Aksel Alpay
LightNews LightNews
illuhad.bsky.social
Aksel Alpay
@illuhad.bsky.social
62 followers 110 following 10 posts
Working at Uni Heidelberg. I do AdaptiveCpp and HPC. Making things go fast!
Posts Replies Media Videos
illuhad.bsky.social
#AdaptiveCpp 25.10 released! Introducing a deployment model (compile once, ship to NVIDIA/AMD/Intel GPUs and CPUs), performance improvements and more! github.com/AdaptiveCpp/...
November 5, 2025 at 4:42 PM
illuhad.bsky.social
My latest paper from @iwocl.bsky.social is now available. Here's one of #AdaptiveCpp's secrets revealed which explains why it often manages to outperform vendor compilers: dl.acm.org/doi/full/10....
Adaptivity in AdaptiveCpp: Optimizing Performance by Leveraging Runtime Information During JIT-Compilation | Proceedings of the 13th International Workshop on OpenCL and SYCL
dl.acm.org
July 10, 2025 at 4:10 PM
illuhad.bsky.social
#AdaptiveCpp 25.02 released! Introducing AdaptiveCpp portable CUDA (PCUDA):direct support for the CUDA and HIP languages in AdaptiveCpp's compiler. Compile CUDA code, obtain single binary that runs on CPU, Intel GPU, NVIDIA GPU, AMD GPU. And perhaps even faster than nvcc. github.com/AdaptiveCpp/...
May 9, 2025 at 6:29 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
So you want to support #SYCL? We’re finding out on the last day of #IWOCL in the invited talk from Ben Ashbaugh (Intel) on his personal perspective for supporting SYCL with #OpenCL.
April 11, 2025 at 7:05 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Next up at #IWOCL, Julian Oppermann (Codeplay Software) talks about
fast compilation of #SYCL for code generated at runtime.
April 11, 2025 at 7:37 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Virtual functions are important to C++ programs. Alexey Sachkov (Intel) proposes how #SYCL can by updated to remove the current restriction on this to make it work for those (like Kokkos) asking for it.
#IWOCL
April 11, 2025 at 8:10 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Just running compute with #SYCL is great but you always want to *SEE* the output. Duncan Brawley (Codeplay Software) is presenting interoperability for images with the DirectX and Vulkan Graphics APIs.
#IWOCL
April 11, 2025 at 9:21 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
The #IWOCL 2025 Outstanding Full Paper Award Winner, Peter Thoman (UIBK) presents their work on optimising strided data movement across GPUs with Celerity and #SYCL
April 11, 2025 at 9:44 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
See you at #IWOCL 2026!
Keep an eye on iwocl.org for updates, and the archive for this and previous conferences.
#OpenCL #SYCL
April 11, 2025 at 10:19 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
More debugging talks at #IWOCL! This time for #OpenCL and #SYCL on CPUs where the code contains barriers. Presenting work from the team at Intel is Alexey Sachkov.
April 10, 2025 at 1:21 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Tooling is a vital part of any programming ecosystem, and is particularly important for heterogeneous systems. Here at #IWOCL Rakesh Ganesh (Intel) shares work on debugging #SYCL on Intel GPUs in Visual Studio and VS Code.
April 10, 2025 at 12:50 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Next up at #IWOCL, Aksel Alpay (Heidelberg University) talks about new Just-in-Time Optimisations in AdaptiveCpp for #SYCL. Ready to try in the latest compiler at github.com/AdaptiveCpp/...
April 10, 2025 at 12:19 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Next up at #IWOCL, Moritz Heckmann (Saarland University) shows his approaches for implementing #SYCL sub-groups on CPUs.
April 10, 2025 at 10:20 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
The #SYCL backend to Kokkos was shared at #IWOCL last year, and Daniel Arndt (ORNL) is back this year to share experiences in targeting Frontier’s AMD GPUs with SYCL through Kokkos.
April 10, 2025 at 9:43 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Alexey Sachkov (Intel) shares how they are working on SPIR-V backend support in LLVM towards a future where #SYCL can be supported upstream in LLVM directly.
#IWOCL
April 10, 2025 at 9:26 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Next, Alexey Sachkov (Intel) presents a vision for improving #SYCL compile times through changes to the SYCL header.
#IWOCL
April 10, 2025 at 8:05 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Day 2 of #IWOCL begins with State of the Union updates from #SYCL and #SYCL SC working groups. Here Lukas Sommer shares their update from SYCL SC bringing standard heterogeneous programming to safety critical systems.
April 10, 2025 at 7:35 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Finishing today, Christoph Alt (Paderborn University) presents work evaluating #oneAPI I/O pipes in #SYCL for a Jacobi Solver on a FPGA Cluster.
#IWOCL
April 9, 2025 at 3:24 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Xavier Hallade and Stefan Werner present Blender and the Cycles renderer implemented in #SYCL for cross-vendor support for 3D rendering, shipping in Blender today.
#IWOCL
April 9, 2025 at 2:38 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
GROMACS have been using #SYCL for a while now, and today, Andrey Alekseenko (KTH) and Ewan Crawford (Codeplay) present work using a SYCL-Graph extension (early versions of this have been presented at previous IWOCLs). This DPC++ extension is also used in Kokkos and Llama.cpp.
#IWOCL
April 9, 2025 at 1:15 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Timothée David Cléris presents Shamrock: a Hydrodynamics (mainly Smooth Particle Hydro) Astrophysics code implemented using #SYCL
#IWOCL
April 9, 2025 at 12:59 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Next up at #IWOCL, Siva Rama Krishna Reddy B and colleagues from Qualcomm share their work on Deep Learning Compilers on Adreno GPUs
April 9, 2025 at 12:24 PM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Hongqiang Wang (Qualcomm) presenting work at #IWOCL on their optimised #OpenCL backend for the open-source llama.cpp large language model (LLM).
April 9, 2025 at 10:25 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
Jan Solanti (Tampere University) now shares his work on improving latency in PoCL-Remote at #IWOCL
#OpenCL
April 9, 2025 at 9:56 AM
Reposted by Aksel Alpay
tomdeakin.bsky.social
A key part of #IWOCL are State of the Union updates on the programming models from the Khronos Working Groups. First up, #OpenCL, presented by Ben Ashbaugh (Intel).
April 9, 2025 at 9:27 AM