Compiled LLVMpipe/Softpipe for Windows

[lowenz]

Anyone can upload a binary/compiled version of last LLVMpipe/Softpipe from Mesa3D 11.2.0 ? (sw "opengl32.dll" )

I'm trying to do it by myself but I encounter a scons error :/

 

And hey, now Mesa3D got OpenGL 4.x compliance, so we can test TDM with a software rasterizer (and it got Intel OpenSWR too: http://openswr.org/ https://github.com/OpenSWR/openswr-mesa )

Edited April 12, 2016 by lowenz

[lowenz]

...or help me understand (I use MSVC 2015+Clang)

 

I followed all the steps here: https://lists.freedesktop.org/archives/mesa-dev/2015-April/081220.html

 

MSVC and Flex/Bison checks are OK but the compiling script seems to have some problems:

TypeError: 'tuple' object is not callable:
  File "C:\mesa\SConstruct", line 138:
    duplicate = 0 # http://www.scons.org/doc/0.97/HTML/scons-user/x2261.html
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Script\SConscript.py", line 604:
    return method(*args, **kw)
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Script\SConscript.py", line 541:
    return _SConscript(self.fs, *files, **subst_kw)
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Script\SConscript.py", line 250:
    exec _file_ in call_stack[-1].globals
  File "C:\mesa\src\SConscript", line 7:
    SConscript('util/SConscript')
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Script\SConscript.py", line 604:
    return method(*args, **kw)
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Script\SConscript.py", line 541:
    return _SConscript(self.fs, *files, **subst_kw)
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Script\SConscript.py", line 250:
    exec _file_ in call_stack[-1].globals
  File "C:\mesa\src\util\SConscript", line 25:
    command = python_cmd + ' $SCRIPT > $TARGET'
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Environment.py", line 224:
    return self.method(*nargs, **kwargs)
  File "C:\mesa\scons\custom.py", line 143:
    deps += script_src.get_implicit_deps(env, python_scanner, path)
  File "C:\Python27\Scripts\..\Lib\site-packages\scons-2.5.0\SCons\Node\__init__.py", line 941:
    path = path_func(scanner)

Edited April 13, 2016 by lowenz

[lowenz]

Found a way.....directly with MSVC 2015

 

Problem now is for LLVMPipe only (I got SofPipe compiling and linking in a nice DLL )

 

Linking build\windows-x86-debug\util\roundeven_test.exe ...
LINK : fatal error LNK1181: cannot open input file 'LLVMBitWriter.lib'
scons: *** [build\windows-x86-debug\util\roundeven_test.exe] Error 1181
scons: building terminated because of errors.

 

LLVMBitWriter.lib is existing and in the right place, damn SCONS.....

Now Mesa (12) officially got this nice alternative SW rasterizer too, as driver for gallium: http://openswr.org/

Edited May 31, 2016 by lowenz

[lowenz]

Well, temporary solution: prebuild DLL from MSYS2

 

If you want to try (maybe with TDM): https://sourceforge.net/projects/msys2/files/REPOS/MINGW/i686/mingw-w64-i686-mesa-11.2.0-1-any.pkg.tar.xz/download

Edited June 2, 2016 by lowenz

[New Horizon]

Sorry, my tech knowledge is not up to date. What does this do exactly?

[lowenz]

Sorry, my tech knowledge is not up to date. What does this do exactly?

LLVMPipe is a LLVM-optimized Gallium implementation (aka driver) and Gallium is an OpenGL software rasterizer of MESA package

It can execute OpenGL code *without* an OpenGL compliant GPU or execute old code not compatible with new GPU drivers (retrogamer best friend )

 

The next version (12.x) will have another Gallium implementation (OpenSWR), made by Intel itself for its new AVX/AVX2 CPUs.

So we can choose LLVMPipe or OpenSWR.

 

LLVMPipe is really good, it can run Unreal @1280x960 with lastest OpenGL renderer (unofficial patch 227i) @28/30 FPS. Yes, CPU-only.

But TDM seems to not like Mesa

Edited June 2, 2016 by lowenz

[lowenz]

QT Creator 4.1b is released today, with LLVM 3.8 support and bundled LLVMPipe compiled with LLVM 3.8 also

So I can run OpenGL 3.0 (waiting for Mesa 12.x!) code with my CPUs.....the old one, an i3 530@3.7 Ghz, and the shiny new i3 6300 stock ('cause only LLVM 3.8 is compatible it!)

