Rafal Bielski

> This clangd is from the apt repo Only `clangd` from `intel/llvm` can parse SYCL headers from `intel/llvm`. The workaround for this issue from https://github.com/intel/llvm/pull/11795 cannot be upstreamed because it...

I tested this PR with https://gist.github.com/rafbiels/2b584584cfd6412e6b255adab4c264d6 on NVIDIA sm_86 GPU. Here's the outcomes: ``` Release Time [ms] CUDA gridXYZ CUDA blockXYZ ----------------------------------------------------------- 2023.2.1 63 {2, 1888, 1} {394, 1, 1}...

Many thanks for the prompt follow-up! Some replies / extra info: * The cuda driver is not needed for plugin installation or for compilation, just the cuda libs. AFAIK the...

There was a discussion of this in the original PR: https://github.com/llvm/llvm-project/pull/81173#discussion_r1483561988 Regardless of whether it was this PR or another in the upstream pull-down which changed the behaviour, it looks...

Does `-ffast-math` conform to the SYCL standard otherwise? BTW, this is (rightly) triggering the `-Wnan-infinity-disabled` warning which is hidden by including SYCL headers as system headers. Including the headers as...

Could we separate the discussion about multi-device context from this single-device context fix? This PR brings back the (relatively uncontroversial) behaviour broken recently and we have a customer waiting for...

The test was added back with https://github.com/intel/llvm/pull/16544 now properly marked as `UNSUPPORTED: (opencl && cpu)`. The issue is still there and the test can be used to validate future fixes...

It's been shown to fail also on the `opencl:fpga` target and on a second thought there is no reason at UR level why it wouldn't affect GPU as well, even...

This is not fixed, the reproducer still shows the issue with intel/llvm `nightly-2025-05-07` and the following OpenCL CPU device: ``` [opencl:cpu][opencl:0] Intel(R) OpenCL, 12th Gen Intel(R) Core(TM) i9-12900K OpenCL 3.0...

Hi @martygrant, as the issue lies in memory alignment, any small change to host allocation prior to the to-be-filled memory may hide or bring up the crash. I found that...