merge master and resolve conflicts

ROCm · May 29, 2024 · 048bcda · 048bcda
2 parents 6dc309c + dc028dd
commit 048bcda
Show file tree

Hide file tree

Showing 976 changed files with 22,533 additions and 12,347 deletions.
diff --git a/.azuredevops/rocm-ci.yml b/.azuredevops/rocm-ci.yml
@@ -0,0 +1,43 @@
+resources:
+  repositories:
+  - repository: pipelines_repo
+    type: github
+    endpoint: ROCm
+    name: ROCm/ROCm
+
+variables:
+- group: common
+- template: /.azuredevops/variables-global.yml@pipelines_repo
+
+trigger:
+  batch: true
+  branches:
+    include:
+    - develop
+  paths:
+    exclude:
+    - .githooks
+    - .github
+    - docs
+    - '.*.y*ml'
+    - '*.md'
+    - Jenkinsfile
+    - LICENSE
+
+pr:
+  autoCancel: true
+  branches:
+    include:
+    - develop
+  paths:
+    exclude:
+    - .github
+    - docs
+    - '.*.y*ml'
+    - '*.md'
+    - Jenkinsfile
+    - LICENSE
+  drafts: false
+
+jobs:
+  - template: ${{ variables.CI_COMPONENT_PATH }}/AMDMIGraphX.yml@pipelines_repo
diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
@@ -356,7 +356,7 @@ jobs:
       with:
         python-version: 3.8
     - name: run License Check
-      run: python3 tools/check_stamped.py
+      run: python3 tools/check_stamped.py ${GITHUB_HEAD_REF:-${GITHUB_REF#refs/heads/}}
 
   linux:
 

diff --git a/.github/workflows/config.md b/.github/workflows/config.md
@@ -0,0 +1,29 @@
+#=====ROCM INFO=====
+ROCM_VERSION : '6.0.2'
+#default ROCm version to be used
+ROCM_BASE_IMAGE : 'rocm/dev-ubuntu-20.04'
+#base image from dockerhub to be used
+ROCM_BUILT_IMAGE : 'rocm-migraphx'
+#name of the docker image built upon ROCm base
+USE_NAVI : '0'
+#disable NAVI in image build
+OVERWRITE_EXISTING : 'true'
+#building new ROCm image overwrites old with same version
+
+#=====REPOS INFO=====
+ORGANIZATION_REPO : 'AMD'
+BENCHMARK_UTILS_REPO : 'ROCm/migraphx-benchmark-utils'
+PERFORMANCE_REPORTS_REPO : 'ROCm/migraphx-reports'
+PERFORMANCE_BACKUP_REPO : 'migraphx-benchmark/performance-backup'
+
+#=====PERFORMANCE SCRIPT PARAMETERS=====
+RESULTS_TO_COMPARE : '10'
+#number of previous performance results to be used in calculations
+CALCULATION_METHOD_FLAG : '-r'
+#calculation method used in reporting, -m for Max value; -s for Std dev; -r for Threshold file
+PERFORMANCE_TEST_TIMEOUT : '30m'
+#timeout for each model after which test is aborted
+
+#===== W A R N I N G =====
+#VARIABLE NAMES NOT TO BE CHANGED, VALUES ONLY!
+#VALUES MUST BE ENGLOSED IN SINGLE QUOTES!
diff --git a/.github/workflows/performance.yaml b/.github/workflows/performance.yaml
@@ -5,7 +5,7 @@ on:
     branches: [develop]
     types: [opened, synchronize, closed]
   schedule:
-    - cron: "0 6 * * 1-6"
+    - cron: "0 7 * * 1-6"
 
