[gambit-list] Can I imply a FORCE all over? (**Prio Q for this weekend or so :D )

Adam adam.mlmb at gmail.com
Thu Jun 9 23:14:21 EDT 2016


I guess

1) Correct (no freezing)
2) Yes, the force is implied exactly everywhere, and
3) Something like 20% on a good CPU.

Please correct me if I would be wrong.


2016-06-10 10:46 GMT+08:00 Adam <adam.mlmb at gmail.com>:

> Aha.
>
> Brief Q:s:
>
> 1) The force code is not blocking the thread scheduler, right - so the
> lazy expression can do (thread-sleep! alot) without freezing the host,
> right?
>
> 2) The auto-forcing is universal, right?
>
> 3) Any guess about how much slower the generated code becomes, do you
> think it's rather 20% or 400%?
>
>     I guess if on the C/assembly level the check is really simple, then a
> good CPU will be able to optimize through already-forced values via branch
> prediction so rather 20% .. ?
>
>
>
> 2016-06-10 10:40 GMT+08:00 Marc Feeley <feeley at iro.umontreal.ca>:
>
>> When using --enable-auto-force, it is OK for test1 to fail this way
>> (because test1 introspects the structure of the generated code, and,
>> unfortunately, using auto forcing changes the structure of the code).
>>
>> Marc
>>
>> > On Jun 9, 2016, at 4:00 PM, Bradley Lucier <lucier at math.purdue.edu>
>> wrote:
>> >
>> > I'm taking the liberty of adding the Gambit mail list to the CC, since
>> this may be of general interest.
>> >
>> > "make check" fails after configuring with
>> >
>> > ./configure 'CC=/usr/local/gcc-6.1.0/bin/gcc -march=native'
>> '--enable-single-host' '--enable-multiple-versions' '--enable-shared'
>> --enable-auto-forcing
>> >
>> > "make -i check" (to ignore errors and continue) gives
>> >
>> > ------------ TEST 1 (debugging support)
>> >
>> LD_LIBRARY_PATH=../lib:../gsi:../gsc:/usr/local/Gambit/current/lib:/usr/local/netpbm10/lib:/usr/local/gcc-6.1.0/lib/:
>> ../gsi/gsi -:~~bin=../bin,~~lib=../lib,~~include=../include -f debug.scm >
>> test1.out
>> > diff test1.ok test1.out && rm -f test1.out
>> > 81a82
>> >> #return(##cprc-if2 3) : ($code rte)
>> > 85a87
>> >> #return(##cprc-if3 3) : ($code rte)
>> > 108a111
>> >> #return(##cprc-cond-if 3) : ($code rte)
>> > 112a116
>> >> #return(##cprc-cond-or 3) : ($code rte)
>> > 117,120c121,126
>> > < #return(##cprc-cond-send-red 3) : ($code rte pred)
>> > < #procedure(##cprc-cond-send-red 4)
>> > < #return(##cprc-cond-send-red 5) : ($code rte oper)
>> > < #return(##cprc-cond-send-red 6) : ($code rte)
>> > ---
>> >> #return(##cprc-cond-send-red 3) : ($code rte)
>> >> #return(##cprc-cond-send-red 4) : ($code rte pred)
>> >> #return(##cprc-cond-send-red 5) : ($code rte pred)
>> >> #procedure(##cprc-cond-send-red 6)
>> >> #return(##cprc-cond-send-red 7) : ($code rte oper)
>> >> #return(##cprc-cond-send-red 8) : ($code rte)
>> > 125,127c131,135
>> > < #return(##cprc-cond-send-sub 3) : ($code rte pred)
>> > < #return(##cprc-cond-send-sub 4) : ($code rte oper)
>> > < #return(##cprc-cond-send-sub 5) : ($code rte)
>> > ---
>> >> #return(##cprc-cond-send-sub 3) : ($code rte)
>> >> #return(##cprc-cond-send-sub 4) : ($code rte pred)
>> >> #return(##cprc-cond-send-sub 5) : ($code rte pred)
>> >> #return(##cprc-cond-send-sub 6) : ($code rte oper)
>> >> #return(##cprc-cond-send-sub 7) : ($code rte)
>> > 131a140
>> >> #return(##cprc-or 3) : ($code rte)
