diff --git a/.github/actions/install-wasi-sdk-wabt/action.yml b/.github/actions/install-wasi-sdk-wabt/action.yml
index c872e4252..6b79087c2 100644
--- a/.github/actions/install-wasi-sdk-wabt/action.yml
+++ b/.github/actions/install-wasi-sdk-wabt/action.yml
@@ -30,14 +30,23 @@ runs:
       if: ${{ startsWith(inputs.os, 'ubuntu') }}
       shell: bash
       run: |
+        echo "Downloading wasi-sdk for Ubuntu..."
         sudo wget -O wasi-sdk.tar.gz --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-x86_64-linux.tar.gz
+
+        echo "Extracting wasi-sdk..."
         sudo tar -xf wasi-sdk.tar.gz
         sudo ln -sf  wasi-sdk-25.0-x86_64-linux/ wasi-sdk
+
+        echo "Downloading wabt for Ubuntu..."
         sudo wget -O wabt.tar.gz     --progress=dot:giga https://github.com/WebAssembly/wabt/releases/download/1.0.37/wabt-1.0.37-ubuntu-20.04.tar.gz
+
+        echo "Extracting wabt..."
         sudo tar -xf wabt.tar.gz
         sudo ln -sf  wabt-1.0.37 wabt
+
         /opt/wasi-sdk/bin/clang --version
         /opt/wabt/bin/wasm-interp --version
+
         echo "::notice::wasi-sdk-25 and wabt-1.0.37 installed on ubuntu"
       working-directory: /opt
 
@@ -45,14 +54,23 @@ runs:
       if: ${{ inputs.os == 'macos-13' }}
       shell: bash
       run: |
+        echo "Downloading wasi-sdk for macOS-13..."
         sudo wget -O wasi-sdk.tar.gz --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-x86_64-macos.tar.gz
+
+        echo "Extracting wasi-sdk..."
         sudo tar -xf wasi-sdk.tar.gz
         sudo ln -sf  wasi-sdk-25.0-x86_64-macos wasi-sdk
+
+        echo "Downloading wabt for macOS-13..."
         sudo wget -O wabt.tar.gz     --progress=dot:giga https://github.com/WebAssembly/wabt/releases/download/1.0.36/wabt-1.0.36-macos-12.tar.gz
+
+        echo "Extracting wabt..."
         sudo tar -xf wabt.tar.gz
         sudo ln -sf  wabt-1.0.36 wabt
+
         /opt/wasi-sdk/bin/clang --version
         /opt/wabt/bin/wasm-interp --version
+
         echo "::notice::wasi-sdk-25 and wabt-1.0.36 installed on macos-13"
       working-directory: /opt
 
@@ -60,21 +78,48 @@ runs:
       if: ${{ inputs.os == 'macos-14' }}
       shell: bash
       run: |
+        echo "Downloading wasi-sdk for macOS-14..."
         sudo wget -O wasi-sdk.tar.gz --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-arm64-macos.tar.gz
+
+        echo "Extracting wasi-sdk..."
         sudo tar -xf wasi-sdk.tar.gz
         sudo ln -sf  wasi-sdk-25.0-arm64-macos wasi-sdk
+
+        echo "Downloading wabt for macOS-14..."
         sudo wget -O wabt.tar.gz     --progress=dot:giga https://github.com/WebAssembly/wabt/releases/download/1.0.37/wabt-1.0.37-macos-14.tar.gz
+
+        echo "Extracting wabt..."
         sudo tar -xf wabt.tar.gz
         sudo ln -sf  wabt-1.0.37 wabt
+
         /opt/wasi-sdk/bin/clang --version
         /opt/wabt/bin/wasm-interp --version
+
         echo "::notice::wasi-sdk-25 and wabt-1.0.37 installed on macos-14"
       working-directory: /opt
 
-    #TODO: Add support for Windows
     - name: Set up wasi-sdk and wabt on Windows
       if: ${{ startsWith(inputs.os, 'windows') }}
-      shell: powershell
+      shell: bash
       run: |
-        echo "::notice::Support for Windows is not implemented yet"
-        exit 1
+        choco install -y wget
+
+        mkdir -p /opt/wasi-sdk
+        mkdir -p /opt/wabt
+
+        echo "Downloading wasi-sdk for Windows..."
+        wget -O wasi-sdk.tar.gz --progress=dot:giga https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-x86_64-windows.tar.gz
+
+        echo "Extracting wasi-sdk..."
+        tar --strip-components=1 -xf wasi-sdk.tar.gz -C /opt/wasi-sdk
+
+        echo "Downloading wabt for Windows..."
+        wget -O wabt.tar.gz --progress=dot:giga https://github.com/WebAssembly/wabt/releases/download/1.0.37/wabt-1.0.37-windows.tar.gz
+
+        echo "Extracting wabt..."
+        tar --strip-components=1 -xf wabt.tar.gz -C /opt/wabt
+
+        /opt/wasi-sdk/bin/clang --version
+        /opt/wabt/bin/wasm-interp --version
+
+        echo "::notice::wasi-sdk-25 and wabt-1.0.37 installed on Windows"
diff --git a/.github/workflows/build_iwasm_release.yml b/.github/workflows/build_iwasm_release.yml
index a975d5807..6acd90fa6 100644
--- a/.github/workflows/build_iwasm_release.yml
+++ b/.github/workflows/build_iwasm_release.yml
@@ -23,7 +23,7 @@ on:
         type: string
         required: true
       upload_url:
-        description: a semantic version number. it is required when `release` is true.
+        description: upload binary assets to the URL of release
         type: string
         required: false
       ver_num:
diff --git a/.github/workflows/build_wamrc.yml b/.github/workflows/build_wamrc.yml
index 55d63f13b..07e3c7cdb 100644
--- a/.github/workflows/build_wamrc.yml
+++ b/.github/workflows/build_wamrc.yml
@@ -23,7 +23,7 @@ on:
         type: string
         required: true
       upload_url:
-        description: a semantic version number. it is required when `release` is true.
+        description: upload binary assets to the URL of release
         type: string
         required: false
       ver_num:
diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml
index 665c1588d..9c3309721 100644
--- a/.github/workflows/codeql.yml
+++ b/.github/workflows/codeql.yml
@@ -53,7 +53,7 @@ jobs:
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@v3.28.18
+      uses: github/codeql-action/init@v3.29.0
       with:
         languages: ${{ matrix.language }}
 
@@ -70,7 +70,7 @@ jobs:
     - run: |
         ./.github/scripts/codeql_buildscript.sh
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@v3.28.18
+      uses: github/codeql-action/analyze@v3.29.0
       with:
         category: "/language:${{matrix.language}}"
         upload: false
@@ -99,7 +99,7 @@ jobs:
         output: ${{ steps.step1.outputs.sarif-output }}/cpp.sarif
 
     - name: Upload CodeQL results to code scanning
-      uses: github/codeql-action/upload-sarif@v3.28.18
+      uses: github/codeql-action/upload-sarif@v3.29.0
       with:
         sarif_file: ${{ steps.step1.outputs.sarif-output }}
         category: "/language:${{matrix.language}}"
diff --git a/.github/workflows/compilation_on_android_ubuntu.yml b/.github/workflows/compilation_on_android_ubuntu.yml
index 65a540747..828773ae0 100644
--- a/.github/workflows/compilation_on_android_ubuntu.yml
+++ b/.github/workflows/compilation_on_android_ubuntu.yml
@@ -621,49 +621,6 @@ jobs:
             llvm_cache_key: ${{ needs.build_llvm_libraries_on_ubuntu_2204.outputs.cache_key }}
             running_mode: aot
             test_option: $WAMR_COMPILER_TEST_OPTIONS
-        exclude:
-          # incompatible modes and features
-          # classic-interp doesn't support simd
-          - running_mode: "classic-interp"
-            test_option: $SIMD_TEST_OPTIONS
-          # llvm jit doesn't support multi module
-          - running_mode: "jit"
-            test_option: $MULTI_MODULES_TEST_OPTIONS
-          # fast-jit doesn't support multi module, simd
-          - running_mode: "fast-jit"
-            test_option: $MULTI_MODULES_TEST_OPTIONS
-          - running_mode: "fast-jit"
-            test_option: $SIMD_TEST_OPTIONS
-          # multi-tier-jit doesn't support multi module, simd
-          - running_mode: "multi-tier-jit"
-            test_option: $MULTI_MODULES_TEST_OPTIONS
-          - running_mode: "multi-tier-jit"
-            test_option: $SIMD_TEST_OPTIONS
-          # fast-jit and multi-tier-jit don't support GC
-          - running_mode: "fast-jit"
-            test_option: $GC_TEST_OPTIONS
-          - running_mode: "multi-tier-jit"
-            test_option: $GC_TEST_OPTIONS
-          # fast-interp, fast-jit, llvm-jit, multi-tier-jit don't support Memory64
-          - running_mode: "fast-interp"
-            test_option: $MEMORY64_TEST_OPTIONS
-          - running_mode: "fast-jit"
-            test_option: $MEMORY64_TEST_OPTIONS
-          - running_mode: "jit"
-            test_option: $MEMORY64_TEST_OPTIONS
-          - running_mode: "multi-tier-jit"
-            test_option: $MEMORY64_TEST_OPTIONS
-          # aot, fast-interp, fast-jit, llvm-jit, multi-tier-jit don't support Multi Memory
-          - running_mode: "aot"
-            test_option: $MULTI_MEMORY_TEST_OPTIONS
-          - running_mode: "fast-interp"
-            test_option: $MULTI_MEMORY_TEST_OPTIONS
-          - running_mode: "fast-jit"
-            test_option: $MULTI_MEMORY_TEST_OPTIONS
-          - running_mode: "jit"
-            test_option: $MULTI_MEMORY_TEST_OPTIONS
-          - running_mode: "multi-tier-jit"
-            test_option: $MULTI_MEMORY_TEST_OPTIONS
 
     steps:
       - name: checkout
@@ -767,123 +724,3 @@ jobs:
           eval $(opam env)
           ./test_wamr.sh ${{ env.X86_32_TARGET_TEST_OPTIONS }} ${{ matrix.test_option }} -t ${{ matrix.running_mode }}
         working-directory: ./tests/wamr-test-suites
-
-  test-wamr-ide:
-    needs:
-      [
-        build_iwasm
-      ]
-    runs-on: ubuntu-22.04
-    env:
-      PYTHON_VERSION: '3.10'
-      PYTHON_UBUNTU_STANDALONE_BUILD: https://github.com/indygreg/python-build-standalone/releases/download/20230507/cpython-3.10.11+20230507-x86_64-unknown-linux-gnu-install_only.tar.gz
-
-    steps:
-      - name: checkout
-        uses: actions/checkout@v4
-
-      - name: install dependencies
-        run: |
-          rustup target add wasm32-wasip1
-          sudo apt update && sudo apt-get install -y lld ninja-build
-          npm install
-        working-directory: test-tools/wamr-ide/VSCode-Extension
-
-      - name: code style check
-        run: |
-          npm install --save-dev prettier
-          npm run prettier-format-check
-        working-directory: test-tools/wamr-ide/VSCode-Extension
-
-      - name: build iwasm with source debugging feature
-        run: |
-          mkdir build
-          cd build
-          cmake .. -DWAMR_BUILD_DEBUG_INTERP=1 -DWAMR_BUILD_REF_TYPES=1
-          make
-        working-directory: product-mini/platforms/linux
-
-      - name: Cache LLDB
-        id: cache-lldb
-        uses: actions/cache@v4
-        env:
-          cache-name: cache-lldb-vscode
-        with:
-          path: test-tools/wamr-ide/VSCode-Extension/resource/debug/linux
-          key: ${{ env.cache-name }}-${{ hashFiles('build-scripts/lldb_wasm.patch') }}-${{ env.PYTHON_UBUNTU_STANDALONE_BUILD }}
-
-      - if: ${{ steps.cache-lldb.outputs.cache-hit != 'true' }}
-        name: get stand-alone python ubuntu
-        run: |
-          wget ${{ env.PYTHON_UBUNTU_STANDALONE_BUILD }} -O python.tar.gz
-          tar -xvf python.tar.gz
-        working-directory: core/deps
-
-      - if: ${{ steps.cache-lldb.outputs.cache-hit != 'true' }}
-        name: download llvm
-        run: |
-          wget https://github.com/llvm/llvm-project/archive/1f27fe6128769f00197925c3b8f6abb9d0e5cd2e.zip
-          unzip -q 1f27fe6128769f00197925c3b8f6abb9d0e5cd2e.zip
-          mv llvm-project-1f27fe6128769f00197925c3b8f6abb9d0e5cd2e llvm-project
-        working-directory: core/deps
-
-      - if: ${{ steps.cache-lldb.outputs.cache-hit != 'true' }}
-        name: apply wamr patch
-        run: |
-          git init
-          git config user.email "action@github.com"
-          git config user.name "github action"
-          git apply ../../../build-scripts/lldb_wasm.patch
-        working-directory: core/deps/llvm-project
-
-      - if: ${{ steps.cache-lldb.outputs.cache-hit != 'true' }}
-        name: build lldb ubuntu
-        run: |
-          echo "start to build lldb..."
-          mkdir -p wamr-lldb
-          cmake -S ./llvm -B build \
-            -G Ninja \
-            -DCMAKE_INSTALL_PREFIX=../wamr-lldb \
-            -DCMAKE_BUILD_TYPE:STRING="Release" \
-            -DCMAKE_EXPORT_COMPILE_COMMANDS=ON \
-            -DLLVM_ENABLE_PROJECTS="clang;lldb" \
-            -DLLVM_TARGETS_TO_BUILD:STRING="X86;WebAssembly" \
-            -DLLVM_BUILD_BENCHMARKS:BOOL=OFF \
-            -DLLVM_BUILD_DOCS:BOOL=OFF \
-            -DLLVM_BUILD_EXAMPLES:BOOL=OFF \
-            -DLLVM_BUILD_LLVM_DYLIB:BOOL=OFF \
-            -DLLVM_BUILD_TESTS:BOOL=OFF \
-            -DLLVM_INCLUDE_BENCHMARKS:BOOL=OFF \
-            -DLLVM_INCLUDE_DOCS:BOOL=OFF \
-            -DLLVM_INCLUDE_EXAMPLES:BOOL=OFF \
-            -DLLVM_INCLUDE_TESTS:BOOL=OFF \
-            -DLLVM_ENABLE_BINDINGS:BOOL=OFF \
-            -DLLVM_ENABLE_LIBXML2:BOOL=ON \
-            -DLLVM_ENABLE_LLD:BOOL=ON \
-            -DLLDB_ENABLE_PYTHON:BOOL=ON \
-            -DLLDB_EMBED_PYTHON_HOME=ON \
-            -DLLDB_PYTHON_HOME=.. \
-            -DLLDB_PYTHON_RELATIVE_PATH=lib/lldb-python \
-            -DPython3_EXECUTABLE="$(pwd)/../python/bin/python${{ env.PYTHON_VERSION }}"
-          cmake --build build --target lldb install --parallel $(nproc)
-        working-directory: core/deps/llvm-project
-
-      - if: ${{ steps.cache-lldb.outputs.cache-hit != 'true' }}
-        name: copy lldb to extension folder
-        run: |
-          mkdir -p bin
-          mkdir -p lib
-          cp ../../../../../../core/deps/llvm-project/lldb/tools/lldb-vscode/package.json ./
-          cp -r ../../../../../../core/deps/llvm-project/lldb/tools/lldb-vscode/syntaxes/ ./
-          cp ../../../../../../core/deps/llvm-project/build/bin/lldb* bin
-          cp ../../../../../../core/deps/llvm-project/build/lib/liblldb*.so lib
-          cp ../../../../../../core/deps/llvm-project/build/lib/liblldb*.so.* lib
-          cp -R ../../../../../../core/deps/llvm-project/build/lib/lldb-python lib
-          cp -R ../../../../../../core/deps/python/lib/python* lib
-          cp ../../../../../../core/deps/python/lib/libpython${{ env.PYTHON_VERSION }}.so.1.0 lib
-        working-directory: test-tools/wamr-ide/VSCode-Extension/resource/debug/linux
-
-      - name: run tests
-        timeout-minutes: 5
-        run: xvfb-run npm run test
-        working-directory: test-tools/wamr-ide/VSCode-Extension
diff --git a/.github/workflows/compilation_on_sgx.yml b/.github/workflows/compilation_on_sgx.yml
index 7577b0bda..ec27fd8ba 100644
--- a/.github/workflows/compilation_on_sgx.yml
+++ b/.github/workflows/compilation_on_sgx.yml
@@ -291,28 +291,6 @@ jobs:
 
       - name: run spec tests
         run: |
-          set +e
           source /opt/intel/sgxsdk/environment
-          attempts=0
-          max_attempts=3
-
-          while [ $attempts -lt $max_attempts ]; do
-            ./test_wamr.sh ${{ matrix.test_option }} -t ${{ matrix.running_mode }}
-            exitcode="$?"
-
-            if [ $exitcode -eq 0 ]; then
-              echo "Spec test passed"
-              exit 0
-            elif [ $exitcode -ne 143 ]; then
-              echo "Spec test failed with error code $exitcode"
-              exit 1
-            fi
-
-            echo "$exitcode is a known GitHub-hosted runner issue"
-            echo "::notice::Re-running the spec test due to error code 143"
-            attempts=$((attempts + 1))
-          done
-
-          echo "::notice::Report an error with code 143 in SGX CI after $max_attempts attempts"
-          exit 143
+          ./test_wamr.sh ${{ matrix.test_option }} -t ${{ matrix.running_mode }}
         working-directory: ./tests/wamr-test-suites
diff --git a/.github/workflows/compilation_on_windows.yml b/.github/workflows/compilation_on_windows.yml
index 21f961cb9..7cee2aa40 100644
--- a/.github/workflows/compilation_on_windows.yml
+++ b/.github/workflows/compilation_on_windows.yml
@@ -172,6 +172,10 @@ jobs:
         run: ./build.sh
         working-directory: ./core/iwasm/libraries/lib-wasi-threads/test/
 
+      - name: install wget
+        shell: bash
+        run: choco install wget
+
       - name: run tests
         shell: bash
         timeout-minutes: 20
diff --git a/.github/workflows/nightly_run.yml b/.github/workflows/nightly_run.yml
index f261c880b..51a80705b 100644
--- a/.github/workflows/nightly_run.yml
+++ b/.github/workflows/nightly_run.yml
@@ -36,15 +36,14 @@ env:
   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"
   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"
   # For Spec Test
-  # FIXME: use binary release(adding -b) instead of building from source after upgrading to 22.04
-  DEFAULT_TEST_OPTIONS: "-s spec -P"
-  MULTI_MODULES_TEST_OPTIONS: "-s spec -M -P"
-  SIMD_TEST_OPTIONS: "-s spec -S -P"
-  EXTENDED_CONST_EXPR_TEST_OPTIONS: "-s spec -N -P"
-  THREADS_TEST_OPTIONS: "-s spec -p -P"
-  X86_32_TARGET_TEST_OPTIONS: "-m x86_32 -P"
+  DEFAULT_TEST_OPTIONS: "-s spec -b -P"
+  MULTI_MODULES_TEST_OPTIONS: "-s spec -b -P -M"
+  SIMD_TEST_OPTIONS: "-s spec -b -P -S"
+  THREADS_TEST_OPTIONS: "-s spec -b -P -p"
+  X86_32_TARGET_TEST_OPTIONS: "-m x86_32"
   WASI_TEST_OPTIONS: "-s wasi_certification -w"
-
+  EXTENDED_CONST_EXPR_TEST_OPTIONS: "-s spec -N -P"
+  
 permissions:
   contents: read
 
@@ -620,24 +619,11 @@ jobs:
             sanitizer: tsan
           - running_mode: "multi-tier-jit"
             sanitizer: tsan
-          # classic-interp and fast-interp don't support simd
-          - running_mode: "classic-interp"
-            test_option: $SIMD_TEST_OPTIONS
+          # simd128.h brings ubsan errors
+          # like: negation of XXXcannot be represented in type 'long int';
+          #       cast to an unsigned type to negate this value to itself
           - running_mode: "fast-interp"
-            test_option: $SIMD_TEST_OPTIONS
-          # llvm jit doesn't support multi module
-          - running_mode: "jit"
-            test_option: $MULTI_MODULES_TEST_OPTIONS
-          # fast-jit doesn't support multi module, simd
-          - running_mode: "fast-jit"
-            test_option: $MULTI_MODULES_TEST_OPTIONS
-          - running_mode: "fast-jit"
-            test_option: $SIMD_TEST_OPTIONS
-          # multi-tier-jit doesn't support multi module, simd
-          - running_mode: "multi-tier-jit"
-            test_option: $MULTI_MODULES_TEST_OPTIONS
-          - running_mode: "multi-tier-jit"
-            test_option: $SIMD_TEST_OPTIONS
+            sanitizer: ubsan
     steps:
       - name: checkout
         uses: actions/checkout@v4
diff --git a/.github/workflows/supply_chain.yml b/.github/workflows/supply_chain.yml
index d9e10a71c..827b300f1 100644
--- a/.github/workflows/supply_chain.yml
+++ b/.github/workflows/supply_chain.yml
@@ -60,6 +60,6 @@ jobs:
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@57eebf61a2246ab60a0c2f5a85766db783ad3553 
+        uses: github/codeql-action/upload-sarif@2847b7f7ab9f48fc49eca90a53fff6007285f399 
         with:
           sarif_file: results.sarif
diff --git a/.github/workflows/wamr_wasi_extensions.yml b/.github/workflows/wamr_wasi_extensions.yml
new file mode 100644
index 000000000..e9d10fe93
--- /dev/null
+++ b/.github/workflows/wamr_wasi_extensions.yml
@@ -0,0 +1,57 @@
+# Copyright (C) 2019 Intel Corporation.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+name: wamr_wasi_extensions
+
+on:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+    paths:
+      - ".github/workflows/wamr_wasi_extensions.yml"
+      - "wamr_wasi_extensios/**"
+      - "core/iwasm/libraries/wasi-nn/include/**"
+      - "core/iwasm/libraries/lib-socket/**"
+  # allow to be triggered manually
+  workflow_dispatch:
+
+# Cancel any in-flight jobs for the same PR/branch so there's only one active
+# at a time
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build_wamr_wasi_extensions:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [ubuntu-22.04, macos-13, macos-14]
+    steps:
+      - name: checkout
+        uses: actions/checkout@v4
+
+      - name: install-wasi-sdk-wabt
+        uses: ./.github/actions/install-wasi-sdk-wabt
+        with:
+          os: ${{ matrix.os }}
+
+      - name: Build wamr-wasi-extensions
+        run: |
+          mkdir dist
+          ./build_libs.sh $(pwd)/dist/wamr-wasi-extensions
+        working-directory: wamr-wasi-extensions
+
+      - name: Build wamr-wasi-extensions samples
+        run: |
+          ./build_samples.sh $(pwd)/dist/wamr-wasi-extensions
+        working-directory: wamr-wasi-extensions
+
+      - name: Upload artifacts
+        if: matrix.os == 'macos-14'
+        uses: actions/upload-artifact@v4
+        with:
+          name: wamr-wasi-extensions
+          path: wamr-wasi-extensions/dist
+          retention-days: 10
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 88a1642b8..4b28fa89c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -99,9 +99,9 @@ if (NOT DEFINED WAMR_BUILD_LIB_WASI_THREADS)
   set (WAMR_BUILD_LIB_WASI_THREADS 0)
 endif ()
 