   workflow_dispatch:
     inputs:
@@ -47,18 +47,53 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  release:
+  get_config:
+    runs-on: ubuntu-latest
+    outputs:
+      rocm_version: ${{ steps.read_config.outputs.rocm_version }}
+      utils_repo: ${{ steps.read_config.outputs.utils_repo }}
+      reports_repo: ${{ steps.read_config.outputs.reports_repo }}
+      backup_repo: ${{ steps.read_config.outputs.backup_repo }}
+      repo_org: ${{ steps.read_config.outputs.repo_org }}
+      perf_number: ${{ steps.read_config.outputs.perf_number }}
+      perf_flag: ${{ steps.read_config.outputs.perf_flag }}
+      perf_timeout: ${{ steps.read_config.outputs.perf_timeout }}
+    steps:
+      - name: checkout
+        uses: actions/[email protected]
+      - name: read_config
+        id: read_config
+        run: |
+          ROCM_VERSION=$(grep 'ROCM_VERSION' .github/workflows/config.md | cut -d "'" -f2)
+          BENCHMARK_UTILS_REPO=$(grep 'BENCHMARK_UTILS_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          PERFORMANCE_REPORTS_REPO=$(grep 'PERFORMANCE_REPORTS_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          PERFORMANCE_BACKUP_REPO=$(grep 'PERFORMANCE_BACKUP_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          ORGANIZATION_REPO=$(grep 'ORGANIZATION_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          RESULTS_TO_COMPARE=$(grep 'RESULTS_TO_COMPARE' .github/workflows/config.md | cut -d "'" -f2)
+          CALCULATION_METHOD_FLAG=$(grep 'CALCULATION_METHOD_FLAG' .github/workflows/config.md | cut -d "'" -f2)
+          PERFORMANCE_TEST_TIMEOUT=$(grep 'PERFORMANCE_TEST_TIMEOUT' .github/workflows/config.md | cut -d "'" -f2)
+          echo "rocm_version=$ROCM_VERSION" >> $GITHUB_OUTPUT
+          echo "utils_repo=$BENCHMARK_UTILS_REPO" >> $GITHUB_OUTPUT
+          echo "reports_repo=$PERFORMANCE_REPORTS_REPO" >> $GITHUB_OUTPUT
+          echo "backup_repo=$PERFORMANCE_BACKUP_REPO" >> $GITHUB_OUTPUT
+          echo "repo_org=$ORGANIZATION_REPO" >> $GITHUB_OUTPUT
+          echo "perf_number=$RESULTS_TO_COMPARE" >> $GITHUB_OUTPUT
+          echo "perf_flag=$CALCULATION_METHOD_FLAG" >> $GITHUB_OUTPUT
+          echo "perf_timeout=$PERFORMANCE_TEST_TIMEOUT" >> $GITHUB_OUTPUT
+      
+  call_reusable:
+    needs: get_config
     uses: ROCm/migraphx-benchmark/.github/workflows/perf-test.yml@main
     with:
-      rocm_release: ${{ github.event.inputs.rocm_release || '6.0.2' }}
-      result_number: ${{ github.event.inputs.result_number || '10' }}
-      flags: ${{ github.event.inputs.flags || '-r' }}
-      performance_reports_repo: ${{ github.event.inputs.performance_reports_repo || 'ROCm/migraphx-reports' }}
-      performance_backup_repo: ${{ github.event.inputs.performance_backup_repo || 'migraphx-benchmark/performance-backup' }}
-      benchmark_utils_repo: ${{ github.event.inputs.benchmark_utils_repo || 'ROCm/migraphx-benchmark-utils' }}
-      organization: ${{ github.event.inputs.organization || 'AMD' }}
-      model_timeout: ${{ github.event.inputs.model_timeout || '30m' }}
+      rocm_release: ${{ github.event.inputs.rocm_release || needs.get_config.outputs.rocm_version }}
+      benchmark_utils_repo: ${{ github.event.inputs.benchmark_utils_repo || needs.get_config.outputs.utils_repo }}
+      performance_reports_repo: ${{ github.event.inputs.performance_reports_repo || needs.get_config.outputs.reports_repo }}
+      performance_backup_repo: ${{ github.event.inputs.performance_backup_repo || needs.get_config.outputs.backup_repo }}
+      organization: ${{ github.event.inputs.organization || needs.get_config.outputs.repo_org }}
+      result_number: ${{ github.event.inputs.result_number || needs.get_config.outputs.perf_number }}
+      flags: ${{ github.event.inputs.flags || needs.get_config.outputs.perf_flag }}
+      model_timeout: ${{ github.event.inputs.model_timeout || needs.get_config.outputs.perf_timeout }}
     secrets:
       gh_token: ${{ secrets.MIGRAPHX_BOT_TOKEN }}
       mail_user: ${{ secrets.MAIL_USERNAME }}
-      mail_pass: ${{ secrets.MAIL_PASSWORD }}
+      mail_pass: ${{ secrets.MAIL_PASSWORD }}
diff --git a/.github/workflows/sync_rocMLIR.yaml b/.github/workflows/sync_rocMLIR.yaml
@@ -0,0 +1,90 @@
+name: rocMLIR sync with extended accuracy
+
+on:
+  schedule:
+    - cron: '0 7 * * sun'
+  pull_request:
+    branches: [develop]
+    types: [synchronize, closed]
+  workflow_dispatch:
+    inputs:
+      rocm_release:
+        type: string
+        description: ROCm release version