>> > 135a145
>> >> #return(##cprc-and 3) : ($code rte)
>> > 140c150,151
>> > < #return(##cprc-case 3) : (rte $code)
>> > ---
>> >> #return(##cprc-case 3) : ($code rte)
>> >> #return(##cprc-case 4) : (rte $code)
>> > 250,251c261,263
>> > < #procedure(##cprc-app0-red 3)
>> > < #return(##cprc-app0-red 4) : ($code rte)
>> > ---
>> >> #return(##cprc-app0-red 3) : ($code rte)
>> >> #procedure(##cprc-app0-red 4)
>> >> #return(##cprc-app0-red 5) : ($code rte)
>> > 257,259c269,272
>> > < #procedure(##cprc-app1-red 4)
>> > < #return(##cprc-app1-red 5) : ($code rte oper)
>> > < #return(##cprc-app1-red 6) : ($code rte)
>> > ---
>> >> #return(##cprc-app1-red 4) : ($code rte oper)
>> >> #procedure(##cprc-app1-red 5)
>> >> #return(##cprc-app1-red 6) : ($code rte oper)
>> >> #return(##cprc-app1-red 7) : ($code rte)
>> > 266,268c279,282
>> > < #procedure(##cprc-app2-red 5)
>> > < #return(##cprc-app2-red 6) : ($code rte oper)
>> > < #return(##cprc-app2-red 7) : ($code rte)
>> > ---
>> >> #return(##cprc-app2-red 5) : ($code rte oper arg1)
>> >> #procedure(##cprc-app2-red 6)
>> >> #return(##cprc-app2-red 7) : ($code rte oper)
>> >> #return(##cprc-app2-red 8) : ($code rte)
>> > 276,278c290,293
>> > < #procedure(##cprc-app3-red 6)
>> > < #return(##cprc-app3-red 7) : ($code rte oper)
>> > < #return(##cprc-app3-red 8) : ($code rte)
>> > ---
>> >> #return(##cprc-app3-red 6) : ($code rte oper arg1 arg2)
>> >> #procedure(##cprc-app3-red 7)
>> >> #return(##cprc-app3-red 8) : ($code rte oper)
>> >> #return(##cprc-app3-red 9) : ($code rte)
>> > 287,289c302,305
>> > < #procedure(##cprc-app4-red 7)
>> > < #return(##cprc-app4-red 8) : ($code rte oper)
>> > < #return(##cprc-app4-red 9) : ($code rte)
>> > ---
>> >> #return(##cprc-app4-red 7) : ($code rte oper arg1 arg2 arg3)
>> >> #procedure(##cprc-app4-red 8)
>> >> #return(##cprc-app4-red 9) : ($code rte oper)
>> >> #return(##cprc-app4-red 10) : ($code rte)
>> > 297,298c313,315
>> > < #procedure(##cprc-app-red 6)
>> > < #return(##cprc-app-red 7) : ($code rte)
>> > ---
>> >> #return(##cprc-app-red 6) : ($code rte oper)
>> >> #procedure(##cprc-app-red 7)
>> >> #return(##cprc-app-red 8) : ($code rte)
>> > 303a321
>> >> #return(##cprc-app0-sub 4) : ($code rte)
>> > 310c328,329
>> > < #return(##cprc-app1-sub 5) : ($code rte)
>> > ---
>> >> #return(##cprc-app1-sub 5) : ($code rte oper)
>> >> #return(##cprc-app1-sub 6) : ($code rte)
>> > 317,318c336,338
>> > < #return(##cprc-app2-sub 5) : ($code rte oper)
>> > < #return(##cprc-app2-sub 6) : ($code rte)
>> > ---
>> >> #return(##cprc-app2-sub 5) : ($code rte oper arg1)
>> >> #return(##cprc-app2-sub 6) : ($code rte oper)
>> >> #return(##cprc-app2-sub 7) : ($code rte)
>> > 326,327c346,348
>> > < #return(##cprc-app3-sub 6) : ($code rte oper)
>> > < #return(##cprc-app3-sub 7) : ($code rte)
>> > ---
>> >> #return(##cprc-app3-sub 6) : ($code rte oper arg1 arg2)
>> >> #return(##cprc-app3-sub 7) : ($code rte oper)
>> >> #return(##cprc-app3-sub 8) : ($code rte)
>> > 336,337c357,359
>> > < #return(##cprc-app4-sub 7) : ($code rte oper)
>> > < #return(##cprc-app4-sub 8) : ($code rte)
>> > ---
>> >> #return(##cprc-app4-sub 7) : ($code rte oper arg1 arg2 arg3)
>> >> #return(##cprc-app4-sub 8) : ($code rte oper)
>> >> #return(##cprc-app4-sub 9) : ($code rte)
>> > 345,346c367,369
