18 Pull requests merged by 15 people

• graph: backend: dnnl: fix genindex for threadpool runtime

  #3415 merged Jun 20, 2025

• xe: remove unused eu_count parameter

  #3401 merged Jun 19, 2025

• xe: add missing status_t checks

  #3400 merged Jun 19, 2025

• graph: utils: pm: correctly handle empty optional subgraph and multi-consumer one

  #3282 merged Jun 19, 2025

• cpu: aarch64: Re-Enable JIT Depthwise Convolution for BF16

  #3441 merged Jun 19, 2025

• graph: backend: dnnl: fix decompose kernel select index check

  #3425 merged Jun 19, 2025

• [GPU] Modify GEMM attr group to support reshape

  #3252 merged Jun 18, 2025

• cpu: ppc64: add gemm and reorder kernels

  #3156 merged Jun 18, 2025

• xe: softmax: restore missing inf_as_zero functionality

  #3318 merged Jun 18, 2025

• ci: aarch64: make ctime regression a warning

  #3416 merged Jun 18, 2025

• cpu: aarch64: prefer brgemm over jit for 1x1 convolutions with sve_256

  #3411 merged Jun 18, 2025

• graph pattern name refactor

  #3362 merged Jun 18, 2025

• doc: readme: update list of verified configurations

  #3431 merged Jun 17, 2025

• cpu: x64: conv: enable scales support for fp8

  #3427 merged Jun 17, 2025

• governance: add Renato Arantes as onednn-cpu-aarch64 codeowner

  #3421 merged Jun 16, 2025

• generic: sycl: RNN Vanilla BWD

  #3015 merged Jun 16, 2025

• cpu: x64: matmul: fix blocking heuristics for l2 set issues

  #3403 merged Jun 16, 2025

• gpu: intel: document workaround for ocl compiler bug in sdpa ukernels

  #2920 merged Jun 13, 2025

22 Pull requests opened by 20 people

• sdpa fma f16

  #3422 opened Jun 13, 2025

• graph: backend: dnnl: backend refactor and sdpa v1 kernel support quantize SDPA

  #3423 opened Jun 16, 2025

• graph: backend: dnnl: backend refactor of adding fusion info attr

  #3424 opened Jun 16, 2025

• github: workflows: bump github/codeql-action from 3.28.18 to 3.29.0

  #3426 opened Jun 16, 2025

• src: remove unnecessary compute device info header dependencies

  #3428 opened Jun 16, 2025

• [GPU] GEMM enable Fp4 weights decompression

  #3430 opened Jun 16, 2025

• cpu: some improvements

  #3432 opened Jun 17, 2025

• fix(benchdnn):: conditionally use assume_buffer_outlives_graph

  #3434 opened Jun 17, 2025

• [backport][rls-v3.8] cpu: x64: matmul: fix blocking heuristics for l2 set issues

  #3436 opened Jun 17, 2025

• [GPU] Add Root Mean Square (RMS) normalization support to lnorm

  #3438 opened Jun 17, 2025

• tests: fix include directory prefix for old shells

  #3439 opened Jun 17, 2025

• cpu: x64: pool: enable u8 type in fwd pooling

  #3440 opened Jun 18, 2025

• tests: benchdnn: inputs: Update gpu fwks inputs

  #3442 opened Jun 18, 2025

• xe: jit: drop proxy classes for nGEN

  #3443 opened Jun 18, 2025

• cpu: x64: enable fp8 support in reorder on NVL

  #3444 opened Jun 18, 2025

• Adding fp8 to ukernel documentation

  #3445 opened Jun 18, 2025

• xe: jit: refactor tensor/tile/coord usage

  #3446 opened Jun 18, 2025

• [GPU] xe: jit: gemm: fix Xe2 FHS strategy regression on LNL

  #3447 opened Jun 19, 2025

• benchdnn: graph: support non-contiguous tensor validation for matmul and strides rewriting

  #3448 opened Jun 19, 2025

• cpu: risc-v: pooling: further optimize the rv64 maxpool

  #3449 opened Jun 19, 2025

• xe: softmax: correct src/dst scale in vectorized kernel

  #3451 opened Jun 19, 2025

• graph: backend: dnnl: cleanup backend code

  #3452 opened Jun 20, 2025

3 Issues opened by 3 people

• Floating point exception in brg_conv_fwd:avx10_1_512_amx with post-ops involving broadcasted tensors

  #3450 opened Jun 19, 2025

• s8s8s32 brgconv_1x1:avx512_core_vnni precision problem

  #3435 opened Jun 17, 2025

• assume_buffer_outlives_graph compilation failed

  #3433 opened Jun 17, 2025

37 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.

