1 ;;;; function call for the x86 VM
3 ;;;; This software is part of the SBCL system. See the README file for
6 ;;;; This software is derived from the CMU CL system, which was
7 ;;;; written at Carnegie Mellon University and released into the
8 ;;;; public domain. The software is in the public domain and is
9 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
10 ;;;; files for more information.
14 (defconstant arg-count-sc
(make-sc+offset any-reg-sc-number rcx-offset
))
15 (defconstant closure-sc
(make-sc+offset any-reg-sc-number rax-offset
))
17 (defconstant return-pc-passing-offset
18 (make-sc+offset sap-stack-sc-number return-pc-save-offset
))
20 (defconstant old-fp-passing-offset
21 (make-sc+offset control-stack-sc-number ocfp-save-offset
))
23 (defun compute-linkage-cell (node name res
)
24 (cond ((sb-c::code-immobile-p node
)
25 (inst lea res
(rip-relative-ea (make-fixup name
:linkage-cell
))))
27 (inst mov res
(thread-slot-ea sb-vm
::thread-linkage-table-slot
))
28 (inst lea res
(ea (make-fixup name
:linkage-cell
) res
)))))
30 ;;; Make the TNs used to hold OLD-FP and RETURN-PC within the current
31 ;;; function. We treat these specially so that the debugger can find
32 ;;; them at a known location.
34 ;;; Without using a save-tn - which does not make much sense if it is
35 ;;; wired to the stack?
36 (defun make-old-fp-save-location ()
37 (let ((tn (make-wired-tn *fixnum-primitive-type
*
38 control-stack-sc-number
40 (setf (tn-kind tn
) :environment
)
42 (defun make-return-pc-save-location ()
43 (let ((tn (make-wired-tn (primitive-type-or-lose 'system-area-pointer
)
44 sap-stack-sc-number return-pc-save-offset
)))
45 (setf (tn-kind tn
) :environment
)
48 ;;; Make a TN for the standard argument count passing location. We only
49 ;;; need to make the standard location, since a count is never passed when we
50 ;;; are using non-standard conventions.
51 (defun make-arg-count-location ()
52 (make-wired-tn *fixnum-primitive-type
* any-reg-sc-number rcx-offset
))
56 ;;; This is used for setting up the Old-FP in local call.
57 (define-vop (current-fp)
58 (:results
(val :scs
(any-reg control-stack
)))
62 ;;; We don't have a separate NFP, so we don't need to do anything here.
63 (define-vop (compute-old-nfp)
69 ;;; Accessing a slot from an earlier stack frame is definite hackery.
70 (define-vop (ancestor-frame-ref)
71 (:args
(frame-pointer :scs
(descriptor-reg))
72 (variable-home-tn :load-if nil
))
73 (:results
(value :scs
(descriptor-reg any-reg
)))
76 (aver (sc-is variable-home-tn control-stack
))
77 (loadw value frame-pointer
78 (frame-word-offset (tn-offset variable-home-tn
)))))
79 (define-vop (ancestor-frame-set)
80 (:args
(frame-pointer :scs
(descriptor-reg))
81 (value :scs
(descriptor-reg any-reg
)))
82 (:results
(variable-home-tn :load-if nil
))
85 (aver (sc-is variable-home-tn control-stack
))
86 (storew value frame-pointer
87 (frame-word-offset (tn-offset variable-home-tn
)))))
89 (macrolet ((define-frame-op
90 (suffix sc stack-sc instruction
91 &optional
(ea `(ea (frame-byte-offset (tn-offset variable-home-tn
))
93 (let ((reffer (symbolicate 'ancestor-frame-ref
'/ suffix
))
94 (setter (symbolicate 'ancestor-frame-set
'/ suffix
)))
96 (define-vop (,reffer ancestor-frame-ref
)
97 (:results
(value :scs
(,sc
)))
99 (aver (sc-is variable-home-tn
,stack-sc
))
100 (inst ,instruction value
102 (define-vop (,setter ancestor-frame-set
)
103 (:args
(frame-pointer :scs
(descriptor-reg))
106 (aver (sc-is variable-home-tn
,stack-sc
))
107 (inst ,instruction
,ea value
)))))))
108 (define-frame-op double-float double-reg double-stack movsd
)
109 (define-frame-op single-float single-reg single-stack movss
)
110 (define-frame-op complex-double-float complex-double-reg complex-double-stack
111 movupd
(ea-for-cdf-data-stack variable-home-tn frame-pointer
))
112 (define-frame-op complex-single-float complex-single-reg complex-single-stack
113 movq
(ea-for-csf-data-stack variable-home-tn frame-pointer
))
114 (define-frame-op signed-byte-64 signed-reg signed-stack mov
)
115 (define-frame-op unsigned-byte-64 unsigned-reg unsigned-stack mov
)
116 (define-frame-op system-area-pointer sap-reg sap-stack mov
))
118 (defun primitive-type-indirect-cell-type (ptype)
119 (declare (type primitive-type ptype
))
120 (macrolet ((foo (&body data
)
121 `(case (primitive-type-name ptype
)
122 ,@(loop for
(name stack-sc ref set
) in data
126 (list (primitive-type-or-lose ',name
)
127 (sc-or-lose ',stack-sc
)
128 (lambda (node block fp value res
)
129 (sb-c::vop
,ref node block
131 (lambda (node block fp new-val value
)
132 (sb-c::vop
,set node block
133 fp new-val value
)))))))))
134 (foo (double-float double-stack
135 ancestor-frame-ref
/double-float
136 ancestor-frame-set
/double-float
)
137 (single-float single-stack
138 ancestor-frame-ref
/single-float
139 ancestor-frame-set
/single-float
)
140 (complex-double-float complex-double-stack
141 ancestor-frame-ref
/complex-double-float
142 ancestor-frame-set
/complex-double-float
)
143 (complex-single-float complex-single-stack
144 ancestor-frame-ref
/complex-single-float
145 ancestor-frame-set
/complex-single-float
)
146 (signed-byte-64 signed-stack
147 ancestor-frame-ref
/signed-byte-64
148 ancestor-frame-set
/signed-byte-64
)
149 (unsigned-byte-64 unsigned-stack
150 ancestor-frame-ref
/unsigned-byte-64
151 ancestor-frame-set
/unsigned-byte-64
)
152 (unsigned-byte-63 unsigned-stack
153 ancestor-frame-ref
/unsigned-byte-64
154 ancestor-frame-set
/unsigned-byte-64
)
155 (system-area-pointer sap-stack
156 ancestor-frame-ref
/system-area-pointer
157 ancestor-frame-set
/system-area-pointer
))))
159 (define-vop (xep-allocate-frame)
163 (shiftf (sb-assem::asmstream-inter-function-padding sb-assem
:*asmstream
*)
165 (emit-alignment n-lowtag-bits
(if (eq nop-kind
:nop
) #x90
0)))
166 (emit-label start-lab
)
167 ;; Skip space for the function header.
168 (inst simple-fun-header-word
)
169 (inst .skip
(* (1- simple-fun-insts-offset
) n-word-bytes
))
170 ;; The start of the actual code.
171 ;; Save the return-pc.
