AMD has recently sent out around a hundred patches, which amount to over 40 thousand lines of code, so as to allow developers to integrate support for it’s upcoming Vega GPU architecture under Linux. The new code is essential towards baking support for Vega under Linux, considering the many changes this architecture entails over AMD’s current-generation Polaris 10 (soon to be rebranded, if sources are correct, to the new RX 500 series.) Also of note is the existence of seven different device IDs for Vega-based products, though this really can’t be extrapolated to the amount of SKUs under the Vega banner. For now, that really is just a number.
However, these lines of code won’t be enough to allow for complete support of Vega under Linux (and this support would be slated for version 4.12 of the operating system). Some more tens of thousands lines of code will be needed for DAL support. It would seem that AMD admits to a lack of resources in rewriting its code for Linux, and is thus looking for a way of sharing code between the two different driver stacks (for Windows and Linux), which would alleviate these issues. AMD would thus be looking towards abstractions. These would allow the company to write one piece of code that would be addressable by both operating systems – thus essentially reducing coding time in half. However, abstractions do come with a problem: these would imply huge amounts of code to be input into Linux, which would bloat an operating system known for it’s finesse. And this means AMD has a lot on its plate until the code needed for full Vega support is baked into Linux’s mainline code. One thing has to be said, though: Linux really isn’t where the money and mainstream support is. So AMD is definitely making the right business decision in putting most of its chops behind Windows.Source: Phoronix, AMD GFX