 

Unreal FlyBy OpenGL (patch 227i) 1280x960:

 

i3 530: 24 FPS

i3 6300: 36 FPS

 

50% speedup with the same frequence and the same core number (2 with HT).

Now this is progress

 

Waiting for MSYS2 Mesa 12.x LLVMPipe for MOAR performance

Edited July 6, 2016 by lowenz

[Bikerdude]

whats is QT Creator 4.1b..? tried googling didn't find anything useful.

[lowenz]

It's an IDE for QT libraries development

 

Here's the press release: http://blog.qt.io/blog/2016/07/06/qt-creator-4-1-0-beta-released/

And here's the 7zip package: http://download.qt.io/development_releases/qtcreator/4.1/4.1.0-beta1/installer_source/windows_vs2013_32/qtcreator.7z

 

Download it, search for "opengl32sw.dll", rename it to "opengl32.dll" and you'll have your MESA-compliant CPU driver for OpenGL 3.0 optimized with LLVM 3.8

 

Put it in your OpenGL 3.0 application/game exe folder and watch your processor do the GPU's work

Edited July 6, 2016 by lowenz

[Bikerdude]

This sounds more like something for Steve or Grayman :-)

[lowenz]

TDM sadly doesn't start with Mesa.....but they can control why and patch it to work with Mesa.

It would be perfect for Linux non-proprietary-drivers-loving users.....and Windows using LLVMPipe too

 

More steps toward a perfectly-compatible OpenGL application

Edited July 6, 2016 by lowenz

[lowenz]

OK, Mesa 12 is out: http://www.phoronix.com/scan.php?page=article&item=mesa-12-released&num=1

OpenGL 4.3 LLVMPipe based on LLVM 3.8 and OpenSWR !

 

Waiting for someone who can compile Windows binaries (32&64 bit "opengl32.dll" for LLVMPipe and for OpenSWR)

Edited July 10, 2016 by lowenz

[lowenz]

I GOT IT!

 

My DLL (MSVC compiled, with no particular optimizations): 42.5 FPS in Unreal

QTCreator DLL (MinGW crosscompiled?): 35.5 FPS in Unreal

 

I'll report the details soon The key was to use the old LLVM 3.7.1!

I'm the boss! Men, it seems to me I landed on the taffin' Moon

 

 

 

It would be really interesting to run TDM on Mesa/LLVMPipe!

Edited September 12, 2016 by lowenz

[lowenz]

RTCW (with widescreen patch ) @16 FPS!

See: GPU 0%

 

Edited September 12, 2016 by lowenz

[Bikerdude]

I havent been following this thread closely, but you look like you have had a breakthrough of sorts.

[lowenz]

Yes, now I got the DLL correctly built and working, as you can see by the RTCW cemetery

Edited September 12, 2016 by lowenz

[lowenz]

It would be really interesting to run TDM on Mesa/LLVMPipe!

 

LOL, TDM runs! It runs!

I got the start screen music playing but I can't see the background and the menu

 

 

Our devs can't fix this problem? It's surely a little incompatibility with Mesa loading the assets!

Edited September 12, 2016 by lowenz

[lowenz]

Loading a map from the console is perfectly working, the problem is the gamma handling and the textures that do not show up/are corrupt

But the geometry is there, you can see it just increasing the gamma over 2/3.....and you can walk around (@2 FPS in a pitch black environment )

Edited September 14, 2016 by lowenz

[nbohr1more]

Still black if you disable texture compression:

 

seta image_useCompression "0"

 

?

[lowenz]

I'll try!

[New Horizon]

Even after the explanation further up, I still don't really get what the benefits are of having this working. lol

[lowenz]

No benefits other than "We got a pseudo-reference OpenGL CPU rasterizer" and "Meeeen, now I can benchmark the CPU with some REAL heavy and optimized task"

 

With compression disable, still black: problem is *loading* the textures:

 

Edited September 15, 2016 by lowenz

[nbohr1more]

Please post a condump and I'll try to examine the failure points.

 

Some of those messages are just console spam about unused native image loading.

 

Though, as I recall, we don't ship all textures with uncompressed versions so to validate it may require

a test map with only uncompressed textures.

 

Of course, if you can ensure that S3TC compression support exists in the llvm-softpipe that would help narrow

things. Typically, modern GPL software has avoided S3TC support as it's considered a proprietary license so

you need to add it via an extension dependency.

[lowenz]

CONDUMP=console dump?

 

Tell me how and you'll have it.

[nbohr1more]

condump <name-of-file.txt> in the console.

 

then grab the file from your darkmod directory and attach it to a post.