[Bug 107510] [GEN8+] up to 10% perf drop on several 3D benchmarks

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Thu Nov 1 00:24:03 UTC 2018


--- Comment #15 from Timothy Arceri <t_arceri at yahoo.com.au> ---
(In reply to Eero Tamminen from comment #14)
> Because exactly same tests were affected both times this perf regression was
> introduced to Mesa, I think it's also the cause for the small regressions in
> the listed fragment shader bound tests:
> - GpuTest v0.7 Julia FP64
> - GfxBench ALU2
> - Unigine Heaven
> - SynMark PSPom
> Not just for the larger regressions in the 2 Tessellation & Geometry shader
> tests.
> Of these, PSPom regression was fixed between these commits:
> 18cc65edf8: 2018-10-15 17:56:12 i965: Drop assert about number of uniforms
> in ARB handling
> 322a919a41: 2018-10-16 11:47:55 anv: Implement VK_EXT_pci_bus_info
> (My guess is that "nir: Copy propagation between blocks" commit fixed it.)
> Timothy's latest patch series:
>   https://patchwork.freedesktop.org/series/51642/
> Fixes just the larger tessellation & geometry shader regressions, not the
> smaller fragment shader ones (expected as latest version disables the
> optimization pass for fragment stage).

Splitting and/or vectorising FS inputs seems to case the scheduler to vary
wildly. I'm seeing up to +/-50% swings in cycle counts in both directions. I'm
not sure what to do about this.

You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20181101/d34cb5e6/attachment.html>

More information about the intel-3d-bugs mailing list