-if (NOT DEFINED WAMR_ENABLE_COPY_CALLSTACK)
+if (NOT DEFINED WAMR_BUILD_COPY_CALL_STACK)
   # Disable copy callstack by default
-  set (WAMR_ENABLE_COPY_CALLSTACK 0)
+  set (WAMR_BUILD_COPY_CALL_STACK 0)
 endif()
 
 if (NOT DEFINED WAMR_BUILD_MINI_LOADER)
diff --git a/RELEASE_NOTES.md b/RELEASE_NOTES.md
index 8b3cfec28..167da4703 100644
--- a/RELEASE_NOTES.md
+++ b/RELEASE_NOTES.md
@@ -497,7 +497,7 @@
 - wasm loader: Fix handling if block without op else (#3404)
 - ref-types: Correct default value for function local variables (#3397)
 - aot compiler: Fix the length type passed to aot_memmove/aot_memset (#3378)
-- Fix loader and mini-loader select potiential error (#3374)
+- Fix loader and mini-loader select potential error (#3374)
 - Fix aot debugger compilation error on windows (#3370)
 - A few native stack detection fixes for macOS/arm64 (#3368)
 - Fix ESP32-S3 compiling error (#3359)
diff --git a/build-scripts/config_common.cmake b/build-scripts/config_common.cmake
index 916499355..d7a211a13 100644
--- a/build-scripts/config_common.cmake
+++ b/build-scripts/config_common.cmake
@@ -338,15 +338,10 @@ if (WAMR_BUILD_SHARED_HEAP EQUAL 1)
   add_definitions (-DWASM_ENABLE_SHARED_HEAP=1)
   message ("     Shared heap enabled")
 endif()
-
-if (WAMR_ENABLE_COPY_CALLSTACK EQUAL 1)
-  add_definitions (-DWAMR_ENABLE_COPY_CALLSTACK=1)
+if (WAMR_BUILD_COPY_CALL_STACK EQUAL 1)
+  add_definitions (-DWASM_ENABLE_COPY_CALL_STACK=1)
   message("     Copy callstack enabled")
-else ()
-  add_definitions (-DWAMR_ENABLE_COPY_CALLSTACK=0)
-  message("     Copy callstack disabled")
 endif()
-
 if (WAMR_BUILD_MEMORY64 EQUAL 1)
   # if native is 32-bit or cross-compiled to 32-bit
   if (NOT WAMR_BUILD_TARGET MATCHES ".*64.*")
@@ -543,6 +538,9 @@ if (WAMR_BUILD_WASI_NN EQUAL 1)
   if (DEFINED WAMR_BUILD_WASI_NN_EXTERNAL_DELEGATE_PATH)
       add_definitions (-DWASM_WASI_NN_EXTERNAL_DELEGATE_PATH="${WAMR_BUILD_WASI_NN_EXTERNAL_DELEGATE_PATH}")
   endif ()
+  if (NOT DEFINED WAMR_BUILD_WASI_EPHEMERAL_NN)
+      set(WAMR_BUILD_WASI_EPHEMERAL_NN 1)
+  endif()
   if (WAMR_BUILD_WASI_EPHEMERAL_NN EQUAL 1)
       message ("     WASI-NN: use 'wasi_ephemeral_nn' instead of 'wasi-nn'")
       add_definitions (-DWASM_ENABLE_WASI_EPHEMERAL_NN=1)
@@ -682,10 +680,8 @@ endif ()
 if (WAMR_BUILD_EXTENDED_CONST_EXPR EQUAL 1)
   message ("     Extended constant expression enabled")
   add_definitions(-DWASM_ENABLE_EXTENDED_CONST_EXPR=1)
-else()
-  message ("     Extended constant expression disabled")
-  add_definitions(-DWASM_ENABLE_EXTENDED_CONST_EXPR=0)
 endif ()
+
 ########################################
 # Show Phase4 Wasm proposals status.
 ########################################
diff --git a/build-scripts/requirements.txt b/build-scripts/requirements.txt
index ef487e06e..480d0c4bb 100644
--- a/build-scripts/requirements.txt
+++ b/build-scripts/requirements.txt
@@ -1 +1 @@
-requests==2.32.3
\ No newline at end of file
+requests==2.32.4
\ No newline at end of file
diff --git a/build-scripts/runtime_lib.cmake b/build-scripts/runtime_lib.cmake
index e538b5d91..f7639f6a6 100644
--- a/build-scripts/runtime_lib.cmake
+++ b/build-scripts/runtime_lib.cmake
@@ -106,6 +106,7 @@ endif ()
 
 if (WAMR_BUILD_WASI_NN EQUAL 1)
     include (${IWASM_DIR}/libraries/wasi-nn/cmake/wasi_nn.cmake)
+    set (WAMR_BUILD_MODULE_INST_CONTEXT 1)
 endif ()
 
 if (WAMR_BUILD_LIB_PTHREAD EQUAL 1)
diff --git a/ci/coding_guidelines_check.py b/ci/coding_guidelines_check.py
index 43c366259..131bca5b6 100644
--- a/ci/coding_guidelines_check.py
+++ b/ci/coding_guidelines_check.py
@@ -4,7 +4,6 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 #
 import argparse
-import re
 from pathlib import Path
 import re
 import shlex
@@ -39,7 +38,7 @@ INVALID_FILE_NAME_SEGMENT = r"([a-zA-Z0-9]+\-[a-zA-Z0-9]+)"
 
 def locate_command(command: str) -> bool:
     if not shutil.which(command):
-        print(f"Command '{command}'' not found")
+        print(f"Command '{command}' not found")
         return False
 
     return True
diff --git a/core/config.h b/core/config.h
index cf8acae65..38af3b029 100644
--- a/core/config.h
+++ b/core/config.h
@@ -193,8 +193,8 @@
 #error "Heap aux stack allocation must be enabled for WASI threads"
 #endif
 
-#ifndef WAMR_ENABLE_COPY_CALLSTACK
-#define WAMR_ENABLE_COPY_CALLSTACK 0
+#ifndef WASM_ENABLE_COPY_CALL_STACK
+#define WASM_ENABLE_COPY_CALL_STACK 0
 #endif
 
 #ifndef WASM_ENABLE_BASE_LIB
diff --git a/core/iwasm/aot/aot_loader.c b/core/iwasm/aot/aot_loader.c
index 9f387034a..358ec5d1d 100644
--- a/core/iwasm/aot/aot_loader.c
+++ b/core/iwasm/aot/aot_loader.c
@@ -1332,6 +1332,13 @@ load_init_expr(const uint8 **p_buf, const uint8 *buf_end, AOTModule *module,
             read_uint32(buf, buf_end, type_idx);
             read_uint32(buf, buf_end, length);
 
+            if (type_idx >= module->type_count
+                || !wasm_type_is_array_type(module->types[type_idx])) {
+                set_error_buf(error_buf, error_buf_size,
+                              "invalid or non-array type index.");
+                goto fail;
+            }
+
             if (init_expr_type == INIT_EXPR_TYPE_ARRAY_NEW_DEFAULT) {
                 expr->u.unary.v.array_new_default.type_index = type_idx;
                 expr->u.unary.v.array_new_default.length = length;
@@ -1778,6 +1785,12 @@ load_types(const uint8 **p_buf, const uint8 *buf_end, AOTModule *module,
             (void)u8;
 
             read_uint32(buf, buf_end, j);
+#if WASM_ENABLE_AOT_VALIDATOR != 0
+            if (j >= module->type_count) {
+                set_error_buf(error_buf, error_buf_size, "invalid type index");
+                goto fail;
+            }
+#endif
             if (module->types[j]->ref_count == UINT16_MAX) {
                 set_error_buf(error_buf, error_buf_size,
                               "wasm type's ref count too large");
@@ -2041,6 +2054,13 @@ load_types(const uint8 **p_buf, const uint8 *buf_end, AOTModule *module,
                 AOTType *cur_type = module->types[j];
                 parent_type_idx = cur_type->parent_type_idx;
                 if (parent_type_idx != (uint32)-1) { /* has parent */
+#if WASM_ENABLE_AOT_VALIDATOR != 0
+                    if (parent_type_idx >= module->type_count) {
+                        set_error_buf(error_buf, error_buf_size,
+                                      "invalid parent type index");
+                        goto fail;
+                    }
+#endif
                     AOTType *parent_type = module->types[parent_type_idx];
 
                     module->types[j]->parent_type = parent_type;
@@ -2064,6 +2084,13 @@ load_types(const uint8 **p_buf, const uint8 *buf_end, AOTModule *module,
                 AOTType *cur_type = module->types[j];
                 parent_type_idx = cur_type->parent_type_idx;
                 if (parent_type_idx != (uint32)-1) { /* has parent */
+#if WASM_ENABLE_AOT_VALIDATOR != 0
+                    if (parent_type_idx >= module->type_count) {
+                        set_error_buf(error_buf, error_buf_size,
+                                      "invalid parent type index");
+                        goto fail;
+                    }
+#endif
                     AOTType *parent_type = module->types[parent_type_idx];
                     /* subtyping has been checked during compilation */
                     bh_assert(wasm_type_is_subtype_of(
@@ -3378,7 +3405,7 @@ do_data_relocation(AOTModule *module, AOTRelocationGroup *group,
     uint8 *data_addr;
     uint32 data_size = 0, i;
     AOTRelocation *relocation = group->relocations;
-    void *symbol_addr;
+    void *symbol_addr = NULL;
     char *symbol, *data_section_name;
 
     if (!strncmp(group->section_name, ".rela.", 6)) {
diff --git a/core/iwasm/aot/aot_runtime.c b/core/iwasm/aot/aot_runtime.c
index a4268dd1a..4b94791ed 100644
--- a/core/iwasm/aot/aot_runtime.c
+++ b/core/iwasm/aot/aot_runtime.c
@@ -4243,9 +4243,9 @@ aot_frame_update_profile_info(WASMExecEnv *exec_env, bool alloc_frame)
 }
 #endif /* end of WASM_ENABLE_AOT_STACK_FRAME != 0 */
 
-#if WAMR_ENABLE_COPY_CALLSTACK != 0
+#if WASM_ENABLE_COPY_CALL_STACK != 0
 uint32
-aot_copy_callstack_tiny_frame(WASMExecEnv *exec_env, wasm_frame_t *buffer,
+aot_copy_callstack_tiny_frame(WASMExecEnv *exec_env, WASMCApiFrame *buffer,
                               const uint32 length, const uint32 skip_n,
                               char *error_buf, uint32 error_buf_size)
 {
@@ -4299,7 +4299,7 @@ aot_copy_callstack_tiny_frame(WASMExecEnv *exec_env, wasm_frame_t *buffer,
 }
 
 uint32
-aot_copy_callstack_standard_frame(WASMExecEnv *exec_env, wasm_frame_t *buffer,
+aot_copy_callstack_standard_frame(WASMExecEnv *exec_env, WASMCApiFrame *buffer,
                                   const uint32 length, const uint32 skip_n,
                                   char *error_buf, uint32_t error_buf_size)
 {
@@ -4349,7 +4349,7 @@ aot_copy_callstack_standard_frame(WASMExecEnv *exec_env, wasm_frame_t *buffer,
 }
 
 uint32
-aot_copy_callstack(WASMExecEnv *exec_env, wasm_frame_t *buffer,
+aot_copy_callstack(WASMExecEnv *exec_env, WASMCApiFrame *buffer,
                    const uint32 length, const uint32 skip_n, char *error_buf,
                    uint32_t error_buf_size)
 {
@@ -4371,7 +4371,7 @@ aot_copy_callstack(WASMExecEnv *exec_env, wasm_frame_t *buffer,
                                              error_buf, error_buf_size);
     }
 }
-#endif // WAMR_ENABLE_COPY_CALLSTACK
+#endif // WASM_ENABLE_COPY_CALL_STACK
 
 #if WASM_ENABLE_DUMP_CALL_STACK != 0
 bool
@@ -4983,8 +4983,8 @@ aot_dump_pgo_prof_data_to_buf(AOTModuleInstance *module_inst, char *buf,
     }
 
     prof_header.magic = 0xFF6C70726F667281LL;
-    /* Version 8 */
-    prof_header.version = 0x0000000000000008LL;
+    /* Version 9 */
+    prof_header.version = 0x0000000000000009LL;
     /* with VARIANT_MASK_IR_PROF (IR Instrumentation) */
     prof_header.version |= 0x1ULL << 56;
     /* with VARIANT_MASK_MEMPROF (Memory Profile) */
@@ -4993,14 +4993,19 @@ aot_dump_pgo_prof_data_to_buf(AOTModuleInstance *module_inst, char *buf,
     prof_header.num_prof_counters = num_prof_counters;
     prof_header.names_size = prof_names_size;
     prof_header.value_kind_last = 1;
+    /* __llvm_prf_bits won't be used in PGO, set dummy value here */
+    prof_header.num_prof_bitmaps = 0;
+    prof_header.bitmap_delta = 0;
 
     if (!is_little_endian()) {
         aot_exchange_uint64((uint8 *)&prof_header.magic);
         aot_exchange_uint64((uint8 *)&prof_header.version);
         aot_exchange_uint64((uint8 *)&prof_header.num_prof_data);
         aot_exchange_uint64((uint8 *)&prof_header.num_prof_counters);
+        aot_exchange_uint64((uint8 *)&prof_header.num_prof_bitmaps);
         aot_exchange_uint64((uint8 *)&prof_header.names_size);
         aot_exchange_uint64((uint8 *)&prof_header.counters_delta);
+        aot_exchange_uint64((uint8 *)&prof_header.bitmap_delta);
         aot_exchange_uint64((uint8 *)&prof_header.value_kind_last);
     }
 
@@ -5018,19 +5023,23 @@ aot_dump_pgo_prof_data_to_buf(AOTModuleInstance *module_inst, char *buf,
             prof_data_64->func_md5 = prof_data->func_md5;
             prof_data_64->func_hash = prof_data->func_hash;
             prof_data_64->offset_counters = prof_data->offset_counters;
+            prof_data_64->offset_bitmaps = prof_data->offset_bitmaps;
             prof_data_64->func_ptr = prof_data->func_ptr;
             prof_data_64->values = (uint64)(uintptr_t)prof_data->values;
             prof_data_64->num_counters = prof_data->num_counters;
+            /* __llvm_prf_bits won't be used in PGO, set dummy value here */
+            prof_data_64->num_bitmaps = 0;
             prof_data_64->num_value_sites[0] = prof_data->num_value_sites[0];
             prof_data_64->num_value_sites[1] = prof_data->num_value_sites[1];
 
             if (!is_little_endian()) {
                 aot_exchange_uint64((uint8 *)&prof_data_64->func_hash);
                 aot_exchange_uint64((uint8 *)&prof_data_64->offset_counters);
-                aot_exchange_uint64((uint8 *)&prof_data_64->offset_counters);
+                aot_exchange_uint64((uint8 *)&prof_data_64->offset_bitmaps);
                 aot_exchange_uint64((uint8 *)&prof_data_64->func_ptr);
                 aot_exchange_uint64((uint8 *)&prof_data_64->values);
                 aot_exchange_uint32((uint8 *)&prof_data_64->num_counters);
+                aot_exchange_uint32((uint8 *)&prof_data_64->num_bitmaps);
                 aot_exchange_uint16((uint8 *)&prof_data_64->num_value_sites[0]);
                 aot_exchange_uint16((uint8 *)&prof_data_64->num_value_sites[1]);
             }
diff --git a/core/iwasm/aot/aot_runtime.h b/core/iwasm/aot/aot_runtime.h
index 5be51c05a..e957f3959 100644
--- a/core/iwasm/aot/aot_runtime.h
+++ b/core/iwasm/aot/aot_runtime.h
@@ -437,6 +437,9 @@ typedef struct AOTFrame {
 } AOTFrame;
 
 #if WASM_ENABLE_STATIC_PGO != 0
+/* The bitmaps fields in LLVMProfileRawHeader, LLVMProfileData,
+ * LLVMProfileData_64 all dummy fields, it's used in MC/DC code coverage
+ * instead of PGO. See https://llvm.org/docs/InstrProfileFormat.html#bitmap */
 typedef struct LLVMProfileRawHeader {
     uint64 magic;
     uint64 version;
@@ -445,8 +448,11 @@ typedef struct LLVMProfileRawHeader {
     uint64 padding_bytes_before_counters;
     uint64 num_prof_counters;
     uint64 padding_bytes_after_counters;
+    uint64 num_prof_bitmaps;
+    uint64 padding_bytes_after_bitmaps;
     uint64 names_size;
     uint64 counters_delta;
+    uint64 bitmap_delta;
     uint64 names_delta;
     uint64 value_kind_last;
 } LLVMProfileRawHeader;
@@ -464,10 +470,12 @@ typedef struct LLVMProfileData {
     uint64 func_md5;
     uint64 func_hash;
     uint64 offset_counters;
+    uint64 offset_bitmaps;
     uintptr_t func_ptr;
     ValueProfNode **values;
     uint32 num_counters;
     uint16 num_value_sites[2];
+    uint32 num_bitmaps;
 } LLVMProfileData;
 
 /* The profiling data for writing to the output file, the width of
@@ -477,10 +485,12 @@ typedef struct LLVMProfileData_64 {
     uint64 func_md5;
     uint64 func_hash;
     uint64 offset_counters;
+    uint64 offset_bitmaps;
     uint64 func_ptr;
     uint64 values;
     uint32 num_counters;
     uint16 num_value_sites[2];
+    uint32 num_bitmaps;
 } LLVMProfileData_64;
 #endif /* end of WASM_ENABLE_STATIC_PGO != 0 */
 
@@ -777,12 +787,12 @@ aot_frame_update_profile_info(WASMExecEnv *exec_env, bool alloc_frame);
 bool
 aot_create_call_stack(struct WASMExecEnv *exec_env);
 
-#if WAMR_ENABLE_COPY_CALLSTACK != 0
+#if WASM_ENABLE_COPY_CALL_STACK != 0
 uint32
-aot_copy_callstack(WASMExecEnv *exec_env, wasm_frame_t *buffer,
+aot_copy_callstack(WASMExecEnv *exec_env, WASMCApiFrame *buffer,
                    const uint32 length, const uint32 skip_n, char *error_buf,
                    uint32_t error_buf_size);
-#endif // WAMR_ENABLE_COPY_CALLSTACK
+#endif // WASM_ENABLE_COPY_CALL_STACK
 
 /**
  * @brief Dump wasm call stack or get the size
diff --git a/core/iwasm/common/gc/gc_type.c b/core/iwasm/common/gc/gc_type.c
index bafa3c86c..8ae12f642 100644
--- a/core/iwasm/common/gc/gc_type.c
+++ b/core/iwasm/common/gc/gc_type.c
@@ -1145,7 +1145,7 @@ wasm_reftype_is_subtype_of(uint8 type1, const WASMRefType *ref_type1,
                 return true;
             else {
                 int32 heap_type = ref_type1->ref_ht_common.heap_type;
-                // We dont care whether type2 is nullable or not. So
+                // We don't care whether type2 is nullable or not. So
                 // we normalize it into its related one-byte type.
                 if (type2 == REF_TYPE_HT_NULLABLE
                     || type2 == REF_TYPE_HT_NON_NULLABLE) {
diff --git a/core/iwasm/common/wasm_runtime_common.c b/core/iwasm/common/wasm_runtime_common.c
index dcee0aeaf..e1d3542b9 100644
--- a/core/iwasm/common/wasm_runtime_common.c
+++ b/core/iwasm/common/wasm_runtime_common.c
@@ -1743,9 +1743,9 @@ wasm_runtime_destroy_exec_env(WASMExecEnv *exec_env)
     wasm_exec_env_destroy(exec_env);
 }
 
-#if WAMR_ENABLE_COPY_CALLSTACK != 0
+#if WASM_ENABLE_COPY_CALL_STACK != 0
 uint32
-wasm_copy_callstack(const wasm_exec_env_t exec_env, wasm_frame_t *buffer,
+wasm_copy_callstack(const wasm_exec_env_t exec_env, WASMCApiFrame *buffer,
                     const uint32 length, const uint32 skip_n, char *error_buf,
                     uint32_t error_buf_size)
 {
@@ -1780,7 +1780,7 @@ wasm_copy_callstack(const wasm_exec_env_t exec_env, wasm_frame_t *buffer,
     strncpy(error_buf, err_msg, error_buf_size);
     return 0;
 }
-#endif // WAMR_ENABLE_COPY_CALLSTACK
+#endif // WASM_ENABLE_COPY_CALL_STACK
 
 bool
 wasm_runtime_init_thread_env(void)
diff --git a/core/iwasm/common/wasm_runtime_common.h b/core/iwasm/common/wasm_runtime_common.h
index 64a6cd793..ad22ea10b 100644
--- a/core/iwasm/common/wasm_runtime_common.h
+++ b/core/iwasm/common/wasm_runtime_common.h
@@ -758,12 +758,12 @@ wasm_runtime_create_exec_env(WASMModuleInstanceCommon *module_inst,
 WASM_RUNTIME_API_EXTERN void
 wasm_runtime_destroy_exec_env(WASMExecEnv *exec_env);
 
-#if WAMR_ENABLE_COPY_CALLSTACK != 0
+#if WASM_ENABLE_COPY_CALL_STACK != 0
 WASM_RUNTIME_API_EXTERN uint32_t
-wasm_copy_callstack(const wasm_exec_env_t exec_env, wasm_frame_t *buffer,
+wasm_copy_callstack(const wasm_exec_env_t exec_env, WASMCApiFrame *buffer,
                     const uint32 length, const uint32 skip_n, char *error_buf,
                     uint32 error_buf_size);
-#endif // WAMR_ENABLE_COPY_CALLSTACK
+#endif // WASM_ENABLE_COPY_CALL_STACK
 
 /* See wasm_export.h for description */
 WASM_RUNTIME_API_EXTERN WASMModuleInstanceCommon *
diff --git a/core/iwasm/compilation/aot_emit_aot_file.c b/core/iwasm/compilation/aot_emit_aot_file.c
index 208498fcb..fec1c34cb 100644
--- a/core/iwasm/compilation/aot_emit_aot_file.c
+++ b/core/iwasm/compilation/aot_emit_aot_file.c
@@ -3419,6 +3419,12 @@ aot_resolve_object_data_sections(AOTObjectData *obj_data)
                     bh_memcpy_s(data_section->name, size, buf, size);
                     data_section->is_name_allocated = true;
                 }
+                else if (obj_data->comp_ctx->enable_llvm_pgo
+                         && !strcmp(name, "__llvm_prf_bits")) {
+                    LOG_WARNING("__llvm_prf_bits section is not supported and "
+                                "shouldn't be used in PGO.");
+                    return false;
+                }
 
                 if (obj_data->comp_ctx->enable_llvm_pgo
                     && !strcmp(name, "__llvm_prf_names")) {
diff --git a/core/iwasm/compilation/aot_llvm.c b/core/iwasm/compilation/aot_llvm.c
index 4f6878148..a36ffb7ee 100644
--- a/core/iwasm/compilation/aot_llvm.c
+++ b/core/iwasm/compilation/aot_llvm.c
@@ -3207,6 +3207,21 @@ aot_create_comp_context(const AOTCompData *comp_data, aot_comp_option_t option)
 
 #if WASM_ENABLE_WAMR_COMPILER != 0
     WASMModule *wasm_module = (WASMModule *)comp_data->wasm_module;
+    bool is_memory64 = false;
+
+    /* TODO: multi-memories for now assuming the memory64 flag of a memory is
+     * consistent across multi-memories */
+    if (wasm_module->import_memory_count > 0)
+        is_memory64 = !!(wasm_module->import_memories[0].u.memory.mem_type.flags
+                         & MEMORY64_FLAG);
+    else if (wasm_module->memory_count > 0)
+        is_memory64 = !!(wasm_module->memories[0].flags & MEMORY64_FLAG);
+
+    if (!(option->bounds_checks == 1 || option->bounds_checks == 0)
+        && is_memory64) {
+        /* For memory64, the boundary check default value is true */
+        comp_ctx->enable_bound_check = true;
+    }
 
     /* Return error if SIMD is disabled by command line but SIMD instructions
      * are used */
@@ -3987,7 +4002,7 @@ aot_get_func_from_table(const AOTCompContext *comp_ctx, LLVMValueRef base,
 
     if (!(func =
               LLVMBuildBitCast(comp_ctx->builder, func, func_type, "func"))) {
-        aot_set_last_error("cast function fialed.");
+        aot_set_last_error("cast function failed.");
         goto fail;
     }
 
@@ -4056,7 +4071,7 @@ aot_load_const_from_table(AOTCompContext *comp_ctx, LLVMValueRef base,
 
     if (!(const_addr = LLVMBuildBitCast(comp_ctx->builder, const_addr,
                                         const_ptr_type, "const_addr"))) {
-        aot_set_last_error("cast const fialed.");
+        aot_set_last_error("cast const failed.");
         return NULL;
     }
 
diff --git a/core/iwasm/include/wasm_export.h b/core/iwasm/include/wasm_export.h
index b4ab34bea..ca96f5824 100644
--- a/core/iwasm/include/wasm_export.h
+++ b/core/iwasm/include/wasm_export.h
@@ -139,8 +139,6 @@ typedef struct wasm_frame_t {
     uint32_t *lp;
 } WASMCApiFrame;
 
-typedef WASMCApiFrame wasm_frame_t;
-
 /* WASM section */
 typedef struct wasm_section_t {
     struct wasm_section_t *next;
@@ -904,7 +902,7 @@ wasm_runtime_destroy_exec_env(wasm_exec_env_t exec_env);
  * @return number of copied frames
  */
 WASM_RUNTIME_API_EXTERN uint32_t
-wasm_copy_callstack(const wasm_exec_env_t exec_env, wasm_frame_t *buffer,
+wasm_copy_callstack(const wasm_exec_env_t exec_env, WASMCApiFrame *buffer,
                     const uint32_t length, const uint32_t skip_n,
                     char *error_buf, uint32_t error_buf_size);
 
diff --git a/core/iwasm/interpreter/wasm_interp_classic.c b/core/iwasm/interpreter/wasm_interp_classic.c
index 17e1085c2..f7de04806 100644
--- a/core/iwasm/interpreter/wasm_interp_classic.c
+++ b/core/iwasm/interpreter/wasm_interp_classic.c
@@ -4088,7 +4088,7 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
                     case WASM_OP_STRING_ENCODE_LOSSY_UTF8_ARRAY:
                     case WASM_OP_STRING_ENCODE_WTF8_ARRAY:
                     {
-                        uint32 start, array_len, count;
+                        uint32 start, array_len;
                         int32 bytes_written;
                         EncodingFlag flag = WTF8;
                         WASMArrayType *array_type;
diff --git a/core/iwasm/interpreter/wasm_loader.c b/core/iwasm/interpreter/wasm_loader.c
index 2c6234d89..12e68c06e 100644
--- a/core/iwasm/interpreter/wasm_loader.c
+++ b/core/iwasm/interpreter/wasm_loader.c
@@ -1001,10 +1001,10 @@ load_init_expr(WASMModule *module, const uint8 **p_buf, const uint8 *buf_end,
             {
                 uint8 type1;
 
+#if WASM_ENABLE_GC == 0
                 CHECK_BUF(p, p_end, 1);
                 type1 = read_uint8(p);
 
-#if WASM_ENABLE_GC == 0
                 cur_value.ref_index = NULL_REF;
                 if (!push_const_expr_stack(&const_expr_ctx, flag, type1,
                                            &cur_value,
@@ -1014,9 +1014,14 @@ load_init_expr(WASMModule *module, const uint8 **p_buf, const uint8 *buf_end,
                                            error_buf, error_buf_size))
                     goto fail;
 #else
+                int32 heap_type;
+                read_leb_int32(p, p_end, heap_type);
+                type1 = (uint8)((int32)0x80 + heap_type);
+
                 cur_value.gc_obj = NULL_REF;
 
                 if (!is_byte_a_type(type1)
+                    || !wasm_is_valid_heap_type(heap_type)
                     || wasm_is_type_multi_byte_type(type1)) {
                     p--;
                     read_leb_uint32(p, p_end, type_idx);
@@ -2267,9 +2272,9 @@ load_type_section(const uint8 *buf, const uint8 *buf_end, WASMModule *module,
                                       "recursive type count too large");
                         return false;
                     }
-                    module->type_count += rec_count - 1;
                     new_total_size =
-                        sizeof(WASMFuncType *) * (uint64)module->type_count;
+                        sizeof(WASMFuncType *)
+                        * (uint64)(module->type_count + rec_count - 1);
                     if (new_total_size > UINT32_MAX) {
                         set_error_buf(error_buf, error_buf_size,
                                       "allocate memory failed");
@@ -2277,6 +2282,7 @@ load_type_section(const uint8 *buf, const uint8 *buf_end, WASMModule *module,
                     }
                     MEM_REALLOC(module->types, (uint32)total_size,
                                 (uint32)new_total_size);
+                    module->type_count += rec_count - 1;
                     total_size = new_total_size;
                 }
 
