Merge remote-tracking branch 'origin/main' into sam2-image

NVIDIA · Nov 25, 2024 · 6324c61 · 6324c61
2 parents 14b53ca + c8ee99f
commit 6324c61
Show file tree

Hide file tree

Showing 194 changed files with 6,149 additions and 2,524 deletions.
diff --git a/.github/workflows/mlir-tensorrt-ci.yml b/.github/workflows/mlir-tensorrt-ci.yml
@@ -6,14 +6,18 @@ on:
       - main
     types: [synchronize, opened, reopened, ready_for_review]
     paths: ["mlir-tensorrt/**"]
+  push:
+    branches:
+      - main
+    paths: ["mlir-tensorrt/**"]
 
 env:
   DEFAULT_IMAGE: ghcr.io/nvidia/tensorrt-incubator/mlir-tensorrt:cuda12.5-ubuntu-llvm17
   REGISTRY: ghcr.io
 
 jobs:
-  mlir-tensorrt-tests:
-    if: github.event.pull_request.draft == false
+  mlir-tensorrt-test-pr:
+    if: github.event_name == 'pull_request' && github.event.pull_request.draft == false
     # `ubuntu-latest` is a CPU runner.
     # If selected, tests requiring GPU are not run.
     runs-on: ubuntu-latest
@@ -128,9 +132,183 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
           run: |
-            export CCACHE_BASEDIR="$PWD"
-            export CCACHE_DIR="$PWD/ccache"
-            export CCACHE_COMPILERCHECK=content
+            export CCACHE_DIR="/ccache"
+            export CCACHE_MAXSIZE=10G
+            ccache --zero-stats || true
+            ccache --show-stats || true
+
+            cd mlir-tensorrt
+            cat > build_and_test.sh <<EOF
+            #!/bin/bash
+            set -e
+
+            python3 -m pip install -r python/requirements-dev.txt
+
+            cmake -B ./build -S . -G Ninja \
+              -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+              -DMLIR_TRT_PACKAGE_CACHE_DIR=/.cache.cpm \
+              -DMLIR_TRT_ENABLE_ASSERTIONS=ON \
+              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=10.5 \
+              -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
+              -DMLIR_TRT_USE_LINKER=lld \
+              -DMLIR_EXECUTOR_ENABLE_GPU_INTEGRATION_TESTS=OFF
+
+            ninja -C build all
+
+            ninja -C build check-mlir-executor
+            ninja -C build check-mlir-tensorrt-dialect
+            ninja -C build check-mlir-tensorrt
+
+            cd ..
+            ccache --show-stats || true
+            EOF
+
+            bash build_and_test.sh
+
+      # Run LIT tests with TensorRT 10 & ASAN
+      - name: Run MLIR-TensorRT lit tests with TensorRT 10, ASAN enabled
+        uses: addnab/docker-run-action@v3
+        with:
+          image: ${{ env.DEFAULT_IMAGE }}
+          options: -v ${{ github.workspace }}/mlir-tensorrt:/mlir-tensorrt -v ${{ github.workspace }}/ccache:/ccache -v ${{ github.workspace }}/.ccache.cpm:/.ccache.cpm
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+          run: |
+            export CCACHE_DIR="/ccache"
+            export CCACHE_MAXSIZE=10G
+            ccache --zero-stats || true
+            ccache --show-stats || true
+
+            cd mlir-tensorrt
+            cat > build_and_test.sh <<EOF
+            #!/bin/bash
+            set -e
+
+            python3 -m pip install -r python/requirements-dev.txt
+
+            cmake -B ./build -S . -G Ninja \
+              -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+              -DMLIR_TRT_PACKAGE_CACHE_DIR=/.cache.cpm \
+              -DMLIR_TRT_ENABLE_ASSERTIONS=ON \
+              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=10.5 \
+              -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
+              -DMLIR_TRT_USE_LINKER=lld \
+              -DMLIR_EXECUTOR_ENABLE_GPU_INTEGRATION_TESTS=OFF \
+              -DENABLE_ASAN=ON
+
+            ninja -C build all
+
+            ninja -C build check-mlir-executor
+            ninja -C build check-mlir-tensorrt-dialect
+            ninja -C build check-mlir-tensorrt
+
+            cd ..
+            ccache --show-stats || true
