2064 commits

Author	SHA1	Message	Date
daanx	498c92e348	update guarded implementation to use block tags	2024-11-16 21:15:50 -08:00
daanx	0e76fe3798	add address hint to primitive allocation API	2024-11-16 16:57:37 -08:00
daanx	b27d154ffd	add virtual address bits and physical memory to the mem config	2024-11-16 15:37:08 -08:00
David Carlier	9f0a7ac0be	_mi_memcpy/_mi_memzero: tighten criteria for intrinsics for windows. ... FSRM is better used for buffer <= 128 bytes and ERMS, if supported for larger chunks.	2024-11-16 10:05:14 +00:00
Daan	4f46cf7d5a	ensure we dont reclaim a segment on a free if that would go above the target segment count	2024-10-29 22:40:58 -07:00
Daan	deba02e7be	Merge branch 'dev' into dev-slice	2024-10-29 20:08:45 -07:00
Daan	b5ae6fc555	remove wrong assertion	2024-10-29 20:08:36 -07:00
Daan	eda16d7c91	remove wrong assertion	2024-10-29 20:07:35 -07:00
Daan	471bc768e3	Merge branch 'dev-slice' into dev-slice-steal	2024-10-27 22:20:19 -07:00
Daan	c0e1132674	merge from dev	2024-10-27 22:17:23 -07:00
Daan	30640233a1	Merge pull request #943 from asdf-bro/patch-1 ... Musl needs __libc* functions too	2024-10-27 22:15:43 -07:00
Daan	c59b0cd7fc	Merge pull request #946 from ArtSin/fix-vsnprintf-int-long-intmax_t ... Fix int and long handling and the use of (u)intptr_t in _mi_vsnprintf	2024-10-27 22:12:29 -07:00
Daan	b3828bba9e	disable aligned hinting or SV39 mmu's, issue #939, and pr #949	2024-10-27 21:58:20 -07:00
Daan	5f35933331	add 0 byte to canary to prevent spurious read overflow to read the canary (issue #951, pr #953)	2024-10-27 21:39:07 -07:00
Daan	afba03145c	Merge pull request #954 from jimwang118/fix-linux ... Fix build error on x64-android	2024-10-27 21:12:58 -07:00
Daan	06ee1d7949	Merge branch 'dev' into dev-slice	2024-10-27 21:11:04 -07:00
Daan	f126b50382	update comments, set constructor priority to 101 on macOS	2024-10-27 21:10:46 -07:00
Jim-Wang	17802e6528	fix build error on linux	2024-10-28 02:06:58 +00:00
Daan	d3455ea29e	Merge branch 'dev' into dev-slice	2024-10-27 18:17:38 -07:00
Daan	ee92b337b9	do not reclaim segments if free-ing from a thread with an already abandoned heap (issue #944)	2024-10-24 00:13:07 -07:00
Daan	2b0d039cf3	fix assertion check	2024-10-23 01:21:41 -07:00
Daan	925efaeac9	improve windows static library initialization to account for thread local destructors (issue #944)	2024-10-23 01:10:00 -07:00
Daan	d951b4dd23	add missing mi_thread_done definition	2024-10-23 00:53:17 -07:00
Daan	6e9b38ac12	fix issue where searching for abandoned blocks would skip the first one	2024-10-22 18:58:55 -07:00
Daan	dfdb9cb877	cleanup process init/done	2024-10-22 06:52:34 -07:00
Daan	104e821709	fix fast divisor for 32-bit platforms	2024-10-22 06:08:56 -07:00
Daan	46e9e7fdd0	fix win32 compilation	2024-10-22 06:06:15 -07:00
Daan	e55ae0aeb7	fix duplicate definition on windows	2024-10-21 23:09:14 -07:00
Daan	f971bd6d74	fix build on windows	2024-10-21 23:05:51 -07:00
Daan	aa881733d7	reorganize primitives for process initialization; use special data segment on Windows for thread termination by default on Windows now (issue #869)	2024-10-21 22:56:59 -07:00
Daan	e58e41c8da	merge from dev	2024-10-21 05:07:12 -07:00
Daan	638ea539de	allow certain options to have defaults set via the pre-processor at build time -- see issue #945	2024-10-21 05:04:01 -07:00
ArtSin	bf251b27b1	Fix int and long handling and the use of (u)intptr_t in _mi_vsnprintf ... 'va_arg' with type 'long' was used for both 'int' and 'long' arguments, but x86_64 Linux has 64-bit 'long', so passing an 'int' argument to '_mi_vsnprintf` results in undefined behaviour. 'intptr_t' was used as the largest integer type, but on 32-bit systems it is 32-bit wide, while 'long long' is 64-bit. 'intmax_t' can be used instead.	2024-10-15 13:39:28 +04:00
Daan	81da26d7d3	make target test for stealing one less since we are about to reclaim_or_alloc a fresh segment	2024-10-11 10:52:35 -07:00
Daan	19ce2c6461	restore randomization when trying to reclaim abandoned segments	2024-10-11 10:44:43 -07:00
daanx	723869014f	add ability to abandon segments after a threshold	2024-10-09 21:24:20 -07:00
Daan Leijen	4913c2c65b	Merge branch 'dev-steal' into dev-slice-steal	2024-10-09 15:16:13 -07:00
Daan Leijen	a7e7cbac89	use enqueue_from_full, and keep inserting at the end	2024-10-09 15:15:57 -07:00
daanx	428e3b9ee0	Merge branch 'dev-steal' into dev-slice-steal	2024-10-09 15:05:48 -07:00
daanx	0316cb1d88	reduce page search to 8	2024-10-09 15:05:35 -07:00
daanx	933ac5c14c	Merge branch 'dev-steal' into dev-slice-steal	2024-10-09 14:41:25 -07:00
daanx	cbc0e19809	revert back to unfull at the end of queues as it slows down some benchmarks (like alloc-test1)	2024-10-09 14:41:12 -07:00
daanx	03fbaedec5	Merge branch 'dev-steal' into dev-slice-steal	2024-10-09 14:35:49 -07:00
daanx	96877159c2	insert full pages that became unfull, at the start of the page queue to increase potential reuse	2024-10-09 14:35:33 -07:00
daanx	a05b5ab0a1	search N pages for a best fit	2024-10-09 14:21:50 -07:00
daanx	b3d502179a	Merge branch 'dev-steal' into dev-slice-steal	2024-10-09 11:28:33 -07:00
daanx	ffa8bce581	prefer pages that do not expand	2024-10-09 11:27:57 -07:00
Philip Brown	f38816d4ed	Musl needs __libc* functions too	2024-10-06 15:42:46 -05:00
Daan Leijen	3ec3aaf858	merge from dev	2024-08-21 11:33:11 -07:00
Daan Leijen	6a21db1017	Merge branch 'dev-guarded' into dev-slice-guarded	2024-08-21 11:31:10 -07:00