common-lisp/sbcl.sbcl
1
0
Fork 0
mirror of git://git.code.sf.net/p/sbcl/sbcl synced 2026-09-10 07:26:40 -04:00
Code Issues Actions 8 Packages Projects Releases Wiki Activity
23732 commits 35 branches 284 tags 116 MiB
Commit graph

23732 commits

This branch
This branch
All branches
Author SHA1 Message Date
Douglas Katzman 973cafa3c7 Update math journal again 2024-01-17 17:36:03 -05:00
Stas Boukarev b5688d2616 join-equality-constraints: account all constraints within a block. 
To avoid order dependencies.
 2024-01-18 01:15:08 +03:00
Douglas Katzman f7920517d0 Update math journal 2024-01-17 17:02:53 -05:00
Stas Boukarev 8321c5a3e7 check-range<: use single-float-widetag as the lowest bignum address. 2024-01-17 23:34:15 +03:00
Stas Boukarev 21b8df7190 range<: don't use check-range< if it's already an integer. 2024-01-17 23:14:58 +03:00
Stas Boukarev 805c42675d Fix hide-packages 2024-01-17 23:01:57 +03:00
Stas Boukarev 5e292dd5fc Replace fixnum-mod-p with range<=. 
It can do everything fixnum-mod-p does and then some.

For arm64 so far.
 2024-01-17 22:37:47 +03:00
Stas Boukarev 066f1cee3a type-from-constraints: ignore hairy leaf-types. 
Fixes lp#2049631
 2024-01-17 19:54:28 +03:00
Stas Boukarev cf6f04447b range<: don't check for fixnum for small integers. 
A bignum address can't be lower than some limit, like
+backend-page-bytes+ or lower than 32-bits on darwin-arm64.
 2024-01-17 19:54:28 +03:00
Stas Boukarev cc2f71d569 Wrong file committed. 2024-01-17 18:01:30 +03:00
Stas Boukarev 7f014c3245 range<=/integer: better (<= 0 x power-of-two) 
Same as fixnum-mod-p
 2024-01-17 17:03:16 +03:00
Stas Boukarev 4633c1050c Resolve a style warning on CCL. 2024-01-17 16:05:55 +03:00
Stas Boukarev 4a2d9beb67 Skip typecase-cache.pure.lisp when no multiway-branch-if-eq. 2024-01-17 15:55:37 +03:00
Stas Boukarev 31d6e75925 perfecthash: don't write files. 
Windows has weird tmpfile() behavior.
Windows also doesn't have fmemopen(), so add a simple in-memory printf.

Fixes lp#2049614
 2024-01-17 15:08:12 +03:00
Stas Boukarev 2455a8a837 Add bit-index to +BACKEND-INTERNAL-ERRORS+. 2024-01-17 11:39:50 +03:00
Stas Boukarev 3adabaac21 Base bit-index on maximum-bignum-length. 2024-01-17 11:36:41 +03:00
Douglas Katzman d95ec6af96 Fix bugs from prior TYPECASE change 
- the fallback routine did not store the layout of the object in question,
  but the layout being tested against. This didn't have the desired effect
  of building up a list of "exact" types that will hit subsequently.

- when expanded, the cache was not reassigned at the call site

- remove a nonsense assignment at the (very old) comment saying "Blow [a]way
  the old vector first" when the old vector is still assigned as the vector
  of the cache being copied, so it can't possibly be freed.
 2024-01-16 19:47:07 -05:00
Douglas Katzman ff8ba39661 Transform TYPECASE on unsealed structure hierarchies 
Users aren't in the habit of freezing their types, so this should be
an improvement over a series of type tests.
 2024-01-16 18:07:01 -05:00
Douglas Katzman 480c1539db Update math journal even more 2024-01-16 17:21:59 -05:00
Douglas Katzman 53e40308be Freeze all the things sooner 
Thereby addressing a FIXME from src/code/typep as it turns out
 2024-01-16 16:58:32 -05:00
Douglas Katzman 1ecd28adb3 Take pcl/caches out of make-host-2 
Someone more intrepid than I can figure out the host-sensitivity,
and a few minor rearrangements will solve the problem of compiling
cache-based TYPECASE by ensuring that it's never needed in self-build.
 2024-01-16 16:43:58 -05:00
Stas Boukarev 9de9d194ab Stop looking for constraints in dead blocks. 
Fixes lp#2049531
 2024-01-16 21:37:56 +03:00
Douglas Katzman 343d7ddc88 Compile pcl/cache in make-host-2 2024-01-16 12:44:58 -05:00
Douglas Katzman 550f12bf52 Update math journal more 2024-01-16 12:33:29 -05:00
Douglas Katzman d6ef86bea6 Express ctype->hashset-sym differently for make-host-2 
I started to improve TYPECASE on structure types that are not sealed
(so subtypes could be added) using a PCL cache. The cross-compiler wanted to
use that transform on this function, but it failed in cold-init (after making
caches themselves viable, of course). NOTINLINE TYPEP made things worse,
because parsing would have work sooner. This change avoids relying on an
abstract rendering of TYPEP by making it concrete albeit ad-hoc.
 2024-01-16 12:30:23 -05:00
Douglas Katzman 698ad0f76b Update math journal 2024-01-16 12:25:14 -05:00
Stas Boukarev e6e34b35a6 word=>byte 2024-01-16 12:36:37 +03:00
Stas Boukarev a68d6724f9 Lower array-dimension-limit. 
A power of two INDEX makes for more compact type checks.
Subsequent arithmetic on array lengths is better.
 2024-01-16 12:21:18 +03:00
Stas Boukarev 3c497d4ac9 Fix logtest-memref. 
Check for the actual chosen SC, not just the type.

