caramelli
commented
1 year ago Thank you for your consideration on DirectFB.
What output do you get when running the modetest tool provided by the libdrm library?
Closed kjngineering closed 1 year ago
caramelli
commented
1 year ago Thank you for your consideration on DirectFB.
What output do you get when running the modetest tool provided by the libdrm library?
kjngineering
commented
1 year ago sorry, I did do a modetest previously, I should have included it.
But I didn't look specifically at the colour modes - would I be right understanding that none of the modes listed for the planes support a RGB colour mode? (even if they do?)
# modetest
trying to open device 'i915'...failed
trying to open device 'amdgpu'...failed
trying to open device 'radeon'...failed
trying to open device 'nouveau'...failed
trying to open device 'vmwgfx'...failed
trying to open device 'omapdrm'...failed
trying to open device 'exynos'...failed
trying to open device 'tilcdc'...failed
trying to open device 'msm'...failed
trying to open device 'sti'...failed
trying to open device 'tegra'...failed
trying to open device 'imx-drm'...failed
trying to open device 'rockchip'...failed
trying to open device 'atmel-hlcdc'...failed
trying to open device 'fsl-dcu-drm'...failed
trying to open device 'vc4'...failed
trying to open device 'virtio_gpu'...failed
trying to open device 'mediatek'...failed
trying to open device 'meson'...failed
trying to open device 'pl111'...failed
trying to open device 'stm'...failed
trying to open device 'sun4i-drm'...done
Encoders:
id crtc type possible crtcs possible clones
46 0 none 0x00000001 0x00000001
Connectors:
id encoder status name size (mm) modes encoders
47 0 connected Unknown-1 154x86 1 46
modes:
index name refresh (Hz) hdisp hss hse htot vdisp vss vse vtot
#0 800x480 60.06 800 1010 1012 1056 480 502 504 525 33300 flags: nhsync, nvsync; type: preferred, driver
props:
1 EDID:
flags: immutable blob
blobs:
value:
2 DPMS:
flags: enum
enums: On=0 Standby=1 Suspend=2 Off=3
value: 0
5 link-status:
flags: enum
enums: Good=0 Bad=1
value: 0
6 non-desktop:
flags: immutable range
values: 0 1
value: 0
4 TILE:
flags: immutable blob
blobs:
value:
CRTCs:
id fb pos size
45 0 (0,0) (0x0)
#0 nan 0 0 0 0 0 0 0 0 0 flags: ; type:
props:
24 VRR_ENABLED:
flags: range
values: 0 1
value: 0
Planes:
id crtc fb CRTC x,y x,y gamma size possible crtcs
31 0 0 0,0 0,0 0 0x00000001
formats: BG16 BG24 BX12 BX15 BX24 RG16 RG24 RX12 RX15 RX24 XB15 XB12 XB24 XR15 XR12 XR24 NV16 NV12 NV21 NV61 UYVY VYUY YUYV YVYU YU11 YU12 YU16 YV11 YV12 YV16
props:
8 type:
flags: immutable enum
enums: Overlay=0 Primary=1 Cursor=2
value: 0
30 IN_FORMATS:
flags: immutable blob
blobs:
value:
01000000000000001e00000018000000
01000000900000004247313642473234
42583132425831354258323452473136
52473234525831325258313552583234
58423135584231325842323458523135
58523132585232344e5631364e563132
4e5632314e5636315559565956595559
59555956595659555955313159553132
59553136595631315956313259563136
ffffff3f000000000000000000000000
0000000000000000
