| 1 |
On 2019-12-17 21:21, Matt Turner wrote: |
| 2 |
|
| 3 |
>> What do you think, guys? |
| 4 |
> |
| 5 |
> I don't love it. |
| 6 |
> |
| 7 |
> I don't like the mess that has become VIDEO_CARDS=... either. radeon |
| 8 |
> vs radeonsi vs amdgpu. |
| 9 |
[...] |
| 10 |
|
| 11 |
I hear you and very much agree, especially regarding the Radeon mess. |
| 12 |
That said, it seems what we are complaining about is the whole |
| 13 |
VIDEO_CARDS system rather than the introduction of 'iris' into |
| 14 |
virtual/opencl. |
| 15 |
|
| 16 |
The current situation with OpenCL providers for Intel GPUs ugly even |
| 17 |
disregarding the above. Basically, it is: |
| 18 |
|
| 19 |
1. Set VIDEO_CARDS: i965 |
| 20 |
|
| 21 |
2a. If you do not set ABI_X86: 32 for virtual/opencl, it will pull in |
| 22 |
dev-libs/intel-neo |
| 23 |
|
| 24 |
2a.1. NEO may or may not work but with the former more likely than the |
| 25 |
latter (it's been a few years since Broadwell architecture came out); |
| 26 |
|
| 27 |
2b. If, however, you DO enable 32-bit x86 ABI it will pull |
| 28 |
dev-libs/beignet instead |
| 29 |
|
| 30 |
2b.1. Beignet at least for the time being is likely to work even on |
| 31 |
modern systems but nowhere nearly as well as NEO would and without |
| 32 |
official upstream support. |
| 33 |
|
| 34 |
|
| 35 |
In contrast, reassigning NEO to video_cards_iris would make the |
| 36 |
Intel-GPU provider tree much more straightforward: |
| 37 |
|
| 38 |
1. For i965, you would get dev-libs/beignet regardless of whether you |
| 39 |
want 32-bit x86 ABI or not; |
| 40 |
|
| 41 |
2. For iris, you would get dev-libs/intel-neo for native 64 bits and |
| 42 |
nothing (okay, technically it would fall back to media-libs/mesa - but |
| 43 |
the fact Mesa is considered a fallback OpenCL provider for all GPUs is a |
| 44 |
completely different can of worms) for 32-bit x86 ABI. |
| 45 |
|
| 46 |
|
| 47 |
-- |
| 48 |
MS |
Archives