func_interp = module_interp->import_functions + func_idx_rt;
     bh_assert(imported_func_interp);
+    bh_assert(imported_func_interp->kind == IMPORT_KIND_FUNC);
+
+    /* it is a placeholder and let's skip it*/
+    if (!import->type)
+        return true;
 
     /* type comparison */
     if (!wasm_functype_same_internal(
@@ -4300,6 +4365,8 @@ interp_link_func(const wasm_instance_t *inst, const WASMModule *module_interp,
         imported_func_interp->u.function.func_ptr_linked = import->u.cb_env.cb;
     else
         imported_func_interp->u.function.func_ptr_linked = import->u.cb;
+    bh_assert(imported_func_interp->u.function.func_ptr_linked);
+
     import->func_idx_rt = func_idx_rt;
 
     (void)inst;
@@ -4318,12 +4385,19 @@ interp_link_global(const WASMModule *module_interp, uint16 global_idx_rt,
 
     imported_global_interp = module_interp->import_globals + global_idx_rt;
     bh_assert(imported_global_interp);
+    bh_assert(imported_global_interp->kind == IMPORT_KIND_GLOBAL);
 
+    /* it is a placeholder and let's skip it*/
+    if (!import->type)
+        return true;
+
+    /* type comparison */
     if (!cmp_val_kind_with_val_type(wasm_valtype_kind(import->type->val_type),
                                     imported_global_interp->u.global.type))
         return false;
 
     /* set init value */
+    bh_assert(import->init);
     switch (wasm_valtype_kind(import->type->val_type)) {
         case WASM_I32:
             imported_global_interp->u.global.global_data_linked.i32 =
@@ -4350,58 +4424,6 @@ interp_link_global(const WASMModule *module_interp, uint16 global_idx_rt,
     return true;
 }
 
-static bool
-interp_link(const wasm_instance_t *inst, const WASMModule *module_interp,
-            wasm_extern_t *imports[])
-{
-    uint32 i = 0;
-    uint32 import_func_i = 0;
-    uint32 import_global_i = 0;
-
-    bh_assert(inst && module_interp && imports);
-
-    for (i = 0; i < module_interp->import_count; ++i) {
-        wasm_extern_t *import = imports[i];
-        WASMImport *import_rt = module_interp->imports + i;
-
-        switch (import_rt->kind) {
-            case IMPORT_KIND_FUNC:
-            {
-                if (!interp_link_func(inst, module_interp, import_func_i,
-                                      wasm_extern_as_func(import))) {
-                    LOG_WARNING("link #%d function failed", import_func_i);
-                    goto failed;
-                }
-                import_func_i++;
-                break;
-            }
-            case IMPORT_KIND_GLOBAL:
-            {
-                if (!interp_link_global(module_interp, import_global_i,
-                                        wasm_extern_as_global(import))) {
-                    LOG_WARNING("link #%d global failed", import_global_i);
-                    goto failed;
-                }
-                import_global_i++;
-                break;
-            }
-            case IMPORT_KIND_MEMORY:
-            case IMPORT_KIND_TABLE:
-            default:
-                ASSERT_NOT_IMPLEMENTED();
-                LOG_WARNING("%s meets unsupported kind: %d", __FUNCTION__,
-                            import_rt->kind);
-                goto failed;
-        }
-    }
-
-    return true;
-
-failed:
-    LOG_DEBUG("%s failed", __FUNCTION__);
-    return false;
-}
-
 static bool
 interp_process_export(wasm_store_t *store,
                       const WASMModuleInstance *inst_interp,
@@ -4501,6 +4523,10 @@ aot_link_func(const wasm_instance_t *inst, const AOTModule *module_aot,
     import_aot_func = module_aot->import_funcs + import_func_idx_rt;
     bh_assert(import_aot_func);
 
+    /* it is a placeholder and let's skip it*/
+    if (!import->type)
+        return true;
+
     /* type comparison */
     if (!wasm_functype_same_internal(import->type, import_aot_func->func_type))
         return false;
@@ -4513,6 +4539,8 @@ aot_link_func(const wasm_instance_t *inst, const AOTModule *module_aot,
         import_aot_func->func_ptr_linked = import->u.cb_env.cb;
     else
         import_aot_func->func_ptr_linked = import->u.cb;
+    bh_assert(import_aot_func->func_ptr_linked);
+
     import->func_idx_rt = import_func_idx_rt;
 
     return true;
@@ -4530,6 +4558,10 @@ aot_link_global(const AOTModule *module_aot, uint16 global_idx_rt,
     import_aot_global = module_aot->import_globals + global_idx_rt;
     bh_assert(import_aot_global);
 
+    /* it is a placeholder and let's skip it*/
+    if (!import->type)
+        return true;
+
     val_type = wasm_globaltype_content(import->type);
     bh_assert(val_type);
 
@@ -4537,6 +4569,7 @@ aot_link_global(const AOTModule *module_aot, uint16 global_idx_rt,
                                     import_aot_global->type))
         return false;
 
+    bh_assert(import->init);
     switch (wasm_valtype_kind(val_type)) {
         case WASM_I32:
             import_aot_global->global_data_linked.i32 = import->init->of.i32;
@@ -4557,62 +4590,6 @@ aot_link_global(const AOTModule *module_aot, uint16 global_idx_rt,
     import->global_idx_rt = global_idx_rt;
     import_aot_global->is_linked = true;
     return true;
-
-failed:
-    LOG_DEBUG("%s failed", __FUNCTION__);
-    return false;
-}
-
-static bool
-aot_link(const wasm_instance_t *inst, const AOTModule *module_aot,
-         wasm_extern_t *imports[])
-{
-    uint32 i = 0;
-    uint32 import_func_i = 0;
-    uint32 import_global_i = 0;
-    wasm_extern_t *import = NULL;
-    wasm_func_t *func = NULL;
-    wasm_global_t *global = NULL;
-
-    bh_assert(inst && module_aot && imports);
-
-    while (import_func_i < module_aot->import_func_count
-           || import_global_i < module_aot->import_global_count) {
-        import = imports[i++];
-
-        bh_assert(import);
-
-        switch (wasm_extern_kind(import)) {
-            case WASM_EXTERN_FUNC:
-                bh_assert(import_func_i < module_aot->import_func_count);
-                func = wasm_extern_as_func((wasm_extern_t *)import);
-                if (!aot_link_func(inst, module_aot, import_func_i, func)) {
-                    LOG_WARNING("link #%d function failed", import_func_i);
-                    goto failed;
-                }
-                import_func_i++;
-
-                break;
-            case WASM_EXTERN_GLOBAL:
-                bh_assert(import_global_i < module_aot->import_global_count);
-                global = wasm_extern_as_global((wasm_extern_t *)import);
-                if (!aot_link_global(module_aot, import_global_i, global)) {
-                    LOG_WARNING("link #%d global failed", import_global_i);
-                    goto failed;
-                }
-                import_global_i++;
-
-                break;
-            case WASM_EXTERN_MEMORY:
-            case WASM_EXTERN_TABLE:
-            default:
-                ASSERT_NOT_IMPLEMENTED();
-                goto failed;
-        }
-    }
-
-    return true;
-
 failed:
     LOG_DEBUG("%s failed", __FUNCTION__);
     return false;
@@ -4693,7 +4670,7 @@ aot_process_export(wasm_store_t *store, const AOTModuleInstance *inst_aot,
             goto failed;
         }
 
-        wasm_name_new_from_string(external->name, export->name);
+        wasm_name_new_from_string_nt(external->name, export->name);
         if (strlen(export->name) && !external->name->data) {
             goto failed;
         }
@@ -4711,6 +4688,95 @@ failed:
 }
 #endif /* WASM_ENABLE_AOT */
 
+static bool
+do_link(const wasm_instance_t *inst, const wasm_module_t *module,
+        const wasm_extern_vec_t *imports)
+{
+    uint32 i, import_func_i, import_global_i;
+
+    bh_assert(inst && module);
+
+    /* we have run a module_type check before. */
+
+    for (i = 0, import_func_i = 0, import_global_i = 0; i < imports->num_elems;
+         i++) {
+        wasm_extern_t *import = imports->data[i];
+
+        if (!import) {
+            LOG_ERROR("imports[%d] is NULL and it is fatal\n", i);
+            goto failed;
+        }
+
+        switch (wasm_extern_kind(import)) {
+            case WASM_EXTERN_FUNC:
+            {
+                bool ret = false;
+#if WASM_ENABLE_INTERP != 0
+                if ((*module)->module_type == Wasm_Module_Bytecode) {
+                    ret = interp_link_func(inst, MODULE_INTERP(module),
+                                           import_func_i,
+                                           wasm_extern_as_func(import));
+                }
+#endif
+#if WASM_ENABLE_AOT != 0
+                if ((*module)->module_type == Wasm_Module_AoT) {
+                    ret = aot_link_func(inst, MODULE_AOT(module), import_func_i,
+                                        wasm_extern_as_func(import));
+                }
+#endif
+                if (!ret) {
+                    LOG_WARNING("link function  #%d failed", import_func_i);
+                    goto failed;
+                }
+
+                import_func_i++;
+                break;
+            }
+            case WASM_EXTERN_GLOBAL:
+            {
+                bool ret = false;
+#if WASM_ENABLE_INTERP != 0
+                if ((*module)->module_type == Wasm_Module_Bytecode) {
+                    ret = interp_link_global(MODULE_INTERP(module),
+                                             import_global_i,
+                                             wasm_extern_as_global(import));
+                }
+#endif
+#if WASM_ENABLE_AOT != 0
+                if ((*module)->module_type == Wasm_Module_AoT) {
+                    ret = aot_link_global(MODULE_AOT(module), import_global_i,
+                                          wasm_extern_as_global(import));
+                }
+#endif
+                if (!ret) {
+                    LOG_WARNING("link global #%d failed", import_global_i);
+                    goto failed;
+                }
+
+                import_global_i++;
+                break;
+            }
+            case WASM_EXTERN_MEMORY:
+            case WASM_EXTERN_TABLE:
+            {
+                LOG_WARNING("doesn't support import memories and tables for "
+                            "now, ignore them");
+                break;
+            }
+            default:
+            {
+                UNREACHABLE();
+                break;
+            }
+        }
+    }
+
+    return true;
+failed:
+    LOG_DEBUG("%s failed", __FUNCTION__);
+    return false;
+}
+
 wasm_instance_t *
 wasm_instance_new(wasm_store_t *store, const wasm_module_t *module,
                   const wasm_extern_vec_t *imports, own wasm_trap_t **trap)
@@ -4719,57 +4785,6 @@ wasm_instance_new(wasm_store_t *store, const wasm_module_t *module,
                                        KILOBYTE(32), KILOBYTE(32));
 }
 
-static bool
-compare_imports(const wasm_module_t *module, const wasm_extern_vec_t *imports)
-{
-    unsigned import_func_count = 0;
-    unsigned import_global_count = 0;
-    unsigned import_memory_count = 0;
-    unsigned import_table_count = 0;
-    unsigned i = 0;
-
-    for (i = 0; imports && i < imports->num_elems; i++) {
-        wasm_extern_t *import = imports->data[i];
-        switch (wasm_extern_kind(import)) {
-            case WASM_EXTERN_FUNC:
-                import_func_count++;
-                break;
-            case WASM_EXTERN_GLOBAL:
-                import_global_count++;
-                break;
-            case WASM_EXTERN_MEMORY:
-                import_memory_count++;
-                break;
-            case WASM_EXTERN_TABLE:
-                import_table_count++;
-                break;
-            default:
-                UNREACHABLE();
-                return false;
-        }
-    }
-
-#if WASM_ENABLE_INTERP != 0
-    if ((*module)->module_type == Wasm_Module_Bytecode)
-        return import_func_count == MODULE_INTERP(module)->import_function_count
-               && import_global_count
-                      == MODULE_INTERP(module)->import_global_count
-               && import_memory_count
-                      == MODULE_INTERP(module)->import_memory_count
-               && import_table_count
-                      == MODULE_INTERP(module)->import_table_count;
-#endif
-#if WASM_ENABLE_AOT != 0
-    if ((*module)->module_type == Wasm_Module_AoT)
-        return import_func_count == MODULE_AOT(module)->import_func_count
-               && import_global_count == MODULE_AOT(module)->import_global_count
-               && import_memory_count == MODULE_AOT(module)->import_memory_count
-               && import_table_count == MODULE_AOT(module)->import_table_count;
-#endif
-
-    return false;
-}
-
 wasm_instance_t *
 wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
                             const wasm_extern_vec_t *imports,
@@ -4779,7 +4794,6 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
     char sub_error_buf[128] = { 0 };
     char error_buf[256] = { 0 };
     wasm_instance_t *instance = NULL;
-    WASMModuleInstance *inst_rt;
     CApiFuncImport *func_import = NULL, **p_func_imports = NULL;
     uint32 i = 0, import_func_count = 0;
     uint64 total_size;
@@ -4790,11 +4804,9 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
     if (!module)
         return NULL;
 
-    if (!compare_imports(module, imports)) {
+    /*
-        snprintf(sub_error_buf, sizeof(sub_error_buf),
+     * will do the check at the end of wasm_runtime_instantiate
-                 "Failed to match imports");
+     */
-        goto failed;
-    }
 
     WASM_C_DUMP_PROC_MEM();
 
@@ -4805,43 +4817,17 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
         goto failed;
     }
 
-    /* link module and imports */
+    /* executes the instantiate-time linking if provided */
-    if (imports && imports->num_elems) {
+    if (imports) {
-        bool link = false;
+        if (!do_link(instance, module, imports)) {
-#if WASM_ENABLE_INTERP != 0
-        if ((*module)->module_type == Wasm_Module_Bytecode) {
-            if (!interp_link(instance, MODULE_INTERP(module),
-                             (wasm_extern_t **)imports->data)) {
-                snprintf(sub_error_buf, sizeof(sub_error_buf),
-                         "Failed to validate imports");
-                goto failed;
-            }
-            link = true;
-        }
-#endif
-
-#if WASM_ENABLE_AOT != 0
-        if ((*module)->module_type == Wasm_Module_AoT) {
-            if (!aot_link(instance, MODULE_AOT(module),
-                          (wasm_extern_t **)imports->data)) {
-                snprintf(sub_error_buf, sizeof(sub_error_buf),
-                         "Failed to validate imports");
-                goto failed;
-            }
-            link = true;
-        }
-#endif
-
-        /*
-         * a wrong combination of module filetype and compilation flags
-         * also leads to below branch
-         */
-        if (!link) {
             snprintf(sub_error_buf, sizeof(sub_error_buf),
-                     "Failed to verify import count");
+                     "Failed to validate imports");
             goto failed;
         }
     }
+    /*
+     * will do the linking result check at the end of wasm_runtime_instantiate
+     */
 
     instance->inst_comm_rt = wasm_runtime_instantiate(
         *module, stack_size, heap_size, sub_error_buf, sizeof(sub_error_buf));
@@ -4856,18 +4842,22 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
     }
 
     /* create the c-api func import list */
-    inst_rt = (WASMModuleInstance *)instance->inst_comm_rt;
 #if WASM_ENABLE_INTERP != 0
     if (instance->inst_comm_rt->module_type == Wasm_Module_Bytecode) {
-        p_func_imports = &inst_rt->e->c_api_func_imports;
+        WASMModuleInstanceExtra *e =
-        import_func_count = inst_rt->module->import_function_count;
+            ((WASMModuleInstance *)instance->inst_comm_rt)->e;
+        p_func_imports = &(e->c_api_func_imports);
+        import_func_count = MODULE_INTERP(module)->import_function_count;
     }
 #endif
 #if WASM_ENABLE_AOT != 0
     if (instance->inst_comm_rt->module_type == Wasm_Module_AoT) {
-        p_func_imports =
+        AOTModuleInstanceExtra *e =
-            &((AOTModuleInstanceExtra *)inst_rt->e)->c_api_func_imports;
+            (AOTModuleInstanceExtra *)((AOTModuleInstance *)
-        import_func_count = ((AOTModule *)inst_rt->module)->import_func_count;
+                                           instance->inst_comm_rt)
+                ->e;
+        p_func_imports = &(e->c_api_func_imports);
+        import_func_count = MODULE_AOT(module)->import_func_count;
     }
 #endif
     bh_assert(p_func_imports);
@@ -4880,16 +4870,21 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
         goto failed;
     }
 
-    /* fill in c-api func import list */
+    /* fill in module_inst->e->c_api_func_imports */
     for (i = 0; imports && i < imports->num_elems; i++) {
-        wasm_func_t *func_host;
+        wasm_func_t *func_host = NULL;
-        wasm_extern_t *in;
+        wasm_extern_t *in = imports->data[i];
+        bh_assert(in);
 
-        in = imports->data[i];
         if (wasm_extern_kind(in) != WASM_EXTERN_FUNC)
             continue;
 
         func_host = wasm_extern_as_func(in);
+        /* it is a placeholder and let's skip it*/
+        if (!func_host->type) {
+            func_import++;
+            continue;
+        }
 
         func_import->with_env_arg = func_host->with_env;
         if (func_host->with_env) {
@@ -4900,6 +4895,7 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
             func_import->func_ptr_linked = func_host->u.cb;
             func_import->env_arg = NULL;
         }
+        bh_assert(func_import->func_ptr_linked);
 
         func_import++;
     }
@@ -4907,6 +4903,8 @@ wasm_instance_new_with_args(wasm_store_t *store, const wasm_module_t *module,
     /* fill with inst */
     for (i = 0; imports && imports->data && i < imports->num_elems; ++i) {
         wasm_extern_t *import = imports->data[i];
+        bh_assert(import);
+
         switch (import->kind) {
             case WASM_EXTERN_FUNC:
                 wasm_extern_as_func(import)->inst_comm_rt =
@@ -5002,7 +5000,7 @@ failed:
              sub_error_buf);
     if (trap != NULL) {
         wasm_message_t message = { 0 };
-        wasm_name_new_from_string(&message, error_buf);
+        wasm_name_new_from_string_nt(&message, error_buf);
         *trap = wasm_trap_new(store, &message);
         wasm_byte_vec_delete(&message);
     }
@@ -5202,3 +5200,16 @@ BASIC_FOUR_LIST(WASM_EXTERN_AS_OTHER_CONST)
 
 BASIC_FOUR_LIST(WASM_OTHER_AS_EXTERN_CONST)
 #undef WASM_OTHER_AS_EXTERN_CONST
+
+wasm_extern_t *
+wasm_extern_new_empty(wasm_store_t *store, wasm_externkind_t extern_kind)
+{
+    if (extern_kind == WASM_EXTERN_FUNC)
+        return wasm_func_as_extern(wasm_func_new_empty(store));
+
+    if (extern_kind == WASM_EXTERN_GLOBAL)
+        return wasm_global_as_extern(wasm_global_new_empty(store));
+
+    LOG_ERROR("Don't support linking table and memory for now");
+    return NULL;
+}

core/iwasm/common/wasm_exec_env.c

@@ -211,6 +211,7 @@ wasm_exec_env_set_thread_info(WASMExecEnv *exec_env)
     exec_env->handle = os_self_thread();
     exec_env->native_stack_boundary =
         stack_boundary ? stack_boundary + WASM_STACK_GUARD_SIZE : NULL;
+    exec_env->native_stack_top_min = (void *)UINTPTR_MAX;
 }
 
 #if WASM_ENABLE_THREAD_MGR != 0

core/iwasm/common/wasm_exec_env.h

@@ -84,6 +84,12 @@ typedef struct WASMExecEnv {
     void **native_symbol;
 #endif
 
+    /*
+     * The lowest stack pointer value observed.
+     * Assumption: native stack grows to the lower address.
+     */
+    uint8 *native_stack_top_min;
+
 #if WASM_ENABLE_FAST_JIT != 0
     /**
      * Cache for
@@ -165,6 +171,17 @@ typedef struct WASMExecEnv {
     } wasm_stack;
 } WASMExecEnv;
 
+#if WASM_ENABLE_MEMORY_PROFILING != 0
+#define RECORD_STACK_USAGE(e, p)               \
+    do {                                       \
+        if ((e)->native_stack_top_min > (p)) { \
+            (e)->native_stack_top_min = (p);   \
+        }                                      \
+    } while (0)
+#else
+#define RECORD_STACK_USAGE(e, p) (void)0
+#endif
+
 WASMExecEnv *
 wasm_exec_env_create_internal(struct WASMModuleInstanceCommon *module_inst,
                               uint32 stack_size);

core/iwasm/common/wasm_native.c

@@ -250,6 +250,10 @@ lookup_symbol(NativeSymbol *native_symbols, uint32 n_native_symbols,
     return NULL;
 }
 
+/**
+ * allow func_type and all outputs, like p_signature, p_attachment and
+ * p_call_conv_raw to be NULL
+ */
 void *
 wasm_native_resolve_symbol(const char *module_name, const char *field_name,
                            const WASMType *func_type, const char **p_signature,
@@ -275,10 +279,13 @@ wasm_native_resolve_symbol(const char *module_name, const char *field_name,
         node = node_next;
     }
 
+    if (!p_signature || !p_attachment || !p_call_conv_raw)
+        return func_ptr;
+
     if (func_ptr) {
         if (signature && signature[0] != '\0') {
             /* signature is not empty, check its format */
-            if (!check_symbol_signature(func_type, signature)) {
+            if (!func_type || !check_symbol_signature(func_type, signature)) {
 #if WASM_ENABLE_WAMR_COMPILER == 0
                 /* Output warning except running aot compiler */
                 LOG_WARNING("failed to check signature '%s' and resolve "

core/iwasm/common/wasm_runtime_common.c

@@ -7,6 +7,7 @@
 #include "bh_common.h"
 #include "bh_assert.h"
 #include "bh_log.h"
+#include "wasm_native.h"
 #include "wasm_runtime_common.h"
 #include "wasm_memory.h"
 #if WASM_ENABLE_INTERP != 0
@@ -128,6 +129,12 @@ runtime_malloc(uint64 size, WASMModuleInstanceCommon *module_inst,
 static JitCompOptions jit_options = { 0 };
 #endif
 
+#if WASM_ENABLE_JIT != 0
+static LLVMJITOptions llvm_jit_options = { 3, 3 };
+#endif
+
+static RunningMode runtime_running_mode = Mode_Default;
+
 #ifdef OS_ENABLE_HW_BOUND_CHECK
 /* The exec_env of thread local storage, set before calling function
    and used in signal handler, as we cannot get it from the argument
@@ -514,6 +521,20 @@ wasm_runtime_destroy()
     wasm_runtime_memory_destroy();
 }
 
+RunningMode
+wasm_runtime_get_default_running_mode(void)
+{
+    return runtime_running_mode;
+}
+
+#if WASM_ENABLE_JIT != 0
+LLVMJITOptions
+wasm_runtime_get_llvm_jit_options(void)
+{
+    return llvm_jit_options;
+}
+#endif
+
 bool
 wasm_runtime_full_init(RuntimeInitArgs *init_args)
 {
@@ -521,10 +542,20 @@ wasm_runtime_full_init(RuntimeInitArgs *init_args)
                                   &init_args->mem_alloc_option))
         return false;
 
+    if (!wasm_runtime_set_default_running_mode(init_args->running_mode)) {
+        wasm_runtime_memory_destroy();
+        return false;
+    }
+
 #if WASM_ENABLE_FAST_JIT != 0
     jit_options.code_cache_size = init_args->fast_jit_code_cache_size;
 #endif
 
+#if WASM_ENABLE_JIT != 0
+    llvm_jit_options.size_level = init_args->llvm_jit_size_level;
+    llvm_jit_options.opt_level = init_args->llvm_jit_opt_level;
+#endif
+
     if (!wasm_runtime_env_init()) {
         wasm_runtime_memory_destroy();
         return false;
@@ -554,6 +585,47 @@ wasm_runtime_full_init(RuntimeInitArgs *init_args)
     return true;
 }
 
+bool
+wasm_runtime_is_running_mode_supported(RunningMode running_mode)
+{
+    if (running_mode == Mode_Default) {
+        return true;
+    }
+    else if (running_mode == Mode_Interp) {
+#if WASM_ENABLE_INTERP != 0
+        return true;
+#endif
+    }
+    else if (running_mode == Mode_Fast_JIT) {
+#if WASM_ENABLE_FAST_JIT != 0
+        return true;
+#endif
+    }
+    else if (running_mode == Mode_LLVM_JIT) {
+#if WASM_ENABLE_JIT != 0
+        return true;
+#endif
+    }
+    else if (running_mode == Mode_Multi_Tier_JIT) {
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
+        return true;
+#endif
+    }
+
+    return false;
+}
+
+bool
+wasm_runtime_set_default_running_mode(RunningMode running_mode)
+{
+    if (wasm_runtime_is_running_mode_supported(running_mode)) {
+        runtime_running_mode = running_mode;
+        return true;
+    }
+    return false;
+}
+
 PackageType
 get_package_type(const uint8 *buf, uint32 size)
 {
@@ -1171,6 +1243,41 @@ wasm_runtime_deinstantiate_internal(WASMModuleInstanceCommon *module_inst,
 #endif
 }
 
+bool
+wasm_runtime_set_running_mode(wasm_module_inst_t module_inst,
+                              RunningMode running_mode)
+{
+#if WASM_ENABLE_AOT != 0
+    if (module_inst->module_type == Wasm_Module_AoT)
+        return true;
+#endif
+
+#if WASM_ENABLE_INTERP != 0
+    if (module_inst->module_type == Wasm_Module_Bytecode) {
+        WASMModuleInstance *module_inst_interp =
+            (WASMModuleInstance *)module_inst;
+
+        return wasm_set_running_mode(module_inst_interp, running_mode);
+    }
+#endif
+
+    return false;
+}
+
+RunningMode
+wasm_runtime_get_running_mode(wasm_module_inst_t module_inst)
+{
+#if WASM_ENABLE_INTERP != 0
+    if (module_inst->module_type == Wasm_Module_Bytecode) {
+        WASMModuleInstance *module_inst_interp =
+            (WASMModuleInstance *)module_inst;
+        return module_inst_interp->e->running_mode;
+    }
+#endif
+
+    return Mode_Default;
+}
+
 void
 wasm_runtime_deinstantiate(WASMModuleInstanceCommon *module_inst)
 {
@@ -1399,6 +1506,22 @@ wasm_runtime_dump_mem_consumption(WASMExecEnv *exec_env)
     else
         os_printf("Total aux stack used: no enough info to profile\n");
 
