RadeonSI OpenGL Compute Shader Patches Published

[darkbasic]

You applied the compute shader patches on top of mesa master or you simply compiled that outdated branch?

[Nille_kungen]

With this there's not much left until 4.3 for radeonsi only GL_ARB_robust_buffer_access_behavior
There's patches out for GLES3.1 and OpenGL4.2 and the other stuff but GL_ARB_robust_buffer_access_behavior for 4.3
https://lists.freedesktop.org/archiv...il/111653.html GL_ARB_compute_shader
https://lists.freedesktop.org/archiv...ch/111103.html GL_ARB_framebuffer_no_attachments
https://lists.freedesktop.org/archiv...ch/110698.html GL_ARB_shader_storage_buffer_object GL_ARB_shader_atomic_counters (also adds 2 GLES3.1 extensions that wasn't in an earlier patch GL_ARB_shader_image_load_store GL_ARB_shader_image_size).

[Ancurio]

I guess the thing is whether this is one of those extensions no one actually uses so you can just lie it exists through environment variable if blocking 4.3

Robustness is used exclusively for running WebGL, it's irrelevant for anything gaming or desktop related.

[haagch]

You applied the compute shader patches on top of mesa master or you simply compiled that outdated branch?

I compiled llvm from https://github.com/BNieuwenhuizen/llvm
and merged https://github.com/BNieuwenhuizen/mesa into mainline git master

[pal666]

Nobody has GL_ARB_robust_buffer_access_behavior yet, and it seems like it might be a PITA to implement rigorous bounds checking on buffers like that throughout the driver stack.

it is easy but nobody needs it

[justmy2cents]

Oh that made me actually look at his branch: https://github.com/BNieuwenhuizen/me...d4fd321855a518

So the actual news here is: We're almost at OpenGL 4.3.

Edit: EVERY TIME.
I made a normal release build of mesa and this happened when I starting Alien Isolation, right after initial loading screen:

[FONT=monospace][COLOR=#000000]Thread 36 "WinMain" received signal SIGSEGV, Segmentation fault. [/COLOR]
[Switching to Thread 0x7fffd406c700 (LWP 3966)]
0x00007fffe9808727 in ?? () from /usr/lib/xorg/modules/dri/radeonsi_dri.so
(gdb) bt
#0 0x00007fffe9808727 in ?? () from /usr/lib/xorg/modules/dri/radeonsi_dri.so
#1 0x0000000000000000 in ?? ()[/FONT]

Yea, that's the complete backtrace.

Then I made a debug build of mesa and guess what. IT DOES NOT CRASH.

So right now I only know it happens here:
>│0x7fffe9808727 orq $0x0,(%rsp)

Anyway, with a debug build of mesa I could try Alien Isolation and it still looks like this:

https://www.youtube.com/watch?v=Y0aqhazR3WE

So sadly I can say: It's not because of missing compute shaders, there's still another problem in radeonsi. Reference: https://bugs.freedesktop.org/show_bug.cgi?id=93144

lol, that is some trippy result. but, funny thing is that if you stop on frames, certain frames look just awesome. like some abstract painting.

[CrystalGamma]

Robustness is used exclusively for running WebGL, it's irrelevant for anything gaming or desktop related.

Can you sum up what it actually does? I'm guessing process isolation has always been there, considering the security implications. So what does implementing this extension actually guarantee?

[Oxalin]

Oh that made me actually look at his branch: https://github.com/BNieuwenhuizen/me...d4fd321855a518

So the actual news here is: We're almost at OpenGL 4.3.

Edit: EVERY TIME.
I made a normal release build of mesa and this happened when I starting Alien Isolation, right after initial loading screen:

[FONT=monospace][COLOR=#000000]Thread 36 "WinMain" received signal SIGSEGV, Segmentation fault. [/COLOR]
[Switching to Thread 0x7fffd406c700 (LWP 3966)]
0x00007fffe9808727 in ?? () from /usr/lib/xorg/modules/dri/radeonsi_dri.so
(gdb) bt
#0 0x00007fffe9808727 in ?? () from /usr/lib/xorg/modules/dri/radeonsi_dri.so
#1 0x0000000000000000 in ?? ()[/FONT]

Yea, that's the complete backtrace.

Then I made a debug build of mesa and guess what. IT DOES NOT CRASH.

So right now I only know it happens here:
>│0x7fffe9808727 orq $0x0,(%rsp)

Anyway, with a debug build of mesa I could try Alien Isolation and it still looks like this:

https://www.youtube.com/watch?v=Y0aqhazR3WE

So sadly I can say: It's not because of missing compute shaders, there's still another problem in radeonsi. Reference: https://bugs.freedesktop.org/show_bug.cgi?id=93144

Did you build an updated LLVM toolchain with the needed patches (also published by Nieuwenhuizen)? Did you verify the extension was now available (glxinfo)?

About your segfault, were you building mesa from a fresh folder or were you recompiling a previous build?

[pal666]

I'm guessing process isolation has always been there, considering the security implications. So what does implementing this extension actually guarantee?

isolation of browser from javascript

[haagch]

Did you build an updated LLVM toolchain with the needed patches (also published by Nieuwenhuizen)? Did you verify the extension was now available (glxinfo)?

LIBGL_DEBUG=verbose DRI_PRIME=1 glxinfo -s | grep -i compute
libGL: pci id for fd 5: 1002:6800, driver radeonsi
libGL: OpenDriver: trying /usr/lib/xorg/modules/dri/tls/radeonsi_dri.so
libGL: OpenDriver: trying /usr/lib/xorg/modules/dri/radeonsi_dri.so
libGL: Using DRI3 for screen 0
GL_ARB_compute_shader
GL_ARB_compute_shader

LIBGL_DEBUG=verbose DRI_PRIME=1 glxinfo | grep OpenGL
libGL: pci id for fd 5: 1002:6800, driver radeonsi
libGL: OpenDriver: trying /usr/lib/xorg/modules/dri/tls/radeonsi_dri.so
libGL: OpenDriver: trying /usr/lib/xorg/modules/dri/radeonsi_dri.so
libGL: Using DRI3 for screen 0
OpenGL vendor string: X.Org
OpenGL renderer string: Gallium 0.4 on AMD PITCAIRN (DRM 2.43.0, LLVM 3.9.0)
OpenGL core profile version string: 4.2 (Core Profile) Mesa 11.3.0-devel (git-b4964b9)
OpenGL core profile shading language version string: 4.20
OpenGL core profile context flags: (none)
OpenGL core profile profile mask: core profile

About your segfault, were you building mesa from a fresh folder or were you recompiling a previous build?

Incremental build. Maybe I'll try a full build later.