28829 commits

This branch

This branch

All branches

Author	SHA1	Message	Date
Douglas Katzman	9c4d4347c2	Avoid SETF of stmt-op and stmt-operands ... Assembler backends typically treat these as immutable entities because mutation tends to be a more bug-prone approach. The current design still allows mutation, but we can mostly hide it.	2026-09-02 23:01:52 +00:00
Douglas Katzman	252a0b200c	Rename STMT-MNEMONIC to STMT-OP ... Easier to type, and matches standard assembler terminology	2026-09-02 19:24:28 +00:00
Douglas Katzman	386706b438	Update crossbuild makefile	2026-09-02 19:09:44 +00:00
Douglas Katzman	3dfe0b3e65	Fix sb-manual build when SBCL_CONTRIB_BLOCKLIST is nonempty	2026-09-02 16:07:22 +00:00
Stas Boukarev	087cb81385	arm64/return: avoid overwriting tmp-tn by load-store-offset ... Load store offsets are good for 4096 values, but it's better to ensure an error is signaled.	2026-09-02 13:52:20 +03:00
Stas Boukarev	974e29da3b	arm64/return: don't put descriptors into tmp-tn	2026-09-02 12:44:56 +03:00
Stas Boukarev	0b7c732828	arm64/return: store to (thread-mv-count) first	2026-09-02 10:48:00 +03:00
Stas Boukarev	c8d0ca77a9	Use CMP instead of SUBS when possible	2026-09-02 10:41:59 +03:00
Douglas Katzman	4ab4654f68	arm64: implement tls-based-mv-return ... Authored by Antigravity	2026-09-01 22:06:41 +00:00
Douglas Katzman	78e706bf96	Generalize the number of MV return regs in gc-common ... And treat the values as movable roots, not pinned roots	2026-09-01 21:37:55 +00:00
Stas Boukarev	1c95a90c3f	arm64: fix passing empty alien structs	2026-09-02 00:30:03 +03:00
Stas Boukarev	ce99730828	x86-64: fix passing packed structs by value ... Reported by Andrew Wolven.	2026-09-02 00:24:05 +03:00
Stas Boukarev	b746310040	Skip tests on #+interpreter	2026-09-01 18:01:23 +03:00
Douglas Katzman	9daea02bac	win32: disable agressive stack scrub	2026-09-01 09:59:48 -04:00
Stas Boukarev	303db288a5	Correctly not check ftypes of unused optionals ... Fixes lp#2165974	2026-09-01 16:53:10 +03:00
Charles Zhang	9685c98e78	Auto-stack-allocate self and mutually-referential functions as well. ... Teach environment analysis how to detect when functions which have references in other functions (potentially itself) are automatically stack allocatable. This is done by marrying the closure environment analysis with some of the existing escape analysis utilities, and by combining those analysis with a step to annotate the correct derived dynamic extent lifetime, which further required some IR1 flow-graph hacking utilities to revoke tail call semantics when necessary. Additionally, the escape analysis performed for functions is also now somewhat unified with how values are handled, so such functions are now also stack allocated when all their references deliver values to e.g. dx-safe functions. In fact, dx-safe as an ir1-info annotation should probably just be removed in favor of user-defined dx declarations. We have thus deleted the arbitrary restriction on auto-dxification of functions that made us just give up the moment a reference to a potentially dxable function was in a different environment. This fixes the existing expected test failures we had regarding being able to auto stack allocate self-referential functions. We also add a bunch of new tests to exercise stack-allocatability and correctness of mutually referential functions which close over each other as well. We can now remove some explicit dynamic extent declarations in the system that are now automatically inferred by the compiler, since it is safer that way in case the functions are changed. It may work to remove more declarations but these are clearly the instances that the compiler now handles. Some test cases provided by Gemini. Everything else written by me.	2026-09-01 13:09:51 +02:00
Charles Zhang	ee1e63f0fb	Be even more conservative around cons and non-simple array types. ... While it could be argued that (lambda (x) (declare (optimize speed)) (declare (type (array * (4 4)) x)) (adjust-array x '(3 5)) (array-dimension (the (array t) x) 0)) should be able to return 4 (I think this is rather dubious though), clearly it is not correct for the attached test case to be able to do so. Because declared types do not know anything about variable subsitution, do not try and squeeze more type information from a declared type if there is more than one reference to the variable. While it is unfortunate that a test had to be disabled as a worse type is derived, I don't see how to safely derive a better type without making other cases incorrect. This could be fixed as a follow-up. Notably (lambda (x) (declare (type (or null (cons fixnum)) x)) (setf (car x) 'foo) (if x (+ (car x) 4) 0)) simply crashes without this fix and is not safe, which is not good even if it's undefined behavior. The corresponding test case with a variable substitution is not only unsafe but incorrect despite being defined behavior. Fixes lp#2165835.	2026-09-01 13:09:51 +02:00
