Video Core Source – 1 Year On

[itsmedoofer]

Hi,

Being as it’s about a year now since the nice chaps at Broadcom made the VC source available I was wondering how much of an impact its had on the average Pi user......

The Quake bounty caused a bit of a stir but I could be wrong has not gone much further and I’ve see projects using the GPU for FFT type things, I cant think or see where else it’s been used for though.

Perhaps people closer to the coalface could give us an insight into what's already been done and where things are going ?

Ta,

[jamesh]

Not heard of much going on. There is Eric Anholt's work on MESA drivers that should have a good impact, but he is just one person (and employed by Broadcom!). Herman Hermitage has been doing good work, but I don't know what the end results of that will be.

As for other stuff, it does beg the question, why were people clamouring for release and full openness when only a few are using the stuff that actually has been released....

[ghans]

Eric Anholts work should have a big impact though. It wold give us "real" OpenGL , not just OpenGL ES. This would automatically accelerate everything which uses GLX behind the scenes.

ghans

[Fidelius]

Eric Anholts work should have a big impact though. It wold give us "real" OpenGL , not just OpenGL ES. This would automatically accelerate everything which uses GLX behind the scenes.

... that sounds very well.

So also Blender would finally work? (It's in the Raspbian repository but uses OpenGL and so gives an error currently)

Would the Wayland project for the Pi also profit from a full MESA, or does it already work well with the Pi as it is? (There's some nice Wayland demo videos on Youtube with, I think Mr. Upton, and the Collabra developers.)

[jamesh]

Eric Anholts work should have a big impact though. It wold give us "real" OpenGL , not just OpenGL ES. This would automatically accelerate everything which uses GLX behind the scenes.

... that sounds very well.

So also Blender would finally work? (It's in the Raspbian repository but uses OpenGL and so gives an error currently)

Would the Wayland project for the Pi also profit from a full MESA, or does it already work well with the Pi as it is? (There's some nice Wayland demo videos on Youtube with, I think Mr. Upton, and the Collabra developers.)

Actually, Eric's work might mean less requirement for Wayland as it would, I think, allow integration of OpenGL with X. Swings and roundabouts.

[ghans]

I think that might be the reason we haven't heard much about Mayard since May last yer ..

ghans

[Fidelius]

Actually, Eric's work might mean less requirement for Wayland as it would, I think, allow integration of OpenGL with X. Swings and roundabouts.

Nice!

So now that the Pi1 to Pi2 upgraders are about to get used hardware-wise to the amazing speed of the four small but mighty ARM7 cores, software-wise the next visual speed-ups do appear on the horizon (Mess/OpenGL, Wayland) for the Pi 1 and 2.

This little machine named Raspbeery Pi is one nice hardware and software thing.

[itsmedoofer]

Hi James,

Thanks for taking the time to reply, appreciated...

Not heard of much going on. There is Eric Anholt's work on MESA drivers that should have a good impact, but he is just one person (and employed by Broadcom!). Herman Hermitage has been doing good work, but I don't know what the end results of that will be.

I’m aware of the work Eric has been doing, unfortunately though his live journal was last updated in October, digging about I can see there has been activity but unless you know what you are looking at its hard to gauge progress, I guess its a big project for one person as well....

For those interested:- http://anholt.livejournal.com/calendar

Maybe this is the key that unlocks the door ?

As for other stuff, it does beg the question, why were people clamouring for release and full openness when only a few are using the stuff that actually has been released....

Well I was thinking the same thing, I was hoping for a flurry of improved “things” based around this but apart from some niche things nothing seems to stand out.... Peter Warden is the only person who I can think of who as grasped the nettle:-

http://petewarden.com/2014/08/07/how-to ... g-its-gpu/

[ghans]

http://cgit.freedesktop.org/mesa/mesa/l ... grep&q=vc4
https://github.com/anholt/linux/commits/vc4-3.19

I think you have no choice instead of regularly pulling and rebuilding the sources for testing ... i'm pretty sure Eben himself will want to make the big
announcment when it's ready for primetime.

ghans

[jamesh]

http://cgit.freedesktop.org/mesa/mesa/l ... grep&q=vc4
https://github.com/anholt/linux/commits/vc4-3.19

I think you have no choice instead of regularly pulling and rebuilding the sources for testing ... i'm pretty sure Eben himself will want to make the big
announcment when it's ready for primetime.

ghans

Yup. I don't think it is far off IIRC my last conversation with Eben correctly!