@@ -2813,7 +2819,8 @@ load_table_import(const uint8 **p_buf, const uint8 *buf_end,
                             error_buf_size)) {
         return false;
     }
-    if (wasm_is_reftype_htref_non_nullable(ref_type.ref_type)) {
+    if (!wasm_is_type_reftype(ref_type.ref_type)
+        || wasm_is_reftype_htref_non_nullable(ref_type.ref_type)) {
         set_error_buf(error_buf, error_buf_size, "type mismatch");
         return false;
     }
@@ -3339,6 +3346,15 @@ load_table(const uint8 **p_buf, const uint8 *buf_end, WASMModule *module,
                             error_buf_size)) {
         return false;
     }
+    /*
+     * TODO: add this validator
+     *   `wasm_is_reftype_htref_non_nullable(ref_type.ref_type)`
+     * after sync up with the latest GC spec
+     */
+    if (!wasm_is_type_reftype(ref_type.ref_type)) {
+        set_error_buf(error_buf, error_buf_size, "type mismatch");
+        return false;
+    }
     table->table_type.elem_type = ref_type.ref_type;
     if (need_ref_type_map) {
         if (!(table->table_type.elem_ref_type =
@@ -3566,7 +3582,8 @@ load_import_section(const uint8 *buf, const uint8 *buf_end, WASMModule *module,
                     /* valtype */
                     CHECK_BUF(p, p_end, 1);
                     global_type = read_uint8(p);
-                    if (wasm_is_reftype_htref_nullable(global_type)) {
+                    if (wasm_is_reftype_htref_nullable(global_type)
+                        || wasm_is_reftype_htref_non_nullable(global_type)) {
                         int32 heap_type;
                         read_leb_int32(p, p_end, heap_type);
                         (void)heap_type;
@@ -3927,7 +3944,7 @@ load_function_section(const uint8 *buf, const uint8 *buf_end,
              * we shall make a copy of code body [p_code, p_code + code_size]
              * when we are worrying about inappropriate releasing behaviour.
              * all code bodies are actually in a buffer which user allocates in
-             * his embedding environment and we don't have power on them.
+             * their embedding environment and we don't have power over them.
              * it will be like:
              * code_body_cp = malloc(code_size);
              * memcpy(code_body_cp, p_code, code_size);
@@ -15253,8 +15270,6 @@ re_scan:
                     case WASM_OP_STRING_NEW_LOSSY_UTF8:
                     case WASM_OP_STRING_NEW_WTF8:
                     {
-                        uint32 memidx;
-
 #if WASM_ENABLE_JIT != 0 || WASM_ENABLE_WAMR_COMPILER != 0
                         func->has_memory_operations = true;
 #endif
@@ -15266,7 +15281,6 @@ re_scan:
                         POP_I32();
                         POP_I32();
                         PUSH_REF(REF_TYPE_STRINGREF);
-                        (void)memidx;
                         break;
                     }
                     case WASM_OP_STRING_CONST:
@@ -15294,8 +15308,6 @@ re_scan:
                     case WASM_OP_STRING_ENCODE_LOSSY_UTF8:
                     case WASM_OP_STRING_ENCODE_WTF8:
                     {
-                        uint32 memidx;
-
 #if WASM_ENABLE_JIT != 0 || WASM_ENABLE_WAMR_COMPILER != 0
                         func->has_memory_operations = true;
 #endif
@@ -15307,7 +15319,6 @@ re_scan:
                         POP_I32();
                         POP_STRINGREF();
                         PUSH_I32();
-                        (void)memidx;
                         break;
                     }
                     case WASM_OP_STRING_CONCAT:
@@ -15348,8 +15359,6 @@ re_scan:
                     case WASM_OP_STRINGVIEW_WTF8_ENCODE_LOSSY_UTF8:
                     case WASM_OP_STRINGVIEW_WTF8_ENCODE_WTF8:
                     {
-                        uint32 memidx;
-
 #if WASM_ENABLE_JIT != 0 || WASM_ENABLE_WAMR_COMPILER != 0
                         func->has_memory_operations = true;
 #endif
@@ -15364,7 +15373,6 @@ re_scan:
                         POP_REF(REF_TYPE_STRINGVIEWWTF8);
                         PUSH_I32();
                         PUSH_I32();
-                        (void)memidx;
                         break;
                     }
                     case WASM_OP_STRINGVIEW_WTF8_SLICE:
@@ -15396,8 +15404,6 @@ re_scan:
                     }
                     case WASM_OP_STRINGVIEW_WTF16_ENCODE:
                     {
-                        uint32 memidx;
-
 #if WASM_ENABLE_JIT != 0 || WASM_ENABLE_WAMR_COMPILER != 0
                         func->has_memory_operations = true;
 #endif
@@ -15411,7 +15417,6 @@ re_scan:
                         POP_I32();
                         POP_REF(REF_TYPE_STRINGVIEWWTF16);
                         PUSH_I32();
-                        (void)memidx;
                         break;
                     }
                     case WASM_OP_STRINGVIEW_WTF16_SLICE:
diff --git a/core/iwasm/interpreter/wasm_mini_loader.c b/core/iwasm/interpreter/wasm_mini_loader.c
index 094ba4d1f..771538a14 100644
--- a/core/iwasm/interpreter/wasm_mini_loader.c
+++ b/core/iwasm/interpreter/wasm_mini_loader.c
@@ -1395,7 +1395,7 @@ load_function_section(const uint8 *buf, const uint8 *buf_end,
              * we shall make a copy of code body [p_code, p_code + code_size]
              * when we are worrying about inappropriate releasing behaviour.
              * all code bodies are actually in a buffer which user allocates in
-             * his embedding environment and we don't have power on them.
+             * their embedding environment and we don't have power over them.
              * it will be like:
              * code_body_cp = malloc(code_size);
              * memcpy(code_body_cp, p_code, code_size);
diff --git a/core/iwasm/interpreter/wasm_runtime.c b/core/iwasm/interpreter/wasm_runtime.c
index 8045d35e3..2b7b11f90 100644
--- a/core/iwasm/interpreter/wasm_runtime.c
+++ b/core/iwasm/interpreter/wasm_runtime.c
@@ -2751,7 +2751,7 @@ wasm_instantiate(WASMModule *module, WASMModuleInstance *parent,
                         }
                         STORE_PTR((void **)global_data, func_obj);
                         global_data += sizeof(void *);
-                        /* Also update the inital_value since other globals may
+                        /* Also update the initial_value since other globals may
                          * refer to this */
                         global->initial_value.gc_obj = (wasm_obj_t)func_obj;
                         break;
@@ -4256,9 +4256,9 @@ wasm_get_module_inst_mem_consumption(const WASMModuleInstance *module_inst,
 #endif /* end of (WASM_ENABLE_MEMORY_PROFILING != 0) \
                  || (WASM_ENABLE_MEMORY_TRACING != 0) */
 
-#if WAMR_ENABLE_COPY_CALLSTACK != 0
+#if WASM_ENABLE_COPY_CALL_STACK != 0
 uint32
-wasm_interp_copy_callstack(WASMExecEnv *exec_env, wasm_frame_t *buffer,
+wasm_interp_copy_callstack(WASMExecEnv *exec_env, WASMCApiFrame *buffer,
                            uint32 length, uint32 skip_n, char *error_buf,
                            uint32_t error_buf_size)
 {
@@ -4303,7 +4303,7 @@ wasm_interp_copy_callstack(WASMExecEnv *exec_env, wasm_frame_t *buffer,
     }
     return count >= skip_n ? count - skip_n : 0;
 }
-#endif // WAMR_ENABLE_COPY_CALLSTACK
+#endif // WASM_ENABLE_COPY_CALL_STACK
 
 #if WASM_ENABLE_DUMP_CALL_STACK != 0
 bool
diff --git a/core/iwasm/interpreter/wasm_runtime.h b/core/iwasm/interpreter/wasm_runtime.h
index 8d38c8831..3d91d8b60 100644
--- a/core/iwasm/interpreter/wasm_runtime.h
+++ b/core/iwasm/interpreter/wasm_runtime.h
@@ -731,12 +731,12 @@ wasm_get_table_inst(const WASMModuleInstance *module_inst, uint32 tbl_idx)
 
 #if WASM_ENABLE_DUMP_CALL_STACK != 0
 
-#if WAMR_ENABLE_COPY_CALLSTACK != 0
+#if WASM_ENABLE_COPY_CALL_STACK != 0
 uint32
-wasm_interp_copy_callstack(WASMExecEnv *exec_env, wasm_frame_t *buffer,
+wasm_interp_copy_callstack(WASMExecEnv *exec_env, WASMCApiFrame *buffer,
                            uint32 length, uint32 skip_n, char *error_buf,
                            uint32_t error_buf_size);
-#endif // WAMR_ENABLE_COPY_CALLSTACK
+#endif // WASM_ENABLE_COPY_CALL_STACK
 
 bool
 wasm_interp_create_call_stack(struct WASMExecEnv *exec_env);
diff --git a/core/iwasm/libraries/debug-engine/debug_engine.c b/core/iwasm/libraries/debug-engine/debug_engine.c
index 340e657e8..24d57d706 100644
--- a/core/iwasm/libraries/debug-engine/debug_engine.c
+++ b/core/iwasm/libraries/debug-engine/debug_engine.c
@@ -743,7 +743,7 @@ wasm_debug_instance_get_obj_mem(WASMDebugInstance *instance, uint64 offset,
     module_inst = (WASMModuleInstance *)exec_env->module_inst;
 
     if (offset + *size > module_inst->module->load_size) {
-        LOG_VERBOSE("wasm_debug_instance_get_data_mem size over flow!\n");
+        LOG_VERBOSE("wasm_debug_instance_get_data_mem size overflow!\n");
         *size = module_inst->module->load_size >= offset
                     ? module_inst->module->load_size - offset
                     : 0;
@@ -797,7 +797,7 @@ wasm_debug_instance_get_linear_mem(WASMDebugInstance *instance, uint64 offset,
         num_bytes_per_page = memory->num_bytes_per_page;
         linear_mem_size = num_bytes_per_page * memory->cur_page_count;
         if (offset + *size > linear_mem_size) {
-            LOG_VERBOSE("wasm_debug_instance_get_linear_mem size over flow!\n");
+            LOG_VERBOSE("wasm_debug_instance_get_linear_mem size overflow!\n");
             *size = linear_mem_size >= offset ? linear_mem_size - offset : 0;
         }
         bh_memcpy_s(buf, (uint32)*size, memory->memory_data + offset,
@@ -830,7 +830,7 @@ wasm_debug_instance_set_linear_mem(WASMDebugInstance *instance, uint64 offset,
         num_bytes_per_page = memory->num_bytes_per_page;
         linear_mem_size = num_bytes_per_page * memory->cur_page_count;
         if (offset + *size > linear_mem_size) {
-            LOG_VERBOSE("wasm_debug_instance_get_linear_mem size over flow!\n");
+            LOG_VERBOSE("wasm_debug_instance_get_linear_mem size overflow!\n");
             *size = linear_mem_size >= offset ? linear_mem_size - offset : 0;
         }
         bh_memcpy_s(memory->memory_data + offset, (uint32)*size, buf,
diff --git a/core/iwasm/libraries/debug-engine/handler.c b/core/iwasm/libraries/debug-engine/handler.c
index 743165dd9..14c7fae6e 100644
--- a/core/iwasm/libraries/debug-engine/handler.c
+++ b/core/iwasm/libraries/debug-engine/handler.c
@@ -175,6 +175,19 @@ process_wasm_global(WASMGDBServer *server, char *args)
     os_mutex_unlock(&tmpbuf_lock);
 }
 
+/* TODO: let server send an empty/error reply.
+   Original issue: 4265
+   Not tested yet, but it should work.
+ */
+static void
+send_reply(WASMGDBServer *server, const char *err)
+{
+    if (!err || !*err)
+        write_packet(server, "");
+    else
+        write_packet(server, err);
+}
+
 void
 handle_general_query(WASMGDBServer *server, char *payload)
 {
@@ -214,6 +227,7 @@ handle_general_query(WASMGDBServer *server, char *payload)
 
         if (!args) {
             LOG_ERROR("payload parse error during handle_general_query");
+            send_reply(server, "");
             return;
         }
 
@@ -384,7 +398,7 @@ send_thread_stop_status(WASMGDBServer *server, uint32 status, korp_tid tid)
     if (status == 0) {
         os_mutex_lock(&tmpbuf_lock);
         (void)snprintf(tmpbuf, MAX_PACKET_SIZE, "W%02" PRIx32, status);
-        write_packet(server, tmpbuf);
+        send_reply(server, tmpbuf);
         os_mutex_unlock(&tmpbuf_lock);
         return;
     }
@@ -403,6 +417,7 @@ send_thread_stop_status(WASMGDBServer *server, uint32 status, korp_tid tid)
                    "T%02" PRIx32 "thread:%" PRIx64 ";name:%s;", gdb_status,
                    (uint64)(uintptr_t)tid, "nobody");
     if (len < 0 || len >= MAX_PACKET_SIZE) {
+        send_reply(server, "E01");
         os_mutex_unlock(&tmpbuf_lock);
         return;
     }
@@ -410,6 +425,7 @@ send_thread_stop_status(WASMGDBServer *server, uint32 status, korp_tid tid)
     if (tids_count > 0) {
         int n = snprintf(tmpbuf + len, MAX_PACKET_SIZE - len, "threads:");
         if (n < 0 || n >= MAX_PACKET_SIZE - len) {
+            send_reply(server, "E01");
             os_mutex_unlock(&tmpbuf_lock);
             return;
         }
@@ -426,6 +442,7 @@ send_thread_stop_status(WASMGDBServer *server, uint32 status, korp_tid tid)
             }
 
             if (n < 0 || n >= MAX_PACKET_SIZE - len) {
+                send_reply(server, "E01");
                 os_mutex_unlock(&tmpbuf_lock);
                 return;
             }
@@ -452,6 +469,7 @@ send_thread_stop_status(WASMGDBServer *server, uint32 status, korp_tid tid)
                      "thread-pcs:%" PRIx64 ";00:%s;reason:%s;description:", pc,
                      pc_string, "exception");
         if (n < 0 || n >= MAX_PACKET_SIZE - len) {
+            send_reply(server, "E01");
             os_mutex_unlock(&tmpbuf_lock);
             return;
         }
@@ -462,6 +480,7 @@ send_thread_stop_status(WASMGDBServer *server, uint32 status, korp_tid tid)
             n = snprintf(tmpbuf + len, MAX_PACKET_SIZE - len, "%02x",
                          exception[i]);
             if (n < 0 || n >= MAX_PACKET_SIZE - len) {
+                send_reply(server, "E01");
                 os_mutex_unlock(&tmpbuf_lock);
                 return;
             }
@@ -592,7 +611,7 @@ handle_get_register(WASMGDBServer *server, char *payload)
     int32 i = strtol(payload, NULL, 16);
 
     if (i != 0) {
-        write_packet(server, "E01");
+        send_reply(server, "E01");
         return;
     }
     regdata = wasm_debug_instance_get_pc(
@@ -748,7 +767,7 @@ handle_add_break(WASMGDBServer *server, char *payload)
     if ((arg_c = sscanf(payload, "%zx,%" SCNx64 ",%zx", &type, &addr, &length))
         != 3) {
         LOG_ERROR("Unsupported number of add break arguments %d", arg_c);
-        write_packet(server, "");
+        send_reply(server, "");
         return;
     }
 
@@ -783,7 +802,7 @@ handle_remove_break(WASMGDBServer *server, char *payload)
     if ((arg_c = sscanf(payload, "%zx,%" SCNx64 ",%zx", &type, &addr, &length))
         != 3) {
         LOG_ERROR("Unsupported number of remove break arguments %d", arg_c);
-        write_packet(server, "");
+        send_reply(server, "");
         return;
     }
 
@@ -835,6 +854,7 @@ handle_malloc(WASMGDBServer *server, char *payload)
     }
     else {
         LOG_ERROR("Payload parse error during handle malloc");
+        send_reply(server, "");
         return;
     }
 
diff --git a/core/iwasm/libraries/lib-socket/src/wasi/wasi_socket_ext.c b/core/iwasm/libraries/lib-socket/src/wasi/wasi_socket_ext.c
index 1172d0a77..f573d35b8 100644
--- a/core/iwasm/libraries/lib-socket/src/wasi/wasi_socket_ext.c
+++ b/core/iwasm/libraries/lib-socket/src/wasi/wasi_socket_ext.c
@@ -12,6 +12,26 @@
 #include <wasi/api.h>
 #include <wasi_socket_ext.h>
 
+/*
+ * Avoid direct TLS access to allow a single library to be
+ * linked to both of threaded and non-threaded applications.
+ *
+ * wasi-libc's errno is a TLS variable, exposed directly via
+ * errno.h. if we use it here, LLVM may lower it differently,
+ * depending on enabled features like atomcs and bulk-memory.
+ * we tweak the way to access errno here in order to make us
+ * compatible with both of threaded and non-threaded applications.
+ * __errno_location() should be reasonably stable because
+ * it was introduced as an alternative ABI for non-C software.
+ * https://github.com/WebAssembly/wasi-libc/pull/347
+ */
+#if defined(errno)
+#undef errno
+#endif
+int *
+__errno_location(void);
+#define errno (*__errno_location())
+
 #define HANDLE_ERROR(error)              \
     if (error != __WASI_ERRNO_SUCCESS) { \
         errno = error;                   \
diff --git a/core/iwasm/libraries/wasi-nn/include/wasi_ephemeral_nn.h b/core/iwasm/libraries/wasi-nn/include/wasi_ephemeral_nn.h
new file mode 100644
index 000000000..f76295a1e
--- /dev/null
+++ b/core/iwasm/libraries/wasi-nn/include/wasi_ephemeral_nn.h
@@ -0,0 +1,12 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#define WASM_ENABLE_WASI_EPHEMERAL_NN 1
+#define WASI_NN_NAME(name) wasi_ephemeral_nn_##name
+
+#include "wasi_nn.h"
+
+#undef WASM_ENABLE_WASI_EPHEMERAL_NN
+#undef WASI_NN_NAME
diff --git a/core/iwasm/libraries/wasi-nn/include/wasi_nn.h b/core/iwasm/libraries/wasi-nn/include/wasi_nn.h
index 35b2d9bf0..cda26324e 100644
--- a/core/iwasm/libraries/wasi-nn/include/wasi_nn.h
+++ b/core/iwasm/libraries/wasi-nn/include/wasi_nn.h
@@ -21,6 +21,7 @@
 #else
 #define WASI_NN_IMPORT(name) \
     __attribute__((import_module("wasi_nn"), import_name(name)))
+#warning You are using "wasi_nn", which is a legacy WAMR-specific ABI. It's deperecated and will likely be removed in future versions of WAMR. Please use "wasi_ephemeral_nn" instead. (For a WASM module, use the wasi_ephemeral_nn.h header instead. For the runtime configurations, enable WASM_ENABLE_WASI_EPHEMERAL_NN/WAMR_BUILD_WASI_EPHEMERAL_NN.)
 #endif
 
 /**
@@ -34,17 +35,22 @@
  * @return wasi_nn_error    Execution status.
  */
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
-wasi_nn_error
-load(graph_builder *builder, uint32_t builder_len, graph_encoding encoding,
-     execution_target target, graph *g) WASI_NN_IMPORT("load");
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(load)
+(WASI_NN_NAME(graph_builder) * builder, uint32_t builder_len,
+ WASI_NN_NAME(graph_encoding) encoding, WASI_NN_NAME(execution_target) target,
+ WASI_NN_NAME(graph) * g) WASI_NN_IMPORT("load");
 #else
-wasi_nn_error
-load(graph_builder_array *builder, graph_encoding encoding,
-     execution_target target, graph *g) WASI_NN_IMPORT("load");
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(load)
+(WASI_NN_NAME(graph_builder_array) * builder,
+ WASI_NN_NAME(graph_encoding) encoding, WASI_NN_NAME(execution_target) target,
+ WASI_NN_NAME(graph) * g) WASI_NN_IMPORT("load");
 #endif
 
-wasi_nn_error
-load_by_name(const char *name, uint32_t name_len, graph *g)
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(load_by_name)
+(const char *name, uint32_t name_len, WASI_NN_NAME(graph) * g)
     WASI_NN_IMPORT("load_by_name");
 
 /**
@@ -59,8 +65,9 @@ load_by_name(const char *name, uint32_t name_len, graph *g)
  * @param ctx       Execution context.
  * @return wasi_nn_error    Execution status.
  */
-wasi_nn_error
-init_execution_context(graph g, graph_execution_context *ctx)
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(init_execution_context)
+(WASI_NN_NAME(graph) g, WASI_NN_NAME(graph_execution_context) * ctx)
     WASI_NN_IMPORT("init_execution_context");
 
 /**
@@ -71,9 +78,10 @@ init_execution_context(graph g, graph_execution_context *ctx)
  * @param tensor    Input tensor.
  * @return wasi_nn_error    Execution status.
  */
-wasi_nn_error
-set_input(graph_execution_context ctx, uint32_t index, tensor *tensor)
-    WASI_NN_IMPORT("set_input");
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(set_input)
+(WASI_NN_NAME(graph_execution_context) ctx, uint32_t index,
+ WASI_NN_NAME(tensor) * tensor) WASI_NN_IMPORT("set_input");
 
 /**
  * @brief Compute the inference on the given inputs.
@@ -81,8 +89,9 @@ set_input(graph_execution_context ctx, uint32_t index, tensor *tensor)
  * @param ctx       Execution context.
  * @return wasi_nn_error    Execution status.
  */
-wasi_nn_error
-compute(graph_execution_context ctx) WASI_NN_IMPORT("compute");
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(compute)
+(WASI_NN_NAME(graph_execution_context) ctx) WASI_NN_IMPORT("compute");
 
 /**
  * @brief Extract the outputs after inference.
@@ -97,15 +106,16 @@ compute(graph_execution_context ctx) WASI_NN_IMPORT("compute");
  * @return wasi_nn_error                Execution status.
  */
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
-wasi_nn_error
-get_output(graph_execution_context ctx, uint32_t index,
-           tensor_data output_tensor, uint32_t output_tensor_max_size,
-           uint32_t *output_tensor_size) WASI_NN_IMPORT("get_output");
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(get_output)
+(WASI_NN_NAME(graph_execution_context) ctx, uint32_t index,
+ uint8_t *output_tensor, uint32_t output_tensor_max_size,
+ uint32_t *output_tensor_size) WASI_NN_IMPORT("get_output");
 #else
-wasi_nn_error
-get_output(graph_execution_context ctx, uint32_t index,
-           tensor_data output_tensor, uint32_t *output_tensor_size)
-    WASI_NN_IMPORT("get_output");
+WASI_NN_ERROR_TYPE
+WASI_NN_NAME(get_output)
+(graph_execution_context ctx, uint32_t index, uint8_t *output_tensor,
+ uint32_t *output_tensor_size) WASI_NN_IMPORT("get_output");
 #endif
 
 #endif
diff --git a/core/iwasm/libraries/wasi-nn/include/wasi_nn_types.h b/core/iwasm/libraries/wasi-nn/include/wasi_nn_types.h
index dd6b8f14a..952fb65e2 100644
--- a/core/iwasm/libraries/wasi-nn/include/wasi_nn_types.h
+++ b/core/iwasm/libraries/wasi-nn/include/wasi_nn_types.h
@@ -13,6 +13,23 @@
 extern "C" {
 #endif
 
+/* our host logic doesn't use any prefix. neither legacy wasi_nn.h does. */
+
+#if !defined(__wasm__) || !defined(WASI_NN_NAME)
+#define WASI_NN_NAME(name) name
+#define WASI_NN_ERROR_NAME(name) name
+#define WASI_NN_TYPE_NAME(name) name
+#define WASI_NN_ENCODING_NAME(name) name
+#define WASI_NN_TARGET_NAME(name) name
+#define WASI_NN_ERROR_TYPE wasi_nn_error
+#else
+#define WASI_NN_ERROR_NAME(name) WASI_NN_NAME(error_##name)
+#define WASI_NN_TYPE_NAME(name) WASI_NN_NAME(type_##name)
+#define WASI_NN_ENCODING_NAME(name) WASI_NN_NAME(encoding_##name)
+#define WASI_NN_TARGET_NAME(name) WASI_NN_NAME(target_##name)
+#define WASI_NN_ERROR_TYPE WASI_NN_NAME(error);
+#endif
+
 /**
  * ERRORS
  *
@@ -22,22 +39,22 @@ extern "C" {
 // https://github.com/WebAssembly/wasi-nn/blob/71320d95b8c6d43f9af7f44e18b1839db85d89b4/wasi-nn.witx#L5-L17
 // Error codes returned by functions in this API.
 typedef enum {
-    success = 0,
-    invalid_argument,
-    invalid_encoding,
-    missing_memory,
-    busy,
-    runtime_error,
-    unsupported_operation,
-    too_large,
-    not_found,
+    WASI_NN_ERROR_NAME(success) = 0,
+    WASI_NN_ERROR_NAME(invalid_argument),
+    WASI_NN_ERROR_NAME(invalid_encoding),
+    WASI_NN_ERROR_NAME(missing_memory),
+    WASI_NN_ERROR_NAME(busy),
+    WASI_NN_ERROR_NAME(runtime_error),
+    WASI_NN_ERROR_NAME(unsupported_operation),
+    WASI_NN_ERROR_NAME(too_large),
+    WASI_NN_ERROR_NAME(not_found),
 