in_formats blob decoded:
BG16: LINEAR
BG24: LINEAR
BX12: LINEAR
BX15: LINEAR
BX24: LINEAR
RG16: LINEAR
RG24: LINEAR
RX12: LINEAR
RX15: LINEAR
RX24: LINEAR
XB15: LINEAR
XB12: LINEAR
XB24: LINEAR
XR15: LINEAR
XR12: LINEAR
XR24: LINEAR
NV16: LINEAR
NV12: LINEAR
NV21: LINEAR
NV61: LINEAR
UYVY: LINEAR
VYUY: LINEAR
YUYV: LINEAR
YVYU: LINEAR
YU11: LINEAR
YU12: LINEAR
YU16: LINEAR
YV11: LINEAR
YV12: LINEAR
YV16: LINEAR
33 zpos:
flags: range
values: 0 2
value: 0
34 COLOR_ENCODING:
flags: enum
enums: ITU-R BT.601 YCbCr=0 ITU-R BT.709 YCbCr=1
value: 1
35 COLOR_RANGE:
flags: enum
enums: YCbCr limited range=0 YCbCr full range=1
value: 0
36 0 0 0,0 0,0 0 0x00000001
formats: BG16 BG24 BX12 BX15 BX24 RG16 RG24 RX12 RX15 RX24 XB15 XB12 XB24 XR15 XR12 XR24 NV16 NV12 NV21 NV61 UYVY VYUY YUYV YVYU YU11 YU12 YU16 YV11 YV12 YV16
props:
8 type:
flags: immutable enum
enums: Overlay=0 Primary=1 Cursor=2
value: 0
30 IN_FORMATS:
flags: immutable blob
blobs:
value:
01000000000000001e00000018000000
01000000900000004247313642473234
42583132425831354258323452473136
52473234525831325258313552583234
58423135584231325842323458523135
58523132585232344e5631364e563132
4e5632314e5636315559565956595559
59555956595659555955313159553132
59553136595631315956313259563136
ffffff3f000000000000000000000000
0000000000000000
in_formats blob decoded:
BG16: LINEAR
BG24: LINEAR
BX12: LINEAR
BX15: LINEAR
BX24: LINEAR
RG16: LINEAR
RG24: LINEAR
RX12: LINEAR
RX15: LINEAR
RX24: LINEAR
XB15: LINEAR
XB12: LINEAR
XB24: LINEAR
XR15: LINEAR
XR12: LINEAR
XR24: LINEAR
NV16: LINEAR
NV12: LINEAR
NV21: LINEAR
NV61: LINEAR
UYVY: LINEAR
VYUY: LINEAR
YUYV: LINEAR
YVYU: LINEAR
YU11: LINEAR
YU12: LINEAR
YU16: LINEAR
YV11: LINEAR
YV12: LINEAR
YV16: LINEAR
38 zpos:
flags: range
values: 0 2
value: 1
39 COLOR_ENCODING:
flags: enum
enums: ITU-R BT.601 YCbCr=0 ITU-R BT.709 YCbCr=1
value: 1
40 COLOR_RANGE:
flags: enum
enums: YCbCr limited range=0 YCbCr full range=1
value: 0
41 0 0 0,0 0,0 0 0x00000001
formats: AB15 AB12 AB24 AR15 AR12 AR24 BG16 BG24 BA15 BA12 BA24 BX24 RG16 RG24 RA12 RA15 RA24 RX24 XB24 XR24
props:
8 type:
flags: immutable enum
enums: Overlay=0 Primary=1 Cursor=2
value: 1
30 IN_FORMATS:
flags: immutable blob
blobs:
value:
01000000000000001400000018000000
01000000680000004142313541423132
41423234415231354152313241523234
42473136424732344241313542413132
42413234425832345247313652473234
52413132524131355241323452583234
5842323458523234ffff0f0000000000
00000000000000000000000000000000
in_formats blob decoded:
AB15: LINEAR
AB12: LINEAR
AB24: LINEAR
AR15: LINEAR
AR12: LINEAR
AR24: LINEAR
BG16: LINEAR
BG24: LINEAR
BA15: LINEAR
BA12: LINEAR
BA24: LINEAR
BX24: LINEAR
RG16: LINEAR
RG24: LINEAR
RA12: LINEAR
RA15: LINEAR
RA24: LINEAR
RX24: LINEAR
XB24: LINEAR
XR24: LINEAR
43 alpha:
flags: range
values: 0 65535
value: 65535
44 zpos:
flags: range
values: 0 2
value: 2
Frame buffers:
id size pitchOK, you can try with this change https://github.com/directfb2/DirectFB2/pull/109
kjngineering
commented
1 year ago That commit fixed it! Good work!
