1;;;; function call for the x86 VM
2
3;;;; This software is part of the SBCL system. See the README file for
4;;;; more information.
5;;;;
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.
11
12(in-package "SB!VM")
13
14(defconstant arg-count-sc (make-sc-offset any-reg-sc-number ecx-offset))
15(defconstant closure-sc (make-sc-offset descriptor-reg-sc-number eax-offset))
16
17;;; Make a passing location TN for a local call return PC.
18;;;
19;;; Always wire the return PC location to the stack in its standard
20;;; location.
21(defun make-return-pc-passing-location (standard)
22  (declare (ignore standard))
23  (make-wired-tn (primitive-type-or-lose 'system-area-pointer)
24                 sap-stack-sc-number return-pc-save-offset))
25
26(defconstant return-pc-passing-offset
27  (make-sc-offset sap-stack-sc-number return-pc-save-offset))
28
29;;; This is similar to MAKE-RETURN-PC-PASSING-LOCATION, but makes a
30;;; location to pass OLD-FP in.
31;;;
32;;; This is wired in both the standard and the local-call conventions,
33;;; because we want to be able to assume it's always there. Besides,
34;;; the x86 doesn't have enough registers to really make it profitable
35;;; to pass it in a register.
36(defun make-old-fp-passing-location (standard)
37  (declare (ignore standard))
38  (make-wired-tn *fixnum-primitive-type* control-stack-sc-number
39                 ocfp-save-offset))
40
41(defconstant old-fp-passing-offset
42  (make-sc-offset control-stack-sc-number ocfp-save-offset))
43
44;;; Make the TNs used to hold OLD-FP and RETURN-PC within the current
45;;; function. We treat these specially so that the debugger can find
46;;; them at a known location.
47;;;
48;;; Without using a save-tn - which does not make much sense if it is
49;;; wired to the stack?
50(defun make-old-fp-save-location (physenv)
51  (physenv-debug-live-tn (make-wired-tn *fixnum-primitive-type*
52                                        control-stack-sc-number
53                                        ocfp-save-offset)
54                         physenv))
55(defun make-return-pc-save-location (physenv)
56  (physenv-debug-live-tn
57   (make-wired-tn (primitive-type-or-lose 'system-area-pointer)
58                  sap-stack-sc-number return-pc-save-offset)
59   physenv))
60
61;;; Make a TN for the standard argument count passing location. We only
62;;; need to make the standard location, since a count is never passed when we
63;;; are using non-standard conventions.
64(defun make-arg-count-location ()
65  (make-wired-tn *fixnum-primitive-type* any-reg-sc-number ecx-offset))
66
67;;;; frame hackery
68
69;;; This is used for setting up the Old-FP in local call.
70(define-vop (current-fp)
71  (:results (val :scs (any-reg control-stack)))
72  (:generator 1
73    (move val ebp-tn)))
74
75;;; We don't have a separate NFP, so we don't need to do anything here.
76(define-vop (compute-old-nfp)
77  (:results (val))
78  (:ignore val)
79  (:generator 1
80    nil))
81
82;;; Accessing a slot from an earlier stack frame is definite hackery.
83(define-vop (ancestor-frame-ref)
84  (:args (frame-pointer :scs (descriptor-reg))
85         (variable-home-tn :load-if nil))
86  (:results (value :scs (descriptor-reg any-reg)))
87  (:policy :fast-safe)
88  (:generator 4
89    (aver (sc-is variable-home-tn control-stack))
90    (loadw value frame-pointer
91           (frame-word-offset (tn-offset variable-home-tn)))))
92(define-vop (ancestor-frame-set)
93  (:args (frame-pointer :scs (descriptor-reg))
94         (value :scs (descriptor-reg any-reg)))
95  (:results (variable-home-tn :load-if nil))
96  (:policy :fast-safe)
97  (:generator 4
98    (aver (sc-is variable-home-tn control-stack))
99    (storew value frame-pointer
100            (frame-word-offset (tn-offset variable-home-tn)))))
101
102(macrolet ((define-frame-op
103               (suffix sc stack-sc instruction
104                &optional (ea
105                           `(make-ea :dword
106                                     :base frame-pointer
107                                     :disp (frame-byte-offset
108                                            (tn-offset variable-home-tn)))))
109               (let ((reffer (symbolicate 'ancestor-frame-ref '/ suffix))
110                     (setter (symbolicate 'ancestor-frame-set '/ suffix)))
111                 `(progn
112                    (define-vop (,reffer ancestor-frame-ref)
113                      (:results (value :scs (,sc)))
114                      (:generator 4
115                        (aver (sc-is variable-home-tn ,stack-sc))
116                        (inst ,instruction value
117                              ,ea)))
118                    (define-vop (,setter ancestor-frame-set)
119                      (:args (frame-pointer :scs (descriptor-reg))
120                             (value :scs (,sc)))
121                      (:generator 4
122                        (aver (sc-is variable-home-tn ,stack-sc))
123                        (inst ,instruction ,ea value))))))
124           (define-x87-frame-op
125               (suffix sc stack-sc (load set)
126                &optional (ea
127                           `(make-ea :dword
128                                     :base frame-pointer
129                                     :disp (frame-byte-offset
130                                            (tn-offset variable-home-tn)))))
131               (let ((reffer (symbolicate 'ancestor-frame-ref '/ suffix))
132                     (setter (symbolicate 'ancestor-frame-set '/ suffix)))
133                 `(progn
134                    (define-vop (,reffer ancestor-frame-ref)