+    /*
+     * Report the native stack usage estimation.
+     *
+     * Unlike the aux stack above, we report the amount unused
+     * because we don't know the stack "bottom".
+     *
+     * Note that this is just about what the runtime itself observed.
+     * It doesn't cover host func implementations, signal handlers, etc.
+     */
+    if (exec_env->native_stack_top_min != (void *)UINTPTR_MAX)
+        os_printf("Native stack left: %zd\n",
+                  exec_env->native_stack_top_min
+                      - exec_env->native_stack_boundary);
+    else
+        os_printf("Native stack left: no enough info to profile\n");
+
     os_printf("Total app heap used: %u\n", app_heap_peak_size);
 }
 #endif /* end of (WASM_ENABLE_MEMORY_PROFILING != 0) \
@@ -2195,6 +2318,12 @@ wasm_set_exception(WASMModuleInstance *module_inst, const char *exception)
     if (exec_env) {
         wasm_cluster_spread_exception(exec_env, exception ? false : true);
     }
+#if WASM_ENABLE_SHARED_MEMORY
+    if (exception) {
+        notify_stale_threads_on_exception(
+            (WASMModuleInstanceCommon *)module_inst);
+    }
+#endif
 #else
     (void)exec_env;
 #endif
@@ -2219,9 +2348,7 @@ static const char *exception_msgs[] = {
     "wasm auxiliary stack underflow", /* EXCE_AUX_STACK_UNDERFLOW */
     "out of bounds table access",     /* EXCE_OUT_OF_BOUNDS_TABLE_ACCESS */
     "wasm operand stack overflow",    /* EXCE_OPERAND_STACK_OVERFLOW */
-#if WASM_ENABLE_FAST_JIT != 0
     "failed to compile fast jit function", /* EXCE_FAILED_TO_COMPILE_FAST_JIT_FUNC */
-#endif
     "",                               /* EXCE_ALREADY_THROWN */
 };
 /* clang-format on */
@@ -5205,3 +5332,24 @@ wasm_runtime_get_version(uint32_t *major, uint32_t *minor, uint32_t *patch)
     *minor = WAMR_VERSION_MINOR;
     *patch = WAMR_VERSION_PATCH;
 }
+
+bool
+wasm_runtime_is_import_func_linked(const char *module_name,
+                                   const char *func_name)
+{
+    return wasm_native_resolve_symbol(module_name, func_name, NULL, NULL, NULL,
+                                      NULL);
+}
+
+bool
+wasm_runtime_is_import_global_linked(const char *module_name,
+                                     const char *global_name)
+{
+#if WASM_ENABLE_LIBC_BUILTIN != 0
+    WASMGlobalImport global = { 0 };
+    return wasm_native_lookup_libc_builtin_global(module_name, global_name,
+                                                  &global);
+#else
+    return false;
+#endif
+}

core/iwasm/common/wasm_runtime_common.h

@@ -25,6 +25,9 @@
 extern "C" {
 #endif
 
+/* Internal use for setting default running mode */
+#define Mode_Default 0
+
 #if WASM_CPU_SUPPORTS_UNALIGNED_ADDR_ACCESS != 0
 
 #define PUT_I64_TO_ADDR(addr, value)       \
@@ -413,6 +416,13 @@ typedef struct wasm_frame_t {
     const char *func_name_wp;
 } WASMCApiFrame;
 
+#ifdef WASM_ENABLE_JIT
+typedef struct LLVMJITOptions {
+    uint32 opt_level;
+    uint32 size_level;
+} LLVMJITOptions;
+#endif
+
 #ifdef OS_ENABLE_HW_BOUND_CHECK
 /* Signal info passing to interp/aot signal handler */
 typedef struct WASMSignalInfo {
@@ -437,10 +447,28 @@ wasm_runtime_get_exec_env_tls(void);
 WASM_RUNTIME_API_EXTERN bool
 wasm_runtime_init(void);
 
+/* Internal API */
+RunningMode
+wasm_runtime_get_default_running_mode(void);
+
+#if WASM_ENABLE_JIT != 0
+/* Internal API */
+LLVMJITOptions
+wasm_runtime_get_llvm_jit_options(void);
+#endif
+
 /* See wasm_export.h for description */
 WASM_RUNTIME_API_EXTERN bool
 wasm_runtime_full_init(RuntimeInitArgs *init_args);
 
+/* See wasm_export.h for description */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_is_running_mode_supported(RunningMode running_mode);
+
+/* See wasm_export.h for description */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_set_default_running_mode(RunningMode running_mode);
+
 /* See wasm_export.h for description */
 WASM_RUNTIME_API_EXTERN void
 wasm_runtime_destroy(void);
@@ -484,6 +512,15 @@ wasm_runtime_instantiate(WASMModuleCommon *module, uint32 stack_size,
                          uint32 heap_size, char *error_buf,
                          uint32 error_buf_size);
 
+/* See wasm_export.h for description */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_set_running_mode(wasm_module_inst_t module_inst,
+                              RunningMode running_mode);
+
+/* See wasm_export.h for description */
+WASM_RUNTIME_API_EXTERN RunningMode
+wasm_runtime_get_running_mode(wasm_module_inst_t module_inst);
+
 /* See wasm_export.h for description */
 WASM_RUNTIME_API_EXTERN void
 wasm_runtime_deinstantiate(WASMModuleInstanceCommon *module_inst);
@@ -956,6 +993,14 @@ void
 wasm_runtime_destroy_custom_sections(WASMCustomSection *section_list);
 #endif
 
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_is_import_func_linked(const char *module_name,
+                                   const char *func_name);
+
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_is_import_global_linked(const char *module_name,
+                                     const char *global_name);
+
 #ifdef __cplusplus
 }
 #endif

core/iwasm/common/wasm_shared_memory.c

@@ -30,6 +30,11 @@ typedef struct AtomicWaitNode {
     korp_cond wait_cond;
 } AtomicWaitNode;
 
+typedef struct AtomicWaitAddressArgs {
+    uint32 index;
+    void **addr;
+} AtomicWaitAddressArgs;
+
 /* Atomic wait map */
 static HashMap *wait_map;
 
@@ -87,6 +92,61 @@ search_module(WASMModuleCommon *module)
     return NULL;
 }
 
+static void
+wait_map_address_count_callback(void *key, void *value,
+                                void *p_total_elem_count)
+{
+    *(uint32 *)p_total_elem_count = *(uint32 *)p_total_elem_count + 1;
+}
+
+static void
+create_list_of_waiter_addresses(void *key, void *value, void *user_data)
+{
+    AtomicWaitAddressArgs *data = (AtomicWaitAddressArgs *)user_data;
+    data->addr[data->index++] = key;
+}
+
+void
+notify_stale_threads_on_exception(WASMModuleInstanceCommon *module_inst)
+{
+    AtomicWaitAddressArgs args = { 0 };
+    uint32 i = 0, total_elem_count = 0;
+    uint64 total_elem_count_size = 0;
+
+    os_mutex_lock(&shared_memory_list_lock);
+
+    /* count number of addresses in wait_map */
+    bh_hash_map_traverse(wait_map, wait_map_address_count_callback,
+                         (void *)&total_elem_count);
+
+    if (!total_elem_count) {
+        os_mutex_unlock(&shared_memory_list_lock);
+        return;
+    }
+
+    /* allocate memory */
+    total_elem_count_size = (uint64)sizeof(void *) * total_elem_count;
+    if (total_elem_count_size >= UINT32_MAX
+        || !(args.addr = wasm_runtime_malloc((uint32)total_elem_count_size))) {
+        LOG_ERROR(
+            "failed to allocate memory for list of atomic wait addresses");
+        os_mutex_unlock(&shared_memory_list_lock);
+        return;
+    }
+
+    /* set values in list of addresses */
+    bh_hash_map_traverse(wait_map, create_list_of_waiter_addresses, &args);
+    os_mutex_unlock(&shared_memory_list_lock);
+
+    /* notify */
+    for (i = 0; i < args.index; i++) {
+        wasm_runtime_atomic_notify(module_inst, args.addr[i], UINT32_MAX);
+    }
+
+    /* free memory allocated to args data */
+    wasm_runtime_free(args.addr);
+}
+
 WASMSharedMemNode *
 wasm_module_get_shared_memory(WASMModuleCommon *module)
 {
@@ -120,6 +180,7 @@ shared_memory_dec_reference(WASMModuleCommon *module)
             bh_list_remove(shared_memory_list, node);
             os_mutex_unlock(&shared_memory_list_lock);
 
+            os_mutex_destroy(&node->shared_mem_lock);
             os_mutex_destroy(&node->lock);
             wasm_runtime_free(node);
         }
@@ -148,7 +209,14 @@ shared_memory_set_memory_inst(WASMModuleCommon *module,
     node->module = module;
     node->memory_inst = memory;
     node->ref_count = 1;
+
+    if (os_mutex_init(&node->shared_mem_lock) != 0) {
+        wasm_runtime_free(node);
+        return NULL;
+    }
+
     if (os_mutex_init(&node->lock) != 0) {
+        os_mutex_destroy(&node->shared_mem_lock);
         wasm_runtime_free(node);
         return NULL;
     }
@@ -322,6 +390,10 @@ wasm_runtime_atomic_wait(WASMModuleInstanceCommon *module, void *address,
     bh_assert(module->module_type == Wasm_Module_Bytecode
               || module->module_type == Wasm_Module_AoT);
 
+    if (wasm_get_exception(module_inst)) {
+        return -1;
+    }
+
     /* Currently we have only one memory instance */
     if (!module_inst->memories[0]->is_shared) {
         wasm_runtime_set_exception(module, "expected shared memory");

core/iwasm/common/wasm_shared_memory.h

@@ -26,6 +26,8 @@ typedef struct WASMSharedMemNode {
     WASMModuleCommon *module;
     /* The memory information */
     WASMMemoryInstanceCommon *memory_inst;
+    /* Lock used for atomic operations */
+    korp_mutex shared_mem_lock;
 
     /* reference count */
     uint32 ref_count;
@@ -37,6 +39,9 @@ wasm_shared_memory_init();
 void
 wasm_shared_memory_destroy();
 
+void
+notify_stale_threads_on_exception(WASMModuleInstanceCommon *module);
+
 WASMSharedMemNode *
 wasm_module_get_shared_memory(WASMModuleCommon *module);
 

core/iwasm/compilation/aot_compiler.h

@@ -259,6 +259,7 @@ check_type_compatible(uint8 src_type, uint8 dst_type)
 #define I32_SIX LLVM_CONST(i32_six)
 #define I32_SEVEN LLVM_CONST(i32_seven)
 #define I32_EIGHT LLVM_CONST(i32_eight)
+#define I32_NINE LLVM_CONST(i32_nine)
 #define I32_NEG_ONE LLVM_CONST(i32_neg_one)
 #define I64_NEG_ONE LLVM_CONST(i64_neg_one)
 #define I32_MIN LLVM_CONST(i32_min)

core/iwasm/compilation/aot_emit_function.c

@@ -366,6 +366,87 @@ fail:
 #endif /* end of (WASM_ENABLE_DUMP_CALL_STACK != 0) \
                  || (WASM_ENABLE_PERF_PROFILING != 0) */
 
+static bool
+record_stack_usage(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
+                   uint32 callee_cell_num)
+{
+    LLVMBasicBlockRef block_curr = LLVMGetInsertBlock(comp_ctx->builder);
+    LLVMBasicBlockRef block_update;
+    LLVMBasicBlockRef block_after_update;
+    LLVMValueRef callee_local_size, new_sp, cmp;
+    LLVMValueRef native_stack_top_min;
+    LLVMTypeRef ptrdiff_type;
+    if (comp_ctx->pointer_size == sizeof(uint64_t)) {
+        ptrdiff_type = I64_TYPE;
+    }
+    else {
+        ptrdiff_type = I32_TYPE;
+    }
+
+    /*
+     * new_sp = last_alloca - callee_local_size;
+     * if (*native_stack_top_min_addr > new_sp) {
+     *    *native_stack_top_min_addr = new_sp;
+     * }
+     */
+
+    if (!(callee_local_size = LLVMConstInt(
+              ptrdiff_type, -(int64_t)callee_cell_num * 4, true))) {
+        aot_set_last_error("llvm build const failed.");
+        return false;
+    }
+    if (!(new_sp = LLVMBuildInBoundsGEP2(comp_ctx->builder, INT8_TYPE,
+                                         func_ctx->last_alloca,
+                                         &callee_local_size, 1, "new_sp"))) {
+        aot_set_last_error("llvm build gep failed");
+        return false;
+    }
+    if (!(native_stack_top_min = LLVMBuildLoad2(
+              comp_ctx->builder, OPQ_PTR_TYPE,
+              func_ctx->native_stack_top_min_addr, "native_stack_top_min"))) {
+        aot_set_last_error("llvm build load failed");
+        return false;
+    }
+    if (!(cmp = LLVMBuildICmp(comp_ctx->builder, LLVMIntULT, new_sp,
+                              native_stack_top_min, "cmp"))) {
+        aot_set_last_error("llvm build icmp failed.");
+        return false;
+    }
+
+    if (!(block_update = LLVMAppendBasicBlockInContext(
+              comp_ctx->context, func_ctx->func, "block_update"))) {
+        aot_set_last_error("llvm add basic block failed.");
+        return false;
+    }
+    if (!(block_after_update = LLVMAppendBasicBlockInContext(
+              comp_ctx->context, func_ctx->func, "block_after_update"))) {
+        aot_set_last_error("llvm add basic block failed.");
+        return false;
+    }
+    LLVMMoveBasicBlockAfter(block_update, block_curr);
+    LLVMMoveBasicBlockAfter(block_after_update, block_update);
+
+    if (!LLVMBuildCondBr(comp_ctx->builder, cmp, block_update,
+                         block_after_update)) {
+        aot_set_last_error("llvm build cond br failed.");
+        return false;
+    }
+
+    LLVMPositionBuilderAtEnd(comp_ctx->builder, block_update);
+    if (!LLVMBuildStore(comp_ctx->builder, new_sp,
+                        func_ctx->native_stack_top_min_addr)) {
+        aot_set_last_error("llvm build store failed");
+        return false;
+    }
+    if (!LLVMBuildBr(comp_ctx->builder, block_after_update)) {
+        aot_set_last_error("llvm build br failed.");
+        return false;
+    }
+
+    LLVMPositionBuilderAtEnd(comp_ctx->builder, block_after_update);
+    return true;
+}
+
 static bool
 check_stack_boundary(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                      uint32 callee_cell_num)
@@ -409,6 +490,19 @@ check_stack_boundary(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
     return true;
 }
 
+static bool
+check_stack(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
+            uint32 callee_cell_num)
+{
+    if (comp_ctx->enable_stack_estimation
+        && !record_stack_usage(comp_ctx, func_ctx, callee_cell_num))
+        return false;
+    if (comp_ctx->enable_stack_bound_check
+        && !check_stack_boundary(comp_ctx, func_ctx, callee_cell_num))
+        return false;
+    return true;
+}
+
 /**
  * Check whether the app address and its buffer are inside the linear memory,
  * if no, throw exception
@@ -852,8 +946,7 @@ aot_compile_op_call(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
         callee_cell_num =
             aot_func->param_cell_num + aot_func->local_cell_num + 1;
 
-        if (comp_ctx->enable_stack_bound_check
+        if (!check_stack(comp_ctx, func_ctx, callee_cell_num))
-            && !check_stack_boundary(comp_ctx, func_ctx, callee_cell_num))
             goto fail;
 
 #if LLVM_VERSION_MAJOR >= 14
@@ -1467,12 +1560,11 @@ aot_compile_op_call_indirect(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
     /* Translate call non-import block */
     LLVMPositionBuilderAtEnd(comp_ctx->builder, block_call_non_import);
 
-    if (comp_ctx->enable_stack_bound_check
+    if (!check_stack(comp_ctx, func_ctx,
-        && !check_stack_boundary(comp_ctx, func_ctx,
+                     param_cell_num + ext_cell_num
-                                 param_cell_num + ext_cell_num
+                         + 1
-                                     + 1
+                         /* Reserve some local variables */
-                                     /* Reserve some local variables */
+                         + 16))
-                                     + 16))
         goto fail;
 
     /* Load function pointer */

core/iwasm/compilation/aot_llvm.c

@@ -286,6 +286,21 @@ create_native_stack_bound(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
     return true;
 }
 
+static bool
+create_native_stack_top_min(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
+{
+    LLVMValueRef offset = I32_NINE;
+
+    if (!(func_ctx->native_stack_top_min_addr = LLVMBuildInBoundsGEP2(
+              comp_ctx->builder, OPQ_PTR_TYPE, func_ctx->exec_env, &offset, 1,
+              "native_stack_top_min_addr"))) {
+        aot_set_last_error("llvm build in bounds gep failed");
+        return false;
+    }
+
+    return true;
+}
+
 static bool
 create_aux_stack_info(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
 {
@@ -434,7 +449,8 @@ create_local_variables(AOTCompData *comp_data, AOTCompContext *comp_ctx,
         }
     }
 
-    if (comp_ctx->enable_stack_bound_check) {
+    if (comp_ctx->enable_stack_bound_check
+        || comp_ctx->enable_stack_estimation) {
         if (aot_func_type->param_count + func->local_count > 0) {
             func_ctx->last_alloca = func_ctx->locals[aot_func_type->param_count
                                                      + func->local_count - 1];
@@ -963,6 +979,10 @@ aot_create_func_context(AOTCompData *comp_data, AOTCompContext *comp_ctx,
         && !create_native_stack_bound(comp_ctx, func_ctx)) {
         goto fail;
     }
+    if (comp_ctx->enable_stack_estimation
+        && !create_native_stack_top_min(comp_ctx, func_ctx)) {
+        goto fail;
+    }
 
     /* Get auxiliary stack info */
     if (wasm_func->has_op_set_global_aux_stack
@@ -1622,6 +1642,9 @@ aot_create_comp_context(AOTCompData *comp_data, aot_comp_option_t option)
     if (option->disable_llvm_lto)
         comp_ctx->disable_llvm_lto = true;
 
+    if (option->enable_stack_estimation)
+        comp_ctx->enable_stack_estimation = true;
+
     comp_ctx->opt_level = option->opt_level;
     comp_ctx->size_level = option->size_level;
 

core/iwasm/compilation/aot_llvm.h

@@ -163,6 +163,7 @@ typedef struct AOTFuncContext {
     LLVMValueRef aot_inst;
     LLVMValueRef argv_buf;
     LLVMValueRef native_stack_bound;
+    LLVMValueRef native_stack_top_min_addr;
     LLVMValueRef aux_stack_bound;
     LLVMValueRef aux_stack_bottom;
     LLVMValueRef native_symbol;
@@ -313,6 +314,9 @@ typedef struct AOTCompContext {
     /* Native stack bounday Check */
     bool enable_stack_bound_check;
 
+    /* Native stack usage estimation */
+    bool enable_stack_estimation;
+
     /* 128-bit SIMD */
     bool enable_simd;
 
@@ -403,6 +407,7 @@ typedef struct AOTCompOption {
     bool enable_aux_stack_frame;
     bool disable_llvm_intrinsics;
     bool disable_llvm_lto;
+    bool enable_stack_estimation;
     uint32 opt_level;
     uint32 size_level;
     uint32 output_format;

core/iwasm/fast-jit/fe/jit_emit_memory.c

@@ -135,7 +135,8 @@ check_and_seek(JitCompContext *cc, JitReg addr, uint32 offset, uint32 bytes)
 #ifndef OS_ENABLE_HW_BOUND_CHECK
     /* ---------- check ---------- */
     /* 1. shortcut if the memory size is 0 */
-    if (0 == cc->cur_wasm_module->memories[mem_idx].init_page_count) {
+    if (cc->cur_wasm_module->memories != NULL
+        && 0 == cc->cur_wasm_module->memories[mem_idx].init_page_count) {
         JitReg module_inst, cur_page_count;
         uint32 cur_page_count_offset =
             (uint32)offsetof(WASMModuleInstance, global_table_data.bytes)
@@ -176,6 +177,18 @@ fail:
     return 0;
 }
 
+#define CHECK_ALIGNMENT(maddr, memory_data, offset1)                   \
+    do {                                                               \
+        GEN_INSN(ADD, maddr, memory_data, offset1);                    \
+        JitReg align_mask = NEW_CONST(I64, ((uint64)1 << align) - 1);  \
+        JitReg AND_res = jit_cc_new_reg_I64(cc);                       \
+        GEN_INSN(AND, AND_res, maddr, align_mask);                     \
+        GEN_INSN(CMP, cc->cmp_reg, AND_res, NEW_CONST(I64, 0));        \
+        if (!jit_emit_exception(cc, EXCE_UNALIGNED_ATOMIC, JIT_OP_BNE, \
+                                cc->cmp_reg, NULL))                    \
+            goto fail;                                                 \
+    } while (0)
+
 bool
 jit_compile_op_i32_load(JitCompContext *cc, uint32 align, uint32 offset,
                         uint32 bytes, bool sign, bool atomic)
@@ -779,6 +792,51 @@ bool
 jit_compile_op_atomic_wait(JitCompContext *cc, uint8 op_type, uint32 align,
                            uint32 offset, uint32 bytes)
 {
+    bh_assert(op_type == VALUE_TYPE_I32 || op_type == VALUE_TYPE_I64);
+
+    // Pop atomic.wait arguments
+    JitReg timeout, expect, expect_64, addr;
+    POP_I64(timeout);
+    if (op_type == VALUE_TYPE_I32) {
+        POP_I32(expect);
+        expect_64 = jit_cc_new_reg_I64(cc);
+        GEN_INSN(I32TOI64, expect_64, expect);
+    }
+    else {
+        POP_I64(expect_64);
+    }
+    POP_I32(addr);
+
+    // Get referenced address and store it in `maddr`
+    JitReg memory_data = get_memory_data_reg(cc->jit_frame, 0);
+    JitReg offset1 = check_and_seek(cc, addr, offset, bytes);
+    if (!offset1)
+        goto fail;
+    JitReg maddr = jit_cc_new_reg_I64(cc);
+    CHECK_ALIGNMENT(maddr, memory_data, offset1);
+
+    // Prepare `wasm_runtime_atomic_wait` arguments
+    JitReg res = jit_cc_new_reg_I32(cc);
+    JitReg args[5] = { 0 };
+    args[0] = get_module_inst_reg(cc->jit_frame);
+    args[1] = maddr;
+    args[2] = expect_64;
+    args[3] = timeout;
+    args[4] = NEW_CONST(I32, false);
+
+    if (!jit_emit_callnative(cc, wasm_runtime_atomic_wait, res, args,
+                             sizeof(args) / sizeof(args[0])))
+        goto fail;
+
+    // Handle return code
+    GEN_INSN(CMP, cc->cmp_reg, res, NEW_CONST(I32, -1));
+    if (!jit_emit_exception(cc, EXCE_ALREADY_THROWN, JIT_OP_BEQ, cc->cmp_reg,
+                            NULL))
+        goto fail;
+
+    PUSH_I32(res);
+    return true;
+fail:
     return false;
 }
 
@@ -786,6 +844,39 @@ bool
 jit_compiler_op_atomic_notify(JitCompContext *cc, uint32 align, uint32 offset,
                               uint32 bytes)
 {
+    // Pop atomic.notify arguments
+    JitReg notify_count, addr;
+    POP_I32(notify_count);
+    POP_I32(addr);
+
+    // Get referenced address and store it in `maddr`
+    JitReg memory_data = get_memory_data_reg(cc->jit_frame, 0);
+    JitReg offset1 = check_and_seek(cc, addr, offset, bytes);
+    if (!offset1)
+        goto fail;
+    JitReg maddr = jit_cc_new_reg_I64(cc);
+    CHECK_ALIGNMENT(maddr, memory_data, offset1);
+
+    // Prepare `wasm_runtime_atomic_notify` arguments
+    JitReg res = jit_cc_new_reg_I32(cc);
+    JitReg args[3] = { 0 };
+    args[0] = get_module_inst_reg(cc->jit_frame);
+    args[1] = maddr;
+    args[2] = notify_count;
+
+    if (!jit_emit_callnative(cc, wasm_runtime_atomic_notify, res, args,
+                             sizeof(args) / sizeof(args[0])))
+        goto fail;
+
+    // Handle return code
+    GEN_INSN(CMP, cc->cmp_reg, res, NEW_CONST(I32, 0));
+    if (!jit_emit_exception(cc, EXCE_ALREADY_THROWN, JIT_OP_BLTS, cc->cmp_reg,
+                            NULL))
+        goto fail;
+
+    PUSH_I32(res);
+    return true;
+fail:
     return false;
 }
 #endif

core/iwasm/fast-jit/jit_codecache.c

@@ -56,9 +56,31 @@ jit_code_cache_free(void *ptr)
 bool
 jit_pass_register_jitted_code(JitCompContext *cc)
 {
-    uint32 jit_func_idx =
+    WASMModuleInstance *instance;
-        cc->cur_wasm_func_idx - cc->cur_wasm_module->import_function_count;
+    WASMModule *module = cc->cur_wasm_module;
-    cc->cur_wasm_module->fast_jit_func_ptrs[jit_func_idx] =
+    WASMFunction *func = cc->cur_wasm_func;
-        cc->cur_wasm_func->fast_jit_jitted_code = cc->jitted_addr_begin;
+    uint32 jit_func_idx = cc->cur_wasm_func_idx - module->import_function_count;
+
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
+    os_mutex_lock(&module->instance_list_lock);
+#endif
+
+    module->fast_jit_func_ptrs[jit_func_idx] = func->fast_jit_jitted_code =
+        cc->jitted_addr_begin;
+
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
+    instance = module->instance_list;
+    while (instance) {
+        if (instance->e->running_mode == Mode_Fast_JIT)
+            instance->fast_jit_func_ptrs[jit_func_idx] = cc->jitted_addr_begin;
+        instance = instance->e->next;
+    }
+
+    os_mutex_unlock(&module->instance_list_lock);
+#else
+    (void)instance;
+#endif
     return true;
 }

core/iwasm/fast-jit/jit_compiler.c

@@ -157,8 +157,16 @@ jit_compiler_compile(WASMModule *module, uint32 func_idx)
     /* Apply compiler passes */
     if (!apply_compiler_passes(cc) || jit_get_last_error(cc)) {
         last_error = jit_get_last_error(cc);
+
+#if WASM_ENABLE_CUSTOM_NAME_SECTION != 0
+        char *function_name = cc->cur_wasm_func->field_name;
+        os_printf("fast jit compilation failed: %s (function_name=%s)\n",
+                  last_error ? last_error : "unknown error", function_name);
+#else
         os_printf("fast jit compilation failed: %s\n",
                   last_error ? last_error : "unknown error");
+#endif
+
         goto fail;
     }
 
@@ -246,6 +254,8 @@ jit_compiler_set_call_to_fast_jit(WASMModule *module, uint32 func_idx)
 
     func_ptr = jit_codegen_compile_call_to_fast_jit(module, func_idx);
     if (func_ptr) {
+        uint32 i = func_idx - module->import_function_count;
+        module->functions[i]->call_to_fast_jit_from_llvm_jit = func_ptr;
         jit_compiler_set_llvm_jit_func_ptr(module, func_idx, func_ptr);
     }
 
@@ -259,12 +269,14 @@ jit_compiler_set_llvm_jit_func_ptr(WASMModule *module, uint32 func_idx,
     WASMModuleInstance *instance;
     uint32 i = func_idx - module->import_function_count;
 
-    module->functions[i]->llvm_jit_func_ptr = module->func_ptrs[i] = func_ptr;
-
     os_mutex_lock(&module->instance_list_lock);
+
+    module->func_ptrs[i] = func_ptr;
+
     instance = module->instance_list;
     while (instance) {
-        instance->func_ptrs[func_idx] = func_ptr;
+        if (instance->e->running_mode == Mode_Multi_Tier_JIT)
+            instance->func_ptrs[func_idx] = func_ptr;
         instance = instance->e->next;
     }
     os_mutex_unlock(&module->instance_list_lock);

core/iwasm/include/aot_export.h

@@ -55,6 +55,7 @@ typedef struct AOTCompOption {
     bool enable_aux_stack_frame;
     bool disable_llvm_intrinsics;
     bool disable_llvm_lto;
+    bool enable_stack_estimation;
     uint32_t opt_level;
     uint32_t size_level;
     uint32_t output_format;

core/iwasm/include/wasm_c_api.h

@@ -354,6 +354,7 @@ WASM_API_EXTERN own wasm_importtype_t* wasm_importtype_new(
 WASM_API_EXTERN const wasm_name_t* wasm_importtype_module(const wasm_importtype_t*);
 WASM_API_EXTERN const wasm_name_t* wasm_importtype_name(const wasm_importtype_t*);
 WASM_API_EXTERN const wasm_externtype_t* wasm_importtype_type(const wasm_importtype_t*);
+WASM_API_EXTERN bool wasm_importtype_is_linked(const wasm_importtype_t*);
 
