ChanServ changed the topic of #lima to: Development channel for open source lima driver for ARM Mali4** GPUs - Kernel driver has landed in mainline, userspace driver is part of mesa - Logs at https://oftc.irclog.whitequark.org/lima/
chewitt has joined #lima
chewitt has quit [Quit: Zzz..]
chewitt has joined #lima
mripard has joined #lima
<enunes>
anarsoul: did some airport seat reviews, it's a lot of improvements by now, nice to see :)
stilbruch has quit [Read error: Connection reset by peer]
<anarsoul[m]>
yeah, I had to redo nickserv auth, since bridge lost connection
<anarsoul[m]>
I have a few more ideas to implement. 1) branching optimization. We have one extra branch in loops if they have a break
<anarsoul[m]>
2) mov coalescing. It looks like vec produces a series of scalar movs that can be coalesced into a vector mov on some cases
<anarsoul[m]>
3) eliminate mov for store output if the source is register. If it is not a single node in block, we can remove it and mark source reg as output reg. If it is a single node, whole block can be removed, predecessor blocks marked as end blocks and source reg is again marked as an output reg