Charles Zhang	2ca5d27079	Delete random macrolet.	2026-09-01 13:09:51 +02:00
Charles Zhang	b69b11cc26	stack: Fix preservation of intervening stack lifetimes, again^n. ... We were sometimes merging dynamic extent lifetimes in such a way that the stack lvar of a dynamic extent later in the flowgraph was used instead of the earliest one. Fix this by checking the state of the stack and making sure that dynamic extents always end up taking on the stack lvars of dynamic extents that happen before them in the flowgraph when merging their lifetimes. Fix found with assistance from Gemini Pro after much heavy hinting from me. Extended commentary and test case reduction by me. Fixes lp#2156347.	2026-09-01 13:01:00 +02:00
Charles Zhang	2f5e20b7af	Endow DX semantics to SET values at IR1tran time. ... Instead of waiting for environment analysis to scan set nodes for dxable values, we can just create a dynamic extent per lambda and associate it with the value of a SET during IR1 translation. Environment analysis can then run as normal and only needs to worry about checking whether the environments of the value and the dynamic extent are the same. Doing so provides the following benefits (notably fixing a bug): * We make handling of DX and sets more uniform, so that we no longer need to handle a case where let conversion has happened or not. This has the following benefits: ** Assignment lambdas don't confuse DX anymore. This fixes lp#2163043. ** Flushed initial values don't cause pessimization anymore, fixing a FIXME. * It is now straightforward/more clear how to make the compiler handle free dynamic-extent declarations. * We can optimize or transform away SET nodes (think copy propagation or SSA type optimizations) during IR1 optimization without worrying about additional DX bookkeeping. Also, revise some commentary about dynamic extent handling across locall and envanal.	2026-09-01 13:00:58 +02:00
Stas Boukarev	8e964fd583	Add shortcuts for bignum * and /	2026-09-01 07:08:06 +03:00
Stas Boukarev	695553ac46	Improve intexp around ratios and bignums ... Avoiding unnecessary negations.	2026-09-01 07:08:06 +03:00
Douglas Katzman	310019ac0e	Add a hack to make GC weak-hash-table tests reliable ... The optimization of not scrubbing the part of the control stack that you think you don't have to scrub is contrary to the requirement of some tests. Real use would not often care. As mentioned in the prior commit, the scrubbing logic was actually deficient going back nearly 15 years, so the new logic is merely differently deficient but geared toward avoiding touching all memory of all stacks of all threads on every GC.	2026-09-01 00:04:57 -04:00
Douglas Katzman	c0ab42f560	Add a fallback path for arch_scrub_control_stack	2026-08-31 22:59:19 -04:00
Douglas Katzman	ba8cedba53	Simplify and speed up arch_scrub_control_stack ... It now uses a heuristic to decide where to stop clearing memory, which is less technically accurate than before, but the key observation is that this code was wrong for 15 years and therefore it's hard to say it takes less time to do more work. i.e. because it never cleared enough, it stands to reason that it might take more CPU time, though algorithmically it's better. Hopefully this is also in practice better at reducing floating garbage. Reviewed by Gemini	2026-08-31 21:53:35 -04:00
Stas Boukarev	774ad4de28	Speed up gmp-intexp ... Faster dispatch to orig-intexp. Faster (expt ratio -power)	2026-09-01 03:03:25 +03:00
Ilia Gradina	c998d48297	sb-gmp: hand a non-rational base back to the original INTEXP ... EXPT dispatches to INTEXP for a (COMPLEX RATIONAL) base as well as for fixnums, bignums and ratios: (((foreach fixnum (or bignum ratio) (complex rational)) integer) (intexp base power)) GMP-INTEXP handles integers, ratios and 0/+-1, then sends everything else to MPZ-POW, which only takes an integer. So merely loading the contrib - LOAD-GMP calls INSTALL-GMP-FUNS itself, and again from an init hook - turns a working EXPT into a type error: * (expt #c(1/5 1) 1) ; Evaluation aborted on #<TYPE-ERROR expected-type: INTEGER ; datum: #C(1/5 1)> Constant folding hides it from a toplevel REPL form, so it only shows up in compiled code with a computed base. Maxima hits it through bernstein_poly, where it fails one of the regression tests on any build linked against sb-gmp. Send non-rational bases to ORIG-INTEXP, which handles them by repeated squaring; GMP keeps the integer and ratio paths it was installed for.	2026-09-01 03:03:25 +03:00
