Intel Arc GPU status

[artyom-beilis]

I got recently arc Intel card (Specifically Sparkle ELF a380)

• I got Intel XPU working (intel extension for pytorch) so I could run the benchmarks - while it isn't brilliant card it works
• Pytorch backend works as is but performance... isn't good at all.
• I pulled an old branch where I tried integration with oneDNN (Intel's cudnn for intel - but open-source) and unlike previous attempt it works quite well with decent performance and this is something than can be used with pytorch/dlprimitives - I tested only forward convolution implementation replacement by Intel's oneDNN calls - and it gave a promising results.