172 (popw rbp-tn
(frame-word-offset return-pc-save-offset
))))
174 (defun emit-lea (target source disp
)
176 (inst mov target source
)
177 (inst lea target
(ea disp source
))))
179 (define-vop (xep-setup-sp)
181 (emit-lea rsp-tn rbp-tn
(- (* n-word-bytes
182 (- (sb-allocated-size 'stack
)
185 ;;; This is emitted directly before either a known-call-local, call-local,
186 ;;; or a multiple-call-local. All it does is allocate stack space for the
187 ;;; callee (who has the same size stack as us).
188 (define-vop (allocate-frame)
189 (:results
(res :scs
(any-reg))
194 (inst lea res
(ea (- (* sp-
>fp-offset n-word-bytes
)) rsp-tn
))
195 (inst sub rsp-tn
(* n-word-bytes
(sb-allocated-size 'stack
)))))
197 (defun make-stack-pointer-tn (&optional nargs
)
198 ;; Avoid using a temporary register if the new frame pointer will be
199 ;; at the same location as the new stack pointer
201 (= (* sp-
>fp-offset n-word-bytes
)
202 (* (max (if (> nargs register-arg-count
)
205 (sb-c::sb-size
(sb-or-lose 'stack
)))
207 (make-wired-tn *fixnum-primitive-type
* any-reg-sc-number rsp-offset
)
208 (make-normal-tn *fixnum-primitive-type
*)))
210 ;;; Allocate a partial frame for passing stack arguments in a full
211 ;;; call. NARGS is the number of arguments passed. We allocate at
212 ;;; least 2 slots, because the XEP noise is going to want to use them
213 ;;; before it can extend the stack.
214 (define-vop (allocate-full-call-frame)
216 (:results
(res :scs
(any-reg)))
218 (let ((fp-offset (* sp-
>fp-offset n-word-bytes
))
219 (stack-size (* (max (if (> nargs register-arg-count
)
222 (sb-c::sb-size
(sb-or-lose 'stack
)))
224 (cond ((= fp-offset stack-size
)
225 (inst sub rsp-tn stack-size
)
228 (inst lea res
(ea (- fp-offset
) rsp-tn
))
229 (inst sub rsp-tn stack-size
))))))
231 ;;; Emit code needed at the return-point from an unknown-values call
232 ;;; for a fixed number of values. Values is the head of the TN-REF
233 ;;; list for the locations that the values are to be received into.
234 ;;; Nvals is the number of values that are to be received (should
235 ;;; equal the length of Values).
237 ;;; If 0 or 1 values are expected, then we just emit an instruction to
238 ;;; reset the SP (which will only be executed when other than 1 value
241 ;;; In the general case we have to do three things:
242 ;;; -- Default unsupplied register values. This need only be done
243 ;;; when a single value is returned, since register values are
244 ;;; defaulted by the called in the non-single case.
245 ;;; -- Default unsupplied stack values. This needs to be done whenever
246 ;;; there are stack values.
247 ;;; -- Reset SP. This must be done whenever other than 1 value is
248 ;;; returned, regardless of the number of values desired.
249 (defun default-unknown-values (vop values nvals node rbx move-temp
)
250 (declare (type (or tn-ref null
) values
)
251 (type unsigned-byte nvals
))
252 (multiple-value-bind (type name leaf
) (sb-c::lvar-fun-type
(sb-c::basic-combination-fun node
))
253 (let* ((verify (and leaf
254 (policy node
(and (>= safety
1)
256 (memq (sb-c::leaf-where-from leaf
) '(:declared-verify
:defined-here
))))
258 (if (fun-type-p type
)
259 (fun-type-returns type
)
261 (sb-c::node-derived-type node
)))
262 (min-values (values-type-min-value-count type
))
263 (max-values (values-type-max-value-count type
))
264 (trust (or (and (= min-values
0)
265 (= max-values call-arguments-limit
))
267 (flet ((check-nargs ()
269 (let* ((*location-context
* (list* name
270 (type-specifier type
)
271 (make-restart-location SKIP
)))
272 (err-lab (generate-error-code vop
'invalid-arg-count-error
))
274 (max (and (< max-values call-arguments-limit
)
278 (inst test
:dword rcx-tn rcx-tn
)
279 (inst cmp
:dword rcx-tn
(fixnumize max
)))
280 (inst jmp
:ne err-lab
))
282 (let ((nargs move-temp
))
285 (inst lea
:dword move-temp
(ea (fixnumize (- min
)) rcx-tn
)))
286 (inst cmp
:dword nargs
(fixnumize (- max min
)))
287 (inst jmp
:a err-lab
)))
290 (inst test
:dword rcx-tn rcx-tn
)
291 (inst jmp
:e err-lab
))
293 (inst cmp
:dword rcx-tn
(fixnumize min
))
294 (inst jmp
:b err-lab
))))))
298 (note-this-location vop
:single-value-return
)
301 (<= (sb-kernel:values-type-max-value-count type
)
302 register-arg-count
)))
304 (not (sb-kernel:values-type-may-be-single-value-p type
)))
305 (inst mov rsp-tn rbx
))
307 (inst cmov
:c rsp-tn rbx
)
309 (inst mov move-temp
(fixnumize 1))
310 (inst cmov
:nc rcx-tn move-temp
)
312 ((<= nvals register-arg-count
)
313 (note-this-location vop
:unknown-return
)
314 (when (or (not trust
)
315 (sb-kernel:values-type-may-be-single-value-p type
))
317 (inst jmp
:c regs-defaulted
)
318 ;; Default the unsupplied registers.
319 (let* ((2nd-tn-ref (tn-ref-across values
))
320 (2nd-tn (tn-ref-tn 2nd-tn-ref
))
321 (2nd-tn-live (neq (tn-kind 2nd-tn
) :unused
)))
323 (inst mov
2nd-tn nil-value
))
326 for tn-ref
= (tn-ref-across 2nd-tn-ref
)
327 then
(tn-ref-across tn-ref
)
328 for count from
2 below register-arg-count
329 unless
(eq (tn-kind (tn-ref-tn tn-ref
)) :unused
)
331 (inst mov
:dword
(tn-ref-tn tn-ref
)
332 (if 2nd-tn-live
2nd-tn nil-value
)))))
333 (inst mov rbx rsp-tn
)
336 (when (or (not trust
)
337 (< register-arg-count
338 (sb-kernel:values-type-max-value-count type
)))
339 (inst mov rsp-tn rbx
))
341 (inst mov move-temp
(fixnumize 1))
342 (inst cmov
:nc rcx-tn move-temp
)
345 (collect ((defaults))
346 (let ((default-stack-slots (gen-label))
348 (loop for i from
1 below register-arg-count
349 for tn
= (tn-ref-tn (setf values
(tn-ref-across values
)))
350 unless
(eq (tn-kind tn
) :unused
)
352 finally
(setf values
(tn-ref-across values
))))
354 (loop for ref
= values then
(tn-ref-across ref
)
356 thereis
(neq (tn-kind (tn-ref-tn ref
)) :unused
))))
358 (note-this-location vop
:unknown-return
)
360 (inst mov move-temp
(fixnumize 1))
361 (inst cmov
:nc rcx-tn move-temp
))
362 ;; If it returned exactly one value the registers and the
363 ;; stack slots need to be filled with NIL.
