Mesa 26.1 adds VK_KHR_copy_memory_indirect support to RADV, enabling GPU-accelerated memory copies using indirect parameters for improved performance in Vulkan applications.
The Radeon "RADV" Vulkan driver has finally gained support for the VK_KHR_copy_memory_indirect extension in Mesa 26.1, marking a significant milestone for AMD's open-source graphics stack. This extension, introduced to the Vulkan API last September as part of Vulkan 1.4.328, enables GPU-accelerated memory copies using indirect parameters read from buffers during execution.
What VK_KHR_copy_memory_indirect Enables
The VK_KHR_copy_memory_indirect extension allows developers to perform copies between memory and image regions using parameters that are read by the device from a buffer during execution. This capability is particularly valuable for scenarios where copy operations need to be dynamic or data-driven, eliminating the need for CPU intervention in setting up copy parameters.
For RADV users, this means improved performance in applications that leverage indirect memory operations, as these operations can now be handled entirely on the GPU without CPU overhead. The extension is especially relevant for compute workloads, game engines, and other graphics applications that require flexible memory management.
Development Timeline and Contributors
The extension's journey to RADV support was a collaborative effort spanning several months. Introduced in September 2025, VK_KHR_copy_memory_indirect was developed by a cross-industry team including Valve's Mike Blumenkrantz and Alyssa Rosenzweig, along with engineers from NVIDIA, AMD, Collabora, Qualcomm, and LunarG.
A Mesa merge request was opened on the same day the extension was announced, but the implementation required extensive work to reach completion. The feature has now landed in Mesa 26.1-devel, with the final merge occurring in March 2026.
Current Support Status
With RADV's implementation in Mesa 26.1, VK_KHR_copy_memory_indirect is now supported across multiple Vulkan drivers:
- RADV (Mesa 26.1+): AMD Radeon GPUs
- Lavapipe: Mesa's software Vulkan driver
- NVK: Mesa's NVIDIA driver (already supported)
- NVIDIA Proprietary: The official NVIDIA Vulkan driver
Intel's ANV driver is also working on support, with an open merge request currently in progress.
Performance Implications
While specific benchmarks for RADV's implementation aren't yet available, the extension's impact on performance can be significant. By enabling GPU-side indirect memory operations, applications can reduce CPU-GPU synchronization overhead and potentially achieve better memory bandwidth utilization.
For developers targeting Vulkan, this means RADV now offers feature parity with other major Vulkan implementations for indirect memory operations. This is particularly important for cross-platform applications that need consistent behavior across different hardware vendors.
Availability
The feature will be available in Mesa 26.1, which is scheduled for release in the coming months. Users can expect to see this support in distributions that ship Mesa 26.1 or later, with backports potentially available for some stable branches depending on vendor policies.
For those interested in testing the feature early, development snapshots of Mesa 26.1-devel are available, though users should be aware that development versions may contain regressions or instability.
This implementation represents another step forward in closing the feature gap between RADV and proprietary Vulkan drivers, continuing the trend of rapid improvement in Mesa's Vulkan support over recent years.

Comments
Please log in or register to join the discussion