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
26723 commits 35 branches 284 tags 116 MiB
Commit graph

26723 commits

This branch
This branch
All branches
Author SHA1 Message Date
Stas Boukarev ec111830aa Inline (the fixnum (* x y)) 2025-09-04 02:58:17 +03:00
Stas Boukarev 0e7d8d574d Inline (the fixnum (* x fixnum)) 
Either fully, if X is an integer, or only the fixnum case.
 2025-09-04 02:31:07 +03:00
Stas Boukarev ba1256ce8f Allow %primitive to be used on conditional VOPs. 2025-09-03 21:35:25 +03:00
Stas Boukarev 9ce2d4e0c4 Avoid boxing in (the fixnum (truncate signed signed)) 
Issue an error directly for the bad unsigned word value.
 2025-09-03 21:35:25 +03:00
Douglas Katzman 10e4956715 Add type decls to do-packed-varints 2025-09-03 14:34:28 -04:00
Douglas Katzman 3349f46065 Move atomic-push outside scope of TLS lock on thread exit 
A per-thread lock can not possibly guard sb-sprof::**sprof-data*.
Also, move LISPTHREAD() macro where perform_backtrace can see it.
 2025-09-03 09:39:50 -04:00
Stas Boukarev c8d41e7943 Ensure that (ceiling signed-word signed-word) is transformed. 
Don't rely on the truncate type deriver, look at the surrounding
truly-the as well.
 2025-09-03 07:16:49 +03:00
Stas Boukarev 09cf721f14 Don't transforming ceiling for 2^k on signed-word 
It can overflow the word, rely on the transform to truncate instead.
 2025-09-03 07:16:49 +03:00
Stas Boukarev bc900ffd3d Don't pretend to be able to restart compile-time type errors. 2025-09-03 01:13:43 +03:00
Douglas Katzman 2ce54539b1 Fix typo in unfixnumize_tid macro 2025-09-02 17:34:59 -04:00
Stas Boukarev 0c9408f84d x86-64, signed-byte-64-p-move-to-word: accept unsigned-reg result. 
Fixes lp#2121880
 2025-09-02 23:32:37 +03:00
Douglas Katzman d9ced28312 Remove tid from primitive object 2025-09-02 16:24:41 -04:00
Douglas Katzman 99a46a4b8c Account for possibly untagged tid slot of sb-thread:thread 2025-09-02 16:01:16 -04:00
Douglas Katzman ffc3ec0fbb Remove incorrect line 
If it were to remain, it would have needed make_fixnum,
but as the remark says, it is a redundant assignment.
 2025-09-02 14:56:39 -04:00
Douglas Katzman 11fc027b21 Read thread's kernel thread ID from lisp without a mutex 2025-09-02 14:41:16 -04:00
Stas Boukarev decc429bcd Pick a more appropriate sc-offset-immediate type. 
Fixes lp#2097152
 2025-09-02 21:09:04 +03:00
Douglas Katzman ff9064c9c8 Add sb_GetTID for NetBSD 
patch by Masatoshi SANO
 2025-09-02 13:14:05 -04:00
Douglas Katzman cd95e5ac61 Express more pessimism 2025-09-02 12:07:16 -04:00
Douglas Katzman 11414bbc90 Remove #if'ed out function 2025-09-02 09:31:59 -04:00
Stas Boukarev a832e6887c Reorder transforms. 2025-09-02 07:55:18 +03:00
Stas Boukarev 624fa777ba delay-ir1-transform: prefer :constraint to :ir1-phases 
:constraint was ignored if :ir1-phases was done first.
 2025-09-02 07:29:15 +03:00
Stas Boukarev c38f487f69 Transform (the fixnum (truncate integer 2)) 
Can be done using signed-word arithmetic.
 2025-09-02 05:35:56 +03:00
Stas Boukarev 1c82528eac Avoid warnings from the FLOAT transform. 2025-09-02 05:35:56 +03:00
Stas Boukarev 174c5ac261 Turn (the integer (the (real 0 5))) into (the (integer 0 5)) 2025-09-02 01:20:23 +03:00
Douglas Katzman 0ea74b3dac Avoid aliens in memmove by restricting args to be SAPs 
on x86-64 it shaves off more than half the machine code.
 2025-09-01 15:44:50 -04:00
Douglas Katzman 3b5041d5a5 build-all-cores 2025-09-01 15:41:30 -04:00
Stas Boukarev 3732f97498 Adjust a test some more. 2025-09-01 22:30:25 +03:00
Stas Boukarev c4ab267d3a Adjust a test. 