# df_andi
~~~~~~~~~~~~~~~~~~~~~~~~~~| DirectFB 2.0.0 |~~~~~~~~~~~~~~~~~~~~~~~~~~
(c) 2017-2023 DirectFB2 Open Source Project (fork of DirectFB)
(c) 2012-2016 DirectFB integrated media GmbH
(c) 2001-2016 The world wide DirectFB Open Source Community
(c) 2000-2004 Convergence (integrated media) GmbH
----------------------------------------------------------------
(*) DirectFB/Core: Single Application Core. (2023-04-10 01:02)
(*) DRMKMS/System: Using device /dev/dri/card0 (default)
(*) DRMKMS/System: Found 1 connectors, 1 encoders, 3 planes
(*) DirectFB/Input: Hot-plug detection enabled with Linux Input
(*) DirectFB/Genefx: NEON enabled
(*) DirectFB/Graphics: Genefx Software Rasterizer 0.7 (DirectFB)
(*) DRMKMS/Screen: Default mode is 800x480 (1 modes in total)
(*) DRMKMS/Layer: Supported properties for layer id 36
(*) zpos
(*) DRMKMS/Layer: Supported properties for layer id 41
(*) alpha
(*) zpos
(*) DirectFB/Core/WM: Default 0.3 (DirectFB)
(*) Direct/Interface: Loaded 'DGIFF' implementation of 'IDirectFBFont'
(*) Direct/Interface: Loaded 'DFIFF' implementation of 'IDirectFBImageProvider'Performance is not great - dfandi is getting ~9.2FPS 0.0% CPU Idle_ on DRM/KMS
FBDev implementation (using DRM FBDEV Emulation layer) is ~43FPS 0.0% CPU Idle
DirectFB(1.7) using FBDev (again DRM emulation layer) is ~27FPS 30.0% CPU Idle
kjngineering
commented
1 year ago Benchmarks using df_dok - no other changes other than one compiled for fbdev, one for DRM/KMS.
# df_dok
~~~~~~~~~~~~~~~~~~~~~~~~~~| DirectFB 2.0.0 |~~~~~~~~~~~~~~~~~~~~~~~~~~
(c) 2017-2023 DirectFB2 Open Source Project (fork of DirectFB)
(c) 2012-2016 DirectFB integrated media GmbH
(c) 2001-2016 The world wide DirectFB Open Source Community
(c) 2000-2004 Convergence (integrated media) GmbH
----------------------------------------------------------------
(*) DirectFB/Core: Single Application Core. (2023-04-10 01:02)
(*) DRMKMS/System: Using device /dev/dri/card0 (default)
(*) DRMKMS/System: Found 1 connectors, 1 encoders, 3 planes
(*) DirectFB/Input: Hot-plug detection enabled with Linux Input
(*) DirectFB/Genefx: NEON enabled
(*) DirectFB/Graphics: Genefx Software Rasterizer 0.7 (DirectFB)
(*) DRMKMS/Screen: Default mode is 800x480 (1 modes in total)
(*) DRMKMS/Layer: Supported properties for layer id 36
(*) zpos
(*) DRMKMS/Layer: Supported properties for layer id 41
(*) alpha
(*) zpos
(*) DirectFB/Core/WM: Default 0.3 (DirectFB)
(*) Direct/Interface: Loaded 'DGIFF' implementation of 'IDirectFBFont'
(*) Direct/Interface: Loaded 'DFIFF' implementation of 'IDirectFBImageProvider'
Benchmarking 256x256 on 800x459 RGB16 (16bit)...
Anti-aliased Text 3.091 secs ( 29.116 KChars/sec) [100.0%]
Anti-aliased Text (blend) 3.155 secs ( 20.538 KChars/sec) [100.0%]
Fill Rectangle 3.000 secs ( 443.460 MPixel/sec) [100.0%]
Fill Rectangle (blend) 3.330 secs ( 7.872 MPixel/sec) [100.0%]
Fill Rectangles [10] 3.018 secs ( 456.015 MPixel/sec) [100.3%]
Fill Rectangles [10] (blend) 8.257 secs ( 7.937 MPixel/sec) [100.0%]
Fill Triangles 3.004 secs ( 278.157 MPixel/sec) [100.3%]
Fill Triangles (blend) 3.054 secs ( 7.510 MPixel/sec) [100.0%]
Draw Rectangle 3.002 secs ( 30.046 KRects/sec) [100.0%]
Draw Rectangle (blend) 3.008 secs ( 2.227 KRects/sec) [100.3%]
Draw Lines [10] 3.008 secs ( 104.720 KLines/sec) [100.0%]
Draw Lines [10] (blend) 3.019 secs ( 9.274 KLines/sec) [100.0%]
Fill Spans 3.009 secs ( 413.819 MPixel/sec) [100.0%]
Fill Spans (blend) 3.385 secs ( 7.744 MPixel/sec) [100.0%]
Fill Trapezoids [10] 3.104 secs ( 380.041 MPixel/sec) [100.3%]
Blit 3.353 secs ( 11.727 MPixel/sec) [100.0%]
