[Bug 103847] [GEN8]=?UTF-8?Q?=C2=A0Large=20perf=20drop=20in=20windowed?=/composited tests performance

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Wed Nov 22 16:53:50 UTC 2017


https://bugs.freedesktop.org/show_bug.cgi?id=103847

            Bug ID: 103847
           Summary: [GEN8] Large perf drop in windowed/composited tests
                    performance
           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

SW setup:
- Mesa git
- X server git
  - modesetting driver
- drm-tip kernel v4.14 (which has somewhat worse texture read perf than 4.13)
- everything else: Ubuntu 16.04 (Unity/compiz)
- whole system is using the Mesa git version, not just test programs

Between following Mesa commits:
2017-11-13  257edb5b9aedc9fc5d5c13eb2f48a0c11d15456f  glx/dri3: Fix passing
renderType into glXCreateContext
2017-11-14  6ac2d16901927013393f873a34c717ece5014c1a  i965/fs: Fix
extract_i8/u8 to a 64-bit destination

On 1-channel BSW (N3050 NUC), performance in all windowed/composited tests
dropped very clearly.  Drop scales directly with FPS, so it was largest in test
with highest FPS:
- 25-30% in GpuTest Triangle
- 6-7% GLB 2.7 Egypt

Same effect is seen also on BDW GT3, but it's much smaller, and even smaller on
 BDW GT2. ->  I think the issue scales also depending on how memory bandwidth
limited the GPU is.

There's no change in fullscreen benchmarks, only in windowed / composited ones.

>From the Mesa change log between these commits, external BO and MOCS changes
look most likely candinates.

(Not bisecting this, just reporting it in hopes that somebody else can
reproduce  & bisect it.)

-- 
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/20171122/d4569041/attachment.html>


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