Try not to leave threads behind.
 2025-09-01 21:50:44 +03:00
Douglas Katzman 940ece0aaf Add decls to sb-unix:unix-read 
Contains no alien manipulations now, just the calls to read() and
os_get_errno() for 40% reduction in asm code size.
 2025-09-01 13:46:52 -04:00
Stas Boukarev 3eb549d69e cut-to-width: erase the derived type of a cast. 
Which won't much its new values.
 2025-09-01 20:43:50 +03:00
Douglas Katzman 1f279ebd29 Require non-NULL thread arg in protect_{thing}_page 2025-09-01 12:34:15 -04:00
Douglas Katzman f112bf8dee Delete irrelevant logic 
There are no hardwired space for x86-64 which was the only platform
where ALLOCATE_LOW was meaningful (and still is for fixedobj space)
 2025-09-01 12:25:51 -04:00
Douglas Katzman dff0d4242d Remove once-used never-changed variable 
Back in rev e365f2f7a9 it might have made sense (being used on both sides
of an #if) but even then it didn't make a ton of sense.
 2025-09-01 11:15:49 -04:00
Stas Boukarev 2da1ea9584 cut-to-width: cut casts with multiple uses. 2025-09-01 18:11:41 +03:00
Douglas Katzman 8ee91012a1 Move a ppc64 hack near other ppc64-specific code 2025-09-01 10:58:43 -04:00
Stas Boukarev 237d391b85 Move stuff from cast-externally-checkable-p into transforms. 2025-09-01 04:12:55 +03:00
Stas Boukarev 0165087f49 Remove :important t as an option to deftransform 
It's not used anywhere, shouldn't be used anywhere. No code merits an
efficiency note without being asked.
 2025-09-01 03:48:40 +03:00
Douglas Katzman 69b22eeca9 Streamline UNIX-WRITE a litte 
(ADDR (DEREF ...)) is a very roundabout way of computing the same result
that SAP+ does, except the former allocate a local alien at runtime.
Despite the new code always binding *PINNED-OBJECTS* on the precise backends,
it's smaller and quicker, especially as 20x more calls to UNIX-WRITE get
an array instead of a SAP in typical use, meaning a pin had to occur.
(Profiling shows array of base-char and UB8 as the commonest cases)
 2025-08-31 16:20:54 -04:00
Douglas Katzman 98d71ca070 Stop calling clock-gettime to gather data that users never see 
It was an attempt to track changes in regression test timing which
segregate compile vs execution times.
 2025-08-31 15:58:11 -04:00
Stas Boukarev 0be32b400b Correct the usage of delete-cast in transforms. 
Can't delete just any cast coming into a transform. While often it
will be an automatically inserted cast for the function, it also might
be a more specific user-supplied one.
 2025-08-31 21:59:30 +03:00
Stas Boukarev ff56ef20a6 Replace the overflow-ash-t-signed VOP with lisp code. 
Which can generate the same code. Unlike the other overflow-t
transforms, it doesn't actually perform any flag checking for overflow
so there's no benefit in it being a VOP.
 2025-08-31 21:58:49 +03:00
Douglas Katzman afb2967d9f Fix build with #+sb-show 2025-08-31 12:41:14 -04:00
Douglas Katzman 0ee871396f Allow minor variance of cl:*features* in snapshot-diff 2025-08-31 12:37:04 -04:00
Douglas Katzman c45469ea77 Fix compilation of standalone ldb if #+permgen 2025-08-31 12:16:39 -04:00
Douglas Katzman 11504281e3 Conditionalize bogus stub 
which if lacking a vop compiles to an infinite loop
 2025-08-31 12:08:26 -04:00
Stas Boukarev f3257aa890 Allow immediate arguments for the error VOPs. 2025-08-30 21:02:21 +03:00
Stas Boukarev 03b730cde7 splice-fun-args: don't get confused by multiple value lvars. 
Reported by Eric Marsden.
 2025-08-30 20:57:13 +03:00
Stas Boukarev 145ec9ac62 Resolve cross-build reproduciblity. 2025-08-30 16:49:19 +03:00
Stas Boukarev 7c81a6098a Transform (the fixnum (ash x -1)) 
(ash (signed-byte 64) -1) can be inlined.
 2025-08-30 06:50:51 +03:00
Stas Boukarev 37f7e8eed4 Allow encoding negative immediate numbers in error breaks. 2025-08-30 06:50:47 +03:00