Performance Optimizations

Intel 64/AMD64 Processors

• Improved fp32 matmul performance with fp4 compressed weights.
• Improved fp32 matmul performance for cases when one of the tensors has a trivial dimension on processors with Intel AVX-512 instruction set support.

Intel Graphics

• Improved fp16/bf16 matmul performance for large tensor cases on Intel Graphics for Intel Core Ultra processor Series 3 (formerly Panther Lake).
• Improved matmul performance for cases with 4-byte alignment on Intel GPUs based on Xe2 architecture.
• Improved performance of fp16/bf16 matmul with mxfp4 weights.
• Improved convolution performance with host-side scalar scales and zero points.
• Improved matmul performance for LLM inference workloads on Intel GPUs based on Xe2/Xe3 architectures.
• Improved f32 SDPA performance for small head sizes.

AArch64 Processors

• Improved performance of bf16 matmul.
• Improved performance of bf16/int8 convolutions.
• Improved matmul performance for cases when one of the tensor has a trivial dimension.
• Improved performance of s8/u8 eltwise post-ops on Arm(R) Neoverse(TM) V1 processors.
• Improved f16 and bf16 eltwise performance with abs, relu, square, sqrt, clip, and clip_v2 algorithms.
• Improved eltwise exp algorithm performance on Arm(R) Neoverse(TM) N1 processors.
• Improved reorder primitive performance.

RISC-V Processors

• Improved f32 matmul, inner product, convolution, softmax, batch normalization, layer normalization, and group normalization primitives performance.
• Improved eltwise and binary primitives performance.
• Improved f32 and fp16 pooling primitive performance.
• Improved fp32 to u8 reorder primitive performance.

Functionality

Functional API

• Introduced destination tensor dynamic quantization in matmul primitive following Open Compute Microscaling (MX) formats specification. See MXFP8 matmul tutorial for quick introduction into MX-capabilities in oneDNN.
• Introduced support for NVFP4 quantization scheme. The changes include support for fp8_e4m3 grouped scales and dynamic quantization support for destination tensor with NVFP4-specific formula for scales computation.
• Introduced support for dropout as a primitive attribute for matmul, softmax and eltwise primitives.

Graph API

• Introduced support for RMS Normalization operation.
• Introduced support for output gradient of attention mask for SDPA and GQA training.

Intel Graphics

• Introduced support for convolution with u8 weights.
• Introduced support for 2D grouped scales in fp8 and dual zero points in matmul.
• Extended support for 5D and 6D tensors in matmul with post-ops.

Intel 64/AMD64 Processors

• Introduced support for different data types of source and destination in pooling forward propagation.

AArch64 Processors

• Added limited support for the BRGEMM Microkernel API
• Added limited support for Windows on Arm builds with MSVC

Usability

Common

• Extended quantization attributes documentation to cover all quantization schemes supported by the library.
• Added matmul fp8 quantization example demonstrating use of matmul primitive with fp8 source, destination, and weights.
• Enabled ONEDNN_ENABLE_GRAPH_DUMP knob by default.

Intel 64/AMD64 Processors

• Extended oneDNN threadpool runtime with an option to support asynchronous execution and updated all CPU implementations accordingly. This extension makes oneDNN compatible with OpenXLA "thunk" runtime.
• Introduced ONEDNN_SAFE_RBP build knob that instructs x64 implementations to preserve value of rbp register for tools that rely on stack unwinding. This option may have visible performance impact on some workloads.

AArch64 Processors

• Fixed a potential overflow on AArch64 builds with Arm Compute Library.
• Significantly reduced memory consumption of convolution primitive with large spatial filters during primitive creation.

Intel Graphics

• Removed build time dependency on OpenCL runtime in SYCL build configuration.

Validation

Deprecated Functionality

• BLAS-like API including dnnl::sgemm, dnnl::gemm_u8s8s32, and dnnl::gemm_s8s8s32 functions is deprecated
  and will be removed in future releases. If you are using this API consider switching to matmul primitive.

Thanks to our Contributors

This release contains contributions from the project core team as well as Andrei Hutu @Anndrey24, Anna Sztukowska @asztukow, Arseniy Obolenskiy @aobolensk, Avanish Tiwari @Tiwari-Avanish, czekun @ZackyLake, Deeksha Kasture @kasturedeeksha, Fadi Arafeh @fadara01, Gassan Salama @gassan-arm, Henry Gardiner @henry-gar, @jstachowintel, Keanu Czirjak @keanucz, Krishna Sai @krishnasai-mcw, Murray Steele @murste01, Narendra Bagria @narenbagria, Joseph Kuo @PershingSquare, @pmanczak, @vishwascm, Yejing Lai @Yejing-Lai, 夏卓昭 @xiazhuozhao