367 (inst jmp
:nc default-stack-slots
))
369 (inst jmp
:c regs-defaulted
)
370 (loop for null
= nil-value then
(car used-registers
)
371 for reg in used-registers
372 do
(inst mov
:dword reg null
))
375 (do ((i register-arg-count
(1+ i
))
376 (val values
(tn-ref-across val
)))
378 (let ((tn (tn-ref-tn val
)))
379 (unless (eq (tn-kind tn
) :unused
)
380 (when (or (not trust
)
382 (let ((default-lab (gen-label)))
383 (defaults (cons default-lab tn
))
384 ;; Note that the max number of values received
385 ;; is assumed to fit in a :dword register.
386 (inst cmp
:dword rcx-tn
(fixnumize i
))
387 (inst jmp
:be default-lab
)))
390 (loadw move-temp rbx
(frame-word-offset (+ sp-
>fp-offset i
)))
391 (inst mov tn move-temp
))
393 (loadw tn rbx
(frame-word-offset (+ sp-
>fp-offset i
))))))))
399 (let ((defaults (defaults)))
401 (assemble (:elsewhere
)
402 (when (or (not trust
)
404 (emit-label default-stack-slots
)
405 (loop for null
= nil-value then
(car used-registers
)
406 for reg in used-registers
407 do
(inst mov
:dword reg null
))
409 (dolist (default defaults
)
410 (emit-label (car default
))
411 (inst mov
(cdr default
) nil-value
))
412 (inst jmp defaulting-done
)))))))))))))
414 ;;;; unknown values receiving
416 ;;; Emit code needed at the return point for an unknown-values call
417 ;;; for an arbitrary number of values.
419 ;;; We do the single and non-single cases with no shared code: there
420 ;;; doesn't seem to be any potential overlap, and receiving a single
421 ;;; value is more important efficiency-wise.
423 ;;; When there is a single value, we just push it on the stack,
424 ;;; returning the old SP and 1.
426 ;;; When there is a variable number of values, we move all of the
427 ;;; argument registers onto the stack, and return ARGS and NARGS.
429 ;;; ARGS and NARGS are TNs wired to the named locations. We must
430 ;;; explicitly allocate these TNs, since their lifetimes overlap with
431 ;;; the results start and count. (Also, it's nice to be able to target
433 (defun receive-unknown-values (args nargs start count node
)
434 (declare (type tn args nargs start count
))
435 (let ((type (sb-c::basic-combination-derived-type node
))
436 (variable-values (gen-label))
437 (stack-values (gen-label))
439 (unused-count-p (eq (tn-kind count
) :unused
)))
440 (when (sb-kernel:values-type-may-be-single-value-p type
)
441 (inst jmp
:c variable-values
)
442 (cond ((eq (tn-kind start
) :unused
)
443 (inst push
(first *register-arg-tns
*)))
444 ((location= start
(first *register-arg-tns
*))
445 (inst push
(first *register-arg-tns
*))
446 (inst lea start
(ea n-word-bytes rsp-tn
)))
447 (t (inst mov start rsp-tn
)
448 (inst push
(first *register-arg-tns
*))))
449 (unless unused-count-p
450 (inst mov count
(fixnumize 1)))
452 (emit-label variable-values
))
453 ;; The stack frame is burnt and RETurned from if there are no
454 ;; stack values. In this case quickly reallocate sufficient space.
455 (when (<= (sb-kernel:values-type-min-value-count type
)
457 (inst cmp nargs
(fixnumize register-arg-count
))
458 (inst jmp
:g stack-values
)
459 #+#.
(cl:if
(cl:= sb-vm
:word-shift sb-vm
:n-fixnum-tag-bits
) '(and) '(or))
460 (inst sub rsp-tn nargs
)
461 #-
#.
(cl:if
(cl:= sb-vm
:word-shift sb-vm
:n-fixnum-tag-bits
) '(and) '(or))
463 (unless unused-count-p
464 (inst mov
:dword
(setf sub rax-tn
) nargs
))
465 (inst shl
:dword sub
(- word-shift n-fixnum-tag-bits
))
466 (inst sub rsp-tn sub
))
467 (emit-label stack-values
))
468 ;; dtc: this writes the registers onto the stack even if they are
469 ;; not needed, only the number specified in rcx are used and have
470 ;; stack allocated to them. No harm is done.
472 for arg in
*register-arg-tns
*
474 for j below
(sb-kernel:values-type-max-value-count type
)
475 do
(storew arg args i
))
476 (unless (eq (tn-kind start
) :unused
)
478 (unless unused-count-p
484 ;;; VOP that can be inherited by unknown values receivers. The main thing this
485 ;;; handles is allocation of the result temporaries.
486 (define-vop (unknown-values-receiver)
487 (:temporary
(:sc descriptor-reg
:offset rbx-offset
488 :from
:eval
:to
(:result
0))
490 (:temporary
(:sc any-reg
:offset rcx-offset
491 :from
:eval
:to
(:result
1))
493 (:results
(start :scs
(any-reg control-stack
))
494 (count :scs
(any-reg control-stack
))))
496 ;;;; local call with unknown values convention return
498 (defun check-ocfp-and-return-pc (old-fp return-pc
)
500 (format t
"*known-return: old-fp ~S, tn-kind ~S; ~S ~S~%"
501 old-fp
(tn-kind old-fp
) (sb-c::tn-save-tn old-fp
)
502 (tn-kind (sb-c::tn-save-tn old-fp
)))
504 (format t
"*known-return: return-pc ~S, tn-kind ~S; ~S ~S~%"
505 return-pc
(tn-kind return-pc
)
506 (sb-c::tn-save-tn return-pc
)
507 (tn-kind (sb-c::tn-save-tn return-pc
)))
508 (unless (and (sc-is old-fp control-stack
)
509 (= (tn-offset old-fp
) ocfp-save-offset
))
510 (error "ocfp not on stack in standard save location?"))
511 (unless (and (sc-is return-pc sap-stack
)
512 (= (tn-offset return-pc
) return-pc-save-offset
))
513 (error "return-pc not on stack in standard save location?")))
515 ;;; The local call convention doesn't fit that well with x86-style
516 ;;; calls. Emit a header for local calls to pop the return address
517 ;;; in the right place.
518 (defun emit-block-header (start-label trampoline-label fall-thru-p alignp
)
519 (when (and fall-thru-p trampoline-label
)
520 (inst jmp start-label
))
521 (when trampoline-label
522 (emit-label trampoline-label
)
523 (popw rbp-tn
(frame-word-offset return-pc-save-offset
)))
525 (emit-alignment n-lowtag-bits alignp
))
526 (emit-label start-label
))
528 ;;; Non-TR local call for a fixed number of values passed according to
529 ;;; the unknown values convention.
531 ;;; FP is the frame pointer in install before doing the call.
533 ;;; NFP would be the number-stack frame pointer if we had a separate
536 ;;; Args are the argument passing locations, which are specified only
537 ;;; to terminate their lifetimes in the caller.
539 ;;; VALUES are the return value locations (wired to the standard
540 ;;; passing locations). NVALS is the number of values received.