[mimi123]

http://cgit.freedesktop.org/mesa/mesa/l ... grep&q=vc4
https://github.com/anholt/linux/commits/vc4-3.19

I think you have no choice instead of regularly pulling and rebuilding the sources for testing ... i'm pretty sure Eben himself will want to make the big
announcment when it's ready for primetime.

ghans

Yup. I don't think it is far off IIRC my last conversation with Eben correctly!

It is good enough to run Unity
I may try Blender soon.

[eriktheitalian]

Hi James,

Thanks for taking the time to reply, appreciated...

Not heard of much going on. There is Eric Anholt's work on MESA drivers that should have a good impact, but he is just one person (and employed by Broadcom!). Herman Hermitage has been doing good work, but I don't know what the end results of that will be.

I’m aware of the work Eric has been doing, unfortunately though his live journal was last updated in October, digging about I can see there has been activity but unless you know what you are looking at its hard to gauge progress, I guess its a big project for one person as well....

For those interested:- http://anholt.livejournal.com/calendar

Maybe this is the key that unlocks the door ?

As for other stuff, it does beg the question, why were people clamouring for release and full openness when only a few are using the stuff that actually has been released....

Well I was thinking the same thing, I was hoping for a flurry of improved “things” based around this but apart from some niche things nothing seems to stand out.... Peter Warden is the only person who I can think of who as grasped the nettle:-

http://petewarden.com/2014/08/07/how-to ... g-its-gpu/

http://cgit.freedesktop.org/mesa/mesa/l ... rivers/vc4

Big impact coming...

[hayderOICO]

can't wait. any ETA for us mere mortals?

[ghans]

Propably not. The foundation is very tight-lipped about ETAs.

ghans

[jamesh]

I look at it like this. The Foundation could announce a release date, and it would be released on that date. Or they could just release on that date. Either way, you get the end result on that date.

So knowing the release date in advance really has no influence on when you get it. You know it's going to turn up at some point.

[mung]

I think I made comment years back about how few people there probably are that actually need or would be able to use full open access to the raspberry pi processor details.

I guessed there are probably only a couple of hundred people that would really be interested, and a small percentage of those would actually get time to do anything.

I have had a very brief look and messed around with some of the demo example work that others have released, but really not had the time or inclination to put any real effort in. Now the rpi2 is available I probably do not need to use the extra processing of the vc4 so will not look further.

I was sort of considering creating realtime motor control driver to run on parts of the vc4, still not really sure how suitable the vc4 would be for that, but it sems definitly possible.

I think the problem is most people want an easy quick fix and learning a whole new architexture and assembly language that is not well supported seems too much work unless it has great benefits and is not achievable any other way.

Most people probably just say they will buy a more powerful ARM board and save on development costs if the project is low volume, there are probably a few hobbiests that will play with things just out of interest but they probably will not be putting much time into it so its very slow to see anything appear.

Most of the open source zealots probably have other agendas than actually creating any software when they demand access to processor details.

I hope the anholt opengl stuff makes it into raspbian soon, not sure what other infrastructure work needs to be done maybe opencl (not even sure about how that works?), or something that would allow generalised access to the 'dsp' which the vc4 supposedly has?

[mimi123]

I think I made comment years back about how few people there probably are that actually need or would be able to use full open access to the raspberry pi processor details.

I guessed there are probably only a couple of hundred people that would really be interested, and a small percentage of those would actually get time to do anything.

I have had a very brief look and messed around with some of the demo example work that others have released, but really not had the time or inclination to put any real effort in. Now the rpi2 is available I probably do not need to use the extra processing of the vc4 so will not look further.

I was sort of considering creating realtime motor control driver to run on parts of the vc4, still not really sure how suitable the vc4 would be for that, but it sems definitly possible.

