Progress On The ATI R600g Gallium3D Driver

[phoronix]

Phoronix: Progress On The ATI R600g Gallium3D Driver

As was just talked about in announcing the open-source 2D and 3D support for ATI Evergreen GPUs, the R600g driver has been gaining lots of momentum in the past few weeks. Ever since this open-source Gallium3D driver that aims to provide hardware-acceleration for ATI Radeon HD 2000/3000/4000 (and potentially Radeon HD 5000 series) hardware took a shader compiler shortcut a few weeks back, it seems almost every time our RSS feed for the Mesa Git change-log refreshes there is a new R600g driver change...

http://www.phoronix.com/vr.php?view=ODUzMA

[droidhacker]

For what its worth, GALLIUM3D should be the focus.

[bridgman]

Yep, that is our current thinking as well. Now that we have gone through the pain and figured out how to make the Evergreen chips run (more or less) it shouldn't be too hard to add EG support to the Gallium3D driver.

That's what we are hoping, anyways

[V!NCENT]

Yep, that is our current thinking as well. Now that we have gone through the pain and figured out how to make the Evergreen chips run (more or less) it shouldn't be too hard to add EG support to the Gallium3D driver.

That's what we are hoping, anyways

I have a nasty feeling you had to work around added DRM, am I correct?

[Chewi]

I suspect many people (like me) have bought into this open ATI thing already but are not buying newer cards (I'm sticking with 4670) simply because they won't work yet. So you might as well focus on Gallium3D.

[LiquidAcid]

Just tried ut2003 on gallium with my RV740 and it kinda works. Visuals are looking good, but from time to time the framebuffer content gets messed up (looks like color inversion to me) and the GPU gets reset quite frequently.

From the kernel log when such a GPU reset happens:

radeon 0000:01:00.0: GPU lockup CP stall for more than 1020msec
------------[ cut here ]------------
WARNING: at drivers/gpu/drm/radeon/radeon_fence.c:235 radeon_fence_wait+0x35d/0x3c0 [radeon]()
Hardware name: To Be Filled By O.E.M.
GPU lockup (waiting for 0x0001A62C last fence id 0x0001A623)
Modules linked in: cpufreq_ondemand lrw gf128mul w83627ehf hwmon_vid snd_seq_midi snd_emu10k1_synth snd_emux_synth snd_seq_virmidi snd_seq_midi_event snd_seq_midi_emul snd_seq snd_pcm_oss snd_mixer_oss reiserfs aes_x86_64 aes_generic cbc dm_crypt radeon ttm drm_kms_helper drm i2c_algo_bit cfbcopyarea cfbimgblt cfbfillrect powernow_k8 mperf snd_emu10k1 snd_rawmidi snd_ac97_codec ac97_bus snd_pcm r8169 i2c_piix4 snd_seq_device i2c_core ehci_hcd ohci_hcd sg mii k10temp snd_timer snd_page_alloc snd_util_mem snd_hwdep
Pid: 3279, comm: X Tainted: G        W   2.6.35.2 #2
Call Trace:
 [<ffffffff8103b65a>] warn_slowpath_common+0x7a/0xb0
 [<ffffffff8103b731>] warn_slowpath_fmt+0x41/0x50
 [<ffffffffa0176b1d>] radeon_fence_wait+0x35d/0x3c0 [radeon]
 [<ffffffff81053530>] ? autoremove_wake_function+0x0/0x40
 [<ffffffffa01773ec>] radeon_sync_obj_wait+0xc/0x10 [radeon]
 [<ffffffffa0135f67>] ttm_bo_wait+0xf7/0x1b0 [ttm]
 [<ffffffffa018c332>] radeon_gem_wait_idle_ioctl+0x92/0x130 [radeon]
 [<ffffffffa00f279a>] drm_ioctl+0x3ba/0x460 [drm]
 [<ffffffff8104b371>] ? get_signal_to_deliver+0x341/0x460
 [<ffffffffa018c2a0>] ? radeon_gem_wait_idle_ioctl+0x0/0x130 [radeon]
 [<ffffffff8100c6ce>] ? save_i387_xstate+0x18e/0x1e0
 [<ffffffff810018fd>] ? do_signal+0x17d/0x780
 [<ffffffff810c25f8>] vfs_ioctl+0x38/0xd0
 [<ffffffff81002c8e>] ? apic_timer_interrupt+0xe/0x20
 [<ffffffff810c27a0>] do_vfs_ioctl+0x80/0x520
 [<ffffffff8100c4d0>] ? restore_i387_xstate+0x110/0x180
 [<ffffffff810c2cb9>] sys_ioctl+0x79/0x90
 [<ffffffff810023ab>] system_call_fastpath+0x16/0x1b
---[ end trace 68d2c582426de957 ]---
[drm] Disabling audio support
radeon 0000:01:00.0: ffff88012fef0600 unpin not necessary
radeon 0000:01:00.0: GPU softreset 
radeon 0000:01:00.0:   R_008010_GRBM_STATUS=0xA2733028
radeon 0000:01:00.0:   R_008014_GRBM_STATUS2=0x00000102
radeon 0000:01:00.0:   R_000E50_SRBM_STATUS=0x200000C0
radeon 0000:01:00.0:   R_008020_GRBM_SOFT_RESET=0x00007FEE
radeon 0000:01:00.0: R_008020_GRBM_SOFT_RESET=0x00000001
radeon 0000:01:00.0:   R_008010_GRBM_STATUS=0x00003028
radeon 0000:01:00.0:   R_008014_GRBM_STATUS2=0x00000002
radeon 0000:01:00.0:   R_000E50_SRBM_STATUS=0x200000C0
radeon 0000:01:00.0: GPU reset succeed
[drm] Clocks initialized !
[drm] ring test succeeded in 1 usecs
[drm] ib test succeeded in 1 usecs
[drm] Enabling audio support

This is with:
Linux voodoomaster 2.6.35.2 #2 SMP PREEMPT Thu Aug 19 23:43:06 CEST 2010 x86_64 AMD Athlon(tm) 7750 Dual-Core Processor AuthenticAMD GNU/Linux

I'm currently back to vanilla 2.6.35.2 since the d-r-t branch seems to have some issue with my Promise PATA controller (I get constant ATA bus errors and eventually the main filesystem is remounted ro).

[LiquidAcid]

Forgot to mention this: Performance is of course really slow and I had to patch r600g to put these gallium.bof dump files into /tmp and not into the local directory. Without this gallium this to create the bof dumps in the ut2003 System directory, to which the normal user doesn't have write access - this results in a segfault and gallium dies

[nanonyme]

Performance is of course really slow

Did you try after that commit that added occlusion queries, btw?

[LiquidAcid]

commit id is 121aa3cfcb106be1ecaae102177bb882bfbce9d9, so yes.

[monkeynut]

It's just a shame we won't get texture compression.