+        required: true
+        default: '6.0.2'
+      base_image:
+        type: string
+        description: Base image for ROCm Docker build
+        required: true
+        default: 'rocm/dev-ubuntu-20.04'
+      docker_image:
+        type: string
+        description: Docker image name for rocm Docker build
+        required: true
+        default: 'rocm-migraphx'
+      build_navi:
+        type: string
+        description: Build navi number
+        required: true
+        default: '0'
+      benchmark_utils_repo:
+        type: string
+        description: Repository where benchmark utils are stored
+        required: true
+        default: 'ROCm/migraphx-benchmark-utils'
+      performance_reports_repo:
+        description: Repository where performance reports are stored
+        required: true
+        default: 'ROCm/migraphx-reports'
+      organization:
+        type: string
+        description: Organization based on which location of files will be different
+        required: true
+        default: 'AMD'
+
+jobs:
+  get_config:
+    runs-on: ubuntu-latest
+    outputs:
+      rocm_version: ${{ steps.read_config.outputs.rocm_version }}
+      rocm_base_image: ${{ steps.read_config.outputs.rocm_base_image }}
+      rocm_built_image: ${{ steps.read_config.outputs.rocm_built_image }}
+      use_navi: ${{ steps.read_config.outputs.use_navi }}
+      utils_repo: ${{ steps.read_config.outputs.utils_repo }}
+      reports_repo: ${{ steps.read_config.outputs.reports_repo }}
+      repo_org: ${{ steps.read_config.outputs.repo_org }}
+    steps:
+      - name: checkout
+        uses: actions/[email protected]
+      - name: read_config
+        id: read_config
+        run: |
+          ROCM_VERSION=$(grep 'ROCM_VERSION' .github/workflows/config.md | cut -d "'" -f2)
+          ROCM_BASE_IMAGE=$(grep 'ROCM_BASE_IMAGE' .github/workflows/config.md | cut -d "'" -f2)
+          ROCM_BUILT_IMAGE=$(grep 'ROCM_BUILT_IMAGE' .github/workflows/config.md | cut -d "'" -f2)
+          BENCHMARK_UTILS_REPO=$(grep 'BENCHMARK_UTILS_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          PERFORMANCE_REPORTS_REPO=$(grep 'PERFORMANCE_REPORTS_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          ORGANIZATION_REPO=$(grep 'ORGANIZATION_REPO' .github/workflows/config.md | cut -d "'" -f2)
+          USE_NAVI=$(grep 'USE_NAVI' .github/workflows/config.ymd | cut -d "'" -f2)
+          echo "rocm_version=$ROCM_VERSION" >> $GITHUB_OUTPUT
+          echo "rocm_base_image=$ROCM_BASE_IMAGE" >> $GITHUB_OUTPUT
+          echo "rocm_built_image=$ROCM_BUILT_IMAGE" >> $GITHUB_OUTPUT
+          echo "use_navi=$USE_NAVI" >> $GITHUB_OUTPUT
+          echo "utils_repo=$BENCHMARK_UTILS_REPO" >> $GITHUB_OUTPUT
+          echo "reports_repo=$PERFORMANCE_REPORTS_REPO" >> $GITHUB_OUTPUT
+          echo "repo_org=$ORGANIZATION_REPO" >> $GITHUB_OUTPUT
+           
+  call_reusable:
+    needs: get_config
+    uses: ROCm/migraphx-benchmark/.github/workflows/rocMLIR_sync.yml@main
+    with:
+      rocm_release: ${{ github.event.inputs.rocm_release || needs.get_config.outputs.rocm_version }}
+      base_image: ${{ github.event.inputs.base_image || needs.get_config.outputs.rocm_base_image }}
+      docker_image: ${{ github.event.inputs.docker_image || needs.get_config.outputs.rocm_built_image }}
+      build_navi: ${{ github.event.inputs.build_navi || needs.get_config.outputs.use_navi }}
+      benchmark_utils_repo: ${{ github.event.inputs.benchmark_utils_repo || needs.get_config.outputs.utils_repo }}
+      performance_reports_repo: ${{ github.event.inputs.performance_reports_repo || needs.get_config.outputs.reports_repo }}
+      organization: ${{ github.event.inputs.organization || needs.get_config.outputs.repo_org }}
+    secrets:
+      gh_token: ${{ secrets.MIGRAPHX_BOT_TOKEN }}
diff --git a/.github/workflows/weekly_master_sync.yaml b/.github/workflows/weekly_master_sync.yaml
@@ -0,0 +1,23 @@
+name: Master weekly sync
+
+on:
+  schedule:
+   - cron: '0 15 * * sun'
+  workflow_dispatch:
+
+jobs:
+  SyncAndMerge:
+    name: Sync master and merge develop
+    runs-on: ubuntu-latest
+    steps:
+
+      - uses: actions/[email protected]
+        with: 
+          ref: develop
+          fetch-depth: '0'
+
+      - name: Merge Fast Forward Only
+        run: |
+          git checkout master
+          git merge origin/develop --ff-only
+          git push origin HEAD
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -7,7 +7,7 @@ Full documentation for MIGraphX is available at
 
