asciilifeform: it's 'optimal' in the sense that it fucks the branch predictor less than an always-computed-jump

asciilifeform: for larger, does a computed jump

asciilifeform: mircea_popescu: it's what gcc does for small (i dun recall threshhold) computed switch

asciilifeform prolly oughta have read the orig adb prior to doing this

asciilifeform: they're diana_coman's exception handlers.

asciilifeform: aah lol i'm a tard :

asciilifeform: the rest, i will omit, but also end up in the various unwindisms, aborts, stubs.

asciilifeform: 490 is various mechanics, ends with _Unwind_SjLj_Resume .

asciilifeform: var_B8 contains some flag. valid values are 0, 1, 2, 3. if 0, we go to loc_490. if 1, loc_601. if 2, loc_62d. if 3, loc_659. if above 4, program dies, cpu executes ud2 (guaranteed bomb) .

asciilifeform: want expand ?

asciilifeform: i suspect mircea_popescu dun habitually read gcc barf

asciilifeform: i.e. loox like is a flag that triggers an unwind

asciilifeform: jz somewhere

asciilifeform: test eax, eax

asciilifeform: cuz it's the 2nd half of a conditional ?

asciilifeform: aa ljmp

asciilifeform: in which

asciilifeform is prolly doomed to actually vivisect the gnat backend at some pt, prolly sooner rather than later

asciilifeform: i'ma guess it is for word-alignment (why not ordinary nop ?? )

asciilifeform: ( which is why asciilifeform never saw it in gnat disasms prior )

asciilifeform: it is only in ljmp-variant

asciilifeform currently wondering wtf https://www.felixcloutier.com/x86/ud is doing in there

asciilifeform: ( will guess, tho i do not presently know, that these trigger unwind of stack )

asciilifeform: also loox like the ljmp variant puts abortism stub in erry proc (that appears in a task, that is)

asciilifeform: yw mircea_popescu

asciilifeform: ^ in classic intelistic form, and with decoded debug symbols

asciilifeform: http://www.loper-os.org/pub/misc/zcx_procs.asm http://www.loper-os.org/pub/misc/ljmp_procs.asm ☟︎

asciilifeform: 1s

asciilifeform: cuz it keeps the where to longjmp in'ere.

asciilifeform: sub $0xXX,%rsp

asciilifeform: mircea_popescu: the stack frame.

asciilifeform: i was discussing diana_coman's much earlier empirical find, that on ljmp stack fills faster per same # of calls. this here is why.

asciilifeform: and see above

asciilifeform: then yes

asciilifeform: aa your were speaking of coad mass

asciilifeform: ( to 0 measurable diff in ffa, oddly enuff, but on e.g. tiny micros might make a diff.. )

asciilifeform: but indeed the ljmp variant craps out slightly bulkier coad across the board

asciilifeform: plus moar killed regs

asciilifeform: the stack frame, that is

asciilifeform: 0x28 == 40 , 0xb8 == 184

asciilifeform: mircea_popescu: where 50 and 60 ?

asciilifeform: ( before anyone asks, the 'unwind resume' variants are extern stdlib symbols, and i haven't looked to see how they differ yet )

asciilifeform: for thread-completeness : http://www.loper-os.org/pub/misc/zcx_proc_flow.png http://www.loper-os.org/pub/misc/ljmp_proc_flow.png

asciilifeform: d: 48 81 ec b8 00 00 00 sub $0xb8,%rsp

asciilifeform: c: 53 push %rbx

asciilifeform: a: 41 54 push %r12

asciilifeform: 8: 41 55 push %r13

asciilifeform: 6: 41 56 push %r14

asciilifeform: 4: 41 57 push %r15

asciilifeform: 1: 48 89 e5 mov %rsp,%rbp

asciilifeform: 0: 55 push %rbp

asciilifeform: 0000000000000000 <procs__a>:

asciilifeform: longjmpistic:

asciilifeform: 9: 48 83 ec 28 sub $0x28,%rsp

asciilifeform: 8: 53 push %rbx

asciilifeform: 6: 41 54 push %r12

asciilifeform: 4: 41 55 push %r13

asciilifeform: 1: 48 89 e5 mov %rsp,%rbp

asciilifeform: 0: 55 push %rbp

asciilifeform: 0000000000000000 <procs__a>:

asciilifeform: zcxistic :

asciilifeform: at the risk of log clutter, will put ftr :

asciilifeform: err, 'procs' and 'mt'

asciilifeform: though, interestingly, only in the unit 'procs' , which actually contains exceptionisms

asciilifeform: diana_coman: btw your 'lick the 9v' intuitive observation earlier was correct, on ljmp variant the default stack frame indeed longer, 184byte vs 40

asciilifeform: aa ok

asciilifeform: diana_coman: the 'zcx' tarball contains a 'ljmp_calls' dir, same as other 1. which is correct ?

asciilifeform: ty diana_coman ! i'ma look

asciilifeform: this does not contradict the hypothesis re handlers tho ( i have only the 'last chance' handler ). it does suggest that sjlj does not speed up ordinary calls substantially tho.

asciilifeform finds that (using pre-ave1 gnat, where i currently can --RTS=sjlj ) no detectable diff in mod ex

asciilifeform goes to test..

asciilifeform nao wonders if there's a seekrit chest fulla ffa speedup in this dig

asciilifeform: iirc on some kernels 'ulimit -s unlimited' worked

asciilifeform: ( and it's lower than installed ram?! )

asciilifeform: ugh is there a hard max ?!

asciilifeform: mircea_popescu: i've had to up stack depth on 1 occasion before -- when testing ffa with ridiculously wide bitnesses ( recall, item runs 100% stackistically )

asciilifeform: ( asciilifeform found that ye olde box cannot actually sustain 100% 8-core for 2+hrs. trips sensor. quite annoying. )

asciilifeform: *overheat

asciilifeform not yet burned down, but managed to overhead $box recently

asciilifeform: diana_coman: fwiw the noise floor on e.g my test box, is 0.003 (i.e. a proggy with empty main)

asciilifeform: a aaa.

asciilifeform: mircea_popescu: i cannot resist to ask, is it 'ragassa' and not 'ragazza' ? ( was it a sicilianism or wat )

asciilifeform: ty

asciilifeform: ( we aren't so much concerned with the 3MB standard lib )

asciilifeform: diana_coman: ideally when you do this, tar up not only the final bin but the contents of obj dir

asciilifeform: aite, i'ma do it to the variant mircea_popescu goes with

asciilifeform: ( from ancient gnat will suffice for time being , for that )

asciilifeform: diana_coman: if you have time, plox post a pair of bins so that i can http://btcbase.org/log/2019-02-14#1896414 tonight ☝︎

asciilifeform: ok

asciilifeform: diana_coman: how quickly is 'very' ? ( can haz numeric ? e.g. 'zxc eats 1kb per level of depth, sjlj - 2kb' )

asciilifeform briefly wondered if ben_vulpes

asciilifeform: !!down drunk_foxx

asciilifeform: aanybody have any idea who ^ is ?

asciilifeform: drunk_foxx: better be quick

asciilifeform: !!up drunk_foxx

asciilifeform: ( and ftr i dun hate winter, either 'bitcoin' or of the ordinary kind, winter is a-ok for so long as you aint stuck in the wind w/out a coat )

asciilifeform: i dun read 'soshulmeadia', dun have any data other than 'no one seems to be offering 20k orc dubloons for coin like yr ago, just nao'

asciilifeform: mircea_popescu: could even say nao is 'bitcoin winter', the sorts of folx inclined to stuff head up arse are stuffing deeper than ever and boasting

asciilifeform: ohai phf

asciilifeform: 'pioneer is the fella with arrow in his back' eh