+            EOF
+
+            bash build_and_test.sh
+
+      # Run LIT tests with TensorRT 8
+      - name: Run MLIR-TensorRT lit tests with TensorRT 8
+        uses: addnab/docker-run-action@v3
+        with:
+          image: ${{ env.DEFAULT_IMAGE }}
+          options: -v ${{ github.workspace }}/mlir-tensorrt:/mlir-tensorrt -v ${{ github.workspace }}/ccache:/ccache -v ${{ github.workspace }}/.ccache.cpm:/.ccache.cpm
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+          run: |
+            export CCACHE_DIR="/ccache"
+            export CCACHE_MAXSIZE=10G
+            ccache --zero-stats || true
+            ccache --show-stats || true
+
+            cd mlir-tensorrt
+            cat > build_and_test.sh <<EOF
+            #!/bin/bash
+            set -e
+
+            python3 -m pip install -r python/requirements-dev.txt
+
+            cmake -B ./build -S . -G Ninja \
+              -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+              -DMLIR_TRT_PACKAGE_CACHE_DIR=/.cache.cpm \
+              -DMLIR_TRT_ENABLE_ASSERTIONS=ON \
+              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=8.6.1.6 \
+              -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
+              -DMLIR_TRT_USE_LINKER=lld \
+              -DMLIR_EXECUTOR_ENABLE_GPU_INTEGRATION_TESTS=OFF
+
+            ninja -C build all
+
+            ninja -C build check-mlir-executor
+            ninja -C build check-mlir-tensorrt-dialect
+            ninja -C build check-mlir-tensorrt
+
+            cd ..
+            ccache --show-stats || true
+            EOF
+
+            bash build_and_test.sh
+
+  mlir-tensorrt-test-main:
+    if: github.event_name == 'push'
+    # `ubuntu-latest` is a CPU runner.
+    # If selected, tests requiring GPU are not run.
+    runs-on: ubuntu-latest
+
+    steps:
+      # Free some disk space, otherwise we get OOM error.
+      - name: Free disk space
+        run: |
+          sudo rm -rf \
+            /usr/share/dotnet "$AGENT_TOOLSDIRECTORY" /usr/local/lib/android /opt/ghc \
+            /usr/local/share/powershell /usr/share/swift /usr/local/.ghcup \
+            /usr/lib/jvm
+
+          sudo apt-get purge microsoft-edge-stable || true
+          sudo apt-get purge google-cloud-cli || true
+          sudo apt-get purge dotnet-sdk-* || true
+          sudo apt-get purge google-chrome-stable || true
+
+          sudo apt-get autoremove -y
+          sudo apt-get autoclean -y
+
+      # Value of `github.workspace` is /home/runner/work/{repo_name}/{repo-name}
+      # i.e. /home/runner/work/TensorRT-Incubator/TensorRT-Incubator in our case.
+      # After this action, repo is cloned inside above path.
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 5
+
+      # Create cache folders
+      - name: Create cache folder
+        run: |
+          mkdir -p ${{ github.workspace }}/ccache
+          mkdir -p ${{ github.workspace }}/.ccache.cpm
+
+      # Create cache action
+      - name: Create cache action
+        id: core-build-cache
+        uses: actions/cache@v4
+        with:
+          key: ${{ runner.os }}-mlir-tensorrt-core-build
+          path: |
+            ${{ github.workspace }}/ccache
+            ${{ github.workspace }}/.ccache.cpm
+
+      # Run LIT tests with TensorRT 10
+      - name: Run MLIR-TensorRT lit tests with TensorRT 10
+        uses: addnab/docker-run-action@v3
+        with:
+          image: ${{ env.DEFAULT_IMAGE }}
+          options: -v ${{ github.workspace }}/mlir-tensorrt:/mlir-tensorrt -v ${{ github.workspace }}/ccache:/ccache -v ${{ github.workspace }}/.ccache.cpm:/.ccache.cpm
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+          run: |
+            export CCACHE_DIR="/ccache"
             export CCACHE_MAXSIZE=10G
             ccache --zero-stats || true
             ccache --show-stats || true
@@ -144,9 +322,9 @@ jobs:
 
