IRC channel logs

<jleightcap>out of curiosity:

<jleightcap>how often does something taking prohibitively long to run come up?

<jleightcap>seem to remember in live-bootstrap there were multiple hour runs

<jleightcap>has there been any work axed because of unrealistic runtime?

<ekaitz>jleightcap: if it helps I'm trying to make Mes faster

<ekaitz>the goal is to improve one order of magnitude or so

<ekaitz>(i don't think I'm getting there, but we'll see)

<jleightcap>ekaitz: i'm mostly curious about what kind of upper bounds are considered realistic

<jleightcap>or at least tolerable!

<jleightcap>not at all commenting on existing effort

<jleightcap>is your optimization work motivated by a speed bottleneck?

<ekaitz>jleightcap: i worked on the riscv bootstrap and it took looooong time

<ekaitz>so i decided to work on this

<jleightcap>oh nice! that's my idealized end point

<jleightcap>i'm thinking about hardware waaaay slower than even that

<lanodan>old sparc? ^^ (oldest machine I have here)

<lanodan>Ah yeah that's likely much worse

<jleightcap>i need to figure out if a full build is, like, weeks? months? ...years?

<ekaitz>in a not very recent laptop it takes hours (many) but not days

<matrix_bridge><Andrius Štikonas> But that's qemu emulated syscalls

<matrix_bridge><Andrius Štikonas> On baremetal riscv hardware it's quite a bit slower

<matrix_bridge><Andrius Štikonas> Maybe 1.5 week or so on jh7110 soc

<matrix_bridge><Andrius Štikonas> Generally live-bootstrap gets to the first gcc in 30 min or so on x86 linux system

<matrix_bridge><Andrius Štikonas> A bit slower if kernel bootstrap is involved

<lanodan>Wow that's slow

<matrix_bridge><Andrius Štikonas> Well, mostly due to very slow memory on visionfive2 (jh7110)

<lanodan>yeah, got a VF2 here as well

<matrix_bridge><Andrius Štikonas> So mes gets botglenecked

<matrix_bridge><Andrius Štikonas> Bottlenecked there

<matrix_bridge><Andrius Štikonas> Mes is very RAM bandwidth intensive

<matrix_bridge><Andrius Štikonas> once you get to tcc, the speed is acceptable

<jleightcap>this is about the hardware integration scale i'm looking at

<jleightcap>although i presume build time >>> boot time of ~5d

<jleightcap>i suppose executing the kernel is a reasonable but ultimately arbitrary end point

<qbit>there needs to be a new term: bootlenecked

<lanodan>Or bognecked

<stikonas>qbit: well, to my defence, I was typing on the phone

<stikonas>jleightcap: well, in order to build stuff we also need more RAM

2025-10-17.log