Optimize non-tail full calls followed by "easy" cleanups that do not execute
arbitrary forms. "Easy" includes DX allocations, special var unbinding, popping
the CATCH block stack. Then the unknown values can be directly returned as if
by tail call, bypassing the RETURN-MULTIPLE helper routine.
We annotate return LVAR as :pass-through, suppress register restoration
after the call (:save-p :compute-only), and emit return as LEAVE+RET.
Cleanup VOPs (unbind-n, %catch-breakup) must preserve the carry flag, and use
only clobberable temporary registers.
For CATCH, decouple the normal return path from the NLX path. The normal path
eliminates use of the stack for buffering the mv-save area and then restoring
it from the stack.
Don't maintain a vector with character sizes.
Calculate the size from the byte buffer up to the current character
buffer position.
Reduces code duplication, reduces work if no file-position is needed.
The technique is to find "equivalent" loads in between which there is no
computation that affects the result of the load. Also it needs a surrounding
LET, which is augmented with a new temp variable as if the user did that.
This is slightly deficient for various reasons:
1. it won't do anything without the LET
2. the more kinds of common subexpressions we allow (such as math),
the slower IR1-OPTIMIZE-COMBINATION is going to run
3. there are possibly other node types that should be allowed to intervene
On the plus side, it's not all that hard to extend the logic to accept
other functions as participants in common subexpressions.
All test cases plus a little bit of assistance from Gemini
concatenated-stream wasn't processing buffered input correctly.
two-way-stream and echo-stream should just return NIL, as
file-position can't apply to both streams.
It doesn't work to omit avx512 from the darwin config which may have been
inadvertent. It's suspicious that a new non-working Makefile was checked in.
And I'm not sure why there are 5 features keywords to do one thing
if all of them are in fact mandatory as of now.
This change has the overall effect of having the evaluator properly
separate out top level code from any defined functions. See the new
test :EVAL-TOP-LEVEL-CODE-SEPARATE-COMPONENT and the existing tests
:STRUCTURE-SLOT-VALUE-IN-METHOD-NO-CALLEES and
:STRUCTURE-MISSING-SLOT-VALUE-IN-METHOD-SINGLE-CALLEE. As a bonus,
self call recognition works for COMPILE now. As a side effect,
we (correctly) get more redefinition warnings as well (for example for
WARN during bootstrap), which were missing before for some reason.
In order to achieve this, the following needs to be done:
* COMPILE does not have its own special way of doing ir1 conversion
for lambdas anymore, instead relying on the top level lambda machinery
to separate out the actual compiled code. This gives the correct
semantics for policy and compilation etc. for free without special or
duplicate logic, and removes all the stuff around
HAS-EXTERNAL-REFERENCES-P, which was an idea that was never really
finished. This allows us to extend COMPILE-IN-LEXENV to evaluate forms
directly and hence allow %SIMPLE-EVAL to not have to manually wrap the
lambdas either, so evaluation-via-compilation behaves like top-level
evaluation by the file compiler (of course without file compilation
semantics). LTV processing and fasteval also can just evaluate forms
directly with EVAL-WITH-COMPILE-IN-LEXENV.
* Storing source forms should always be off for PCL generated
functions, revealed by the clos-arena test. This was probably not
detected before because the policy inheritance was subtly wrong.
* Fix some tests to not rely on questionable behavior of how named
COMPILE interacts with global type information. Such behavior should
only work in compile file mode.
* The TL-XEP debug name had nothing to do with the actual TL-XEP
functional kind, which was confusing.
* TIME now counts all lambda conversions. Just ignoring the entry
points produced by COMPILE for the top level lambda didn't make much
sense since all other kinds of entry points get counted. It's there
purely to give a hint to the user how much work the compiler is doing,
not to count anything in the source being compiled.
* Core components need to have patch tables again. The fact that they
are exercised in warm bootstrap now means that we are likely actually
using the load time code stripping functionality not just in theory,
vut in practice as well.
* Add post ir1-toplevel processing smashing of names onto compiled
functions again to get names right. As an added bonus, this allows us
to get recognize self calls working for COMPILE functions.
* core debug source infos never actually used the function slot, and
it's not clear what function to give it when evaluating top level
forms, so delete it.
* Also fix > 3 decade old bug in MAKE-LISP-SOURCE-INFO in which a
plain vector was used when one with a fill pointer was expected. I
guess this code path hadn't been exercised in a long time.
Add a new function, make-numeric-type, which looks like
(make-numeric-type 'integer low high), not needing to construct a
list as for (specifier-type `(integer ,low ,high))
