Merge branch 'develop' into add_parity_check_ci

.github/workflows/benchmark.yaml

@@ -7,7 +7,7 @@ jobs:
   benchmark:
     uses: ROCmSoftwarePlatform/actions/.github/workflows/benchmarks.yml@main
     with:
-      rocm_version: 5.2
+      rocm_version: 5.7
       script_repo: migraphx-benchmark/benchmark-utils
       result_path: /usr/share/migraphx/test-results
       result_repo: ROCmSoftwarePlatform/comparison-results

.github/workflows/ci.yaml

@@ -8,6 +8,12 @@ on:
       - master
       - 'release/**'
 
+env:
+  DOCKER_USER: ${{secrets.DOCKERHUB_USERID}}
+  DOCKER_TOKEN: ${{secrets.DOCKERHUB_TOKEN}}
+  DOCKER_IMAGE_UBUNTU: "rocm/migraphx-ci-ubuntu"
+  DOCKER_IMAGE_SLES: "rocm/migraphx-ci-sles"
+
 
 jobs:
   cancel:
@@ -17,22 +23,102 @@ jobs:
         uses: styfle/cancel-workflow-action@0.11.0
         with:
           access_token: ${{ github.token }}
-  tidy:
+
+  check_image:
+    name: Check if image exists in registry
+    runs-on: ubuntu-latest
+    outputs:
+      imageexists:  ${{ steps.check_image.outputs.imageexists }}
+      imagetag:  ${{ steps.image_hash.outputs.imagetag }}
+      imageexists_sles:  ${{ steps.check_image.outputs.imageexists_sles }}
+      imagetag_sles:  ${{ steps.image_hash.outputs.imagetag_sles }}
+
+    steps: 
+      - name: Checkout Code
+        uses: actions/checkout@v3
+
+      - name: Create Image Tag
+        id: image_hash
+        run: |
+          echo "imagetag=hip-clang-${{hashFiles('**/hip-clang.docker', '**/*requirements.txt', '**/install_prereqs.sh', '**/rbuild.ini')}}" >> $GITHUB_OUTPUT
+          echo "imagetag_sles=hip-clang-${{hashFiles('**/tools/docker/sles.docker', '**/*requirements.txt', '**/install_prereqs.sh', '**/rbuild.ini')}}" >> $GITHUB_OUTPUT
+
+      - name: Check if image is built already
+        id: check_image
+        env:
+          DOCKER_TAG_UBUNTU: ${{ steps.image_hash.outputs.imagetag }}
+          DOCKER_TAG_SLES: ${{ steps.image_hash.outputs.imagetag_sles }}
+        run: |
+          if [[ "$(docker manifest inspect $DOCKER_IMAGE_UBUNTU:$DOCKER_TAG_UBUNTU 2> /dev/null)" != "" ]]; then
+            echo "imageexists=true" >> $GITHUB_OUTPUT
+            echo "Image already exists, skip building available"
+          else
+            echo "imageexists=false" >> $GITHUB_OUTPUT
+            echo "Tag does not exist, build and publishing required"
+          fi
+
+          if [[ "$(docker manifest inspect $DOCKER_IMAGE_SLES:$DOCKER_TAG_SLES 2> /dev/null)" != "" ]]; then
+            echo "imageexists_sles=true" >> $GITHUB_OUTPUT
+            echo "SLES Image already exists, skip building available"
+          else
+            echo "imageexists_sles=false" >> $GITHUB_OUTPUT
+            echo "SLES Tag does not exist, build and publishing required"
+          fi
+
+
+  build_image:
+    name: Build image
     runs-on: ROCM-Ubuntu
+    needs: check_image
+    if: ${{ needs.check_image.outputs.imageexists != 'true' }}
     steps:
     - uses: actions/checkout@v3
 