 
 // Export Types
@@ -797,6 +798,9 @@ static inline void* wasm_val_ptr(const wasm_val_t* val) {
 
 #define KILOBYTE(n) ((n) * 1024)
 
+// Create placeholders filled in `wasm_externvec_t* imports` for `wasm_instance_new()`
+WASM_API_EXTERN wasm_extern_t *wasm_extern_new_empty(wasm_store_t *,  wasm_externkind_t);
+
 ///////////////////////////////////////////////////////////////////////////////
 
 #undef own

core/iwasm/include/wasm_export.h

@@ -131,6 +131,14 @@ typedef struct mem_alloc_info_t {
     uint32_t highmark_size;
 } mem_alloc_info_t;
 
+/* Running mode of runtime and module instance*/
+typedef enum RunningMode {
+    Mode_Interp = 1,
+    Mode_Fast_JIT,
+    Mode_LLVM_JIT,
+    Mode_Multi_Tier_JIT,
+} RunningMode;
+
 /* WASM runtime initialize arguments */
 typedef struct RuntimeInitArgs {
     mem_alloc_type_t mem_alloc_type;
@@ -152,6 +160,13 @@ typedef struct RuntimeInitArgs {
 
     /* Fast JIT code cache size */
     uint32_t fast_jit_code_cache_size;
+
+    /* Default running mode of the runtime */
+    RunningMode running_mode;
+
+    /* LLVM JIT opt and size level */
+    uint32_t llvm_jit_opt_level;
+    uint32_t llvm_jit_size_level;
 } RuntimeInitArgs;
 
 #ifndef WASM_VALKIND_T_DEFINED
@@ -195,9 +210,9 @@ WASM_RUNTIME_API_EXTERN bool
 wasm_runtime_init(void);
 
 /**
- * Initialize the WASM runtime environment, and also initialize
+ * Initialize the WASM runtime environment, WASM running mode,
- * the memory allocator and register native symbols, which are specified
+ * and also initialize the memory allocator and register native symbols,
- * with init arguments
+ * which are specified with init arguments
  *
  * @param init_args specifies the init arguments
  *
@@ -206,6 +221,28 @@ wasm_runtime_init(void);
 WASM_RUNTIME_API_EXTERN bool
 wasm_runtime_full_init(RuntimeInitArgs *init_args);
 
+/**
+ * Query whether a certain running mode is supported for the runtime
+ *
+ * @param running_mode the running mode to query
+ *
+ * @return true if this running mode is supported, false otherwise
+ */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_is_running_mode_supported(RunningMode running_mode);
+
+/**
+ * Set the default running mode for the runtime. It is inherited
+ * to set the running mode of a module instance when it is instantiated,
+ * and can be changed by calling wasm_runtime_set_running_mode
+ *
+ * @param running_mode the running mode to set
+ *
+ * @return true if success, false otherwise
+ */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_set_default_running_mode(RunningMode running_mode);
+
 /**
  * Destroy the WASM runtime environment.
  */
@@ -450,6 +487,34 @@ wasm_runtime_instantiate(const wasm_module_t module,
                          uint32_t stack_size, uint32_t heap_size,
                          char *error_buf, uint32_t error_buf_size);
 
+/**
+ * Set the running mode of a WASM module instance, override the
+ * default running mode of the runtime. Note that it only makes sense when
+ * the input is a wasm bytecode file: for the AOT file, runtime always runs
+ * it with AOT engine, and this function always returns true.
+ *
+ * @param module_inst the WASM module instance to set running mode
+ * @param running_mode the running mode to set
+ *
+ * @return true if success, false otherwise
+ */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_set_running_mode(wasm_module_inst_t module_inst,
+                              RunningMode running_mode);
+
+/**
+ * Get the running mode of a WASM module instance, if no running mode
+ * is explicitly set the default running mode of runtime will
+ * be used and returned. Note that it only makes sense when the input is a
+ * wasm bytecode file: for the AOT file, this function always returns 0.
+ *
+ * @param module_inst the WASM module instance to query for running mode
+ *
+ * @return the running mode this module instance currently use
+ */
+WASM_RUNTIME_API_EXTERN RunningMode
+wasm_runtime_get_running_mode(wasm_module_inst_t module_inst);
+
 /**
  * Deinstantiate a WASM module instance, destroy the resources.
  *
@@ -1259,6 +1324,22 @@ wasm_runtime_get_custom_section(wasm_module_t const module_comm,
  */
 WASM_RUNTIME_API_EXTERN void
 wasm_runtime_get_version(uint32_t *major, uint32_t *minor, uint32_t *patch);
+
+/**
+ * Check whether an import func `(import <module_name> <func_name> (func ...))` is linked or not
+ * with runtime registered natvie functions
+ */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_is_import_func_linked(const char *module_name,
+                                   const char *func_name);
+
+/**
+ * Check whether an import global `(import <module_name> <global_name> (global ...))` is linked or not
+ * with runtime registered natvie globals
+ */
+WASM_RUNTIME_API_EXTERN bool
+wasm_runtime_is_import_global_linked(const char *module_name,
+                                     const char *global_name);
 /* clang-format on */
 
 #ifdef __cplusplus

core/iwasm/interpreter/wasm.h

@@ -278,9 +278,14 @@ struct WASMFunction {
 #endif
 
 #if WASM_ENABLE_FAST_JIT != 0
+    /* The compiled fast jit jitted code block of this function */
     void *fast_jit_jitted_code;
 #if WASM_ENABLE_JIT != 0 && WASM_ENABLE_LAZY_JIT != 0
+    /* The compiled llvm jit func ptr of this function */
     void *llvm_jit_func_ptr;
+    /* Code block to call fast jit jitted code of this function
+       from the llvm jit jitted code */
+    void *call_to_fast_jit_from_llvm_jit;
 #endif
 #endif
 };
@@ -505,15 +510,14 @@ struct WASMModule {
     uint64 load_size;
 #endif
 
-#if WASM_ENABLE_DEBUG_INTERP != 0                    \
+#if WASM_ENABLE_DEBUG_INTERP != 0                         \
-    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT \
+    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
         && WASM_ENABLE_LAZY_JIT != 0)
     /**
      * List of instances referred to this module. When source debugging
      * feature is enabled, the debugger may modify the code section of
      * the module, so we need to report a warning if user create several
-     * instances based on the same module. Sub instances created by
+     * instances based on the same module.
-     * lib-pthread or spawn API won't be added into the list.
      *
      * Also add the instance to the list for Fast JIT to LLVM JIT
      * tier-up, since we need to lazily update the LLVM func pointers
@@ -533,7 +537,22 @@ struct WASMModule {
 #endif
 
 #if WASM_ENABLE_FAST_JIT != 0
-    /* func pointers of Fast JITed (un-imported) functions */
+    /**
+     * func pointers of Fast JITed (un-imported) functions
+     * for non Multi-Tier JIT mode:
+     *   (1) when lazy jit is disabled, each pointer is set to the compiled
+     *       fast jit jitted code
+     *   (2) when lazy jit is enabled, each pointer is firstly inited as
+     *       jit_global->compile_fast_jit_and_then_call, and then set to the
+     *       compiled fast jit jitted code when it is called (the stub will
+     *       compile the jit function and then update itself)
+     * for Multi-Tier JIT mode:
+     *   each pointer is firstly inited as compile_fast_jit_and_then_call,
+     *   and then set to the compiled fast jit jitted code when it is called,
+     *   and when the llvm jit func ptr of the same function is compiled, it
+     *   will be set to call_to_llvm_jit_from_fast_jit of this function type
+     *   (tier-up from fast-jit to llvm-jit)
+     */
     void **fast_jit_func_ptrs;
     /* locks for Fast JIT lazy compilation */
     korp_mutex fast_jit_thread_locks[WASM_ORC_JIT_BACKEND_THREAD_NUM];
@@ -543,7 +562,16 @@ struct WASMModule {
 #if WASM_ENABLE_JIT != 0
     struct AOTCompData *comp_data;
     struct AOTCompContext *comp_ctx;
-    /* func pointers of LLVM JITed (un-imported) functions */
+    /**
+     * func pointers of LLVM JITed (un-imported) functions
+     * for non Multi-Tier JIT mode:
+     *   each pointer is set to the lookuped llvm jit func ptr, note that it
+     *   is a stub and will trigger the actual compilation when it is called
+     * for Multi-Tier JIT mode:
+     *   each pointer is inited as call_to_fast_jit code block, when the llvm
+     *   jit func ptr is actually compiled, it is set to the compiled llvm jit
+     *   func ptr
+     */
     void **func_ptrs;
     /* whether the func pointers are compiled */
     bool *func_ptrs_compiled;
@@ -568,6 +596,12 @@ struct WASMModule {
     korp_tid llvm_jit_init_thread;
     /* whether the llvm jit is initialized */
     bool llvm_jit_inited;
+    /* Whether to enable llvm jit compilation:
+       it is set to true only when there is a module instance starts to
+       run with running mode Mode_LLVM_JIT or Mode_Multi_Tier_JIT,
+       since no need to enable llvm jit compilation for Mode_Interp and
+       Mode_Fast_JIT, so as to improve performance for them */
+    bool enable_llvm_jit_compilation;
 #endif
 };
 

core/iwasm/interpreter/wasm_interp_classic.c

@@ -696,28 +696,28 @@ trunc_f64_to_int(WASMModuleInstance *module, uint32 *frame_sp, float64 src_min,
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint32)(*(uint8 *)maddr);                       \
             *(uint8 *)maddr = (uint8)(readv op sval);                \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else if (opcode == WASM_OP_ATOMIC_RMW_I32_##OP_NAME##16_U) { \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint32)LOAD_U16(maddr);                         \
             STORE_U16(maddr, (uint16)(readv op sval));               \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else {                                                       \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = LOAD_I32(maddr);                                 \
             STORE_U32(maddr, readv op sval);                         \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         PUSH_I32(readv);                                             \
         break;                                                       \
@@ -736,39 +736,39 @@ trunc_f64_to_int(WASMModuleInstance *module, uint32 *frame_sp, float64 src_min,
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)(*(uint8 *)maddr);                       \
             *(uint8 *)maddr = (uint8)(readv op sval);                \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else if (opcode == WASM_OP_ATOMIC_RMW_I64_##OP_NAME##16_U) { \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)LOAD_U16(maddr);                         \
             STORE_U16(maddr, (uint16)(readv op sval));               \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else if (opcode == WASM_OP_ATOMIC_RMW_I64_##OP_NAME##32_U) { \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)LOAD_U32(maddr);                         \
             STORE_U32(maddr, (uint32)(readv op sval));               \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else {                                                       \
             uint64 op_result;                                        \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS();                            \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)LOAD_I64(maddr);                         \
             op_result = readv op sval;                               \
             STORE_I64(maddr, op_result);                             \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         PUSH_I64(readv);                                             \
         break;                                                       \
@@ -1151,6 +1151,11 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
     uint32 cache_index, type_index, param_cell_num, cell_num;
     uint8 value_type;
 
+#if WASM_ENABLE_SHARED_MEMORY != 0
+    WASMSharedMemNode *node =
+        wasm_module_get_shared_memory((WASMModuleCommon *)module->module);
+#endif
+
 #if WASM_ENABLE_DEBUG_INTERP != 0
     uint8 *frame_ip_orig = NULL;
     WASMDebugInstance *debug_instance = wasm_exec_env_get_instance(exec_env);
@@ -3458,23 +3463,23 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I32_LOAD8_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)(*(uint8 *)maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I32_LOAD16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)LOAD_U16(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = LOAD_I32(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
 
                         PUSH_I32(readv);
@@ -3493,30 +3498,30 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I64_LOAD8_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)(*(uint8 *)maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_LOAD16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U16(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_LOAD32_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U32(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = LOAD_I64(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
 
                         PUSH_I64(readv);
@@ -3535,23 +3540,23 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I32_STORE8) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             *(uint8 *)maddr = (uint8)sval;
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I32_STORE16) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U16(maddr, (uint16)sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U32(maddr, frame_sp[1]);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         break;
                     }
@@ -3569,31 +3574,31 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I64_STORE8) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             *(uint8 *)maddr = (uint8)sval;
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_STORE16) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U16(maddr, (uint16)sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_STORE32) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U32(maddr, (uint32)sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             PUT_I64_TO_ADDR((uint32 *)maddr,
                                             GET_I64_FROM_ADDR(frame_sp + 1));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         break;
                     }
@@ -3613,32 +3618,32 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
                             expect = (uint8)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)(*(uint8 *)maddr);
                             if (readv == expect)
                                 *(uint8 *)maddr = (uint8)(sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_RMW_I32_CMPXCHG16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
                             expect = (uint16)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)LOAD_U16(maddr);
                             if (readv == expect)
                                 STORE_U16(maddr, (uint16)(sval));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = LOAD_I32(maddr);
                             if (readv == expect)
                                 STORE_U32(maddr, sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         PUSH_I32(readv);
                         break;
@@ -3659,44 +3664,44 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
                             expect = (uint8)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)(*(uint8 *)maddr);
                             if (readv == expect)
                                 *(uint8 *)maddr = (uint8)(sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_RMW_I64_CMPXCHG16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
                             expect = (uint16)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U16(maddr);
                             if (readv == expect)
                                 STORE_U16(maddr, (uint16)(sval));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_RMW_I64_CMPXCHG32_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
                             expect = (uint32)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U32(maddr);
                             if (readv == expect)
                                 STORE_U32(maddr, (uint32)(sval));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS();
 
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_I64(maddr);
                             if (readv == expect) {
                                 STORE_I64(maddr, sval);
                             }
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         PUSH_I64(readv);
                         break;
@@ -4150,6 +4155,7 @@ wasm_interp_call_wasm(WASMModuleInstance *module_inst, WASMExecEnv *exec_env,
     }
     argc = function->param_cell_num;
 
+    RECORD_STACK_USAGE(exec_env, (uint8 *)&prev_frame);
 #if !(defined(OS_ENABLE_HW_BOUND_CHECK) \
       && WASM_DISABLE_STACK_HW_BOUND_CHECK == 0)
     if ((uint8 *)&prev_frame < exec_env->native_stack_boundary) {
@@ -4194,58 +4200,51 @@ wasm_interp_call_wasm(WASMModuleInstance *module_inst, WASMExecEnv *exec_env,
         }
     }
     else {
-#if WASM_ENABLE_LAZY_JIT != 0
+        RunningMode running_mode =
+            wasm_runtime_get_running_mode((wasm_module_inst_t)module_inst);
 
-        /* Fast JIT to LLVM JIT tier-up is enabled */
+        if (running_mode == Mode_Interp) {
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0
+            wasm_interp_call_func_bytecode(module_inst, exec_env, function,
-        /* Fast JIT and LLVM JIT are both enabled, call llvm jit function
+                                           frame);
-           if it is compiled, else call fast jit function */
+        }
-        uint32 func_idx = (uint32)(function - module_inst->e->functions);
+#if WASM_ENABLE_FAST_JIT != 0
-        if (module_inst->module->func_ptrs_compiled
+        else if (running_mode == Mode_Fast_JIT) {
-                [func_idx - module_inst->module->import_function_count]) {
+            fast_jit_call_func_bytecode(module_inst, exec_env, function, frame);
+        }
+#endif
+#if WASM_ENABLE_JIT != 0
+        else if (running_mode == Mode_LLVM_JIT) {
             llvm_jit_call_func_bytecode(module_inst, exec_env, function, argc,
                                         argv);
             /* For llvm jit, the results have been stored in argv,
                no need to copy them from stack frame again */
             copy_argv_from_frame = false;
         }
-        else {
+#endif
-            fast_jit_call_func_bytecode(module_inst, exec_env, function, frame);
+#if WASM_ENABLE_LAZY_JIT != 0 && WASM_ENABLE_FAST_JIT != 0 \
+    && WASM_ENABLE_JIT != 0
+        else if (running_mode == Mode_Multi_Tier_JIT) {
+            /* Tier-up from Fast JIT to LLVM JIT, call llvm jit function
+               if it is compiled, else call fast jit function */
+            uint32 func_idx = (uint32)(function - module_inst->e->functions);
+            if (module_inst->module->func_ptrs_compiled
+                    [func_idx - module_inst->module->import_function_count]) {
+                llvm_jit_call_func_bytecode(module_inst, exec_env, function,
+                                            argc, argv);
+                /* For llvm jit, the results have been stored in argv,
+                   no need to copy them from stack frame again */
+                copy_argv_from_frame = false;
+            }
+            else {
+                fast_jit_call_func_bytecode(module_inst, exec_env, function,
+                                            frame);
+            }
         }
-#elif WASM_ENABLE_JIT != 0
-        /* Only LLVM JIT is enabled */
-        llvm_jit_call_func_bytecode(module_inst, exec_env, function, argc,
-                                    argv);
-        /* For llvm jit, the results have been stored in argv,
-           no need to copy them from stack frame again */
-        copy_argv_from_frame = false;
-#elif WASM_ENABLE_FAST_JIT != 0
-        /* Only Fast JIT is enabled */
-        fast_jit_call_func_bytecode(module_inst, exec_env, function, frame);
-#else
-        /* Both Fast JIT and LLVM JIT are disabled */
-        wasm_interp_call_func_bytecode(module_inst, exec_env, function, frame);
 #endif
-
+        else {
-#else /* else of WASM_ENABLE_LAZY_JIT != 0 */
+            /* There should always be a supported running mode selected */
-
+            bh_assert(0);
-        /* Fast JIT to LLVM JIT tier-up is enabled */
+        }
-#if WASM_ENABLE_JIT != 0
-        /* LLVM JIT is enabled */
-        llvm_jit_call_func_bytecode(module_inst, exec_env, function, argc,
-                                    argv);
-        /* For llvm jit, the results have been stored in argv,
-           no need to copy them from stack frame again */
-        copy_argv_from_frame = false;
-#elif WASM_ENABLE_FAST_JIT != 0
-        /* Fast JIT is enabled */
-        fast_jit_call_func_bytecode(module_inst, exec_env, function, frame);
-#else
-        /* Both Fast JIT and LLVM JIT are disabled */
-        wasm_interp_call_func_bytecode(module_inst, exec_env, function, frame);
-#endif
-
-#endif /* end of WASM_ENABLE_LAZY_JIT != 0 */
 
         (void)wasm_interp_call_func_bytecode;
 #if WASM_ENABLE_FAST_JIT != 0

core/iwasm/interpreter/wasm_interp_fast.c

@@ -469,28 +469,28 @@ LOAD_PTR(void *addr)
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(1);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint32)(*(uint8 *)maddr);                       \
             *(uint8 *)maddr = (uint8)(readv op sval);                \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else if (opcode == WASM_OP_ATOMIC_RMW_I32_##OP_NAME##16_U) { \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(2);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint32)LOAD_U16(maddr);                         \
             STORE_U16(maddr, (uint16)(readv op sval));               \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else {                                                       \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(4);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = LOAD_I32(maddr);                                 \
             STORE_U32(maddr, readv op sval);                         \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         PUSH_I32(readv);                                             \
         break;                                                       \
@@ -509,39 +509,39 @@ LOAD_PTR(void *addr)
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(1);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)(*(uint8 *)maddr);                       \
             *(uint8 *)maddr = (uint8)(readv op sval);                \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else if (opcode == WASM_OP_ATOMIC_RMW_I64_##OP_NAME##16_U) { \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(2);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)LOAD_U16(maddr);                         \
             STORE_U16(maddr, (uint16)(readv op sval));               \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else if (opcode == WASM_OP_ATOMIC_RMW_I64_##OP_NAME##32_U) { \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(4);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)LOAD_U32(maddr);                         \
             STORE_U32(maddr, (uint32)(readv op sval));               \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         else {                                                       \
             uint64 op_result;                                        \
             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);     \
             CHECK_ATOMIC_MEMORY_ACCESS(8);                           \
                                                                      \
-            os_mutex_lock(&module->e->mem_lock);                     \
+            os_mutex_lock(&node->shared_mem_lock);                   \
             readv = (uint64)LOAD_I64(maddr);                         \
             op_result = readv op sval;                               \
             STORE_I64(maddr, op_result);                             \
-            os_mutex_unlock(&module->e->mem_lock);                   \
+            os_mutex_unlock(&node->shared_mem_lock);                 \
         }                                                            \
         PUSH_I64(readv);                                             \
         break;                                                       \
@@ -1183,6 +1183,11 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
     uint32 local_idx, local_offset, global_idx;
     uint8 opcode, local_type, *global_addr;
 
+#if WASM_ENABLE_SHARED_MEMORY != 0
+    WASMSharedMemNode *node =
+        wasm_module_get_shared_memory((WASMModuleCommon *)module->module);
+#endif
+
 #if WASM_ENABLE_LABELS_AS_VALUES != 0
 #define HANDLE_OPCODE(op) &&HANDLE_##op
     DEFINE_GOTO_TABLE(const void *, handle_table);
@@ -3296,23 +3301,23 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I32_LOAD8_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(1);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)(*(uint8 *)maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I32_LOAD16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(2);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)LOAD_U16(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(4);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = LOAD_I32(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
 
                         PUSH_I32(readv);
@@ -3331,30 +3336,30 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I64_LOAD8_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(1);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)(*(uint8 *)maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_LOAD16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(2);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U16(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_LOAD32_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(4);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U32(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(8);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = LOAD_I64(maddr);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
 
                         PUSH_I64(readv);
@@ -3372,23 +3377,23 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I32_STORE8) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(1);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             *(uint8 *)maddr = (uint8)sval;
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I32_STORE16) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(2);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U16(maddr, (uint16)sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(4);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U32(maddr, sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         break;
                     }
@@ -3406,30 +3411,30 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                         if (opcode == WASM_OP_ATOMIC_I64_STORE8) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 1, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(1);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             *(uint8 *)maddr = (uint8)sval;
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_STORE16) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(2);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U16(maddr, (uint16)sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_I64_STORE32) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(4);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_U32(maddr, (uint32)sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(8);
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             STORE_I64(maddr, sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         break;
                     }
@@ -3449,32 +3454,32 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                             CHECK_ATOMIC_MEMORY_ACCESS(1);
 
                             expect = (uint8)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)(*(uint8 *)maddr);
                             if (readv == expect)
                                 *(uint8 *)maddr = (uint8)(sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_RMW_I32_CMPXCHG16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(2);
 
                             expect = (uint16)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint32)LOAD_U16(maddr);
                             if (readv == expect)
                                 STORE_U16(maddr, (uint16)(sval));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(4);
 
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = LOAD_I32(maddr);
                             if (readv == expect)
                                 STORE_U32(maddr, sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         PUSH_I32(readv);
                         break;
@@ -3495,44 +3500,44 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                             CHECK_ATOMIC_MEMORY_ACCESS(1);
 
                             expect = (uint8)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)(*(uint8 *)maddr);
                             if (readv == expect)
                                 *(uint8 *)maddr = (uint8)(sval);
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_RMW_I64_CMPXCHG16_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 2, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(2);
 
                             expect = (uint16)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U16(maddr);
                             if (readv == expect)
                                 STORE_U16(maddr, (uint16)(sval));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else if (opcode == WASM_OP_ATOMIC_RMW_I64_CMPXCHG32_U) {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 4, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(4);
 
                             expect = (uint32)expect;
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_U32(maddr);
                             if (readv == expect)
                                 STORE_U32(maddr, (uint32)(sval));
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         else {
                             CHECK_BULK_MEMORY_OVERFLOW(addr + offset, 8, maddr);
                             CHECK_ATOMIC_MEMORY_ACCESS(8);
 
-                            os_mutex_lock(&module->e->mem_lock);
+                            os_mutex_lock(&node->shared_mem_lock);
                             readv = (uint64)LOAD_I64(maddr);
                             if (readv == expect) {
                                 STORE_I64(maddr, sval);
                             }
-                            os_mutex_unlock(&module->e->mem_lock);
+                            os_mutex_unlock(&node->shared_mem_lock);
                         }
                         PUSH_I64(readv);
                         break;
@@ -3901,6 +3906,7 @@ wasm_interp_call_wasm(WASMModuleInstance *module_inst, WASMExecEnv *exec_env,
     }
     argc = function->param_cell_num;
 
+    RECORD_STACK_USAGE(exec_env, (uint8 *)&prev_frame);
 #if !(defined(OS_ENABLE_HW_BOUND_CHECK) \
       && WASM_DISABLE_STACK_HW_BOUND_CHECK == 0)
     if ((uint8 *)&prev_frame < exec_env->native_stack_boundary) {

core/iwasm/interpreter/wasm_loader.c

@@ -1399,6 +1399,7 @@ load_global_import(const uint8 **p_buf, const uint8 *buf_end,
     WASMModule *sub_module = NULL;
     WASMGlobal *linked_global = NULL;
 #endif
+    bool ret = false;
 
     CHECK_BUF(p, p_end, 2);
     declare_type = read_uint8(p);
@@ -1411,15 +1412,16 @@ load_global_import(const uint8 **p_buf, const uint8 *buf_end,
     }
 
 #if WASM_ENABLE_LIBC_BUILTIN != 0
-    global->is_linked = wasm_native_lookup_libc_builtin_global(
+    ret = wasm_native_lookup_libc_builtin_global(sub_module_name, global_name,
-        sub_module_name, global_name, global);
+                                                 global);
-    if (global->is_linked) {
+    if (ret) {
         if (global->type != declare_type
             || global->is_mutable != declare_mutable) {
             set_error_buf(error_buf, error_buf_size,
                           "incompatible import type");
             return false;
         }
+        global->is_linked = true;
     }
 #endif
 #if WASM_ENABLE_MULTI_MODULE != 0
@@ -1449,6 +1451,7 @@ load_global_import(const uint8 **p_buf, const uint8 *buf_end,
     global->is_mutable = (declare_mutable == 1);
 
     (void)parent_module;
+    (void)ret;
     return true;
 fail:
     return false;
@@ -2989,6 +2992,7 @@ static bool
 init_llvm_jit_functions_stage1(WASMModule *module, char *error_buf,
                                uint32 error_buf_size)
 {
+    LLVMJITOptions llvm_jit_options = wasm_runtime_get_llvm_jit_options();
     AOTCompOption option = { 0 };
     char *aot_last_error;
     uint64 size;
@@ -3027,8 +3031,11 @@ init_llvm_jit_functions_stage1(WASMModule *module, char *error_buf,
     }
 
     option.is_jit_mode = true;
-    option.opt_level = 3;
+
-    option.size_level = 3;
+    llvm_jit_options = wasm_runtime_get_llvm_jit_options();
+    option.opt_level = llvm_jit_options.opt_level;
+    option.size_level = llvm_jit_options.size_level;
+
 #if WASM_ENABLE_BULK_MEMORY != 0
     option.enable_bulk_memory = true;
 #endif
@@ -3048,6 +3055,9 @@ init_llvm_jit_functions_stage1(WASMModule *module, char *error_buf,
 #if (WASM_ENABLE_PERF_PROFILING != 0) || (WASM_ENABLE_DUMP_CALL_STACK != 0)
     option.enable_aux_stack_frame = true;
 #endif
+#if WASM_ENABLE_MEMORY_PROFILING != 0
+    option.enable_stack_estimation = true;
+#endif
 
     module->comp_ctx = aot_create_comp_context(module->comp_data, &option);
     if (!module->comp_ctx) {
@@ -3109,6 +3119,8 @@ init_llvm_jit_functions_stage2(WASMModule *module, char *error_buf,
         module->func_ptrs[i] = (void *)func_addr;
 
