Update oneAPI dependencies to 2024.1, add some SYCL runtime variables, change default device to 'arc' and documentation updates.

Author: simonlui

Dockerfile

@@ -21,9 +21,9 @@ RUN no_proxy=$no_proxy wget --progress=dot:giga -O- https://apt.repos.intel.com/
    | tee /etc/apt/sources.list.d/oneAPI.list
 
 # Define and install oneAPI runtime libraries for less space.
-ARG DPCPP_VER=2024.0.0-49819
-ARG MKL_VER=2024.0.0-49656
-ARG CMPLR_COMMON_VER=2024.0
+ARG DPCPP_VER=2024.1.0-963
+ARG MKL_VER=2024.1.0-691
+ARG CMPLR_COMMON_VER=2024.1
 # intel-oneapi-compiler-shared-common provides `sycl-ls`
 RUN apt-get update && \
     apt-get install -y --no-install-recommends --fix-missing \
@@ -34,7 +34,7 @@ RUN apt-get update && \
     rm -rf /var/lib/apt/lists/*
 
 # Add and prepare Intel Graphics driver index. This is dependent on being able to pass your GPU with a working driver on the host side where the image will run.
-ARG DEVICE=flex
+ARG DEVICE=arc
 # hadolint ignore=DL4006
 RUN no_proxy=$no_proxy wget --progress=dot:giga -qO - https://repositories.intel.com/graphics/intel-graphics.key | \
     gpg --dearmor --output /usr/share/keyrings/intel-graphics.gpg
@@ -48,7 +48,7 @@ FROM ubuntu:${UBUNTU_VERSION}
 # Copy all the files from the oneAPI runtime libraries image into the actual final image.
 RUN mkdir -p /oneapi-lib
 COPY --from=oneapi-lib-installer /opt/intel/oneapi/redist/lib/ /oneapi-lib/
-ARG CMPLR_COMMON_VER=2024.0
+ARG CMPLR_COMMON_VER=2024.1
 COPY --from=oneapi-lib-installer /opt/intel/oneapi/compiler/${CMPLR_COMMON_VER}/bin/sycl-ls /bin/
 COPY --from=oneapi-lib-installer /usr/share/keyrings/intel-graphics.gpg /usr/share/keyrings/intel-graphics.gpg
 COPY --from=oneapi-lib-installer /etc/apt/sources.list.d/intel.gpu.jammy.list /etc/apt/sources.list.d/intel.gpu.jammy.list
@@ -64,6 +64,7 @@ ENV LD_LIBRARY_PATH=/oneapi-lib:/oneapi-lib/intel64:$LD_LIBRARY_PATH
 RUN apt-get update && \
     apt-get install -y --no-install-recommends --fix-missing \
     ca-certificates \
+    fonts-noto \
     gnupg2 \
     gpg-agent \
     software-properties-common && \
@@ -153,12 +154,17 @@ ENV LD_PRELOAD=${ALLOCATOR_LD_PRELOAD}
 ENV NEOReadDebugKeys=1
 ENV ClDeviceGlobalMemSizeAvailablePercent=100
 
+# Enable SYCL variables for cache reuse and single threaded mode.
+# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md
+ENV SYCL_CACHE_PERSISTENT=1
+ENV SYCL_PI_LEVEL_ZERO_SINGLE_THREAD_MODE=1
+
 # Enable double precision emulation just in case.
 # See https://github.com/intel/compute-runtime/blob/master/opencl/doc/FAQ.md#feature-double-precision-emulation-fp64
 ENV OverrideDefaultFP64Settings=1
 ENV IGC_EnableDPEmulation=1
 
-# Set variable for better training performance
+# Set variable for better training performance in case.
 # See https://github.com/intel/intel-extension-for-pytorch/issues/296#issuecomment-1461118993
 ENV IPEX_XPU_ONEDNN_LAYOUT=1
 

README.md

@@ -35,7 +35,7 @@ cd Docker_IPEX_ComfyUI
 docker build -t ipex-arc-comfy:latest -f Dockerfile .
 ```
 #### Temporary workaround with Intel Extension for Pytorch
-Go to the releases page and download either Python 3.10 or 3.11 packages versions for intel_extension_for_pytorch, intel_extension_for_pytorch_deepspeed, torch, torchaudio, and torchvision and put the package files in your ComfyUI directory. The startup script will pick them up and install them the first time you start the container.
+Go to the releases page and download the Python 3.11 packages versions for intel_extension_for_pytorch, intel_extension_for_pytorch_deepspeed, torch, torchaudio, and torchvision and put the package files in your ComfyUI directory. This will replace installation of these packages from Intel's repositories until a new version of IPEX releases. The startup script will pick them up and install them the first time you start the container.
 
 Once the image build is complete, then run the following if using Linux in terminal or Docker Desktop.
 ```console
@@ -92,4 +92,4 @@ docker:
 * `ipexrun` is a launcher script to use Intel's Extension For Pytorch without code changes with optimizations enabled. There may be issues running ComfyUI through the launcher with some of the arguments you can use so it is not enabled by default. To use the XPU path that uses your GPU, add in `-e UseIPEXRUN=true` to the argument string above. Additionally, if one wants to run it in CPU mode, you should additionally add in `-e UseXPU=false` to that list. You should also then set the environment variable for passing arguments to `ipexrun` adding `-e IPEXRUNArgs=<Your arguments here>`. A reference to all the `ipexrun` arguments can be found [here](https://intel.github.io/intel-extension-for-pytorch/xpu/latest/tutorials/performance_tuning/launch_script.html)
 * You can change between `tcmalloc` (default) and `jemalloc` if using CPU `ipexrun`, add in `--build-arg="ALLOCATOR=jemalloc"` when building the image in the first step to switch between the two allocators for `ipexrun`.
 
-Please refer to the [Dockerfile](./Dockerfile) for all available build arguments and environment variables not mentioned here and documented.
+Please refer to the [Dockerfile](./Dockerfile) for all available build arguments and environment variables not mentioned here and documented.