-    # In this step, this action saves a list of existing images,
-    # the cache is created without them in the post run.
-    # It also restores the cache if it exists.
-    - name: Docker layer cache
-      uses: jpribyl/action-docker-layer-caching@v0.1.1
-      with:
-        key: docker-layer-caching-migraphx-${{hashFiles('hip-clang.docker', '**/*requirements.txt', '**/install_prereqs.sh', 'rbuild.ini')}}
-        restore-keys:
-          docker-layer-caching-migraphx-
-      # Ignore the failure of a step and avoid terminating the job.
-      continue-on-error: true
+    - name: Build and publish 
+      env:            
+        DOCKER_TAG_UBUNTU: ${{ needs.check_image.outputs.imagetag }}
+      run: |
+        # The TOKEN and USERID are github secrets, Action failures at this step
+        # can come from a PR from a fork changing a file which forces a rebuild
+        # Resolve by making an internal PR of the Forked PR
+        echo $DOCKER_TOKEN | docker login -u $DOCKER_USER --password-stdin
+
+        docker pull $DOCKER_IMAGE_UBUNTU:latest || true
+        docker build . --file hip-clang.docker --cache-from $DOCKER_IMAGE_UBUNTU:latest --tag $DOCKER_IMAGE_UBUNTU:$DOCKER_TAG_UBUNTU --tag $DOCKER_IMAGE_UBUNTU:latest;
+        docker push $DOCKER_IMAGE_UBUNTU:$DOCKER_TAG_UBUNTU;
+        docker push $DOCKER_IMAGE_UBUNTU:latest;
+
+  build_SLES_image:
+    name: Build SLES image
+    runs-on: ROCM-Ubuntu
+    needs: check_image
+    if: ${{ needs.check_image.outputs.imageexists_sles != 'true' }}
+    steps:
+    - uses: actions/checkout@v3
+    - name: Build and publish SLES
+      env:            
+        DOCKER_TAG_SLES: ${{ needs.check_image.outputs.imagetag_sles }}
+      run: |
+        # The TOKEN and USERID are github secrets, Action failures at this step
+        # can come from a PR from a fork changing a file wichi forces a rebuild
+        # Resolve by making an internal PR of the Forked PR
+        echo $DOCKER_TOKEN | docker login -u $DOCKER_USER --password-stdin
+
+        docker pull $DOCKER_IMAGE_SLES:latest || true
+        docker build . --file ./tools/docker/sles.docker --cache-from $DOCKER_IMAGE_SLES:latest --tag $DOCKER_IMAGE_SLES:$DOCKER_TAG_SLES --tag $DOCKER_IMAGE_SLES:latest;
+        docker push $DOCKER_IMAGE_SLES:$DOCKER_TAG_SLES;
+        docker push $DOCKER_IMAGE_SLES:latest;
+
+
+  tidy:
+    runs-on: ROCM-Ubuntu
+    needs: [ build_image, check_image ]
+    env:
+      DOCKER_TAG_UBUNTU: ${{ needs.check_image.outputs.imagetag }}
+
+    if: ${{ !cancelled() && (needs.build_image.result == 'success' || needs.build_image.result == 'skipped') }}
+    steps: 
+    - uses: actions/checkout@v3
 
     - name: Restore cache files for tidy
       uses: actions/cache/restore@v3 
@@ -42,12 +128,8 @@ jobs:
         key: tidy-cache-${{ github.ref }}
         restore-keys: tidy-cache-
 
-    - name: Build the Docker image
-      run: |
-        docker build . --file hip-clang.docker --tag migraphx
-
-    - name: Clang tidy
-      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data migraphx bash < {0}"
+    - name: Clang Tidy
+      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data $DOCKER_IMAGE_UBUNTU:$DOCKER_TAG_UBUNTU bash < {0}"
       run: |
         mkdir build
         cd build
@@ -60,6 +142,7 @@ jobs:
           -DROCM_ENABLE_GH_ANNOTATIONS=On \
           -DCLANG_TIDY_DEPEND_ON_TARGET=Off \
           -DCLANG_TIDY_CACHE=/data/tidy-cache \
+          -DGPU_TARGETS=gfx908 \
           ..
         make -j$(nproc) -k onnx-proto tf-proto tidy
 
@@ -84,21 +167,14 @@ jobs:
 
   cppcheck:
     runs-on: ROCM-Ubuntu
+    needs: [ build_image, check_image ]
+    env:
+      DOCKER_TAG_UBUNTU: ${{ needs.check_image.outputs.imagetag }}
+
+    if: ${{ !cancelled() && (needs.build_image.result == 'success' || needs.build_image.result == 'skipped') }}
     steps:
     - uses: actions/checkout@v3
 