542 ;;; Save is the save info, which we can ignore since saving has been
545 ;;; TARGET is a continuation pointing to the start of the called
547 (define-vop (call-local)
551 (:results
(values :more t
))
553 (:move-args
:local-call
)
554 (:info arg-locs callee target nvals
)
556 (:ignore nfp arg-locs args callee
)
558 (:temporary
(:sc any-reg
) move-temp
)
561 (note-this-location vop
:call-site
)
563 (default-unknown-values vop values nvals node rbx-tn move-temp
)))
565 ;;; Non-TR local call for a variable number of return values passed according
566 ;;; to the unknown values convention. The results are the start of the values
567 ;;; glob and the number of values received.
568 (define-vop (multiple-call-local unknown-values-receiver
)
573 (:move-args
:local-call
)
574 (:info save callee target
)
575 (:ignore args save nfp callee
)
580 (note-this-location vop
:call-site
)
582 (note-this-location vop
:unknown-return
)
583 (receive-unknown-values values-start nvals start count node
)))
585 ;;;; local call with known values return
587 ;;; Non-TR local call with known return locations. Known-value return
588 ;;; works just like argument passing in local call.
590 ;;; Note: we can't use normal load-tn allocation for the fixed args,
591 ;;; since all registers may be tied up by the more operand. Instead,
592 ;;; we use MAYBE-LOAD-STACK-TN.
593 (define-vop (known-call-local)
597 (:results
(res :more t
))
598 (:move-args
:local-call
)
600 (:info save callee target
)
601 (:ignore args res save nfp callee
)
605 (note-this-location vop
:call-site
)
607 (note-this-location vop
:known-return
)))
609 ;;; From Douglas Crosher
610 ;;; Return from known values call. We receive the return locations as
611 ;;; arguments to terminate their lifetimes in the returning function. We
612 ;;; restore FP and CSP and jump to the Return-PC.
613 (define-vop (known-return)
617 (:move-args
:known-return
)
619 (:ignore val-locs vals
)
622 (check-ocfp-and-return-pc old-fp return-pc
)
623 ;; Zot all of the stack except for the old-fp and return-pc.
629 ;;; There is something of a cross-product effect with full calls.
630 ;;; Different versions are used depending on whether we know the
631 ;;; number of arguments or the name of the called function, and
632 ;;; whether we want fixed values, unknown values, or a tail call.
634 ;;; In full call, the arguments are passed creating a partial frame on
635 ;;; the stack top and storing stack arguments into that frame. On
636 ;;; entry to the callee, this partial frame is pointed to by FP.
638 ;;; This macro helps in the definition of full call VOPs by avoiding
639 ;;; code replication in defining the cross-product VOPs.
641 ;;; NAME is the name of the VOP to define.
643 ;;; NAMED is true if the first argument is an fdefinition object whose
644 ;;; definition is to be called.
646 ;;; RETURN is either :FIXED, :UNKNOWN or :TAIL:
647 ;;; -- If :FIXED, then the call is for a fixed number of values, returned in
648 ;;; the standard passing locations (passed as result operands).
649 ;;; -- If :UNKNOWN, then the result values are pushed on the stack, and the
650 ;;; result values are specified by the Start and Count as in the
651 ;;; unknown-values continuation representation.
652 ;;; -- If :TAIL, then do a tail-recursive call. No values are returned.
653 ;;; The Old-Fp and Return-PC are passed as the second and third arguments.
655 ;;; In non-tail calls, the pointer to the stack arguments is passed as
656 ;;; the last fixed argument. If Variable is false, then the passing
657 ;;; locations are passed as a more arg. Variable is true if there are
658 ;;; a variable number of arguments passed on the stack. Variable
659 ;;; cannot be specified with :TAIL return. TR variable argument call
660 ;;; is implemented separately.
662 ;;; In tail call with fixed arguments, the passing locations are
663 ;;; passed as a more arg, but there is no new-FP, since the arguments
664 ;;; have been set up in the current frame.
665 (defmacro define-full-call
(vop-name named return variable
&optional args
)
666 (aver (not (and variable
(eq return
:tail
))))
667 `(define-vop (,vop-name
,@(when (eq return
:unknown
) '(unknown-values-receiver)))
668 (:args
,@(unless (eq return
:tail
)
669 '((new-fp :scs
(any-reg) :to
(:argument
1))))
671 ,@(unless named
; FUN is an info argument for named call
672 '((fun :scs
(descriptor-reg control-stack
)
673 :target rax
:to
(:argument
0))))
675 ,@(when (eq return
:tail
)
680 `((args :more t
,@(unless (eq args
:fixed
)
681 '(:scs
(descriptor-reg control-stack
)))))))
683 ,@(when (memq return
'(:fixed
:unboxed
)) '((:results
(values :more t
))))
685 (:save-p
,(if (eq return
:tail
) :compute-only t
))
687 ,@(unless (or (eq return
:tail
) variable
)
688 `((:move-args
,(if (eq args
:fixed
) :fixed
:full-call
))))
692 (:info
,@(unless (or variable
(eq return
:tail
)) '(arg-locs))
693 ,@(unless variable
'(nargs))
694 ;; Intuitively you might want FUN to be the first codegen arg,
695 ;; but that won't work, because EMIT-ARG-MOVES wants the
696 ;; passing locs in (FIRST (vop-codegen-info vop)).
697 ,@(when named
'(fun))
698 ,@(when (eq return
:fixed
) '(nvals))
700 ,@(unless named
'(fun-type)))
702 (:ignore
,@(unless (or variable
(eq return
:tail
)) '(arg-locs))
703 ,@(unless variable
'(args))
704 ,@(when (eq return
:unboxed
) '(values)))
706 ;; For anonymous call, RAX is the function. For named call, RAX will be the linkage
707 ;; table base if not stepping, or the linkage cell itself if stepping.
708 ;; Calls from immobile-space without stepping avoid using RAX, and instead
709 ;; access the linkage table relative to RIP.
710 (:temporary
(:sc descriptor-reg
:offset rax-offset
:from
(:argument
0) :to
:eval
) rax
)
712 ;; We pass the number of arguments in RCX.
714 (:sc unsigned-reg
:offset rcx-offset
:to
,(if (eq return
:fixed
) :save
:eval
))
717 ,@(when (eq return
:fixed
)
718 ;; Save it for DEFAULT-UNKNOWN-VALUES to work
719 `((:temporary
(:sc unsigned-reg
:offset rbx-offset
:from
:result
) rbx
)
720 (:temporary
(:sc any-reg
) move-temp
)))
722 ;; With variable call, we have to load the
723 ;; register-args out of the (new) stack frame before
724 ;; doing the call. Therefore, we have to tell the
725 ;; lifetime stuff that we need to use them.
727 (mapcar (lambda (name offset
)
728 `(:temporary
(:sc descriptor-reg
733 *register-arg-names
* *register-arg-offsets
*))
735 ,@(when (eq return
:tail
)
736 '((:temporary
(:sc unsigned-reg
:from
(:argument
1) :to
(:argument
2))
738 ,@(unless (eq return
:tail
) '((:node-var node
)))
740 (:generator
,(+ (if named
5 0)
742 (if (eq return
:tail
) 0 10)
744 (if (eq return
:unknown
) 25 0))
746 (progn node
) ; always "use" it
748 ;; This has to be done before the frame pointer is
749 ;; changed! RAX stores the 'lexical environment' needed
751 ,@(unless named
'((move rax fun
)))
754 ;; For variable call, compute the number of
755 ;; arguments and move some of the arguments to
757 `((inst mov rcx new-fp
)
758 (inst sub rcx rsp-tn
)
759 (inst shr rcx
,(- word-shift n-fixnum-tag-bits
))
760 ;; Move the necessary args to registers,
761 ;; this moves them all even if they are
763 ,@(loop for name in
*register-arg-names
*
764 for index downfrom -
1
765 collect
`(loadw ,name new-fp
,index
)))
766 '((cond ((listp nargs
)) ;; no-verify-arg-count
770 (inst mov rcx
(fixnumize nargs
))))))
771 ,@(cond ((eq return
:tail
)
772 '( ;; Python has figured out what frame we should
773 ;; return to so might as well use that clue.
