[Bug 101064] 2-3% performance drop in GpuTest v0.7 FurMark with "nir/copy_prop: Respect the source's number of components"

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Tue May 16 16:22:54 UTC 2017


            Bug ID: 101064
           Summary: 2-3% performance drop in GpuTest v0.7 FurMark with
                    "nir/copy_prop: Respect the source's number of
           Product: Mesa
           Version: git
          Hardware: Other
                OS: All
            Status: NEW
          Severity: normal
          Priority: medium
         Component: Drivers/DRI/i965
          Assignee: intel-3d-bugs at lists.freedesktop.org
          Reporter: eero.t.tamminen at intel.com
        QA Contact: intel-3d-bugs at lists.freedesktop.org

This commit regressed GpuTest v0.7 FurMark performance by 2-3% on GEN9+:
commit 3c312be7b3e95ec7540e98abed1b6f3cc8d31b2a
Author:     Jason Ekstrand <jason.ekstrand at intel.com>
AuthorDate: Thu Mar 2 17:10:24 2017 -0800
Commit:     Jason Ekstrand <jason.ekstrand at intel.com>
CommitDate: Tue Mar 14 07:36:20 2017 -0700

    nir/copy_prop: Respect the source's number of components

    In the near future we are going to require that the num_components in a
    src dereference match the num_components of the SSA value being
    dereferenced.  To do that, we need copy_prop to not remove our MOVs from
    a larger SSA value into an instruction that uses fewer channels.

    Because we suddenly have to know how many components each source has,
    this makes the pass a bit more complicated.  Fortunately, copy
    propagation is the only pass that cares about the number of components
    are read by any given source so it's fairly contained.

    Shader-db results on Sky Lake:

       total instructions in shared programs: 13318947 -> 13320265 (0.01%)
       instructions in affected programs: 260633 -> 261951 (0.51%)
       helped: 324
       HURT: 1027

On few of the GEN8+ platforms, that commit also:
* Regressed SynMark v7 PSBump2 performance by ~1%
* Increased SynMark v7 PSBump8 performance by ~1%

Timothy's commits 2 months later:
* i965: remove GLSL IR optimisation loop
* nir: shuffle constants to the top

Compensated for the FurMark drop (SynMark tests were not affected by them).

If those commits fixed what this regressed, this bug should be closed.  If they
improved something else, it may be better to keep this open.

(Sorry that it took so long to bisect this, the change was too small to be
bisected on devices with slightly larger variance.)

You are receiving this mail because:
You are the assignee for the bug.
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/20170516/686bfea8/attachment.html>

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