-    # In this step, this action saves a list of existing images,
-    # the cache is created without them in the post run.
-    # It also restores the cache if it exists.
-    - name: Docker layer cache
-      uses: jpribyl/action-docker-layer-caching@v0.1.1
-      with:
-        key: docker-layer-caching-migraphx-${{hashFiles('hip-clang.docker', '**/*requirements.txt', '**/install_prereqs.sh', 'rbuild.ini')}}
-        restore-keys:
-          docker-layer-caching-migraphx-
-      # Ignore the failure of a step and avoid terminating the job.
-      continue-on-error: true
-
     - name: Restore cache files for cppcheck
       id: cppcheck_restore
       uses: actions/cache/restore@v3
@@ -107,11 +183,8 @@ jobs:
         key: cppcheck-cache-${{ hashFiles('cppcheck.rules', 'CMakeLists.txt') }}-${{ github.ref }}
         restore-keys: cppcheck-cache-${{ hashFiles('cppcheck.rules', 'CMakeLists.txt') }}-
 
-    - name: Build the Docker image
-      run: docker build . --file hip-clang.docker --tag migraphx
-
     - name: Cppcheck
-      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data migraphx bash < {0}"
+      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data $DOCKER_IMAGE_UBUNTU:$DOCKER_TAG_UBUNTU bash < {0}"
       run: |
         mkdir build
         cd build
@@ -119,6 +192,7 @@ jobs:
           -DCPPCHECK_BUILD_DIR=/data/cppcheck-cache \
           -DBUILD_DEV=On \
           -DROCM_ENABLE_GH_ANNOTATIONS=On \
+          -DGPU_TARGETS=gfx908 \
           ..
         make -j$(nproc) cppcheck
 
@@ -142,29 +216,31 @@ jobs:
 
 
   format:
-    runs-on: ROCM-Ubuntu
+    runs-on: ubuntu-latest
+    needs: [ build_image, check_image ]
+    env:
+      DOCKER_TAG_UBUNTU: ${{ needs.check_image.outputs.imagetag }}
+
+    if: ${{ !cancelled() && (needs.build_image.result == 'success' || needs.build_image.result == 'skipped') }}
     steps:
     - uses: actions/checkout@v3
       with:
         fetch-depth: 0
 
-    # In this step, this action saves a list of existing images,
-    # the cache is created without them in the post run.
-    # It also restores the cache if it exists.
-    - name: Docker layer cache
-      uses: jpribyl/action-docker-layer-caching@v0.1.1
-      with:
-        key: docker-layer-caching-migraphx-${{hashFiles('hip-clang.docker', '**/*requirements.txt', '**/install_prereqs.sh', 'rbuild.ini')}}
-        restore-keys:
-          docker-layer-caching-migraphx-
-      # Ignore the failure of a step and avoid terminating the job.
+    - name: Free space
+      uses: jlumbroso/free-disk-space@main
       continue-on-error: true
-
-    - name: Build the Docker image
-      run: docker build . --file hip-clang.docker --tag migraphx
+      with:
+        tool-cache: true
+        android: true
+        dotnet: true
+        haskell: true
+        large-packages: true
+        swap-storage: true
+        docker-images: true
 
     - name: Check formatting
-      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data migraphx bash < {0}"
+      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data $DOCKER_IMAGE_UBUNTU:$DOCKER_TAG_UBUNTU bash < {0}"
       run: |
         set -e
         git config --global --add safe.directory /data
@@ -172,26 +248,16 @@ jobs:
 
   sles:
     runs-on: ROCM-Ubuntu
+    needs: [ build_SLES_image, check_image ]
+    env:
+      DOCKER_TAG_SLES: ${{ needs.check_image.outputs.imagetag_sles }}
+    
+    if: ${{ !cancelled() && (needs.build_SLES_image.result == 'success' || needs.build_SLES_image.result == 'skipped') }}      
     steps:
     - uses: actions/checkout@v3
       with:
         fetch-depth: 0
 