774 ;; This seems really important to the
775 ;; implementation of things like
776 ;; (without-interrupts ...)
778 ;; dtc; Could be doing a tail call from a
779 ;; known-call-local etc in which the old-fp
780 ;; or ret-pc are in regs or in non-standard
781 ;; places. If the passing location were
782 ;; wired to the stack in standard locations
783 ;; then these moves will be un-necessary;
784 ;; this is probably best for the x86.
787 (unless (= ocfp-save-offset
(tn-offset old-fp
))
788 ;; FIXME: FORMAT T for stale
789 ;; diagnostic output (several of
790 ;; them around here), ick
791 (error "** tail-call old-fp not S0~%")
792 (move old-fp-tmp old-fp
)
793 (storew old-fp-tmp rbp-tn
(frame-word-offset ocfp-save-offset
))))
794 ((any-reg descriptor-reg
)
795 (error "** tail-call old-fp in reg not S0~%")
796 (storew old-fp rbp-tn
(frame-word-offset ocfp-save-offset
))))
798 ;; For tail call, we have to push the
799 ;; return-pc so that it looks like we CALLed
800 ;; despite the fact that we are going to JMP.
801 (inst push return-pc
)))
803 ;; For non-tail call, we have to save our
804 ;; frame pointer and install the new frame
805 ;; pointer. We can't load stack tns after this
807 `( ;; Python doesn't seem to allocate a frame
808 ;; here which doesn't leave room for the
811 ;; The variable args are on the stack and
812 ;; become the frame, but there may be <3
813 ;; args and 3 stack slots are assumed
814 ;; allocate on the call. So need to ensure
815 ;; there are at least 3 slots. This hack
818 '(inst sub rsp-tn
(* 3 n-word-bytes
)))
820 ;; Bias the new-fp for use as an fp
822 '(inst sub new-fp
(* sp-
>fp-offset n-word-bytes
)))
825 (storew rbp-tn new-fp
(frame-word-offset ocfp-save-offset
))
826 (move rbp-tn new-fp
)))) ; NB - now on new stack frame.
828 (when step-instrumenting
829 ,@(when named
'((compute-linkage-cell node fun rax
)))
830 (emit-single-step-test)
832 (inst break single-step-around-trap
))
834 (note-this-location vop
:call-site
)
836 `(emit-direct-call fun
',(if (eq return
:tail
) 'jmp
'call
)
837 node step-instrumenting
))
839 `(tail-call-unnamed rax fun-type vop
))
841 `(call-unnamed rax fun-type vop
)))
843 (:fixed
'((default-unknown-values vop values nvals node rbx move-temp
)))
845 '((note-this-location vop
:unknown-return
)
846 (receive-unknown-values values-start nvals start count node
)))
847 ((:tail
:unboxed
))))))
849 (define-full-call call nil
:fixed nil
)
850 (define-full-call call-named t
:fixed nil
)
851 (define-full-call multiple-call nil
:unknown nil
)
852 (define-full-call multiple-call-named t
:unknown nil
)
853 (define-full-call tail-call nil
:tail nil
)
854 (define-full-call tail-call-named t
:tail nil
)
856 (define-full-call call-variable nil
:fixed t
)
857 (define-full-call multiple-call-variable nil
:unknown t
)
858 (define-full-call fixed-call-named t
:fixed nil
:fixed
)
859 (define-full-call fixed-tail-call-named t
:tail nil
:fixed
)
861 (define-full-call unboxed-call-named t
:unboxed nil
)
862 (define-full-call fixed-unboxed-call-named t
:unboxed nil
:fixed
)
864 ;;; Call NAME "directly" meaning in a single JMP or CALL instruction,
865 ;;; if possible (without loading RAX)
866 (defun emit-direct-call (name instruction node step-instrumenting
)
867 (cond (step-instrumenting
868 ;; If step-instrumenting, then RAX points to the linkage table cell
869 (inst* instruction
(ea rax-tn
)))
870 ((sb-c::code-immobile-p node
)
871 (inst* instruction
(rip-relative-ea (make-fixup name
:linkage-cell
))))
873 ;; get the linkage table base into RAX
874 (inst mov rax-tn
(thread-slot-ea sb-vm
::thread-linkage-table-slot
))
875 (inst* instruction
(ea (make-fixup name
:linkage-cell
) rax-tn
)))))
877 ;;; Invoke the function-designator FUN.
878 (defun tail-call-unnamed (fun type vop
)
879 (let ((relative-call (sb-c::code-immobile-p vop
))
880 (fun-ea (ea (- (* closure-fun-slot n-word-bytes
) fun-pointer-lowtag
)
885 (%lea-for-lowtag-test rbx-tn fun fun-pointer-lowtag
)
886 (inst test
:byte rbx-tn lowtag-mask
)
887 (inst jmp
:nz
(if relative-call
888 (make-fixup 'call-symbol
:assembly-routine
)
892 (unless relative-call
893 (invoke-asm-routine 'jmp
'call-symbol vop
))))
895 (invoke-asm-routine 'jmp
'call-symbol vop
))
897 (inst jmp fun-ea
)))))
899 (defun call-unnamed (fun type vop
)
902 (invoke-asm-routine 'call
'call-symbol vop
))
905 (when (eq type
:designator
)
906 (%lea-for-lowtag-test rbx-tn fun fun-pointer-lowtag
)
907 (inst test
:byte rbx-tn lowtag-mask
)
909 (invoke-asm-routine 'call
'call-symbol vop
)
912 (inst call
(ea (- (* closure-fun-slot n-word-bytes
) fun-pointer-lowtag
)
916 ;;; This is defined separately, since it needs special code that BLT's
917 ;;; the arguments down. All the real work is done in the assembly
918 ;;; routine. We just set things up so that it can find what it needs.
919 (define-vop (tail-call-variable)
920 (:args
(args :scs
(any-reg control-stack
) :target rsi
)
921 (function :scs
(descriptor-reg control-stack
) :target rax
)
925 (:temporary
(:sc unsigned-reg
:offset rsi-offset
:from
(:argument
0)) rsi
)
926 (:temporary
(:sc unsigned-reg
:offset rax-offset
:from
(:argument
1)) rax
)
929 (check-ocfp-and-return-pc old-fp return-pc
)
930 ;; Move these into the passing locations if they are not already there.
933 ;; And jump to the assembly routine.
934 (invoke-asm-routine 'jmp
(if (eq fun-type
:function
)
936 'tail-call-callable-variable
)
939 ;;;; unknown values return
941 ;;; Return a single-value using the Unknown-Values convention.
