Message from @Tervy

Discord ID: 644513099210293248


2019-11-14 11:50:32 UTC  

Twitter link goes nowhere

2019-11-14 11:51:04 UTC  

ahh apparently <> adds shit to the end some reason

2019-11-14 11:51:06 UTC  

what in the fuck

2019-11-14 11:51:20 UTC  

now it works

2019-11-14 11:51:22 UTC  

you added a space in front of the link bruh

2019-11-14 11:51:33 UTC  

between < and the link

2019-11-14 11:51:33 UTC  

might be

2019-11-14 11:51:38 UTC  
2019-11-14 11:51:43 UTC  

well color me suprised

2019-11-14 11:51:48 UTC  

it thinks lone > in the end is part of the link and translates it .. well thats retarded too

2019-11-14 11:52:04 UTC  

yes

2019-11-14 12:16:05 UTC  

@Tervy I don't know of anything that actually uses HLE

2019-11-14 12:16:17 UTC  

mainly because AMD doesn't even have it so you have to be compiling for intel in the first place

2019-11-14 12:19:26 UTC  

yeah so any larger server systems who actualy need all juice they can get

2019-11-14 12:19:40 UTC  

what actualy do computing and heavy lifting :D

2019-11-14 12:19:57 UTC  

but indeed for normal users the HLE hit aint __that__ bad expect on probly some rare cases

2019-11-14 12:24:08 UTC  

@Goz3rr and aint HLE anyway C/C++ only thing or ?

2019-11-14 12:24:27 UTC  

HLE is about machine code

2019-11-14 12:24:30 UTC  

it's an instruction

2019-11-14 12:24:57 UTC  

yeah but all thing i know have had problems/used it have had something to do with C so i was just quessing

2019-11-14 12:25:02 UTC  

HLE adds two instruction prefixes

2019-11-14 12:25:09 UTC  

which through some black magic fuckery are backwards compatible

2019-11-14 12:25:17 UTC  

so they can remove HLE without breaking anything

2019-11-14 12:25:25 UTC  

since it'll just fall back

2019-11-14 12:25:42 UTC  

the performance loss that tweet is talking about is unrelated to HLE instructions itself

2019-11-14 12:26:24 UTC  

the microcode update changed something relating to how jump instructions are decoded

2019-11-14 12:26:48 UTC  

the problem this tweet is talking about

2019-11-14 12:26:56 UTC  

compilers can know about this issue and generate different code

2019-11-14 12:27:12 UTC  

to avoid crossing 32 byte boundaries or ending up on a 32 byte boundary

2019-11-14 12:27:30 UTC  

for negligible side effects (basically just add a nop or reorder code)

2019-11-14 12:27:33 UTC  

but the problem is

2019-11-14 12:27:43 UTC  

everything has to be recompiled for that to happen

2019-11-14 12:27:53 UTC  

hence things compiled before now might try to do that and see reduced performance

2019-11-14 12:30:17 UTC  

and the way almost all modern software works

2019-11-14 12:30:39 UTC  

no one actually compiles for different cpus

2019-11-14 12:30:51 UTC  

unless you're some arch turboautist

2019-11-14 12:32:27 UTC  

gentoo*

2019-11-14 12:32:51 UTC  

Same difference

2019-11-14 12:34:09 UTC  

it's not, you don't build from source on arch unless outside package manageer

2019-11-14 12:50:03 UTC  

Performance in games seems to be in line with the 0-4% decrease Intel states