-    # In this step, this action saves a list of existing images,
-    # the cache is created without them in the post run.
-    # It also restores the cache if it exists.
-    - name: Docker layer cache
-      uses: jpribyl/action-docker-layer-caching@v0.1.1
-      with:
-        key: docker-layer-caching-migraphx-sles-${{hashFiles('hip-clang.docker', '**/*requirements.txt', '**/install_prereqs.sh', 'rbuild.ini')}}
-        restore-keys:
-          docker-layer-caching-migraphx-sles-
-      # Ignore the failure of a step and avoid terminating the job.
-      continue-on-error: true
-
-    - name: Build the Docker image
-      run: docker build . --file tools/docker/sles.docker --tag migraphx-sles
-
     - name: Restore cache files for ccache
       uses: actions/cache/restore@v3 
       id: ccache_restore
@@ -201,7 +267,7 @@ jobs:
         restore-keys: ccache-sles-
 
     - name: Build migraphx
-      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data migraphx-sles bash < {0}"
+      shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data $DOCKER_IMAGE_SLES:$DOCKER_TAG_SLES bash < {0}"
       run: |
         set -e
         export CCACHE_COMPRESSLEVEL=10
@@ -216,6 +282,7 @@ jobs:
           -DBUILD_DEV=On \
           -DCMAKE_CXX_COMPILER_LAUNCHER=/usr/local/bin/ccache \
           -DCMAKE_C_COMPILER_LAUNCHER=/usr/local/bin/ccache \
+          -DGPU_TARGETS=gfx908 \
           ..
         make -j$(nproc) tests driver
 
@@ -242,6 +309,7 @@ jobs:
     steps:
     - name: Free space
       uses: jlumbroso/free-disk-space@main
+      continue-on-error: true
       with:
         tool-cache: true
         android: true
@@ -272,6 +340,7 @@ jobs:
     steps:
     - name: Free space
       uses: jlumbroso/free-disk-space@main
+      continue-on-error: true
       with:
         tool-cache: true
         android: true
@@ -312,6 +381,7 @@ jobs:
     steps:
     - name: Free space
       uses: jlumbroso/free-disk-space@main
+      continue-on-error: true
       with:
         tool-cache: true
         android: true
@@ -430,6 +500,7 @@ jobs:
     steps:
     - name: Free space
       uses: jlumbroso/free-disk-space@main
+      continue-on-error: true
       with:
         tool-cache: true
         android: true

.github/workflows/performance.yaml

@@ -12,7 +12,7 @@ on:
       rocm_release:
         description: ROCm Version
         required: true
-        default: '5.6'
+        default: '5.7'
       performance_reports_repo:
         description: Repository where performance reports are stored
         required: true
@@ -50,7 +50,7 @@ jobs:
   release:
     uses: ROCmSoftwarePlatform/migraphx-benchmark/.github/workflows/perf-test.yml@main
     with:
-      rocm_release: ${{ github.event.inputs.rocm_release || '5.6' }}
+      rocm_release: ${{ github.event.inputs.rocm_release || '5.7' }}
       result_number: ${{ github.event.inputs.result_number || '10' }}
       flags: ${{ github.event.inputs.flags || '-r' }}
       performance_reports_repo: ${{ github.event.inputs.performance_reports_repo || 'ROCmSoftwarePlatform/migraphx-reports' }}

.gitignore

@@ -80,3 +80,6 @@ docs/html
 cmake-build*/
 build*/
 
+# Recommended location to install rbuild dependencies from README.md
+depend
+

.readthedocs.yaml

@@ -9,6 +9,10 @@ sphinx:
 formats: [htmlzip]
 
 python:
-   version: "3.8"
    install:
    - requirements: docs/.sphinx/requirements.txt
+
+build:
+   os: ubuntu-20.04
+   tools:
+      python: "3.8"

