Merge branch 'main' into bose_fx2trt_converters

Author: apbose

.circleci/config.yml

@@ -263,7 +263,7 @@ commands:
     parameters:
       torch-build:
         type: string
-        default: "2.0.0.dev20230219+cu117"
+        default: "2.1.0.dev20230314+cu117"
       torch-build-index:
         type: string
         default: "https://download.pytorch.org/whl/nightly/cu117"
@@ -524,15 +524,47 @@ commands:
       - store_artifacts:
           path: /tmp/testlogs
 
-  test-fx_converters:
-    description: "Test the fx converters"
+  test-fx_converters_acc:
+    description: "Test the fx acc converters"
     steps:
       - run:
           name: Run FX converter tests
           command: |
             cd py/torch_tensorrt/fx/test
-            pushd converters/
-            pytest --junitxml=/tmp/artifacts/test_results/fx/converters/test_results.xml
+            pushd converters/acc_op/
+            pytest --junitxml=/tmp/artifacts/test_results/fx/converters/acc_op/test_results.xml
+            popd
+
+      - store_test_results:
+          path: /tmp/artifacts
+      - store_artifacts:
+          path: /tmp/testlogs
+
+  test-fx_converters_aten:
+    description: "Test the fx aten converters"
+    steps:
+      - run:
+          name: Run FX converter tests
+          command: |
+            cd py/torch_tensorrt/fx/test
+            pushd converters/aten_op/
+            pytest --junitxml=/tmp/artifacts/test_results/fx/converters/aten_op/test_results.xml
+            popd
+
+      - store_test_results:
+          path: /tmp/artifacts
+      - store_artifacts:
+          path: /tmp/testlogs
+
+  test-fx_converters_vanilla:
+    description: "Test the fx vanilla converters"
+    steps:
+      - run:
+          name: Run FX converter tests
+          command: |
+            cd py/torch_tensorrt/fx/test
+            pushd converters/vanilla/
+            pytest --junitxml=/tmp/artifacts/test_results/fx/converters/vanilla/test_results.xml
             popd
 
       - store_test_results:
@@ -587,7 +619,7 @@ commands:
           path: /tmp/testlogs
 
   test-fx_tracer:
-    description: "Test the fx tracer"
+    description: "Test all fx tracers"
     steps:
       - run:
           name: Run FX tracer
@@ -602,6 +634,22 @@ commands:
       - store_artifacts:
           path: /tmp/testlogs
 
+  test-fx_tracer_acc:
+    description: "Test the fx acc tracer only"
+    steps:
+      - run:
+          name: Run FX tracer
+          command: |
+            cd py/torch_tensorrt/fx/test
+            pushd tracer
+            list_tracer=$(ls | grep test_acc)
+            pytest $list_tracer --junitxml=/tmp/artifacts/test_results/fx/tracer/test_results.xml
+            popd
+      - store_test_results:
+          path: /tmp/artifacts
+      - store_artifacts:
+          path: /tmp/testlogs
+
   test-fx_quant:
     description: "Test the fx quant"
     steps:
@@ -625,7 +673,9 @@ commands:
           name: Run fx tests
           command: |
             mkdir -p /tmp/artifacts/test_results
-      - test-fx_converters
+      - test-fx_converters_acc
+      - test-fx_converters_aten
+      - test-fx_converters_vanilla
       - test-fx_passes
       - test-fx_tools
       - test-fx_trt_lower
@@ -637,6 +687,26 @@ commands:
       - store_artifacts:
           path: /tmp/testlogs
 
+  test-fx-no-aten:
+    description: "Test the fx backend without aten operators"
+    steps:
+      - run:
+          name: Run fx tests without aten ops
+          command: |
+            mkdir -p /tmp/artifacts/test_results
+      - test-fx_converters_acc
+      - test-fx_converters_vanilla
+      - test-fx_passes
+      - test-fx_tools
+      - test-fx_trt_lower
+      - test-fx_tracer_acc
+      - test-fx_core
+      - test-fx_quant
+      - store_test_results:
+          path: /tmp/artifacts
+      - store_artifacts:
+          path: /tmp/testlogs
+
 # Define a job to be invoked later in a workflow.
 # See: https://circleci.com/docs/2.0/configuration-reference/#jobs
 jobs:
@@ -782,6 +852,37 @@ jobs:
       - dump-test-env
       - test-fx
 
