4 Pull requests merged by 3 people

• Enable IGC release on CI

  #367 merged May 9, 2025

• Add flash attention prefill shapes to benchmarks

  #330 merged May 9, 2025

• Break the benchmarks to have separate executables

  #349 merged May 6, 2025

• Use intel-graphics-staging on CI

  #346 merged May 5, 2025

8 Pull requests opened by 4 people

• Reenable BMG examples in CI testing

  #358 opened May 6, 2025

• Fix Variable Sequence Length Support for Flash Attention Prefill + KV Cache

  #359 opened May 6, 2025

• Fix variable length for Flash Attention prefill

  #360 opened May 6, 2025

• Added the second release required FP16 fine-tuned GEMM kernels.

  #361 opened May 7, 2025

• Fix Variable Sequence Length Support for Flash Attention Decode

  #362 opened May 7, 2025

• Add benchmark for Flash Attention Decode

  #363 opened May 7, 2025

• Fix case where matrix size exceeds max uint32

  #364 opened May 8, 2025

• Avoid warnings about Int (alias template)

  #365 opened May 8, 2025

2 Issues opened by 2 people

• [QST] Will it be merged into nvidia cutlass?

  #366 opened May 9, 2025

• [BUG] `XE_2D_U8x32x32_LD_N` for copying A doesn't work for FP8 GEMM & for mixed dtype GEMM examples

  #357 opened May 5, 2025

10 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Input alignment

  #323 commented on May 9, 2025 • 9 new comments

• Rename files and classes for Flash Attention

  #354 commented on May 9, 2025 • 3 new comments

• Pass IGC options on cmd line

  #353 commented on May 8, 2025 • 2 new comments

• Support for bf16 C and D in GEMM

  #356 commented on May 8, 2025 • 2 new comments

• Switch to SPIRV APIs from internal built-in APIs

  #255 commented on May 9, 2025 • 0 new comments

• Enable batch tests for streamK

  #258 commented on May 5, 2025 • 0 new comments

• Pure FP8 (W8A8) GEMM support (draft)

  #306 commented on May 6, 2025 • 0 new comments

• Avoid the cacheline alignment requirement for batches

  #325 commented on May 5, 2025 • 0 new comments

• enable splitk for mixed precision gemm

  #339 commented on May 6, 2025 • 0 new comments

• Enable FP8_E5M2 GEMM and unify FP8 GEMM implementation with xe_mma.hpp

  #352 commented on May 7, 2025 • 0 new comments