Add documentation for cumem_utils (pytorch#2232)

Summary:
- Update documentation on adding Python and C++ documentation
- Add extensive documentation for `cumem_utils`

Pull Request resolved: pytorch#2232

Reviewed By: spcyppt

Differential Revision: D52393909

Pulled By: q10

fbshipit-source-id: 8d4561135b79d1e5b791e1e9204d8c8b81d3be4e

.github/scripts/fbgemm_gpu_docs.bash

@@ -34,13 +34,20 @@ install_docs_tools () {
   # shellcheck disable=SC2155
   local env_prefix=$(env_name_or_prefix "${env_name}")
 
-  echo "[INSTALL] Installing docs tools ..."
+  echo "[INSTALL] Installing Doxygen ..."
+
   # shellcheck disable=SC2086
   (exec_with_retries 3 conda install ${env_prefix} -c conda-forge -y \
-    doxygen) || return 1
+    doxygen \
+    make) || return 1
 
   # Check binaries are visible in the PATH
   (test_binpath "${env_name}" doxygen) || return 1
+  (test_binpath "${env_name}" make) || return 1
+
+  echo "[BUILD] Installing docs-build dependencies ..."
+  # shellcheck disable=SC2086
+  (exec_with_retries 3 conda run ${env_prefix} python -m pip install -r requirements.txt) || return 1
 
   echo "[INSTALL] Successfully installed all the docs tools"
 }
@@ -69,10 +76,6 @@ build_fbgemm_gpu_docs () {
   # shellcheck disable=SC2155
   local env_prefix=$(env_name_or_prefix "${env_name}")
 
-  echo "[BUILD] Installing docs-build dependencies ..."
-  # shellcheck disable=SC2086
-  (exec_with_retries 3 conda run ${env_prefix} python -m pip install -r requirements.txt) || return 1
-
   echo "[BUILD] Running Doxygen build ..."
   # shellcheck disable=SC2086
   (exec_with_retries 3 conda run ${env_prefix} doxygen Doxyfile.in) || return 1

.github/scripts/utils_build.bash

@@ -166,12 +166,14 @@ install_build_tools () {
     cmake \
     hypothesis \
     jinja2 \
+    make \
     ninja \
     numpy \
     scikit-build \
     wheel) || return 1
 
   # Check binaries are visible in the PAATH
+  (test_binpath "${env_name}" make) || return 1
   (test_binpath "${env_name}" cmake) || return 1
   (test_binpath "${env_name}" ninja) || return 1
 

.github/workflows/fbgemm_gpu_docs.yml

@@ -67,7 +67,7 @@ jobs:
       run: . $PRELUDE; install_build_tools $BUILD_ENV
 
     - name: Install Documentation Tools
-      run: . $PRELUDE; install_docs_tools $BUILD_ENV
+      run: . $PRELUDE; cd fbgemm_gpu/docs; install_docs_tools $BUILD_ENV
 
     - name: Install PyTorch-CPU Nightly
       run: . $PRELUDE; install_pytorch_pip $BUILD_ENV nightly cpu

.gitignore

@@ -90,3 +90,6 @@ install_manifest.txt
 compile_commands.json
 CTestTestfile.cmake
 _deps
+
+# Sphinx repos
+pytorch-sphinx-theme/

fbgemm_gpu/README.md

@@ -1,8 +1,8 @@
 # FBGEMM_GPU
 
-[![FBGEMM_GPU CI](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci.yml/badge.svg)](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci.yml)
-[![FBGEMM_GPU-CPU Nightly Build](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_cpu_nightly.yml/badge.svg?event=schedule)](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_cpu_nightly.yml)
-[![FBGEMM_GPU-CUDA Nightly Build](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_cuda_nightly.yml/badge.svg?event=schedule)](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_cuda_nightly.yml)
+[![FBGEMM_GPU-CPU CI](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci_cpu.yml/badge.svg)](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci_cpu.yml)
+[![FBGEMM_GPU-CUDA CI](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci_cuda.yml/badge.svg)](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci_cuda.yml)
+[![FBGEMM_GPU-ROCm CI](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci_rocm.yml/badge.svg)](https://github.com/pytorch/FBGEMM/actions/workflows/fbgemm_gpu_ci_rocm.yml)
 
 FBGEMM_GPU (FBGEMM GPU Kernels Library) is a collection of high-performance PyTorch
 GPU operator libraries for training and inference.  The library provides efficient

fbgemm_gpu/codegen/embedding_forward_template_helpers.cuh

@@ -130,7 +130,7 @@ __device__ __forceinline__ void cp_async_fence() {
 
 /// Partial specialization
 
-/// Blocks until all but <N> previous cp.async.commit_group operations have
+/// Blocks until all but N previous cp.async.commit_group operations have
 /// committed.
 
 template <int N>