943 ;;; pfw--get wired-tn conflicts sometimes if register sc specd for args
944 ;;; having problems targeting args to regs -- using temps instead.
946 ;;; First off, modifying the return-pc defeats the branch-prediction
947 ;;; optimizations on modern CPUs quite handily. Second, we can do all
948 ;;; this without needing a temp register. Fixed the latter, at least.
949 ;;; -- AB 2006/Feb/04
950 (define-vop (return-single)
956 (check-ocfp-and-return-pc old-fp return-pc
)
957 ;; Drop stack above old-fp and restore old frame pointer
959 ;; Clear the multiple-value return flag
964 ;;; Do unknown-values return of a fixed (other than 1) number of
965 ;;; values. The VALUES are required to be set up in the standard
966 ;;; passing locations. NVALS is the number of values returned.
968 ;;; Basically, we just load RCX with the number of values returned and
969 ;;; RBX with a pointer to the values, set RSP to point to the end of
970 ;;; the values, and jump directly to return-pc.
973 (return-pc :to
(:eval
1))
977 ;; In the case of other than one value, we need these registers to
978 ;; tell the caller where they are and how many there are.
979 (:temporary
(:sc unsigned-reg
:offset rbx-offset
) rbx
)
980 (:temporary
(:sc unsigned-reg
:offset rcx-offset
) rcx
)
981 ;; We need to stretch the lifetime of return-pc past the argument
982 ;; registers so that we can default the argument registers without
983 ;; trashing return-pc.
984 (:temporary
(:sc unsigned-reg
:offset
(first *register-arg-offsets
*)
986 (:temporary
(:sc unsigned-reg
:offset
(second *register-arg-offsets
*)
988 (:temporary
(:sc unsigned-reg
:offset
(third *register-arg-offsets
*)
992 (check-ocfp-and-return-pc old-fp return-pc
)
994 ;; This is handled in RETURN-SINGLE.
995 (error "nvalues is 1"))
996 ;; Establish the values pointer and values count.
997 (inst lea rbx
(ea (* sp-
>fp-offset n-word-bytes
) rbp-tn
))
999 (zeroize rcx
) ; smaller
1000 (inst mov rcx
(fixnumize nvals
)))
1001 ;; Pre-default any argument register that need it.
1002 (when (< nvals register-arg-count
)
1003 (let* ((arg-tns (nthcdr nvals
(list a0 a1 a2
)))
1004 (first (first arg-tns
)))
1005 (inst mov first nil-value
)
1006 (dolist (tn (cdr arg-tns
))
1007 (inst mov tn first
))))
1008 ;; Set the multiple value return flag.
1010 ;; And away we go. Except that return-pc is still on the
1011 ;; stack and we've changed the stack pointer. So we have to
1012 ;; tell it to index off of RBX instead of RBP.
1013 (cond ((<= nvals register-arg-count
)
1017 ;; Some values are on the stack after RETURN-PC and OLD-FP,
1018 ;; can't return normally and some slots of the frame will
1019 ;; be used as temporaries by the receiver.
1021 ;; Clear as much of the stack as possible, but not past the
1022 ;; old frame address.
1024 (ea (frame-byte-offset (1- nvals
)) rbp-tn
))
1025 (move rbp-tn old-fp
)
1026 (inst push
(ea (frame-byte-offset
1027 (+ sp-
>fp-offset
(tn-offset return-pc
)))
1031 ;;; Do unknown-values return of an arbitrary number of values (passed
1032 ;;; on the stack.) We check for the common case of a single return
1033 ;;; value, and do that inline using the normal single value return
1034 ;;; convention. Otherwise, we branch off to code that calls an
1035 ;;; assembly-routine.
1037 ;;; The assembly routine takes the following args:
1038 ;;; RCX -- number of values to find there.
1039 ;;; RSI -- pointer to where to find the values.
1040 (define-vop (return-multiple)
1043 (vals :scs
(any-reg) :target rsi
)
1044 (nvals :scs
(any-reg) :target rcx
))
1045 (:temporary
(:sc unsigned-reg
:offset rsi-offset
:from
(:argument
2)) rsi
)
1046 (:temporary
(:sc unsigned-reg
:offset rcx-offset
:from
(:argument
3)) rcx
)
1047 (:temporary
(:sc descriptor-reg
:offset
(first *register-arg-offsets
*)
1048 :from
(:eval
0)) a0
)
1052 (check-ocfp-and-return-pc old-fp return-pc
)
1053 (unless (policy node
(> space speed
))
1054 ;; Check for the single case.
1055 (let ((not-single (gen-label)))
1056 (inst cmp nvals
(fixnumize 1))
1057 (inst jmp
:ne not-single
)
1058 ;; Return with one value.
1060 ;; Clear the stack until ocfp.
1062 ;; clear the multiple-value return flag
1066 ;; Nope, not the single case. Jump to the assembly routine.
1067 (emit-label not-single
)))
1070 (invoke-asm-routine 'jmp
'return-multiple vop
)))
1074 ;;; Get the lexical environment from its passing location.
1075 (define-vop (setup-closure-environment)
1076 (:results
(closure :scs
(descriptor-reg)))
1081 (move closure rax-tn
)))
1083 ;;; Copy a &MORE arg from the argument area to the end of the current
1084 ;;; frame. FIXED is the number of non-&MORE arguments.
1085 (define-vop (copy-more-arg)
1086 (:temporary
(:sc any-reg
:offset r8-offset
) copy-index
)
1087 (:temporary
(:sc any-reg
:offset r9-offset
) source
)
1088 (:temporary
(:sc descriptor-reg
:offset r10-offset
) temp
)
1089 (:info fixed min-verified
)
1091 ;; Avoid the copy if there are no more args.
1092 (cond ((zerop fixed
)
1093 (inst test
:dword rcx-tn rcx-tn
)
1094 (inst jmp
:z JUST-ALLOC-FRAME
))
1095 ((and (eql min-verified fixed
)
1097 ;; verify-arg-count will do a CMP
1098 (inst jmp
:e JUST-ALLOC-FRAME
))
1100 (inst cmp
:dword rcx-tn
(fixnumize fixed
))
1101 (inst jmp
:be JUST-ALLOC-FRAME
)))
1103 ;; Create a negated copy of the number of arguments to allow us to
1104 ;; use EA calculations in order to do scaled subtraction.
1105 (inst mov
:dword temp rcx-tn
)
1108 ;; Allocate the space on the stack.
1109 ;; stack = rbp + sp->fp-offset - frame-size - (nargs - fixed)
1110 ;; if we'd move SP backward, swap the meaning of rsp and source;
1111 ;; otherwise, we'd be accessing values below SP, and that's no good
1112 ;; if a signal interrupts this code sequence. In that case, store
1113 ;; the final value in rsp after the stack-stack memmove loop.
1114 (let* ((delta (- fixed
(sb-allocated-size 'stack
)))
1116 (fixnum->word
(ash 1 (- word-shift n-fixnum-tag-bits
)))
1117 (below (plusp delta
)))
1118 (inst lea
(if below source rsp-tn
)
1119 (ea (* n-word-bytes
(+ sp-
>fp-offset delta
))
1120 rbp-tn temp fixnum-
>word
))
1122 ;; Now: nargs>=1 && nargs>fixed
1124 (cond ((< fixed register-arg-count
)
1125 ;; the code above only moves the final value of rsp in
1126 ;; rsp directly if that condition is satisfied. Currently,
1127 ;; r-a-c is 3, so the aver is OK. If the calling convention
1128 ;; ever changes, the logic above with LEA will have to be
1130 (aver (<= fixed
(sb-allocated-size 'stack
)))
1131 ;; We must stop when we run out of stack args, not when we
1132 ;; run out of more args.