CMakeLists.txt

@@ -162,6 +162,8 @@ rocm_enable_clang_tidy(
         -cppcoreguidelines-pro-type-vararg
         -cppcoreguidelines-special-member-functions
         -cppcoreguidelines-virtual-class-destructor
+        -cppcoreguidelines-avoid-capture-default-when-capturing-this
+        -cppcoreguidelines-rvalue-reference-param-not-moved
         -google-readability-*
         -google-runtime-int
         -google-runtime-references
@@ -262,6 +264,7 @@ rocm_enable_cppcheck(
 enable_testing()
 
 include(ROCMCreatePackage)
+include(ROCMTest)
 
 set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
@@ -269,6 +272,7 @@ set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/bin)
 add_subdirectory(src)
 add_subdirectory(docs)
 if(BUILD_TESTING)
+    rocm_enable_test_package(migraphx)
     add_subdirectory(test)
 endif()
 add_subdirectory(tools)

Dockerfile

@@ -10,7 +10,7 @@ RUN apt-get update && apt-get install -y gnupg2 --no-install-recommends curl &&
     curl -sL http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add -
 
 # Add rocm repository
-RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.6/ focal main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.7/ focal main > /etc/apt/sources.list.d/rocm.list'
 
 # From docs.amd.com for installing rocm. Needed to install properly
 RUN sh -c "echo 'Package: *\nPin: release o=repo.radeon.com\nPin-priority: 600' > /etc/apt/preferences.d/rocm-pin-600"

Jenkinsfile

@@ -107,11 +107,11 @@ rocmtest clang_debug: rocmnode('cdna') { cmake_build ->
     stage('hipRTC Debug') {
         def sanitizers = "undefined"
         def debug_flags = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize-recover=${sanitizers}"
-        cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}' -DMIGRAPHX_USE_HIPRTC=On", gpu_debug: true)
+        cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}' -DMIGRAPHX_USE_HIPRTC=On -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')", gpu_debug: true)
     }
-}, clang_release: rocmnode('cdna') { cmake_build ->
+}, clang_release: rocmnode('mi100+') { cmake_build ->
     stage('Hip Clang Release') {
-        cmake_build(flags: "-DCMAKE_BUILD_TYPE=release")
+        cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')")
         stash includes: 'build/*.deb', name: 'migraphx-package'
     }
 // }, hidden_symbols: rocmnode('cdna') { cmake_build ->
@@ -120,7 +120,7 @@ rocmtest clang_debug: rocmnode('cdna') { cmake_build ->
 //     }
 }, all_targets_debug : rocmnode('cdna') { cmake_build ->
     stage('All targets Release') {
-        cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DMIGRAPHX_ENABLE_GPU=On -DMIGRAPHX_ENABLE_CPU=On -DMIGRAPHX_ENABLE_FPGA=On") 
+        cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DMIGRAPHX_ENABLE_GPU=On -DMIGRAPHX_ENABLE_CPU=On -DMIGRAPHX_ENABLE_FPGA=On -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')") 
     }
 }, mlir_debug: rocmnode('cdna') { cmake_build ->
     stage('MLIR Debug') {
@@ -129,13 +129,13 @@ rocmtest clang_debug: rocmnode('cdna') { cmake_build ->
             // Note: the -fno-sanitize= is copied from upstream LLVM_UBSAN_FLAGS.
             def debug_flags_cxx = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize=vptr,function -fno-sanitize-recover=${sanitizers}"
             def debug_flags = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize=vptr -fno-sanitize-recover=${sanitizers}"
-            cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DMIGRAPHX_ENABLE_MLIR=On -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags_cxx}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}'")
+            cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DMIGRAPHX_ENABLE_MLIR=On -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags_cxx}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}' -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')")
         }
     }
 }, ck_release: rocmnode('mi100+') { cmake_build ->
     stage('CK Release') {
         withEnv(['MIGRAPHX_ENABLE_CK=1', 'MIGRAPHX_TUNE_CK=1']) {
-            cmake_build(flags: "-DCMAKE_BUILD_TYPE=release")
+            cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')")
         }
     }
 }, clang_asan: rocmnode('nogpu') { cmake_build ->
@@ -159,7 +159,6 @@ def onnxnode(name, body) {
     }
 }
 
