[alpaka] Add support for the SYCL back-end #407
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
additional sources are necessary to use gdb-oneapi the same is true for the other tools (advisor, inspector, vtune...) but since they are useless there is no point in sourcing them as well
The application is compiled once for the CPU(s) and once for the Intel GPU(s). The flags for AOT compilation have been added and icpx is used as the default compiler for the SYCL backend. The linking step is performed as well with icpx when the SYCL backend is enabled.
commit 819974ddc5b2eb4b33e709bd317701793cdb7d15 Author: Jan Stephan <[email protected]> Date: Thu Aug 3 14:20:59 2023 +0200 Always use std::size_t for CUDA pitch calculations
Changed the order to make the ONEAPI compiler happy
For the other backends it is mapped to ALPAKA_STATIC_ACC_MEM_CONSTANT, but global variables are not supported yet in the SYCL backend. However, in this case a `constexpr` is enough to obtain the same result
- added the allocator policy: Caching or Synchronous - added allocCachedBuf: TODO implement pitch in SYCL - fixed cout because SYCL events in alpaka are not shared pointers - do not cache SYCL events - Changed size of CountersOnly to 1 because SYCL kernels do not support zero-lenght arrays - implemented HostOnlyTask using `alpaka/core/CallbackThread` - adapt `prefixScan` and `radixSort` to SYCL: work around the function pointer not supported in SYCl kernels - implemented the work division - add the `--syclcpu` and `--syclgpu` options for the backends
Math functions are defined in the `namespace math` and are taken from the `sycl namespace` for the SYCL backend, from the global namespace for the CUDA and HIP backends and from the `std namespace` in every other case
added the ifdef for SYCL
Add trait to set the warp size to 32 for the kernels that requires it. Implemented in alpaka only for the SYCL backend at the moment, does nothing for the other backends
There is a bug with `any_of_group` / `all_of_group` in the OpenCL runtime that can be worked around setting the sub-group size equal to the block size. The bug has been solved in the latest runtimes, but the application hangs with these..
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
With #1845 and other following contributions, alpaka now supports also SYCL/oneAPI as a back-end to target CPUs, Intel GPUs and FPGAs.
This PR propagates the support for SYCL also in pixeltrack.
Some comments:
--syclcpuand--syclgpuall_of_group/any_of_groupon CPU, SYCL kernels do not support zero-lenght arrays, device global variables are not supported in the SYCL back-end yet)sycl namespace)In addition, the vendor-specific RNG support in alpaka has been disabled.