 #if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_LAZY_JIT != 0
+        module->functions[i]->llvm_jit_func_ptr = (void *)func_addr;
+
         if (module->orcjit_stop_compiling)
             return false;
 #endif
@@ -3199,9 +3211,9 @@ orcjit_thread_callback(void *arg)
 
     /* Wait until init_llvm_jit_functions_stage2 finishes */
     os_mutex_lock(&module->tierup_wait_lock);
-    while (!module->llvm_jit_inited) {
+    while (!(module->llvm_jit_inited && module->enable_llvm_jit_compilation)) {
         os_cond_reltimedwait(&module->tierup_wait_cond,
-                             &module->tierup_wait_lock, 10);
+                             &module->tierup_wait_lock, 10000);
         if (module->orcjit_stop_compiling) {
             /* init_llvm_jit_functions_stage2 failed */
             os_mutex_unlock(&module->tierup_wait_lock);
@@ -3852,8 +3864,8 @@ create_module(char *error_buf, uint32 error_buf_size)
     bh_assert(ret == BH_LIST_SUCCESS);
 #endif
 
-#if WASM_ENABLE_DEBUG_INTERP != 0                    \
+#if WASM_ENABLE_DEBUG_INTERP != 0                         \
-    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT \
+    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
         && WASM_ENABLE_LAZY_JIT != 0)
     if (os_mutex_init(&module->instance_list_lock) != 0) {
         set_error_buf(error_buf, error_buf_size,
@@ -4148,10 +4160,8 @@ check_wasi_abi_compatibility(const WASMModule *module,
 
     /* should have one at least */
     if (module->import_wasi_api && !start && !initialize) {
-        set_error_buf(
+        LOG_WARNING("warning: a module with WASI apis should be either "
-            error_buf, error_buf_size,
+                    "a command or a reactor");
-            "a module with WASI apis must be either a command or a reactor");
-        return false;
     }
 
     /*
@@ -4256,7 +4266,8 @@ wasm_loader_unload(WASMModule *module)
     if (!module)
         return;
 
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT && WASM_ENABLE_LAZY_JIT != 0
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
     module->orcjit_stop_compiling = true;
     if (module->llvm_jit_init_thread)
         os_thread_join(module->llvm_jit_init_thread, NULL);
@@ -4277,7 +4288,8 @@ wasm_loader_unload(WASMModule *module)
         aot_destroy_comp_data(module->comp_data);
 #endif
 
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT && WASM_ENABLE_LAZY_JIT != 0
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
     if (module->tierup_wait_lock_inited) {
         os_mutex_destroy(&module->tierup_wait_lock);
         os_cond_destroy(&module->tierup_wait_cond);
@@ -4312,9 +4324,9 @@ wasm_loader_unload(WASMModule *module)
                         module->functions[i]->fast_jit_jitted_code);
                 }
 #if WASM_ENABLE_JIT != 0 && WASM_ENABLE_LAZY_JIT != 0
-                if (module->functions[i]->llvm_jit_func_ptr) {
+                if (module->functions[i]->call_to_fast_jit_from_llvm_jit) {
                     jit_code_cache_free(
-                        module->functions[i]->llvm_jit_func_ptr);
+                        module->functions[i]->call_to_fast_jit_from_llvm_jit);
                 }
 #endif
 #endif
@@ -4406,8 +4418,8 @@ wasm_loader_unload(WASMModule *module)
     }
 #endif
 
-#if WASM_ENABLE_DEBUG_INTERP != 0                    \
+#if WASM_ENABLE_DEBUG_INTERP != 0                         \
-    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT \
+    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
         && WASM_ENABLE_LAZY_JIT != 0)
     os_mutex_destroy(&module->instance_list_lock);
 #endif

core/iwasm/interpreter/wasm_mini_loader.c

@@ -1835,6 +1835,7 @@ static bool
 init_llvm_jit_functions_stage1(WASMModule *module, char *error_buf,
                                uint32 error_buf_size)
 {
+    LLVMJITOptions llvm_jit_options = wasm_runtime_get_llvm_jit_options();
     AOTCompOption option = { 0 };
     char *aot_last_error;
     uint64 size;
@@ -1873,8 +1874,9 @@ init_llvm_jit_functions_stage1(WASMModule *module, char *error_buf,
     }
 
     option.is_jit_mode = true;
-    option.opt_level = 3;
+    option.opt_level = llvm_jit_options.opt_level;
-    option.size_level = 3;
+    option.size_level = llvm_jit_options.size_level;
+
 #if WASM_ENABLE_BULK_MEMORY != 0
     option.enable_bulk_memory = true;
 #endif
@@ -1894,6 +1896,9 @@ init_llvm_jit_functions_stage1(WASMModule *module, char *error_buf,
 #if (WASM_ENABLE_PERF_PROFILING != 0) || (WASM_ENABLE_DUMP_CALL_STACK != 0)
     option.enable_aux_stack_frame = true;
 #endif
+#if WASM_ENABLE_MEMORY_PROFILING != 0
+    option.enable_stack_estimation = true;
+#endif
 
     module->comp_ctx = aot_create_comp_context(module->comp_data, &option);
     if (!module->comp_ctx) {
@@ -1957,6 +1962,8 @@ init_llvm_jit_functions_stage2(WASMModule *module, char *error_buf,
         module->func_ptrs[i] = (void *)func_addr;
 
 #if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_LAZY_JIT != 0
+        module->functions[i]->llvm_jit_func_ptr = (void *)func_addr;
+
         if (module->orcjit_stop_compiling)
             return false;
 #endif
@@ -2047,9 +2054,9 @@ orcjit_thread_callback(void *arg)
 
     /* Wait until init_llvm_jit_functions_stage2 finishes */
     os_mutex_lock(&module->tierup_wait_lock);
-    while (!module->llvm_jit_inited) {
+    while (!(module->llvm_jit_inited && module->enable_llvm_jit_compilation)) {
         os_cond_reltimedwait(&module->tierup_wait_cond,
-                             &module->tierup_wait_lock, 10);
+                             &module->tierup_wait_lock, 10000);
         if (module->orcjit_stop_compiling) {
             /* init_llvm_jit_functions_stage2 failed */
             os_mutex_unlock(&module->tierup_wait_lock);
@@ -2718,7 +2725,8 @@ create_module(char *error_buf, uint32 error_buf_size)
     bh_assert(ret == BH_LIST_SUCCESS);
 #endif
 
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT && WASM_ENABLE_LAZY_JIT != 0
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
     if (os_mutex_init(&module->instance_list_lock) != 0) {
         set_error_buf(error_buf, error_buf_size,
                       "init instance list lock failed");
@@ -2939,7 +2947,8 @@ wasm_loader_unload(WASMModule *module)
     if (!module)
         return;
 
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT && WASM_ENABLE_LAZY_JIT != 0
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
     module->orcjit_stop_compiling = true;
     if (module->llvm_jit_init_thread)
         os_thread_join(module->llvm_jit_init_thread, NULL);
@@ -2960,7 +2969,8 @@ wasm_loader_unload(WASMModule *module)
         aot_destroy_comp_data(module->comp_data);
 #endif
 
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT && WASM_ENABLE_LAZY_JIT != 0
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
     if (module->tierup_wait_lock_inited) {
         os_mutex_destroy(&module->tierup_wait_lock);
         os_cond_destroy(&module->tierup_wait_cond);
@@ -2995,9 +3005,9 @@ wasm_loader_unload(WASMModule *module)
                         module->functions[i]->fast_jit_jitted_code);
                 }
 #if WASM_ENABLE_JIT != 0 && WASM_ENABLE_LAZY_JIT != 0
-                if (module->functions[i]->llvm_jit_func_ptr) {
+                if (module->functions[i]->call_to_fast_jit_from_llvm_jit) {
                     jit_code_cache_free(
-                        module->functions[i]->llvm_jit_func_ptr);
+                        module->functions[i]->call_to_fast_jit_from_llvm_jit);
                 }
 #endif
 #endif
@@ -3056,7 +3066,8 @@ wasm_loader_unload(WASMModule *module)
     }
 #endif
 
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT && WASM_ENABLE_LAZY_JIT != 0
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
     os_mutex_destroy(&module->instance_list_lock);
 #endif
 

core/iwasm/interpreter/wasm_runtime.c

@@ -737,13 +737,12 @@ functions_instantiate(const WASMModule *module, WASMModuleInstance *module_inst,
 
         function++;
     }
+    bh_assert((uint32)(function - functions) == function_count);
 
 #if WASM_ENABLE_FAST_JIT != 0
     module_inst->fast_jit_func_ptrs = module->fast_jit_func_ptrs;
 #endif
 
-    bh_assert((uint32)(function - functions) == function_count);
-    (void)module_inst;
     return functions;
 }
 
@@ -1288,9 +1287,8 @@ init_func_ptrs(WASMModuleInstance *module_inst, WASMModule *module,
         *func_ptrs = import_func->func_ptr_linked;
     }
 
-    /* Set defined function pointers */
+    /* The defined function pointers will be set in
-    bh_memcpy_s(func_ptrs, sizeof(void *) * module->function_count,
+       wasm_runtime_set_running_mode, no need to set them here */
-                module->func_ptrs, sizeof(void *) * module->function_count);
     return true;
 }
 #endif /* end of WASM_ENABLE_JIT != 0 */
@@ -1336,6 +1334,173 @@ init_func_type_indexes(WASMModuleInstance *module_inst, char *error_buf,
 }
 #endif /* end of WASM_ENABLE_FAST_JIT != 0 || WASM_ENABLE_JIT != 0 */
 
+static bool
+set_running_mode(WASMModuleInstance *module_inst, RunningMode running_mode,
+                 bool first_time_set)
+{
+    WASMModule *module = module_inst->module;
+
+    if (running_mode == Mode_Default) {
+#if WASM_ENABLE_FAST_JIT == 0 && WASM_ENABLE_JIT == 0
+        running_mode = Mode_Interp;
+#elif WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT == 0
+        running_mode = Mode_Fast_JIT;
+#elif WASM_ENABLE_FAST_JIT == 0 && WASM_ENABLE_JIT != 0
+        running_mode = Mode_LLVM_JIT;
+#else /* WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 */
+#if WASM_ENABLE_LAZY_JIT == 0
+        running_mode = Mode_LLVM_JIT;
+#else
+        running_mode = Mode_Multi_Tier_JIT;
+#endif
+#endif
+    }
+
+    if (!wasm_runtime_is_running_mode_supported(running_mode))
+        return false;
+
+#if !(WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+      && WASM_ENABLE_LAZY_JIT != 0) /* No possible multi-tier JIT */
+    module_inst->e->running_mode = running_mode;
+
+    if (running_mode == Mode_Interp) {
+        /* Do nothing for Mode_Interp */
+    }
+    else if (running_mode == Mode_Fast_JIT) {
+        /* Do nothing for Mode_Fast_JIT since
+           module_inst->fast_jit_func_ptrs is same as
+           module->fast_jit_func_ptrs */
+    }
+#if WASM_ENABLE_JIT != 0
+    else if (running_mode == Mode_LLVM_JIT) {
+        /* Set defined function pointers */
+        bh_memcpy_s(module_inst->func_ptrs + module->import_function_count,
+                    sizeof(void *) * module->function_count, module->func_ptrs,
+                    sizeof(void *) * module->function_count);
+    }
+#endif
+    else {
+        bh_assert(0);
+    }
+#else /* Possible multi-tier JIT */
+    os_mutex_lock(&module->instance_list_lock);
+
+    module_inst->e->running_mode = running_mode;
+
+    if (running_mode == Mode_Interp) {
+        /* Do nothing for Mode_Interp */
+    }
+#if WASM_ENABLE_FAST_JIT != 0
+    else if (running_mode == Mode_Fast_JIT) {
+        JitGlobals *jit_globals = jit_compiler_get_jit_globals();
+        uint32 i;
+
+        /* Allocate memory for fast_jit_func_ptrs if needed */
+        if (!module_inst->fast_jit_func_ptrs
+            || module_inst->fast_jit_func_ptrs == module->fast_jit_func_ptrs) {
+            uint64 total_size = (uint64)sizeof(void *) * module->function_count;
+            if (!(module_inst->fast_jit_func_ptrs =
+                      runtime_malloc(total_size, NULL, 0))) {
+                os_mutex_unlock(&module->instance_list_lock);
+                return false;
+            }
+        }
+
+        for (i = 0; i < module->function_count; i++) {
+            if (module->functions[i]->fast_jit_jitted_code) {
+                /* current fast jit function has been compiled */
+                module_inst->fast_jit_func_ptrs[i] =
+                    module->functions[i]->fast_jit_jitted_code;
+            }
+            else {
+                module_inst->fast_jit_func_ptrs[i] =
+                    jit_globals->compile_fast_jit_and_then_call;
+            }
+        }
+    }
+#endif
+#if WASM_ENABLE_JIT != 0
+    else if (running_mode == Mode_LLVM_JIT) {
+        void **llvm_jit_func_ptrs;
+        uint32 i;
+
+        /* Notify backend threads to start llvm jit compilation */
+        module->enable_llvm_jit_compilation = true;
+
+        /* Wait until llvm jit finishes initialization */
+        os_mutex_lock(&module->tierup_wait_lock);
+        while (!module->llvm_jit_inited) {
+            os_cond_reltimedwait(&module->tierup_wait_cond,
+                                 &module->tierup_wait_lock, 10);
+            if (module->orcjit_stop_compiling) {
+                /* init_llvm_jit_functions_stage2 failed */
+                os_mutex_unlock(&module->tierup_wait_lock);
+                os_mutex_unlock(&module->instance_list_lock);
+                return false;
+            }
+        }
+        os_mutex_unlock(&module->tierup_wait_lock);
+
+        llvm_jit_func_ptrs =
+            module_inst->func_ptrs + module->import_function_count;
+        for (i = 0; i < module->function_count; i++) {
+            llvm_jit_func_ptrs[i] = module->functions[i]->llvm_jit_func_ptr;
+        }
+    }
+#endif
+    else if (running_mode == Mode_Multi_Tier_JIT) {
+        /* Notify backend threads to start llvm jit compilation */
+        module->enable_llvm_jit_compilation = true;
+
+        /* Free fast_jit_func_ptrs if it is allocated before */
+        if (module_inst->fast_jit_func_ptrs
+            && module_inst->fast_jit_func_ptrs != module->fast_jit_func_ptrs) {
+            wasm_runtime_free(module_inst->fast_jit_func_ptrs);
+        }
+        module_inst->fast_jit_func_ptrs = module->fast_jit_func_ptrs;
+
+        /* Copy all llvm jit func ptrs from the module */
+        bh_memcpy_s(module_inst->func_ptrs + module->import_function_count,
+                    sizeof(void *) * module->function_count, module->func_ptrs,
+                    sizeof(void *) * module->function_count);
+    }
+    else {
+        bh_assert(0);
+    }
+
+    /* Add module instance into module's instance list if not added */
+    if (first_time_set) {
+        bool found = false;
+        WASMModuleInstance *node = module->instance_list;
+
+        while (node) {
+            if (node == module_inst) {
+                found = true;
+                break;
+            }
+            node = node->e->next;
+        }
+
+        if (!found) {
+            module_inst->e->next = module->instance_list;
+            module->instance_list = module_inst;
+        }
+    }
+
+    os_mutex_unlock(&module->instance_list_lock);
+#endif /* end of !(WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+                   && WASM_ENABLE_LAZY_JIT != 0) */
+
+    (void)module;
+    return true;
+}
+
+bool
+wasm_set_running_mode(WASMModuleInstance *module_inst, RunningMode running_mode)
+{
+    return set_running_mode(module_inst, running_mode, false);
+}
+
 /**
  * Instantiate module
  */
@@ -1422,15 +1587,6 @@ wasm_instantiate(WASMModule *module, bool is_sub_inst, uint32 stack_size,
     module_inst->e =
         (WASMModuleInstanceExtra *)((uint8 *)module_inst + extra_info_offset);
 
-#if WASM_ENABLE_SHARED_MEMORY != 0
-    if (os_mutex_init(&module_inst->e->mem_lock) != 0) {
-        set_error_buf(error_buf, error_buf_size,
-                      "create shared memory lock failed");
-        goto fail;
-    }
-    module_inst->e->mem_lock_inited = true;
-#endif
-
 #if WASM_ENABLE_MULTI_MODULE != 0
     module_inst->e->sub_module_inst_list =
         &module_inst->e->sub_module_inst_list_head;
@@ -1803,33 +1959,39 @@ wasm_instantiate(WASMModule *module, bool is_sub_inst, uint32 stack_size,
     }
 #endif
 
-#if WASM_ENABLE_DEBUG_INTERP != 0                         \
+#if WASM_ENABLE_WASI_NN != 0
-    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    if (!is_sub_inst) {
-        && WASM_ENABLE_LAZY_JIT != 0)
+        if (!(module_inst->e->wasi_nn_ctx = wasi_nn_initialize())) {
+            set_error_buf(error_buf, error_buf_size,
+                          "wasi nn initialization failed");
+            goto fail;
+        }
+    }
+#endif
+
+#if WASM_ENABLE_DEBUG_INTERP != 0
     if (!is_sub_inst) {
         /* Add module instance into module's instance list */
         os_mutex_lock(&module->instance_list_lock);
-#if WASM_ENABLE_DEBUG_INTERP != 0
         if (module->instance_list) {
             LOG_WARNING(
                 "warning: multiple instances referencing to the same module "
                 "may cause unexpected behaviour during debugging");
         }
-#endif
-#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
-    && WASM_ENABLE_LAZY_JIT != 0
-        /* Copy llvm func ptrs again in case that they were updated
-           after the module instance was created */
-        bh_memcpy_s(module_inst->func_ptrs + module->import_function_count,
-                    sizeof(void *) * module->function_count, module->func_ptrs,
-                    sizeof(void *) * module->function_count);
-#endif
         module_inst->e->next = module->instance_list;
         module->instance_list = module_inst;
         os_mutex_unlock(&module->instance_list_lock);
     }
 #endif
 
+    /* Set running mode before executing wasm functions */
+    if (!set_running_mode(module_inst, wasm_runtime_get_default_running_mode(),
+                          true)) {
+        set_error_buf(error_buf, error_buf_size,
+                      "set instance running mode failed");
+        goto fail;
+    }
+
     if (module->start_function != (uint32)-1) {
         /* TODO: fix start function can be import function issue */
         if (module->start_function >= module->import_function_count)
@@ -1895,11 +2057,48 @@ wasm_deinstantiate(WASMModuleInstance *module_inst, bool is_sub_inst)
     if (!module_inst)
         return;
 
+#if WASM_ENABLE_DEBUG_INTERP != 0                         \
+    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+        && WASM_ENABLE_LAZY_JIT != 0)
+    /* Remove instance from module's instance list before freeing
+       func_ptrs and fast_jit_func_ptrs of the instance, to avoid
+       accessing the freed memory in the jit backend compilation
+       threads */
+    if (!is_sub_inst) {
+        WASMModule *module = module_inst->module;
+        WASMModuleInstance *instance_prev = NULL, *instance;
+        os_mutex_lock(&module->instance_list_lock);
+
+        instance = module->instance_list;
+        while (instance) {
+            if (instance == module_inst) {
+                if (!instance_prev)
+                    module->instance_list = instance->e->next;
+                else
+                    instance_prev->e->next = instance->e->next;
+                break;
+            }
+            instance_prev = instance;
+            instance = instance->e->next;
+        }
+
+        os_mutex_unlock(&module->instance_list_lock);
+    }
+#endif
+
 #if WASM_ENABLE_JIT != 0
     if (module_inst->func_ptrs)
         wasm_runtime_free(module_inst->func_ptrs);
 #endif
 
+#if WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
+    && WASM_ENABLE_LAZY_JIT != 0
+    if (module_inst->fast_jit_func_ptrs
+        && module_inst->fast_jit_func_ptrs
+               != module_inst->module->fast_jit_func_ptrs)
+        wasm_runtime_free(module_inst->fast_jit_func_ptrs);
+#endif
+
 #if WASM_ENABLE_FAST_JIT != 0 || WASM_ENABLE_JIT != 0
     if (module_inst->func_type_indexes)
         wasm_runtime_free(module_inst->func_type_indexes);
@@ -1951,39 +2150,17 @@ wasm_deinstantiate(WASMModuleInstance *module_inst, bool is_sub_inst)
     }
 #endif
 
-#if WASM_ENABLE_DEBUG_INTERP != 0                         \
-    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
-        && WASM_ENABLE_LAZY_JIT != 0)
-    if (!is_sub_inst) {
-        WASMModule *module = module_inst->module;
-        WASMModuleInstance *instance_prev = NULL, *instance;
-        os_mutex_lock(&module->instance_list_lock);
-
-        instance = module->instance_list;
-        while (instance) {
-            if (instance == module_inst) {
-                if (!instance_prev)
-                    module->instance_list = instance->e->next;
-                else
-                    instance_prev->e->next = instance->e->next;
-                break;
-            }
-            instance_prev = instance;
-            instance = instance->e->next;
-        }
-
-        os_mutex_unlock(&module->instance_list_lock);
-    }
-#endif
-
-#if WASM_ENABLE_SHARED_MEMORY != 0
-    if (module_inst->e->mem_lock_inited)
-        os_mutex_destroy(&module_inst->e->mem_lock);
-#endif
-
     if (module_inst->e->c_api_func_imports)
         wasm_runtime_free(module_inst->e->c_api_func_imports);
 
+#if WASM_ENABLE_WASI_NN != 0
+    if (!is_sub_inst) {
+        WASINNContext *wasi_nn_ctx = module_inst->e->wasi_nn_ctx;
+        if (wasi_nn_ctx)
+            wasi_nn_destroy(wasi_nn_ctx);
+    }
+#endif
+
     wasm_runtime_free(module_inst);
 }
 
@@ -2056,6 +2233,7 @@ call_wasm_with_hw_bound_check(WASMModuleInstance *module_inst,
     /* Check native stack overflow firstly to ensure we have enough
        native stack to run the following codes before actually calling
        the aot function in invokeNative function. */
+    RECORD_STACK_USAGE(exec_env, (uint8 *)&exec_env_tls);
     if ((uint8 *)&exec_env_tls < exec_env->native_stack_boundary
                                      + page_size * (guard_page_count + 1)) {
         wasm_set_exception(module_inst, "native stack overflow");

core/iwasm/interpreter/wasm_runtime.h

@@ -11,6 +11,10 @@
 #include "../common/wasm_runtime_common.h"
 #include "../common/wasm_exec_env.h"
 
+#if WASM_ENABLE_WASI_NN != 0
+#include "../libraries/wasi-nn/src/wasi_nn_private.h"
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -59,9 +63,7 @@ typedef enum WASMExceptionID {
     EXCE_AUX_STACK_UNDERFLOW,
     EXCE_OUT_OF_BOUNDS_TABLE_ACCESS,
     EXCE_OPERAND_STACK_OVERFLOW,
-#if WASM_ENABLE_FAST_JIT != 0
     EXCE_FAILED_TO_COMPILE_FAST_JIT_FUNC,
-#endif
     EXCE_ALREADY_THROWN,
     EXCE_NUM,
 } WASMExceptionID;
@@ -219,12 +221,7 @@ typedef struct WASMModuleInstanceExtra {
     WASMFunctionInstance *retain_function;
 
     CApiFuncImport *c_api_func_imports;
-
+    RunningMode running_mode;
-#if WASM_ENABLE_SHARED_MEMORY != 0
-    /* lock for shared memory atomic operations */
-    korp_mutex mem_lock;
-    bool mem_lock_inited;
-#endif
 
 #if WASM_ENABLE_MULTI_MODULE != 0
     bh_list sub_module_inst_list_head;
@@ -237,11 +234,15 @@ typedef struct WASMModuleInstanceExtra {
     uint32 max_aux_stack_used;
 #endif
 
-#if WASM_ENABLE_DEBUG_INTERP != 0                    \
+#if WASM_ENABLE_DEBUG_INTERP != 0                         \
-    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT \
+    || (WASM_ENABLE_FAST_JIT != 0 && WASM_ENABLE_JIT != 0 \
         && WASM_ENABLE_LAZY_JIT != 0)
     WASMModuleInstance *next;
 #endif
+
+#if WASM_ENABLE_WASI_NN != 0
+    WASINNContext *wasi_nn_ctx;
+#endif
 } WASMModuleInstanceExtra;
 
 struct AOTFuncPerfProfInfo;
@@ -298,7 +299,11 @@ struct WASMModuleInstance {
        not available in AOTModuleInstance */
     DefPointer(void **, import_func_ptrs);
     /* Array of function pointers to fast jit functions,
-       not available in AOTModuleInstance */
+       not available in AOTModuleInstance:
+       Only when the multi-tier JIT macros are all enabled and the running
+       mode of current module instance is set to Mode_Fast_JIT, runtime
+       will allocate new memory for it, otherwise it always points to the
+       module->fast_jit_func_ptrs */
     DefPointer(void **, fast_jit_func_ptrs);
     /* The custom data that can be set/get by wasm_{get|set}_custom_data */
     DefPointer(void *, custom_data);
@@ -402,6 +407,10 @@ wasm_dump_perf_profiling(const WASMModuleInstance *module_inst);
 void
 wasm_deinstantiate(WASMModuleInstance *module_inst, bool is_sub_inst);
 
+bool
+wasm_set_running_mode(WASMModuleInstance *module_inst,
+                      RunningMode running_mode);
+
 WASMFunctionInstance *
 wasm_lookup_function(const WASMModuleInstance *module_inst, const char *name,
                      const char *signature);

core/iwasm/libraries/lib-pthread/lib_pthread_wrapper.c

@@ -565,6 +565,7 @@ pthread_create_wrapper(wasm_exec_env_t exec_env,
 #if WASM_ENABLE_LIBC_WASI != 0
     WASIContext *wasi_ctx;
 #endif
+    CApiFuncImport **new_c_api_func_imports = NULL;
 
     bh_assert(module);
     bh_assert(module_inst);
@@ -597,6 +598,46 @@ pthread_create_wrapper(wasm_exec_env_t exec_env,
         wasm_runtime_set_wasi_ctx(new_module_inst, wasi_ctx);
 #endif
 