135                      (:results (value :scs (,sc)))
136                      (:generator 4
137                        (aver (sc-is variable-home-tn ,stack-sc))
138                        ,(if (symbolp load)
139                             `(with-empty-tn@fp-top (value)
140                                (inst ,load ,ea))
141                             load)))
142                    (define-vop (,setter ancestor-frame-set)
143                      (:args (frame-pointer :scs (descriptor-reg))
144                             (value :scs (,sc)))
145                      (:generator 4
146                        (aver (sc-is variable-home-tn ,stack-sc))
147                        ,(if (symbolp set)
148                             `(with-tn@fp-top (value)
149                                (inst ,set ,ea))
150                             set)))))))
151  (define-frame-op signed-byte-32 signed-reg signed-stack mov)
152  (define-frame-op unsigned-byte-32 unsigned-reg unsigned-stack mov)
153  (define-frame-op system-area-pointer sap-reg sap-stack mov)
154
155  (define-x87-frame-op double-float double-reg double-stack
156    (fldd fstd) (make-ea :dword
157                         :base frame-pointer
158                         :disp (frame-byte-offset
159                                (1+ (tn-offset variable-home-tn)))))
160  (define-x87-frame-op single-float single-reg single-stack
161    (fld fst))
162
163  (define-x87-frame-op complex-double-float complex-double-reg
164    complex-double-stack
165    ((let ((real (complex-double-reg-real-tn value))
166           (imag (complex-double-reg-imag-tn value)))
167       (with-empty-tn@fp-top (real)
168         (inst fldd (ea-for-cdf-real-stack variable-home-tn frame-pointer)))
169       (with-empty-tn@fp-top (imag)
170         (inst fldd (ea-for-cdf-imag-stack variable-home-tn frame-pointer))))
171     (let ((real (complex-double-reg-real-tn value))
172           (imag (complex-double-reg-imag-tn value)))
173       (with-tn@fp-top (real)
174         (inst fstd (ea-for-cdf-real-stack variable-home-tn frame-pointer)))
175       (with-tn@fp-top (imag)
176         (inst fstd (ea-for-cdf-imag-stack variable-home-tn frame-pointer))))))
177  (define-x87-frame-op complex-single-float complex-single-reg
178    complex-single-stack
179    ((let ((real (complex-single-reg-real-tn value))
180           (imag (complex-single-reg-imag-tn value)))
181       (with-empty-tn@fp-top (real)
182         (inst fld (ea-for-csf-real-stack variable-home-tn frame-pointer)))
183       (with-empty-tn@fp-top (imag)
184         (inst fld (ea-for-csf-imag-stack variable-home-tn frame-pointer))))
185     (let ((real (complex-single-reg-real-tn value))
186           (imag (complex-single-reg-imag-tn value)))
187       (with-tn@fp-top (real)
188         (inst fst (ea-for-csf-real-stack variable-home-tn frame-pointer)))
189       (with-tn@fp-top (imag)
190         (inst fst (ea-for-csf-imag-stack variable-home-tn frame-pointer)))))))
191
192(defun primitive-type-indirect-cell-type (ptype)
193  (declare (type primitive-type ptype))
194  (macrolet ((foo (&body data)
195                 `(case (primitive-type-name ptype)
196                    ,@(loop for (name stack-sc ref set) in data
197                            collect
198                            `(,name
199                               (load-time-value
200                                (list (primitive-type-or-lose ',name)
201                                      (sc-or-lose ',stack-sc)
202                                      (lambda (node block fp value res)
203                                        (sb!c::vop ,ref node block
204                                                   fp value res))
205                                      (lambda (node block fp new-val value)
206                                        (sb!c::vop ,set node block
207                                                   fp new-val value)))))))))
208    (foo (double-float double-stack
209                       ancestor-frame-ref/double-float
210                       ancestor-frame-set/double-float)
211         (single-float single-stack
212                       ancestor-frame-ref/single-float
213                       ancestor-frame-set/single-float)
214         (complex-double-float complex-double-stack
215                               ancestor-frame-ref/complex-double-float
216                               ancestor-frame-set/complex-double-float)
217         (complex-single-float complex-single-stack
218                               ancestor-frame-ref/complex-single-float
219                               ancestor-frame-set/complex-single-float)
220         (signed-byte-32 signed-stack
221                         ancestor-frame-ref/signed-byte-32
222                         ancestor-frame-set/signed-byte-32)
223         (unsigned-byte-32 unsigned-stack
224                           ancestor-frame-ref/unsigned-byte-32
225                           ancestor-frame-set/unsigned-byte-32)
226         (unsigned-byte-31 unsigned-stack
227                           ancestor-frame-ref/unsigned-byte-32
228                           ancestor-frame-set/unsigned-byte-32)
229         (system-area-pointer sap-stack
230                              ancestor-frame-ref/system-area-pointer
231                              ancestor-frame-set/system-area-pointer))))
232
233(define-vop (xep-allocate-frame)
234  (:info start-lab)
235  (:generator 1
236    (emit-alignment n-lowtag-bits)
237    (emit-label start-lab)
238    ;; Skip space for the function header.