-
 rocmtest onnxbuild: onnxnode('anygpu') { cmake_build ->
     stage("Onnxruntime Build and Install") {
         sh '''
@@ -167,6 +166,7 @@ rocmtest onnxbuild: onnxnode('anygpu') { cmake_build ->
             #ls -lR
             md5sum ./build/*.deb
             dpkg -i ./build/*.deb
+            env
             cd /onnxruntime && ./build_and_install_onnxrt.sh
         '''
         stash includes:'/onnxruntime/build/Linux/Release/dist/*.whl', name:'onnxruntime-wheel'
@@ -189,12 +189,14 @@ def onnxtestnode(name, body) {
 rocmtest onnxtests: onnxtestnode('anygpu') {cmake_build ->
     stage("Onnxruntime Unit tests") {
         sh '''
+            env
             cd /onnxruntime && ./test_onnxrt_unit_tests.sh 
         '''
     }
 }, onnx_parity: onnxtestnode('anygpu') {cmake_build ->
     stage("Parity tests Stable Pytorch") {
         sh '''
+            env
             cd /onnxruntime && ./test_onnxrt_parity_tests.sh 
         '''
     }

cmake/PythonModules.cmake

@@ -86,7 +86,7 @@ function(py_add_module NAME)
     )
 
 endfunction()
-set(PYTHON_SEARCH_VERSIONS 2.7 3.5 3.6 3.7 3.8 3.9 3.10)
+set(PYTHON_SEARCH_VERSIONS 3.5 3.6 3.7 3.8 3.9 3.10)
 set(PYTHON_DISABLE_VERSIONS "" CACHE STRING "")
 foreach(PYTHON_DISABLE_VERSION ${PYTHON_DISABLE_VERSIONS})
     list(REMOVE_ITEM PYTHON_SEARCH_VERSIONS ${PYTHON_DISABLE_VERSION})

dev-requirements.txt

@@ -26,5 +26,5 @@ facebook/zstd@v1.4.5 -X subdir -DCMAKE_DIR=build/cmake
 ccache@v4.1 -DENABLE_TESTING=OFF
 pcre,pfultz2/pcre@8.45 -H sha256:d6f7182602a775a7d500a0cedca6449af0400c6493951513046d17615ed0bf11
 danmar/cppcheck@bb2711c22a0be09efe7f1a8da3030876471026c8 -DHAVE_RULES=1 # 2.11
-RadeonOpenCompute/rocm-cmake@189d497ed185683154ae9766393b9a10ff21201f --build
+RadeonOpenCompute/rocm-cmake@5a34e72d9f113eb5d028e740c2def1f944619595 --build
 -f requirements.txt

docs/.sphinx/requirements.txt

@@ -21,7 +21,7 @@ charset-normalizer==3.1.0
     # via requests
 click==8.1.3
     # via sphinx-external-toc
-cryptography==41.0.3
+cryptography==41.0.4
     # via pyjwt
 deprecated==1.2.13
     # via pygithub
@@ -87,7 +87,7 @@ requests==2.28.2
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core>=0.20.0
+rocm-docs-core==0.24.2
     # via -r requirements.in
 smmap==5.0.0
     # via gitdb

docs/dev_intro.rst

@@ -131,7 +131,7 @@ In this case, we can create `argument <migraphx::argument>` objects directly fro
     std::vector<float> results_vector(64);
     result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
 
-    EXPECT(migraphx::verify::verify_range(results_vector, sol));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, sol));
 
 An `argument <migraphx::argument>` can handle memory buffers from either the GPU or the CPU.
 By default when running the `program <migraphx::program>`, buffers are allocated on the corresponding target.

docs/driver.rst

@@ -50,9 +50,17 @@ Runs reference and CPU or GPU implementations and checks outputs for consistency
 
 .. include:: ./driver/compile.rst
 
-.. option::  --tolerance [double]
+.. option::  --rms-tol [double]
 
-Tolerance for errors (Default: 80)
+Tolerance for RMS error (Default: 0.001)
+
+.. option::  --atol [double]
+
+Tolerance for elementwise absolute difference (Default: 0.001)
+
+.. option::  --rtol [double]
+
+Tolerance for elementwise relative difference (Default: 0.001)
 