I think the problem is most people want an easy quick fix and learning a whole new architexture and assembly language that is not well supported seems too much work unless it has great benefits and is not achievable any other way.

Most people probably just say they will buy a more powerful ARM board and save on development costs if the project is low volume, there are probably a few hobbiests that will play with things just out of interest but they probably will not be putting much time into it so its very slow to see anything appear.

Most of the open source zealots probably have other agendas than actually creating any software when they demand access to processor details.

I hope the anholt opengl stuff makes it into raspbian soon, not sure what other infrastructure work needs to be done maybe opencl (not even sure about how that works?), or something that would allow generalised access to the 'dsp' which the vc4 supposedly has?

Not supposedly but I'm using it right now.
IT IS NOT A DSP BUT A FULL CPU WITH DSP EXTENSIONS, DUAL CORE WITH SMP RUNNING A FULL RTOS.

[ktb]

mimi123,

What Pi model did you get it to work on? Which kernel? vc4-3.18 or vc4-3.19? Based on commits, it looked like Pi2 support was only added to vc4-3.19. I've built both, but haven't been successful in trying to use the driver on my Pi2. Can you give me any tips?

http://www.raspberrypi.org/forums/viewt ... 16#p719816

BTW, this seemed interesting https://github.com/anholt/linux/blob/bc ... -rpi-b.dts.

Code: Select all

/dts-v1/;
/include/ "bcm2836-rpi.dtsi"

/ {
	compatible = "raspberrypi,model-b-plus", "brcm,bcm2836";
	model = "Raspberry Pi 2 Model B+";

It's probably just a mistake.

[jamesh]

Just FYI, there is a dual scaler/vector core on the Pi2 VC4, plus 12 Quad processors. The Vector/Scaler cores run ThreadX as the RTOS, and control all the HW blocks in the cores. The cores can move between scaler and vectors operations at any time. Usually the cores are programmed in C for the scaler code, and assembler for the vectors. The quads have a particularly unpleasant instruction set and are not for the faint hearted.

[mimi123]

mimi123,

What Pi model did you get it to work on? Which kernel? vc4-3.18 or vc4-3.19? Based on commits, it looked like Pi2 support was only added to vc4-3.19. I've built both, but haven't been successful in trying to use the driver on my Pi2. Can you give me any tips?

http://www.raspberrypi.org/forums/viewt ... 16#p719816

BTW, this seemed interesting https://github.com/anholt/linux/blob/bc ... -rpi-b.dts.

Code: Select all

/dts-v1/;
/include/ "bcm2836-rpi.dtsi"

/ {
	compatible = "raspberrypi,model-b-plus", "brcm,bcm2836";
	model = "Raspberry Pi 2 Model B+";

It's probably just a mistake.

Pi2 with both 3.18 and 3.19. That is just a mistake. Did you enable the interrupt mask and build the DRM module as included (not a module)? Is your screen 1080p(or you have to "correct" one source file)?

[ktb]

mimi123,

What Pi model did you get it to work on? Which kernel? vc4-3.18 or vc4-3.19? Based on commits, it looked like Pi2 support was only added to vc4-3.19. I've built both, but haven't been successful in trying to use the driver on my Pi2. Can you give me any tips?

http://www.raspberrypi.org/forums/viewt ... 16#p719816

BTW, this seemed interesting https://github.com/anholt/linux/blob/bc ... -rpi-b.dts.

Code: Select all

/dts-v1/;
/include/ "bcm2836-rpi.dtsi"

/ {
	compatible = "raspberrypi,model-b-plus", "brcm,bcm2836";
	model = "Raspberry Pi 2 Model B+";

It's probably just a mistake.

Pi2 with both 3.18 and 3.19. That is just a mistake. Did you enable the interrupt mask and build the DRM module as included (not a module)? Is your screen 1080p(or you have to "correct" one source file)?

1. Yes. I added mask_gpu_interrupt0=0x400 to /boot/config.txt and the DRM module is built-in.

Code: Select all

"You need CONFIG_DRM_VC4=y (not m, or it won't load)."