     // for WasmEdge-wasi-nn
-    end_of_sequence = 100,  // End of Sequence Found.
-    context_full = 101,     // Context Full.
-    prompt_tool_long = 102, // Prompt Too Long.
-    model_not_found = 103,  // Model Not Found.
-} wasi_nn_error;
+    WASI_NN_ERROR_NAME(end_of_sequence) = 100,  // End of Sequence Found.
+    WASI_NN_ERROR_NAME(context_full) = 101,     // Context Full.
+    WASI_NN_ERROR_NAME(prompt_tool_long) = 102, // Prompt Too Long.
+    WASI_NN_ERROR_NAME(model_not_found) = 103,  // Model Not Found.
+} WASI_NN_ERROR_TYPE;
 
 /**
  * TENSOR
@@ -51,15 +68,27 @@ typedef enum {
 typedef struct {
     uint32_t *buf;
     uint32_t size;
-} tensor_dimensions;
+} WASI_NN_NAME(tensor_dimensions);
 
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
 // sync up with
 // https://github.com/WebAssembly/wasi-nn/blob/71320d95b8c6d43f9af7f44e18b1839db85d89b4/wasi-nn.witx#L19-L28
 // The type of the elements in a tensor.
-typedef enum { fp16 = 0, fp32, fp64, u8, i32, i64 } tensor_type;
+typedef enum {
+    WASI_NN_TYPE_NAME(fp16) = 0,
+    WASI_NN_TYPE_NAME(fp32),
+    WASI_NN_TYPE_NAME(fp64),
+    WASI_NN_TYPE_NAME(u8),
+    WASI_NN_TYPE_NAME(i32),
+    WASI_NN_TYPE_NAME(i64),
+} WASI_NN_NAME(tensor_type);
 #else
-typedef enum { fp16 = 0, fp32, up8, ip32 } tensor_type;
+typedef enum {
+    WASI_NN_TYPE_NAME(fp16) = 0,
+    WASI_NN_TYPE_NAME(fp32),
+    WASI_NN_TYPE_NAME(up8),
+    WASI_NN_TYPE_NAME(ip32),
+} WASI_NN_NAME(tensor_type);
 #endif /* WASM_ENABLE_WASI_EPHEMERAL_NN != 0 */
 
 // The tensor data.
@@ -70,7 +99,14 @@ typedef enum { fp16 = 0, fp32, up8, ip32 } tensor_type;
 // 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;
+#if !defined(__wasm__) || WASM_ENABLE_WASI_EPHEMERAL_NN != 0
+typedef struct {
+    uint8_t *buf;
+    uint32_t size;
+} WASI_NN_NAME(tensor_data);
+#else
+typedef uint8_t *WASI_NN_NAME(tensor_data);
+#endif
 
 // A tensor.
 typedef struct {
@@ -78,16 +114,16 @@ typedef struct {
     // represent a tensor containing a single value, use `[1]` for the tensor
     // dimensions.
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0 && defined(__wasm__)
-    tensor_dimensions dimensions;
+    WASI_NN_NAME(tensor_dimensions) dimensions;
 #else
-    tensor_dimensions *dimensions;
+    WASI_NN_NAME(tensor_dimensions) * dimensions;
 #endif
     // Describe the type of element in the tensor (e.g., f32).
     uint8_t type;
     uint8_t _pad[3];
     // Contains the tensor data.
-    tensor_data data;
-} tensor;
+    WASI_NN_NAME(tensor_data) data;
+} WASI_NN_NAME(tensor);
 
 /**
  * GRAPH
@@ -102,15 +138,15 @@ typedef struct {
 typedef struct {
     uint8_t *buf;
     uint32_t size;
-} graph_builder;
+} WASI_NN_NAME(graph_builder);
 
 typedef struct {
-    graph_builder *buf;
+    WASI_NN_NAME(graph_builder) * buf;
     uint32_t size;
-} graph_builder_array;
+} WASI_NN_NAME(graph_builder_array);
 
 // An execution graph for performing inference (i.e., a model).
-typedef uint32_t graph;
+typedef uint32_t WASI_NN_NAME(graph);
 
 // sync up with
 // https://github.com/WebAssembly/wasi-nn/blob/main/wit/wasi-nn.wit#L75
@@ -118,23 +154,25 @@ typedef uint32_t graph;
 // various backends that encode (i.e., serialize) their graph IR with different
 // formats.
 typedef enum {
-    openvino = 0,
-    onnx,
-    tensorflow,
-    pytorch,
-    tensorflowlite,
-    ggml,
-    autodetect,
-    unknown_backend,
-} graph_encoding;
+    WASI_NN_ENCODING_NAME(openvino) = 0,
+    WASI_NN_ENCODING_NAME(onnx),
+    WASI_NN_ENCODING_NAME(tensorflow),
+    WASI_NN_ENCODING_NAME(pytorch),
+    WASI_NN_ENCODING_NAME(tensorflowlite),
+    WASI_NN_ENCODING_NAME(ggml),
+    WASI_NN_ENCODING_NAME(autodetect),
+    WASI_NN_ENCODING_NAME(unknown_backend),
+} WASI_NN_NAME(graph_encoding);
 
 // Define where the graph should be executed.
-typedef enum execution_target { cpu = 0, gpu, tpu } execution_target;
+typedef enum WASI_NN_NAME(execution_target) {
+    WASI_NN_TARGET_NAME(cpu) = 0,
+    WASI_NN_TARGET_NAME(gpu),
+    WASI_NN_TARGET_NAME(tpu),
+} WASI_NN_NAME(execution_target);
 
 // Bind a `graph` to the input and output tensors for an inference.
-typedef uint32_t graph_execution_context;
-
-/* Definition of 'wasi_nn.h' structs in WASM app format (using offset) */
+typedef uint32_t WASI_NN_NAME(graph_execution_context);
 
 #ifdef __cplusplus
 }
diff --git a/core/iwasm/libraries/wasi-nn/src/utils/wasi_nn_app_native.c b/core/iwasm/libraries/wasi-nn/src/utils/wasi_nn_app_native.c
index 6e91c949b..4d56fed93 100644
--- a/core/iwasm/libraries/wasi-nn/src/utils/wasi_nn_app_native.c
+++ b/core/iwasm/libraries/wasi-nn/src/utils/wasi_nn_app_native.c
@@ -99,7 +99,8 @@ graph_builder_array_app_native(wasm_module_inst_t instance,
 
 static wasi_nn_error
 tensor_data_app_native(wasm_module_inst_t instance, uint32_t total_elements,
-                       tensor_wasm *input_tensor_wasm, tensor_data *data)
+                       tensor_wasm *input_tensor_wasm, void **data,
+                       uint32_t *size)
 {
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
 #define data_size input_tensor_wasm->data_size
@@ -113,8 +114,9 @@ tensor_data_app_native(wasm_module_inst_t instance, uint32_t total_elements,
         NN_ERR_PRINTF("input_tensor_wasm->data_offset is invalid");
         return invalid_argument;
     }
-    *data = (tensor_data)wasm_runtime_addr_app_to_native(
+    *data = wasm_runtime_addr_app_to_native(
         instance, (uint64)input_tensor_wasm->data_offset);
+    *size = data_size;
     return success;
 #undef data_size
 }
@@ -188,16 +190,19 @@ tensor_app_native(wasm_module_inst_t instance, tensor_wasm *input_tensor_wasm,
     NN_DBG_PRINTF("Tensor type: %d", input_tensor_wasm->type);
     NN_DBG_PRINTF("Total number of elements: %d", total_elements);
 
-    tensor_data data = NULL;
+    void *data = NULL;
+    uint32_t datasize;
     if (success
-        != (res = tensor_data_app_native(instance, total_elements,
-                                         input_tensor_wasm, &data))) {
+        != (res =
+                tensor_data_app_native(instance, total_elements,
+                                       input_tensor_wasm, &data, &datasize))) {
         wasm_runtime_free(dimensions);
         return res;
     }
 
     input_tensor->type = input_tensor_wasm->type;
     input_tensor->dimensions = dimensions;
-    input_tensor->data = data;
+    input_tensor->data.buf = data;
+    input_tensor->data.size = datasize;
     return success;
 }
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn.c b/core/iwasm/libraries/wasi-nn/src/wasi_nn.c
index 1a8ad03c6..5c865c5be 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn.c
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn.c
@@ -20,6 +20,10 @@
 #include "wasi_nn_types.h"
 #include "wasm_export.h"
 
+#if WASM_ENABLE_WASI_EPHEMERAL_NN == 0
+#warning You are using "wasi_nn", which is a legacy WAMR-specific ABI. It's deperecated and will likely be removed in future versions of WAMR. Please use "wasi_ephemeral_nn" instead. (For a WASM module, use the wasi_ephemeral_nn.h header instead. For the runtime configurations, enable WASM_ENABLE_WASI_EPHEMERAL_NN/WAMR_BUILD_WASI_EPHEMERAL_NN.)
+#endif
+
 #define HASHMAP_INITIAL_SIZE 20
 #if defined(__APPLE__)
 #define LIB_EXTENTION ".dylib"
@@ -51,65 +55,36 @@ struct backends_api_functions {
             NN_ERR_PRINTF("Error %s() -> %d", #func, wasi_error);          \
     } while (0)
 
-/* HashMap utils */
-static HashMap *hashmap;
-
-static uint32
-hash_func(const void *key)
-{
-    // fnv1a_hash
-    const uint32 FNV_PRIME = 16777619;
-    const uint32 FNV_OFFSET_BASIS = 2166136261U;
-
-    uint32 hash = FNV_OFFSET_BASIS;
-    const unsigned char *bytes = (const unsigned char *)key;
-
-    for (size_t i = 0; i < sizeof(uintptr_t); ++i) {
-        hash ^= bytes[i];
-        hash *= FNV_PRIME;
-    }
-
-    return hash;
-}
-
-static bool
-key_equal_func(void *key1, void *key2)
-{
-    return key1 == key2;
-}
-
-static void
-key_destroy_func(void *key1)
-{
-    /* key type is wasm_module_inst_t*. do nothing */
-}
+static void *wasi_nn_key;
 
 static void
 wasi_nn_ctx_destroy(WASINNContext *wasi_nn_ctx)
 {
-    NN_DBG_PRINTF("[WASI NN] DEINIT...");
-
     if (wasi_nn_ctx == NULL) {
-        NN_ERR_PRINTF(
-            "Error when deallocating memory. WASI-NN context is NULL");
         return;
     }
+    NN_DBG_PRINTF("[WASI NN] DEINIT...");
     NN_DBG_PRINTF("Freeing wasi-nn");
     NN_DBG_PRINTF("-> is_model_loaded: %d", wasi_nn_ctx->is_model_loaded);
     NN_DBG_PRINTF("-> current_encoding: %d", wasi_nn_ctx->backend);
 
-    /* deinit() the backend */
-    wasi_nn_error res;
-    call_wasi_nn_func(wasi_nn_ctx->backend, deinit, res,
-                      wasi_nn_ctx->backend_ctx);
+    bh_assert(!wasi_nn_ctx->busy);
 
+    /* deinit() the backend */
+    if (wasi_nn_ctx->is_backend_ctx_initialized) {
+        wasi_nn_error res;
+        call_wasi_nn_func(wasi_nn_ctx->backend, deinit, res,
+                          wasi_nn_ctx->backend_ctx);
+    }
+
+    os_mutex_destroy(&wasi_nn_ctx->lock);
     wasm_runtime_free(wasi_nn_ctx);
 }
 
 static void
-value_destroy_func(void *value)
+dtor(wasm_module_inst_t inst, void *ctx)
 {
-    wasi_nn_ctx_destroy((WASINNContext *)value);
+    wasi_nn_ctx_destroy(ctx);
 }
 
 bool
@@ -122,12 +97,9 @@ wasi_nn_initialize()
         return false;
     }
 
-    // hashmap { instance: wasi_nn_ctx }
-    hashmap = bh_hash_map_create(HASHMAP_INITIAL_SIZE, true, hash_func,
-                                 key_equal_func, key_destroy_func,
-                                 value_destroy_func);
-    if (hashmap == NULL) {
-        NN_ERR_PRINTF("Error while initializing hashmap");
+    wasi_nn_key = wasm_runtime_create_context_key(dtor);
+    if (wasi_nn_key == NULL) {
+        NN_ERR_PRINTF("Failed to create context key");
         os_mutex_destroy(&wasi_nn_lock);
         return false;
     }
@@ -148,6 +120,11 @@ wasi_nn_initialize_context()
     }
 
     memset(wasi_nn_ctx, 0, sizeof(WASINNContext));
+    if (os_mutex_init(&wasi_nn_ctx->lock)) {
+        NN_ERR_PRINTF("Error when initializing a lock for WASI-NN context");
+        wasm_runtime_free(wasi_nn_ctx);
+        return NULL;
+    }
     return wasi_nn_ctx;
 }
 
@@ -156,29 +133,59 @@ static WASINNContext *
 wasm_runtime_get_wasi_nn_ctx(wasm_module_inst_t instance)
 {
     WASINNContext *wasi_nn_ctx =
-        (WASINNContext *)bh_hash_map_find(hashmap, (void *)instance);
+        wasm_runtime_get_context(instance, wasi_nn_key);
     if (wasi_nn_ctx == NULL) {
-        wasi_nn_ctx = wasi_nn_initialize_context();
-        if (wasi_nn_ctx == NULL)
-            return NULL;
-
-        bool ok =
-            bh_hash_map_insert(hashmap, (void *)instance, (void *)wasi_nn_ctx);
-        if (!ok) {
-            NN_ERR_PRINTF("Error while storing context");
-            wasi_nn_ctx_destroy(wasi_nn_ctx);
+        WASINNContext *newctx = wasi_nn_initialize_context();
+        if (newctx == NULL)
             return NULL;
+        os_mutex_lock(&wasi_nn_lock);
+        wasi_nn_ctx = wasm_runtime_get_context(instance, wasi_nn_key);
+        if (wasi_nn_ctx == NULL) {
+            wasm_runtime_set_context_spread(instance, wasi_nn_key, newctx);
+            wasi_nn_ctx = newctx;
+            newctx = NULL;
+        }
+        os_mutex_unlock(&wasi_nn_lock);
+        if (newctx != NULL) {
+            wasi_nn_ctx_destroy(newctx);
         }
     }
-
     return wasi_nn_ctx;
 }
 
+static WASINNContext *
+lock_ctx(wasm_module_inst_t instance)
+{
+    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        return NULL;
+    }
+    os_mutex_lock(&wasi_nn_ctx->lock);
+    if (wasi_nn_ctx->busy) {
+        os_mutex_unlock(&wasi_nn_ctx->lock);
+        return NULL;
+    }
+    wasi_nn_ctx->busy = true;
+    os_mutex_unlock(&wasi_nn_ctx->lock);
+    return wasi_nn_ctx;
+}
+
+static void
+unlock_ctx(WASINNContext *wasi_nn_ctx)
+{
+    if (wasi_nn_ctx == NULL) {
+        return;
+    }
+    os_mutex_lock(&wasi_nn_ctx->lock);
+    bh_assert(wasi_nn_ctx->busy);
+    wasi_nn_ctx->busy = false;
+    os_mutex_unlock(&wasi_nn_ctx->lock);
+}
+
 void
 wasi_nn_destroy()
 {
-    // destroy hashmap will destroy keys and values
-    bh_hash_map_destroy(hashmap);
+    wasm_runtime_destroy_context_key(wasi_nn_key);
 
     // close backends' libraries and registered functions
     for (unsigned i = 0; i < sizeof(lookup) / sizeof(lookup[0]); i++) {
@@ -397,6 +404,41 @@ detect_and_load_backend(graph_encoding backend_hint,
     return ret;
 }
 
+static wasi_nn_error
+ensure_backend(wasm_module_inst_t instance, graph_encoding encoding,
+               WASINNContext *wasi_nn_ctx)
+{
+    wasi_nn_error res;
+
+    graph_encoding loaded_backend = autodetect;
+    if (!detect_and_load_backend(encoding, &loaded_backend)) {
+        res = invalid_encoding;
+        NN_ERR_PRINTF("load backend failed");
+        goto fail;
+    }
+
+    if (wasi_nn_ctx->is_backend_ctx_initialized) {
+        if (wasi_nn_ctx->backend != loaded_backend) {
+            res = unsupported_operation;
+            goto fail;
+        }
+    }
+    else {
+        wasi_nn_ctx->backend = loaded_backend;
+
+        /* init() the backend */
+        call_wasi_nn_func(wasi_nn_ctx->backend, init, res,
+                          &wasi_nn_ctx->backend_ctx);
+        if (res != success)
+            goto fail;
+
+        wasi_nn_ctx->is_backend_ctx_initialized = true;
+    }
+    return success;
+fail:
+    return res;
+}
+
 /* WASI-NN implementation */
 
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
@@ -410,6 +452,8 @@ wasi_nn_load(wasm_exec_env_t exec_env, graph_builder_array_wasm *builder,
              graph_encoding encoding, execution_target target, graph *g)
 #endif /* WASM_ENABLE_WASI_EPHEMERAL_NN != 0 */
 {
+    wasi_nn_error res;
+
     NN_DBG_PRINTF("[WASI NN] LOAD [encoding=%d, target=%d]...", encoding,
                   target);
 
@@ -417,18 +461,23 @@ wasi_nn_load(wasm_exec_env_t exec_env, graph_builder_array_wasm *builder,
     if (!instance)
         return runtime_error;
 
-    wasi_nn_error res;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
+    }
+
     graph_builder_array builder_native = { 0 };
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
     if (success
         != (res = graph_builder_array_app_native(
                 instance, builder, builder_wasm_size, &builder_native)))
-        return res;
+        goto fail;
 #else  /* WASM_ENABLE_WASI_EPHEMERAL_NN == 0 */
     if (success
         != (res = graph_builder_array_app_native(instance, builder,
                                                  &builder_native)))
-        return res;
+        goto fail;
 #endif /* WASM_ENABLE_WASI_EPHEMERAL_NN != 0 */
 
     if (!wasm_runtime_validate_native_addr(instance, g,
@@ -438,19 +487,7 @@ wasi_nn_load(wasm_exec_env_t exec_env, graph_builder_array_wasm *builder,
         goto fail;
     }
 
-    graph_encoding loaded_backend = autodetect;
-    if (!detect_and_load_backend(encoding, &loaded_backend)) {
-        res = invalid_encoding;
-        NN_ERR_PRINTF("load backend failed");
-        goto fail;
-    }
-
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-    wasi_nn_ctx->backend = loaded_backend;
-
-    /* init() the backend */
-    call_wasi_nn_func(wasi_nn_ctx->backend, init, res,
-                      &wasi_nn_ctx->backend_ctx);
+    res = ensure_backend(instance, encoding, wasi_nn_ctx);
     if (res != success)
         goto fail;
 
@@ -465,6 +502,7 @@ fail:
     // XXX: Free intermediate structure pointers
     if (builder_native.buf)
         wasm_runtime_free(builder_native.buf);
+    unlock_ctx(wasi_nn_ctx);
 
     return res;
 }
@@ -473,6 +511,8 @@ wasi_nn_error
 wasi_nn_load_by_name(wasm_exec_env_t exec_env, char *name, uint32_t name_len,
                      graph *g)
 {
+    wasi_nn_error res;
+
     wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
     if (!instance) {
         return runtime_error;
@@ -496,29 +536,26 @@ wasi_nn_load_by_name(wasm_exec_env_t exec_env, char *name, uint32_t name_len,
 
     NN_DBG_PRINTF("[WASI NN] LOAD_BY_NAME %s...", name);
 
-    graph_encoding loaded_backend = autodetect;
-    if (!detect_and_load_backend(autodetect, &loaded_backend)) {
-        NN_ERR_PRINTF("load backend failed");
-        return invalid_encoding;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
     }
 
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-    wasi_nn_ctx->backend = loaded_backend;
-
-    wasi_nn_error res;
-    /* init() the backend */
-    call_wasi_nn_func(wasi_nn_ctx->backend, init, res,
-                      &wasi_nn_ctx->backend_ctx);
+    res = ensure_backend(instance, autodetect, wasi_nn_ctx);
     if (res != success)
-        return res;
+        goto fail;
 
     call_wasi_nn_func(wasi_nn_ctx->backend, load_by_name, res,
                       wasi_nn_ctx->backend_ctx, name, name_len, g);
     if (res != success)
-        return res;
+        goto fail;
 
     wasi_nn_ctx->is_model_loaded = true;
-    return success;
+    res = success;
+fail:
+    unlock_ctx(wasi_nn_ctx);
+    return res;
 }
 
 wasi_nn_error
@@ -526,6 +563,8 @@ wasi_nn_load_by_name_with_config(wasm_exec_env_t exec_env, char *name,
                                  int32_t name_len, char *config,
                                  int32_t config_len, graph *g)
 {
+    wasi_nn_error res;
+
     wasm_module_inst_t instance = wasm_runtime_get_module_inst(exec_env);
     if (!instance) {
         return runtime_error;
@@ -554,30 +593,28 @@ wasi_nn_load_by_name_with_config(wasm_exec_env_t exec_env, char *name,
 
     NN_DBG_PRINTF("[WASI NN] LOAD_BY_NAME_WITH_CONFIG %s %s...", name, config);
 
-    graph_encoding loaded_backend = autodetect;
-    if (!detect_and_load_backend(autodetect, &loaded_backend)) {
-        NN_ERR_PRINTF("load backend failed");
-        return invalid_encoding;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
     }
 
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-    wasi_nn_ctx->backend = loaded_backend;
-
-    wasi_nn_error res;
-    /* init() the backend */
-    call_wasi_nn_func(wasi_nn_ctx->backend, init, res,
-                      &wasi_nn_ctx->backend_ctx);
+    res = ensure_backend(instance, autodetect, wasi_nn_ctx);
     if (res != success)
-        return res;
+        goto fail;
+    ;
 
     call_wasi_nn_func(wasi_nn_ctx->backend, load_by_name_with_config, res,
                       wasi_nn_ctx->backend_ctx, name, name_len, config,
                       config_len, g);
     if (res != success)
-        return res;
+        goto fail;
 
     wasi_nn_ctx->is_model_loaded = true;
-    return success;
+    res = success;
+fail:
+    unlock_ctx(wasi_nn_ctx);
+    return res;
 }
 
 wasi_nn_error
@@ -591,20 +628,27 @@ wasi_nn_init_execution_context(wasm_exec_env_t exec_env, graph g,
         return runtime_error;
     }
 
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-
     wasi_nn_error res;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
+    }
+
     if (success != (res = is_model_initialized(wasi_nn_ctx)))
-        return res;
+        goto fail;
 
     if (!wasm_runtime_validate_native_addr(
             instance, ctx, (uint64)sizeof(graph_execution_context))) {
         NN_ERR_PRINTF("ctx is invalid");
-        return invalid_argument;
+        res = invalid_argument;
+        goto fail;
     }
 
     call_wasi_nn_func(wasi_nn_ctx->backend, init_execution_context, res,
                       wasi_nn_ctx->backend_ctx, g, ctx);
+fail:
+    unlock_ctx(wasi_nn_ctx);
     return res;
 }
 
@@ -619,17 +663,21 @@ wasi_nn_set_input(wasm_exec_env_t exec_env, graph_execution_context ctx,
         return runtime_error;
     }
 
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-
     wasi_nn_error res;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
+    }
+
     if (success != (res = is_model_initialized(wasi_nn_ctx)))
-        return res;
+        goto fail;
 
     tensor input_tensor_native = { 0 };
     if (success
         != (res = tensor_app_native(instance, input_tensor,
                                     &input_tensor_native)))
-        return res;
+        goto fail;
 
     call_wasi_nn_func(wasi_nn_ctx->backend, set_input, res,
                       wasi_nn_ctx->backend_ctx, ctx, index,
@@ -637,7 +685,8 @@ wasi_nn_set_input(wasm_exec_env_t exec_env, graph_execution_context ctx,
     // XXX: Free intermediate structure pointers
     if (input_tensor_native.dimensions)
         wasm_runtime_free(input_tensor_native.dimensions);
-
+fail:
+    unlock_ctx(wasi_nn_ctx);
     return res;
 }
 
@@ -651,26 +700,32 @@ wasi_nn_compute(wasm_exec_env_t exec_env, graph_execution_context ctx)
         return runtime_error;
     }
 
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-
     wasi_nn_error res;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
+    }
+
     if (success != (res = is_model_initialized(wasi_nn_ctx)))
-        return res;
+        goto fail;
 
     call_wasi_nn_func(wasi_nn_ctx->backend, compute, res,
                       wasi_nn_ctx->backend_ctx, ctx);
+fail:
+    unlock_ctx(wasi_nn_ctx);
     return res;
 }
 
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
 wasi_nn_error
 wasi_nn_get_output(wasm_exec_env_t exec_env, graph_execution_context ctx,
-                   uint32_t index, tensor_data output_tensor,
+                   uint32_t index, void *output_tensor,
                    uint32_t output_tensor_len, uint32_t *output_tensor_size)
 #else  /* WASM_ENABLE_WASI_EPHEMERAL_NN == 0 */
 wasi_nn_error
 wasi_nn_get_output(wasm_exec_env_t exec_env, graph_execution_context ctx,
-                   uint32_t index, tensor_data output_tensor,
+                   uint32_t index, void *output_tensor,
                    uint32_t *output_tensor_size)
 #endif /* WASM_ENABLE_WASI_EPHEMERAL_NN != 0 */
 {
@@ -681,28 +736,36 @@ wasi_nn_get_output(wasm_exec_env_t exec_env, graph_execution_context ctx,
         return runtime_error;
     }
 