 .. option::  -i, --per-instruction
 

examples/migraphx/migraphx_driver/README.md

@@ -55,7 +55,9 @@ See below for a comprehensive list of commands and option arguments, as well as
 | --exhaustive-tune                        | Enable exhaustive search to find fastest kernel           |
 | --fp16                                   | Quantize for fp16                                         |
 | --int8                                   | Quantize for int8                                         |
-| --tolerance                              | Tolerance for errors                                      |
+| --rms-tol                                | Tolerance for the RMS error (Default: 0.001)              |
+| --atol                                   | Tolerance for elementwise absolute difference (Default: 0.001) |
+| --rtol                                   | Tolerance for elementwise relative difference (Default: 0.001) |
 | --per-instruction \| -i                  | Verify each instruction                                   |
 | --reduce \| -r                           | Reduce program and verify                                 |
 | --iterations \| -n                       | Number of iterations to run for perf report               |

hip-clang.docker

@@ -6,7 +6,7 @@ ARG PREFIX=/usr/local
 RUN dpkg --add-architecture i386
 
 # Add rocm repository
-RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.6/ focal main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.7/ focal main > /etc/apt/sources.list.d/rocm.list'
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \

rbuild.ini

@@ -29,3 +29,12 @@ define =
     CMAKE_CXX_COMPILER_LAUNCHER=${deps_dir}/bin/ccache
     MIGRAPHX_ENABLE_CPU=On
     BUILD_DEV=On
+
+[cibuild]
+cxx = ${rocm_path}/llvm/bin/clang++
+cc = ${rocm_path}/llvm/bin/clang
+deps =
+    -f dev-requirements.txt
+define =
+    CMAKE_C_COMPILER_LAUNCHER=${deps_dir}/bin/ccache
+    CMAKE_CXX_COMPILER_LAUNCHER=${deps_dir}/bin/ccache

src/CMakeLists.txt

@@ -96,6 +96,7 @@ add_library(migraphx
     serialize.cpp
     shape.cpp
     simplify_algebra.cpp
+    simplify_dyn_ops.cpp
     simplify_reshapes.cpp
     split_single_dyn_dim.cpp
     target.cpp
@@ -196,6 +197,7 @@ register_migraphx_ops(
     reduce_sum
     relu
     reshape
+    reshape_lazy
     reverse
     rnn
     rnn_last_cell_output

src/auto_contiguous.cpp

@@ -25,7 +25,6 @@
 #include <migraphx/program.hpp>
 #include <migraphx/instruction.hpp>
 #include <migraphx/make_op.hpp>
-
 #include <migraphx/iterator_for.hpp>
 
 namespace migraphx {

src/driver/main.cpp

@@ -536,13 +536,19 @@ struct params : command<params>
 struct verify : command<verify>
 {
     compiler c;
-    double tolerance     = 80;
+    migraphx::verify::tolerance tols;
     bool per_instruction = false;
     bool reduce          = false;
     void parse(argument_parser& ap)
     {
         c.parse(ap);
-        ap(tolerance, {"--tolerance"}, ap.help("Tolerance for errors"));
+        ap(tols.rms_tol, {"--rms-tol"}, ap.help("Tolerance for the RMS error (Default: 0.001)"));
+        ap(tols.atol,
+           {"--atol"},
+           ap.help("Tolerance for the elementwise absolute difference (Default: 0.001)"));
+        ap(tols.rtol,
+           {"--rtol"},
+           ap.help("Tolerance for the elementwise relative difference (Default: 0.001)"));
         ap(per_instruction,
            {"-i", "--per-instruction"},
            ap.help("Verify each instruction"),
@@ -567,15 +573,15 @@ struct verify : command<verify>
 
         if(per_instruction)
         {
-            verify_instructions(p, t, c.co, quantize, tolerance);
+            verify_instructions(p, t, c.co, quantize, tols);
         }
         else if(reduce)
         {
-            verify_reduced_program(p, t, c.co, quantize, m, tolerance);
+            verify_reduced_program(p, t, c.co, quantize, m, tols);
         }
         else
         {
-            verify_program(c.l.file, p, t, c.co, quantize, m, tolerance);
+            verify_program(c.l.file, p, t, c.co, quantize, m, tols);
         }
     }
 };