Blit 180 4.365 secs ( 4.504 MPixel/sec) [100.2%]
Blit colorkeyed 3.381 secs ( 5.815 MPixel/sec) [100.0%]
Blit with format conversion 3.247 secs ( 8.073 MPixel/sec) [100.0%]
Blit with colorizing 3.475 secs ( 11.315 MPixel/sec) [100.2%]
Blit from 32bit (blend) 5.627 secs ( 2.329 MPixel/sec) [100.0%]
Blit from 32bit (blend) with colorizing 3.533 secs ( 3.709 MPixel/sec) [100.0%]
Blit SrcOver (premultiplied source) 3.291 secs ( 3.982 MPixel/sec) [100.0%]
Blit SrcOver (premultiply source) 3.594 secs ( 3.646 MPixel/sec) [100.2%]
Stretch Blit 7.160 secs ( 4.487 MPixel/sec) [100.0%]
Stretch Blit colorkeyed 7.234 secs ( 4.441 MPixel/sec) [100.0%]# df_dok
~~~~~~~~~~~~~~~~~~~~~~~~~~| DirectFB 2.0.0 |~~~~~~~~~~~~~~~~~~~~~~~~~~
(c) 2017-2023 DirectFB2 Open Source Project (fork of DirectFB)
(c) 2012-2016 DirectFB integrated media GmbH
(c) 2001-2016 The world wide DirectFB Open Source Community
(c) 2000-2004 Convergence (integrated media) GmbH
----------------------------------------------------------------
(*) DirectFB/Core: Single Application Core. (2023-04-09 22:23)
(*) FBDev/System: Using device /dev/fb0 (default)
(*) FBDev/System: Found 'sun4i-drmdrmfb' (ID 0) with framebuffer at 0x00000000, 1500k (MMIO 0x00000000, 0k)
(*) DirectFB/Input: Hot-plug detection enabled with Linux Input
(*) DirectFB/Genefx: NEON enabled
(*) DirectFB/Graphics: Generic Software Rasterizer 0.7 (DirectFB)
(*) FBDev/Screen: Default mode is 800x480 (0 modes in total)
(*) DirectFB/Core/WM: Default 0.3 (DirectFB)
(*) FBDev/Mode: Setting 800x480 RGB32
(*) FBDev/Mode: Switched to 800x480 (virtual 800x480) at 32 bits (RGB32), pitch 3200
(*) Direct/Interface: Loaded 'DGIFF' implementation of 'IDirectFBFont'
(*) Direct/Interface: Loaded 'DFIFF' implementation of 'IDirectFBImageProvider'
(*) FBDev/Mode: Setting 800x480 RGB32
(*) FBDev/Mode: Switched to 800x480 (virtual 800x480) at 32 bits (RGB32), pitch 3200
Benchmarking 256x256 on 800x459 RGB32 (32bit)...
Anti-aliased Text 3.000 secs ( 92.400 KChars/sec) [100.0%]
Anti-aliased Text (blend) 3.159 secs ( 20.512 KChars/sec) [100.3%]
Fill Rectangle 3.007 secs ( 217.944 MPixel/sec) [100.3%]
Fill Rectangle (blend) 3.511 secs ( 3.733 MPixel/sec) [100.0%]
Fill Rectangles [10] 3.254 secs ( 221.541 MPixel/sec) [100.3%]
Fill Rectangles [10] (blend) 17.525 secs ( 3.739 MPixel/sec) [100.0%]
Fill Triangles 3.017 secs ( 170.519 MPixel/sec) [100.3%]
Fill Triangles (blend) 3.539 secs ( 3.703 MPixel/sec) [100.2%]
Draw Rectangle 3.003 secs ( 28.837 KRects/sec) [100.0%]
Draw Rectangle (blend) 3.046 secs ( 2.035 KRects/sec) [100.0%]
Draw Lines [10] 3.006 secs ( 99.800 KLines/sec) [100.3%]
Draw Lines [10] (blend) 3.038 secs ( 9.874 KLines/sec) [100.3%]
Fill Spans 3.006 secs ( 252.900 MPixel/sec) [100.3%]
Fill Spans (blend) 3.525 secs ( 3.718 MPixel/sec) [100.0%]
Fill Trapezoids [10] 3.218 secs ( 203.654 MPixel/sec) [100.0%]
Blit 3.040 secs ( 107.789 MPixel/sec) [100.0%]
Blit 180 3.001 secs ( 96.087 MPixel/sec) [100.0%]
Blit colorkeyed 3.049 secs ( 90.275 MPixel/sec) [100.0%]
Blit with format conversion 3.062 secs ( 34.244 MPixel/sec) [100.0%]
Blit with colorizing 3.283 secs ( 19.962 MPixel/sec) [100.0%]
Blit from 32bit (blend) 3.229 secs ( 4.059 MPixel/sec) [100.3%]
Blit from 32bit (blend) with colorizing 4.110 secs ( 3.189 MPixel/sec) [100.0%]
Blit SrcOver (premultiplied source) 3.600 secs ( 7.281 MPixel/sec) [100.0%]
Blit SrcOver (premultiply source) 3.016 secs ( 21.729 MPixel/sec) [100.0%]
Stretch Blit 3.000 secs ( 112.313 MPixel/sec) [100.0%]
Stretch Blit colorkeyed 3.007 secs ( 82.687 MPixel/sec) [100.0%]Blit performance and alpha seems to really struggle in DRM/KMS.
