RadeonSI Compute Shaders Land In Mesa Git

[jf33]

Are there any plans to implement GL_ARB_bindless_texture in mesa? I think it's the only missing extension from the AZDO family, so it would bring some neat performance improvements in some games.

[finalzone]

Note that RadeonSI also supports OpenGL ES 3.1 as well.

[Veerappan]

maybe even summer 2016
amd will opensource their proprietary opencl

AMD has already open-sourced enough of their opencl built-in library to finish all of the non-image functions for at least opencl 1.2. We just don't have enough manpower at the moment to port those functions over to libclc.

AMD Also opened up the image portion of their library, but it uses enough LLVM intrinsics that those will probably end up re-written in a generic manner.

But yeah, need more warm bodies to write unit tests and port the implementations. We've been very close to having enough implemented for a good portion of michael's CL testsuite for a while, and even things like the C++AMP driver that is out there only need half a dozen new functions implemented... I have/had been working on trying to get that going, but real life got in the way of my volunteer programming efforts for a while.

[pal666]

Are there any plans to implement GL_ARB_bindless_texture in mesa?

yes

https://www.google.com.ua/webhp?sourceid=chrome-instant&ion=1&espv=2&ie=UTF-8#q=bindless%20site%3Apeople.freedesktop.org%2F~cbrill%2Fdri-log

[bridgman]

It's still a long road for OpenCL (CUDA is probably not an option?), but I'd guess it's just a matter of time.

How about this ?

Convert your CUDA Code to C++ Using AMD’s New HIP Tool
The world of HPC as it exists today consists of many GPU-accelerated applications that use the
proprietary CUDA language and infrastructure. The problem with using proprietary software is
that it is almost always tightly controlled and its source code is most often kept secret, and with
CUDA, the hardware options are limited to one vendor.
AMD, a strong proponent of open source and open standards, has created a new tool that will
allow developers to convert CUDA code to common C++. The resulting C++ code can run through
either CUDA NVCC or AMD HCC compilers. This new Heterogeneous-compute Interface for
Portability, or HIP, is a tool that provides customers with more choice in hardware and
development tools.

http://www.amd.com/Documents/HIP-Datasheet.pdf

[haagch]

Bug report for Alien Isolation segfaulting on start: https://bugs.freedesktop.org/show_bug.cgi?id=95026
Can anyone reproduce this?

Edit: Of course I got it wrong.
*compiles some more to add correct info to the bug report*
*sigh*

[thelongdivider]

This is really great work. As a radeonSI user, I think it has given me the best overall GPU experience on linux.

[DanielPower]

Does anyone know what version of OpenGL Divinity Original Sin: Enhanced Edition uses?

It's the only game I still can't run on open source drivers as of the RadeonSI with OpenGL 4.2 release. Looking forward to trying with the new 4.3 release.

As soon as Divinity runs, I'll be removing catalyst.

[aandreif]

Does anyone know what version of OpenGL Divinity Original Sin: Enhanced Edition uses?

It's the only game I still can't run on open source drivers as of the RadeonSI with OpenGL 4.2 release. Looking forward to trying with the new 4.3 release.

As soon as Divinity runs, I'll be removing catalyst.

Hi,

DOS:EE needs 4.2, but does some weird stuff see here:https://bugs.freedesktop.org/show_bug.cgi?id=93551 and the forum post in the last comment there, also have emailed them directly, and Raze from Larian said that the issue will be brought to the lead Linux prog. still no luck though.

Maybe now that we have this momentum with Mesa, they'll reconsider supporting the OSS drivers. Shame because DOS:EE is really nice.

[andre30correia]

FPS for Shadow of Mordor on R9-390:



$ MESA_GLSL_VERSION_OVERRIDE=430 steam.sh

Note: R9 390 requires "echo high > /sys/class/drm/card0/device/power_dpm_force_performance_level", otherwise it will hang.

horrible performance, is a game problem or drivers?