1133 ;; Number to copy = nargs-3
1134 ;; Save the original count of args.
1135 (inst mov rbx-tn rcx-tn
)
1136 (inst sub rbx-tn
(fixnumize register-arg-count
))
1137 ;; Everything of interest in registers.
1138 (inst jmp
:be DO-REGS
))
1140 ;; Number to copy = nargs-fixed
1141 (inst lea rbx-tn
(ea (- (fixnumize fixed
)) rcx-tn
))))
1143 ;; Initialize R8 to be the end of args.
1144 ;; Swap with SP if necessary to mirror the previous condition
1145 (unless (zerop delta
)
1146 (inst lea
(if below rsp-tn source
)
1147 (ea (* sp-
>fp-offset n-word-bytes
)
1148 rbp-tn temp fixnum-
>word
)))
1150 ;; src: rbp + temp + sp->fp
1151 ;; dst: rbp + temp + sp->fp + (fixed - [stack-size])
1152 (cond ((zerop delta
)) ; no-op move
1154 ;; dst is lower than src, copy forward
1155 (zeroize copy-index
)
1156 ;; We used to use REP MOVS here, but on modern x86 it performs
1157 ;; much worse than an explicit loop for small blocks.
1160 (inst mov temp
(ea source copy-index
))
1161 (inst mov
(ea rsp-tn copy-index
) temp
)
1162 (inst add copy-index n-word-bytes
)
1163 (inst sub rbx-tn
(fixnumize 1))
1164 (inst jmp
:nz loop
))
1166 ;; dst is higher than src; copy backward
1168 (inst sub rbx-tn
(fixnumize 1))
1169 (inst mov temp
(ea rsp-tn rbx-tn fixnum-
>word
))
1170 (inst mov
(ea source rbx-tn fixnum-
>word
) temp
)
1172 ;; done with the stack--stack copy. Reset RSP to its final
1174 (inst mov rsp-tn source
))))
1177 ;; Here: nargs>=1 && nargs>fixed
1178 (when (< fixed register-arg-count
)
1179 ;; Now we have to deposit any more args that showed up in
1183 ;; Store it relative to rbp
1184 (inst mov
(ea (* n-word-bytes
1186 (+ 1 (- i fixed
) (sb-allocated-size 'stack
))))
1188 (nth i
*register-arg-tns
*))
1191 (when (>= i register-arg-count
)
1194 ;; Don't deposit any more than there are.
1195 #.
(assert (= register-arg-count
3))
1197 (inst cmp
:dword rcx-tn
(fixnumize i
))
1198 (inst jmp
:eq DONE
))
1199 ;; Use a single comparison for 1 and 2
1201 (inst cmp
:dword rcx-tn
(fixnumize 2))
1204 (inst jmp
:eq DONE
)))))
1209 (emit-lea rsp-tn rbp-tn
1212 (sb-allocated-size 'stack
))))
1217 (:translate sb-c
::%more-kw-arg
)
1218 (:policy
:fast-safe
)
1219 (:args
(object :scs
(descriptor-reg) :to
(:result
1))
1220 (index :scs
(any-reg) :to
(:result
1) :target keyword
))
1221 (:arg-types
* tagged-num
)
1222 (:results
(value :scs
(descriptor-reg any-reg
))
1223 (keyword :scs
(descriptor-reg any-reg
)))
1226 (inst mov value
(ea object index
(ash 1 (- word-shift n-fixnum-tag-bits
))))
1227 (inst mov keyword
(ea n-word-bytes object index
1228 (ash 1 (- word-shift n-fixnum-tag-bits
))))))
1230 (define-vop (more-arg/c
)
1231 (:translate sb-c
:%more-arg
)
1232 (:policy
:fast-safe
)
1233 (:args
(object :scs
(descriptor-reg) :to
(:result
1)))
1235 (:arg-types
* (:constant
(signed-byte #.
(- 32 word-shift
))))
1236 (:results
(value :scs
(descriptor-reg any-reg
)))
1239 (inst mov value
(ea (- (* index n-word-bytes
)) object
))))
1241 (define-vop (more-arg)
1242 (:translate sb-c
:%more-arg
)
1243 (:policy
:fast-safe
)
1244 (:args
(object :scs
(descriptor-reg) :to
(:result
1))
1245 (index :scs
(any-reg) :to
(:result
1) :target value
))
1246 (:arg-types
* tagged-num
)
1247 (:results
(value :scs
(descriptor-reg any-reg
)))
1252 (inst mov value
(ea object value
1253 (ash 1 (- word-shift n-fixnum-tag-bits
))))))
1255 (define-vop (more-arg-or-nil)
1256 (:policy
:fast-safe
)
1257 (:args
(object :scs
(descriptor-reg) :to
(:result
1))
1258 (count :scs
(any-reg) :to
(:result
1)))
1259 (:arg-types
* tagged-num
)
1261 (:results
(value :scs
(descriptor-reg any-reg
)))
1264 (inst mov value nil-value
)
1265 (inst cmp count
(fixnumize index
))
1267 (inst mov value
(ea (- (* index n-word-bytes
)) object
))
1270 ;;; Turn more arg (context, count) into a list.
1271 ;;; Cons cells will be filled in right-to-left.
1272 ;;; This has a slight advantage in code size, and eliminates an initial
1273 ;;; forward jump into the loop. it also admits an interesting possibility
1274 ;;; to reduce the scope of the pseudo-atomic section so as not to
1275 ;;; encompass construction of the list. To do that, we will need to invent
1276 ;;; a new widetag for "contiguous CONS block" which has a header conveying
1277 ;;; the total payload length. Initially we would store that into the CAR of the
1278 ;;; first cons cell. Upon seeing such header, GC shall treat that entire object
1279 ;;; as a boxed payload of specified length. It will be implicitly pinned
1280 ;;; (if conservative) or transported as a whole (if precise). Then when the CAR
1281 ;;; of the first cons is overwritten, the object changes to a linked list.
1283 (:translate %listify-rest-args
)
1285 ;; CONTEXT is used throughout the copying loop
1286 (:args
(context :scs
(descriptor-reg) :to
:save
)
1287 (count :scs
(any-reg) :target rcx
))
1288 (:arg-types
* tagged-num
)
1289 ;; The only advantage to specifying RCX here is that JRCXZ can be used
1290 ;; in one place, and then only in the unlikely scenario that CONTEXT is not
1291 ;; in RCX. If it was, SHL sets/clears the Z flag, but LEA doesn't.
1292 ;; Not much of an advantage, but why not.
