RadeonSI Compute Shaders Land In Mesa Git

[Eliasvan]

Cool!
Anyone here owning an AMD card to test the famous 64KB OpenGL4.3 demo (https://www.phoronix.com/scan.php?pa...tem&px=MTY2ODc) with the open source drivers?

If Wine manages to run this demo, I'd be interested to know the results.

[haagch]

Cool!
Anyone here owning an AMD card to test the famous 64KB OpenGL4.3 demo (https://www.phoronix.com/scan.php?pa...tem&px=MTY2ODc) with the open source drivers?

Same with intel and radeonsi:

[FONT=monospace][COLOR=#000000]Unhandled exception: page fault on read access to 0x00000000 in 32-bit code (0xf69e665f). [/COLOR]
err:dbghelp_msc:pe_load_debug_directory Got a page fault while loading symbols
Register dump:
CS:0023 SS:002b DS:002b ES:002b FS:0063 GS:006b
EIP:f69e665f ESP:0033f970 EBP:0033fb68 EFLAGS:00210246(  R- --  I  Z- -P- )
EAX:00000000 EBX:f6a2e000 ECX:ffffffff EDX:ffffffff
ESI:00000000 EDI:00000000
Stack dump:
0x0033f970:  00001f03 7dc82c18 7dc82c18 00000000
0x0033f980:  00110014 7bce7000 0033fa08 7bc54951
0x0033f990:  0033f9b0 f715a000 4d430003 ffffffff
0x0033f9a0:  0033f9cc 00000000 7dd9af6c 00000110
0x0033f9b0:  09800062 00000000 7dc82c18 7dc812f0
0x0033f9c0:  0033f9e0 7bce7000 0033fb88 00000004
Backtrace:
=>0 0xf69e665f in winex11 (+0x3665f) (0x0033fb68)
 1 0xf69e9675 in winex11 (+0x39674) (0x0033fb78)
 2 0xf69d8ccd in winex11 (+0x28ccc) (0x0033fb98)
 3 0xf70c9c9f in gdi32 (+0x19c9e) (0x0033fbb8)
 4 0xf71208be __wine_get_wgl_driver+0x5d() in gdi32 (0x0033fbe8)
 5 0x7a8b1235 wglDescribePixelFormat+0x44() in opengl32 (0x0033fc38)
 6 0x7a8b12c1 wglChoosePixelFormat+0x50() in opengl32 (0x0033fcd8)
 7 0xf7120922 ChoosePixelFormat+0x41() in gdi32 (0x0033fd18)
 8 0x0040149a in timeless (+0x21499) (0x0033fe28)
 9 0x004015e7 in timeless (+0x215e6) (0x093c18f2)
 10 0x003fdb68 in timeless (+0x1db67) (0x003fdaba)
 11 0x0c181c10 (0x08081004)
0xf69e665f: repne scasb %es:(%edi)[/FONT]

Not sure if it's wine or the application, but it's using the compatibility profile. Funny thing is, when not using the mesa version override, then the shaders fail to compile and it actually starts, but obviously doesn't display anything useful: https://www.youtube.com/watch?v=55rtVqKHsfA

[bridgman]

Not sure if it's wine or the application, but it's using the compatibility profile.

Hmm... someone needs to have a word with the developers... compatibility profiles are a dead end for gaming (and for everything else IMO although there is less agreement there).

EDIT - hold on, you're talking about that 64K demo... a little less black & white because taking a chance on compatibility profiles *might* have saved them a bit of space.

[haagch]

Unfortunately it still happens all the time.

The worst is when there is core profile support, but it is guarded with #ifdef __APPLE__... E.g. OSVR render manager: https://github.com/sensics/OSVR-Rend....cpp#L282-L287

Well, https://github.com/sensics/OSVR-RenderManager/issues/68

[Eliasvan]

Same with intel and radeonsi:

[FONT=monospace][COLOR=#000000]Unhandled exception: page fault on read access to 0x00000000 in 32-bit code (0xf69e665f). [/COLOR]
err:dbghelp_msc:pe_load_debug_directory Got a page fault while loading symbols
Register dump:
CS:0023 SS:002b DS:002b ES:002b FS:0063 GS:006b
EIP:f69e665f ESP:0033f970 EBP:0033fb68 EFLAGS:00210246( R- -- I Z- -P- )
EAX:00000000 EBX:f6a2e000 ECX:ffffffff EDX:ffffffff
ESI:00000000 EDI:00000000
Stack dump:
0x0033f970: 00001f03 7dc82c18 7dc82c18 00000000
0x0033f980: 00110014 7bce7000 0033fa08 7bc54951
0x0033f990: 0033f9b0 f715a000 4d430003 ffffffff
0x0033f9a0: 0033f9cc 00000000 7dd9af6c 00000110
0x0033f9b0: 09800062 00000000 7dc82c18 7dc812f0
0x0033f9c0: 0033f9e0 7bce7000 0033fb88 00000004
Backtrace:
=>0 0xf69e665f in winex11 (+0x3665f) (0x0033fb68)
1 0xf69e9675 in winex11 (+0x39674) (0x0033fb78)
2 0xf69d8ccd in winex11 (+0x28ccc) (0x0033fb98)
3 0xf70c9c9f in gdi32 (+0x19c9e) (0x0033fbb8)
4 0xf71208be __wine_get_wgl_driver+0x5d() in gdi32 (0x0033fbe8)
5 0x7a8b1235 wglDescribePixelFormat+0x44() in opengl32 (0x0033fc38)
6 0x7a8b12c1 wglChoosePixelFormat+0x50() in opengl32 (0x0033fcd8)
7 0xf7120922 ChoosePixelFormat+0x41() in gdi32 (0x0033fd18)
8 0x0040149a in timeless (+0x21499) (0x0033fe28)
9 0x004015e7 in timeless (+0x215e6) (0x093c18f2)
10 0x003fdb68 in timeless (+0x1db67) (0x003fdaba)
11 0x0c181c10 (0x08081004)
0xf69e665f: repne scasb %es:(%edi)[/FONT]

Not sure if it's wine or the application, but it's using the compatibility profile. Funny thing is, when not using the mesa version override, then the shaders fail to compile and it actually starts, but obviously doesn't display anything useful: https://www.youtube.com/watch?v=55rtVqKHsfA

Thanks for testing this!
At least the music plays nicely ;-)

Hmm... someone needs to have a word with the developers... compatibility profiles are a dead end for gaming (and for everything else IMO although there is less agreement there).

EDIT - hold on, you're talking about that 64K demo... a little less black & white because taking a chance on compatibility profiles *might* have saved them a bit of space.

Does this mean that, without modifications to the 64KB binary, this demo will never be able to be run on Mesa?

[geearf]

I've been playing a bit of Mordor, and it plays fine on Ultra.... not sure what my fps are, but since it seems good enough I have not checked.
I did not expect that at all..

So a little update.
I was wrong, I was playing in between High and Ultra, when I switched from High to Ultra some stuff required a restart that I didn't do at the time, and the full ultra crashes SOM at times so I went back to High.
It played about fine apart from 2 stages:
- One mission where in some location it was quite slow, but just turn around and it's good again.
- the last boss fight, somehow the stage there was quite sluggish, which made me do it more times than needed.

[bridgman]

Does this mean that, without modifications to the 64KB binary, this demo will never be able to be run on Mesa?

That's off past the end of my OpenGL knowledge. If the demo just happens to be asking for a compatibility profile (which AFAIK Mesa will not provide) but isn't actually using any deprecated functions then it *might* work... not sure unfortunately.

[osmotr]

Cool!
Anyone here owning an AMD card to test the famous 64KB OpenGL4.3 demo (https://www.phoronix.com/scan.php?pa...tem&px=MTY2ODc) with the open source drivers?

If Wine manages to run this demo, I'd be interested to know the results.

i manage to run demo with enabling compatiblity profile on radeonsi
i run with MESA_GL_VERSION_OVERRIDE=4.3COMPAT MESA_GLSL_VERSION_OVERRIDE=430

[haagch]

i run with MESA_GL_VERSION_OVERRIDE=4.3COMPAT

Oh cool, I did not know that 4.3 and 4.3COMPAT have different effects. With 4.3COMPAT it also runs for me.

Video with some gallium hud info:

https://youtu.be/EeVX2zAyoss

Low FPS, but interesting that GPU usage is at 100% all the time and no buffer wait time, so I guess it runs with "good" performance.

Anyway, that all is in 64k? Even with heavy procedural generation that's not something you'd think is possible. Wow.

Edit: Oh, one of the shaders fails to compile: https://gist.github.com/ChristophHaa...93aa94fce883da
0:36(446): error: no matching function for call to `jl(vec3)'; candidates are:
0:36(446): error: float jl(float)

Anyone knows if this is valid in GLSL 430?
floatjl(infloat c){
returnfloat(cu(int(c)))/1.07374e+09-.5;
}
jl(l.xxy*1439)

Sure, it could be defined that the function is called for each of the 3 elements of the vec3 and then returns the results as a vec3, but is it?