Tags: intel/llvm
Tags
Merged to sycl-web with no conflict or build issue
HALT Conflicts detected on merge .: (from main) CONFLICT (content): Merge conflict in llvm/lib/Target/NVPTX/NVPTXLowerArgs.cpp commit 21d93c2 Author: Alex MacLean <amaclean@nvidia.com> Date: Fri Feb 20 09:40:21 2026 -0800 [NVPTX] Cleanup NVPTXLowerArgs, simplifying logic and improving alignment propagation (#180286) Affected files: llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp llvm/lib/Target/NVPTX/NVPTXLowerArgs.cpp llvm/test/CodeGen/NVPTX/bug21465.ll llvm/test/CodeGen/NVPTX/lower-args-alignment.ll llvm/test/CodeGen/NVPTX/lower-args-gridconstant.ll llvm/test/CodeGen/NVPTX/lower-args.ll llvm/test/CodeGen/NVPTX/lower-byval-args.ll Notify: icl.xmain.pulldown.coordinators@intel.com
Merged to sycl-web with no conflict or build issue
[SYCL][Doc] E5M2 stochastic rounding is for float (#21316) Fix a mistake in the spec for E5M2 stochastic rounding. The supported datatypes are `half`, `bfloat16`, and `float`. This is what the comment said, but the synopsis incorrectly showed `double` instead of `float`.
[SYCL][E2E] Fix validation layers for Vulkan in bindless images inter… …op tests (#21271) This PR fixes issue which we always had false for `return layer.layerName == "VK_LAYER_KHRONOS_validation"` because it's comparison of C-style char array to a string literal. The content is ommited. I removed that whole check. We need for CI always Vulkan validation layers to check for issues on Vulkan side in bindless interop tests.
[SYCL][Doc] Update proposed specs for FP4-6-8 (#21286) After discussing with the PyTorch team, we decided to make the following changes to these specifications: * Limit the vector length `N` to the values 1 and 2 for all types. * Remove stochastic rounding and no-saturation modes from E4M3.
PreviousNext