-    WASINNContext *wasi_nn_ctx = wasm_runtime_get_wasi_nn_ctx(instance);
-
     wasi_nn_error res;
+    WASINNContext *wasi_nn_ctx = lock_ctx(instance);
+    if (wasi_nn_ctx == NULL) {
+        res = busy;
+        goto fail;
+    }
+
     if (success != (res = is_model_initialized(wasi_nn_ctx)))
-        return res;
+        goto fail;
 
     if (!wasm_runtime_validate_native_addr(instance, output_tensor_size,
                                            (uint64)sizeof(uint32_t))) {
         NN_ERR_PRINTF("output_tensor_size is invalid");
-        return invalid_argument;
+        res = invalid_argument;
+        goto fail;
     }
 
+    tensor_data tensor = {
+        .buf = output_tensor,
 #if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
+        .size = output_tensor_len,
+#else
+        .size = *output_tensor_size,
+#endif
+    };
     call_wasi_nn_func(wasi_nn_ctx->backend, get_output, res,
-                      wasi_nn_ctx->backend_ctx, ctx, index, output_tensor,
-                      &output_tensor_len);
-    *output_tensor_size = output_tensor_len;
-#else  /* WASM_ENABLE_WASI_EPHEMERAL_NN == 0 */
-    call_wasi_nn_func(wasi_nn_ctx->backend, get_output, res,
-                      wasi_nn_ctx->backend_ctx, ctx, index, output_tensor,
+                      wasi_nn_ctx->backend_ctx, ctx, index, &tensor,
                       output_tensor_size);
-#endif /* WASM_ENABLE_WASI_EPHEMERAL_NN != 0 */
+fail:
+    unlock_ctx(wasi_nn_ctx);
     return res;
 }
 
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn_llamacpp.c b/core/iwasm/libraries/wasi-nn/src/wasi_nn_llamacpp.c
index 23c867b0a..572a5bf33 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn_llamacpp.c
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn_llamacpp.c
@@ -2,6 +2,9 @@
  * Copyright (C) 2019 Intel Corporation.  All rights reserved.
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  */
+
+#include <stdlib.h>
+
 #include "wasi_nn_types.h"
 #include "utils/logger.h"
 #include "llama.h"
@@ -286,7 +289,7 @@ deinit_backend(void *ctx)
 
     llama_backend_free();
 
-    os_free(backend_ctx);
+    free(backend_ctx);
     return success;
 }
 
@@ -381,18 +384,18 @@ set_input(void *ctx, graph_execution_context exec_ctx, uint32_t index,
           tensor *wasi_nn_tensor)
 {
     struct LlamaContext *backend_ctx = (struct LlamaContext *)ctx;
-    // tensor->data is the prompt string. ends with \0
-    char *prompt_text = (char *)wasi_nn_tensor->data;
+    // tensor->data is the prompt string.
+    char *prompt_text = (char *)wasi_nn_tensor->data.buf;
+    uint32_t prompt_text_len = wasi_nn_tensor->data.size;
 
 #ifndef NDEBUG
     NN_DBG_PRINTF("--------------------------------------------------");
-    NN_DBG_PRINTF("prompt_text: %s", prompt_text);
+    NN_DBG_PRINTF("prompt_text: %.*s", (int)prompt_text_len, prompt_text);
     NN_DBG_PRINTF("--------------------------------------------------");
 #endif
 
     // tokenize the prompt
     uint32_t n_token_max = llama_n_ctx(backend_ctx->ctx);
-    uint32_t prompt_text_len = strlen(prompt_text);
 
     if (backend_ctx->prompt == NULL) {
         backend_ctx->prompt = calloc(n_token_max, sizeof(llama_token));
@@ -477,7 +480,6 @@ compute(void *ctx, graph_execution_context exec_ctx)
 
     // main loop
     int32_t n_cur = batch.n_tokens;
-    int n_decode = 0;
     int32_t n_vocab = llama_n_vocab(backend_ctx->model);
     llama_token_data *candidates = NULL;
 
@@ -528,7 +530,6 @@ compute(void *ctx, graph_execution_context exec_ctx)
         // push this new token for next evaluation
         llama_batch_add(&batch, new_token_id, n_cur, seq_ids,
                         sizeof(seq_ids) / sizeof(seq_ids[0]), true);
-        n_decode++;
         n_cur++;
 
         if (llama_decode(backend_ctx->ctx, batch) != 0) {
@@ -549,7 +550,7 @@ fail:
 
 __attribute__((visibility("default"))) wasi_nn_error
 get_output(void *ctx, graph_execution_context exec_ctx, uint32_t index,
-           tensor_data output_tensor, uint32_t *output_tensor_size)
+           tensor_data *output_tensor, uint32_t *output_tensor_size)
 {
     struct LlamaContext *backend_ctx = (struct LlamaContext *)ctx;
 
@@ -568,7 +569,7 @@ get_output(void *ctx, graph_execution_context exec_ctx, uint32_t index,
             printf("%s\n", output_metadata);
         }
 
-        memcpy(output_tensor, output_metadata, strlen(output_metadata));
+        memcpy(output_tensor->buf, output_metadata, strlen(output_metadata));
         *output_tensor_size = strlen(output_metadata);
         return success;
     }
@@ -588,7 +589,7 @@ get_output(void *ctx, graph_execution_context exec_ctx, uint32_t index,
             printf("%s", buf);
         }
 
-        memcpy(output_tensor + end_pos, buf, strlen(buf));
+        memcpy(output_tensor->buf + end_pos, buf, strlen(buf));
         end_pos += strlen(buf);
     }
 
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.c b/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.c
index dcfec1ccb..ae3a8572c 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.c
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.c
@@ -26,17 +26,25 @@
  * from 4. to 6. is the Inference Loop
  */
 
+/* these limits are arbitrary. */
+#define MAX_GRAPHS 4
+#define MAX_EXECUTION_CONTEXTS 4
+
 typedef struct {
     ov_core_t *core;
     /* keep input model files */
-    void *weight_data;
-    ov_tensor_t *weights_tensor;
-    ov_model_t *model;
-    /* add prepostprocess */
-    ov_model_t *new_model;
-    ov_compiled_model_t *compiled_model;
-    ov_infer_request_t *infer_request;
-    ov_tensor_t *input_tensor;
+    struct OpenVINOGraph {
+        void *weight_data;
+        ov_tensor_t *weights_tensor;
+        ov_model_t *model;
+        ov_compiled_model_t *compiled_model;
+    } graphs[MAX_GRAPHS];
+    struct OpenVINOExecutionContext {
+        struct OpenVINOGraph *graph;
+        ov_infer_request_t *infer_request;
+    } execution_contexts[MAX_EXECUTION_CONTEXTS];
+    unsigned int n_graphs;
+    unsigned int n_execution_contexts;
 } OpenVINOContext;
 
 /*
@@ -181,6 +189,29 @@ wasi_nn_tensor_type_to_openvino_element_type(tensor_type wasi_nn_type)
     return UNDEFINED;
 }
 
+static void
+free_graph(struct OpenVINOGraph *graph)
+{
+    if (graph->weight_data)
+        os_free(graph->weight_data);
+
+    if (graph->weights_tensor)
+        ov_tensor_free(graph->weights_tensor);
+
+    if (graph->model)
+        ov_model_free(graph->model);
+
+    if (graph->compiled_model)
+        ov_compiled_model_free(graph->compiled_model);
+}
+
+static void
+free_execution_context(struct OpenVINOExecutionContext *c)
+{
+    if (c->infer_request)
+        ov_infer_request_free(c->infer_request);
+}
+
 static wasi_nn_error
 uint32_array_to_int64_array(uint32_t array_size, uint32_t *src, int64_t **dst)
 {
@@ -200,6 +231,8 @@ load(void *ctx, graph_builder_array *builder, graph_encoding encoding,
      execution_target target, graph *g)
 {
     OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    struct OpenVINOGraph *graph;
+    unsigned int graph_idx;
     wasi_nn_error ret = unsupported_operation;
 
     if (encoding != openvino) {
@@ -225,39 +258,47 @@ load(void *ctx, graph_builder_array *builder, graph_encoding encoding,
     graph_builder xml = builder->buf[0];
     graph_builder weight = builder->buf[1];
 
-    /* if xml is a String with a model in IR */
-    if (!(xml.buf[xml.size] == '\0' && xml.buf[xml.size - 1] != '\0')) {
-        NN_ERR_PRINTF("Invalid xml string.");
-        return invalid_argument;
+    graph_idx = ov_ctx->n_graphs;
+    if (graph_idx >= MAX_GRAPHS) {
+        return runtime_error;
     }
+    graph = &ov_ctx->graphs[graph_idx];
+    memset(graph, 0, sizeof(*graph));
 
     /* transfer weight to an ov tensor */
     {
-        ov_ctx->weight_data = os_malloc(weight.size);
-        if (!ov_ctx->weight_data)
+        graph->weight_data = os_malloc(weight.size);
+        if (!graph->weight_data)
             goto fail;
-        memcpy(ov_ctx->weight_data, weight.buf, weight.size);
+        memcpy(graph->weight_data, weight.buf, weight.size);
 
         ov_element_type_e type = U8;
         int64_t dims[1] = { weight.size };
         ov_shape_t shape = { 1, dims };
         CHECK_OV_STATUS(ov_tensor_create_from_host_ptr(type, shape,
-                                                       ov_ctx->weight_data,
-                                                       &ov_ctx->weights_tensor),
+                                                       graph->weight_data,
+                                                       &graph->weights_tensor),
                         ret);
     }
 
     /* load model from buffer */
     CHECK_OV_STATUS(ov_core_read_model_from_memory_buffer(
                         ov_ctx->core, (char *)xml.buf, xml.size,
-                        ov_ctx->weights_tensor, &ov_ctx->model),
+                        graph->weights_tensor, &graph->model),
                     ret);
 #ifndef NDEBUG
-    print_model_input_output_info(ov_ctx->model);
+    print_model_input_output_info(graph->model);
 #endif
 
-    ret = success;
+    CHECK_OV_STATUS(ov_core_compile_model(ov_ctx->core, graph->model, "CPU", 0,
+                                          &graph->compiled_model),
+                    ret);
+
+    *g = graph_idx;
+    ov_ctx->n_graphs++;
+    return success;
 fail:
+    free_graph(graph);
     return ret;
 }
 
@@ -265,20 +306,62 @@ __attribute__((visibility("default"))) wasi_nn_error
 load_by_name(void *ctx, const char *filename, uint32_t filename_len, graph *g)
 {
     OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    struct OpenVINOGraph *graph;
+    unsigned int graph_idx;
     wasi_nn_error ret = unsupported_operation;
 
-    CHECK_OV_STATUS(
-        ov_core_read_model(ov_ctx->core, filename, NULL, &ov_ctx->model), ret);
+    graph_idx = ov_ctx->n_graphs;
+    if (graph_idx >= MAX_GRAPHS) {
+        return runtime_error;
+    }
+    graph = &ov_ctx->graphs[graph_idx];
 
-    ret = success;
+    memset(graph, 0, sizeof(*graph));
+    CHECK_OV_STATUS(
+        ov_core_read_model(ov_ctx->core, filename, NULL, &graph->model), ret);
+
+    CHECK_OV_STATUS(ov_core_compile_model(ov_ctx->core, graph->model, "CPU", 0,
+                                          &graph->compiled_model),
+                    ret);
+
+    *g = graph_idx;
+    ov_ctx->n_graphs++;
+    return success;
 fail:
+    free_graph(graph);
     return ret;
 }
 
 __attribute__((visibility("default"))) wasi_nn_error
 init_execution_context(void *ctx, graph g, graph_execution_context *exec_ctx)
 {
+    OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    struct OpenVINOGraph *graph;
+    struct OpenVINOExecutionContext *exec;
+    unsigned int exec_idx;
+    wasi_nn_error ret;
+
+    if (g >= ov_ctx->n_graphs)
+        return runtime_error;
+    graph = &ov_ctx->graphs[g];
+
+    exec_idx = ov_ctx->n_execution_contexts;
+    if (exec_idx >= MAX_EXECUTION_CONTEXTS)
+        return runtime_error;
+    exec = &ov_ctx->execution_contexts[exec_idx];
+
+    memset(exec, 0, sizeof(*exec));
+    exec->graph = graph;
+
+    CHECK_OV_STATUS(ov_compiled_model_create_infer_request(
+                        graph->compiled_model, &exec->infer_request),
+                    ret);
+
+    *exec_ctx = exec_idx;
+    ov_ctx->n_execution_contexts++;
     return success;
+fail:
+    return ret;
 }
 
 __attribute__((visibility("default"))) wasi_nn_error
@@ -286,19 +369,15 @@ set_input(void *ctx, graph_execution_context exec_ctx, uint32_t index,
           tensor *wasi_nn_tensor)
 {
     OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    struct OpenVINOExecutionContext *exec;
     wasi_nn_error ret = unsupported_operation;
     ov_shape_t input_shape = { 0 };
+    ov_tensor_t *input_tensor = NULL;
     int64_t *ov_dims = NULL;
 
-    ov_preprocess_prepostprocessor_t *ppp = NULL;
-    ov_preprocess_input_info_t *input_info = NULL;
-    ov_preprocess_input_tensor_info_t *input_tensor_info = NULL;
-    ov_layout_t *input_layout = NULL;
-    ov_preprocess_preprocess_steps_t *input_process = NULL;
-    ov_preprocess_input_model_info_t *p_input_model = NULL;
-    ov_layout_t *model_layout = NULL;
-    ov_preprocess_output_info_t *output_info = NULL;
-    ov_preprocess_output_tensor_info_t *output_tensor_info = NULL;
+    if (exec_ctx >= ov_ctx->n_execution_contexts)
+        return runtime_error;
+    exec = &ov_ctx->execution_contexts[exec_ctx];
 
     /* wasi_nn_tensor -> ov_tensor */
     {
@@ -308,17 +387,6 @@ set_input(void *ctx, graph_execution_context exec_ctx, uint32_t index,
         if (ret != success)
             goto fail;
 
-        /* NCHW -> NHWC */
-        if (wasi_nn_tensor->dimensions->size == 4 || ov_dims[1] == 3) {
-            /* N */
-            /* H */
-            ov_dims[1] = ov_dims[2];
-            /* W */
-            ov_dims[2] = ov_dims[3];
-            /* C */
-            ov_dims[3] = 3;
-        }
-
         CHECK_OV_STATUS(ov_shape_create(wasi_nn_tensor->dimensions->size,
                                         ov_dims, &input_shape),
                         ret);
@@ -334,101 +402,22 @@ set_input(void *ctx, graph_execution_context exec_ctx, uint32_t index,
                       shape_info);
 
         CHECK_OV_STATUS(ov_tensor_create_from_host_ptr(input_type, input_shape,
-                                                       wasi_nn_tensor->data,
-                                                       &ov_ctx->input_tensor),
+                                                       wasi_nn_tensor->data.buf,
+                                                       &input_tensor),
                         ret);
     }
 
-    /* set preprocess based on wasi_nn_tensor */
-    {
-        CHECK_OV_STATUS(
-            ov_preprocess_prepostprocessor_create(ov_ctx->model, &ppp), ret);
-
-        /* reuse user' created tensor's info */
-        CHECK_OV_STATUS(ov_preprocess_prepostprocessor_get_input_info_by_index(
-                            ppp, index, &input_info),
-                        ret);
-        CHECK_OV_STATUS(ov_preprocess_input_info_get_tensor_info(
-                            input_info, &input_tensor_info),
-                        ret);
-        CHECK_OV_STATUS(ov_preprocess_input_tensor_info_set_from(
-                            input_tensor_info, ov_ctx->input_tensor),
-                        ret);
-        /* ! HAS TO BE NHWC. Match previous layout conversion */
-        CHECK_OV_STATUS(ov_layout_create("NHWC", &input_layout), ret);
-        CHECK_OV_STATUS(ov_preprocess_input_tensor_info_set_layout(
-                            input_tensor_info, input_layout),
-                        ret);
-
-        /* add RESIZE */
-        CHECK_OV_STATUS(ov_preprocess_input_info_get_preprocess_steps(
-                            input_info, &input_process),
-                        ret);
-        CHECK_OV_STATUS(
-            ov_preprocess_preprocess_steps_resize(input_process, RESIZE_LINEAR),
-            ret);
-
-        /* input model */
-        CHECK_OV_STATUS(
-            ov_preprocess_input_info_get_model_info(input_info, &p_input_model),
-            ret);
-        // TODO: what if not?
-        CHECK_OV_STATUS(ov_layout_create("NCHW", &model_layout), ret);
-        CHECK_OV_STATUS(ov_preprocess_input_model_info_set_layout(p_input_model,
-                                                                  model_layout),
-                        ret);
-
-        /* output -> F32(possibility) */
-        CHECK_OV_STATUS(ov_preprocess_prepostprocessor_get_output_info_by_index(
-                            ppp, index, &output_info),
-                        ret);
-        CHECK_OV_STATUS(ov_preprocess_output_info_get_tensor_info(
-                            output_info, &output_tensor_info),
-                        ret);
-        CHECK_OV_STATUS(
-            ov_preprocess_output_set_element_type(output_tensor_info, F32),
-            ret);
-
-        CHECK_OV_STATUS(
-            ov_preprocess_prepostprocessor_build(ppp, &ov_ctx->new_model), ret);
-    }
-
-    CHECK_OV_STATUS(ov_core_compile_model(ov_ctx->core, ov_ctx->new_model,
-                                          "CPU", 0, &ov_ctx->compiled_model),
-                    ret);
-
-    CHECK_OV_STATUS(ov_compiled_model_create_infer_request(
-                        ov_ctx->compiled_model, &ov_ctx->infer_request),
-                    ret);
-
     /* install ov_tensor -> infer_request */
     CHECK_OV_STATUS(ov_infer_request_set_input_tensor_by_index(
-                        ov_ctx->infer_request, index, ov_ctx->input_tensor),
+                        exec->infer_request, index, input_tensor),
                     ret);
     ret = success;
-
 fail:
     if (ov_dims)
         os_free(ov_dims);
+    if (input_tensor)
+        ov_tensor_free(input_tensor);
     ov_shape_free(&input_shape);
-    if (ppp)
-        ov_preprocess_prepostprocessor_free(ppp);
-    if (input_info)
-        ov_preprocess_input_info_free(input_info);
-    if (input_tensor_info)
-        ov_preprocess_input_tensor_info_free(input_tensor_info);
-    if (input_layout)
-        ov_layout_free(input_layout);
-    if (input_process)
-        ov_preprocess_preprocess_steps_free(input_process);
-    if (p_input_model)
-        ov_preprocess_input_model_info_free(p_input_model);
-    if (model_layout)
-        ov_layout_free(model_layout);
-    if (output_info)
-        ov_preprocess_output_info_free(output_info);
-    if (output_tensor_info)
-        ov_preprocess_output_tensor_info_free(output_tensor_info);
 
     return ret;
 }
@@ -437,9 +426,14 @@ __attribute__((visibility("default"))) wasi_nn_error
 compute(void *ctx, graph_execution_context exec_ctx)
 {
     OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    struct OpenVINOExecutionContext *exec;
     wasi_nn_error ret = unsupported_operation;
 
-    CHECK_OV_STATUS(ov_infer_request_infer(ov_ctx->infer_request), ret);
+    if (exec_ctx >= ov_ctx->n_execution_contexts)
+        return runtime_error;
+    exec = &ov_ctx->execution_contexts[exec_ctx];
+
+    CHECK_OV_STATUS(ov_infer_request_infer(exec->infer_request), ret);
     ret = success;
 fail:
     return ret;
@@ -447,23 +441,33 @@ fail:
 
 __attribute__((visibility("default"))) wasi_nn_error
 get_output(void *ctx, graph_execution_context exec_ctx, uint32_t index,
-           tensor_data output_tensor, uint32_t *output_tensor_size)
+           tensor_data *output_tensor, uint32_t *output_tensor_size)
 {
     OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    struct OpenVINOExecutionContext *exec;
     wasi_nn_error ret = unsupported_operation;
     ov_tensor_t *ov_tensor = NULL;
     void *data = NULL;
     size_t byte_size = 0;
 
+    if (exec_ctx >= ov_ctx->n_execution_contexts)
+        return runtime_error;
+    exec = &ov_ctx->execution_contexts[exec_ctx];
+
     CHECK_OV_STATUS(ov_infer_request_get_output_tensor_by_index(
-                        ov_ctx->infer_request, index, &ov_tensor),
+                        exec->infer_request, index, &ov_tensor),
                     ret);
 
     CHECK_OV_STATUS(ov_tensor_get_byte_size(ov_tensor, &byte_size), ret);
 
+    if (byte_size > output_tensor->size) {
+        ret = too_large;
+        goto fail;
+    }
+
     CHECK_OV_STATUS(ov_tensor_data(ov_tensor, &data), ret);
 
-    memcpy(output_tensor, data, byte_size);
+    memcpy(output_tensor->buf, data, byte_size);
 
     *output_tensor_size = (uint32_t)byte_size;
 
@@ -517,27 +521,16 @@ __attribute__((visibility("default"))) wasi_nn_error
 deinit_backend(void *ctx)
 {
     OpenVINOContext *ov_ctx = (OpenVINOContext *)ctx;
+    unsigned int i;
 
     if (!ov_ctx)
         return invalid_argument;
 
-    if (ov_ctx->weight_data)
-        os_free(ov_ctx->weight_data);
+    for (i = 0; i < ov_ctx->n_execution_contexts; i++)
+        free_execution_context(&ov_ctx->execution_contexts[i]);
 
-    if (ov_ctx->weights_tensor)
-        ov_tensor_free(ov_ctx->weights_tensor);
-
-    if (ov_ctx->input_tensor)
-        ov_tensor_free(ov_ctx->input_tensor);
-
-    if (ov_ctx->infer_request)
-        ov_infer_request_free(ov_ctx->infer_request);
-
-    if (ov_ctx->compiled_model)
-        ov_compiled_model_free(ov_ctx->compiled_model);
-
-    if (ov_ctx->model)
-        ov_model_free(ov_ctx->model);
+    for (i = 0; i < ov_ctx->n_graphs; i++)
+        free_graph(&ov_ctx->graphs[i]);
 
     if (ov_ctx->core)
         ov_core_free(ov_ctx->core);
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.h b/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.h
index ea03a226f..0233568c0 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.h
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn_openvino.h
@@ -24,7 +24,7 @@ compute(void *ctx, graph_execution_context exec_ctx);
 
 __attribute__((visibility("default"))) wasi_nn_error
 get_output(void *ctx, graph_execution_context exec_ctx, uint32_t index,
-           tensor_data output_tensor, uint32_t *output_tensor_size);
+           tensor_data *output_tensor, uint32_t *output_tensor_size);
 
 __attribute__((visibility("default"))) wasi_nn_error
 init_backend(void **ctx);
@@ -32,4 +32,4 @@ init_backend(void **ctx);
 __attribute__((visibility("default"))) wasi_nn_error
 deinit_backend(void *ctx);
 
-#endif /* WASI_NN_OPENVINO_HPP */
\ No newline at end of file
+#endif /* WASI_NN_OPENVINO_HPP */
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn_private.h b/core/iwasm/libraries/wasi-nn/src/wasi_nn_private.h
index bb56f72fb..466f2cef4 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn_private.h
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn_private.h
@@ -9,7 +9,12 @@
 #include "wasi_nn_types.h"
 #include "wasm_export.h"
 
+#include "bh_platform.h"
+
 typedef struct {
+    korp_mutex lock;
+    bool busy;
+    bool is_backend_ctx_initialized;
     bool is_model_loaded;
     graph_encoding backend;
     void *backend_ctx;
@@ -27,7 +32,7 @@ typedef wasi_nn_error (*SET_INPUT)(void *, graph_execution_context, uint32_t,
                                    tensor *);
 typedef wasi_nn_error (*COMPUTE)(void *, graph_execution_context);
 typedef wasi_nn_error (*GET_OUTPUT)(void *, graph_execution_context, uint32_t,
-                                    tensor_data, uint32_t *);
+                                    tensor_data *, uint32_t *);
 /* wasi-nn general APIs */
 typedef wasi_nn_error (*BACKEND_INITIALIZE)(void **);
 typedef wasi_nn_error (*BACKEND_DEINITIALIZE)(void *);
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.cpp b/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.cpp
index 09e12f0d2..0ca323b70 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.cpp
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.cpp
@@ -281,6 +281,11 @@ set_input(void *tflite_ctx, graph_execution_context ctx, uint32_t index,
 {
     TFLiteContext *tfl_ctx = (TFLiteContext *)tflite_ctx;
 
+    if (input_tensor->type != fp32) {
+        NN_ERR_PRINTF("unsupported input tensor type %u", input_tensor->type);
+        return runtime_error;
+    }
+
     wasi_nn_error res;
     if (success != (res = is_valid_graph_execution_context(tfl_ctx, ctx)))
         return res;
@@ -319,7 +324,7 @@ set_input(void *tflite_ctx, graph_execution_context ctx, uint32_t index,
                 index);
 
         int size = model_tensor_size * sizeof(float);
-        bh_memcpy_s(it, size, input_tensor->data, size);
+        bh_memcpy_s(it, size, input_tensor->data.buf, size);
     }
     else { // TODO: Assuming uint8 quantized networks.
         TfLiteAffineQuantization *quant_info =
@@ -337,7 +342,7 @@ set_input(void *tflite_ctx, graph_execution_context ctx, uint32_t index,
         NN_DBG_PRINTF("input tensor: (scale, offset) = (%f, %f)", scale,
                       zero_point);
 
-        float *input_tensor_f = (float *)input_tensor->data;
+        float *input_tensor_f = (float *)input_tensor->data.buf;
         for (uint32_t i = 0; i < model_tensor_size; ++i) {
             it[i] = (uint8_t)(input_tensor_f[i] / scale + zero_point);
         }
@@ -361,7 +366,7 @@ compute(void *tflite_ctx, graph_execution_context ctx)
 
