Commit Graph

16 Commits

Author SHA1 Message Date
Matthew Flatt
d4981dd8c3 less code for trap checks
When a proceudre starts with a trap check, move the check to the very
beginning, even before checking the argument count. That way, event
detection can turn into a compact jump to an event handler, instead of
inserting a general call to `$event` in the procedure body.

original commit: 06b12d505698a2378734689370bb9e0f8eda06b9
2020-02-07 10:56:15 -07:00
Matthew Flatt
540c58bbe8 use POPCNT instruction when available on x86_64
On x86_64, a POPCNT instruction is usually available, and it can speed
up `fxpopcount` operations by a factor of 2-3.

Since POPCNT isn't always available, code using `fxpopcount` is
compiled to a call to a generic implementation. The linker substitutes
a POPCNT instruction when it determines at runtime that POPCNT is
available.

Some measurements on a 2018 MacBook Pro (2.7 GHz Core i7) using the
program below:

 popcnt = this implementation, POPCNT discovered
 nocnt  = this implementation, POPCNT considered unavailable
 optcnt = compile to use POPCNT directly (no linker work)
 cpcnt  = compile to inlined generic (no linker work, no POPCNT)

Since the generic implementation is always a 64-bit popcount, it's not
as good as an inlined version for `fxpopcount32`, but otherwise the
link-edit approach to POPCNT works well:

            fxpopcount      fxpopcount32
 popcnt:       0.098s
 nocnt:        0.284s
 optcnt        0.109s  [slower means noise?]
 cpcnt:        0.279s         0.188s

 (optimize-level 3)
 (time
  (let loop ([v #f] [i 100000000])
    (if (fx= i 0)
        v
        (loop (fxpopcount i) (fx- i 1)))))

original commit: 5f090e509f8fe5edc777ed9f0463b20c2e571336
2020-01-11 11:04:48 -07:00
Matthew Flatt
8070a7b910 Merge branch 'eqfl' of github.com:mflatt/ChezScheme
original commit: 8b36396eacb139e0fff70efcd2c9dc842815324f
2019-01-22 05:57:17 -07:00
dyb
ee9a4b3f59 profile counts are now maintained even for code that has been
reclaimed by the collector and must be released explicitly by the
programmer via (profile-release-counters).
  pdhtml.ss, primdata.ss,
  globals.h, externs.h, fasl.c, prim5.c, prim.c, alloc.c, scheme.c,
  misc.ms,
  release_notes.stex, system.stex

original commit: 68e20f721618dbaf4c1634067c2bee24a493a750
2019-01-17 09:43:18 -08:00
Matthew Flatt
3e297e025e adjust make-arity-wrapper to enforce the supplied arity mask
original commit: a9ec7da3ea3b8edc665b060bcba675248119d260
2019-01-15 11:56:03 -07:00
Matthew Flatt
3658fad6b7 Merge branch 'count' of github.com:mflatt/ChezScheme
original commit: 9d0a647e87089c91195cc49480d848212025a7ed
2019-01-05 09:55:07 -07:00
Matthew Flatt
aaaa5fefa1 add compute-size-increments
Also adds `get-initial-thread`, since threa values are useful with
`compute-size[-increments]`.

Changes the compiler to inline `weak-pair?` and `ephemeron-pair?`,
since that provides better performance for `compute-size-increments`.

original commit: 57d0cc13f8e932972cba3837b4f54e9c86786091
2019-01-05 09:49:12 -07:00
Matthew Flatt
48228739fe add object-references to reflect GC's tracing of objects
The `object-references` function is intended to support debugging of
memory leaks by providing a mapping from each live object to the
object that retained it.

original commit: 61f6602b7e6c388c529f3c5995dcf71a7c42e005
2018-07-16 18:08:48 -06:00
Andy Keep
1f0e42f0ec Merge branch 'pthread' of https://github.com/mflatt/ChezScheme into mflatt-pthread-2
original commit: 2230e5adcb61ac8b27c9deee682270f4e17fbafb
2018-04-21 22:09:05 -04:00
Matthew Flatt
9aa1fc4caa add collect-rendezvous
original commit: f7cb82d97e34b14bfbafe635b0d4a294527b02c3
2018-04-02 13:52:03 -04:00
Matthew Flatt
22d4fd9978 Add __thread foreign-call convention
See the `foreign-callable` docs for a good example use.

original commit: e3463c78c511ad861dfa49865bb447e9777f9eb8
2018-03-14 17:20:33 -06:00
Matthew Flatt
743800bbb5 support struct args to and results from foreign procedures
original commit: f0a94bdb9f57c1bf7ffbb66693fb5476a6f0e65b
2018-03-12 21:01:47 -06:00
Bob Burger
831ea8ad18 changed copyright year to 2017
7.ss, scheme.1.in, comments of many files

original commit: 06f858f9a505b9d6fb6ca1ac97234927cb2dc641
2017-04-06 11:41:33 -04:00
Matthew Flatt
21fe925d06 add procedure-arity-mask
original commit: 4bd061000ab903feb3fe8e3b96ecbcb10c59dba9
2017-02-22 07:16:53 -07:00
dybvig
b4d452cc71 - eliminated a couple of thread-safety issues and limitations on the
sizes of pathnames produced by expansion of tilde (home-directory)
  prefixes by replacing S_pathname, S_pathname_impl, and S_homedir
  with S_malloc_pathname, which always mallocs space for the result.
  one thread-safety issue involved the use of static strings for expanded
  pathnames and affected various file-system operations.  the other
  affected the file open routines and involved use of the incoming
  pathname while deactivated.  the incoming pathname is sometimes if not
  always a pointer into a Scheme bytevector, which can be overwritten if a
  collection occurs while the thread is deactivated.  the size limitation
  corresponded to the use of the static strings, which were limited to
  PATH_MAX bytes.  (PATH_MAX typically isn't actually the maximum path
  length in contemporary operating systems.)  eliminated similar issues
  for wide pathnames under Windows by adding S_malloc_wide_pathname.
  consumers of the old routines have been modified to use the new
  routines and to free the result strings.  the various file operations
  now consistently treat a pathname with an unresolvable home directory
  as a pathname that happens to start with a tilde.  eliminated unused
  foreign-symbol binding of "(cs)pathname" to S_pathname.
    io.c, externs.h, new_io.c, prim5.c, scheme.c, prim.c
- various places where a call to close or gzclose was retried when
  the close operation was interrupted no longer do so, since this can
  cause problems when another thread has reallocated the same file
  descriptor.
    new_io.c
- now using vcvarsall type x86_amd64 rather than amd64 when the
  former appears to supported and the latter does not, as is the
  case with VS Express 2015.
    c/Mf-a6nt, c/Mf-ta6nt
- commented out one of the thread mats that consistently causes
  indefinite delays under Windows and OpenBSD due to starvation.
    thread.ms
- increased wait time for a couple of subprocess responses
    6.ms
- added call to collector to close files opened during iconv mats
  specifically for when mats are run under Windows with no iconv dll.
    io.ms

original commit: ad44924307c576eb2fc92e7958afe8b615a7f48b
2016-06-16 23:04:32 -04:00
dyb
1356af91b3 initial upload of open-source release
original commit: 47a210c15c63ba9677852269447bd2f2598b51fe
2016-04-26 10:04:54 -04:00