For Now Intel's Beignet Seems Better Off Than Radeon Gallium3D Clover

[phoronix]

Phoronix: For Now Intel's Beignet Seems Better Off Than Radeon Gallium3D Clover

While Intel's Beignet project for providing open-source OpenCL support for their hardware on Linux was widely criticized upon its debut for being a new project rather than basing the work on Gallium3D's "Clover" OpenCL state tracker, Beignet has matured much more quickly and for now at least seems to be better off than the Gallium3D OpenCL support...

http://www.phoronix.com/vr.php?view=MTc4NzA

[Veerappan]

I'm not too surprised that Luxmark, JuliaGPU, and MandelBulb are all failing... They all have CL code which is written assuming that your CL runtime library provides CL version 1.2 and that your device has also exposed support for it. The static (and other) storage class specifier(s) was added as a CL 1.2 feature, and none of these applications actually bother checking for support for this feature before they use it. It may be possible for us to bump clover to reporting 1.2 support, but I'm not sure if all of the required CL APIs that were introduced in 1.2 are in place or if we'd be introducing regressions for applications that currently do work.

With regards to x264, you're right, it's not using OpenCL on radeon/clover. x264 requires support for the image objects/apis in OpenCL, which the Radeon cards and clover do not yet support. In x264's case, it at least gracefully falls back to CPU encoding instead of crashing. Luxmark doesn't necessarily do that same check for support before it tries to use it (although it does have a fallback path that CAN be enabled manually).

[artivision]

I'm not too surprised that Luxmark, JuliaGPU, and MandelBulb are all failing... They all have CL code which is written assuming that your CL runtime library provides CL version 1.2 and that your device has also exposed support for it. The static (and other) storage class specifier(s) was added as a CL 1.2 feature, and none of these applications actually bother checking for support for this feature before they use it. It may be possible for us to bump clover to reporting 1.2 support, but I'm not sure if all of the required CL APIs that were introduced in 1.2 are in place or if we'd be introducing regressions for applications that currently do work.

With regards to x264, you're right, it's not using OpenCL on radeon/clover. x264 requires support for the image objects/apis in OpenCL, which the Radeon cards and clover do not yet support. In x264's case, it at least gracefully falls back to CPU encoding instead of crashing. Luxmark doesn't necessarily do that same check for support before it tries to use it (although it does have a fallback path that CAN be enabled manually).

When they will have support for the right state tracker (D3D9), and when they will have 4x the GPU flops with small and mediocre CPUs (like Atoms and Haswell-Celerons_Pentiums, then they will have a client, me (for allot chips).

[rikkinho]

normal

this is normal, intel have the best opensource drivers, and now opencl too

[jrch2k8]

BTW luxmark run in latest git code on RadeonSI, oibaf is just too old

Arch + lcarlier is best way to try

[jrch2k8]

BTW luxmark run in latest git code on RadeonSI, oibaf is just too old

Arch + lcarlier is best way to try

evidence is faster than beignet XD

[rikkinho]

lol?

evidence is faster than beignet XD

looks like a high-end card what about the same level card?

[Pontostroy]

BTW luxmark run in latest git code on RadeonSI, oibaf is just too old

Arch + lcarlier is best way to try

Awesome

opensuse+pontostroy:x11 just works too

[marco26]

Never seen an intel gpu capable of doing this...

[Pontostroy]

http://openbenchmarking.org/result/1409139-KH-RADEONSIL32,1408304-KH-IVYBRIDGE38&compare

beignet intel hd4000 vs clover radeon hd7790