Stas Boukarev	39303703db	with-interrupt-bindings: mv-count is a fixnum	2026-09-01 00:53:12 +03:00
Stas Boukarev	de01b60ddb	Try more values in mv-return.impure ... Some backends use 4 or 6 registers	2026-09-01 00:36:13 +03:00
Stas Boukarev	1dc3d56d15	Redo a test without using threads	2026-09-01 00:23:12 +03:00
Douglas Katzman	17d379157f	Forgot a DX	2026-08-31 16:45:30 -04:00
Douglas Katzman	3a3d112d7e	Fix interrupt-thread for #+tls-based-mv-return	2026-08-31 16:35:26 -04:00
Gabor Melis	9c2cbbd7d1	sb-manual: don't use version controlled symlinks ... They are a hassle on Windows.	2026-08-31 10:20:39 +02:00
Stas Boukarev	ce95b400c9	Resolve a GC issue with #+tls-based-mv-return ... Write the high water mark before storing things on the return stack.	2026-08-31 10:34:45 +03:00
Stas Boukarev	66e972ebc0	#+tls-based-mv-return return: don't write RCX before it's needed	2026-08-31 03:13:58 +03:00
Stas Boukarev	2a3976842d	Better return value tn movement with #+tls-based-mv-return ... E.g. (values 1 2 3 4 5 7 8 9 10 11 12 13) doesn't move into stack locations before moving into its passing place.	2026-08-31 03:02:22 +03:00
Stas Boukarev	565e932d70	Reduce DISPATCH-RATIO expansion size ... Don't check for fixnum ratio components if the integer it'll be added to is a bignum.	2026-08-31 01:00:58 +03:00
Stas Boukarev	8b499bda00	More ratio arithmetic shortcuts ... Like dividing by -1 adding 0, etc.	2026-08-31 00:45:27 +03:00
Douglas Katzman	c5d1a5ce8e	Enable a crossbuild test of :tls-based-mv-return	2026-08-30 16:57:03 -04:00
Douglas Katzman	4abf6ef5f7	x86-64: Implement "possible win" cited in calling-convention.texinfo ... Create a dedicated area of 'struct thread' to return multiple-values in excess of the arg-passing registers. Gating feature is :TLS-BASED-MV-RETURN and not enabled by default. The limit of 64 (inclusive) values is the minimum to get most existing tests to pass unchanged. After the optimizations outlined below are done, the feature will be more compelling, and we can see about making it opt-out rather than opt-in, or making MULTIPLE-VALUES-LIMIT configurable. (I'm sure that some user will claim to need 1000 values.) The benefit will be fully realized in some subsequent changes, namely: freeing a DX alloc, unbinding a special var, and breaking up a catch block will all return from their frame without invoking RETURN-MULTIPLE. Similarly, some "simple" unwind-protect forms such as release-mutex could too- anything that does not clobber the thread-local area during a cleanup. Also this alleviates a big impediment to creating an LLVM-based backend which could explore performance differences among the call conventions offered, since micro-management of the stack becomes less critical. Co-authored with Antigravity	2026-08-30 15:51:25 -04:00
Stas Boukarev	b36bdc5855	More compact fop-simd-pack	2026-08-30 21:19:41 +03:00
Douglas Katzman	e5daa56dc3	build-all-cores	2026-08-30 12:28:02 -04:00
Stas Boukarev	71c43e5aa0	Don't define very large simd-pack functions for printing ... Just use sap-refs from memory. Saves 1.1MB of core space.	2026-08-30 16:41:43 +03:00
Stas Boukarev	e60afcbe4e	arm64: use a flagged :conditional %digit-0-or-plusp VOP	2026-08-30 12:44:12 +03:00
Stas Boukarev	ba19daa024	Don't define a no-op splat as a VOP ... It creates non-transparent moves.	2026-08-30 12:38:11 +03:00
Stas Boukarev	01bd7882b1	More partial result modification false positives	2026-08-30 12:06:55 +03:00
Stas Boukarev	2e9abed30d	arm64: more optimizable integer-length ... arm64 has leading sign/zeros instructions, to get integer-length they need to be subtracted from 63/64. Doing it in VOPs is opaque, move to a lisp transform and two cls/clz VOPs. That way (- 63 (integer-length x)) is just one instruction.	2026-08-30 09:57:09 +03:00
Stas Boukarev	94b6820242	expt: use (= (ftruncate n) n) to check for integers ... It's no slower, and can be faster if #+round-float is available. It's certainly simpler and more compact.	2026-08-30 08:59:42 +03:00
Stas Boukarev	c33aaa910b	arm64: better complex-float-=	2026-08-30 08:06:30 +03:00
Stas Boukarev	5b897710fc	arm64: use zip1 for creating complex floats	2026-08-30 08:06:30 +03:00