+    /* workaround about passing instantiate-linking information */
+    {
+        CApiFuncImport *c_api_func_imports;
+        uint32 import_func_count = 0;
+        uint32 size_in_bytes = 0;
+
+#if WASM_ENABLE_INTERP != 0
+        if (module_inst->module_type == Wasm_Module_Bytecode) {
+            new_c_api_func_imports = &(
+                ((WASMModuleInstance *)new_module_inst)->e->c_api_func_imports);
+            c_api_func_imports =
+                ((WASMModuleInstance *)module_inst)->e->c_api_func_imports;
+            import_func_count = ((WASMModule *)module)->import_function_count;
+        }
+#endif
+#if WASM_ENABLE_AOT != 0
+        if (module_inst->module_type == Wasm_Module_AoT) {
+            AOTModuleInstanceExtra *e =
+                (AOTModuleInstanceExtra *)((AOTModuleInstance *)new_module_inst)
+                    ->e;
+            new_c_api_func_imports = &(e->c_api_func_imports);
+
+            e = (AOTModuleInstanceExtra *)((AOTModuleInstance *)module_inst)->e;
+            c_api_func_imports = e->c_api_func_imports;
+
+            import_func_count = ((AOTModule *)module)->import_func_count;
+        }
+#endif
+
+        if (import_func_count != 0 && c_api_func_imports) {
+            size_in_bytes = sizeof(CApiFuncImport *) * import_func_count;
+            *new_c_api_func_imports = wasm_runtime_malloc(size_in_bytes);
+            if (!(*new_c_api_func_imports))
+                goto fail;
+
+            bh_memcpy_s(*new_c_api_func_imports, size_in_bytes,
+                        c_api_func_imports, size_in_bytes);
+        }
+    }
+
     if (!(info_node = wasm_runtime_malloc(sizeof(ThreadInfoNode))))
         goto fail;
 

core/iwasm/libraries/libc-wasi/libc_wasi_wrapper.c

@@ -63,6 +63,12 @@ typedef struct WASIContext {
 wasi_ctx_t
 wasm_runtime_get_wasi_ctx(wasm_module_inst_t module_inst);
 
+static inline size_t
+min(size_t a, size_t b)
+{
+    return a > b ? b : a;
+}
+
 static inline struct fd_table *
 wasi_ctx_get_curfds(wasm_module_inst_t module_inst, wasi_ctx_t wasi_ctx)
 {
@@ -951,6 +957,97 @@ wasi_path_remove_directory(wasm_exec_env_t exec_env, wasi_fd_t fd,
     return wasmtime_ssp_path_remove_directory(curfds, fd, path, path_len);
 }
 
+#if WASM_ENABLE_THREAD_MGR != 0
+static __wasi_timestamp_t
+get_timeout_for_poll_oneoff(const wasi_subscription_t *in,
+                            uint32 nsubscriptions)
+{
+    __wasi_timestamp_t timeout = (__wasi_timestamp_t)-1;
+    uint32 i = 0;
+
+    for (i = 0; i < nsubscriptions; ++i) {
+        const __wasi_subscription_t *s = &in[i];
+        if (s->u.type == __WASI_EVENTTYPE_CLOCK
+            && (s->u.u.clock.flags & __WASI_SUBSCRIPTION_CLOCK_ABSTIME) == 0) {
+            timeout = min(timeout, s->u.u.clock.timeout);
+        }
+    }
+    return timeout;
+}
+
+static void
+update_clock_subscription_data(wasi_subscription_t *in, uint32 nsubscriptions,
+                               const wasi_timestamp_t new_timeout)
+{
+    uint32 i = 0;
+    for (i = 0; i < nsubscriptions; ++i) {
+        __wasi_subscription_t *s = &in[i];
+        if (s->u.type == __WASI_EVENTTYPE_CLOCK) {
+            s->u.u.clock.timeout = new_timeout;
+        }
+    }
+}
+
+static wasi_errno_t
+execute_interruptible_poll_oneoff(wasm_module_inst_t module_inst,
+#if !defined(WASMTIME_SSP_STATIC_CURFDS)
+                                  struct fd_table *curfds,
+#endif
+                                  const __wasi_subscription_t *in,
+                                  __wasi_event_t *out, size_t nsubscriptions,
+                                  size_t *nevents)
+{
+    if (nsubscriptions == 0) {
+        *nevents = 0;
+        return __WASI_ESUCCESS;
+    }
+
+    wasi_errno_t err;
+    __wasi_timestamp_t elapsed = 0;
+
+    const __wasi_timestamp_t timeout = get_timeout_for_poll_oneoff(
+                                 in, nsubscriptions),
+                             time_quant = 1e9;
+    const uint64 size_to_copy =
+        nsubscriptions * (uint64)sizeof(wasi_subscription_t);
+    __wasi_subscription_t *in_copy = NULL;
+
+    if (size_to_copy >= UINT32_MAX
+        || !(in_copy = (__wasi_subscription_t *)wasm_runtime_malloc(
+                 (uint32)size_to_copy))) {
+        return __WASI_ENOMEM;
+    }
+
+    bh_memcpy_s(in_copy, size_to_copy, in, size_to_copy);
+
+    while (timeout == (__wasi_timestamp_t)-1 || elapsed <= timeout) {
+        elapsed += time_quant;
+
+        /* update timeout for clock subscription events */
+        update_clock_subscription_data(in_copy, nsubscriptions,
+                                       min(time_quant, timeout - elapsed));
+        err = wasmtime_ssp_poll_oneoff(curfds, in_copy, out, nsubscriptions,
+                                       nevents);
+        if (err) {
+            wasm_runtime_free(in_copy);
+            return err;
+        }
+
+        if (wasm_runtime_get_exception(module_inst) || *nevents > 0) {
+            wasm_runtime_free(in_copy);
+
+            if (*nevents) {
+                return __WASI_ESUCCESS;
+            }
+            return EINTR;
+        }
+    }
+
+    wasm_runtime_free(in_copy);
+    return __WASI_ESUCCESS;
+}
+#endif
+
 static wasi_errno_t
 wasi_poll_oneoff(wasm_exec_env_t exec_env, const wasi_subscription_t *in,
                  wasi_event_t *out, uint32 nsubscriptions, uint32 *nevents_app)
@@ -958,7 +1055,7 @@ wasi_poll_oneoff(wasm_exec_env_t exec_env, const wasi_subscription_t *in,
     wasm_module_inst_t module_inst = get_module_inst(exec_env);
     wasi_ctx_t wasi_ctx = get_wasi_ctx(module_inst);
     struct fd_table *curfds = wasi_ctx_get_curfds(module_inst, wasi_ctx);
-    size_t nevents;
+    size_t nevents = 0;
     wasi_errno_t err;
 
     if (!wasi_ctx)
@@ -969,7 +1066,12 @@ wasi_poll_oneoff(wasm_exec_env_t exec_env, const wasi_subscription_t *in,
         || !validate_native_addr(nevents_app, sizeof(uint32)))
         return (wasi_errno_t)-1;
 
+#if WASM_ENABLE_THREAD_MGR == 0
     err = wasmtime_ssp_poll_oneoff(curfds, in, out, nsubscriptions, &nevents);
+#else
+    err = execute_interruptible_poll_oneoff(module_inst, curfds, in, out,
+                                            nsubscriptions, &nevents);
+#endif
     if (err)
         return err;
 
@@ -1861,12 +1963,6 @@ allocate_iovec_app_buffer(wasm_module_inst_t module_inst,
     return __WASI_ESUCCESS;
 }
 
-static inline size_t
-min(size_t a, size_t b)
-{
-    return a > b ? b : a;
-}
-
 static wasi_errno_t
 copy_buffer_to_iovec_app(wasm_module_inst_t module_inst, uint8 *buf_begin,
                          uint32 buf_size, iovec_app_t *data, uint32 data_len,

core/iwasm/libraries/libc-wasi/sandboxed-system-primitives/src/gnuc.h

@@ -0,0 +1,14 @@
+/*
+ * Copyright (C) 2023 Amazon.com, Inc. or its affiliates. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#if !defined(__GNUC_PREREQ) && (defined(__GNUC__) || defined(__GNUG__)) \
+    && !defined(__clang__) && defined(__GNUC_MINOR__)
+/* Depending on the platform the macro is defined in sys/features.h or
+   features.h Given the macro is simple, we re-implement it here instead of
+   dealing with two different paths.
+ */
+#define __GNUC_PREREQ(maj, min) \
+    ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
+#endif

core/iwasm/libraries/libc-wasi/sandboxed-system-primitives/src/numeric_limits.h

@@ -1,42 +0,0 @@
-// Part of the Wasmtime Project, under the Apache License v2.0 with LLVM
-// Exceptions. See
-// https://github.com/bytecodealliance/wasmtime/blob/main/LICENSE for license
-// information.
-//
-// Significant parts of this file are derived from cloudabi-utils. See
-// https://github.com/bytecodealliance/wasmtime/blob/main/lib/wasi/sandboxed-system-primitives/src/LICENSE
-// for license information.
-//
-// The upstream file contains the following copyright notice:
-//
-// Copyright (c) 2015 Nuxi, https://nuxi.nl/
-
-#ifndef COMMON_LIMITS_H
-#define COMMON_LIMITS_H
-
-#define NUMERIC_MIN(t)                                    \
-    _Generic((t)0, char                                   \
-             : CHAR_MIN, signed char                      \
-             : SCHAR_MIN, unsigned char : 0, short        \
-             : SHRT_MIN, unsigned short : 0, int          \
-             : INT_MIN, unsigned int : 0, long            \
-             : LONG_MIN, unsigned long : 0, long long     \
-             : LLONG_MIN, unsigned long long : 0, default \
-             : (void)0)
-
-#define NUMERIC_MAX(t)                       \
-    _Generic((t)0, char                      \
-             : CHAR_MAX, signed char         \
-             : SCHAR_MAX, unsigned char      \
-             : UCHAR_MAX, short              \
-             : SHRT_MAX, unsigned short      \
-             : USHRT_MAX, int                \
-             : INT_MAX, unsigned int         \
-             : UINT_MAX, long                \
-             : LONG_MAX, unsigned long       \
-             : ULONG_MAX, long long          \
-             : LLONG_MAX, unsigned long long \
-             : ULLONG_MAX, default           \
-             : (void)0)
-
-#endif

core/iwasm/libraries/libc-wasi/sandboxed-system-primitives/src/posix.c

@@ -15,7 +15,6 @@
 #include "bh_platform.h"
 #include "wasmtime_ssp.h"
 #include "locking.h"
-#include "numeric_limits.h"
 #include "posix.h"
 #include "random.h"
 #include "refcount.h"
@@ -2257,8 +2256,7 @@ convert_timestamp(__wasi_timestamp_t in, struct timespec *out)
     in /= 1000000000;
 
     // Clamp to the maximum in case it would overflow our system's time_t.
-    out->tv_sec =
+    out->tv_sec = (time_t)in < BH_TIME_T_MAX ? (time_t)in : BH_TIME_T_MAX;
-        (time_t)in < NUMERIC_MAX(time_t) ? (time_t)in : NUMERIC_MAX(time_t);
 }
 
 // Converts the provided timestamps and flags to a set of arguments for
@@ -3226,6 +3224,7 @@ wasi_ssp_sock_get_reuse_port(
 #else
     errno = ENOTSUP;
     ret = BHT_ERROR;
+    optval = 0;
 #endif /* defined(SO_REUSEPORT) */
 
     fd_object_release(fo);

core/iwasm/libraries/libc-wasi/sandboxed-system-primitives/src/refcount.h

@@ -16,6 +16,7 @@
 
 #include "bh_platform.h"
 #include "locking.h"
+#include "gnuc.h"
 
 #define PRODUCES(...) LOCKS_SHARED(__VA_ARGS__) NO_LOCK_ANALYSIS
 #define CONSUMES(...) UNLOCKS(__VA_ARGS__) NO_LOCK_ANALYSIS
@@ -95,6 +96,42 @@ refcount_release(struct refcount *r)
     return old == 1;
 }
 
+#elif defined(__GNUC_PREREQ)
+
+#if __GNUC_PREREQ(4, 7)
+
+struct refcount {
+    unsigned int count;
+};
+
+/* Initialize the reference counter. */
+static inline void
+refcount_init(struct refcount *r, unsigned int count)
+{
+    __atomic_store_n(&r->count, count, __ATOMIC_SEQ_CST);
+}
+
+/* Increment the reference counter. */
+static inline void
+refcount_acquire(struct refcount *r)
+{
+    __atomic_fetch_add(&r->count, 1, __ATOMIC_ACQUIRE);
+}
+
+/* Decrement the reference counter, returning whether the reference
+   dropped to zero. */
+static inline bool
+refcount_release(struct refcount *r)
+{
+    int old = (int)__atomic_fetch_sub(&r->count, 1, __ATOMIC_RELEASE);
+    bh_assert(old != 0 && "Reference count becoming negative");
+    return old == 1;
+}
+
+#else /* else of __GNUC_PREREQ (4.7) */
+#error "Reference counter isn't implemented"
+#endif /* end of __GNUC_PREREQ (4.7) */
+
 #else /* else of CONFIG_HAS_STD_ATOMIC */
 #error "Reference counter isn't implemented"
 #endif /* end of CONFIG_HAS_STD_ATOMIC */

core/iwasm/libraries/libc-wasi/sandboxed-system-primitives/src/ssp_config.h

@@ -14,6 +14,7 @@
 #ifndef SSP_CONFIG_H
 #define SSP_CONFIG_H
 
+#include "gnuc.h"
 #include <stdlib.h>
 
 #if defined(__FreeBSD__) || defined(__APPLE__) \
@@ -107,10 +108,31 @@
 #endif
 
 #if !defined(BH_PLATFORM_LINUX_SGX)
+/* Clang's __GNUC_PREREQ macro has a different meaning than GCC one,
+so we have to handle this case specially */
+#if defined(__clang__)
+/* Clang provides stdatomic.h since 3.6.0
+See https://releases.llvm.org/3.6.0/tools/clang/docs/ReleaseNotes.html */
+#if __clang_major__ > 3 || (__clang_major__ == 3 && __clang_minor__ >= 6)
 #define CONFIG_HAS_STD_ATOMIC 1
 #else
 #define CONFIG_HAS_STD_ATOMIC 0
 #endif
+#elif defined(__GNUC_PREREQ)
+/* Even though older versions of GCC support C11, atomics were
+not implemented until 4.9. See
+https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58016 */
+#if __GNUC_PREREQ(4, 9)
+#define CONFIG_HAS_STD_ATOMIC 1
+#else /* else of __GNUC_PREREQ(4, 9) */
+#define CONFIG_HAS_STD_ATOMIC 0
+#endif /* end of __GNUC_PREREQ(4, 9) */
+#else  /* else of defined(__GNUC_PREREQ) */
+#define CONFIG_HAS_STD_ATOMIC 1
+#endif /* end of defined(__GNUC_PREREQ) */
+#else  /* else of !defined(BH_PLATFORM_LINUX_SGX) */
+#define CONFIG_HAS_STD_ATOMIC 0
+#endif /* end of !defined(BH_PLATFORM_LINUX_SGX) */
 
 #if !defined(__NuttX__)
 #define CONFIG_HAS_D_INO 1

core/iwasm/libraries/thread-mgr/thread_manager.c

@@ -76,6 +76,58 @@ traverse_list(bh_list *l, list_visitor visitor, void *user_data)
     }
 }
 
+/* Assumes cluster->lock is locked */
+static bool
+safe_traverse_exec_env_list(WASMCluster *cluster, list_visitor visitor,
+                            void *user_data)
+{
+    Vector proc_nodes;
+    void *node;
+    bool ret = true;
+
+    if (!bh_vector_init(&proc_nodes, cluster->exec_env_list.len, sizeof(void *),
+                        false)) {
+        ret = false;
+        goto final;
+    }
+
+    node = bh_list_first_elem(&cluster->exec_env_list);
+
+    while (node) {
+        bool already_processed = false;
+        void *proc_node;
+        for (size_t i = 0; i < bh_vector_size(&proc_nodes); i++) {
+            if (!bh_vector_get(&proc_nodes, i, &proc_node)) {
+                ret = false;
+                goto final;
+            }
+            if (proc_node == node) {
+                already_processed = true;
+                break;
+            }
+        }
+        if (already_processed) {
+            node = bh_list_elem_next(node);
+            continue;
+        }
+
+        os_mutex_unlock(&cluster->lock);
+        visitor(node, user_data);
+        os_mutex_lock(&cluster->lock);
+        if (!bh_vector_append(&proc_nodes, &node)) {
+            ret = false;
+            goto final;
+        }
+
+        node = bh_list_first_elem(&cluster->exec_env_list);
+    }
+
+final:
+    bh_vector_destroy(&proc_nodes);
+
+    return ret;
+}
+
 /* The caller must lock cluster->lock */
 static bool
 allocate_aux_stack(WASMExecEnv *exec_env, uint32 *start, uint32 *size)
@@ -344,7 +396,6 @@ wasm_cluster_del_exec_env(WASMCluster *cluster, WASMExecEnv *exec_env)
         os_mutex_unlock(&cluster->debug_inst->wait_lock);
     }
 #endif
-
     if (bh_list_remove(&cluster->exec_env_list, exec_env) != 0)
         ret = false;
 
@@ -478,7 +529,7 @@ fail4:
     /* free the allocated aux stack space */
     free_aux_stack(exec_env, aux_stack_start);
 fail3:
-    wasm_exec_env_destroy(new_exec_env);
+    wasm_exec_env_destroy_internal(new_exec_env);
 fail2:
     wasm_runtime_deinstantiate_internal(new_module_inst, true);
 fail1:
@@ -616,7 +667,7 @@ fail3:
     if (alloc_aux_stack)
         free_aux_stack(exec_env, aux_stack_start);
 fail2:
-    wasm_exec_env_destroy(new_exec_env);
+    wasm_exec_env_destroy_internal(new_exec_env);
 fail1:
     os_mutex_unlock(&cluster->lock);
 
@@ -786,16 +837,22 @@ wasm_cluster_join_thread(WASMExecEnv *exec_env, void **ret_val)
     korp_tid handle;
 
     os_mutex_lock(&cluster_list_lock);
+    os_mutex_lock(&exec_env->cluster->lock);
+
     if (!clusters_have_exec_env(exec_env) || exec_env->thread_is_detached) {
         /* Invalid thread, thread has exited or thread has been detached */
         if (ret_val)
             *ret_val = NULL;
+        os_mutex_unlock(&exec_env->cluster->lock);
         os_mutex_unlock(&cluster_list_lock);
         return 0;
     }
     exec_env->wait_count++;
     handle = exec_env->handle;
+
+    os_mutex_unlock(&exec_env->cluster->lock);
     os_mutex_unlock(&cluster_list_lock);
+
     return os_thread_join(handle, ret_val);
 }
 
@@ -878,15 +935,22 @@ int32
 wasm_cluster_cancel_thread(WASMExecEnv *exec_env)
 {
     os_mutex_lock(&cluster_list_lock);
+    os_mutex_lock(&exec_env->cluster->lock);
+
+    if (!exec_env->cluster) {
+        goto final;
+    }
     if (!clusters_have_exec_env(exec_env)) {
         /* Invalid thread or the thread has exited */
-        os_mutex_unlock(&cluster_list_lock);
+        goto final;
-        return 0;
     }
-    os_mutex_unlock(&cluster_list_lock);
 
     set_thread_cancel_flags(exec_env);
 
+final:
+    os_mutex_unlock(&exec_env->cluster->lock);
+    os_mutex_unlock(&cluster_list_lock);
+
     return 0;
 }
 
@@ -908,11 +972,9 @@ wasm_cluster_terminate_all(WASMCluster *cluster)
 {
     os_mutex_lock(&cluster->lock);
     cluster->processing = true;
-    os_mutex_unlock(&cluster->lock);
 
-    traverse_list(&cluster->exec_env_list, terminate_thread_visitor, NULL);
+    safe_traverse_exec_env_list(cluster, terminate_thread_visitor, NULL);
 
-    os_mutex_lock(&cluster->lock);
     cluster->processing = false;
     os_mutex_unlock(&cluster->lock);
 }
@@ -923,12 +985,10 @@ wasm_cluster_terminate_all_except_self(WASMCluster *cluster,
 {
     os_mutex_lock(&cluster->lock);
     cluster->processing = true;
-    os_mutex_unlock(&cluster->lock);
 
-    traverse_list(&cluster->exec_env_list, terminate_thread_visitor,
+    safe_traverse_exec_env_list(cluster, terminate_thread_visitor,
-                  (void *)exec_env);
+                                (void *)exec_env);
 
-    os_mutex_lock(&cluster->lock);
     cluster->processing = false;
     os_mutex_unlock(&cluster->lock);
 }
@@ -950,11 +1010,9 @@ wams_cluster_wait_for_all(WASMCluster *cluster)
 {
     os_mutex_lock(&cluster->lock);
     cluster->processing = true;
-    os_mutex_unlock(&cluster->lock);
 
-    traverse_list(&cluster->exec_env_list, wait_for_thread_visitor, NULL);
+    safe_traverse_exec_env_list(cluster, wait_for_thread_visitor, NULL);
 
-    os_mutex_lock(&cluster->lock);
     cluster->processing = false;
     os_mutex_unlock(&cluster->lock);
 }
@@ -965,12 +1023,10 @@ wasm_cluster_wait_for_all_except_self(WASMCluster *cluster,
 {
     os_mutex_lock(&cluster->lock);
     cluster->processing = true;
-    os_mutex_unlock(&cluster->lock);
 
-    traverse_list(&cluster->exec_env_list, wait_for_thread_visitor,
+    safe_traverse_exec_env_list(cluster, wait_for_thread_visitor,
-                  (void *)exec_env);
+                                (void *)exec_env);
 
-    os_mutex_lock(&cluster->lock);
     cluster->processing = false;
     os_mutex_unlock(&cluster->lock);
 }

core/iwasm/libraries/wasi-nn/.dockerignore

@@ -1 +0,0 @@
-**/Dockerfile

core/iwasm/libraries/wasi-nn/README.md

@@ -17,27 +17,76 @@ By only including this file in your WASM application you will bind WASI-NN into
 To run the tests we assume that the current directory is the root of the repository.
 
 
-1. Build the docker image,
+### Build the runtime
+
+Build the runtime base image,
 
 ```
-docker build -t wasi-nn -f core/iwasm/libraries/wasi-nn/test/Dockerfile .
+docker build -t wasi-nn-base -f core/iwasm/libraries/wasi-nn/test/Dockerfile.base .
 ```
 
-2. Run the container
+Build the runtime image for your execution target type.
+
+`EXECUTION_TYPE` can be:
+* `cpu`
+* `nvidia-gpu`
 
 ```
-docker run wasi-nn
+EXECUTION_TYPE=cpu
+docker build -t wasi-nn-${EXECUTION_TYPE} -f core/iwasm/libraries/wasi-nn/test/Dockerfile.${EXECUTION_TYPE} .
 ```
 
+
+### Build wasm app
+
+```
+docker build -t wasi-nn-compile -f core/iwasm/libraries/wasi-nn/test/Dockerfile.compile .
+```
+
+```
+docker run -v $PWD/core/iwasm/libraries/wasi-nn:/wasi-nn wasi-nn-compile
+```
+
+
+### Run wasm app
+
 If all the tests have run properly you will the the following message in the terminal,
 
 ```
 Tests: passed!
 ```
 
+* CPU
+
+```
+docker run \
+    -v $PWD/core/iwasm/libraries/wasi-nn/test:/assets wasi-nn-cpu \
+    --dir=/assets \
+    --env="TARGET=cpu" \
+    /assets/test_tensorflow.wasm
+```
+
+* (NVIDIA) GPU
+
+```
+docker run \
+    --runtime=nvidia \
+    -v $PWD/core/iwasm/libraries/wasi-nn/test:/assets wasi-nn-nvidia-gpu \
+    --dir=/assets \
+    --env="TARGET=gpu" \
+    /assets/test_tensorflow.wasm
+```
+
+Requirements:
+* [NVIDIA docker](https://github.com/NVIDIA/nvidia-docker).
+
 ## What is missing
 
-* Only 1 model at a time is supported.
+Supported:
+
+* Only 1 WASM app at a time.
+* Only 1 model at a time.
     * `graph` and `graph-execution-context` are ignored.
-* Only `tensorflow` (lite) is supported.
+* Graph encoding: `tensorflowlite`.
-* Only `cpu` is supported.
+* Execution target: `cpu` and `gpu`.
+* Tensor type: `fp32`.