             cmake -B ./build -S . -G Ninja \
               -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-              -DMLIR_TRT_PACKAGE_CACHE_DIR=${PWD}/.cache.cpm \
+              -DMLIR_TRT_PACKAGE_CACHE_DIR=/.cache.cpm \
               -DMLIR_TRT_ENABLE_ASSERTIONS=ON \
-              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=10.2 \
+              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=10.5 \
               -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
               -DMLIR_TRT_USE_LINKER=lld \
               -DMLIR_EXECUTOR_ENABLE_GPU_INTEGRATION_TESTS=OFF
@@ -173,9 +351,7 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
           run: |
-            export CCACHE_BASEDIR="$PWD"
-            export CCACHE_DIR="$PWD/ccache"
-            export CCACHE_COMPILERCHECK=content
+            export CCACHE_DIR="/ccache"
             export CCACHE_MAXSIZE=10G
             ccache --zero-stats || true
             ccache --show-stats || true
@@ -189,9 +365,9 @@ jobs:
 
             cmake -B ./build -S . -G Ninja \
               -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-              -DMLIR_TRT_PACKAGE_CACHE_DIR=${PWD}/.cache.cpm \
+              -DMLIR_TRT_PACKAGE_CACHE_DIR=/.cache.cpm \
               -DMLIR_TRT_ENABLE_ASSERTIONS=ON \
-              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=10.2 \
+              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=10.5 \
               -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
               -DMLIR_TRT_USE_LINKER=lld \
               -DMLIR_EXECUTOR_ENABLE_GPU_INTEGRATION_TESTS=OFF \
@@ -209,8 +385,8 @@ jobs:
 
             bash build_and_test.sh
 
-      # Run LIT tests with TensorRT 9
-      - name: Run MLIR-TensorRT lit tests with TensorRT 9
+      # Run LIT tests with TensorRT 8
+      - name: Run MLIR-TensorRT lit tests with TensorRT 8
         uses: addnab/docker-run-action@v3
         with:
           image: ${{ env.DEFAULT_IMAGE }}
@@ -219,9 +395,7 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
           run: |
-            export CCACHE_BASEDIR="$PWD"
-            export CCACHE_DIR="$PWD/ccache"
-            export CCACHE_COMPILERCHECK=content
+            export CCACHE_DIR="/ccache"
             export CCACHE_MAXSIZE=10G
             ccache --zero-stats || true
             ccache --show-stats || true
@@ -235,9 +409,9 @@ jobs:
 
             cmake -B ./build -S . -G Ninja \
               -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-              -DMLIR_TRT_PACKAGE_CACHE_DIR=${PWD}/.cache.cpm \
+              -DMLIR_TRT_PACKAGE_CACHE_DIR=/.cache.cpm \
               -DMLIR_TRT_ENABLE_ASSERTIONS=ON \
-              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=9.2.0.5 \
+              -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION=8.6.1.6 \
               -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
               -DMLIR_TRT_USE_LINKER=lld \
               -DMLIR_EXECUTOR_ENABLE_GPU_INTEGRATION_TESTS=OFF

diff --git a/mlir-tensorrt/CMakeLists.txt b/mlir-tensorrt/CMakeLists.txt
@@ -55,7 +55,7 @@ mtrt_option(MLIR_TRT_ENABLE_EXECUTOR "Build the Executor dialect and MLIR-Tensor
 mtrt_option(MLIR_TRT_ENABLE_NCCL "Enable the NCCL runtime module" ON)
 
 set(MLIR_TRT_TENSORRT_DIR "" CACHE STRING "Path to TensorRT install directory")
-set(MLIR_TRT_DOWNLOAD_TENSORRT_VERSION "10.2" CACHE STRING
+set(MLIR_TRT_DOWNLOAD_TENSORRT_VERSION "10.5" CACHE STRING
    "Version of TensorRT to download and use. It overrides MLIR_TRT_TENSORRT_DIR.")
 set(MLIR_TRT_PACKAGE_CACHE_DIR "" CACHE STRING "Directory where to cache downloaded C++ packages")
 set(MLIR_TRT_USE_LINKER "" CACHE STRING "Specify a linker to use (e.g. LLD); this is just an alias for LLVM_USE_LINKER")

diff --git a/mlir-tensorrt/Version.cmake b/mlir-tensorrt/Version.cmake
@@ -1,6 +1,6 @@
 set(MLIR_TENSORRT_VERSION_MAJOR "0")
 set(MLIR_TENSORRT_VERSION_MINOR "1")
-set(MLIR_TENSORRT_VERSION_PATCH "36")
+set(MLIR_TENSORRT_VERSION_PATCH "37")
 set(MLIR_TENSORRT_VERSION
   "${MLIR_TENSORRT_VERSION_MAJOR}.${MLIR_TENSORRT_VERSION_MINOR}.${MLIR_TENSORRT_VERSION_PATCH}")
 