>> > < #procedure(##cprc-app-sub 6)
>> > < #return(##cprc-app-sub 7) : ($code rte)
>> > ---
>> >> #return(##cprc-app-sub 6) : ($code rte oper)
>> >> #procedure(##cprc-app-sub 7)
>> >> #return(##cprc-app-sub 8) : ($code rte)
>> > makefile:113: recipe for target 'test1' failed
>> > make[1]: [test1] Error 1 (ignored)
>> >
>> >
>> > I don't know whether this is important.
>> >
>> > Brad
>> >
>> > On 06/09/2016 02:11 PM, Mikael More wrote:
>> >> Awesomeness, thanks!!!!!
>> >>
>> >> Fantastic, this means tight integration with a functional object
>> >> database is possible.
>> >>
>> >> I guess the expression code is running under the scheduler too, opening
>> >> for storage of the database on non-low-latency facilities too, great!
>> >>
>> >>
>> >>
>> >> 2016-06-10 1:30 GMT+08:00 Bradley Lucier <lucier at math.purdue.edu
>> >> <mailto:lucier at math.purdue.edu>>:
>> >>
>> >>   Try adding --enable-auto-forcing to your configure of Gambit.
>> >>
>> >>
>> >>   On 06/09/2016 01:20 PM, Adam wrote:
>> >>
>> >>       Hi Marc!
>> >>
>> >>       I have a practical question:
>> >>
>> >>       I am thinking about the possibility of realizing a TIGHTLY
>> >>       integrated
>> >>       persistent object store ("object database").
>> >>
>> >>       (Tightly integrated, that is, with the Scheme heap.)
>> >>
>> >>       The object store would be functional, so the host system can
>> cache
>> >>       object references as much as it wants.
>> >>
>> >>
>> >>       My question is:
>> >>
>> >>       The best thing would be to
>> >>
>> >>       1) Have lazy loading of objects into the Scheme heap from the
>> >>       database
>> >>       (because the stored structures may have wather depth and shape
>> >>       meaning
>> >>       that precaching algorithms never would have an idea about what
>> >>       really
>> >>       needs to be loaded, so better let the code show that simply
>> >>       through what
>> >>       slots etc. it accesses), *AND* to at the same time
>> >>
>> >>       2) *NOT* need to customize the Scheme code for the lazy loading,
>> >>       i.e.,
>> >>       have implicit |force| all over the code?
>> >>
>> >>
>> >>       A final wish here would be
>> >>
>> >>       3) The option that the |force| would be run as a procedure
>> >>       within the
>> >>       green threading system, so that other green threads would
>> continue
>> >>       executing while the lazy value is being loaded. Not sure about
>> >>       this one
>> >>       though - maybe for code-performance reasons it would be better
>> >>       to have
>> >>       it block.
>> >>
>> >>
>> >>       Even runtime procedures would need to do the implicit |force|.
>> >>       So the
>> >>       values that not are loaded right now, would be represented by a
>> >>       loading
>> >>       closure, and the result of closure invocation would be the value
>> >>       that
>> >>       should take the place of the respective accessed slot.
>> >>
>> >>
>> >>       Can this be done in Gambit, in its current form or with some
>> >>       very slight
>> >>       modification?
>> >>
>> >>       Thanks!
>> >>
>> >>
>> >>
>> >
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.iro.umontreal.ca/pipermail/gambit-list/attachments/20160610/d44fc6e5/attachment.htm>


More information about the Gambit-list mailing list