Intel's Vulkan Driver Is Working On A NIR Cache
As a possible performance win, Jason Ekstrand as the lead developer of the Intel ANV open-source Vulkan driver has been developing a NIR cache.
NIR, of course, being the intermediate representation used by Intel's OpenGL and Vulkan drivers. This NIR cache sits between after the SPIR-V-to-NIR conversion but before the linking stage. The intention of this new cache is to help with the overhead of link-time optimizations thanks to caching more of this optimized NIR state before the linking operation.
Unfortunately there are no performance numbers to share at this time on the NIR cache performance. Ekstrand wrote, "I have no numbers to back this series up; just intuition."
This NIR cache for now can be found on the Mesa mailing list.
NIR, of course, being the intermediate representation used by Intel's OpenGL and Vulkan drivers. This NIR cache sits between after the SPIR-V-to-NIR conversion but before the linking stage. The intention of this new cache is to help with the overhead of link-time optimizations thanks to caching more of this optimized NIR state before the linking operation.
Unfortunately there are no performance numbers to share at this time on the NIR cache performance. Ekstrand wrote, "I have no numbers to back this series up; just intuition."
This NIR cache for now can be found on the Mesa mailing list.
2 Comments