AMD Pushes Out New R600/700 3D Code

[monraaf]

So I was experimenting a little more, added the following options to xorg.conf:

Option "EXANoComposite" "true"
Option "EXANoUploadToScreen"
Option "EXANoDownloadFromScreen"
Option "MigrationHeuristic" "greedy"

...and added a printf statement after the glFlush() from hello.c. Now I launched the hello program from a terminal and alt-tabbed between the terminal and the hello window. Each time I uncovered the hello window the text I put in the printf statement was printed to the terminal but besides the decoration put on by the window manager the hello window remained completely transparent. After a while the hello program segfaulted and about a half minute later the X server finally locked up.

[monraaf]

I do have the latest drm from agd5f's 3d branch installed. AccelMethod none doesn't do anything here, I still get normal 2D acceleration and Xv and a lock up with the redbook hello program. Maybe Richard is using radeonHD?

Instead of immediately rebooting after a lock up, this time I waited some time hoping that the server would resume normal operation after a while
It didn't...but if it's of any use my dmesg got spammed with a lot of:

[drm] wait for fifo failed status : 0xA02034AC 0x00000003

Edit: RadeonHD doesn't seem to work with agd5f's drm. No acceleration at all and Mesa Software Rasterizer.

[bridgman]

Hmm... I know it still needs work but I don't think it should be locking up the X server. I have pinged the devs for comment...

EDIT - Richard said that glxgears *does* hang the system (he's working on that) but he wouldn't expect the Redbook "Hello" program to do that. There is some corruption in "Hello" because the clear code is stubbed out though.

Just checking, are you using the latest drm from the r6xx-r7xx-3d branch of ~agd5f/drm ? Last change was about 6 days ago but Bad Things will happen with older code.

EDIT - Richard also mentioned that he is using Option "AccelMethod" "None" for testing right now, in case that helps.

[monraaf]

Richard pushed a couple of updates earlier today.

Yeah, I noticed. I have subscribed to the git rss feed The headline "redbook hello can run" grabbed my attention. Unfortunately just like glxgears it locks up the X server here.

[bridgman]

Richard pushed a couple of updates earlier today. The "clear" code is still broken but that's next on his list.

[monraaf]

Yeah, I got glxinfo running:

OpenGL vendor string: Advanced Micro Devices, Inc.
OpenGL renderer string: Mesa DRI R600 20090101 x86/MMX+/3DNow!+/SSE2 TCL
OpenGL version string: 1.4 Mesa 7.5-devel

For some reason the preprocessor didn't replace DRI_CONF_FP_OPTIMIZATION_SPEED with what it's defined to be, so I just replaced it myself with 0.
Also I had to add my chip id to radeon_set_screen_flags(). Note to AMD: 0x9610 is also an RS780!

glxgears now locks up the X server after printing some numbers related to pixel and vertex shaders .

[nanonyme]

Unfortunately can't edit my previous message anymore: the mixup with libdrm was my own fault. Solved in a fresh installation.

[nanonyme]

Anyone success with the latest commit? I had to disable gallium to get it to build. But running glxinfo now gives me:

Fatal error in __driConfigOptions line 284, column 0: illegal default value: DRI_CONF_FP_OPTIMIZATION_SPEED
Aborted

Having played around with things a bit last night, I suspect you did the same mistake I did. Are you sure you told it --with-dri-drivers=radeon,r600? Since when I did that, I ran into compiling problems that I could only solve by combining some parts from agd5f's DRM with modesetting-gem libdrm. (I don't describe here exactly what I did nor do I recommend it because the end result was a lockup in starting X)
Edit: As in, I found out that not only should it not work, it also should not compile.

[nanonyme]

Which branch did you build ? Richard's latest push in 6xx-rewrite was just work-in-progress, still a bit more to do before it runs.

We're trying to keep the ported code visible even if it's not actually making triangles yet.

Also that same macro seems to be defined in r300_fragprog.h but seems to be missing from r700_fragprog.h altogether. (Or am I missing something?)

[nanonyme]

Just a note so I get this written down somewhere: anyone else getting their num lock, caps lock and scroll lock lights blinked randomly on and off with the r6xx-r7xx-3d DRM? Happens for me, I only get rid of it by rebooting.