Fbdev is a huge performance increase on DirectFB(1.7)
Let me know if you would like any more tests run, I can build and test quite quickly now.
Also massive shoutout to @fifteenhex for the Buildroot packages for DFB2 - I would highly recommend switching to official sources and committing them to the buildroot repository.
caramelli
commented
1 year ago Interesting, these differences are consistent with results I get.
I guess we can get good performance if the drmkms system module is combined with a GFX driver module for hardware acceleration. But comparing to the fbdev system module like you did, I'm also convinced that this drmkms implementation could be improved when no GFX driver module is available. It's on my TODO list, and any help making it is welcome.
Note that if you can access a 3D GPU via OpenGL ES, the DirectFB2-eglgbm system module combined with the DirectFB2-gles2 GFX driver provide great performance with the df_andi and df_dok examples.
kjngineering
commented
1 year ago @caramelli
But comparing to the fbdev system module like you did, I'm also convinced that this drmkms implementation could be ? improved when no GFX driver module is available. It's on my TODO list, and any help making it is welcome.
Well in this test case the implementation as I understand is actually a FBDEV emulation shim (DRM_FBDEV_EMULATUON) provided by DRM, so I had expected DRM to be quicker on the basis that it is theoretically a shorter data pathway/less data manipulation.
/dev/fb0 ----> [emulation shim] ----> /dev/card0 ----> [display pipeline/mixer] ----> panel etc.
Either way the fbdev is pretty quick regardless, so its at least a means to an end. At the moment its the only practical option to use SDL2 in small unaccelerated devices without running a full mesa stack with the software rateriser option enabled, since they depreciated a fbdev endpoint.
This particular processor doesn't have any GPU, but I have another devkit coming which has a Mali 400 MP2 (LIMA) which I will test in due course and try and provide some results.
fifteenhex
commented
1 year ago Also massive shoutout to @fifteenhex for the Buildroot packages for DFB2 - I would highly recommend switching to official sources and committing them to the buildroot repository.
Yeah, that would make my life easier as getting SDL/SDL2 to link against DFB2 with DFB2 being external is a pain. If you want to have a go submitting to buildroot be my guest.. If I do it we're talking a few months until I have time.
BTW.. I think most Allwinner chips have a 2D blitter? You could write a driver for that like I did for the MStar/Sigmastar GE and that should help a lot.
Also from what I remember of the code there seems to be code paths for handling stereo displays etc. I think there might be some gains in making all of that stuff configurable or maybe just dropping it as I doubt anyone will step up to test it works.
kjngineering
commented
1 year ago BTW.. I think most Allwinner chips have a 2D blitter? You could write a driver for that like I did for the MStar/Sigmastar GE and that should help a lot.
The Display Engine 2.0 in my test chip has some mixer support in mainline (but not the full capability of the peripheral), based on the mode test above it supports the primary plane, plus two overlays. I assume these are the hardware blit layers? Would love to give a crack at expanding the driver, but I am most definitely hardware guy, not software.
kjngineering
commented
1 year ago Resolved with pull #109 .
Issue #110 created regarding DRM/KMS v FBDev performance issues.
caramelli
commented
3 months ago
Hi,
Thanks for reviving this project, I used DirectFB a few years ago on a project to glue SDL to a cheap SPI display on an embedded system, it wasn't the most performant thing in the world, but got the job done. The revival and improvements made thus far make this worth considering again - keep up the good work!
I've just done my first environment test and tried to do a test using direct DRM/KMS (e.g. card0) but I am now getting an segmentation fault, and it is not clear what the issue is.
Display is a RGB666/18-bit "MIPI-DPI" display which is running on an Allwinner sun4i-drm - DE2 interface.
The above test works fine if directfb2 is compiled with the fbdev option and piped to dev/fb0 (using DRM_FBDEV_EMULATUON) - and almost 2x the performance of DirectFB(1).