AMD RDNA4 Paired Context Reg Feature Merged For RADV To Potentially Help Performance
([Radeon] 6 Hours Ago
AMD RDNA4)
- Reference: 0001540512
- News link: https://www.phoronix.com/news/AMD-RDNA4-Paired-Context-Regs
- Source link:
After various commits to Mesa 25.1 in recent days in preparing for a new paired context registers feature of AMD RDNA4 (GFX12 graphics), the feature today is wired up for the Radeon Vulkan driver "RADV" with the upcoming Mesa 25.1 release. This feature has the possibility of helping performance for new AMD RDNA4 hardware but the performance benefits at this time aren't clear.
RADV lead developer Samuel Pitoiset of Valve's Linux graphics driver team has been working on this new performance feature with AMD RDNA4 graphics with the Radeon RX 9070 series. Pitoiset explains of it:
"CP is very slow on GFX12 and parsing the packet header is the main bottleneck. Using paired context regs reduce the number of packet headers and it should be more optimal.
It doesn't seem worth when only one context reg is emitted (one packet header and same number of DWORDS) or when consecutive context regs are emitted (would increase the number of DWORDS)."
So the command processor (CP) is very slow on the new AMD GPUs and the paired context registers should help reduce the main bottleneck.
[1]This merge request today landed the support for the RADV Vukan driver to use paired context regs when in an optimal state. But as brought up in the benchmarks on that merge request, in the few tests done so far there doesn't appear to be any solid benefit of this feature -- at least in its current state as wired up for RADV.
We'll see if future driver improvements make this paired context registers feature more worthwhile or not. In any event the initial enablement is now in place for Mesa 25.1 due out later this quarter.
Separately, also [2]merged today is a fix for RDNA4 / RX 9070 series to avoid potential GPU hangs or corruption "garbage" issues after suspend and resume on Linux.
[1] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34421
[2] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34347/diffs?commit_id=410f7f9f6ed4c42646be9d19e74350453f34069d
RADV lead developer Samuel Pitoiset of Valve's Linux graphics driver team has been working on this new performance feature with AMD RDNA4 graphics with the Radeon RX 9070 series. Pitoiset explains of it:
"CP is very slow on GFX12 and parsing the packet header is the main bottleneck. Using paired context regs reduce the number of packet headers and it should be more optimal.
It doesn't seem worth when only one context reg is emitted (one packet header and same number of DWORDS) or when consecutive context regs are emitted (would increase the number of DWORDS)."
So the command processor (CP) is very slow on the new AMD GPUs and the paired context registers should help reduce the main bottleneck.
[1]This merge request today landed the support for the RADV Vukan driver to use paired context regs when in an optimal state. But as brought up in the benchmarks on that merge request, in the few tests done so far there doesn't appear to be any solid benefit of this feature -- at least in its current state as wired up for RADV.
We'll see if future driver improvements make this paired context registers feature more worthwhile or not. In any event the initial enablement is now in place for Mesa 25.1 due out later this quarter.
Separately, also [2]merged today is a fix for RDNA4 / RX 9070 series to avoid potential GPU hangs or corruption "garbage" issues after suspend and resume on Linux.
[1] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34421
[2] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34347/diffs?commit_id=410f7f9f6ed4c42646be9d19e74350453f34069d
kiffmet