 __attribute__((visibility("default"))) wasi_nn_error
 get_output(void *tflite_ctx, 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)
 {
     TFLiteContext *tfl_ctx = (TFLiteContext *)tflite_ctx;
 
@@ -384,23 +389,34 @@ get_output(void *tflite_ctx, graph_execution_context ctx, uint32_t index,
         return too_large;
     }
 
-    uint32_t model_tensor_size = 1;
-    for (int i = 0; i < (int)tensor->dims->size; ++i)
-        model_tensor_size *= (uint32_t)tensor->dims->data[i];
-
-    if (*output_tensor_size < model_tensor_size) {
-        NN_ERR_PRINTF("Insufficient memory to copy tensor %d", index);
-        return too_large;
-    }
-
     if (tensor->quantization.type == kTfLiteNoQuantization) {
         NN_DBG_PRINTF("No quantization information");
-        float *ot =
-            tfl_ctx->interpreters[ctx].interpreter->typed_output_tensor<float>(
-                index);
-
-        int size = model_tensor_size * sizeof(float);
-        bh_memcpy_s(output_tensor, size, ot, size);
+#if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
+        if (output_tensor->size < tensor->bytes) {
+            NN_ERR_PRINTF("Insufficient memory to copy tensor %d", index);
+            return too_large;
+        }
+#else
+        /*
+         * for now, maintain the bug-to-bug compatibility with the old abi,
+         * where the size here is the number of fp32, not bytes.
+         */
+        if (output_tensor->size < tensor->bytes / sizeof(float)) {
+            NN_ERR_PRINTF("Insufficient memory to copy tensor %d", index);
+            return too_large;
+        }
+#endif
+        bh_memcpy_s(output_tensor->buf, output_tensor->size, tensor->data.data,
+                    tensor->bytes);
+#if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
+        *output_tensor_size = tensor->bytes;
+#else
+        /*
+         * for now, maintain the bug-to-bug compatibility with the old abi,
+         * where the size here is the number of fp32, not bytes.
+         */
+        *output_tensor_size = tensor->bytes / sizeof(float);
+#endif
     }
     else { // TODO: Assuming uint8 quantized networks.
         TfLiteAffineQuantization *quant_info =
@@ -409,6 +425,27 @@ get_output(void *tflite_ctx, graph_execution_context ctx, uint32_t index,
             NN_ERR_PRINTF("Quantization per channel is not supported");
             return runtime_error;
         }
+
+        uint32_t model_tensor_size = 1;
+        for (int i = 0; i < (int)tensor->dims->size; ++i)
+            model_tensor_size *= (uint32_t)tensor->dims->data[i];
+
+#if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
+        if (output_tensor->size / sizeof(float) < model_tensor_size) {
+            NN_ERR_PRINTF("Insufficient memory to copy tensor %d", index);
+            return too_large;
+        }
+#else
+        /*
+         * for now, maintain the bug-to-bug compatibility with the old abi,
+         * where the size here is the number of fp32, not bytes.
+         */
+        if (output_tensor->size < model_tensor_size) {
+            NN_ERR_PRINTF("Insufficient memory to copy tensor %d", index);
+            return too_large;
+        }
+#endif
+
         uint8_t *ot = tfl_ctx->interpreters[ctx]
                           .interpreter->typed_output_tensor<uint8_t>(index);
 
@@ -417,13 +454,22 @@ get_output(void *tflite_ctx, graph_execution_context ctx, uint32_t index,
         NN_DBG_PRINTF("output tensor: (scale, offset) = (%f, %f)", scale,
                       zero_point);
 
-        float *output_tensor_f = (float *)output_tensor;
+        float *output_tensor_f = (float *)output_tensor->buf;
         for (uint32_t i = 0; i < model_tensor_size; ++i) {
             output_tensor_f[i] = (ot[i] - zero_point) * scale;
         }
+
+#if WASM_ENABLE_WASI_EPHEMERAL_NN != 0
+        *output_tensor_size = model_tensor_size * sizeof(float);
+#else
+        /*
+         * for now, maintain the bug-to-bug compatibility with the old abi,
+         * where the size here is the number of fp32, not bytes.
+         */
+        *output_tensor_size = model_tensor_size;
+#endif
     }
 
-    *output_tensor_size = model_tensor_size;
     return success;
 }
 
diff --git a/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.hpp b/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.hpp
index d6e04ab0e..4ded6e407 100644
--- a/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.hpp
+++ b/core/iwasm/libraries/wasi-nn/src/wasi_nn_tensorflowlite.hpp
@@ -32,7 +32,7 @@ compute(void *tflite_ctx, graph_execution_context ctx);
 
 __attribute__((visibility("default"))) wasi_nn_error
 get_output(void *tflite_ctx, 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);
 
 __attribute__((visibility("default"))) wasi_nn_error
 init_backend(void **tflite_ctx);
diff --git a/core/iwasm/libraries/wasi-nn/test/build.sh b/core/iwasm/libraries/wasi-nn/test/build.sh
index dda400f16..14f6b9e09 100755
--- a/core/iwasm/libraries/wasi-nn/test/build.sh
+++ b/core/iwasm/libraries/wasi-nn/test/build.sh
@@ -3,6 +3,17 @@
 # Copyright (C) 2019 Intel Corporation.  All rights reserved.
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
+# on intel mac, this ends up with a lot of the following error.
+#
+#  AttributeError: 'Sequential' object has no attribute '_get_save_spec'.
+#
+# * "pip install tensorflow" installs tensorflow 2.16.2 on intel mac.
+#   (because it's the last version before tf deprecated the target.)
+# * keras 3 support in the version seems incomplete (thus the error)
+# * a workaround: use keras 2 as mentioned in:
+#   https://github.com/tensorflow/tensorflow/releases/tag/v2.16.1
+#   https://blog.tensorflow.org/2024/03/whats-new-in-tensorflow-216.html
+
 CURR_PATH=$(cd $(dirname $0) && pwd -P)
 
 # WASM application that uses WASI-NN
diff --git a/core/iwasm/libraries/wasi-nn/test/models/mult_outputs.py b/core/iwasm/libraries/wasi-nn/test/models/mult_outputs.py
index 98a50129c..8506e1806 100755
--- a/core/iwasm/libraries/wasi-nn/test/models/mult_outputs.py
+++ b/core/iwasm/libraries/wasi-nn/test/models/mult_outputs.py
@@ -3,7 +3,7 @@
 
 import tensorflow as tf
 import numpy as np
-from keras.layers import AveragePooling2D, Conv2D
+from tensorflow.keras.layers import AveragePooling2D, Conv2D
 
 from tensorflow.keras import Input, Model
 
diff --git a/core/shared/platform/esp-idf/espidf_platform.c b/core/shared/platform/esp-idf/espidf_platform.c
index d5f821d07..045c3a5f6 100644
--- a/core/shared/platform/esp-idf/espidf_platform.c
+++ b/core/shared/platform/esp-idf/espidf_platform.c
@@ -201,10 +201,20 @@ openat(int fd, const char *pathname, int flags, ...)
     int ret;
     char dir_path[DIR_PATH_LEN];
     char *full_path;
+    mode_t mode = 0;
+    bool has_mode = false;
+
+    if (flags & O_CREAT) {
+        va_list ap;
+        va_start(ap, flags);
+        mode = (mode_t)va_arg(ap, int);
+        va_end(ap);
+        has_mode = true;
+    }
 
     ret = fcntl(fd, F_GETPATH, dir_path);
     if (ret != 0) {
-        errno = -EINVAL;
+        errno = EINVAL;
         return -1;
     }
 
@@ -214,7 +224,7 @@ openat(int fd, const char *pathname, int flags, ...)
         return -1;
     }
 
-    new_fd = open(full_path, flags);
+    new_fd = has_mode ? open(full_path, flags, mode) : open(full_path, flags);
     free(full_path);
 
     return new_fd;
diff --git a/core/shared/utils/bh_vector.c b/core/shared/utils/bh_vector.c
index 352ce7192..7f0c74b9b 100644
--- a/core/shared/utils/bh_vector.c
+++ b/core/shared/utils/bh_vector.c
@@ -35,8 +35,8 @@ extend_vector(Vector *vector, size_t length)
     if (length <= vector->max_elems)
         return true;
 
-    if (length < vector->size_elem * 3 / 2)
-        length = vector->size_elem * 3 / 2;
+    if (length < vector->max_elems * 3 / 2)
+        length = vector->max_elems * 3 / 2;
 
     if (!(data = alloc_vector_data(length, vector->size_elem))) {
         return false;
@@ -194,12 +194,12 @@ bh_vector_append(Vector *vector, const void *elem_buf)
         goto just_return;
     }
 
-    /* make sure one more slot is used by the thread who allocas it */
+    /* make sure one more slot is used by the thread who allocates it */
     if (vector->lock)
         os_mutex_lock(vector->lock);
 
     if (!extend_vector(vector, vector->num_elems + 1)) {
-        LOG_ERROR("Append ector elem failed: extend vector failed.\n");
+        LOG_ERROR("Append vector elem failed: extend vector failed.\n");
         goto unlock_return;
     }
 
diff --git a/doc/build_wamr.md b/doc/build_wamr.md
index d8c64eb50..78e0711ec 100644
--- a/doc/build_wamr.md
+++ b/doc/build_wamr.md
@@ -102,6 +102,7 @@ cmake -DWAMR_BUILD_PLATFORM=linux -DWAMR_BUILD_TARGET=ARM
 
 ### **Enable lib wasi-nn**
 - **WAMR_BUILD_WASI_NN**=1/0, default to disable if not set
+> Note: WAMR_BUILD_WASI_NN without WAMR_BUILD_WASI_EPHEMERAL_NN is deprecated and will likely be removed in future versions of WAMR. Please consider to enable WAMR_BUILD_WASI_EPHEMERAL_NN as well.
 > Note: See [WASI-NN](../core/iwasm/libraries/wasi-nn) for more details.
 
 ### **Enable lib wasi-nn GPU mode**
@@ -113,7 +114,7 @@ cmake -DWAMR_BUILD_PLATFORM=linux -DWAMR_BUILD_TARGET=ARM
 - **WAMR_BUILD_WASI_NN_EXTERNAL_DELEGATE_PATH**=Path to the external delegate shared library (e.g. `libedgetpu.so.1.0` for Coral USB)
 
 ### **Enable lib wasi-nn with `wasi_ephemeral_nn` module support**
-- **WAMR_BUILD_WASI_EPHEMERAL_NN**=1/0, default to disable if not set
+- **WAMR_BUILD_WASI_EPHEMERAL_NN**=1/0, default to enable if not set
 
 ### **Disable boundary check with hardware trap**
 - **WAMR_DISABLE_HW_BOUND_CHECK**=1/0, default to enable if not set and supported by platform
@@ -364,4 +365,4 @@ For Valgrind, begin with the following configurations and add additional ones as
   -DWAMR_DISABLE_HW_BOUND_CHECK=0 \
   -DWAMR_DISABLE_WRITE_GS_BASE=0
   #...
-```
\ No newline at end of file
+```
diff --git a/language-bindings/python/wamr-api/README.md b/language-bindings/python/wamr-api/README.md
index b2ef1e105..58229da42 100644
--- a/language-bindings/python/wamr-api/README.md
+++ b/language-bindings/python/wamr-api/README.md
@@ -6,7 +6,7 @@
 
 ### Pre-requisites
 #### Install requirements
-Before proceeding it is necessary to make sure your Python environment is correctly configured. To do ths open a terminal session in this directory and perfom the following:
+Before proceeding it is necessary to make sure your Python environment is correctly configured. To do this open a terminal session in this directory and perform the following:
 
 
 ```shell
diff --git a/language-bindings/python/wasm-c-api/docs/design.md b/language-bindings/python/wasm-c-api/docs/design.md
index a952731d2..3478ad021 100644
--- a/language-bindings/python/wasm-c-api/docs/design.md
+++ b/language-bindings/python/wasm-c-api/docs/design.md
@@ -353,12 +353,12 @@ writable and needs to be copied into a ctype array.
 
 #### variable arguments
 
-A function with _variable arugments_ makes it hard to specify the required
+A function with _variable arguments_ makes it hard to specify the required
 argument types for the function prototype. It leaves us one way to call it
 directly without any arguments type checking.
 
 ```python
-libc.printf(b"Hello, an int %d, a float %f, a string %s\n", c_int(1), c_doulbe(3.14), "World!")
+libc.printf(b"Hello, an int %d, a float %f, a string %s\n", c_int(1), c_double(3.14), "World!")
 ```
 
 #### Use `c_bool` to represent `wasm_mutability_t `
@@ -373,7 +373,7 @@ libc.printf(b"Hello, an int %d, a float %f, a string %s\n", c_int(1), c_doulbe(3
 
 ### bindgen.py
 
-`bindge.py` is a tool to create WAMR python binding automatically. `binding.py`
+`bindgen.py` is a tool to create WAMR python binding automatically. `binding.py`
 is generated. We should avoid modification on it. Additional helpers should go
 to `ffi.py`.
 
diff --git a/product-mini/platforms/darwin/CMakeLists.txt b/product-mini/platforms/darwin/CMakeLists.txt
index cd7c8bc88..8d1994281 100644
--- a/product-mini/platforms/darwin/CMakeLists.txt
+++ b/product-mini/platforms/darwin/CMakeLists.txt
@@ -114,6 +114,12 @@ set (CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}")
 
 set (CMAKE_MACOSX_RPATH True)
 
+# if enable wasi-nn, both wasi-nn-backends and iwasm
+# need to use same WAMR (dynamic) libraries
+if (WAMR_BUILD_WASI_NN EQUAL 1)
+  set (BUILD_SHARED_LIBS ON)
+endif ()
+
 set (WAMR_ROOT_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../../..)
 
 include (${WAMR_ROOT_DIR}/build-scripts/runtime_lib.cmake)
diff --git a/product-mini/platforms/linux-sgx/enclave-sample/App/pal_api.h b/product-mini/platforms/linux-sgx/enclave-sample/App/pal_api.h
index 2db1fbb25..9b8077c04 100644
--- a/product-mini/platforms/linux-sgx/enclave-sample/App/pal_api.h
+++ b/product-mini/platforms/linux-sgx/enclave-sample/App/pal_api.h
@@ -79,7 +79,7 @@ struct wamr_pal_create_process_args {
     // Untrusted environment variable array pass to new process.
     //
     // The untrusted env vars to the command. And the last element of the array
-    // must be NULL to indicate the length of array.
+    // must be NULL to indicate the end of the array.
     //
     // Optional field.
     const char **env;
diff --git a/tests/benchmarks/README.md b/tests/benchmarks/README.md
index 2112829e0..95d85b1c0 100644
--- a/tests/benchmarks/README.md
+++ b/tests/benchmarks/README.md
@@ -8,6 +8,8 @@ Refer to the `README.md` under each folder for how to build and run the benchmar
 
 ## Install `llvm-profdata`
 
+> PS: the `llvm-profdata` vesion needs to be the same major version with llvm libraries used to build wamrc.
+
 The tool `llvm-profdata` is used when running the `test_pgo.sh` script under the benchmark folder. There are two ways to install it:
 
 1. Refer to https://apt.llvm.org/, e.g. in Ubuntu 20.04, add lines below to /etc/apt/source.list
@@ -18,19 +20,22 @@ deb-src http://apt.llvm.org/focal/ llvm-toolchain-focal main
 # 15
 deb http://apt.llvm.org/focal/ llvm-toolchain-focal-15 main
 deb-src http://apt.llvm.org/focal/ llvm-toolchain-focal-15 main
+# 18
+deb http://apt.llvm.org/focal/ llvm-toolchain-focal-18 main
+deb-src http://apt.llvm.org/focal/ llvm-toolchain-focal-18 main
 ```
 
 Then run `sudo apt update`, `sudo apt install llvm`. And after installing:
 
 ```bash
 cd /usr/bin
-sudo ln -s llvm-profdata-15 llvm-profdata
+sudo ln -s llvm-profdata-18 llvm-profdata
 ```
 
