hexagon

This release addresses a CPY fence bug in the hexagon backend. It also provides pre-compiled binaries for various platforms and configurations.

This release introduces significant enhancements to the Hexagon backend, including multi-NPU device support, a fully asynchronous backend, and improved performance for various operations like ALLREDUCE and matrix multiplications. It also includes numerous bug fixes and optimizations across different Hexagon modules.

This release includes a fix for a Windows crash related to op_poll. It also provides pre-compiled binaries for various platforms and configurations.

This release includes a bug fix for the Hexagon backend, improving tensor type checking when reusing descriptors. It also provides pre-compiled binaries for various platforms and hardware accelerators.

This release introduces a new CLAMP operation for hexagon and provides updated pre-built binaries for various platforms including macOS, Linux, Android, and Windows with support for multiple hardware accelerators and software stacks.

This release introduces a significant rework of the L2 cache handling in the hexagon backend, focusing on dirty bit tracking and lazy flushing for improved performance. It also includes numerous updates and fixes across various hexagon and HMX components, enhancing thread management, DMA operations, and overall robustness.

This release introduces significant enhancements to the hexagon backend, specifically adding VISION RoPE support and improving handling of strided and non-contiguous memory operations within hex-rope.

This release focuses heavily on internal optimizations and robustness improvements for the hexagon backend, particularly for matrix multiplication and attention kernels, alongside asynchronous queue handling and workpool enhancements.

This release introduces a new GELU Quick implementation for the hexagon backend and provides updated pre-built binaries for various platforms including macOS, Linux, Android, and Windows.

This release introduces significant performance enhancements for Hexagon devices by adding Q4_1 quantization support for matrix operations. Several related bugs in buffer sizing and fp16 calculations were also resolved.

This release introduces support for the TRI operation within the Hexagon backend, alongside various cleanups and fixes related to merge conflicts and configuration errors in the Hexagon and GGML components.

This release focuses on internal optimizations within the Hexagon and HMX backends, specifically eliminating scalar VTCM loads and improving scale handling.

This release introduces support for non-contiguous row tensors in hexagon backend unary operations and provides numerous pre-compiled binaries for macOS, Linux, Android, Windows, and openEuler platforms.

This release enhances the hexagon backend by making vmem size configurable, adding vmem space measurement, and fixing related scripts. It also provides numerous pre-built binaries for diverse hardware and software stacks.

This release introduces a guard for the HMX clock request targeting v75+ platforms and provides numerous pre-compiled binaries across various operating systems and hardware configurations.

This release updates the Hexagon module by bumping the HMX frequency to its maximum corner setting and resolves a logging error in hex-mm.

This release introduces the SOLVE_TRI operation to the Hexagon backend and includes various performance and cleanup improvements within the Hexagon and GGML layers. Numerous pre-built binaries are provided for broad platform compatibility.

This release introduces support for the FILL operation on the hexagon backend and provides numerous pre-compiled binaries for diverse operating systems and hardware configurations.

This release introduces the DIAG operation and HVX support for Hexagon, alongside various bug fixes and extensive pre-compiled binaries for multiple platforms.

This release focuses heavily on expanding hardware and OS support, particularly adding Linux on Snapdragon support via the hexagon backend, and updating build configurations and documentation.

This release significantly improves hexagon performance through op request batching, buffer management rewrite, and explicit L2 cache control. It also removes the deprecated GGML_HEXAGON_EXPERIMENTAL environment variable.

This release includes a minor optimization for argosrt output initialization within the hexagon component and provides updated binary distributions for macOS, Linux, Windows, and openEuler targeting various CPU/GPU architectures.

This release introduces cumsum operation support for the hexagon backend, including DMA enablement, and resolves minor line-ending issues.

This release focuses on general DMA and Binary Operation fixes for the Hexagon backend, improving handling of large strides and resolving issues in ssm-conv. It also introduces platform-specific binary updates for broad compatibility.

This release focuses on Q4_0 and MXFP4 repack fixes within the hexagon backend, optimizing matrix multiplication kernels and resolving tail corruption issues.

This release focuses heavily on backend improvements for the hexagon architecture, specifically adding and refining the f32 ssm_conv operation and its HVX kernel implementation. It also includes various pre-compiled binaries for broad platform support.

This release introduces fp16 support for core binary operations on hexagon and includes fixes for older hardware compatibility and documentation links.

This release focuses on significant optimization and tuning of matmul and dot kernels within the hexagon backend, including new 2x2 kernel implementations and internal refactoring of quantization and matmul entry points.

This release introduces several new operations (ARGSORT, DIV, SQR, SQRT, SUM_ROWS, GEGLU) to the hexagon backend and optimizes existing binary operations using DMA.

This release focuses heavily on enabling and improving Hexagon offloading support on Windows on Snapdragon (WoS), including build system enhancements, driver refactoring, and auto-signing capabilities. A bug fix was also implemented regarding quantization checks for src1 reuse logic.