Announcement

Collapse
No announcement yet.

AMD Lands OpenMAX State Tracker In Mesa Gallium3D

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • agd5f
    replied
    Originally posted by droste View Post
    Similar symptom (smc rejects the request), but not likely related as the hw has changed significantly between these hw generations.

    Leave a comment:


  • droste
    replied
    Originally posted by Pontostroy View Post
    Code:
    echo "high" > /sys/class/drm/card0/device/power_dpm_force_performance_level 
    bash: echo: write error: Invalid argument
    
    echo "low" > /sys/class/drm/card0/device/power_dpm_force_performance_level 
    bash: echo: write error: Invalid argument
    Looks familiar to me:
    https://bugzilla.kernel.org/show_bug.cgi?id=60523

    Leave a comment:


  • agd5f
    replied
    Originally posted by Deathsimple View Post
    I think the problem is rather that before using VCE the hardware is scaling up and down depending on 3D load as expected, but after using VCE for the first time that doesn't seem to work any more.

    Ideas?
    Ah, ok. Not sure off hand. I'll need to dig into it a bit more.

    Leave a comment:


  • Pontostroy
    replied
    Originally posted by agd5f View Post
    Those parameters just adjust the valid ranges of clocks. The clock level still scales with GPU (shader) load. For decode or encode, there is fairly minimal 3d work, so the clocks may not scale up very much. If you want to force the highest clocks, regardless of state (battery/performance), you need to force the performance level (low, high, auto). Auto is the default and allows the GPU to scale based on load.

    Code:
    each high > /sys/class/drm/card0/device/power_dpm_force_performance_level
    Code:
    echo "performance" > /sys/class/drm/card0/device/power_dpm_state
    
    cat /sys/class/drm/card0/device/power_dpm_state
    performance
    
    echo "high" > /sys/class/drm/card0/device/power_dpm_force_performance_level 
    bash: echo: write error: Invalid argument
    
    echo "low" > /sys/class/drm/card0/device/power_dpm_force_performance_level 
    bash: echo: write error: Invalid argument
    
    
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 30000 mclk: 15000

    Leave a comment:


  • Deathsimple
    replied
    Originally posted by agd5f View Post
    Those parameters just adjust the valid ranges of clocks. The clock level still scales with GPU (shader) load. For decode or encode, there is fairly minimal 3d work, so the clocks may not scale up very much. If you want to force the highest clocks, regardless of state (battery/performance), you need to force the performance level (low, high, auto). Auto is the default and allows the GPU to scale based on load.
    I think the problem is rather that before using VCE the hardware is scaling up and down depending on 3D load as expected, but after using VCE for the first time that doesn't seem to work any more.

    Ideas?

    Leave a comment:


  • agd5f
    replied
    Originally posted by Pontostroy View Post
    Code:
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 30000 mclk: 15000
    
    echo "performance" > /sys/class/drm/card0/device/power_dpm_state
    
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 30714 mclk: 15000
    
    
    @rec
    cat /sys/kernel/debug/dri/0/radeon_pm_info                                                                                                         
    power level avg    sclk: 35273 mclk: 150000        
    
    
    @rec off
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 35149 mclk: 150000
    
    
    cat /sys/class/drm/card0/device/power_dpm_state
    performance
    
    
    echo "battery" > /sys/class/drm/card0/device/power_dpm_state
    
    
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 35149 mclk: 150000
    Those parameters just adjust the valid ranges of clocks. The clock level still scales with GPU (shader) load. For decode or encode, there is fairly minimal 3d work, so the clocks may not scale up very much. If you want to force the highest clocks, regardless of state (battery/performance), you need to force the performance level (low, high, auto). Auto is the default and allows the GPU to scale based on load.

    Code:
    each high > /sys/class/drm/card0/device/power_dpm_force_performance_level

    Leave a comment:


  • Pontostroy
    replied
    Code:
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 30000 mclk: 15000
    
    echo "performance" > /sys/class/drm/card0/device/power_dpm_state
    
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 30714 mclk: 15000
    
    
    @rec
    cat /sys/kernel/debug/dri/0/radeon_pm_info                                                                                                         
    power level avg    sclk: 35273 mclk: 150000        
    
    
    @rec off
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 35149 mclk: 150000
    
    
    cat /sys/class/drm/card0/device/power_dpm_state
    performance
    
    
    echo "battery" > /sys/class/drm/card0/device/power_dpm_state
    
    
    cat /sys/kernel/debug/dri/0/radeon_pm_info
    power level avg    sclk: 35149 mclk: 150000

    Leave a comment:


  • Pontostroy
    replied
    Originally posted by Pontostroy View Post
    Lighsmark 2008 fullhd after restart Finished, average fps = 357.11
    Lighsmark 2008 fullhd with recording Finished, average fps = 41.52 (rec_2014-02-10_105800.avi is emty after Lighsmark run, glsgears in little window grabs fine )
    Lighsmark 2008 fullhd after recording Finished, average fps = 121.12


    I just add use-damage=0 to ximagesrc

    Lighsmark 2008 fullhd with recording Finished, average fps = 171.97 (video is fine)
    Lighsmark 2008 fullhd after recording Finished, average fps = 121.77!!!!!

    Leave a comment:


  • Pontostroy
    replied
    Originally posted by Deathsimple View Post
    Quite nice, considering how often we copy the data from VRAM to system memory and back, but I guess you didn't tried any full screen 3D application. There is still allot of things we can optimize.



    Thx, for the info. What was the last good version? Can you bisect?
    Lighsmark 2008 fullhd after restart Finished, average fps = 357.11
    Lighsmark 2008 fullhd with recording Finished, average fps = 41.52 (rec_2014-02-10_105800.avi is emty after Lighsmark run, glsgears in little window grabs fine )
    Lighsmark 2008 fullhd after recording Finished, average fps = 121.12

    Leave a comment:


  • Deathsimple
    replied
    Originally posted by Pontostroy View Post
    Now all perfect. I can grab my screen in FULLHD H264 with 5-10% cpu usage.
    Quite nice, considering how often we copy the data from VRAM to system memory and back, but I guess you didn't tried any full screen 3D application. There is still allot of things we can optimize.

    Originally posted by Pontostroy View Post
    But this kernel http://cgit.freedesktop.org/~deathsi...?h=vce-release has very poor 3d performance, up to 2-3 times slower.
    Thx, for the info. What was the last good version? Can you bisect?

    Leave a comment:

Working...
X