00:07 <HdkR> macc24: use ninja and make sure your build directory is not root of the source. Seems like it's messing up weirdly

00:44 <HdkR> I could test that configuration in a bit, not sure if we managed to break it

00:48 <HdkR> macc24: Confirmed, it's a problem of using make instead of Ninja

00:48 <HdkR> Not sure when that broke

00:54 <HdkR> Looks like Stef forgot to generate the folder structure correctly and ninja just did that for us

01:48 <HdkR> macc24: https://github.com/FEX-Emu/FEX/pull/525 PR opened to fix that problem

01:57 <alyssa> HdkR: what's PR #524 about? :o

01:58 <alyssa> (static RA)

01:59 <HdkR> alyssa: It statically allocates all or most of the x86/x86-64 GPRs in to host Arm64 GPRs to remove memory loadstores

02:00 <HdkR> Puts them in callee saved registers so calling out to helper code is okayish

02:02 <alyssa> I guess I don't understand what static means here?

02:05 <HdkR> Something like guest register RAX will be statically bound to host register x19. So any usage of RAX in the guest code only uses the host GPR directly instead of having to load it from memory first, and then storing to memory at the end of its usage

02:06 <HdkR> So it breaks the dynamic nature of RA with an SSA IR

02:06 <alyssa> Ahhh

02:11 <HdkR> It's a bit necessary for maximum perf

02:17 <HdkR> Especially since x86 compiled code is very branch happy and if you can't see the branch target you lose perf for "no reason"

02:18 <alyssa> :<

02:25 <HdkR> Although we also need more cross-block optimizations so there is plenty of work to do :P

08:43 <icecream95> alyssa: Another idea for improving panfrost_pool_alloc_aligned: Store the allocation size in the top bits of the GPU VA

16:50 <macc24> HdkR: how do i run fex? i can't get hello world to work

16:50 <macc24> i won't run away this time, i promise

16:53 <macc24> there should be more documentation...

17:07 <macc24> oh and firefox rendering is sometimes garbage on G72

19:50 <macc24> time to air out the room and fix the sound