core/iwasm/libraries/wasi-nn/logger.h

@@ -1,55 +0,0 @@
-/*
- * Copyright (C) 2019 Intel Corporation.  All rights reserved.
- * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
- */
-
-#ifndef WASI_NN_LOGGER_H
-#define WASI_NN_LOGGER_H
-
-#include <stdio.h>
-#include <string.h>
-
-#define __FILENAME__ \
-    (strrchr(__FILE__, '/') ? strrchr(__FILE__, '/') + 1 : __FILE__)
-
-/* Disable a level by removing the define */
-#define ENABLE_ERR_LOG
-#define ENABLE_WARN_LOG
-#define ENABLE_DBG_LOG
-#define ENABLE_INFO_LOG
-
-// Definition of the levels
-#ifdef ENABLE_ERR_LOG
-#define NN_ERR_PRINTF(fmt, ...)                                    \
-    printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
-    printf("\n");                                                  \
-    fflush(stdout)
-#else
-#define NN_ERR_PRINTF(fmt, ...)
-#endif
-#ifdef ENABLE_WARN_LOG
-#define NN_WARN_PRINTF(fmt, ...)                                   \
-    printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
-    printf("\n");                                                  \
-    fflush(stdout)
-#else
-#define NN_WARN_PRINTF(fmt, ...)
-#endif
-#ifdef ENABLE_DBG_LOG
-#define NN_DBG_PRINTF(fmt, ...)                                    \
-    printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
-    printf("\n");                                                  \
-    fflush(stdout)
-#else
-#define NN_DBG_PRINTF(fmt, ...)
-#endif
-#ifdef ENABLE_INFO_LOG
-#define NN_INFO_PRINTF(fmt, ...)                                   \
-    printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
-    printf("\n");                                                  \
-    fflush(stdout)
-#else
-#define NN_INFO_PRINTF(fmt, ...)
-#endif
-
-#endif

core/iwasm/libraries/wasi-nn/src/utils/logger.h

@@ -0,0 +1,63 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#ifndef WASI_NN_LOGGER_H
+#define WASI_NN_LOGGER_H
+
+#include <stdio.h>
+#include <string.h>
+
+#define __FILENAME__ \
+    (strrchr(__FILE__, '/') ? strrchr(__FILE__, '/') + 1 : __FILE__)
+
+/* Disable a level by removing the define */
+#define ENABLE_ERR_LOG
+#define ENABLE_WARN_LOG
+#define ENABLE_DBG_LOG
+#define ENABLE_INFO_LOG
+
+// Definition of the levels
+#ifdef ENABLE_ERR_LOG
+#define NN_ERR_PRINTF(fmt, ...)                                        \
+    do {                                                               \
+        printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
+        printf("\n");                                                  \
+        fflush(stdout);                                                \
+    } while (0)
+#else
+#define NN_ERR_PRINTF(fmt, ...)
+#endif
+#ifdef ENABLE_WARN_LOG
+#define NN_WARN_PRINTF(fmt, ...)                                       \
+    do {                                                               \
+        printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
+        printf("\n");                                                  \
+        fflush(stdout);                                                \
+    } while (0)
+#else
+#define NN_WARN_PRINTF(fmt, ...)
+#endif
+#ifdef ENABLE_DBG_LOG
+#define NN_DBG_PRINTF(fmt, ...)                                        \
+    do {                                                               \
+        printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
+        printf("\n");                                                  \
+        fflush(stdout);                                                \
+    } while (0)
+#else
+#define NN_DBG_PRINTF(fmt, ...)
+#endif
+#ifdef ENABLE_INFO_LOG
+#define NN_INFO_PRINTF(fmt, ...)                                       \
+    do {                                                               \
+        printf("[%s:%d] " fmt, __FILENAME__, __LINE__, ##__VA_ARGS__); \
+        printf("\n");                                                  \
+        fflush(stdout);                                                \
+    } while (0)
+#else
+#define NN_INFO_PRINTF(fmt, ...)
+#endif
+
+#endif

core/iwasm/libraries/wasi-nn/src/utils/wasi_nn_app_native.c

@@ -0,0 +1,163 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include "wasi_nn_app_native.h"
+
+static error
+graph_builder_app_native(wasm_module_inst_t instance,
+                         graph_builder_wasm *builder_wasm,
+                         graph_builder *builder)
+{
+    if (!wasm_runtime_validate_app_addr(instance, builder_wasm->buf_offset,
+                                        builder_wasm->size * sizeof(uint8_t))) {
+        NN_ERR_PRINTF("builder_wasm->buf_offset is invalid");
+        return invalid_argument;
+    }
+
+    builder->buf = (uint8_t *)wasm_runtime_addr_app_to_native(
+        instance, builder_wasm->buf_offset);
+    builder->size = builder_wasm->size;
+    return success;
+}
+
+error
+graph_builder_array_app_native(wasm_module_inst_t instance,
+                               graph_builder_array_wasm *builder_array_wasm,
+                               graph_builder_array *builder_array)
+{
+    if (!wasm_runtime_validate_native_addr(instance, builder_array_wasm,
+                                           sizeof(graph_builder_array_wasm))) {
+        NN_ERR_PRINTF("builder_array_wasm is invalid");
+        return invalid_argument;
+    }
+
+    NN_DBG_PRINTF("Graph builder array contains %d elements",
+                  builder_array_wasm->size);
+
+    if (!wasm_runtime_validate_app_addr(
+            instance, builder_array_wasm->buf_offset,
+            builder_array_wasm->size * sizeof(graph_builder_wasm))) {
+        NN_ERR_PRINTF("builder_array_wasm->buf_offset is invalid");
+        return invalid_argument;
+    }
+
+    graph_builder_wasm *builder_wasm =
+        (graph_builder_wasm *)wasm_runtime_addr_app_to_native(
+            instance, builder_array_wasm->buf_offset);
+
+    graph_builder *builder = (graph_builder *)wasm_runtime_malloc(
+        builder_array_wasm->size * sizeof(graph_builder));
+    if (builder == NULL)
+        return missing_memory;
+
+    for (uint32_t i = 0; i < builder_array_wasm->size; ++i) {
+        error res;
+        if (success
+            != (res = graph_builder_app_native(instance, &builder_wasm[i],
+                                               &builder[i]))) {
+            wasm_runtime_free(builder);
+            return res;
+        }
+
+        NN_DBG_PRINTF("Graph builder %d contains %d elements", i,
+                      builder->size);
+    }
+
+    builder_array->buf = builder;
+    builder_array->size = builder_array_wasm->size;
+    return success;
+}
+
+static error
+tensor_data_app_native(wasm_module_inst_t instance, uint32_t total_elements,
+                       tensor_wasm *input_tensor_wasm, tensor_data *data)
+{
+    if (!wasm_runtime_validate_app_addr(
+            instance, input_tensor_wasm->data_offset, total_elements)) {
+        NN_ERR_PRINTF("input_tensor_wasm->data_offset is invalid");
+        return invalid_argument;
+    }
+    *data = (tensor_data)wasm_runtime_addr_app_to_native(
+        instance, input_tensor_wasm->data_offset);
+    return success;
+}
+
+static error
+tensor_dimensions_app_native(wasm_module_inst_t instance,
+                             tensor_wasm *input_tensor_wasm,
+                             tensor_dimensions **dimensions)
+{
+    if (!wasm_runtime_validate_app_addr(instance,
+                                        input_tensor_wasm->dimensions_offset,
+                                        sizeof(tensor_dimensions_wasm))) {
+        NN_ERR_PRINTF("input_tensor_wasm->dimensions_offset is invalid");
+        return invalid_argument;
+    }
+
+    tensor_dimensions_wasm *dimensions_wasm =
+        (tensor_dimensions_wasm *)wasm_runtime_addr_app_to_native(
+            instance, input_tensor_wasm->dimensions_offset);
+
+    if (!wasm_runtime_validate_app_addr(instance, dimensions_wasm->buf_offset,
+                                        sizeof(tensor_dimensions))) {
+        NN_ERR_PRINTF("dimensions_wasm->buf_offset is invalid");
+        return invalid_argument;
+    }
+
+    *dimensions =
+        (tensor_dimensions *)wasm_runtime_malloc(sizeof(tensor_dimensions));
+    if (dimensions == NULL)
+        return missing_memory;
+
+    (*dimensions)->size = dimensions_wasm->size;
+    (*dimensions)->buf = (uint32_t *)wasm_runtime_addr_app_to_native(
+        instance, dimensions_wasm->buf_offset);
+
+    NN_DBG_PRINTF("Number of dimensions: %d", (*dimensions)->size);
+    return success;
+}
+
+error
+tensor_app_native(wasm_module_inst_t instance, tensor_wasm *input_tensor_wasm,
+                  tensor *input_tensor)
+{
+    NN_DBG_PRINTF("Converting tensor_wasm to tensor");
+    if (!wasm_runtime_validate_native_addr(instance, input_tensor_wasm,
+                                           sizeof(tensor_wasm))) {
+        NN_ERR_PRINTF("input_tensor_wasm is invalid");
+        return invalid_argument;
+    }
+
+    error res;
+
+    tensor_dimensions *dimensions = NULL;
+    if (success
+        != (res = tensor_dimensions_app_native(instance, input_tensor_wasm,
+                                               &dimensions))) {
+        NN_ERR_PRINTF("error when parsing dimensions");
+        return res;
+    }
+
+    uint32_t total_elements = 1;
+    for (uint32_t i = 0; i < dimensions->size; ++i) {
+        total_elements *= dimensions->buf[i];
+        NN_DBG_PRINTF("Dimension %d: %d", i, dimensions->buf[i]);
+    }
+    NN_DBG_PRINTF("Tensor type: %d", input_tensor_wasm->type);
+    NN_DBG_PRINTF("Total number of elements: %d", total_elements);
+
+    tensor_data data = NULL;
+    if (success
+        != (res = tensor_data_app_native(instance, total_elements,
+                                         input_tensor_wasm, &data))) {
+        wasm_runtime_free(dimensions);
+        return res;
+    }
+
+    input_tensor->type = input_tensor_wasm->type;
+    input_tensor->dimensions = dimensions;
+    input_tensor->data = data;
+    return success;
+}

core/iwasm/libraries/wasi-nn/src/utils/wasi_nn_app_native.h

@@ -0,0 +1,51 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#ifndef WASI_NN_APP_NATIVE
+#define WASI_NN_APP_NATIVE
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+
+#include "wasi_nn.h"
+#include "logger.h"
+
+#include "bh_platform.h"
+#include "wasm_export.h"
+
+typedef struct {
+    uint32_t buf_offset;
+    uint32_t size;
+} graph_builder_wasm;
+
+typedef struct {
+    uint32_t buf_offset;
+    uint32_t size;
+} graph_builder_array_wasm;
+
+typedef struct {
+    uint32_t buf_offset;
+    uint32_t size;
+} tensor_dimensions_wasm;
+
+typedef struct {
+    uint32_t dimensions_offset;
+    tensor_type type;
+    uint32_t data_offset;
+} tensor_wasm;
+
+error
+graph_builder_array_app_native(wasm_module_inst_t instance,
+                               graph_builder_array_wasm *builder,
+                               graph_builder_array *builder_native);
+
+error
+tensor_app_native(wasm_module_inst_t instance, tensor_wasm *input_tensor,
+                  tensor *input_tensor_native);
+
+#endif

core/iwasm/libraries/wasi-nn/src/wasi_nn.c

@@ -0,0 +1,302 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdbool.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+
+#include "wasi_nn.h"
+#include "wasi_nn_app_native.h"
+#include "logger.h"
+#include "wasi_nn_tensorflowlite.hpp"
+
+#include "bh_platform.h"
+#include "wasm_export.h"
+#include "wasm_runtime.h"
+#include "aot_runtime.h"
+
+/* Definition of 'wasi_nn.h' structs in WASM app format (using offset) */
+
+typedef error (*LOAD)(graph_builder_array *, graph_encoding, execution_target,
+                      graph *);
+typedef error (*INIT_EXECUTION_CONTEXT)(graph, graph_execution_context *);
+typedef error (*SET_INPUT)(graph_execution_context, uint32_t, tensor *);
+typedef error (*COMPUTE)(graph_execution_context);
+typedef error (*GET_OUTPUT)(graph_execution_context, uint32_t, tensor_data,
+                            uint32_t *);
+
+typedef struct {
+    LOAD load;
+    INIT_EXECUTION_CONTEXT init_execution_context;
+    SET_INPUT set_input;
+    COMPUTE compute;
+    GET_OUTPUT get_output;
+} api_function;
+
+/* Global variables */
+
+static api_function lookup[] = {
+    { NULL, NULL, NULL, NULL, NULL },
+    { NULL, NULL, NULL, NULL, NULL },
+    { NULL, NULL, NULL, NULL, NULL },
+    { NULL, NULL, NULL, NULL, NULL },
+    { tensorflowlite_load, tensorflowlite_init_execution_context,
+      tensorflowlite_set_input, tensorflowlite_compute,
+      tensorflowlite_get_output }
+};
+
+/* Utils */
+
+static bool
+is_encoding_implemented(graph_encoding encoding)
+{
+    return lookup[encoding].load && lookup[encoding].init_execution_context
+           && lookup[encoding].set_input && lookup[encoding].compute
+           && lookup[encoding].get_output;
+}
+
+static error
+is_model_initialized(WASINNContext *wasi_nn_ctx)
+{
+    if (!wasi_nn_ctx->is_initialized) {
+        NN_ERR_PRINTF("Model not initialized.");
+        return runtime_error;
+    }
+    return success;
+}
+
+WASINNContext *
+wasm_runtime_get_wasi_nn_ctx(wasm_module_inst_t instance)
+{
+    WASINNContext *wasi_nn_ctx = NULL;
+#if WASM_ENABLE_INTERP != 0
+    if (instance->module_type == Wasm_Module_Bytecode) {
+        NN_DBG_PRINTF("Getting ctx from WASM");
+        WASMModuleInstance *module_inst = (WASMModuleInstance *)instance;
+        wasi_nn_ctx = ((WASMModuleInstanceExtra *)module_inst->e)->wasi_nn_ctx;
+    }
+#endif
+#if WASM_ENABLE_AOT != 0
+    if (instance->module_type == Wasm_Module_AoT) {
+        NN_DBG_PRINTF("Getting ctx from AOT");
+        AOTModuleInstance *module_inst = (AOTModuleInstance *)instance;
+        wasi_nn_ctx = ((AOTModuleInstanceExtra *)module_inst->e)->wasi_nn_ctx;
+    }
+#endif
+    bh_assert(wasi_nn_ctx != NULL);
+    NN_DBG_PRINTF("Returning ctx");
+    return wasi_nn_ctx;
+}
+
+/* WASI-NN implementation */
+
+error
+wasi_nn_load(wasm_exec_env_t exec_env, graph_builder_array_wasm *builder,
+             graph_encoding encoding, execution_target target, graph *g)
+{
+    NN_DBG_PRINTF("Running wasi_nn_load [encoding=%d, target=%d]...", encoding,
+                  target);
+
+    if (!is_encoding_implemented(encoding)) {
+        NN_ERR_PRINTF("Encoding not supported.");
+        return invalid_encoding;
+    }
+
+    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
+    bh_assert(instance);
+
+    error res;
+    graph_builder_array builder_native = { 0 };
+    if (success
+        != (res = graph_builder_array_app_native(instance, builder,
+                                                 &builder_native)))
+        return res;
+
+    if (!wasm_runtime_validate_native_addr(instance, g, sizeof(graph))) {
+        NN_ERR_PRINTF("graph is invalid");
+        res = invalid_argument;
+        goto fail;
+    }
+
+    res = lookup[encoding].load(&builder_native, encoding, target, g);
+
+    NN_DBG_PRINTF("wasi_nn_load finished with status %d [graph=%d]", res, *g);
+
+    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
+
+    wasi_nn_ctx->current_encoding = encoding;
+    wasi_nn_ctx->is_initialized = true;
+
+fail:
+    // XXX: Free intermediate structure pointers
+    if (builder_native.buf)
+        wasm_runtime_free(builder_native.buf);
+
+    return res;
+}
+
+error
+wasi_nn_init_execution_context(wasm_exec_env_t exec_env, graph g,
+                               graph_execution_context *ctx)
+{
+    NN_DBG_PRINTF("Running wasi_nn_init_execution_context [graph=%d]...", g);
+
+    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
+    bh_assert(instance);
+    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
+
+    error res;
+    if (success != (res = is_model_initialized(wasi_nn_ctx)))
+        return res;
+
+    if (!wasm_runtime_validate_native_addr(instance, ctx,
+                                           sizeof(graph_execution_context))) {
+        NN_ERR_PRINTF("ctx is invalid");
+        return invalid_argument;
+    }
+
+    res = lookup[wasi_nn_ctx->current_encoding].init_execution_context(g, ctx);
+    *ctx = g;
+    NN_DBG_PRINTF(
+        "wasi_nn_init_execution_context finished with status %d [ctx=%d]", res,
+        *ctx);
+    return res;
+}
+
+error
+wasi_nn_set_input(wasm_exec_env_t exec_env, graph_execution_context ctx,
+                  uint32_t index, tensor_wasm *input_tensor)
+{
+    NN_DBG_PRINTF("Running wasi_nn_set_input [ctx=%d, index=%d]...", ctx,
+                  index);
+
+    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
+    bh_assert(instance);
+    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
+
+    error res;
+    if (success != (res = is_model_initialized(wasi_nn_ctx)))
+        return res;
+
+    tensor input_tensor_native = { 0 };
+    if (success
+        != (res = tensor_app_native(instance, input_tensor,
+                                    &input_tensor_native)))
+        return res;
+
+    res = lookup[wasi_nn_ctx->current_encoding].set_input(ctx, index,
+                                                          &input_tensor_native);
+
+    // XXX: Free intermediate structure pointers
+    if (input_tensor_native.dimensions)
+        wasm_runtime_free(input_tensor_native.dimensions);
+
+    NN_DBG_PRINTF("wasi_nn_set_input finished with status %d", res);
+    return res;
+}
+
+error
+wasi_nn_compute(wasm_exec_env_t exec_env, graph_execution_context ctx)
+{
+    NN_DBG_PRINTF("Running wasi_nn_compute [ctx=%d]...", ctx);
+
+    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
+    bh_assert(instance);
+    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
+
+    error res;
+    if (success != (res = is_model_initialized(wasi_nn_ctx)))
+        return res;
+
+    res = lookup[wasi_nn_ctx->current_encoding].compute(ctx);
+    NN_DBG_PRINTF("wasi_nn_compute finished with status %d", res);
+    return res;
+}
+
+error
+wasi_nn_get_output(wasm_exec_env_t exec_env, graph_execution_context ctx,
+                   uint32_t index, tensor_data output_tensor,
+                   uint32_t *output_tensor_size)
+{
+    NN_DBG_PRINTF("Running wasi_nn_get_output [ctx=%d, index=%d]...", ctx,
+                  index);
+
+    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
+    bh_assert(instance);
+    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
+
+    error res;
+    if (success != (res = is_model_initialized(wasi_nn_ctx)))
+        return res;
+
+    if (!wasm_runtime_validate_native_addr(instance, output_tensor_size,
+                                           sizeof(uint32_t))) {
+        NN_ERR_PRINTF("output_tensor_size is invalid");
+        return invalid_argument;
+    }
+
+    res = lookup[wasi_nn_ctx->current_encoding].get_output(
+        ctx, index, output_tensor, output_tensor_size);
+    NN_DBG_PRINTF("wasi_nn_get_output finished with status %d [data_size=%d]",
+                  res, *output_tensor_size);
+    return res;
+}
+
+/* Non-exposed public functions */
+
+WASINNContext *
+wasi_nn_initialize()
+{
+    NN_DBG_PRINTF("Initializing wasi-nn");
+    WASINNContext *wasi_nn_ctx =
+        (WASINNContext *)wasm_runtime_malloc(sizeof(WASINNContext));
+    if (wasi_nn_ctx == NULL) {
+        NN_ERR_PRINTF("Error when allocating memory for WASI-NN context");
+        return NULL;
+    }
+    wasi_nn_ctx->is_initialized = true;
+    wasi_nn_ctx->current_encoding = 3;
+    return wasi_nn_ctx;
+}
+
+void
+wasi_nn_destroy(WASINNContext *wasi_nn_ctx)
+{
+    if (wasi_nn_ctx == NULL) {
+        NN_ERR_PRINTF(
+            "Error when deallocating memory. WASI-NN context is NULL");
+        return;
+    }
+    NN_DBG_PRINTF("Freeing wasi-nn");
+    NN_DBG_PRINTF("-> is_initialized: %d", wasi_nn_ctx->is_initialized);
+    NN_DBG_PRINTF("-> current_encoding: %d", wasi_nn_ctx->current_encoding);
+    tensorflowlite_destroy();
+    wasm_runtime_free(wasi_nn_ctx);
+}
+
+/* Register WASI-NN in WAMR */
+
+/* clang-format off */
+#define REG_NATIVE_FUNC(func_name, signature) \
+    { #func_name, wasi_nn_##func_name, signature, NULL }
+/* clang-format on */
+
+static NativeSymbol native_symbols_wasi_nn[] = {
+    REG_NATIVE_FUNC(load, "(*ii*)i"),
+    REG_NATIVE_FUNC(init_execution_context, "(i*)i"),
+    REG_NATIVE_FUNC(set_input, "(ii*)i"),
+    REG_NATIVE_FUNC(compute, "(i)i"),
+    REG_NATIVE_FUNC(get_output, "(ii**)i"),
+};
+
+uint32_t
+get_wasi_nn_export_apis(NativeSymbol **p_libc_wasi_apis)
+{
+    *p_libc_wasi_apis = native_symbols_wasi_nn;
+    return sizeof(native_symbols_wasi_nn) / sizeof(NativeSymbol);
+}

core/iwasm/libraries/wasi-nn/src/wasi_nn_private.h

@@ -0,0 +1,30 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#ifndef WASI_NN_PRIVATE_H
+#define WASI_NN_PRIVATE_H
+
+#include "wasi_nn_types.h"
+
+typedef struct {
+    bool is_initialized;
+    graph_encoding current_encoding;
+} WASINNContext;
+
+/**
+ * @brief Initialize wasi-nn
+ *
+ */
+WASINNContext *
+wasi_nn_initialize();
+/**
+ * @brief Destroy wasi-nn on app exists
+ *
+ */
+
+void
+wasi_nn_destroy(WASINNContext *wasi_nn_ctx);
+
+#endif

core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.cpp

@@ -3,8 +3,10 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  */
 
-#include "wasi_nn_tensorflow.hpp"
+#include "wasi_nn.h"
-#include "wasi_nn_common.h"
+#include "wasi_nn_tensorflowlite.hpp"
+#include "logger.h"
+
 #include "bh_common.h"
 #include "bh_platform.h"
 #include "platform_common.h"
@@ -14,6 +16,7 @@
 #include <tensorflow/lite/model.h>
 #include <tensorflow/lite/optional_debug_tools.h>
 #include <tensorflow/lite/error_reporter.h>
+#include <tensorflow/lite/delegates/gpu/delegate.h>
 
 /* Global variables */
 
@@ -25,30 +28,30 @@ static char *model_pointer = NULL;
 /* WASI-NN (tensorflow) implementation */
 
 error
-tensorflow_load(graph_builder_array builder, graph_encoding encoding,
+tensorflowlite_load(graph_builder_array *builder, graph_encoding encoding,
-                execution_target target, graph *graph)
+                    execution_target target, graph *g)
 {
     if (model_pointer != NULL) {
         wasm_runtime_free(model_pointer);
         model_pointer = NULL;
     }
 
-    if (builder.size != 1) {
+    if (builder->size != 1) {
         NN_ERR_PRINTF("Unexpected builder format.");
         return invalid_argument;
     }
 
-    if (encoding != tensorflow) {
+    if (encoding != tensorflowlite) {
-        NN_ERR_PRINTF("Encoding is not tensorflow.");
+        NN_ERR_PRINTF("Encoding is not tensorflowlite.");
         return invalid_argument;
     }
 
-    if (target != cpu) {
+    if (target != cpu && target != gpu) {
-        NN_ERR_PRINTF("Only CPU target is supported.");
+        NN_ERR_PRINTF("Only CPU and GPU target is supported.");
         return invalid_argument;
     }
 
-    uint32_t size = builder.buf[0].size;
+    uint32_t size = builder->buf[0].size;
 
     model_pointer = (char *)wasm_runtime_malloc(size);
     if (model_pointer == NULL) {
@@ -56,7 +59,7 @@ tensorflow_load(graph_builder_array builder, graph_encoding encoding,
         return missing_memory;
     }
 
-    bh_memcpy_s(model_pointer, size, builder.buf[0].buf, size);
+    bh_memcpy_s(model_pointer, size, builder->buf[0].buf, size);
 
     model = tflite::FlatBufferModel::BuildFromBuffer(model_pointer, size, NULL);
     if (model == NULL) {
@@ -77,11 +80,34 @@ tensorflow_load(graph_builder_array builder, graph_encoding encoding,
         return missing_memory;
     }
 
+    bool use_default = false;
+    switch (target) {
+        case gpu:
+        {
+            // https://www.tensorflow.org/lite/performance/gpu
+            auto options = TfLiteGpuDelegateOptionsV2Default();
+            options.inference_preference =
+                TFLITE_GPU_INFERENCE_PREFERENCE_SUSTAINED_SPEED;
+            options.inference_priority1 =
+                TFLITE_GPU_INFERENCE_PRIORITY_MIN_LATENCY;
+            auto *delegate = TfLiteGpuDelegateV2Create(&options);
+            if (interpreter->ModifyGraphWithDelegate(delegate) != kTfLiteOk) {
+                NN_ERR_PRINTF("Error when enabling GPU delegate.");
+                use_default = true;
+            }
+            break;
+        }
+        default:
+            use_default = true;
+    }
+    if (use_default)
+        NN_WARN_PRINTF("Default encoding is CPU.");
+
     return success;
 }
 
 error
-tensorflow_init_execution_context(graph graph)
+tensorflowlite_init_execution_context(graph g, graph_execution_context *ctx)
 {
     if (interpreter == NULL) {
         NN_ERR_PRINTF("Non-initialized interpreter.");
@@ -92,8 +118,8 @@ tensorflow_init_execution_context(graph graph)
 }
 
 error
-tensorflow_set_input(graph_execution_context ctx, uint32_t index,
+tensorflowlite_set_input(graph_execution_context ctx, uint32_t index,
-                     tensor *input_tensor)
+                         tensor *input_tensor)
 {
     if (interpreter == NULL) {
         NN_ERR_PRINTF("Non-initialized interpreter.");
@@ -113,11 +139,11 @@ tensorflow_set_input(graph_execution_context ctx, uint32_t index,
     }
 
     uint32_t model_tensor_size = 1;
-    for (int i = 0; i < (int)tensor->dims->size; ++i)
+    for (int i = 0; i < tensor->dims->size; ++i)
         model_tensor_size *= (uint32_t)tensor->dims->data[i];
 
     uint32_t input_tensor_size = 1;
-    for (int i = 0; i < input_tensor->dimensions->size; i++)
+    for (uint32_t i = 0; i < input_tensor->dimensions->size; i++)
         input_tensor_size *= (uint32_t)input_tensor->dimensions->buf[i];
 
     if (model_tensor_size != input_tensor_size) {
@@ -136,7 +162,7 @@ tensorflow_set_input(graph_execution_context ctx, uint32_t index,
 }
 
 error
-tensorflow_compute(graph_execution_context ctx)
+tensorflowlite_compute(graph_execution_context ctx)
 {
     if (interpreter == NULL) {
         NN_ERR_PRINTF("Non-initialized interpreter.");
@@ -147,8 +173,9 @@ tensorflow_compute(graph_execution_context ctx)
 }
 
 error
-tensorflow_get_output(graph_execution_context context, uint32_t index,
+tensorflowlite_get_output(graph_execution_context ctx, uint32_t index,
-                      tensor_data output_tensor, uint32_t *output_tensor_size)
+                          tensor_data output_tensor,
+                          uint32_t *output_tensor_size)
 {
     if (interpreter == NULL) {
         NN_ERR_PRINTF("Non-initialized interpreter.");
@@ -178,7 +205,7 @@ tensorflow_get_output(graph_execution_context context, uint32_t index,
     }
 
     float *tensor_f = interpreter->typed_output_tensor<float>(index);
-    for (int i = 0; i < model_tensor_size; ++i)
+    for (uint32_t i = 0; i < model_tensor_size; ++i)
         NN_DBG_PRINTF("output: %f", tensor_f[i]);
 
