On Fri, 12 Jul 2019, user...@yahoo.com wrote: > > Please see attached "all.tar.xz", which contains the following files: > > > > "x11perf -all" tests: > > > > 1) x11perf_8_fbdev.txt : Debian 8.11, mach64 removed > > 2) x11perf_8_mach64.txt : Debian 8.11, mach64 installed > > 3) x11perf_sid_fbdev.txt : Debian sid, mach64 removed > > 4) x11perf_sid_mach64.txt : Debian sid, mach64 installed > > 5) x11perfcomp.txt : comparison of the above four tests > > > > "/var/log/Xorg.0.log" files: > > > > 6) Xorg_8_fbdev.log : Xorg log file for test (1) > > 7) Xorg_8_mach64.log : Xorg log file for test (2) > > 8) Xorg_sid_fbdev.log : Xorg log file for test (3) > > 9) Xorg_sid_mach64.log : Xorg log file for test (4) > > > > "glxgears" (from mesa-utils) tests: > > > > 10) glxgears.txt > > ... > > I added two tests. Please see the attached "update.tar.xz", which > contains the following six files: > > "x11perf -all" tests: > > 11) x11perf_sid_fbdev-1.txt : Debian sid, mach64 removed, multiuser > 12) x11perf_sid_mach64-1.txt : Debian sid, mach64 installed, multiuser > 13) x11perfcomp_sid.txt : comparison of 3), 4), 11) and 12) > > "/var/log/Xorg.0.log" files: > > 14) Xorg_sid_fbdev-1.log : Xorg log file for 11) > 15) Xorg_sid_mach64-1.log : Xorg log file for 12) > > serial console log: > > 16) console_log.txt : console log file for 11) and 12) >
Stan's x11perf benchmarks using fbdev show some serious regressions between xorg-server 2:1.16.4-1+deb8u2 and xorg-server 2:1.20.4-1. The following test scores went backwards by more than 10 %. 34 % 500x500 tiled rectangle (216x208 tile) 30 % 100-pixel circle 28 % 500x500 tiled rectangle (161x145 tile) 26 % Circulate Unmapped window (200 kids) 26 % 500x500 tiled rectangle (17x15 tile) 25 % 100-pixel partial circle 25 % 100-pixel ellipse 25 % 100-pixel dashed circle 24 % Copy 500x500 from window to window 24 % 500x500 opaque stippled rectangle (17x15 stipple) 24 % 100-pixel wide ellipse 23 % Composite 500x500 from window to window 21 % 100x10 wide horizontal line segment 21 % 100-pixel horizontal line segment 20 % Char in 80-char aa line (Charter 10) 19 % Scroll 500x500 pixels 18 % Fill 100x100 equivalent triangle 17 % Char in 70-char line (8x13) 17 % 100-pixel fill chord partial ellipse 16 % 100-pixel dashed ellipse 15 % 100-pixel double-dashed ellipse 15 % 1-pixel solid circle 14 % 10-pixel vertical line segment 13 % PutImage XY 100x100 square 13 % Fill 300x300 tiled trapezoid (17x15 tile) 13 % Fill 100x100 equivalent complex polygons 12 % ShmGetImage XY 10x10 square 12 % Fill 100x100 trapezoid 12 % Char in 80-char rgb line (Charter 10) 12 % Char in 30-char line (TR 24) 12 % 100-pixel fill slice partial ellipse 11 % PutImage XY 500x500 square 11 % 500x500 rectangle 11 % 500x500 opaque stippled rectangle (161x145 stipple) 11 % 10x10 tiled rectangle (161x145 tile) 10 % Copy 500x500 1-bit deep plane 10 % Char in 30-char aa line (Charter 24) 10 % 500-pixel vertical line segment 10 % 10-pixel dashed segment Those are the worst regressions. In total, 40% of x11perf measurements have regressed (though there's probably some measurement variation). Michel, since this is fbdev not mach64, the bug you found cannot have caused this. Is this problem confined to powerpc or does it also appear in other Xorg regression tests? --
