Further improvements to fl CI in open source

Summary:
Updates CI infrastructure and configuration for flashlight.

Start using `gpu.small` (`p3.2xlarge`) instances on AWS vis-a-vis CircleCI with the built-in `ubuntu-1604:201903-01` image.

> It looks like we're downloading and installing CUDA, Docker, and NVIDIA Docker - why??

Unfortunately, CircleCI's VM images, even if they do have NVIDIA/CUDA drivers on them, don't support running a Docker image with nvidia-docker (flags need to be passed at the Docker invocation at runtime). We're forced to install CUDA and nvidia-docker as a result so we can manually invoke `sudo docker run --runtime=nvidia ...`

Other changes/notes:
- The CPU build still uses a normal VM instance with the CPU base docker image. Reduce the number of threads to 1 since any more causes the compiler to OOM for some reason
- Since tests don't compltely pass with the CPU backend yet (there are still a few unimplemented exceptions), don't run CPU tests
- GPU backend builds now take advantage of as many available threads as possible with `-j$(nproc)`
- Separates the CPU and CUDA builds into two different workflows so they're done in parallel.

Reviewed By: VitaliyLi

Differential Revision: D16806968

fbshipit-source-id: 02dea7034fc78d18a5e4a56735dc1c40cb56a07f

.circleci/config.yml

@@ -6,24 +6,74 @@
 
 version: 2.0
 
+gpu: &gpu
+  machine:
+    image: ubuntu-1604:201903-01
+  resource_class: gpu.small
+
 jobs:
-  build:
-    docker:
-      - image: flml/flashlight:cuda-base-latest
+  build-cuda:
+    <<: *gpu
     steps:
       - checkout
       - run:
+          name: Setup Docker and nvidia-docker
+          command: |
+            # Install CUDA 9.2
+            wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_9.2.148-1_amd64.deb
+            sudo dpkg -i cuda-repo-ubuntu1604_9.2.148-1_amd64.deb
+            sudo apt-get update || true
+            sudo apt-get --yes --force-yes install cuda
+            nvidia-smi
+            # Install Docker
+            sudo apt-get update
+            sudo apt-get install \
+            apt-transport-https \
+            ca-certificates \
+            curl \
+            software-properties-common
+            curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
+            sudo apt-key fingerprint 0EBFCD88
+            sudo add-apt-repository \
+            "deb [arch=amd64] https://download.docker.com/linux/ubuntu \
+            $(lsb_release -cs) \
+            stable"
+            sudo apt-get update
+            sudo apt-get install docker-ce
+            # Install nvidia-docker
+            curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | \
+            sudo apt-key add -
+            distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
+            curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | \
+            sudo tee /etc/apt/sources.list.d/nvidia-docker.list
+            sudo apt-get update
+            # Install nvidia-docker2 and reload the Docker daemon configuration
+            sudo apt-get install -y nvidia-docker2
+            sudo pkill -SIGHUP dockerd
+      - run:
+          name: Build flashlight with CUDA backend inside nvidia-docker
           command: |
-            mkdir -p build && cd build
-            cmake .. -DCMAKE_BUILD_TYPE=Release -DFLASHLIGHT_BACKEND=CUDA
-            make && make install
+            sudo docker run --runtime=nvidia --rm -itd --ipc=host --name flashlight flml/flashlight:cuda-base-latest
+            sudo docker exec -it flashlight bash -c "mkdir /flashlight"
+            sudo docker cp . flashlight:/flashlight
+            sudo docker exec -it flashlight bash -c "\
+            cd /flashlight && pwd && ls && mkdir -p build && cd build && \
+            cmake .. -DCMAKE_BUILD_TYPE=Release -DFLASHLIGHT_BACKEND=CUDA && \
+            make -j$(nproc) && make install && make test"
   build-cpu:
     docker:
       - image: flml/flashlight:cpu-base-latest
     steps:
       - checkout
       - run:
+          name: Build flashlight with CPU backend
           command: |
             export MKLROOT=/opt/intel/mkl && mkdir -p build && cd build
             cmake .. -DCMAKE_BUILD_TYPE=Release -DFLASHLIGHT_BACKEND=CPU
-            make && make install
+            make -j1 && make install
+workflows:
+  version: 2
+  build_and_install:
+    jobs:
+      - build-cuda
+      - build-cpu