diff --git a/mlir-tensorrt/build_tools/cmake/Dependencies.cmake b/mlir-tensorrt/build_tools/cmake/Dependencies.cmake
@@ -86,11 +86,15 @@ function(download_tensorrt)
   if(ARG_VERSION VERSION_EQUAL "10.2")
     set(ARG_VERSION "10.2.0.19")
   endif()
+  # Canonicalize "10.5" version by setting it to the latest public TRT 10.5 version.
+  if(ARG_VERSION VERSION_EQUAL "10.5")
+    set(ARG_VERSION "10.5.0.18")
+  endif()
 
   set(downloadable_versions
-    "9.0.1.4" "9.1.0.4" "9.2.0.5"
+     "8.6.1.6" "9.0.1.4" "9.1.0.4" "9.2.0.5"
     "10.0.0.6" "10.1.0.27"
-    "10.2.0.19"
+    "10.2.0.19" "10.5.0.18"
   )
 
   if(NOT ARG_VERSION IN_LIST downloadable_versions)
@@ -100,6 +104,28 @@ function(download_tensorrt)
 
   set(TRT_VERSION "${ARG_VERSION}")
 
+  # Handle TensorRT 8 versions. These are publicly accessible download links.
+  if(ARG_VERSION VERSION_LESS 9.0.0 AND ARG_VERSION VERSION_GREATER 8.0.0)
+    string(REGEX MATCH "[0-9]+\\.[0-9]+\\.[0-9]+" trt_short_version ${ARG_VERSION})
+    set(CUDA_VERSION "12.0")
+    set(OS "linux")
+    EXECUTE_PROCESS(COMMAND uname -m
+                    COMMAND tr -d '\n'
+                    OUTPUT_VARIABLE ARCH)
+    if(ARCH STREQUAL "arm64")
+      set(ARCH "aarch64")
+      set(OS "Ubuntu-20.04")
+    elseif(ARCH STREQUAL "amd64")
+      set(ARCH "x86_64")
+      set(OS "Linux")
+    elseif(ARCH STREQUAL "aarch64")
+      set(OS "Ubuntu-20.04")
+    elseif(NOT (ARCH STREQUAL "x86_64"))
+      message(FATAL_ERROR "Direct download not available for architecture: ${ARCH}")
+    endif()
+    set(_url "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/${trt_short_version}/tars/TensorRT-${TRT_VERSION}.${OS}.${ARCH}-gnu.cuda-${CUDA_VERSION}.tar.gz") 
+  endif()
+
   # Handle TensorRT 9 versions. These are publicly accessible download links.
   if(ARG_VERSION VERSION_LESS 10.0.0 AND ARG_VERSION VERSION_GREATER 9.0.0)
     string(REGEX MATCH "[0-9]+\\.[0-9]+\\.[0-9]+" trt_short_version ${ARG_VERSION})
@@ -137,19 +163,23 @@ function(download_tensorrt)
     set(_url "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.2.0/tars/TensorRT-10.2.0.19.Linux.x86_64-gnu.cuda-12.5.tar.gz")
   endif()
 
+  if(ARG_VERSION VERSION_EQUAL 10.5.0.18)
+    set(_url "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.5.0/tars/TensorRT-10.5.0.18.Linux.x86_64-gnu.cuda-12.6.tar.gz")
+  endif()
+
   if(NOT _url)
     message(FATAL_ERROR "Could not determine TensorRT download URL")
   endif()
 
   message(STATUS "TensorRT Download URL: ${_url}")
 
   CPMAddPackage(
-    NAME TensorRT9
+    NAME TensorRT
     VERSION "${TRT_VERSION}"
     URL ${_url}
     DOWNLOAD_ONLY
   )
-  set("${ARG_OUT_VAR}" "${TensorRT9_SOURCE_DIR}" PARENT_SCOPE)
+  set("${ARG_OUT_VAR}" "${TensorRT_SOURCE_DIR}" PARENT_SCOPE)
 endfunction()
 
 #-------------------------------------------------------------------------------------