> Device Drivers > Graphics support > Direct Rendering Manager

<*> Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)  --->
    I2C encoder or helper chips  --->
< > PTN3460 DP/LVDS bridge (NEW)
< > DisplayLink (NEW)
< > DRM support for Marvell Armada SoCs (NEW)
< > DRM Support for TI LCDC Display Controller (NEW)
<*> Broadcom VC4 Graphics


"You need CONFIG_CMA_SIZE_MBYTES=64 (or, apparently, cma=64M on the kernel command line), or you'll get "vc4-drm vc4-drm.0: failed to allocate buffer with size 7057408". You also, if you want to see output, need a monitor that can scan out the same 1680x1050 as me, for now."

> Device Drivers > Generic Driver Options

[*] DMA Contiguous Memory Allocator
      *** Default contiguous memory area size: ***
(64)  Size in Mega Bytes
      Selected region size (Use mega bytes value only)  --->
(8)   Maximum PAGE_SIZE order of alignment for contiguous buffers

2. I'm not sure, but maybe this is the problem. I have a 720p Samsung TV/monitor with a preferred mode of 1360x768. It supports some 1920x1080 modes and will display at 1920x1080, however It's scaled and doesn't look good. 1920x1080 appears to work OK in general, but maybe not when trying to use the VC4 driver? So, I'll give it one more attempt after modifying vc4_display.c

Code: Select all

static struct vc4_params params = {
	.width = 1360,
	.height = 768
};
...
module_param_named(width, params.width, int, 0444);
MODULE_PARM_DESC(width, "Display width (default 1360");

module_param_named(height, params.height, int, 0444);
MODULE_PARM_DESC(height, "Display height (default 768)");

I wonder... Do you see the following errors in your Xorg.0.log?

Code: Select all

[   296.024] (II) xfree86: Adding drm device (/dev/dri/card0)
[   296.028] (II) systemd-logind: got fd for /dev/dri/card0 226:0 fd 7 paused 0
[   296.028] (EE) /dev/dri/card0: failed to set DRM interface version 1.4: Invalid argument
[   296.028] (II) systemd-logind: releasing fd for 226:0
[   296.032] (II) no primary bus or device found
...
[   296.103] (EE) modesetting(0): drmSetMaster failed: Permission denied
[   296.103] (EE)
Fatal server error:
[   296.103] (EE) AddScreen/ScreenInit failed for driver 0
[   296.103] (EE)
[   296.103] (EE) 

Thank you for the advice.

[mimi123]

mimi123,

What Pi model did you get it to work on? Which kernel? vc4-3.18 or vc4-3.19? Based on commits, it looked like Pi2 support was only added to vc4-3.19. I've built both, but haven't been successful in trying to use the driver on my Pi2. Can you give me any tips?

http://www.raspberrypi.org/forums/viewt ... 16#p719816

BTW, this seemed interesting https://github.com/anholt/linux/blob/bc ... -rpi-b.dts.

Code: Select all

/dts-v1/;
/include/ "bcm2836-rpi.dtsi"

/ {
	compatible = "raspberrypi,model-b-plus", "brcm,bcm2836";
	model = "Raspberry Pi 2 Model B+";

It's probably just a mistake.

Pi2 with both 3.18 and 3.19. That is just a mistake. Did you enable the interrupt mask and build the DRM module as included (not a module)? Is your screen 1080p(or you have to "correct" one source file)?

1. Yes. I added mask_gpu_interrupt0=0x400 to /boot/config.txt and the DRM module is built-in.

Code: Select all

"You need CONFIG_DRM_VC4=y (not m, or it won't load)."

> Device Drivers > Graphics support > Direct Rendering Manager

<*> Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)  --->
    I2C encoder or helper chips  --->
< > PTN3460 DP/LVDS bridge (NEW)
< > DisplayLink (NEW)
< > DRM support for Marvell Armada SoCs (NEW)
< > DRM Support for TI LCDC Display Controller (NEW)
<*> Broadcom VC4 Graphics


"You need CONFIG_CMA_SIZE_MBYTES=64 (or, apparently, cma=64M on the kernel command line), or you'll get "vc4-drm vc4-drm.0: failed to allocate buffer with size 7057408". You also, if you want to see output, need a monitor that can scan out the same 1680x1050 as me, for now."