Fixes lp#2049477
 2024-01-16 12:19:33 +03:00
Douglas Katzman 80453cb12b Use symbolic constants and fix bug in a comment 2024-01-15 19:04:39 -05:00
Douglas Katzman b506f229d5 Fix #-sb-unicode again 2024-01-15 14:36:25 -05:00
Douglas Katzman dab0cbd402 Convert another hash-table to a perfect hash 2024-01-15 13:44:36 -05:00
Douglas Katzman 45d686370f Fix glitch in manual C-to-Lisp translation 2024-01-15 13:07:00 -05:00
Douglas Katzman 1d4f0f3f32 Fix #-sb-unicode 2024-01-15 12:54:00 -05:00
Stas Boukarev cd6033c7ca No need to use %ash/right directly. 2024-01-15 14:09:25 +03:00
Stas Boukarev f99dd5338c Better MIN/MAX expansion. 
Don't cast to real if it's going to be calling </>.
Don't expand to recursive invocations of MIN/MAX, resulting in faster
compilation.
 2024-01-15 12:36:15 +03:00
Douglas Katzman 6b69b17918 Show off a perfect hash function 2024-01-15 03:21:51 -05:00
Douglas Katzman 5a1f17b980 Implement more %ASH/RIGHT vops 2024-01-14 23:53:17 -05:00
Douglas Katzman 9af3741401 Fix hide-packages test failure 2024-01-14 22:29:44 -05:00
Douglas Katzman 8a29eccc3b Put a usable front-end on the C perfect hash generator 2024-01-14 22:22:56 -05:00
Douglas Katzman 5d0d22cfcb mips: add vops for %ASH/RIGHT 2024-01-14 22:03:08 -05:00
Douglas Katzman 5f1f3ca947 Export NEWCHARSTAR-STRING 
Will be needed for reading an s-expression from the perfect hash generator.

Update the math journal for all builds too.
 2024-01-14 19:42:01 -05:00
Douglas Katzman b4dbe43ba8 Commit perfect hash generator of Bob Jenkins 
From https://burtleburtle.net/bob/hash/perfect.html
Renamed perfect.[ch] -> perfecthash.[ch] and perfhex.c -> perfecthex.inc

Despite being ancient and unmaintained, this bit of code avoids
introducing any C++ (like from gperf) into our runtime.
The top of perfecthash.c describes some motivating ideas and
local changes, such as producing Lisp expressions.
 2024-01-14 18:58:29 -05:00
Douglas Katzman 96a29982fe Try a different spelling 
win32 was complaining about a commented fallthrough
 2024-01-14 18:48:49 -05:00
Stas Boukarev c78d6234e7 Inherit equality constraints into the alternative path. 2024-01-14 19:44:32 +03:00
Stas Boukarev 844ac31df3 min/max transforms: use < instead of <=. 
<= is transformed to < later, except for floats. The out of line
MIN/MAX already use </>, so the handling (or mishandling) of NaN is
more consistent.
 2024-01-14 13:39:13 +03:00
Stas Boukarev 9192c41c7b transform-eq-on-words: transform directly to type predicates. 
Which may be done after constraint propagation has been disabled. And
is faster overall.

Fixes lp#2049281
 2024-01-14 00:44:53 +03:00
Stas Boukarev 2c88919aa1 Avoid a loop in equality constraints. 
Don't loop on SETs that change the direction of an equality
constraint.

Fixes lp#2049263
 2024-01-13 22:39:12 +03:00
Douglas Katzman 7e00a81e35 Allow more slop in array-type-hash-mixer test 2024-01-12 17:01:23 -05:00
Stas Boukarev 05df6b2265 Resolve a KLUDGE. 2024-01-13 00:16:55 +03:00