Misc. improvements to documentation/build setup for first-time builds. (

apache#7840)

- Makefile
  - Target "crttest" ignored OUTPUTDIR variable

- .gitignore
  - Added ignores for download test data/models.

- docs/README.txt
  - Missing quotes on sphinx dep, needs pinned autodocsumm version

- docs/contribute/pull_request.rst
  - Use "ci_lint" docker image
  - Updated C++ test instructions to refer to the from_source installation for gtest.
  - Updated python test instructions with synr package dependency

- docs/langref/relay_expr.rst
  - Updated reference for example usage of TempExpr. `src/relay/pass/alter_op_layout.cc`
    no longer exists, and `src/relay/transforms/alter_op_layout.cc` doesn't use TempExpr.
    Picked a different use case as example.

- tests/scripts/task_cpp_unittest.sh
  - Updated "make crttest" to run only if "USE_MICRO" is enabled.  While USE_MICRO is always enabled
    in the CI builds, task_cpp_unittest.sh is also recommended for use in
    docs/install/from_source.rst, which does not mandate USE_MICRO.

- docs/install/from_source.rst
  - Added -DMAKE_SHARED_LIBS=ON to the google test cmake config.  By default, only static libs are
    generated for gtest, while TVM's build preferentially selects the shared libs.

- tutorials/get_started/auto_tuning_with_python.py
  - Changed norm_img_data to avoid loop, improve readability

- tutorials/get_started/relay_quick_start.py
  - Previous version used different input data passed to the initial and deployed module, then
    asserts that the results should be the same.  Modified so that the same input data are passed in
    both cases.

Co-authored-by: Eric Lunderberg <[email protected]>

.gitignore

@@ -233,3 +233,8 @@ conda/pkg
 # nix files
 .envrc
 *.nix
+
+# Downloaded models/datasets
+.tvm_test_data
+.dgl
+.caffe2

Makefile

@@ -52,7 +52,7 @@ cpptest:
 	@mkdir -p $(OUTPUTDIR) && cd $(OUTPUTDIR) && cmake .. && $(MAKE) cpptest
 
 crttest:
-	@mkdir -p build && cd build && cmake .. && $(MAKE) crttest
+	@mkdir -p $(OUTPUTDIR) && cd $(OUTPUTDIR) && cmake .. && $(MAKE) crttest
 
 # EMCC; Web related scripts
 EMCC_FLAGS= -std=c++11\

docs/README.txt

@@ -3,7 +3,8 @@ TVM Documentations
 This folder contains the source of TVM documents
 