• cpu: aarch64: Update ACL reorder API

  #2992 commented on Jun 19, 2025 • 31 new comments

• xe: sdpa: pass scale as a scalar kernel parameter (host side scalar memory descriptors)

  #3412 commented on Jun 19, 2025 • 24 new comments

• graph: doc, interface, backend: support SDPA training

  #3396 commented on Jun 20, 2025 • 19 new comments

• [Graph| tests, example, doc] Add GQA v2 support for implicit causal mask and example, doc update

  #3409 commented on Jun 20, 2025 • 10 new comments

• graph: backend: dnnl: relax restriction for compressed SDPA dispatch and improve validation

  #3227 commented on Jun 20, 2025 • 8 new comments

• cpu: aarch64: eltwise: make jit eltwise vector length agnostic

  #3378 commented on Jun 18, 2025 • 7 new comments

• doc: build: updates to build from source

  #3380 commented on Jun 19, 2025 • 5 new comments

• cpu: aarch64: extend brgemm conv to sve_128

  #3363 commented on Jun 16, 2025 • 5 new comments

• [GPU]Batched GEMM scale support

  #3387 commented on Jun 18, 2025 • 4 new comments

• graph: handle different engine instances in compied parition cache

  #3413 commented on Jun 18, 2025 • 4 new comments

• cpu: aarch64: brgemm: Add support for int8 in brgemm kernel

  #3414 commented on Jun 19, 2025 • 3 new comments

• gpu: intel: sycl: add support for kernel compilation

  #2988 commented on Jun 18, 2025 • 3 new comments

• xe: jit: gemm: downstream gemmstone

  #3390 commented on Jun 17, 2025 • 3 new comments

• tests: benchdnn: graph: add skip logic for NV GPU

  #3331 commented on Jun 20, 2025 • 3 new comments

• cpu: aarch64: modify acl_pooling for stateless functions

  #2849 commented on Jun 18, 2025 • 2 new comments

• aarch64: matmul: Enabling variable N block sizes for jit int8 matmul

  #3348 commented on Jun 18, 2025 • 2 new comments

• benchdnn: memory: gpu: enable support for RNG memory fill

  #3336 commented on Jun 19, 2025 • 1 new comment

• example: add backward propagation to vanilla rnn example

  #3329 commented on Jun 20, 2025 • 1 new comment

• rfcs: host-side scalars support

  #3236 commented on Jun 17, 2025 • 1 new comment

• common: verbose: asynchronous verbose mode for execution time tracking

  #3055 commented on Jun 16, 2025 • 1 new comment

• xe: conv: check GRF access bounds in release build

  #3394 commented on Jun 18, 2025 • 0 new comments

• rfcs: proposal for an asynchronous verbose mode

  #3393 commented on Jun 16, 2025 • 0 new comments

• graph: backend: dnnl: make matmul use any layout format only for constant cases

  #3398 commented on Jun 20, 2025 • 0 new comments

• xe: jit: enable lazy signal header allocation

  #3402 commented on Jun 19, 2025 • 0 new comments

• xelpg: jit: gemm: additional f16 accumulation strategies

  #3417 commented on Jun 16, 2025 • 0 new comments

• xe: sdpa: fix memory leak in internal sdpa tests

  #3418 commented on Jun 16, 2025 • 0 new comments

• src: gpu: intel: remove gen9-gen11 code as obsolete

  #3392 commented on Jun 17, 2025 • 0 new comments

• graph: backend: dnnl: support conv + xnary + swish pattern

  #3391 commented on Jun 20, 2025 • 0 new comments

• cpu: rnn: remove unnecessary memory allocations

  #3389 commented on Jun 19, 2025 • 0 new comments

• cpu: aarch64: extend brdgmm to support sve_128

  #3388 commented on Jun 19, 2025 • 0 new comments

• Fix performance regressions in RNN

  #3291 commented on Jun 18, 2025 • 0 new comments

• rfcs: graph api: support SDPA training

  #3233 commented on Jun 19, 2025 • 0 new comments

• [PoC, do not merge] src: gpu: intel: add user-supplied precomputed zps to gemm

  #3222 commented on Jun 17, 2025 • 0 new comments

• benchdnn: graph: support validation through select primitive

  #3157 commented on Jun 20, 2025 • 0 new comments

• [GPU] Expand matmul decomp cases

  #2916 commented on Jun 16, 2025 • 0 new comments

• build: removed -Wno-deprecated-declarations option for host compiler

  #2760 commented on Jun 17, 2025 • 0 new comments

• Example of dnnl::vanilla_rnn_backward

  #3257 commented on Jun 20, 2025 • 0 new comments