     *output_tensor_size = model_tensor_size;
@@ -186,3 +213,22 @@ tensorflow_get_output(graph_execution_context context, uint32_t index,
                 model_tensor_size * sizeof(float));
     return success;
 }
+
+void
+tensorflowlite_destroy()
+{
+    /*
+        TensorFlow Lite memory is man
+
+        Related issues:
+        * https://github.com/tensorflow/tensorflow/issues/15880
+    */
+    NN_DBG_PRINTF("Freeing memory.");
+    model.reset(nullptr);
+    model = NULL;
+    interpreter.reset(nullptr);
+    interpreter = NULL;
+    wasm_runtime_free(model_pointer);
+    model_pointer = NULL;
+    NN_DBG_PRINTF("Memory free'd.");
+}

core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.hpp

@@ -0,0 +1,41 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#ifndef WASI_NN_TENSORFLOWLITE_HPP
+#define WASI_NN_TENSORFLOWLITE_HPP
+
+#include "wasi_nn.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+error
+tensorflowlite_load(graph_builder_array *builder, graph_encoding encoding,
+                    execution_target target, graph *g);
+
+error
+tensorflowlite_init_execution_context(graph g, graph_execution_context *ctx);
+
+error
+tensorflowlite_set_input(graph_execution_context ctx, uint32_t index,
+                         tensor *input_tensor);
+
+error
+tensorflowlite_compute(graph_execution_context ctx);
+
+error
+tensorflowlite_get_output(graph_execution_context ctx, uint32_t index,
+                          tensor_data output_tensor,
+                          uint32_t *output_tensor_size);
+
+void
+tensorflowlite_destroy();
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

core/iwasm/libraries/wasi-nn/test/CMakeLists.txt

@@ -7,10 +7,10 @@ project (iwasm)
 
 set (CMAKE_VERBOSE_MAKEFILE OFF)
 # Reset default linker flags
-set (CMAKE_SHARED_LIBRARY_LINK_C_FLAGS "")
-set (CMAKE_SHARED_LIBRARY_LINK_CXX_FLAGS "")
 set (CMAKE_C_STANDARD 99)
 set (CMAKE_CXX_STANDARD 14)
+set (CMAKE_SHARED_LIBRARY_LINK_C_FLAGS "")
+set (CMAKE_SHARED_LIBRARY_LINK_CXX_FLAGS "")
 
 if (NOT DEFINED WAMR_BUILD_PLATFORM)
   set (WAMR_BUILD_PLATFORM "linux")

core/iwasm/libraries/wasi-nn/test/Dockerfile

@@ -1,32 +0,0 @@
-# Copyright (C) 2019 Intel Corporation.  All rights reserved.
-# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-
-FROM ubuntu:22.04
-
-ENV DEBIAN_FRONTEND=noninteractive
-
-RUN apt-get update && apt-get install -y \
-    cmake build-essential git wget python3.10 python3-pip
-
-RUN wget -q https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-14/wasi-sdk-14.0-linux.tar.gz && \
-    tar xf wasi-sdk-*-linux.tar.gz -C /opt && rm -f wasi-sdk-*-linux.tar.gz && \
-    mv /opt/wasi-sdk-14.0 /opt/wasi-sdk
-
-WORKDIR /home/wamr
-
-COPY core core
-COPY build-scripts build-scripts
-COPY product-mini product-mini
-
-RUN pip3 install -r core/iwasm/libraries/wasi-nn/test/requirements.txt
-
-WORKDIR /home/wamr/core/iwasm/libraries/wasi-nn/test/build
-
-RUN cmake -DWAMR_BUILD_WASI_NN=1 ..
-RUN make -j $(grep -c ^processor /proc/cpuinfo)
-
-WORKDIR /home/wamr/core/iwasm/libraries/wasi-nn/test
-
-RUN ./build.sh
-
-ENTRYPOINT [ "./build/iwasm",  "--dir=.", "test_tensorflow.wasm" ]

core/iwasm/libraries/wasi-nn/test/Dockerfile.base

@@ -0,0 +1,22 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+FROM ubuntu:20.04 AS base
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+RUN apt-get update && apt-get install -y \
+    cmake build-essential git
+
+WORKDIR /home/wamr
+
+COPY . .
+
+WORKDIR /home/wamr/core/iwasm/libraries/wasi-nn/test/build
+
+RUN cmake \
+  -DWAMR_BUILD_WASI_NN=1 \
+  -DTFLITE_ENABLE_GPU=ON \
+  ..
+
+RUN make -j $(grep -c ^processor /proc/cpuinfo)

core/iwasm/libraries/wasi-nn/test/Dockerfile.compile

@@ -0,0 +1,23 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+FROM ubuntu:20.04
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+RUN apt-get update && apt-get install -y \
+    cmake build-essential git wget python3.10 python3-pip
+
+ARG WASI_SDK_VER=19
+RUN wget -c --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-${WASI_SDK_VER}/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz -P /opt \
+  && tar xf /opt/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz -C /opt \
+  && ln -fs /opt/wasi-sdk-${WASI_SDK_VER}.0 /opt/wasi-sdk \
+  && rm /opt/wasi-sdk-${WASI_SDK_VER}.0-linux.tar.gz
+
+WORKDIR /wasi-nn/test
+
+COPY core/iwasm/libraries/wasi-nn/test/requirements.txt .
+
+RUN pip3 install -r requirements.txt && rm requirements.txt
+
+ENTRYPOINT [ "bash", "./build.sh" ]

core/iwasm/libraries/wasi-nn/test/Dockerfile.cpu

@@ -0,0 +1,8 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+FROM ubuntu:20.04
+
+COPY --from=wasi-nn-base /home/wamr/core/iwasm/libraries/wasi-nn/test/build/iwasm /run/iwasm
+
+ENTRYPOINT [ "/run/iwasm" ]

core/iwasm/libraries/wasi-nn/test/Dockerfile.nvidia-gpu

@@ -0,0 +1,20 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+FROM nvidia/cuda:11.3.0-runtime-ubuntu20.04
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        ocl-icd-libopencl1 \
+        ocl-icd-opencl-dev \
+        clinfo && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN mkdir -p /etc/OpenCL/vendors && \
+    echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd
+
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+
+COPY --from=wasi-nn-base /home/wamr/core/iwasm/libraries/wasi-nn/test/build/iwasm /run/iwasm
+
+ENTRYPOINT [ "/run/iwasm" ]

core/iwasm/libraries/wasi-nn/test/build.sh

@@ -7,7 +7,7 @@
     -Wl,--allow-undefined \
     -Wl,--strip-all,--no-entry \
     --sysroot=/opt/wasi-sdk/share/wasi-sysroot \
-    -I/home/wamr/core/iwasm/libraries/wasi-nn \
+    -I.. \
     -o test_tensorflow.wasm test_tensorflow.c
 
 # TFLite models to use in the tests

core/iwasm/libraries/wasi-nn/test/test_tensorflow.c

@@ -28,7 +28,7 @@ typedef struct {
 // WASI-NN wrappers
 
 error
-wasm_load(char *model_name, graph *graph)
+wasm_load(char *model_name, graph *g, execution_target target)
 {
     FILE *pFile = fopen(model_name, "r");
     if (pFile == NULL)
@@ -64,7 +64,7 @@ wasm_load(char *model_name, graph *graph)
     arr.buf[0].size = result;
     arr.buf[0].buf = buffer;
 
-    error res = load(&arr, tensorflow, cpu, graph);
+    error res = load(&arr, tensorflowlite, target, g);
 
     fclose(pFile);
     free(buffer);
@@ -73,13 +73,13 @@ wasm_load(char *model_name, graph *graph)
 }
 
 error
-wasm_init_execution_context(graph graph, graph_execution_context *ctx)
+wasm_init_execution_context(graph g, graph_execution_context *ctx)
 {
-    return init_execution_context(graph, ctx);
+    return init_execution_context(g, ctx);
 }
 
 error
-wasm_input(graph_execution_context ctx, float *input_tensor, uint32_t *dim)
+wasm_set_input(graph_execution_context ctx, float *input_tensor, uint32_t *dim)
 {
     tensor_dimensions dims;
     dims.size = INPUT_TENSOR_DIMS;
@@ -115,11 +115,12 @@ wasm_get_output(graph_execution_context ctx, uint32_t index, float *out_tensor,
 // Inference
 
 float *
-run_inference(float *input, uint32_t *input_size, uint32_t *output_size,
+run_inference(execution_target target, float *input, uint32_t *input_size,
-              char *model_name, uint32_t num_output_tensors)
+              uint32_t *output_size, char *model_name,
+              uint32_t num_output_tensors)
 {
     graph graph;
-    if (wasm_load(model_name, &graph) != success) {
+    if (wasm_load(model_name, &graph, target) != success) {
         fprintf(stderr, "Error when loading model.");
         exit(1);
     }
@@ -130,7 +131,7 @@ run_inference(float *input, uint32_t *input_size, uint32_t *output_size,
         exit(1);
     }
 
-    if (wasm_input(ctx, input, input_size) != success) {
+    if (wasm_set_input(ctx, input, input_size) != success) {
         fprintf(stderr, "Error when setting input tensor.");
         exit(1);
     }
@@ -151,7 +152,7 @@ run_inference(float *input, uint32_t *input_size, uint32_t *output_size,
         *output_size = MAX_OUTPUT_TENSOR_SIZE - *output_size;
         if (wasm_get_output(ctx, i, &out_tensor[offset], output_size)
             != success) {
-            fprintf(stderr, "Error when getting input .");
+            fprintf(stderr, "Error when getting output .");
             exit(1);
         }
 
@@ -185,14 +186,14 @@ create_input(int *dims)
 // TESTS
 
 void
-test_sum()
+test_sum(execution_target target)
 {
     int dims[] = { 1, 5, 5, 1 };
     input_info input = create_input(dims);
 
     uint32_t output_size = 0;
-    float *output = run_inference(input.input_tensor, input.dim, &output_size,
+    float *output = run_inference(target, input.input_tensor, input.dim,
-                                  "models/sum.tflite", 1);
+                                  &output_size, "/assets/models/sum.tflite", 1);
 
     assert(output_size == 1);
     assert(fabs(output[0] - 300.0) < EPSILON);
@@ -203,14 +204,14 @@ test_sum()
 }
 
 void
-test_max()
+test_max(execution_target target)
 {
     int dims[] = { 1, 5, 5, 1 };
     input_info input = create_input(dims);
 
     uint32_t output_size = 0;
-    float *output = run_inference(input.input_tensor, input.dim, &output_size,
+    float *output = run_inference(target, input.input_tensor, input.dim,
-                                  "models/max.tflite", 1);
+                                  &output_size, "/assets/models/max.tflite", 1);
 
     assert(output_size == 1);
     assert(fabs(output[0] - 24.0) < EPSILON);
@@ -222,14 +223,15 @@ test_max()
 }
 
 void
-test_average()
+test_average(execution_target target)
 {
     int dims[] = { 1, 5, 5, 1 };
     input_info input = create_input(dims);
 
     uint32_t output_size = 0;
-    float *output = run_inference(input.input_tensor, input.dim, &output_size,
+    float *output =
-                                  "models/average.tflite", 1);
+        run_inference(target, input.input_tensor, input.dim, &output_size,
+                      "/assets/models/average.tflite", 1);
 
     assert(output_size == 1);
     assert(fabs(output[0] - 12.0) < EPSILON);
@@ -241,14 +243,15 @@ test_average()
 }
 
 void
-test_mult_dimensions()
+test_mult_dimensions(execution_target target)
 {
     int dims[] = { 1, 3, 3, 1 };
     input_info input = create_input(dims);
 
     uint32_t output_size = 0;
-    float *output = run_inference(input.input_tensor, input.dim, &output_size,
+    float *output =
-                                  "models/mult_dim.tflite", 1);
+        run_inference(target, input.input_tensor, input.dim, &output_size,
+                      "/assets/models/mult_dim.tflite", 1);
 
     assert(output_size == 9);
     for (int i = 0; i < 9; i++)
@@ -260,14 +263,15 @@ test_mult_dimensions()
 }
 
 void
-test_mult_outputs()
+test_mult_outputs(execution_target target)
 {
     int dims[] = { 1, 4, 4, 1 };
     input_info input = create_input(dims);
 
     uint32_t output_size = 0;
-    float *output = run_inference(input.input_tensor, input.dim, &output_size,
+    float *output =
-                                  "models/mult_out.tflite", 2);
+        run_inference(target, input.input_tensor, input.dim, &output_size,
+                      "/assets/models/mult_out.tflite", 2);
 
     assert(output_size == 8);
     // first tensor check
@@ -285,16 +289,30 @@ test_mult_outputs()
 int
 main()
 {
+    char *env = getenv("TARGET");
+    if (env == NULL) {
+        printf("Usage:\n--env=\"TARGET=[cpu|gpu]\"\n");
+        return 1;
+    }
+    execution_target target;
+    if (strcmp(env, "cpu") == 0)
+        target = cpu;
+    else if (strcmp(env, "gpu") == 0)
+        target = gpu;
+    else {
+        printf("Wrong target!");
+        return 1;
+    }
     printf("################### Testing sum...\n");
-    test_sum();
+    test_sum(target);
     printf("################### Testing max...\n");
-    test_max();
+    test_max(target);
     printf("################### Testing average...\n");
-    test_average();
+    test_average(target);
     printf("################### Testing multiple dimensions...\n");
-    test_mult_dimensions();
+    test_mult_dimensions(target);
     printf("################### Testing multiple outputs...\n");
-    test_mult_outputs();
+    test_mult_outputs(target);
 
     printf("Tests: passed!\n");
     return 0;

core/iwasm/libraries/wasi-nn/wasi_nn.cmake

@@ -5,6 +5,15 @@ set (WASI_NN_DIR ${CMAKE_CURRENT_LIST_DIR})
 
 add_definitions (-DWASM_ENABLE_WASI_NN=1)
 
-set (LIBC_WASI_NN_SOURCE ${WASI_NN_DIR}/wasi_nn_native.c ${WASI_NN_DIR}/wasi_nn_tensorflow.cpp)
+include_directories (${WASI_NN_DIR})
+include_directories (${WASI_NN_DIR}/src)
+include_directories (${WASI_NN_DIR}/src/utils)
+
+set (
+    LIBC_WASI_NN_SOURCE
+    ${WASI_NN_DIR}/src/wasi_nn.c
+    ${WASI_NN_DIR}/src/wasi_nn_tensorflowlite.cpp
+    ${WASI_NN_DIR}/src/utils/wasi_nn_app_native.c
+)
 
 set (TENSORFLOW_LIB tensorflow-lite)

core/iwasm/libraries/wasi-nn/wasi_nn.h

@@ -3,63 +3,17 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  */
 
-#ifndef WASI_NN_WASM_H
-#define WASI_NN_WASM_H
-
-#include "wasi_nn_common.h"
-
 /**
  * Following definition from:
- * [Aug 10th, 2022]
+ * [Oct 25th, 2022]
- * https://github.com/WebAssembly/wasi-nn/blob/e5e1a6c31f424c7cd63026cd270e9746775675a0/wasi-nn.wit.md
+ * https://github.com/WebAssembly/wasi-nn/blob/0f77c48ec195748990ff67928a4b3eef5f16c2de/wasi-nn.wit.md
  */
 
-/* The graph initialization data. */
+#ifndef WASI_NN_H
+#define WASI_NN_H
 
-// This consists of an array of buffers because implementing backends may encode
+#include <stdint.h>
-// their graph IR in parts (e.g., OpenVINO stores its IR and weights
+#include "wasi_nn_types.h"
-// separately).
-typedef struct {
-    uint8_t *buf;
-    uint32_t size;
-} graph_builder;
-
-typedef struct {
-    graph_builder *buf;
-    uint32_t size;
-} graph_builder_array;
-
-/* The dimensions of a tensor. */
-
-// The array length matches the tensor rank and each element in the array
-// describes the size of each dimension.
-typedef struct {
-    uint32_t *buf;
-    uint32_t size;
-} tensor_dimensions;
-
-/* The tensor data. */
-
-// Initially conceived as a sparse representation, each empty cell would be
-// filled with zeros and the array length must match the product of all of the
-// dimensions and the number of bytes in the type (e.g., a 2x2 tensor with
-// 4-byte f32 elements would have a data array of length 16). Naturally, this
-// representation requires some knowledge of how to lay out data in
-// memory--e.g., using row-major ordering--and could perhaps be improved.
-typedef uint8_t *tensor_data;
-
-/* A tensor. */
-
-typedef struct {
-    // Describe the size of the tensor (e.g., 2x2x2x2 -> [2, 2, 2, 2]). To
-    // represent a tensor containing a single value, use `[1]` for the tensor
-    // dimensions.
-    tensor_dimensions *dimensions;
-    // Describe the type of element in the tensor (e.g., f32).
-    tensor_type type;
-    // Contains the tensor data.
-    tensor_data data;
-} tensor;
 
 /**
  * @brief Load an opaque sequence of bytes to use for inference.
@@ -67,25 +21,31 @@ typedef struct {
  * @param builder   Model builder.
  * @param encoding  Model encoding.
  * @param target    Execution target.
- * @param graph     Graph.
+ * @param g         Graph.
  * @return error    Execution status.
  */
 error
 load(graph_builder_array *builder, graph_encoding encoding,
-     execution_target target, graph *graph)
+     execution_target target, graph *g)
-    __attribute__((export_module("wasi_nn")))
     __attribute__((import_module("wasi_nn")));
 
+/**
+ * INFERENCE
+ *
+ */
+
+// Bind a `graph` to the input and output tensors for an inference.
+typedef uint32_t graph_execution_context;
+
 /**
  * @brief Create an execution instance of a loaded graph.
  *
- * @param graph     Graph.
+ * @param g         Graph.
  * @param ctx       Execution context.
  * @return error    Execution status.
  */
 error
-init_execution_context(graph graph, graph_execution_context *ctx)
+init_execution_context(graph g, graph_execution_context *ctx)
-    __attribute__((export_module("wasi_nn")))
     __attribute__((import_module("wasi_nn")));
 
 /**
@@ -98,7 +58,6 @@ init_execution_context(graph graph, graph_execution_context *ctx)
  */
 error
 set_input(graph_execution_context ctx, uint32_t index, tensor *tensor)
-    __attribute__((export_module("wasi_nn")))
     __attribute__((import_module("wasi_nn")));
 
 /**
@@ -108,8 +67,7 @@ set_input(graph_execution_context ctx, uint32_t index, tensor *tensor)
  * @return error    Execution status.
  */
 error
-compute(graph_execution_context ctx) __attribute__((export_module("wasi_nn")))
+compute(graph_execution_context ctx) __attribute__((import_module("wasi_nn")));
-__attribute__((import_module("wasi_nn")));
 
 /**
  * @brief Extract the outputs after inference.
@@ -126,7 +84,6 @@ __attribute__((import_module("wasi_nn")));
 error
 get_output(graph_execution_context ctx, uint32_t index,
            tensor_data output_tensor, uint32_t *output_tensor_size)
-    __attribute__((export_module("wasi_nn")))
     __attribute__((import_module("wasi_nn")));
 
 #endif

core/iwasm/libraries/wasi-nn/wasi_nn_common.h

@@ -1,44 +0,0 @@
-/*
- * Copyright (C) 2019 Intel Corporation.  All rights reserved.
- * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
- */
-
-#ifndef WASI_NN_COMMON_H
-#define WASI_NN_COMMON_H
-
-#include <stdint.h>
-
-// The type of the elements in a tensor.
-typedef enum { fp16 = 0, fp32, up8, ip32 } tensor_type;
-
-// Describes the encoding of the graph. This allows the API to be implemented by
-// various backends that encode (i.e., serialize) their graph IR with different
-// formats.
-typedef enum { openvino = 0, onnx, tensorflow, pytorch } graph_encoding;
-
-// Define where the graph should be executed.
-typedef enum { cpu = 0, gpu, tpu } execution_target;
-
-// Error codes returned by functions in this API.
-typedef enum {
-    // No error occurred.
-    success = 0,
-    // Caller module passed an invalid argument.
-    invalid_argument,
-    // Invalid encoding.
-    invalid_encoding,
-    // Caller module is missing a memory export.
-    missing_memory,
-    // Device or resource busy.
-    busy,
-    // Runtime Error.
-    runtime_error,
-} error;
-
-// An execution graph for performing inference (i.e., a model).
-typedef uint32_t graph;
-
-// Bind a `graph` to the input and output tensors for an inference.
-typedef uint32_t graph_execution_context;
-
-#endif

core/iwasm/libraries/wasi-nn/wasi_nn_native.c

@@ -1,264 +0,0 @@
-/*
- * Copyright (C) 2019 Intel Corporation.  All rights reserved.
- * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
- */
-
-#include <stdio.h>
-#include <assert.h>
-#include <errno.h>
-#include <string.h>
-#include <stdlib.h>
-
-#include "wasi_nn_common.h"
-#include "wasm_export.h"
-#include "bh_platform.h"
-
-#include "wasi_nn.h"
-#include "wasi_nn_tensorflow.hpp"
-#include "logger.h"
-
-/* Definition of 'wasi_nn.h' structs in WASM app format (using offset) */
-
-typedef struct {
-    uint32_t buf_offset;
-    uint32_t size;
-} graph_builder_wasm;
-
-typedef struct {
-    uint32_t buf_offset;
-    uint32_t size;
-} graph_builder_array_wasm;
-
-typedef struct {
-    uint32_t dimensions_offset;
-    tensor_type type;
-    uint32_t data_offset;
-} tensor_wasm;
-
-typedef struct {
-    uint32_t buf_offset;
-    uint32_t size;
-} tensor_dimensions_wasm;
-
-/* Global variables */
-
-static uint8_t _is_initialized;
-static graph_encoding _encoding;
-
-/* Utils */
-
-static error
-check_initialized()
-{
-    if (!_is_initialized) {
-        NN_ERR_PRINTF("Model not initialized.");
-        return invalid_argument;
-    }
-    if (_encoding != tensorflow) {
-        NN_ERR_PRINTF("Model encoding is not tensorflow.");
-        return invalid_argument;
-    }
-    return success;
-}
-
-/* WASI-NN implementation */
-
-error
-wasi_nn_load(wasm_exec_env_t exec_env, graph_builder_array_wasm *builder,
-             graph_encoding encoding, execution_target target, graph *graph)
-{
-    NN_DBG_PRINTF("Running wasi_nn_load [encoding=%d, target=%d]...", encoding,
-                  target);
-
-    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
-    bh_assert(instance);
-
-    if (!wasm_runtime_validate_native_addr(instance, builder,
-                                           sizeof(graph_builder_array_wasm)))
-        return invalid_argument;
-
-    if (!wasm_runtime_validate_app_addr(instance, builder->buf_offset,
-                                        builder->size * sizeof(uint32_t)))
-        return invalid_argument;
-
-    NN_DBG_PRINTF("Graph builder array contains %d elements", builder->size);
-
-    graph_builder_wasm *gb_wasm =
-        (graph_builder_wasm *)wasm_runtime_addr_app_to_native(
-            instance, builder->buf_offset);
-
-    graph_builder *gb_native = (graph_builder *)wasm_runtime_malloc(
-        builder->size * sizeof(graph_builder));
-    if (gb_native == NULL)
-        return missing_memory;
-
-    for (int i = 0; i < builder->size; ++i) {
-        if (!wasm_runtime_validate_app_addr(instance, gb_wasm[i].buf_offset,
-                                            gb_wasm[i].size
-                                                * sizeof(uint8_t))) {
-            wasm_runtime_free(gb_native);
-            return invalid_argument;
-        }
-
-        gb_native[i].buf = (uint8_t *)wasm_runtime_addr_app_to_native(
-            instance, gb_wasm[i].buf_offset);
-        gb_native[i].size = gb_wasm[i].size;
-
-        NN_DBG_PRINTF("Graph builder %d contains %d elements", i,
-                      gb_wasm[i].size);
-    }
-
-    graph_builder_array gba_native = { .buf = gb_native,
-                                       .size = builder->size };
-
-    if (!wasm_runtime_validate_native_addr(instance, graph, sizeof(graph))) {
-        wasm_runtime_free(gb_native);
-        return invalid_argument;
-    }
-
-    switch (encoding) {
-        case tensorflow:
-            break;
-        default:
-            NN_ERR_PRINTF("Only tensorflow is supported.");
-            wasm_runtime_free(gb_native);
-            return invalid_argument;
-    }
-
-    _encoding = encoding;
-    _is_initialized = 1;
-
-    error res = tensorflow_load(gba_native, _encoding, target, graph);
-    NN_DBG_PRINTF("wasi_nn_load finished with status %d [graph=%d]", res,
-                  *graph);
-
-    wasm_runtime_free(gb_native);
-    return res;
-}
-
-error
-wasi_nn_init_execution_context(wasm_exec_env_t exec_env, graph graph,
-                               graph_execution_context *ctx)
-{
-    NN_DBG_PRINTF("Running wasi_nn_init_execution_context [graph=%d]...",
-                  graph);
-    error res;
-    if (success != (res = check_initialized()))
-        return res;
-    res = tensorflow_init_execution_context(graph);
-    *ctx = graph;
-    NN_DBG_PRINTF(
-        "wasi_nn_init_execution_context finished with status %d [ctx=%d]", res,
-        *ctx);
-    return res;
-}
-
-error
-wasi_nn_set_input(wasm_exec_env_t exec_env, graph_execution_context ctx,
-                  uint32_t index, tensor_wasm *input_tensor)
-{
-    NN_DBG_PRINTF("Running wasi_nn_set_input [ctx=%d, index=%d]...", ctx,
-                  index);
-
-    error res;
-    if (success != (res = check_initialized()))
-        return res;
-
-    wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
-    bh_assert(instance);
-
-    if (!wasm_runtime_validate_native_addr(instance, input_tensor,
-                                           sizeof(tensor_wasm)))
-        return invalid_argument;
-
-    if (!wasm_runtime_validate_app_addr(
-            instance, input_tensor->dimensions_offset, sizeof(uint32_t)))
-        return invalid_argument;
-
-    tensor_dimensions_wasm *dimensions_w =
-        (tensor_dimensions_wasm *)wasm_runtime_addr_app_to_native(
-            instance, input_tensor->dimensions_offset);
-
-    if (!wasm_runtime_validate_app_addr(instance, dimensions_w->buf_offset,
-                                        dimensions_w->size * sizeof(uint32_t)))
-        return invalid_argument;
-
-    tensor_dimensions dimensions = {
-        .buf = (uint32_t *)wasm_runtime_addr_app_to_native(
-            instance, dimensions_w->buf_offset),
-        .size = dimensions_w->size
-    };
-
-    NN_DBG_PRINTF("Number of dimensions: %d", dimensions.size);
-    int total_elements = 1;
-    for (int i = 0; i < dimensions.size; ++i) {
-        NN_DBG_PRINTF("Dimension %d: %d", i, dimensions.buf[i]);
-        total_elements *= dimensions.buf[i];
-    }
-    NN_DBG_PRINTF("Tensor type: %d", input_tensor->type);
-
-    if (!wasm_runtime_validate_app_addr(instance, input_tensor->data_offset,
-                                        total_elements))
-        return invalid_argument;
-
-    tensor tensor = { .type = input_tensor->type,
-                      .dimensions = &dimensions,
-                      .data = (uint8_t *)wasm_runtime_addr_app_to_native(
-                          instance, input_tensor->data_offset) };
-
-    res = tensorflow_set_input(ctx, index, &tensor);
-    NN_DBG_PRINTF("wasi_nn_set_input finished with status %d", res);
-    return res;
-}
-
-error
-wasi_nn_compute(wasm_exec_env_t exec_env, graph_execution_context ctx)
-{
-    NN_DBG_PRINTF("Running wasi_nn_compute [ctx=%d]...", ctx);
-    error res;
-    if (success != (res = check_initialized()))
-        return res;
-
-    res = tensorflow_compute(ctx);
-    NN_DBG_PRINTF("wasi_nn_compute finished with status %d", res);
-    return res;
-}
-
-error
-wasi_nn_get_output(wasm_exec_env_t exec_env, graph_execution_context ctx,
-                   uint32_t index, tensor_data output_tensor,
-                   uint32_t *output_tensor_size)
-{
-    NN_DBG_PRINTF("Running wasi_nn_get_output [ctx=%d, index=%d]...", ctx,
-                  index);
-    error res;
-    if (success != (res = check_initialized()))
-        return res;
-
-    res = tensorflow_get_output(ctx, index, output_tensor, output_tensor_size);
-    NN_DBG_PRINTF("wasi_nn_get_output finished with status %d [data_size=%d]",
-                  res, *output_tensor_size);
-    return res;
-}
-
-/* Register WASI-NN in WAMR */
-
-/* clang-format off */
-#define REG_NATIVE_FUNC(func_name, signature) \
-    { #func_name, wasi_nn_##func_name, signature, NULL }
-/* clang-format on */
-
-static NativeSymbol native_symbols_wasi_nn[] = {
-    REG_NATIVE_FUNC(load, "(*ii*)i"),
-    REG_NATIVE_FUNC(init_execution_context, "(i*)i"),
-    REG_NATIVE_FUNC(set_input, "(ii*)i"),
-    REG_NATIVE_FUNC(compute, "(i)i"),
-    REG_NATIVE_FUNC(get_output, "(ii**)i"),
-};
-
-uint32_t
-get_wasi_nn_export_apis(NativeSymbol **p_libc_wasi_apis)
-{
-    *p_libc_wasi_apis = native_symbols_wasi_nn;
-    return sizeof(native_symbols_wasi_nn) / sizeof(NativeSymbol);
-}