+  test-py-fx-x86_64-linux-no-aten:
+    parameters:
+      torch-build:
+        type: string
+      torch-build-index:
+        type: string
+      trt-version-long:
+        type: string
+    machine:
+      image: ubuntu-2004-cuda-11.4:202110-01
+    resource_class: gpu.nvidia.large
+    steps:
+      - checkout
+      - attach_workspace:
+          at: /tmp/dist/
+      - install-torch-from-index:
+          torch-build:  << parameters.torch-build >>
+          torch-build-index: << parameters.torch-build-index >>
+      - create-py-env:
+          trt-version-long: << parameters.trt-version-long >>
+      - install-cudnn
+      # - run:
+      #     name: "Set LD_LIBRARY_PATH path to include the installed CUDNN"
+      #     command: export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu/:$LD_LIBRARY_PATH
+      - run:
+          name: "Install torch-tensorrt"
+          command: pip3 install --pre /tmp/dist/x86_64-linux/*cp39-cp39*.whl
+      # We install torch after torch-trt because pip automatically enforces the version constraint otherwise
+      - dump-test-env
+      - test-fx-no-aten
+
   package-x86_64-linux:
     parameters:
       enabled:
@@ -1070,10 +1171,16 @@ parameters:
   # Nightly platform config
   torch-build:
     type: string
-    default: "2.0.0.dev20230219+cu117"
+    default: "2.1.0.dev20230314+cu117"
   torch-build-index:
     type: string
     default: "https://download.pytorch.org/whl/nightly/cu117"
+  torch-build-legacy:
+    type: string
+    default: "1.13.1+cu117"
+  torch-build-index-legacy:
+    type: string
+    default: "https://download.pytorch.org/whl/cu117"
   cudnn-version:
     type: string
     default: "8.5.0.96"
@@ -1127,6 +1234,7 @@ workflows:
                 - release/**/*
     jobs:
       - build-x86_64-linux:
+          name: build-x86_64-linux
           torch-build: << pipeline.parameters.torch-build >>
           torch-build-index: << pipeline.parameters.torch-build-index >>
 
@@ -1153,6 +1261,36 @@ workflows:
           requires:
             - build-x86_64-linux
 
+      - build-x86_64-linux:
+          name: build-x86_64-linux-legacy
+          torch-build: << pipeline.parameters.torch-build-legacy >>
+          torch-build-index: << pipeline.parameters.torch-build-index-legacy >>
+
+      - test-core-cpp-x86_64-linux:
+          name: test-core-cpp-x86_64-linux-legacy
+          torch-build: << pipeline.parameters.torch-build-legacy >>
+          torch-build-index: << pipeline.parameters.torch-build-index-legacy >>
+          trt-version-short: << pipeline.parameters.trt-version-short >>
+          trt-version-long: << pipeline.parameters.trt-version-long >>
+          cudnn-version: << pipeline.parameters.cudnn-version >>
+          requires:
+            - build-x86_64-linux-legacy
+
+      - test-py-ts-x86_64-linux:
+          name: test-py-ts-x86_64-linux-legacy
+          torch-build: << pipeline.parameters.torch-build-legacy >>
+          torch-build-index: << pipeline.parameters.torch-build-index-legacy >>
+          trt-version-long: << pipeline.parameters.trt-version-long >>
+          requires:
+            - build-x86_64-linux-legacy
+
+      - test-py-fx-x86_64-linux-no-aten:
+          torch-build: << pipeline.parameters.torch-build-legacy >>
+          torch-build-index: << pipeline.parameters.torch-build-index-legacy >>
+          trt-version-long: << pipeline.parameters.trt-version-long >>
+          requires:
+            - build-x86_64-linux-legacy
+
   release:
     when: << pipeline.parameters.enable-packaging >>
     jobs:

