NVIDIA's New Allocator Library Will Need A Lot Of Work In 2018

[Delgarde]

This might end up like the EGLStreams situation...

The entire point of this new allocator is to *fix* the EGLStreams situation, so it seems unlikely that NVidia would be wasting time and money on it if they didn't think it offered a way forward.

Besides, would you really expect their proposed design to be accepted immediately by the DRI/DRM guys without *any* comment, without anyone raising concerns, or asking for clarification about why they've made certain decisions? There's no drama going on here... it's just the usual process of collaboratively evolving a design. As far as I can figure, the responses basically amount to "this aspect of the design would force us to break API compatibility in KMS... what's the justification for that?"

[tildearrow]

So far though many upstream open-source driver developers are not yet convinced about the current design of this Unix Device Memory Allocation library as a potential replacement to GBM.

This might end up like the EGLStreams situation...

[boxie]

Let's hope they come up with a design that everyone it not cranky at

[phoronix]

Phoronix: NVIDIA's New Allocator Library Will Need A Lot Of Work In 2018

Last week NVIDIA sent out an experimental allocator driver for the Nouveau code-base as well as EXT_external_objects support for Nouveau NVC0 in Mesa. So far though many upstream open-source driver developers are not yet convinced about the current design of this Unix Device Memory Allocation library as a potential replacement to GBM...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Alloc-More-Work-2018