src/driver/verify.cpp

@@ -30,6 +30,7 @@
 #include <migraphx/instruction.hpp>
 #include <migraphx/compile_options.hpp>
 #include <migraphx/quantization.hpp>
+#include <migraphx/ranges.hpp>
 
 namespace migraphx {
 namespace driver {
@@ -76,15 +77,25 @@ void verify_program(const std::string& name,
                     compile_options options,
                     precision quantize,
                     const parameter_map& inputs,
-                    double tolerance)
+                    verify::tolerance tols)
 {
-    auto x = run_ref(p, inputs);
-    auto y = run_target(p, t, options, quantize, inputs);
+    auto ref_outs    = run_ref(p, inputs);
+    auto target_outs = run_target(p, t, options, quantize, inputs);
 
-    std::size_t output_num = x.size();
+    std::size_t output_num = ref_outs.size();
     for(std::size_t i = 0; i < output_num; ++i)
     {
-        verify_args(name, x[i], y[i], tolerance);
+        if(ref_outs[i].get_shape().type() != target_outs[i].get_shape().type() or
+           ref_outs[i].get_shape().lens() != target_outs[i].get_shape().lens())
+        {
+            std::cout << "FAILED: " << name << std::endl;
+            std::cout << "Shape mismatch {" << ref_outs[i].get_shape() << "} != {"
+                      << target_outs[i].get_shape() << "}" << std::endl;
+        }
+        else
+        {
+            verify_args(name, target_outs[i], verify::expected{ref_outs[i]}, tols);
+        }
     }
 }
 
@@ -92,7 +103,7 @@ void verify_instructions(const program& prog,
                          const target& t,
                          compile_options options,
                          precision quantize,
-                         double tolerance)
+                         verify::tolerance tols)
 {
     const auto* mm_prog = prog.get_main_module();
     for(auto&& ins : (*mm_prog))
@@ -123,8 +134,7 @@ void verify_instructions(const program& prog,
         {
             std::cout << "Verify: " << ins.name() << std::endl;
             std::cout << p << std::endl;
-            verify_program(
-                ins.name(), p, t, options, quantize, create_param_map(p, false), tolerance);
+            verify_program(ins.name(), p, t, options, quantize, create_param_map(p, false), tols);
         }
         catch(...)
         {
@@ -140,14 +150,22 @@ void verify_reduced(program p,
                     compile_options options,
                     precision quantize,
                     const parameter_map& inputs,
-                    double tolerance)
+                    verify::tolerance tols)
 {
     auto* mm  = p.get_main_module();
-    auto last = std::prev(mm->end(), n + 1);
+    auto last = std::prev(mm->end(), n);
     mm->remove_instructions(last, mm->end());
     std::cout << "Verify: " << n << std::endl;
     std::cout << p << std::endl;
-    verify_program(std::to_string(n), p, t, options, quantize, inputs, tolerance);
+    try
+    {
+        verify_program(std::to_string(n), p, t, options, quantize, inputs, tols);
+    }
+    catch(const std::exception& e)
+    {
+        std::cout << "FAILED: " << n << std::endl;
+        std::cout << "Exception: " << e.what() << std::endl;
+    }
 }
 
 void verify_reduced_program(const program& p,
@@ -155,14 +173,20 @@ void verify_reduced_program(const program& p,
                             compile_options options,
                             precision quantize,
                             const parameter_map& inputs,
-                            double tolerance)
+                            verify::tolerance tols)
 {
     const auto* mm = p.get_main_module();
     auto n         = std::distance(mm->begin(), mm->end());
     std::cout << "Verify steps: " << n << std::endl;
-    for(std::size_t i = 0; i < n; i++)
+    for(std::size_t i = 1; i < n; i++)
     {
-        verify_reduced(p, i, t, options, quantize, inputs, tolerance);
+        auto last = std::prev(mm->end(), i + 1);
+        if(contains({"@literal", "@param"}, last->name()))
+        {
+            std::cout << "Skip: " << i << std::endl;
+            continue;
+        }
+        verify_reduced(p, i, t, options, quantize, inputs, tols);
     }
 }