README.md

@@ -73,6 +73,7 @@ import torch_tensorrt
 ...
 
 trt_ts_module = torch_tensorrt.compile(torch_script_module,
+    # If the inputs to the module are plain Tensors, specify them via the `inputs` argument:
     inputs = [example_tensor, # Provide example tensor for input shape or...
         torch_tensorrt.Input( # Specify input object with shape and dtype
             min_shape=[1, 3, 224, 224],
@@ -81,6 +82,12 @@ trt_ts_module = torch_tensorrt.compile(torch_script_module,
             # For static size shape=[1, 3, 224, 224]
             dtype=torch.half) # Datatype of input tensor. Allowed options torch.(float|half|int8|int32|bool)
     ],
+
+    # For inputs containing tuples or lists of tensors, use the `input_signature` argument:
+    # Below, we have an input consisting of a Tuple of two Tensors (Tuple[Tensor, Tensor])
+    # input_signature = ( (torch_tensorrt.Input(shape=[1, 3, 224, 224], dtype=torch.half),
+    #                      torch_tensorrt.Input(shape=[1, 3, 224, 224], dtype=torch.half)), ),
+
     enabled_precisions = {torch.half}, # Run with FP16
 )
 
@@ -114,7 +121,7 @@ torch.jit.save(trt_ts_module, "trt_torchscript_module.ts") # save the TRT embedd
 These are the following dependencies used to verify the testcases. Torch-TensorRT can work with other versions, but the tests are not guaranteed to pass.
 
 - Bazel 5.2.0
-- Libtorch 2.0.0.dev20230103 (built with CUDA 11.7)
+- Libtorch 2.1.0.dev20230314 (built with CUDA 11.7)
 - CUDA 11.7
 - cuDNN 8.5.0
 - TensorRT 8.5.1.7
@@ -124,7 +131,7 @@ These are the following dependencies used to verify the testcases. Torch-TensorR
 Releases: https://github.com/pytorch/TensorRT/releases
 
 ```
-pip install torch-tensorrt==1.2.0 --find-links https://github.com/pytorch/TensorRT/releases/expanded_assets/v1.2.0
+pip install torch-tensorrt
 ```
 
 ## Compiling Torch-TensorRT

WORKSPACE

@@ -56,17 +56,17 @@ new_local_repository(
 http_archive(
     name = "libtorch",
     build_file = "@//third_party/libtorch:BUILD",
-    sha256 = "8b3b48615169c83c1b643c0efade078ea080b1da598e15fcf01bc59421f3095e",
+    sha256 = "7c4b8754830fef23ec19c5eaf414794cee9597b435df055f5c1d0471d3e81568",
     strip_prefix = "libtorch",
-    urls = ["https://download.pytorch.org/libtorch/nightly/cu117/libtorch-cxx11-abi-shared-with-deps-2.0.0.dev20230219%2Bcu117.zip"],
+    urls = ["https://download.pytorch.org/libtorch/nightly/cu117/libtorch-cxx11-abi-shared-with-deps-2.1.0.dev20230314%2Bcu117.zip"],
 )
 
 http_archive(
     name = "libtorch_pre_cxx11_abi",
     build_file = "@//third_party/libtorch:BUILD",
-    sha256 = "aa7fd06079d260ff83c344d043fb84fbd9cf831cf375ed8b5a1b62416817af31",
+    sha256 = "f1e64a75dd12d0ba4c8c1f61947299e0a9c50684dff64f0cfbf355aa7a13e8cf",
     strip_prefix = "libtorch",
-    urls = ["https://download.pytorch.org/libtorch/nightly/cu117/libtorch-shared-with-deps-2.0.0.dev20230219%2Bcu117.zip"],
+    urls = ["https://download.pytorch.org/libtorch/nightly/cu117/libtorch-shared-with-deps-2.1.0.dev20230314%2Bcu117.zip"],
 )
 
 # Download these tarballs manually from the NVIDIA website