> Device Drivers > Generic Driver Options

[*] DMA Contiguous Memory Allocator
      *** Default contiguous memory area size: ***
(64)  Size in Mega Bytes
      Selected region size (Use mega bytes value only)  --->
(8)   Maximum PAGE_SIZE order of alignment for contiguous buffers

2. I'm not sure, but maybe this is the problem. I have a 720p Samsung TV/monitor with a preferred mode of 1360x768. It supports some 1920x1080 modes and will display at 1920x1080, however It's scaled and doesn't look good. 1920x1080 appears to work OK in general, but maybe not when trying to use the VC4 driver? So, I'll give it one more attempt after modifying vc4_display.c

Code: Select all

static struct vc4_params params = {
	.width = 1360,
	.height = 768
};
...
module_param_named(width, params.width, int, 0444);
MODULE_PARM_DESC(width, "Display width (default 1360");

module_param_named(height, params.height, int, 0444);
MODULE_PARM_DESC(height, "Display height (default 768)");

I wonder... Do you see the following errors in your Xorg.0.log?

Code: Select all

[   296.024] (II) xfree86: Adding drm device (/dev/dri/card0)
[   296.028] (II) systemd-logind: got fd for /dev/dri/card0 226:0 fd 7 paused 0
[   296.028] (EE) /dev/dri/card0: failed to set DRM interface version 1.4: Invalid argument
[   296.028] (II) systemd-logind: releasing fd for 226:0
[   296.032] (II) no primary bus or device found
...
[   296.103] (EE) modesetting(0): drmSetMaster failed: Permission denied
[   296.103] (EE)
Fatal server error:
[   296.103] (EE) AddScreen/ScreenInit failed for driver 0
[   296.103] (EE)
[   296.103] (EE) 

Thank you for the advice.

I don't have errors on the Xorg log.Did you try to run as root? Did you compile anholt's X11 module?

[ktb]

I don't have errors on the Xorg log.Did you try to run as root? Did you compile anholt's X11 module?

I don't think I have tried running the xserver as root, but the user I am using is in the video group. I will give root a try.

What is anholt's X11 module? I don't see that mentioned anywhere. I compiled the kernels from GitHub. I then compiled mesa as well as the latest xserver, xf86-input-evdev, xf86-input-synaptics, xf86-video-modesetting and even xf86-video-fbdev (as a fallback) from freedesktop.org.

[ktb]

I don't have errors on the Xorg log.Did you try to run as root? Did you compile anholt's X11 module?

I don't think I have tried running the xserver as root, but the user I am using is in the video group. I will give root a try.

What is anholt's X11 module? I don't see that mentioned anywhere. I compiled the kernels from GitHub. I then compiled mesa as well as the latest xserver, xf86-input-evdev, xf86-input-synaptics, xf86-video-modesetting and even xf86-video-fbdev (as a fallback) from freedesktop.org.

Alright. This seems interesting. I tried startx as root and it actually appears to be working. The desktop doesn't all fit on my screen (it's 1920x1080 on my 1360x768 screen), but that can probably be fixed by the width and height modification to vc4_display.c.

This still happens
[ 221.900] (EE) /dev/dri/card0: failed to set DRM interface version 1.4: Invalid argument
...
[ 221.905] (II) no primary bus or device found

However, I no longer get this error:
[ 296.103] (EE) modesetting(0): drmSetMaster failed: Permission denied
[ 296.103] (EE)
Fatal server error:
[ 296.103] (EE) AddScreen/ScreenInit failed for driver 0
[ 296.103] (EE)
[ 296.103] (EE)

I'll have to test some things to confirm, but this is exciting. Thank you very much mimi123. I probably wouldn't have tried it as root if not for your suggestion.

[AndyPi]

This is interesting and recent
https://www.youtube.com/watch?feature=p ... XDeketJNdk

came from
http://phoronix.com/scan.php?page=news_ ... 3D-VC4-LCA