From 6c0872ebdeeadf3d667568457a54ad5fe91feace Mon Sep 17 00:00:00 2001
From: Matt Pharr <matt@pharr.org>
Date: Thu, 28 Jan 2021 13:32:32 -0800
Subject: [PATCH] Add CI for GPU builds (#103)

(Only makes sure everything compiles; doesn't run tests.)
---
 .github/workflows/{c-cpp.yml => ci-cpu.yml} | 12 +++-
 .github/workflows/ci-gpu.yml                | 75 +++++++++++++++++++++
 CMakeLists.txt                              |  4 +-
 README.md                                   |  3 +-
 src/pbrt/gpu/pathintegrator.cpp             |  3 -
 5 files changed, 90 insertions(+), 7 deletions(-)
 rename .github/workflows/{c-cpp.yml => ci-cpu.yml} (89%)
 create mode 100644 .github/workflows/ci-gpu.yml

diff --git a/.github/workflows/c-cpp.yml b/.github/workflows/ci-cpu.yml
similarity index 89%
rename from .github/workflows/c-cpp.yml
rename to .github/workflows/ci-cpu.yml
index 7ea02adc..02502485 100644
--- a/.github/workflows/c-cpp.yml
+++ b/.github/workflows/ci-cpu.yml
@@ -1,6 +1,14 @@
-name: ci
+name: cpu-build-and-test
 
-on: [push, pull_request]
+on:
+  push:
+    paths-ignore:
+      - '**.md'
+      - 'images/**'
+  pull_request:
+    paths-ignore:
+      - '**.md'
+      - 'images/**'
 
 jobs:
   build:
diff --git a/.github/workflows/ci-gpu.yml b/.github/workflows/ci-gpu.yml
new file mode 100644
index 00000000..f49dbe64
--- /dev/null
+++ b/.github/workflows/ci-gpu.yml
@@ -0,0 +1,75 @@
+name: gpu-build-only
+
+on:
+  push:
+    paths-ignore:
+      - '**.md'
+      - 'images/**'
+  pull_request:
+    paths-ignore:
+      - '**.md'
+      - 'images/**'
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        optix: [ optix-7.1.0, optix-7.2.0 ]
+        cuda: [ '11.0', '11.1', '11.2' ]
+
+    name: GPU Build Only (CUDA ${{ matrix.cuda }}, ${{ matrix.optix }})
+
+    runs-on: ubuntu-20.04
+
+    steps:
+    - name: Checkout pbrt
+      uses: actions/checkout@v2
+      with:
+        submodules: true
+
+    - name: Checkout rgb2spectrum tables
+      uses: actions/checkout@v2
+      with:
+        repository: mmp/rgb2spectrum
+        path: build
+
+    - name: Checkout OptiX headers
+      uses: actions/checkout@v2
+      with:
+        ssh-key: ${{ secrets.CHECKOUT_KEY }}
+        repository: mmp/optix-headers
+        path: optix
+
+    - name: Get cmake
+      uses: lukka/get-cmake@latest
+
+    - name: Install OpenEXR
+      run: sudo apt-get -y install libopenexr-dev
+
+    - name: Try to restore CUDA install from cache
+      id: cache-cuda
+      uses: actions/cache@v2
+      with:
+        path: /usr/local/cuda-${{ matrix.cuda }}
+        key: cuda-usr-local-${{ matrix.cuda }}
+
+    - name: Install CUDA
+      if: steps.cache-cuda.outputs.cache-hit != 'true'
+      run: |
+        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
+        sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
+        sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
+        sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
+        cuda=`echo ${{ matrix.cuda }} | tr . -`
+        sudo apt install -y cuda-compiler-${cuda} cuda-libraries-${cuda} cuda-libraries-dev-${cuda}
+        echo /usr/local/cuda-${{ matrix.cuda }}/bin >> $GITHUB_PATH
+        
+    - name: Configure
+      run: |
+        cd build
+        cmake .. -DPBRT_USE_PREGENERATED_RGB_TO_SPECTRUM_TABLES=True -DPBRT_OPTIX7_PATH=../optix/${{ matrix.optix }} -DPBRT_GPU_SHADER_MODEL=sm_80
+      
+    - name: Build
+      # We need to limit the number of jobs so that it doesn't OOM 
+      run: cmake --build build --parallel 3 --config Release
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 268672a9..29b01502 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -159,7 +159,9 @@ if (CMAKE_CUDA_COMPILER)
     else ()
         enable_language (CUDA)
         list (APPEND PBRT_DEFINITIONS "PBRT_BUILD_GPU_RENDERER")
-        list (APPEND PBRT_DEFINITIONS "NVTX")
+        if (PBRT_NVTX)
+            list (APPEND PBRT_DEFINITIONS "NVTX")
+        endif ()
         set (PBRT_CUDA_ENABLED ON)
 
         # FIXME
diff --git a/README.md b/README.md
index 4c16aa77..32c0cfcc 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,8 @@
 pbrt, Version 4 (Early Release)
 ===============================
 
-[<img src="https://github.com/mmp/pbrt-v4/workflows/ci/badge.svg">](https://github.com/mmp/pbrt-v4/actions)
+[<img src="https://github.com/mmp/pbrt-v4/workflows/cpu-build-and-test/badge.svg">](https://github.com/mmp/pbrt-v4/actions)
+[<img src="https://github.com/mmp/pbrt-v4/workflows/gpu-build-only/badge.svg">](https://github.com/mmp/pbrt-v4/actions)
 
 ![Transparent Machines frame, via @beeple](images/teaser-transparent-machines.png)
 
diff --git a/src/pbrt/gpu/pathintegrator.cpp b/src/pbrt/gpu/pathintegrator.cpp
index 70b088c7..b07b6caa 100644
--- a/src/pbrt/gpu/pathintegrator.cpp
+++ b/src/pbrt/gpu/pathintegrator.cpp
@@ -33,7 +33,6 @@
 #include <map>
 
 #include <cuda.h>
-#include <cuda_profiler_api.h>
 #include <cuda_runtime.h>
 
 #ifdef NVTX
@@ -603,8 +602,6 @@ void GPURender(ParsedScene &scene) {
 
     LOG_VERBOSE("Total rendering time: %.3f s", timer.ElapsedSeconds());
 
-    CUDA_CHECK(cudaProfilerStop());
-
     if (!Options->quiet) {
         ReportKernelStats();