 2. Build manually
 
 ```bash
-git clone --depth 1 --branch release/15.x https://github.com/llvm/llvm-project.git
+git clone --depth 1 --branch release/18.x https://github.com/llvm/llvm-project.git
 cd llvm-project
 mkdir build && cd build
 cmake ../llvm \
diff --git a/tests/benchmarks/coremark/run.sh b/tests/benchmarks/coremark/run.sh
index 0d308bb68..b24486346 100755
--- a/tests/benchmarks/coremark/run.sh
+++ b/tests/benchmarks/coremark/run.sh
@@ -2,6 +2,7 @@
 
 # Copyright (C) 2019 Intel Corporation.  All rights reserved.
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+set -e
 
 PLATFORM=$(uname -s | tr A-Z a-z)
 
diff --git a/tests/benchmarks/coremark/test_pgo.sh b/tests/benchmarks/coremark/test_pgo.sh
index 1c631312e..25bed2c45 100755
--- a/tests/benchmarks/coremark/test_pgo.sh
+++ b/tests/benchmarks/coremark/test_pgo.sh
@@ -2,6 +2,7 @@
 
 # Copyright (C) 2019 Intel Corporation.  All rights reserved.
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+set -e
 
 PLATFORM=$(uname -s | tr A-Z a-z)
 
diff --git a/tests/fuzz/wasm-mutator-fuzz/server/app/main.py b/tests/fuzz/wasm-mutator-fuzz/server/app/main.py
index 620625dd3..c1ee3bfc4 100644
--- a/tests/fuzz/wasm-mutator-fuzz/server/app/main.py
+++ b/tests/fuzz/wasm-mutator-fuzz/server/app/main.py
@@ -72,7 +72,7 @@ def to_json(inst, cls):
 
 
 class Fuzzing(db.Model):
-    __tablename__ = 'fazzing_task'
+    __tablename__ = 'fuzzing_task'
     id = db.Column(db.Integer, autoincrement=True,
                    primary_key=True, nullable=False)
     repo = db.Column(db.String(200), nullable=False, default='')
@@ -96,7 +96,7 @@ class TaskError(db.Model):
     __tablename__ = 'task_error'
     id = db.Column(db.Integer, autoincrement=True,
                    primary_key=True, nullable=False)
-    fazzing_id = db.Column(db.Integer, db.ForeignKey("fazzing_task.id"))
+    fuzzing_id = db.Column(db.Integer, db.ForeignKey("fuzzing_task.id"))
     name = db.Column(db.String(200), nullable=False, default='')
     std_out = db.Column(db.Text, default='')
     data = db.Column(db.JSON)
@@ -119,9 +119,9 @@ def to_data(data):
 
 def error_count(data):
     error = len(TaskError.query.filter(
-        TaskError.fazzing_id == data.get('id'), TaskError.status.in_([1, 2])).all())
+        TaskError.fuzzing_id == data.get('id'), TaskError.status.in_([1, 2])).all())
     end_error = len(TaskError.query.filter(
-        TaskError.fazzing_id == data.get('id'), TaskError.status == 0).all())
+        TaskError.fuzzing_id == data.get('id'), TaskError.status == 0).all())
     data['error'] = error
     data['end_error'] = end_error
     return data
@@ -159,11 +159,11 @@ def show_fuzz_list():
     id = data.get('id')
     if id:
         all_error = TaskError.query.filter(
-            TaskError.fazzing_id == id).with_entities(TaskError.id, TaskError.fazzing_id,
+            TaskError.fuzzing_id == id).with_entities(TaskError.id, TaskError.fuzzing_id,
                                                       TaskError.create_time, TaskError.data,
                                                       TaskError.name, TaskError.status,
                                                       TaskError.update_time, TaskError.comment).order_by(TaskError.status.desc(), TaskError.update_time.desc(), TaskError.id.desc()).all()
-        data_message = [{'id': error['id'], "fuzzing_id": error['fazzing_id'],
+        data_message = [{'id': error['id'], "fuzzing_id": error['fuzzing_id'],
                          "name": error['name'], "data": error['data'],
                          'create_time': error['create_time'].strftime('%Y-%m-%d %H:%M:%S'),
                          'update_time': error['update_time'].strftime('%Y-%m-%d %H:%M:%S'),
@@ -204,7 +204,7 @@ def New_fuzzing():
         # curd.set_error_status_to(list(map(lambda x: x.id, error_list)), db)
         # Fuzzing.query.filter_by(id=fuzz.id).delete()
         fuzz.data = {'error': "Clone repo Error"}
-        db.commit()
+        db.session.commit()
         return jsonify({"status": 0, "result": "", "msg": "Clone repo Error"})
 
     wamr_path_parent = fuzz_dir.parent.parent
@@ -277,7 +277,7 @@ def scheduler_run_task():
 
     for fuzz in fuzz_query:
         all_error = TaskError.query.filter(
-            TaskError.fazzing_id == fuzz.id).with_entities(TaskError.name).all()
+            TaskError.fuzzing_id == fuzz.id).with_entities(TaskError.name).all()
         fuzz_cmd = wasm_mutator_dir / \
             'workspace' / f'build_{fuzz.id}'
         dir_list = filter(lambda x: x.startswith(
@@ -287,7 +287,7 @@ def scheduler_run_task():
         for dir in dir_list:
             cmd = f'cd {fuzz_cmd} && ./wasm_mutator_fuzz {dir}'
             status, resp = getstatusoutput(cmd)
-            task_error = TaskError(name=dir, std_out=resp, fazzing_id=fuzz.id,
+            task_error = TaskError(name=dir, std_out=resp, fuzzing_id=fuzz.id,
                                    create_time=datetime.utcnow() + timedelta(hours=8))
             db.session.add(task_error)
             db.session.commit()
@@ -312,7 +312,7 @@ def get_error_txt():
         return jsonify({"status": 0, "results": [], 'msg': "Error"})
     error = TaskError.query.get(id)
     fuzz_cmd = wasm_mutator_dir / \
-        'workspace' / f'build_{error.fazzing_id}'
+        'workspace' / f'build_{error.fuzzing_id}'
     file_cmd = fuzz_cmd / error.name
 
     response = send_file(file_cmd, as_attachment=True,
@@ -351,7 +351,7 @@ def get_cases_zip():
     with ZipFile(memory_file, "w", ZIP_DEFLATED) as zf:
         for task_error in task_query:
             fuzz_cmd = wasm_mutator_dir / \
-                'workspace' / f'build_{task_error.fazzing_id}'
+                'workspace' / f'build_{task_error.fuzzing_id}'
             file_cmd = fuzz_cmd / task_error.name
             zf.write(str(file_cmd), arcname=task_error.name)
     memory_file.seek(0)
@@ -399,7 +399,7 @@ def error_restart():
     if run_status:
         return jsonify({"status": 0, "results": [], 'msg': "There are already tasks in progress"})
     task_query = TaskError.query.filter(TaskError.id.in_(id_list)).all()
-    fuzzing_id = task_query[0].fazzing_id
+    fuzzing_id = task_query[0].fuzzing_id
     fuzz_cmd = wasm_mutator_dir / \
         'workspace' / f'build_{fuzzing_id}'
     restart_cmd = wasm_mutator_dir / \
@@ -412,7 +412,7 @@ def error_restart():
     if not Path(restart_cmd / 'wamr').exists():
         print('------ error: clone repo not folder exists ------')
         # fuzz.data = {'error': "Clone repo Error"}
-        db.commit()
+        db.session.commit()
         return jsonify({"status": 0, "result": "", "msg": "Clone repo Error"})
     wamr_path_parent = fuzz_dir.parent.parent
     wamr_path = wamr_path_parent / 'wamr'
diff --git a/tests/requirement-engineering/gc-aot/build_spec_interpreter.sh b/tests/requirement-engineering/gc-aot/build_spec_interpreter.sh
index 48d6343b1..0ecfe93f4 100755
--- a/tests/requirement-engineering/gc-aot/build_spec_interpreter.sh
+++ b/tests/requirement-engineering/gc-aot/build_spec_interpreter.sh
@@ -17,7 +17,7 @@ git apply ../../../wamr-test-suites/spec-test-script/gc_ignore_cases.patch
 # Set OCaml compiler environment
 eval $(opam config env)
 
-echo "compile the reference intepreter"
+echo "compile the reference interpreter"
 pushd interpreter
 make
-popd
\ No newline at end of file
+popd
diff --git a/tests/standalone/test-running-modes/test_c_embed_api_thoroughly.py b/tests/standalone/test-running-modes/test_c_embed_api_thoroughly.py
index 63e871e4a..8dcde7e74 100755
--- a/tests/standalone/test-running-modes/test_c_embed_api_thoroughly.py
+++ b/tests/standalone/test-running-modes/test_c_embed_api_thoroughly.py
@@ -9,7 +9,7 @@ import os
 from collections import OrderedDict
 
 
-def CLI_ARGS_GENREATOR(running_modes_supported: list[str]) -> list[str]:
+def CLI_ARGS_GENERATOR(running_modes_supported: list[str]) -> list[str]:
     res = []
     list_2d = [["--default-running-mode={} --module-running-mode={}".format(i, j)
                 for i in running_modes_supported] for j in running_modes_supported]
@@ -35,16 +35,16 @@ def main():
     ]
 
     # Python 3.7+: Dictionary iteration order is guaranteed to be in order of insertion.
-    # just to be safe, using orderreddict
+    # just to be safe, using OrderedDict
     # key: value -> compile mode, {"compile_flag": CMake compile flag, "iwasm_cli_args": array of CLI args tested}
     test_options = OrderedDict({
-        "INTERP": {"compile_flag": COMPILE_FLAGS[0], "cli_args": CLI_ARGS_GENREATOR(RUNNING_MODES[:1])},
-        "FAST_JIT": {"compile_flag": COMPILE_FLAGS[1], "cli_args": CLI_ARGS_GENREATOR(RUNNING_MODES[:2])},
+        "INTERP": {"compile_flag": COMPILE_FLAGS[0], "cli_args": CLI_ARGS_GENERATOR(RUNNING_MODES[:1])},
+        "FAST_JIT": {"compile_flag": COMPILE_FLAGS[1], "cli_args": CLI_ARGS_GENERATOR(RUNNING_MODES[:2])},
         "LLVM_JIT": {"compile_flag": COMPILE_FLAGS[2],
-                     "cli_args": CLI_ARGS_GENREATOR([RUNNING_MODES[0], RUNNING_MODES[2]])},
-        "MULTI_TIER_JIT": {"compile_flag": COMPILE_FLAGS[3], "cli_args": CLI_ARGS_GENREATOR(RUNNING_MODES)},
+                     "cli_args": CLI_ARGS_GENERATOR([RUNNING_MODES[0], RUNNING_MODES[2]])},
+        "MULTI_TIER_JIT": {"compile_flag": COMPILE_FLAGS[3], "cli_args": CLI_ARGS_GENERATOR(RUNNING_MODES)},
         "EAGER_JIT_WITH_BOTH_JIT": {"compile_flag": COMPILE_FLAGS[4],
-                                    "cli_args": CLI_ARGS_GENREATOR(RUNNING_MODES[:3])}
+                                    "cli_args": CLI_ARGS_GENERATOR(RUNNING_MODES[:3])}
     })
 
     build_cmd = "./build_c_embed.sh \"{build_flag}\""
diff --git a/tests/standalone/test-running-modes/test_iwasm_thoroughly.py b/tests/standalone/test-running-modes/test_iwasm_thoroughly.py
index a5af29101..3c631a6d5 100755
--- a/tests/standalone/test-running-modes/test_iwasm_thoroughly.py
+++ b/tests/standalone/test-running-modes/test_iwasm_thoroughly.py
@@ -29,7 +29,7 @@ def main():
     ]
 
     # Python 3.7+: Dictionary iteration order is guaranteed to be in order of insertion.
-    # just to be safe, using orderreddict
+    # just to be safe, using OrderedDict
     # key: value -> compile mode, {"compile_flag": CMake compile flag, "iwasm_cli_args": array of CLI args tested}
     test_options = OrderedDict({
         "INTERP": {"compile_flag": COMPILE_FLAGS[0], "iwasm_cli_args": IWASM_CLI_ARGS[:1]},
diff --git a/tests/unit/memory64/memory64_atomic_test.cc b/tests/unit/memory64/memory64_atomic_test.cc
index 2f9703890..3b2f679b5 100644
--- a/tests/unit/memory64/memory64_atomic_test.cc
+++ b/tests/unit/memory64/memory64_atomic_test.cc
@@ -31,7 +31,7 @@ class memory64_atomic_test_suite : public testing::TestWithParam<RunningMode>
         return true;
 
     fail:
-        if (!module)
+        if (module)
             wasm_runtime_unload(module);
 
         return false;
@@ -56,11 +56,13 @@ class memory64_atomic_test_suite : public testing::TestWithParam<RunningMode>
         if (exec_env)
             wasm_runtime_destroy_exec_env(exec_env);
         if (module_inst)
+            wasm_runtime_deinstantiate(module_inst);
+        if (module)
             wasm_runtime_unload(module);
         return false;
     }
 
-    void destory_exec_env()
+    void destroy_exec_env()
     {
         wasm_runtime_destroy_exec_env(exec_env);
         wasm_runtime_deinstantiate(module_inst);
@@ -109,7 +111,7 @@ class memory64_atomic_test_suite : public testing::TestWithParam<RunningMode>
     virtual void TearDown()
     {
         if (cleanup) {
-            destory_exec_env();
+            destroy_exec_env();
             wasm_runtime_destroy();
             cleanup = false;
         }
@@ -339,8 +341,8 @@ TEST_P(memory64_atomic_test_suite, atomic_opcodes_i64_rmw_cmpxchg)
     PUT_I64_TO_ADDR(wasm_argv + 2, 0x100F0E0D0C0B0A09);
     // new
     PUT_I64_TO_ADDR(wasm_argv + 4, 0xdeadcafebeefdead);
-    ASSERT_TRUE(wasm_runtime_call_wasm(exec_env, func_map["i64_atomic_rmw_cmpxchg"],
-                                       6, wasm_argv));
+    ASSERT_TRUE(wasm_runtime_call_wasm(
+        exec_env, func_map["i64_atomic_rmw_cmpxchg"], 6, wasm_argv));
     i64 = 0x100F0E0D0C0B0A09;
     ASSERT_EQ(i64, GET_U64_FROM_ADDR(wasm_argv));
 
diff --git a/tests/unit/memory64/memory64_test.cc b/tests/unit/memory64/memory64_test.cc
index af36f308c..2418d70ad 100644
--- a/tests/unit/memory64/memory64_test.cc
+++ b/tests/unit/memory64/memory64_test.cc
@@ -31,7 +31,7 @@ class memory64_test_suite : public testing::TestWithParam<RunningMode>
         return true;
 
     fail:
-        if (!module)
+        if (module)
             wasm_runtime_unload(module);
 
         return false;
@@ -56,11 +56,13 @@ class memory64_test_suite : public testing::TestWithParam<RunningMode>
         if (exec_env)
             wasm_runtime_destroy_exec_env(exec_env);
         if (module_inst)
+            wasm_runtime_deinstantiate(module_inst);
+        if (module)
             wasm_runtime_unload(module);
         return false;
     }
 
-    void destory_exec_env()
+    void destroy_exec_env()
     {
         wasm_runtime_destroy_exec_env(exec_env);
         wasm_runtime_deinstantiate(module_inst);
@@ -201,7 +203,7 @@ TEST_P(memory64_test_suite, memory_8GB)
     i64 = 0xbeefdead;
     ASSERT_EQ(i64, GET_U64_FROM_ADDR(wasm_argv));
 
-    destory_exec_env();
+    destroy_exec_env();
 }
 
 TEST_P(memory64_test_suite, mem64_from_clang)
@@ -228,7 +230,7 @@ TEST_P(memory64_test_suite, mem64_from_clang)
     i32 = 0x109;
     ASSERT_EQ(i32, wasm_argv[0]);
 
-    destory_exec_env();
+    destroy_exec_env();
 }
 
 INSTANTIATE_TEST_CASE_P(RunningMode, memory64_test_suite,
diff --git a/tests/unit/running-modes/wasm_running_modes_test.cc b/tests/unit/running-modes/wasm_running_modes_test.cc
index e18e64fb1..5f370dd64 100644
--- a/tests/unit/running-modes/wasm_running_modes_test.cc
+++ b/tests/unit/running-modes/wasm_running_modes_test.cc
@@ -21,7 +21,7 @@ std::string TEST_WASM1 = "/hello.wasm";
 std::string TEST_WASM2 = "/mytest.wasm";
 char *WASM_FILE_1;
 char *WASM_FILE_2;
-std::vector<RunningMode> running_mode_supportted = { Mode_Interp,
+std::vector<RunningMode> running_mode_supported = { Mode_Interp,
 #if WASM_ENABLE_FAST_JIT != 0
                                                      Mode_Fast_JIT,
 #endif
@@ -76,7 +76,7 @@ class wasm_running_modes_test_suite : public testing::TestWithParam<RunningMode>
         return true;
 
     fail:
-        if (!module)
+        if (module)
             wasm_runtime_unload(module);
 
         return false;
@@ -101,11 +101,13 @@ class wasm_running_modes_test_suite : public testing::TestWithParam<RunningMode>
         if (exec_env)
             wasm_runtime_destroy_exec_env(exec_env);
         if (module_inst)
+            wasm_runtime_deinstantiate(module_inst);
+        if (module)
             wasm_runtime_unload(module);
         return false;
     }
 
-    void destory_exec_env()
+    void destroy_exec_env()
     {
         wasm_runtime_destroy_exec_env(exec_env);
         wasm_runtime_deinstantiate(module_inst);
@@ -139,7 +141,7 @@ class wasm_running_modes_test_suite : public testing::TestWithParam<RunningMode>
         ASSERT_TRUE(ret);
         ASSERT_EQ(10, wasm_argv[0]);
 
-        destory_exec_env();
+        destroy_exec_env();
     }
 
     void run_wasm_complex(char *filename1, char *filename2,
@@ -168,7 +170,7 @@ class wasm_running_modes_test_suite : public testing::TestWithParam<RunningMode>
         ASSERT_TRUE(ret);
         ASSERT_EQ(10, wasm_argv[0]);
 
-        destory_exec_env();
+        destroy_exec_env();
 
         /* run wasm file 2 in running_mode */
         ret = load_wasm_file(filename2);
@@ -184,7 +186,7 @@ class wasm_running_modes_test_suite : public testing::TestWithParam<RunningMode>
         ret = wasm_runtime_call_wasm(exec_env, main, 2, wasm_argv);
         ASSERT_TRUE(ret);
 
-        destory_exec_env();
+        destroy_exec_env();
     }
 
   public:
@@ -246,7 +248,7 @@ TEST_F(wasm_running_modes_test_suite, wasm_runtime_is_running_mode_supported)
     // normal situation
     ASSERT_EQ(true, wasm_runtime_is_running_mode_supported(
                         static_cast<RunningMode>(Mode_Default)));
-    for (auto running_mode : running_mode_supportted) {
+    for (auto running_mode : running_mode_supported) {
         ASSERT_EQ(true, wasm_runtime_is_running_mode_supported(running_mode));
     }
 
@@ -264,7 +266,7 @@ TEST_F(wasm_running_modes_test_suite, wasm_runtime_set_default_running_mode)
     // normal situation: only set up
     ASSERT_EQ(true, wasm_runtime_set_default_running_mode(
                         static_cast<RunningMode>(Mode_Default)));
-    for (auto running_mode : running_mode_supportted) {
+    for (auto running_mode : running_mode_supported) {
         ASSERT_EQ(true, wasm_runtime_set_default_running_mode(running_mode));
     }
 
@@ -296,13 +298,13 @@ TEST_P(wasm_running_modes_test_suite,
        wasm_runtime_set_and_get_running_mode_complex)
 {
     RunningMode default_running_mode = GetParam();
-    for (auto running_mode : running_mode_supportted) {
+    for (auto running_mode : running_mode_supported) {
         run_wasm_complex(WASM_FILE_1, WASM_FILE_2, default_running_mode,
                          running_mode);
     }
 }
 
 INSTANTIATE_TEST_CASE_P(RunningMode, wasm_running_modes_test_suite,
-                        testing::ValuesIn(running_mode_supportted));
+                        testing::ValuesIn(running_mode_supported));
 
-}
\ No newline at end of file
+}
diff --git a/tests/wamr-test-suites/test_wamr.sh b/tests/wamr-test-suites/test_wamr.sh
index 695a25332..6f498653f 100755
--- a/tests/wamr-test-suites/test_wamr.sh
+++ b/tests/wamr-test-suites/test_wamr.sh
@@ -367,39 +367,39 @@ function sightglass_test()
 
 function setup_wabt()
 {
-    WABT_VERSION=1.0.37
+    # please sync with .github/actions/install-wasi-sdk-wabt/action.yml
+    case ${PLATFORM} in
+        cosmopolitan)
+            ;;
+        linux)
+            WABT_URL=https://github.com/WebAssembly/wabt/releases/download/1.0.37/wabt-1.0.37-ubuntu-20.04.tar.gz
+            WABT_VERSION=1.0.37
+            ;;
+        darwin)
+            WABT_URL=https://github.com/WebAssembly/wabt/releases/download/1.0.36/wabt-1.0.36-macos-12.tar.gz
+            WABT_VERSION=1.0.36
+            ;;
+        windows)
+            WABT_URL=https://github.com/WebAssembly/wabt/releases/download/1.0.37/wabt-1.0.37-windows.tar.gz
+            WABT_VERSION=1.0.37
+            ;;
+        *)
+            echo "wabt platform for ${PLATFORM} in unknown"
+            exit 1
+            ;;
+    esac
+
     if [ ${WABT_BINARY_RELEASE} == "YES" ]; then
         echo "download a binary release and install"
         local WAT2WASM=${WORK_DIR}/wabt/out/gcc/Release/wat2wasm
         if [ ! -f ${WAT2WASM} ]; then
-            case ${PLATFORM} in
-                cosmopolitan)
-                    ;;
-                linux)
-                    WABT_PLATFORM=ubuntu-20.04
-                    ;;
-                darwin)
-                    WABT_PLATFORM=macos-12
-                    ;;
-                windows)
-                    WABT_PLATFORM=windows
-                    ;;
-                *)
-                    echo "wabt platform for ${PLATFORM} in unknown"
-                    exit 1
-                    ;;
-            esac
-            if [ ! -f /tmp/wabt-${WABT_VERSION}-${WABT_PLATFORM}.tar.gz ]; then
-                curl -L \
-                    https://github.com/WebAssembly/wabt/releases/download/${WABT_VERSION}/wabt-${WABT_VERSION}-${WABT_PLATFORM}.tar.gz \
-                    -o /tmp/wabt-${WABT_VERSION}-${WABT_PLATFORM}.tar.gz
-            fi
+            pushd /tmp
+            wget -O wabt-tar.gz --progress=dot:giga ${WABT_URL}
+            tar xf wabt-tar.gz
+            popd
 
-            cd /tmp \
-            && tar zxf wabt-${WABT_VERSION}-${WABT_PLATFORM}.tar.gz \
-            && mkdir -p ${WORK_DIR}/wabt/out/gcc/Release/ \
-            && install wabt-${WABT_VERSION}/bin/* ${WORK_DIR}/wabt/out/gcc/Release/ \
-            && cd -
+            mkdir -p ${WORK_DIR}/wabt/out/gcc/Release/
+            cp /tmp/wabt-${WABT_VERSION}/bin/* ${WORK_DIR}/wabt/out/gcc/Release/
         fi
     else
         echo "download source code and compile and install"
@@ -420,7 +420,7 @@ function setup_wabt()
 
 function compile_reference_interpreter()
 {
-    echo "compile the reference intepreter"
+    echo "compile the reference interpreter"
     pushd interpreter
     make
     if [ $? -ne 0 ]
@@ -484,9 +484,9 @@ function spec_test()
         fi
 
         # As of version 1.0.36, wabt is still unable to correctly handle the GC proposal.
-        # 
+        #
         # $ $ /opt/wabt-1.0.36/bin/wast2json --enable-all ../spec/test/core/br_if.wast
-        # 
+        #
         # ../spec/test/core/br_if.wast:670:26: error: unexpected token "null", expected a numeric index or a name (e.g. 12 or $foo).
         #     (func $f (param (ref null $t)) (result funcref) (local.get 0))
         #
@@ -553,6 +553,9 @@ function spec_test()
     popd
     echo $(pwd)
 
+    #TODO: remove it when we can assume wabt is installed
+    # especially for CI Or there is installation script in the project
+    # that we can rely on
     setup_wabt
 
     ln -sf ${WORK_DIR}/../spec-test-script/all.py .
@@ -643,8 +646,8 @@ function spec_test()
 function wamr_compiler_test()
 {
     if [[ $1 != "aot" ]]; then
-        echo "WAMR compiler tests only support AOT mode"
-        exit 1
+        echo "WAMR compiler tests only support AOT mode, skip $1"
+        return 0
     fi
 
     echo  "Now start WAMR compiler tests"
@@ -899,51 +902,12 @@ function do_execute_in_running_mode()
 {
     local RUNNING_MODE="$1"
 
-    if [[ ${ENABLE_MULTI_MEMORY} -eq 1 ]]; then
-        if [[ "${RUNNING_MODE}" != "classic-interp" \
-                && "${RUNNING_MODE}" != "aot" ]]; then
-            echo "support multi-memory in classic-interp mode and aot mode"
-            return 0
-        fi
-    fi
+    # filter out uncompatible running mode based on targeting proposal features
+    # keep alpha order
 
-    if [[ ${ENABLE_MEMORY64} -eq 1 ]]; then
-        if [[ "${RUNNING_MODE}" != "classic-interp" \
-                && "${RUNNING_MODE}" != "aot" ]]; then
-            echo "support memory64(wasm64) in classic-interp mode and aot mode"
-            return 0
-        fi
-    fi
-
-    if [[ ${ENABLE_MULTI_MODULE} -eq 1 ]]; then
-        if [[ "${RUNNING_MODE}" != "classic-interp" \
-                && "${RUNNING_MODE}" != "fast-interp" \
-                && "${RUNNING_MODE}" != "aot" ]]; then
-            echo "support multi-module in both interp modes"
-            return 0
-        fi
-    fi
-
-    if [[ ${SGX_OPT} == "--sgx" ]]; then
-        if [[ "${RUNNING_MODE}" != "classic-interp" \
-                && "${RUNNING_MODE}" != "fast-interp" \
-                && "${RUNNING_MODE}" != "aot" \
-                && "${RUNNING_MODE}" != "fast-jit" ]]; then
-            echo "support sgx in both interp modes, fast-jit mode and aot mode"
-            return 0
-        fi
-    fi
-
-    if [[ ${ENABLE_SIMD} -eq 1 ]]; then
-        if [[ "${RUNNING_MODE}" != "jit" && "${RUNNING_MODE}" != "aot" && "${RUNNING_MODE}" != "fast-interp" ]]; then
-            echo "support simd in llvm-jit, aot and fast-interp mode"
-            return 0;
-        fi
-    fi
-
-    if [[ ${TARGET} == "X86_32" ]]; then
-        if [[ "${RUNNING_MODE}" == "jit" || "${RUNNING_MODE}" == "fast-jit" ]]; then
-            echo "both llvm-jit mode and fast-jit mode do not support X86_32 target"
+    if [[ ${ENABLE_EH} -eq 1 ]]; then
+        if [[ "${RUNNING_MODE}" != "classic-interp" ]]; then
+            echo "support exception handling in classic-interp"
             return 0;
         fi
     fi
@@ -958,9 +922,67 @@ function do_execute_in_running_mode()
         fi
     fi
 
-    if [[ ${ENABLE_EH} -eq 1 ]]; then
+    if [[ ${ENABLE_MEMORY64} -eq 1 ]]; then
+        if [[ "${RUNNING_MODE}" != "classic-interp" \
+                && "${RUNNING_MODE}" != "aot" ]]; then
+            echo "support memory64(wasm64) in classic-interp mode and aot mode"
+            return 0
+        fi
+    fi
+
+    if [[ ${ENABLE_MULTI_MEMORY} -eq 1 ]]; then
         if [[ "${RUNNING_MODE}" != "classic-interp" ]]; then
-            echo "support exception handling in classic-interp"
+            echo "support multi-memory in classic-interp mode mode"
+            return 0
+        fi
+    fi
+
+    if [[ ${ENABLE_MULTI_MODULE} -eq 1 ]]; then
+        if [[ "${RUNNING_MODE}" != "classic-interp" \
+                && "${RUNNING_MODE}" != "fast-interp" \
+                && "${RUNNING_MODE}" != "aot" ]]; then
+            echo "support multi-module in both interp modes"
+            return 0
+        fi
+    fi
+
+    if [[ ${ENABLE_SIMD} -eq 1 ]]; then
+        if [[ "${RUNNING_MODE}" != "jit" && "${RUNNING_MODE}" != "aot" && "${RUNNING_MODE}" != "fast-interp" ]]; then
+            echo "support simd in llvm-jit, aot and fast-interp mode"
+            return 0;
+        fi
+    fi
+
+    # filter out uncompatible running mode based on SGX support
+    if [[ ${SGX_OPT} == "--sgx" ]]; then
+        if [[ "${RUNNING_MODE}" != "classic-interp" \
+                && "${RUNNING_MODE}" != "fast-interp" \
+                && "${RUNNING_MODE}" != "aot" \
+                && "${RUNNING_MODE}" != "fast-jit" ]]; then
+            echo "support sgx in both interp modes, fast-jit mode and aot mode"
+            return 0
+        fi
+    fi
+
+    # filter out uncompatible running mode based on architecture
+    if [[ ${TARGET} == "X86_32" ]]; then
+        if [[ "${RUNNING_MODE}" == "jit" || "${RUNNING_MODE}" == "fast-jit" || "${RUNNING_MODE}" == "multi-tier-jit" ]]; then
+            echo "both llvm-jit, fast-jit and multi-tier-jit mode do not support X86_32 target"
+            return 0;
+        fi
+
+        if [[ ${ENABLE_MEMORY64} -eq 1 ]]; then
+            echo "memory64 does not support X86_32 target"
+            return 0;
+        fi
+
+        if [[ ${ENABLE_MULTI_MEMORY} -eq 1 ]]; then
+            echo "multi-memory does not support X86_32 target"
+            return 0;
+        fi
+
+        if [[ ${ENABLE_SIMD} -eq 1 ]]; then
+            echo "simd does not support X86_32 target"
             return 0;
         fi
     fi
diff --git a/wamr-compiler/main.c b/wamr-compiler/main.c
index 3c1c44ff7..410f4f668 100644
--- a/wamr-compiler/main.c
+++ b/wamr-compiler/main.c
@@ -137,9 +137,12 @@ print_help()
     printf("                              3 - Small code model\n");
     printf("  -sgx                      Generate code for SGX platform (Intel Software Guard Extensions)\n");
     printf("  --bounds-checks=1/0       Enable or disable the bounds checks for memory access:\n");
-    printf("                              by default it is disabled in all 64-bit platforms except SGX and\n");
-    printf("                              in these platforms runtime does bounds checks with hardware trap,\n");
-    printf("                              and by default it is enabled in all 32-bit platforms\n");
+    printf("                              This flag controls bounds checking with a software check. \n"); 
+    printf("                              On 64-bit platforms, it is disabled by default, using a hardware \n"); 
+    printf("                              trap if supported, except when SGX or memory64 is enabled,\n"); 
+    printf("                              which defaults to a software check.\n"); 
+    printf("                              On 32-bit platforms, the flag is enabled by default, using a software check\n");
+    printf("                              due to the lack of hardware support.\n"); 
     printf("                            CAVEAT: --bounds-checks=0 enables some optimizations\n");
     printf("                              which make the compiled AOT module incompatible\n");
     printf("                              with a runtime without the hardware bounds checks.\n");
diff --git a/wamr-wasi-extensions/CMakeLists.txt b/wamr-wasi-extensions/CMakeLists.txt
new file mode 100644
index 000000000..bed9f6780
--- /dev/null
+++ b/wamr-wasi-extensions/CMakeLists.txt
@@ -0,0 +1,9 @@
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+cmake_minimum_required (VERSION 3.14)
+
+project(wamr-wasi-extensions LANGUAGES C)
+
+add_subdirectory(nn)
+add_subdirectory(socket)
diff --git a/wamr-wasi-extensions/build_libs.sh b/wamr-wasi-extensions/build_libs.sh
new file mode 100755
index 000000000..97f2fe104
--- /dev/null
+++ b/wamr-wasi-extensions/build_libs.sh