 ### Additions
 
-* Added FP8 support
+* Added beta version of FP8, functional, not performant
 * Created a dockerfile with MIGraphX+ONNX Runtime EP+Torch
 * Added support for the `Hardmax`, `DynamicQuantizeLinear`, `Qlinearconcat`, `Unique`, `QLinearAveragePool`, `QLinearSigmoid`, `QLinearLeakyRelu`, `QLinearMul`, `IsInf` operators
 * Created web site examples for `Whisper`, `Llama-2`, and `Stable Diffusion 2.1`

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -61,6 +61,8 @@ else()
 option(MIGRAPHX_ENABLE_PYTHON "Enable python bindings" ON)
 endif()
 
+option(MIGRAPHX_USE_ROCBLAS "Enable MIGraphX to use rocBLAS" ON)
+
 # By default build shared libraries
 option(BUILD_SHARED_LIBS "Create shared libraries" ON)
 
@@ -167,6 +169,9 @@ rocm_enable_clang_tidy(
         -cert-dcl51-cpp
         -cert-err33-c
         -cert-str34-c
+        # We seed random numbers with constants for reproducibility
+        -cert-msc32-c
+        -cert-msc51-cpp
         # Disable all alpha checks by default
         -clang-analyzer-alpha*
         # Enable some alpha checks
@@ -328,11 +333,24 @@ foreach(py_file ${backend_files})
     configure_file(${py_file} ${DEST_DIR}/lib/onnx_migraphx/. COPYONLY)
 endforeach(py_file)
 
+if(BUILD_ADDRESS_SANITIZER)
+    set(DEPENDS_HIP_RUNTIME "hip-runtime-amd-asan" )
+else()
+    set(DEPENDS_HIP_RUNTIME "hip-runtime-amd" )
+endif()
+
+if(MIGRAPHX_USE_ROCBLAS)
+    list(APPEND PACKAGE_DEPENDS rocblas)
+endif()
+
+rocm_package_add_deb_dependencies(SHARED_DEPENDS "hip-dev")
+rocm_package_add_rpm_dependencies(SHARED_DEPENDS "hip-devel")
+
 rocm_create_package(
     NAME MIGraphX
     DESCRIPTION "AMD's graph optimizer"
     MAINTAINER "AMDMIGraphX Maintainer <[email protected]>"
     LDCONFIG
     PTH
-    DEPENDS miopen-hip rocblas hip-rocclr hip-base half ${PACKAGE_DEPENDS}
+    DEPENDS miopen-hip ${DEPENDS_HIP_RUNTIME} half ${PACKAGE_DEPENDS}
 )
diff --git a/Dockerfile b/Dockerfile
@@ -6,15 +6,18 @@ ARG PREFIX=/usr/local
 RUN dpkg --add-architecture i386
 
 # Install rocm key
-RUN apt-get update && apt-get install -y gnupg2 --no-install-recommends curl && \
+RUN apt-get update && apt-get install -y software-properties-common 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/6.0.2/ focal main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/6.0.2/ jammy 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"
 
+# rocgdb doesn't work on 22.04, workaround by installing the older python packages that are in 20.04
+RUN add-apt-repository -y ppa:deadsnakes/ppa
+
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     apt-utils \
@@ -32,10 +35,10 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     python3 \
     python3-dev \
     python3-pip \
-    software-properties-common \
+    libpython3.8 \
     wget \
     rocm-device-libs \
-    hip-base \
+    hip-dev \
     libnuma-dev \
     miopen-hip \
     rocblas \

diff --git a/Jenkinsfile b/Jenkinsfile
@@ -155,13 +155,13 @@ rocmtest clang_debug: rocmnode('mi100+') { cmake_build ->
             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='${gpu_targets}'")
         }
     }
-}, ck_hiprtc: rocmnode('mi100+') { cmake_build ->
-    stage('CK hipRTC') {
-        withEnv(['MIGRAPHX_ENABLE_CK=1', 'MIGRAPHX_TUNE_CK=1', 'MIGRAPHX_DISABLE_MLIR=1']) {
-            def gpu_targets = getgputargets()
-            cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DMIGRAPHX_USE_HIPRTC=On -DGPU_TARGETS='${gpu_targets}'")
-        }
-    }
+//}, ck_hiprtc: rocmnode('mi100+') { cmake_build ->
+//   stage('CK hipRTC') {
+//        withEnv(['MIGRAPHX_ENABLE_CK=1', 'MIGRAPHX_TUNE_CK=1', 'MIGRAPHX_DISABLE_MLIR=1']) {
+//            def gpu_targets = getgputargets()
+//            cmake_build(flags: "-DCMAKE_BUILD_TYPE=release -DMIGRAPHX_USE_HIPRTC=On -DGPU_TARGETS='${gpu_targets}'")
+//        }
+//    }
 }, clang_asan: rocmnode('nogpu') { cmake_build ->
     stage('Clang ASAN') {
         def sanitizers = "undefined,address"