Repository navigation
Download links for nvidia and amd pluggins are not valid #9
Description
Activity
Hello, can't find them either.
Hello, me neither.
Both the direct download link https://developer.codeplay.com/api/v1/products/download?product=oneapi&variant=nvidia&filters[]=linux and https://developer.codeplay.com/apt/public.key (for setting up the APT repository) return 404 Errors
I guess they are deprecated, but there is no official announcement. I wonder if compiling DPCPP with AMD/Nvidia support would work.
Hello, I also encountered the same problem. Do you have any good solutions？
FYI, the Intel oneAPI toolkit 2025.3.0 release notes say
CUDA/HIP adapters: Starting in 2025.3 release, CUDA and HIP adapters are no longer available in binary packages. Instead, they will be provided in source format together with the SYCL open-source compiler.
I tried it just now. The instructions are not correct. When I run the first cmake configure I get:
$ pwd /<path-to>/DPCPP/llvm $ cmake -S . -B build -DUR_BUILD_TESTS=OFF -DUR_ENABLE_TRACING=ON -DUR_BUILD_ADAPTER_CUDA=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX="/scratch/project_2001659/cristian/DPCPP/intel" CMake Error: The source directory "/scratch/project_2001659/cristian/DPCPP/llvm" does not appear to contain CMakeLists.txt. Specify --help for usage, or press the help button on the CMake GUI.If I do
cd llvmagain I get:CMake Warning: Manually-specified variables were not used by the project: UR_BUILD_ADAPTER_CUDA UR_BUILD_TESTS UR_ENABLE_TRACINGI tried it just now. The instructions are not correct. When I run the first cmake configure I get:
$ pwd /<path-to>/DPCPP/llvm $ cmake -S . -B build -DUR_BUILD_TESTS=OFF -DUR_ENABLE_TRACING=ON -DUR_BUILD_ADAPTER_CUDA=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX="/scratch/project_2001659/cristian/DPCPP/intel" CMake Error: The source directory "/scratch/project_2001659/cristian/DPCPP/llvm" does not appear to contain CMakeLists.txt. Specify --help for usage, or press the help button on the CMake GUI.If I do
cd llvmagain I get:CMake Warning:
Manually-specified variables were not used by the project:UR_BUILD_ADAPTER_CUDA UR_BUILD_TESTS UR_ENABLE_TRACINGYeah the instructions are definitely wrong. I presume you're supposed to use the
unified-runtimedirectory.Thanks for replying. Unfortunately for me our servers do not have the depndencies, so I will try after the next upgrade.
I tried it just now. The instructions are not correct. When I run the first cmake configure I get:
$ pwd /<path-to>/DPCPP/llvm $ cmake -S . -B build -DUR_BUILD_TESTS=OFF -DUR_ENABLE_TRACING=ON -DUR_BUILD_ADAPTER_CUDA=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX="/scratch/project_2001659/cristian/DPCPP/intel" CMake Error: The source directory "/scratch/project_2001659/cristian/DPCPP/llvm" does not appear to contain CMakeLists.txt. Specify --help for usage, or press the help button on the CMake GUI.If I do
cd llvmagain I get:
CMake Warning:
Manually-specified variables were not used by the project:UR_BUILD_ADAPTER_CUDA UR_BUILD_TESTS UR_ENABLE_TRACINGYeah the instructions are definitely wrong. I presume you're supposed to use the
unified-runtimedirectory.FWIW, I tried using the prebuilt CUDA adapter library from the v6.2.1 release by setting
UR_ADAPTERS_SEARCH_PATHappropriately but ended up getting a segfault with thesycl-lsfrom the 2025.3.1 Intel DPC++ compiler.I got it to work with oneAPI 2025.2 + Unified Runtime (UR) v6.2.1. I reported a bug with oneAPI 2025.3 + Unified Runtime v6.3.0-rc1 at intel/llvm#20945. The segfault I previously mentioned was due to the differing Unified Memory Framework (UMF) versions (v0.11 in UR 6.2.1 and v1.0 in oneAPI 2025.3).
Hello,
Are the Nvidia and AMD pluggins deprecated? can't download anymore. Neither the old links nor the present instructions work.
Best regards,
Cristian-Vasile Achim