@@ -0,0 +1,15 @@
+#! /bin/sh
+
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+set -e
+
+PREFIX=${1:-/tmp/wamr}
+WASI_SDK=${WASI_SDK:-/opt/wasi-sdk}
+
+cmake -B build-lib \
+-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK}/share/cmake/wasi-sdk.cmake \
+-DCMAKE_INSTALL_PREFIX=${PREFIX} \
+.
+cmake --build build-lib -t install
diff --git a/wamr-wasi-extensions/build_samples.sh b/wamr-wasi-extensions/build_samples.sh
new file mode 100755
index 000000000..5dfeeabf2
--- /dev/null
+++ b/wamr-wasi-extensions/build_samples.sh
@@ -0,0 +1,33 @@
+#! /bin/sh
+
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+set -e
+
+PREFIX=${1:-/tmp/wamr}
+WASI_SDK=${WASI_SDK:-/opt/wasi-sdk}
+
+cmake -B build-app-nn \
+-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK}/share/cmake/wasi-sdk.cmake \
+-DCMAKE_PREFIX_PATH=${PREFIX} \
+samples/nn
+cmake --build build-app-nn
+
+cmake -B build-app-nn-cli \
+-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK}/share/cmake/wasi-sdk.cmake \
+-DCMAKE_PREFIX_PATH=${PREFIX} \
+samples/nn-cli
+cmake --build build-app-nn-cli
+
+cmake -B build-app-socket-nslookup \
+-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK}/share/cmake/wasi-sdk-pthread.cmake \
+-DCMAKE_PREFIX_PATH=${PREFIX} \
+samples/socket-nslookup
+cmake --build build-app-socket-nslookup
+
+cmake -B build-app-socket-tcp-udp \
+-DCMAKE_TOOLCHAIN_FILE=${WASI_SDK}/share/cmake/wasi-sdk-pthread.cmake \
+-DCMAKE_PREFIX_PATH=${PREFIX} \
+samples/socket-tcp-udp
+cmake --build build-app-socket-tcp-udp
diff --git a/wamr-wasi-extensions/nn/CMakeLists.txt b/wamr-wasi-extensions/nn/CMakeLists.txt
new file mode 100644
index 000000000..27c22c4b8
--- /dev/null
+++ b/wamr-wasi-extensions/nn/CMakeLists.txt
@@ -0,0 +1,25 @@
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+add_library(wamr-wasi-nn INTERFACE)
+
+set(wasi_nn_header_dir ${CMAKE_CURRENT_SOURCE_DIR}/../../core/iwasm/libraries/wasi-nn/include)
+
+set(headers
+    ${wasi_nn_header_dir}/wasi_ephemeral_nn.h
+    ${wasi_nn_header_dir}/wasi_nn.h
+    ${wasi_nn_header_dir}/wasi_nn_types.h
+)
+
+set_property(TARGET wamr-wasi-nn PROPERTY PUBLIC_HEADER ${headers})
+
+target_include_directories(wamr-wasi-nn
+                           INTERFACE
+                           $<BUILD_INTERFACE:${wasi_nn_header_dir}>
+                           $<INSTALL_INTERFACE:include>)
+
+install(TARGETS wamr-wasi-nn
+        EXPORT wamr-wasi-nn-config
+        PUBLIC_HEADER DESTINATION include/wamr)
+install(EXPORT wamr-wasi-nn-config
+        DESTINATION lib/cmake/wamr-wasi-nn)
diff --git a/wamr-wasi-extensions/samples/nn-cli/CMakeLists.txt b/wamr-wasi-extensions/samples/nn-cli/CMakeLists.txt
new file mode 100644
index 000000000..df90b5ba4
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn-cli/CMakeLists.txt
@@ -0,0 +1,12 @@
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+cmake_minimum_required(VERSION 3.14)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_C_STANDARD_REQUIRED YES)
+
+project(nn-cli LANGUAGES C)
+add_executable(nn-cli main.c fileio.c map.c)
+find_package(wamr-wasi-nn REQUIRED)
+target_link_libraries(nn-cli wamr-wasi-nn)
diff --git a/wamr-wasi-extensions/samples/nn-cli/fileio.c b/wamr-wasi-extensions/samples/nn-cli/fileio.c
new file mode 100644
index 000000000..5d8163ba4
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn-cli/fileio.c
@@ -0,0 +1,73 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+/*
+ * modified copy-and-paste from:
+ * https://github.com/yamt/toywasm/blob/0eaad8cacd0cc7692946ff19b25994f106113be8/lib/fileio.c
+ */
+
+#include <sys/stat.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#include "fileio.h"
+
+int
+map_file(const char *path, void **pp, size_t *sizep)
+{
+    void *p;
+    size_t size;
+    ssize_t ssz;
+    int fd;
+    int ret;
+
+    fd = open(path, O_RDONLY);
+    if (fd == -1) {
+        ret = errno;
+        assert(ret != 0);
+        return ret;
+    }
+    struct stat st;
+    ret = fstat(fd, &st);
+    if (ret == -1) {
+        ret = errno;
+        assert(ret != 0);
+        close(fd);
+        return ret;
+    }
+    size = st.st_size;
+    if (size > 0) {
+        p = malloc(size);
+    }
+    else {
+        /* Avoid a confusing error */
+        p = malloc(1);
+    }
+    if (p == NULL) {
+        close(fd);
+        return ENOMEM;
+    }
+    ssz = read(fd, p, size);
+    if (ssz != size) {
+        ret = errno;
+        assert(ret != 0);
+        close(fd);
+        return ret;
+    }
+    close(fd);
+    *pp = p;
+    *sizep = size;
+    return 0;
+}
+
+void
+unmap_file(void *p, size_t sz)
+{
+    free(p);
+}
diff --git a/wamr-wasi-extensions/samples/nn-cli/fileio.h b/wamr-wasi-extensions/samples/nn-cli/fileio.h
new file mode 100644
index 000000000..e268222bc
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn-cli/fileio.h
@@ -0,0 +1,14 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+/*
+ * modified copy-and-paste from:
+ * https://github.com/yamt/toywasm/blob/0eaad8cacd0cc7692946ff19b25994f106113be8/lib/fileio.h
+ */
+
+int
+map_file(const char *filename, void **pp, size_t *szp);
+void
+unmap_file(void *p, size_t sz);
diff --git a/wamr-wasi-extensions/samples/nn-cli/main.c b/wamr-wasi-extensions/samples/nn-cli/main.c
new file mode 100644
index 000000000..0358158f3
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn-cli/main.c
@@ -0,0 +1,497 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include <sys/stat.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <getopt.h>
+#include <inttypes.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include <wamr/wasi_ephemeral_nn.h>
+
+#include "fileio.h"
+#include "map.h"
+
+static struct map graphs;
+static struct map contexts;
+
+static void
+load_graph(char *options)
+{
+    int target = wasi_ephemeral_nn_target_cpu;
+    int encoding = wasi_ephemeral_nn_encoding_openvino;
+    const char *id = "default";
+    wasi_ephemeral_nn_graph_builder *builders = NULL;
+    size_t nbuilders = 0;
+    enum {
+        opt_id,
+        opt_file,
+        opt_encoding,
+        opt_target,
+    };
+    static char *const keylistp[] = {
+        [opt_id] = "id",
+        [opt_file] = "file",
+        [opt_encoding] = "encoding",
+        [opt_target] = "target",
+        NULL,
+    };
+    while (*options) {
+        extern char *suboptarg;
+        char *value;
+        const char *saved = options;
+        switch (getsubopt(&options, keylistp, &value)) {
+            case opt_id:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                id = value;
+                break;
+            case opt_file:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                builders =
+                    realloc(builders, (nbuilders + 1) * sizeof(*builders));
+                if (builders == NULL) {
+                    exit(1);
+                }
+                wasi_ephemeral_nn_graph_builder *b = &builders[nbuilders++];
+                int ret = map_file(value, (void *)&b->buf, (void *)&b->size);
+                if (ret != 0) {
+                    fprintf(stderr, "map_file \"%s\" failed: %s\n", value,
+                            strerror(ret));
+                    exit(1);
+                }
+                break;
+            case opt_encoding:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                encoding = atoi(value);
+                break;
+            case opt_target:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                target = atoi(value);
+                break;
+            case -1:
+                fprintf(stderr, "unknown subopt %s\n", saved);
+                exit(2);
+        }
+    }
+
+    wasi_ephemeral_nn_error nnret;
+    wasi_ephemeral_nn_graph g;
+    nnret = wasi_ephemeral_nn_load(builders, nbuilders, encoding, target, &g);
+    size_t i;
+    for (i = 0; i < nbuilders; i++) {
+        wasi_ephemeral_nn_graph_builder *b = &builders[i];
+        unmap_file(b->buf, b->size);
+    }
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "load failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    map_set(&graphs, id, g);
+}
+
+static void
+init_execution_context(char *options)
+{
+    const char *id = "default";
+    const char *graph_id = "default";
+    enum {
+        opt_id,
+        opt_graph_id,
+    };
+    static char *const keylistp[] = {
+        [opt_id] = "id",
+        [opt_graph_id] = "graph-id",
+        NULL,
+    };
+    while (*options) {
+        extern char *suboptarg;
+        char *value;
+        const char *saved = options;
+        switch (getsubopt(&options, keylistp, &value)) {
+            case opt_id:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                id = value;
+                break;
+            case opt_graph_id:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                graph_id = value;
+                break;
+            case -1:
+                fprintf(stderr, "unknown subopt %s\n", saved);
+                exit(2);
+        }
+    }
+
+    wasi_ephemeral_nn_graph g = map_get(&graphs, graph_id);
+    wasi_ephemeral_nn_graph_execution_context c;
+    wasi_ephemeral_nn_error nnret;
+    nnret = wasi_ephemeral_nn_init_execution_context(g, &c);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "init_execution_context failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    map_set(&contexts, id, c);
+}
+
+static void
+set_input(char *options)
+{
+    int ret;
+    const char *context_id = "default";
+    uint32_t idx = 0;
+    wasi_ephemeral_nn_tensor tensor = {
+        .dimensions = { .buf = NULL, .size = 0, },
+        .type = wasi_ephemeral_nn_type_fp32,
+        .data = NULL,
+    };
+    void *buf = NULL;
+    size_t sz = 0;
+    enum {
+        opt_context_id,
+        opt_dim,
+        opt_type,
+        opt_idx,
+        opt_file,
+    };
+    static char *const keylistp[] = {
+        [opt_context_id] = "context-id",
+        [opt_dim] = "dim",
+        [opt_type] = "type",
+        [opt_idx] = "idx",
+        [opt_file] = "file",
+        NULL,
+    };
+    while (*options) {
+        extern char *suboptarg;
+        char *value;
+        const char *saved = options;
+        switch (getsubopt(&options, keylistp, &value)) {
+            case opt_context_id:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                context_id = value;
+                break;
+            case opt_dim:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                wasi_ephemeral_nn_tensor_dimensions *dims = &tensor.dimensions;
+
+                dims->buf =
+                    realloc(dims->buf, (dims->size + 1) * sizeof(*dims->buf));
+                if (dims->buf == NULL) {
+                    exit(1);
+                }
+                dims->buf[dims->size++] = atoi(value);
+                break;
+            case opt_type:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                tensor.type = atoi(value);
+                break;
+            case opt_file:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                if (buf != NULL) {
+                    fprintf(stderr, "duplicated tensor data\n");
+                    exit(2);
+                }
+                ret = map_file(value, &buf, &sz);
+                if (ret != 0) {
+                    fprintf(stderr, "map_file \"%s\" failed: %s\n", value,
+                            strerror(ret));
+                    exit(1);
+                }
+                break;
+            case opt_idx:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                idx = atoi(value);
+                break;
+            case -1:
+                fprintf(stderr, "unknown subopt %s\n", saved);
+                exit(2);
+        }
+    }
+
+    if (tensor.dimensions.size == 0) {
+        fprintf(stderr, "no dimension is given\n");
+        exit(2);
+    }
+    if (buf == NULL) {
+        fprintf(stderr, "no tensor is given\n");
+        exit(2);
+    }
+
+    /*
+     * REVISIT: we can check the tensor size against type/dimensions
+     * and warn the user if unexpected.
+     */
+
+    wasi_ephemeral_nn_error nnret;
+    wasi_ephemeral_nn_graph_execution_context c =
+        map_get(&contexts, context_id);
+    tensor.data.buf = buf;
+    tensor.data.size = sz;
+    nnret = wasi_ephemeral_nn_set_input(c, idx, &tensor);
+    unmap_file(buf, sz);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "set_input failed with %d\n", (int)nnret);
+        exit(1);
+    }
+}
+
+static void
+compute(char *options)
+{
+    const char *context_id = "default";
+    enum {
+        opt_context_id,
+    };
+    static char *const keylistp[] = {
+        [opt_context_id] = "context-id",
+        NULL,
+    };
+    while (*options) {
+        extern char *suboptarg;
+        char *value;
+        const char *saved = options;
+        switch (getsubopt(&options, keylistp, &value)) {
+            case opt_context_id:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                context_id = value;
+                break;
+            case -1:
+                fprintf(stderr, "unknown subopt %s\n", saved);
+                exit(2);
+        }
+    }
+
+    wasi_ephemeral_nn_graph_execution_context c =
+        map_get(&contexts, context_id);
+    wasi_ephemeral_nn_error nnret;
+    nnret = wasi_ephemeral_nn_compute(c);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "compute failed with %d\n", (int)nnret);
+        exit(1);
+    }
+}
+
+static void
+get_output(char *options)
+{
+    int ret;
+    const char *outfile = NULL;
+    const char *context_id = "default";
+    uint32_t idx = 0;
+    enum {
+        opt_context_id,
+        opt_idx,
+        opt_file,
+    };
+    static char *const keylistp[] = {
+        [opt_context_id] = "context-id",
+        [opt_idx] = "idx",
+        [opt_file] = "file",
+        NULL,
+    };
+    while (*options) {
+        extern char *suboptarg;
+        char *value;
+        const char *saved = options;
+        switch (getsubopt(&options, keylistp, &value)) {
+            case opt_context_id:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                context_id = value;
+                break;
+            case opt_file:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                outfile = value;
+                break;
+            case opt_idx:
+                if (value == NULL) {
+                    fprintf(stderr, "no value for %s\n", saved);
+                    exit(2);
+                }
+                idx = atoi(value);
+                break;
+            case -1:
+                fprintf(stderr, "unknown subopt %s\n", saved);
+                exit(2);
+        }
+    }
+
+    int outfd = -1;
+    if (outfile != NULL) {
+        outfd = open(outfile, O_CREAT | O_TRUNC | O_WRONLY);
+        if (outfd == -1) {
+            fprintf(stderr, "failed to open output file \"%s\": %s\n", outfile,
+                    strerror(errno));
+            exit(1);
+        }
+    }
+
+    wasi_ephemeral_nn_error nnret;
+    wasi_ephemeral_nn_graph_execution_context c =
+        map_get(&contexts, context_id);
+    void *resultbuf = NULL;
+    size_t resultbufsz = 256;
+    uint32_t resultsz;
+retry:
+    resultbuf = realloc(resultbuf, resultbufsz);
+    if (resultbuf == NULL) {
+        exit(1);
+    }
+    nnret =
+        wasi_ephemeral_nn_get_output(c, 0, resultbuf, resultbufsz, &resultsz);
+    if (nnret == wasi_ephemeral_nn_error_too_large) {
+        resultbufsz *= 2;
+        goto retry;
+    }
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "get_output failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    if (outfd != -1) {
+        ssize_t written = write(outfd, resultbuf, resultsz);
+        if (written == -1) {
+            fprintf(stderr, "failed to write: %s\n", strerror(errno));
+            exit(1);
+        }
+        if (written == -1) {
+            fprintf(stderr, "unexpetecd write length %zu (expected %zu)\n",
+                    written, (size_t)resultsz);
+            exit(1);
+        }
+        ret = close(outfd);
+        if (ret != 0) {
+            fprintf(stderr, "failed to close: %s\n", strerror(errno));
+            exit(1);
+        }
+    }
+    else {
+        fprintf(stderr, "WARNING: discarding %zu bytes output\n",
+                (size_t)resultsz);
+    }
+}
+
+enum longopt {
+    opt_load_graph = 0x100,
+    opt_init_execution_context,
+    opt_set_input,
+    opt_compute,
+    opt_get_output,
+};
+
+static const struct option longopts[] = {
+    {
+        "load-graph",
+        required_argument,
+        NULL,
+        opt_load_graph,
+    },
+    {
+        "init-execution-context",
+        optional_argument,
+        NULL,
+        opt_init_execution_context,
+    },
+    {
+        "set-input",
+        required_argument,
+        NULL,
+        opt_set_input,
+    },
+    {
+        "compute",
+        optional_argument,
+        NULL,
+        opt_compute,
+    },
+    {
+        "get-output",
+        optional_argument,
+        NULL,
+        opt_get_output,
+    },
+    {
+        NULL,
+        0,
+        NULL,
+        0,
+    },
+};
+
+int
+main(int argc, char **argv)
+{
+    extern char *optarg;
+    int ch;
+    int longidx;
+    while ((ch = getopt_long(argc, argv, "", longopts, &longidx)) != -1) {
+        switch (ch) {
+            case opt_load_graph:
+                load_graph(optarg);
+                break;
+            case opt_init_execution_context:
+                init_execution_context(optarg ? optarg : "");
+                break;
+            case opt_set_input:
+                set_input(optarg);
+                break;
+            case opt_compute:
+                compute(optarg ? optarg : "");
+                break;
+            case opt_get_output:
+                get_output(optarg ? optarg : "");
+                break;
+            default:
+                exit(2);
+        }
+    }
+    exit(0);
+}
diff --git a/wamr-wasi-extensions/samples/nn-cli/map.c b/wamr-wasi-extensions/samples/nn-cli/map.c
new file mode 100644
index 000000000..3ed817242
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn-cli/map.c
@@ -0,0 +1,58 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "map.h"
+
+static uintmax_t *
+map_find_slot(struct map *m, const char *name)
+{
+    size_t i;
+    for (i = 0; i < m->nentries; i++) {
+        if (!strcmp(m->entries[i].k, name)) {
+            return &m->entries[i].v;
+        }
+    }
+    return NULL;
+}
+
+static void
+map_append(struct map *m, const char *k, uintmax_t v)
+{
+    m->entries = realloc(m->entries, (m->nentries + 1) * sizeof(*m->entries));
+    if (m->entries == NULL) {
+        exit(1);
+    }
+    struct map_entry *e = &m->entries[m->nentries++];
+    e->k = k;
+    e->v = v;
+}
+
+void
+map_set(struct map *m, const char *k, uintmax_t v)
+{
+    uintmax_t *p = map_find_slot(m, k);
+    if (p != NULL) {
+        fprintf(stderr, "duplicated id \"%s\"\n", k);
+        exit(1);
+    }
+    map_append(m, k, v);
+}
+
+uintmax_t
+map_get(struct map *m, const char *k)
+{
+    uintmax_t *p = map_find_slot(m, k);
+    if (p == NULL) {
+        fprintf(stderr, "id \"%s\" not found\n", k);
+        exit(1);
+    }
+    return *p;
+}
diff --git a/wamr-wasi-extensions/samples/nn-cli/map.h b/wamr-wasi-extensions/samples/nn-cli/map.h
new file mode 100644
index 000000000..0059293c8
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn-cli/map.h
@@ -0,0 +1,19 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include <stdint.h>
+
+struct map {
+    struct map_entry {
+        const char *k;
+        uintmax_t v;
+    } * entries;
+    size_t nentries;
+};
+
+void
+map_set(struct map *m, const char *k, uintmax_t v);
+uintmax_t
+map_get(struct map *m, const char *k);
diff --git a/wamr-wasi-extensions/samples/nn/CMakeLists.txt b/wamr-wasi-extensions/samples/nn/CMakeLists.txt
new file mode 100644
index 000000000..59d607f64
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn/CMakeLists.txt
@@ -0,0 +1,13 @@
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+cmake_minimum_required(VERSION 3.14)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_C_STANDARD_REQUIRED YES)
+set(CMAKE_C_EXTENSIONS NO)
+
+project(nn-classification-openvino LANGUAGES C)
+add_executable(nn-classification-openvino "app.c")
+find_package(wamr-wasi-nn REQUIRED)
+target_link_libraries(nn-classification-openvino wamr-wasi-nn)
diff --git a/wamr-wasi-extensions/samples/nn/app.c b/wamr-wasi-extensions/samples/nn/app.c
new file mode 100644
index 000000000..15f637732
--- /dev/null
+++ b/wamr-wasi-extensions/samples/nn/app.c
@@ -0,0 +1,173 @@
+/*
+ * Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include <sys/stat.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include <wamr/wasi_ephemeral_nn.h>
+
+/*
+ * what this application does is basically same as:
+ * https://github.com/bytecodealliance/wasmtime/tree/efa236e58d09570baaf27865da33fb852fcf40a5/crates/wasi-nn/examples/classification-example
+ *
+ * map_file/unmap_file are copy-and-pasted from:
+ * https://github.com/yamt/toywasm/blob/0eaad8cacd0cc7692946ff19b25994f106113be8/lib/fileio.c
+ */
+
+int
+map_file(const char *path, void **pp, size_t *sizep)
+{
+    void *p;
+    size_t size;
+    ssize_t ssz;
+    int fd;
+    int ret;
+
+    fd = open(path, O_RDONLY);
+    if (fd == -1) {
+        ret = errno;
+        assert(ret != 0);
+        return ret;
+    }
+    struct stat st;
+    ret = fstat(fd, &st);
+    if (ret == -1) {
+        ret = errno;
+        assert(ret != 0);
+        close(fd);
+        return ret;
+    }
+    size = st.st_size;
+    if (size > 0) {
+        p = malloc(size);
+    }
+    else {
+        /* Avoid a confusing error */
+        p = malloc(1);
+    }
+    if (p == NULL) {
+        close(fd);
+        return ENOMEM;
+    }
+    ssz = read(fd, p, size);
+    if (ssz != size) {
+        ret = errno;
+        assert(ret != 0);
+        close(fd);
+        return ret;
+    }
+    close(fd);
+    *pp = p;
+    *sizep = size;
+    return 0;
+}
+
+void
+unmap_file(void *p, size_t sz)
+{
+    free(p);
+}
+
+static void
+print_result(const float *result, size_t sz)
+{
+    /*
+     * just dump the raw result.
+     * you can postprocess the output with eg. "sort -k2nr | head"
+     */
+    int i;
+    for (i = 0; i < sz / sizeof(float); i++) {
+        printf("%d %f\n", i, result[i]);
+    }
+}
+
+int
+main(int argc, char **argv)
+{
+    wasi_ephemeral_nn_error nnret;
+    int ret;
+    void *xml;
+    size_t xmlsz;
+    ret = map_file("fixture/model.xml", &xml, &xmlsz);
+    if (ret != 0) {
+        fprintf(stderr, "failed to load fixture/model.xml: %s\n",
+                strerror(ret));
+        exit(1);
+    }
+    void *weights;
+    size_t weightssz;
+    ret = map_file("fixture/model.bin", &weights, &weightssz);
+    if (ret != 0) {
+        fprintf(stderr, "failed to load fixture/model.bin: %s\n",
+                strerror(ret));
+        exit(1);
+    }
+    /* note: openvino takes two buffers, namely IR and weights */
+    wasi_ephemeral_nn_graph_builder builders[2] = { {
+                                                        .buf = xml,
+                                                        .size = xmlsz,
+                                                    },
+                                                    {
+                                                        .buf = weights,
+                                                        .size = weightssz,
+                                                    } };
+    wasi_ephemeral_nn_graph g;
+    nnret =
+        wasi_ephemeral_nn_load(builders, 2, wasi_ephemeral_nn_encoding_openvino,
+                               wasi_ephemeral_nn_target_cpu, &g);
+    unmap_file(xml, xmlsz);
+    unmap_file(weights, weightssz);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "load failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    wasi_ephemeral_nn_graph_execution_context ctx;
+    nnret = wasi_ephemeral_nn_init_execution_context(g, &ctx);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "init_execution_context failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    void *tensordata;
+    size_t tensordatasz;
+    ret = map_file("fixture/tensor.bgr", &tensordata, &tensordatasz);
+    if (ret != 0) {
+        fprintf(stderr, "failed to load fixture/tensor.bgr: %s\n",
+                strerror(ret));
+        exit(1);
+    }
+    wasi_ephemeral_nn_tensor tensor = {
+        .dimensions = { .buf = (uint32_t[]){1, 3, 224, 224,}, .size = 4, },
+        .type = wasi_ephemeral_nn_type_fp32,
+        .data.buf = tensordata,
+        .data.size = tensordatasz,
+    };
+    nnret = wasi_ephemeral_nn_set_input(ctx, 0, &tensor);
+    unmap_file(tensordata, tensordatasz);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "set_input failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    nnret = wasi_ephemeral_nn_compute(ctx);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "compute failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    float result[1001];
+    uint32_t resultsz;
+    nnret = wasi_ephemeral_nn_get_output(ctx, 0, (void *)result, sizeof(result),
+                                         &resultsz);
+    if (nnret != wasi_ephemeral_nn_error_success) {
+        fprintf(stderr, "get_output failed with %d\n", (int)nnret);
+        exit(1);
+    }
+    print_result(result, resultsz);
+}
diff --git a/wamr-wasi-extensions/samples/socket-nslookup/CMakeLists.txt b/wamr-wasi-extensions/samples/socket-nslookup/CMakeLists.txt
new file mode 100644
index 000000000..3c437524a
--- /dev/null
+++ b/wamr-wasi-extensions/samples/socket-nslookup/CMakeLists.txt
@@ -0,0 +1,11 @@
+cmake_minimum_required(VERSION 3.14)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_C_STANDARD_REQUIRED YES)
+set(CMAKE_C_EXTENSIONS NO)
+
+project(socket-nslookup LANGUAGES C)
+add_executable(socket-nslookup ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/iwasm/libraries/lib-socket/test/nslookup.c)
+find_package(wamr-wasi-socket REQUIRED)
+target_link_libraries(socket-nslookup wamr-wasi-socket)
+target_link_options(socket-nslookup PRIVATE -Wl,--max-memory=262144)
diff --git a/wamr-wasi-extensions/samples/socket-tcp-udp/CMakeLists.txt b/wamr-wasi-extensions/samples/socket-tcp-udp/CMakeLists.txt
new file mode 100644
index 000000000..d166094a7
--- /dev/null
+++ b/wamr-wasi-extensions/samples/socket-tcp-udp/CMakeLists.txt
@@ -0,0 +1,10 @@
+cmake_minimum_required(VERSION 3.14)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_C_STANDARD_REQUIRED YES)
+
+project(socket-tcp-udp LANGUAGES C)
+add_executable(socket-tcp-udp ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/iwasm/libraries/lib-socket/test/tcp_udp.c)
+find_package(wamr-wasi-socket REQUIRED)
+target_link_libraries(socket-tcp-udp wamr-wasi-socket)
+target_link_options(socket-tcp-udp PRIVATE -Wl,--max-memory=262144)
diff --git a/wamr-wasi-extensions/socket/CMakeLists.txt b/wamr-wasi-extensions/socket/CMakeLists.txt
new file mode 100644
index 000000000..74bbf31c4
--- /dev/null
+++ b/wamr-wasi-extensions/socket/CMakeLists.txt
@@ -0,0 +1,26 @@
+set(wasi_socket_dir ${CMAKE_CURRENT_SOURCE_DIR}/../../core/iwasm/libraries/lib-socket)
+set(wasi_socket_header_dir ${wasi_socket_dir}/inc)
+
+set(srcs ${wasi_socket_dir}/src/wasi/wasi_socket_ext.c)
+set(headers
+    ${wasi_socket_header_dir}/wasi_socket_ext.h
+)
+
+add_library(wamr-wasi-socket STATIC ${srcs})
+set_property(TARGET wamr-wasi-socket PROPERTY PUBLIC_HEADER ${headers})
+target_include_directories(wamr-wasi-socket
+                           PUBLIC
+                           $<BUILD_INTERFACE:${wasi_socket_header_dir}>
+                           $<INSTALL_INTERFACE:include>)
+
+# as this is a library, be extra conservative about wasm features
+# to improve compatibilities. as this particular library is just a
+# simple static stub, extra wasm features won't benefit us much anyway.
+# note that LLVM-19 enables reference-types by default.
+target_compile_options(wamr-wasi-socket PRIVATE -mno-reference-types)
+
+install(TARGETS wamr-wasi-socket
+        EXPORT wamr-wasi-socket-config
+        PUBLIC_HEADER DESTINATION include)
+install(EXPORT wamr-wasi-socket-config
+        DESTINATION lib/cmake/wamr-wasi-socket)
diff --git a/wamr-wasi-extensions/test.sh b/wamr-wasi-extensions/test.sh
new file mode 100755
index 000000000..1d366dfe9
--- /dev/null
+++ b/wamr-wasi-extensions/test.sh
@@ -0,0 +1,11 @@
+#! /bin/sh
+
+# Copyright (C) 2025 Midokura Japan KK.  All rights reserved.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+set -e
+
+PREFIX=${1:-/tmp/wamr}
+
+./build_libs.sh ${PREFIX}
+./build_samples.sh ${PREFIX}