Issues

• examples/12 MoE GEMM: reported GFLOP/s is dominated by per-expert fill; suggest making fill a first-class option

  intel/sycl-tla#855 · Filter by author knowrohit opened on Aug 26, 2026

• [Windows][BMG G21] SYCL-TLA Block2D copy causes DEVICE_LOST or hangs on Arc B580

  intel/sycl-tla#851 · Filter by author yzwzhanghao2 opened on Aug 20, 2026
  2

• [BUG] It can get the performance improvement after "lsc_load_block2d.ugm.ca.ca" replaced "lsc_load_block2d.ugm.ca.ca.uc"

  bug
  intel/sycl-tla#850 · Filter by author yihuaxu opened on Aug 19, 2026

• [BUG] FlashAttention performance regression on B580

  bug
  intel/sycl-tla#839 · Filter by author LuFinch opened on Jul 20, 2026

• [QST] Verification failed after running /examples/00_bmg_gemm/00_bmg_gemm in B60

  question
  intel/sycl-tla#696 · Filter by author jiafuzha opened on Jan 20, 2026
  1

• [BUG] Unable to convert output dtype from FP32 to BF16 in Group GEMM epilogue

  bugCUTLASS C++
  intel/sycl-tla#500 · Filter by author sanchitintel opened on Sep 11, 2025
  1

  

• [BUG] FP8 scaled MM has quite high latency

  bug
  intel/sycl-tla#452 · Filter by author sanchitintel opened on Jul 1, 2025
  2

• Set large GRF mode in cmake

  intel/sycl-tla#427 · Filter by author rolandschulz opened on Jun 11, 2025
  1

• Test 2025.2

  intel/sycl-tla#415 · Filter by author rolandschulz opened on Jun 10, 2025

• [FEA] Improve FP8xFP8 GEMM performance, especially for FP8_E4M3

  intel/sycl-tla#394 · Filter by author sanchitintel opened on May 27, 2025
  2

• [QST] Ambiguous error message when tuning the Tileshape

  question
  intel/sycl-tla#376 · Filter by author cfgfung opened on May 14, 2025
  7

• 05_pvc_gemm_with_epilogues fails verification on Windows

  intel/sycl-tla#348 · Filter by author rolandschulz opened on Apr 30, 2025

• Compilation and running tests requires >8GB

  intel/sycl-tla#343 · Filter by author rolandschulz opened on Apr 30, 2025
  1

• [QST] What is the plan for adding intel-cutlass to PyTorch ?

  question
  intel/sycl-tla#286 · Filter by author githubsgi opened on Mar 31, 2025

• Would it make sense to move more into XE_2D_LD_Unpack?

  intel/sycl-tla#59 · Filter by author rolandschulz opened on May 7, 2024