 - A hosted version of doc is at https://tvm.apache.org/docs
-- pip install sphinx>=1.5.5 sphinx-gallery sphinx_rtd_theme matplotlib Image recommonmark "Pillow<7" autodocsumm tlcpack-sphinx-addon
+- pip install "sphinx>=1.5.5" sphinx-gallery sphinx_rtd_theme matplotlib Image recommonmark "Pillow<7" "autodocsumm<0.2.0" tlcpack-sphinx-addon
+- (Versions 0.2.0 to 0.2.2 of autodocsumm are incompatible with sphinx>=3.4, https://github.com/Chilipp/autodocsumm/pull/42 )
 - Build tvm first in the root folder.
 - Run the following command
 ```bash

docs/contribute/pull_request.rst

@@ -41,15 +41,15 @@ This is a quick guide to submit a pull request, please also refer to the detaile
 
     # While the lint commands used should be identical to those run in CI, this command reproduces
     # the CI lint procedure exactly (typically helpful for debugging lint script errors).
-    docker/bash.sh tlcpack/ci-lint ./tests/scripts/task_lint.sh
+    docker/bash.sh ci_lint ./tests/scripts/task_lint.sh
 
   When the clang-format lint check fails, run git-clang-format as follows to automatically reformat
   your code:
 
   .. code:: bash
 
     # Run clang-format check for all the files that changed since upstream/main
-    docker/bash.sh tlcpack/ci-lint ./tests/lint/git-clang-format.sh upstream/main
+    docker/bash.sh ci_lint ./tests/lint/git-clang-format.sh upstream/main
 
 - Add test-cases to cover the new features or bugfix the patch introduces.
 - Document the code you wrote, see more at :ref:`doc_guide`
@@ -88,35 +88,28 @@ Here is the protocol to update CI image:
 
 Testing
 -------
-Even though we have hooks to run unit tests automatically for each pull request, It's always recommended to run unit tests
+Even though we have hooks to run unit tests automatically for each pull request, it's always recommended to run unit tests
 locally beforehand to reduce reviewers' burden and speedup review process.
 
+Running the C++ tests requires installation of gtest, following the instructions in
+:ref:`install-from-source-cpp-tests`
+
 C++
 ^^^
 .. code:: bash
 
   # assume you are in tvm source root
   TVM_ROOT=`pwd`
 
-  # you need to install google test first, gtest will be installed to $TVM_ROOT/lib
-  apt-get install -y libgtest-dev
-  CACHE_PREFIX=. make -f 3rdparty/dmlc-core/scripts/packages.mk gtest
-
-  mkdir build
-  cd build
-  GTEST_LIB=$TVM_ROOT/lib cmake -DUSE_LLVM=ON ..
-  make cpptest -j$(nproc)
-  for test in *_test; do
-    ./$test
-  done
+  ./tests/scripts/task_cpp_unittest.sh
 
 Python
 ^^^^^^
 Necessary dependencies:
 
 .. code:: bash
 
-  pip install --user pytest Cython
+  pip install --user pytest Cython synr
 
 If you want to run all tests:
 

docs/install/from_source.rst

@@ -260,6 +260,8 @@ Install Contrib Libraries
    nnpack
 
 
+.. _install-from-source-cpp-tests:
+
 Enable C++ Tests
 ----------------
 We use `Google Test <https://github.com/google/googletest>`_ to drive the C++
@@ -271,7 +273,7 @@ tests in TVM. The easiest way to install GTest is from source.
        cd googletest
        mkdir build
        cd build
-       cmake ..
+       cmake -DMAKE_SHARED_LIBS=ON ..
        make
        sudo make install
 

docs/langref/relay_expr.rst

@@ -685,9 +685,9 @@ code but may be inserted in a pass. Any :code:`TempExpr` created in a pass
 should ideally be eliminated before the pass is complete, as a
 :code:`TempExpr` only stores internal state and has no semantics of its own.
 
-For an example of :code:`TempExpr` being used in a pass,
-see :code:`src/relay/pass/alter_op_layout.cc`, which uses :code:`TempExpr` nodes
-to store information about operator layouts as the pass tries to rearrange operator
-calls.
+For an example of :code:`TempExpr` being used in a pass, see
+:code:`src/relay/transforms/fold_scale_axis.cc`, which uses
+:code:`TempExpr` nodes to store information about scaling parameters
+as the pass tries to fold these into the weights of a convolution.
 
 See :py:class:`~tvm.relay.expr.TempExpr` for its definition and documentation.

python/tvm/tir/op.py

@@ -264,7 +264,7 @@ def any(*args, span=None):
 
 
 def all(*args, span=None):
-    """Create a new experssion of the intersection of all conditions in the
+    """Create a new expression of the intersection of all conditions in the
       arguments
 
     Parameters

tests/scripts/task_cpp_unittest.sh

@@ -34,7 +34,12 @@ export OMP_NUM_THREADS=1
 rm -f build/*_test
 
 make cpptest -j2
-make crttest  # NOTE: don't parallelize, due to issue with build deps.
+
+# "make crttest" requires USE_MICRO to be enabled, which is not always the case.
+if grep crttest build/Makefile > /dev/null; then
+    make crttest  # NOTE: don't parallelize, due to issue with build deps.
+fi
+
 for test in build/*_test; do
     ./$test
 done

tutorials/frontend/deploy_object_detection_pytorch.py

@@ -44,7 +44,7 @@
 from tvm import relay
 from tvm import relay
 from tvm.runtime.vm import VirtualMachine
-from tvm.contrib.download import download
+from tvm.contrib.download import download_testdata
 
 import numpy as np
 import cv2
@@ -96,11 +96,10 @@ def forward(self, inp):
 ######################################################################
 # Download a test image and pre-process
 # -------------------------------------
-img_path = "test_street_small.jpg"
 img_url = (
     "https://raw.githubusercontent.com/dmlc/web-data/" "master/gluoncv/detection/street_small.jpg"
 )
-download(img_url, img_path)
+img_path = download_testdata(img_url, "test_street_small.jpg", module="data")
 
 img = cv2.imread(img_path).astype("float32")
 img = cv2.resize(img, (in_size, in_size))

tutorials/get_started/auto_tuning_with_python.py

@@ -127,11 +127,9 @@
 img_data = np.transpose(img_data, (2, 0, 1))
 
 # Normalize according to the ImageNet input specification
-imagenet_mean = np.array([0.485, 0.456, 0.406])
-imagenet_stddev = np.array([0.229, 0.224, 0.225])
-norm_img_data = np.zeros(img_data.shape).astype("float32")
-for i in range(img_data.shape[0]):
-    norm_img_data[i, :, :] = (img_data[i, :, :] / 255 - imagenet_mean[i]) / imagenet_stddev[i]
+imagenet_mean = np.array([0.485, 0.456, 0.406]).reshape((3, 1, 1))
+imagenet_stddev = np.array([0.229, 0.224, 0.225]).reshape((3, 1, 1))
+norm_img_data = (img_data / 255 - imagenet_mean) / imagenet_stddev
 
 # Add the batch dimension, as we are expecting 4-dimensional input: NCHW.
 img_data = np.expand_dims(norm_img_data, axis=0)

tutorials/get_started/relay_quick_start.py

@@ -141,7 +141,7 @@
 
 # load the module back.
 loaded_lib = tvm.runtime.load_module(path_lib)
-input_data = tvm.nd.array(np.random.uniform(size=data_shape).astype("float32"))
+input_data = tvm.nd.array(data)
 
 module = graph_executor.GraphModule(loaded_lib["default"](dev))
 module.run(data=input_data)
@@ -151,4 +151,4 @@
 print(out_deploy.flatten()[0:10])
 
 # check whether the output from deployed module is consistent with original one
-tvm.testing.assert_allclose(out_deploy, out, atol=1e-3)
+tvm.testing.assert_allclose(out_deploy, out, atol=1e-5)