core/iwasm/libraries/wasi-nn/wasi_nn_tensorflow.hpp

@@ -1,40 +0,0 @@
-/*
- * Copyright (C) 2019 Intel Corporation.  All rights reserved.
- * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
- */
-
-#ifndef WASI_NN_TENSORFLOW_HPP
-#define WASI_NN_TENSORFLOW_HPP
-
-#include <stdio.h>
-
-#include "wasi_nn.h"
-#include "logger.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-error
-tensorflow_load(graph_builder_array builder, graph_encoding encoding,
-                execution_target target, graph *graph);
-
-error
-tensorflow_init_execution_context(graph graph);
-
-error
-tensorflow_set_input(graph_execution_context ctx, uint32_t index,
-                     tensor *input_tensor);
-
-error
-tensorflow_compute(graph_execution_context ctx);
-
-error
-tensorflow_get_output(graph_execution_context context, uint32_t index,
-                      tensor_data output_tensor, uint32_t *output_tensor_size);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif

core/iwasm/libraries/wasi-nn/wasi_nn_types.h

@@ -0,0 +1,106 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#ifndef WASI_NN_TYPES_H
+#define WASI_NN_TYPES_H
+
+/**
+ * ERRORS
+ *
+ */
+
+// Error codes returned by functions in this API.
+typedef enum {
+    // No error occurred.
+    success = 0,
+    // Caller module passed an invalid argument.
+    invalid_argument,
+    // Invalid encoding.
+    invalid_encoding,
+    // Caller module is missing a memory export.
+    missing_memory,
+    // Device or resource busy.
+    busy,
+    // Runtime Error.
+    runtime_error,
+} error;
+
+/**
+ * TENSOR
+ *
+ */
+
+// The dimensions of a tensor.
+//
+// The array length matches the tensor rank and each element in the array
+// describes the size of each dimension.
+typedef struct {
+    uint32_t *buf;
+    uint32_t size;
+} tensor_dimensions;
+
+// The type of the elements in a tensor.
+typedef enum { fp16 = 0, fp32, up8, ip32 } tensor_type;
+
+// The tensor data.
+//
+// Initially conceived as a sparse representation, each empty cell would be
+// filled with zeros and the array length must match the product of all of the
+// dimensions and the number of bytes in the type (e.g., a 2x2 tensor with
+// 4-byte f32 elements would have a data array of length 16). Naturally, this
+// representation requires some knowledge of how to lay out data in
+// memory--e.g., using row-major ordering--and could perhaps be improved.
+typedef uint8_t *tensor_data;
+
+// A tensor.
+typedef struct {
+    // Describe the size of the tensor (e.g., 2x2x2x2 -> [2, 2, 2, 2]). To
+    // represent a tensor containing a single value, use `[1]` for the tensor
+    // dimensions.
+    tensor_dimensions *dimensions;
+    // Describe the type of element in the tensor (e.g., f32).
+    tensor_type type;
+    // Contains the tensor data.
+    tensor_data data;
+} tensor;
+
+/**
+ * GRAPH
+ *
+ */
+
+// The graph initialization data.
+//
+// This consists of an array of buffers because implementing backends may encode
+// their graph IR in parts (e.g., OpenVINO stores its IR and weights
+// separately).
+typedef struct {
+    uint8_t *buf;
+    uint32_t size;
+} graph_builder;
+
+typedef struct {
+    graph_builder *buf;
+    uint32_t size;
+} graph_builder_array;
+
+// An execution graph for performing inference (i.e., a model).
+typedef uint32_t graph;
+
+// Describes the encoding of the graph. This allows the API to be implemented by
+// various backends that encode (i.e., serialize) their graph IR with different
+// formats.
+typedef enum {
+    openvino = 0,
+    onnx,
+    tensorflow,
+    pytorch,
+    tensorflowlite
+} graph_encoding;
+
+// Define where the graph should be executed.
+typedef enum execution_target { cpu = 0, gpu, tpu } execution_target;
+
+#endif

core/shared/platform/esp-idf/espidf_socket.c

@@ -77,7 +77,7 @@ os_socket_bind(bh_socket_t socket, const char *host, int *port)
     }
 
     socklen = sizeof(addr);
-    if (getsockname(socket, (void *)&addr, &socklen) == -1) {
+    if (getsockname(socket, (struct sockaddr *)&addr, &socklen) == -1) {
         goto fail;
     }
 
@@ -120,7 +120,7 @@ os_socket_accept(bh_socket_t server_sock, bh_socket_t *sock, void *addr,
                  unsigned int *addrlen)
 {
     struct sockaddr addr_tmp;
-    unsigned int len = sizeof(struct sockaddr);
+    socklen_t len = sizeof(struct sockaddr);
 
     *sock = accept(server_sock, (struct sockaddr *)&addr_tmp, &len);
 
@@ -205,7 +205,7 @@ os_socket_addr_remote(bh_socket_t socket, bh_sockaddr_t *sockaddr)
     struct sockaddr_in addr;
     socklen_t addr_len = sizeof(addr);
 
-    if (getpeername(socket, &addr, &addr_len) == -1) {
+    if (getpeername(socket, (struct sockaddr *)&addr, &addr_len) == -1) {
         return BHT_ERROR;
     }
 
@@ -219,7 +219,7 @@ os_socket_addr_local(bh_socket_t socket, bh_sockaddr_t *sockaddr)
     struct sockaddr_in addr;
     socklen_t addr_len = sizeof(addr);
 
-    if (getsockname(socket, &addr, &addr_len) == -1) {
+    if (getsockname(socket, (struct sockaddr *)&addr, &addr_len) == -1) {
         return BHT_ERROR;
     }
 

core/shared/platform/linux-sgx/sgx_ipfs.c

@@ -16,6 +16,11 @@
 #define SGX_ERROR_FILE_LOWEST_ERROR_ID SGX_ERROR_FILE_BAD_STATUS
 #define SGX_ERROR_FILE_HIGHEST_ERROR_ID SGX_ERROR_FILE_CLOSE_FAILED
 
+// Internal buffer filled with zeroes and used when extending the size of
+// protected files.
+#define ZEROES_PADDING_LENGTH 32 * 1024
+char zeroes_padding[ZEROES_PADDING_LENGTH] = { 0 };
+
 // The mapping between file descriptors and IPFS file pointers.
 static HashMap *ipfs_file_list;
 
@@ -78,6 +83,27 @@ ipfs_file_destroy(void *sgx_file)
     sgx_fclose(sgx_file);
 }
 
+// Writes a given number of zeroes in file at the current offset.
+// The return value is zero if successful; otherwise non-zero.
+static int
+ipfs_write_zeroes(void *sgx_file, size_t len)
+{
+    int min_count;
+
+    while (len > 0) {
+        min_count = len < ZEROES_PADDING_LENGTH ? len : ZEROES_PADDING_LENGTH;
+
+        if (sgx_fwrite(zeroes_padding, 1, min_count, sgx_file) == 0) {
+            errno = convert_sgx_errno(sgx_ferror(sgx_file));
+            return -1;
+        }
+
+        len -= min_count;
+    }
+
+    return 0;
+}
+
 int
 ipfs_init()
 {
@@ -104,7 +130,7 @@ ipfs_posix_fallocate(int fd, off_t offset, size_t len)
 
     // The wrapper for fseek takes care of extending the file if sought beyond
     // the end
-    if (ipfs_lseek(fd, offset + len, SEEK_CUR) == -1) {
+    if (ipfs_lseek(fd, offset + len, SEEK_SET) == -1) {
         return errno;
     }
 
@@ -354,7 +380,7 @@ ipfs_fflush(int fd)
 off_t
 ipfs_lseek(int fd, off_t offset, int nwhence)
 {
-    off_t new_offset;
+    off_t cursor_current_location;
     void *sgx_file = fd2file(fd);
     if (!sgx_file) {
         errno = EBADF;
@@ -364,20 +390,20 @@ ipfs_lseek(int fd, off_t offset, int nwhence)
     // Optimization: if the offset is 0 and the whence is SEEK_CUR,
     // this is equivalent of a call to ftell.
     if (offset == 0 && nwhence == SEEK_CUR) {
-        int64_t ftell_result = (off_t)sgx_ftell(sgx_file);
+        cursor_current_location = (off_t)sgx_ftell(sgx_file);
 
-        if (ftell_result == -1) {
+        if (cursor_current_location == -1) {
             errno = convert_sgx_errno(sgx_ferror(sgx_file));
             return -1;
         }
 
-        return ftell_result;
+        return cursor_current_location;
     }
 
     int fseek_result = sgx_fseek(sgx_file, offset, nwhence);
 
     if (fseek_result == 0) {
-        new_offset = (__wasi_filesize_t)sgx_ftell(sgx_file);
+        off_t new_offset = (off_t)sgx_ftell(sgx_file);
 
         if (new_offset == -1) {
             errno = convert_sgx_errno(sgx_ferror(sgx_file));
@@ -405,17 +431,39 @@ ipfs_lseek(int fd, off_t offset, int nwhence)
         // manually.
 
         // Assume the error is raised because the cursor is moved beyond the end
-        // of the file. Try to move the cursor at the end of the file.
+        // of the file.
+
+        // If the whence is the current cursor location, retrieve it
+        if (nwhence == SEEK_CUR) {
+            cursor_current_location = (off_t)sgx_ftell(sgx_file);
+        }
+
+        // Move the cursor at the end of the file
         if (sgx_fseek(sgx_file, 0, SEEK_END) == -1) {
             errno = convert_sgx_errno(sgx_ferror(sgx_file));
             return -1;
         }
 
+        // Compute the number of zeroes to append.
+        int64_t number_of_zeroes;
+        switch (nwhence) {
+            case SEEK_SET:
+                number_of_zeroes = offset - sgx_ftell(sgx_file);
+                break;
+            case SEEK_END:
+                number_of_zeroes = offset;
+                break;
+            case SEEK_CUR:
+                number_of_zeroes =
+                    cursor_current_location + offset - sgx_ftell(sgx_file);
+                break;
+            default:
+                errno = EINVAL;
+                return -1;
+        }
+
         // Write the missing zeroes
-        char zero = 0;
+        if (ipfs_write_zeroes(sgx_file, number_of_zeroes) != 0) {
-        int64_t number_of_zeroes = offset - sgx_ftell(sgx_file);
-        if (sgx_fwrite(&zero, 1, number_of_zeroes, sgx_file) == 0) {
-            errno = convert_sgx_errno(sgx_ferror(sgx_file));
             return -1;
         }
 
@@ -468,9 +516,7 @@ ipfs_ftruncate(int fd, off_t len)
 
     // Increasing the size is equal to writing from the end of the file
     // with null bytes.
-    char null_byte = 0;
+    if (ipfs_write_zeroes(sgx_file, len - file_size) != 0) {
-    if (sgx_fwrite(&null_byte, 1, len - file_size, sgx_file) == 0) {
-        errno = convert_sgx_errno(sgx_ferror(sgx_file));
         return -1;
     }
 

doc/build_wasm_app.md

@@ -3,12 +3,12 @@
 Prepare WASM building environments
 ==================================
 
-For C and C++, WASI-SDK version 12.0+ is the major tool supported by WAMR to build WASM applications. Also, we can use [Emscripten SDK (EMSDK)](https://github.com/emscripten-core/emsdk), but it is not recommended. And there are some other compilers such as the standard clang compiler, which might also work [here](./other_wasm_compilers.md).
+For C and C++, WASI-SDK version 19.0+ is the major tool supported by WAMR to build WASM applications. Also, we can use [Emscripten SDK (EMSDK)](https://github.com/emscripten-core/emsdk), but it is not recommended. And there are some other compilers such as the standard clang compiler, which might also work [here](./other_wasm_compilers.md).
 
 To install WASI SDK, please download the [wasi-sdk release](https://github.com/CraneStation/wasi-sdk/releases) and extract the archive to default path `/opt/wasi-sdk`.
 
 The official *wasi-sdk release* doesn't fully support *latest 128-bit SIMD spec* yet. WAMR provides a script in [build-wasi-sdk](../test-tools/build-wasi-sdk/) to generate
-another wasi-sdk with *llvm-13* from source code and installs it at *../test-tools/wasi-sdk*. If you plan to build WASM applications with *latest 128-bit SIMD*, please use it instead of the official release.
+another wasi-sdk with *llvm-15* from source code and installs it at *../test-tools/wasi-sdk*. If you plan to build WASM applications with *latest 128-bit SIMD*, please use it instead of the official release.
 
 And [sample workloads](../samples/workload) are using the self-compiled wasi-sdk.
 

doc/port_wamr.md

@@ -10,9 +10,13 @@ This document describes how to port WAMR to a new platform "**new-os**"
 # Step 1: Implement platform API layer
 
 -------------------------
-Firstly create the folder **`core/shared/platform/new-os`** for platform API layer implementations. In the folder you just created, you must provide the following files:
+Firstly create the folder for platform API layer implementations:
+* for common platforms, create a folder in **`core/shared/platform/new-os`** in WAMR repository folder, so the implementation can be upstreamed
+* for platforms that are internal and its implementation shouldn't be published, it's recommended to create a folder outside of the WAMR repository folder (e.g. have separate repository for platform API layer implementation)
 
-- `platform_internal.h`: It can be used for any platform specific definitions such as macros, data types and internal APIs.
+In the folder you just created, you must provide the following files:
+
+- `platform_internal.h`: It can be used for any platform-specific definitions such as macros, data types and internal APIs.
 
 - `shared_platform.cmake`: the cmake file will be included by the building script. It is recommended to add a definition for your platform:
 
@@ -20,16 +24,16 @@ Firstly create the folder **`core/shared/platform/new-os`** for platform API lay
     add_definitions(-DBH_PLATFORM_YOUR_NAME)
     ```
 
-Then go to implement the APIs defined in following header files for the platform abstraction layer:
+Then go to implement the APIs defined in the following header files for the platform abstraction layer:
 
-- [`platform_api_vmcore.h`](../core/shared/platform/include/platform_api_vmcore.h):   mandatory for building mini-product (vmcore only). Part of APIs are needed only for Ahead of Time compilation support. 
+- [`platform_api_vmcore.h`](../core/shared/platform/include/platform_api_vmcore.h):   mandatory for building mini-product (vmcore only). Part of the APIs is needed only for Ahead of Time compilation support.
 - [`platform_api_extension.h`](../core/shared/platform/include/platform_api_extension.h): mandatory for app-mgr and app-framework. Given that the app-mgr and app-framework are not required for your target platform, you won't have to implement the API defined in the `platform_api_extension.h`.
 
 
 
 **common/posix:**
 
-There is posix based implementation of the platform API located in the `platform/common/posix` folder. You can include it if your platform support posix API. refer to platform linux implementation.
+There is posix based implementation of the platform API located in the `platform/common/posix` folder. You can include it if your platform supports posix API. refer to platform linux implementation.
 
 
 
@@ -42,7 +46,7 @@ Some platforms such as ZephyrOS don't provide math functions e.g. sqrt, fabs and
 # Step 2: Create the mini product for the platform
 
 -------------------------
-You can build a mini WAMR product which is only the vmcore for you platform. Normally you need to implement the main function which loads a WASM file and run it with the WASM runtime. You don't have to do this step if there is no mini-product need for your platform porting.
+You can build a mini WAMR product which is only the vmcore for your platform. Normally you need to implement the main function which loads a WASM file and run it with the WASM runtime. You don't have to do this step if there is no mini-product need for your platform porting.
 
 
 
@@ -58,6 +62,11 @@ cd build
 cmake .. -DWAMR_BUILD_PLATFORM=new-os
 ```
 
+For platform implementations that are outside of the WAMR repository (e.g. internal platforms), you also need to provide `SHARED_PLATFORM_CONFIG` path:
+
+```
+cmake .. -DWAMR_BUILD_PLATFORM=new-os -DSHARED_PLATFORM_CONFIG=/path/to/new-os/shared_platform.cmake
+```
 
 
 Refer to [build_wamr.md](./build_wamr.md) for the building configurations and parameters.

language-bindings/python/MANIFEST.in

@@ -0,0 +1 @@
+include src/wamr/libs/*

language-bindings/python/README.md

@@ -1,31 +1,34 @@
 # wamr-python
 
+The WAMR Python package contains a set of high-level bindings for WAMR API and WASM-C-API.
+
 ## Installation
 
-### Installing from the source code
+To Install from local source tree in _development mode_ run the following command,
-
-Installing from local source tree is in _development mode_. The package appears to be installed but still is editable from the source tree.
 
 ```bash
-$ python -m pip install -e /path/to/wamr-root/binding/python
+python -m pip install -e .
 ```
 
+In this mode the package appears to be installed but still is editable from the source tree.
+
 ## Usage
 
-```python
+From the same package you can use two set of APIs.
-import wamr.ffi as ffi
+
+To use the WAMR API you can import the symbols as follows,
+
+```py
+from wamr.wamrapi.wamr import Engine, Module, Instance, ExecEnv
 ```
 
-### Preparation
+In the order hand, to use the WASM-C-API,
 
-The binding will load the shared library _libiwasm.so_ from the WAMR repo. So before running the binding, you need to build the library yourself.
+```py
+import wamr.wasmcapi.ffi as ffi
+```
 
-The default compile options are good enough.
+For more information:
 
-Please be aware that `wasm_frame_xxx` and `wasm_trap_xxx` only work well when enabling `WAMR_BUILD_DUMP_CALL_STACK`.
+* [WAMR API](./wamr_api)
-
+* [WASM-C-API](./wasm_c_api)
-### Examples
-
-There is a [simple example](./samples/hello_procedural.py) to show how to use bindings. Actually, the python binding follows C-APIs. There it should be easy if be familiar with _programming with wasm-c-api_.
-
-Unit test cases under _./tests_ could be another but more complete references.

language-bindings/python/setup.py

@@ -8,7 +8,28 @@
 # pylint: disable=missing-function-docstring
 # pylint: disable=missing-module-docstring
 
-from setuptools import setup, find_packages
+import pathlib
+from setuptools import setup
+from setuptools.command.develop import develop
+from setuptools.command.install import install
+from subprocess import check_call
+
+
+def build_library():
+    cur_path = pathlib.Path(__file__).parent
+    check_call(f"{cur_path}/utils/create_lib.sh".split())
+
+class PreDevelopCommand(develop):
+    """Pre-installation for development mode."""
+    def run(self):
+        build_library()
+        develop.run(self)
+
+class PreInstallCommand(install):
+    """Pre-installation for installation mode."""
+    def run(self):
+        build_library()
+        install.run(self)
 
 
 with open("README.md") as f:
@@ -24,7 +45,11 @@ setup(
     long_description=readme,
     author="The WAMR Project Developers",
     author_email="hello@bytecodealliance.org",
-    url="https://github.com/bytecodealliance/wamr-python",
+    url="https://github.com/bytecodealliance/wasm-micro-runtime",
     license=license,
-    packages=["wamr"],
+    include_package_data=True,
+    cmdclass={
+        'develop': PreDevelopCommand,
+        'install': PreInstallCommand,
+    },
 )

language-bindings/python/src/wamr/wamrapi/wamr.py

@@ -0,0 +1,123 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+from ctypes import Array
+from ctypes import c_char
+from ctypes import c_uint
+from ctypes import c_uint8
+from ctypes import c_void_p
+from ctypes import cast
+from ctypes import create_string_buffer
+from ctypes import POINTER
+from ctypes import pointer
+
+from wamr.wamrapi.iwasm import Alloc_With_Pool
+from wamr.wamrapi.iwasm import RuntimeInitArgs
+from wamr.wamrapi.iwasm import wasm_exec_env_t
+from wamr.wamrapi.iwasm import wasm_function_inst_t
+from wamr.wamrapi.iwasm import wasm_module_inst_t
+from wamr.wamrapi.iwasm import wasm_module_t
+from wamr.wamrapi.iwasm import wasm_runtime_call_wasm
+from wamr.wamrapi.iwasm import wasm_runtime_create_exec_env
+from wamr.wamrapi.iwasm import wasm_runtime_deinstantiate
+from wamr.wamrapi.iwasm import wasm_runtime_destroy
+from wamr.wamrapi.iwasm import wasm_runtime_destroy_exec_env
+from wamr.wamrapi.iwasm import wasm_runtime_full_init
+from wamr.wamrapi.iwasm import wasm_runtime_instantiate
+from wamr.wamrapi.iwasm import wasm_runtime_load
+from wamr.wamrapi.iwasm import wasm_runtime_lookup_function
+from wamr.wamrapi.iwasm import wasm_runtime_unload
+
+
+class Engine:
+    def __init__(self):
+        self.init_args = self._get_init_args()
+        wasm_runtime_full_init(pointer(self.init_args))
+
+    def __del__(self):
+        print("deleting Engine")
+        wasm_runtime_destroy()
+
+    def _get_init_args(self, heap_size: int = 1024 * 512) -> RuntimeInitArgs:
+        init_args = RuntimeInitArgs()
+        init_args.mem_alloc_type = Alloc_With_Pool
+        init_args.mem_alloc_option.pool.heap_buf = cast(
+            (c_char * heap_size)(), c_void_p
+        )
+        init_args.mem_alloc_option.pool.heap_size = heap_size
+        return init_args
+
+
+class Module:
+    __create_key = object()
+
+    @classmethod
+    def from_file(cls, engine: Engine, fp: str) -> "Module":
+        return Module(cls.__create_key, engine, fp)
+
+    def __init__(self, create_key: object, engine: Engine, fp: str) -> None:
+        assert (
+            create_key == Module.__create_key
+        ), "Module objects must be created using Module.from_file"
+        self.engine = engine
+        self.module, self.file_data = self._create_module(fp)
+
+    def __del__(self):
+        print("deleting Module")
+        wasm_runtime_unload(self.module)
+
+    def _create_module(self, fp: str) -> tuple[wasm_module_t, Array[c_uint]]:
+        with open(fp, "rb") as f:
+            data = f.read()
+            data = (c_uint8 * len(data))(*data)
+
+        error_buf = create_string_buffer(128)
+        module = wasm_runtime_load(data, len(data), error_buf, len(error_buf))
+        if not module:
+            raise Exception("Error while creating module")
+        return module, data
+
+
+class Instance:
+    def __init__(self, module: Module, stack_size: int = 65536, heap_size: int = 16384):
+        self.module = module
+        self.module_inst = self._create_module_inst(module, stack_size, heap_size)
+
+    def __del__(self):
+        print("deleting Instance")
+        wasm_runtime_deinstantiate(self.module_inst)
+
+    def lookup_function(self, name: str):
+        func = wasm_runtime_lookup_function(self.module_inst, name, None)
+        if not func:
+            raise Exception("Error while looking-up function")
+        return func
+
+    def _create_module_inst(self, module: Module, stack_size: int, heap_size: int) -> wasm_module_inst_t:
+        error_buf = create_string_buffer(128)
+        module_inst = wasm_runtime_instantiate(
+            module.module, stack_size, heap_size, error_buf, len(error_buf)
+        )
+        if not module_inst:
+            raise Exception("Error while creating module instance")
+        return module_inst
+
+
+class ExecEnv:
+    def __init__(self, module_inst: Instance, stack_size: int = 65536):
+        self.module_inst = module_inst
+        self.exec_env = self._create_exec_env(module_inst, stack_size)
+
+    def __del__(self):
+        print("deleting ExecEnv")
+        wasm_runtime_destroy_exec_env(self.exec_env)
+
+    def call(self, func: wasm_function_inst_t, argc: int, argv: "POINTER[c_uint]"):
+        if not wasm_runtime_call_wasm(self.exec_env, func, argc, argv):
+            raise Exception("Error while calling function")
+
+    def _create_exec_env(self, module_inst: Instance, stack_size: int) -> wasm_exec_env_t:
+        exec_env = wasm_runtime_create_exec_env(module_inst.module_inst, stack_size)
+        if not exec_env:
+            raise Exception("Error while creating execution environment")
+        return exec_env

language-bindings/python/src/wamr/wasmcapi/ffi.py

@@ -36,8 +36,8 @@ current_file = Path(__file__)
 if current_file.is_symlink():
     current_file = Path(os.readlink(current_file))
 current_dir = current_file.parent.resolve()
-root_dir = current_dir.parent.parent.parent.parent.resolve()
+root_dir = current_dir.parents[4].resolve()
-wamr_dir = root_dir.joinpath("wasm-micro-runtime").resolve()
+wamr_dir = root_dir.resolve()
 if not wamr_dir.exists():
     raise RuntimeError(f"not found the repo of wasm-micro-runtime under {root_dir}")
 

language-bindings/python/utils/create_lib.sh

@@ -0,0 +1,17 @@
+#!/bin/sh
+
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+CUR_DIR=$(cd $(dirname $0) && pwd -P)
+ROOT_DIR=${CUR_DIR}/../../..
+
+WAMR_BUILD_PLATFORM=${WAMR_BUILD_PLATFORM:-"linux"}
+
+cd ${ROOT_DIR}/product-mini/platforms/${WAMR_BUILD_PLATFORM}
+
+mkdir -p build && cd build
+cmake ..
+make -j
+
+cp libiwasm.so ${CUR_DIR}/../src/wamr/libs

language-bindings/python/wamr-api/README.md

@@ -0,0 +1,25 @@
+# WARM API
+
+## Examples
+
+Copy in `language-bindings/python/wamr/libs` the library `libiwasm` generated from `product-mini/platforms`.
+
+There is a [simple example](./samples/main.py) to show how to use bindings.
+
+```
+python samples/main.py
+```
+
+## Update WAMR API bindings
+
+Install requirements,
+
+```
+pip install -r requirements.txt
+```
+
+Run the following command,
+
+```sh
+ctypesgen ../../../../core/iwasm/include/wasm_export.h -l ../libs/libiwasm.so -o iwasm.py
+```

language-bindings/python/wamr-api/requirements.txt

@@ -0,0 +1 @@
+ctypesgen==1.1.1

language-bindings/python/wamr-api/samples/compile.sh

@@ -0,0 +1,11 @@
+#!/bin/sh
+
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+/opt/wasi-sdk/bin/clang     \
+    -O0 -z stack-size=4096 -Wl,--initial-memory=65536 \
+    -Wl,--strip-all,--no-entry -nostdlib \
+    -Wl,--export=sum\
+    -Wl,--allow-undefined \
+    -o test.wasm sum.c

language-bindings/python/wamr-api/samples/main.py

@@ -0,0 +1,22 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+from wamr.wamrapi.wamr import Engine, Module, Instance, ExecEnv
+from ctypes import c_uint
+import pathlib
+
+def main():
+    engine = Engine()
+    module = Module.from_file(engine, pathlib.Path(__file__).parent / "sum.wasm")
+    module_inst = Instance(module)
+    exec_env = ExecEnv(module_inst)
+
+    func = module_inst.lookup_function("sum")
+
+    argv = (c_uint * 2)(*[10, 11])
+    exec_env.call(func, len(argv), argv)
+    print(argv[0])
+
+
+if __name__ == "__main__":
+    main()

language-bindings/python/wamr-api/samples/sum.c

@@ -0,0 +1,12 @@
+/*
+ * Copyright (C) 2019 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include <stdio.h>
+
+int
+sum(int a, int b)
+{
+    return a + b;
+}

language-bindings/python/wasm-c-api/README.md

@@ -0,0 +1,7 @@
+# WASM-C-API
+
+## Examples
+
+There is a [simple example](./samples/hello_procedural.py) to show how to use bindings. Actually, the python binding follows C-APIs. There it should be easy if be familiar with _programming with wasm-c-api_.
+
+Unit test cases under _./tests_ could be another but more complete references.