1293 (:temporary
(:sc unsigned-reg
:offset rcx-offset
:from
(:argument
1)) rcx
)
1294 ;; Note that DST conflicts with RESULT because we use both as temps
1295 (:temporary
(:sc unsigned-reg
) value dst
)
1296 #+gs-seg
(:temporary
(:sc unsigned-reg
:offset
15) thread-tn
)
1297 (:results
(result :scs
(descriptor-reg)))
1301 ;; TODO: if instrumenting, just revert to the older way of precomputing
1302 ;; a size rather than scaling by 8 in ALLOCATION so that we don't have
1303 ;; to scale and unscale.
1304 ;; Compute the number of bytes to allocate
1305 (let ((shift (- (1+ word-shift
) n-fixnum-tag-bits
)))
1306 (if (location= count rcx
)
1307 (inst shl
:dword rcx shift
)
1308 (inst lea
:dword rcx
(ea nil count
(ash 1 shift
)))))
1310 (move rcx count
:dword
)
1311 ;; Setup for the CDR of the last cons (or the entire result) being NIL.
1312 (inst mov result nil-value
)
1313 (cond ((not (member :allocation-size-histogram sb-xc
:*features
*))
1315 (t ; jumps too far for JRCXZ sometimes
1317 (inst jmp
:z done
)))
1318 (when (and (not (node-stack-allocate-p node
)) (instrument-alloc-policy-p node
))
1319 (inst shl
:dword rcx word-shift
) ; compute byte count
1320 (instrument-alloc +cons-primtype
+ rcx node
(list value dst
) thread-tn
)
1321 (inst shr
:dword rcx word-shift
)) ; undo the computation
1322 (pseudo-atomic (:elide-if
(node-stack-allocate-p node
) :thread-tn thread-tn
)
1323 ;; Produce an untagged pointer into DST
1325 (cond ((node-stack-allocate-p node
)
1326 ;; LEA on RSP would be ok but we'd need to negate RCX first, then un-negate
1327 ;; to compute the final cons, then negate again. So use SHL and SUB instead.
1328 (inst shl
:dword rcx word-shift
)
1329 (stack-allocation rcx
0 dst
)
1332 (allocation +cons-primtype
+ rcx
0 dst node value thread-tn
1339 'call
(if (system-tlab-p 0 node
) 'sys-listify-
&rest
'listify-
&rest
)
1342 (inst jmp alloc-done
)))
1344 ;; Recalculate DST as a tagged pointer to the last cons
1345 (inst lea dst
(ea (- list-pointer-lowtag
(* cons-size n-word-bytes
)) dst rcx scale
))
1346 ;; scale=8 implies RCX counts ncells (as a fixnum) therefore just untag it.
1347 ;; scale=1 implies RCX counts nbytes therefore ncells = RCX/16
1348 (inst shr
:dword rcx
(if (= scale
8) n-fixnum-tag-bits
(1+ word-shift
))))
1349 ;; The rightmost arguments are at lower addresses.
1350 ;; Start by indexing the last argument
1351 (inst neg rcx
) ; :QWORD because it's negative
1353 ;; Grab one value and store into this cons. Use RCX as an index into the
1354 ;; vector of values in CONTEXT, but add 8 because CONTEXT points exactly at
1355 ;; the 0th value, which means that the index is 1 word too low.
1356 ;; (It's -1 if there is exactly 1 value, instead of 0, and so on)
1357 (inst mov value
(ea 8 context rcx
8))
1358 ;; RESULT began as NIL which gives the correct value for the CDR in the final cons.
1359 ;; Subsequently it points to each cons just populated, which is correct all the way
1360 ;; up to and including the final result.
1361 (storew result dst cons-cdr-slot list-pointer-lowtag
)
1362 (storew value dst cons-car-slot list-pointer-lowtag
)
1363 (inst mov result dst
) ; preserve the value to put in the CDR of the preceding cons
1364 (inst sub dst
(* cons-size n-word-bytes
)) ; get the preceding cons
1365 (inst inc rcx
) ; :QWORD because it's negative
1370 ;;; Return the location and size of the &MORE arg glob created by
1371 ;;; COPY-MORE-ARG. SUPPLIED is the total number of arguments supplied
1372 ;;; (originally passed in RCX). FIXED is the number of non-rest
1375 ;;; We must duplicate some of the work done by COPY-MORE-ARG, since at
1376 ;;; that time the environment is in a pretty brain-damaged state,
1377 ;;; preventing this info from being returned as values. What we do is
1378 ;;; compute supplied - fixed, and return a pointer that many words
1379 ;;; below the current stack top.
1381 (:policy
:fast-safe
)
1382 (:translate sb-c
::%more-arg-context
)
1383 (:args
(supplied :scs
(any-reg) :target count
))
1384 (:arg-types positive-fixnum
(:constant fixnum
))
1386 (:results
(context :scs
(descriptor-reg))
1387 (count :scs
(any-reg)))
1388 (:result-types t tagged-num
)
1389 (:note
"more-arg-context")
1391 (move count supplied
)
1392 ;; SP at this point points at the last arg pushed.
1393 ;; Point to the first more-arg, not above it.
1394 (inst lea context
(ea (- (* (1+ fixed
) n-word-bytes
))
1396 (ash 1 (- word-shift n-fixnum-tag-bits
))))
1397 (unless (zerop fixed
)
1398 (inst sub count
(fixnumize fixed
)))))
1400 (define-vop (verify-arg-count)
1401 (:policy
:fast-safe
)
1402 (:args
(nargs :scs
(any-reg)))
1403 (:arg-types positive-fixnum
(:constant t
) (:constant t
))
1404 (:temporary
(:sc unsigned-reg
:offset rbx-offset
) temp
)
1407 (:save-p
:compute-only
)
1409 ;; NOTE: copy-more-arg expects this to issue a CMP for min > 1
1411 (generate-error-code vop
'invalid-arg-count-error nargs
)))
1414 (inst test
:dword nargs nargs
)
1415 (inst cmp
:dword nargs
(fixnumize max
)))
1416 (inst jmp
:ne err-lab
))
1420 (inst lea
:dword temp
(ea (fixnumize (- min
)) nargs
)))
1421 (inst cmp
:dword temp
(fixnumize (- max min
)))
1422 (inst jmp
:a err-lab
))
1425 (inst test
:dword nargs nargs
)
1426 (inst jmp
:e err-lab
))
1428 (inst cmp
:dword nargs
(fixnumize min
))
1429 (inst jmp
:b err-lab
))))))))
1432 (defun emit-single-step-test ()
1433 ;; We use different ways of representing whether stepping is on on
1434 ;; +SB-THREAD / -SB-THREAD: on +SB-THREAD, we use a slot in the
1435 ;; thread structure. On -SB-THREAD we use the value of a static
1436 ;; symbol. Things are done this way, since reading a thread-local
1437 ;; slot from a symbol would require an extra register on +SB-THREAD,
1438 ;; and reading a slot from a thread structure would require an extra
1439 ;; register on -SB-THREAD. While this isn't critical for x86-64,
1440 ;; it's more serious for x86.
1441 #+sb-thread
(inst cmp
:byte
(thread-slot-ea thread-stepping-slot
) 0)
1442 #-sb-thread
(inst cmp
:byte
(static-symbol-value-ea 'sb-impl
::*stepping
*) 0))
1444 (define-vop (step-instrument-before-vop)
1445 (:policy
:fast-safe
)
1448 (emit-single-step-test)
1450 (inst break single-step-before-trap
)
1452 (note-this-location vop
:internal-error
)))