glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-07 Thread Michel Dänzer
Test case: 1. Start 'naked' Xorg with an xterm: xinit =xterm -- :1 -retro & 2. Start xfwm4: DISPLAY=:1 xfwm4 & 3. Maximize and unmaximize xterm window There are at least two regressions introduced by the latest glamor series: > commit 3c34dd3603989c0365654ca1b6809395c7f3b169 > A

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-07 Thread Keith Packard
Michel Dänzer writes: > Test case: > > 1. Start 'naked' Xorg with an xterm: xinit =xterm -- :1 -retro & > 2. Start xfwm4: DISPLAY=:1 xfwm4 & > 3. Maximize and unmaximize xterm window > > > There are at least two regressions introduced by the latest glamor > series: Can you try run

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-07 Thread Michel Dänzer
On Mon, 2014-04-07 at 11:27 -0700, Keith Packard wrote: > Michel Dänzer writes: > > > Test case: > > > > 1. Start 'naked' Xorg with an xterm: xinit =xterm -- :1 -retro & > > 2. Start xfwm4: DISPLAY=:1 xfwm4 & > > 3. Maximize and unmaximize xterm window > > > > > > There are at leas

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-07 Thread Michel Dänzer
On Mon, 2014-04-07 at 18:33 +0900, Michel Dänzer wrote: > Test case: > > 1. Start 'naked' Xorg with an xterm: xinit =xterm -- :1 -retro & > 2. Start xfwm4: DISPLAY=:1 xfwm4 & > 3. Maximize and unmaximize xterm window > > > There are at least two regressions introduced by the lates

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-07 Thread Keith Packard
Michel Dänzer writes: > I've come to realize it's coincidence that the text appears at step 2: > The problem is that a lot of time is spent in glamor_font_get() -> > glTexSubImage2D(). It would probably be more efficient to pack all the > glyphs in malloc()ed memory and pass that to glTexImage2D(

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-07 Thread Michel Dänzer
On Mon, 2014-04-07 at 23:39 -0700, Keith Packard wrote: > Michel Dänzer writes: > > > I've come to realize it's coincidence that the text appears at step 2: > > The problem is that a lot of time is spent in glamor_font_get() -> > > glTexSubImage2D(). It would probably be more efficient to pack al

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-08 Thread Keith Packard
Michel Dänzer writes: > Current xserver master takes on the order of ten seconds to start up > xterm on my Kaveri box. I'm sure you agree that's not acceptable. :) That's gotta be a performance problem in the GL driver then; we're uploading on the order of 200 teeny little textures. Perhaps you

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-08 Thread Michel Dänzer
On Die, 2014-04-08 at 00:00 -0700, Keith Packard wrote: > Michel Dänzer writes: > > > Current xserver master takes on the order of ten seconds to start up > > xterm on my Kaveri box. I'm sure you agree that's not acceptable. :) > > That's gotta be a performance problem in the GL driver then; we'

Re: glamor regressions with radeonsi (Re: xserver: Branch 'master' - 15 commits)

2014-04-08 Thread Keith Packard
Michel Dänzer writes: > I think the glamor code is doing conceptually the wrong thing anyway: > glTexSubImage is intended for changing parts of an existing texture. I'd hate to try and understand the 'intent' of any standard as complex as this, but glTexSubImage seemed to me the 'obvious' way to