239    (inst simple-fun-header-word)
240    (dotimes (i (1- simple-fun-code-offset))
241      (inst dword 0))
242
243    ;; The start of the actual code.
244    ;; Save the return-pc.
245    (popw ebp-tn (frame-word-offset return-pc-save-offset))))
246
247(define-vop (xep-setup-sp)
248  (:generator 1
249    (inst lea esp-tn
250            (make-ea :dword :base ebp-tn
251                     :disp (- (* n-word-bytes
252                                 (- (max 3 (sb-allocated-size 'stack))
253                                    sp->fp-offset)))))))
254
255;;; This is emitted directly before either a known-call-local, call-local,
256;;; or a multiple-call-local. All it does is allocate stack space for the
257;;; callee (who has the same size stack as us).
258(define-vop (allocate-frame)
259  (:results (res :scs (any-reg))
260            (nfp))
261  (:info callee)
262  (:ignore nfp callee)
263  (:generator 2
264    (inst lea res (make-ea :dword :base esp-tn
265                           :disp (- (* sp->fp-offset n-word-bytes))))
266    (inst sub esp-tn (* n-word-bytes (sb-allocated-size 'stack)))))
267
268;;; Allocate a partial frame for passing stack arguments in a full
269;;; call. NARGS is the number of arguments passed. We allocate at
270;;; least 3 slots, because the XEP noise is going to want to use them
271;;; before it can extend the stack.
272(define-vop (allocate-full-call-frame)
273  (:info nargs)
274  (:results (res :scs (any-reg)))
275  (:generator 2
276    (inst lea res (make-ea :dword :base esp-tn
277                           :disp (- (* sp->fp-offset n-word-bytes))))
278    (inst sub esp-tn (* (max nargs 3) n-word-bytes))))
279
280;;; Emit code needed at the return-point from an unknown-values call
281;;; for a fixed number of values. Values is the head of the TN-REF
282;;; list for the locations that the values are to be received into.
283;;; Nvals is the number of values that are to be received (should
284;;; equal the length of Values).
285;;;
286;;; If 0 or 1 values are expected, then we just emit an instruction to
287;;; reset the SP (which will only be executed when other than 1 value
288;;; is returned.)
289;;;
290;;; In the general case we have to do three things:
291;;;  -- Default unsupplied register values. This need only be done
292;;;     when a single value is returned, since register values are
293;;;     defaulted by the called in the non-single case.
294;;;  -- Default unsupplied stack values. This needs to be done whenever
295;;;     there are stack values.
296;;;  -- Reset SP. This must be done whenever other than 1 value is
297;;;     returned, regardless of the number of values desired.
298(defun default-unknown-values (vop values nvals node)
299  (declare (type (or tn-ref null) values)
300           (type unsigned-byte nvals))
301  (let ((type (sb!c::basic-combination-derived-type node)))
302    (cond
303      ((<= nvals 1)
304       (note-this-location vop :single-value-return)
305       (cond
306         ((<= (values-type-max-value-count type)
307              register-arg-count)
308          (when (and (named-type-p type)
309                     (eq nil (named-type-name type)))
310            ;; The function never returns, it may happen that the code
311            ;; ends right here leavig the :SINGLE-VALUE-RETURN note
312            ;; dangling. Let's emit a NOP.
313            (inst nop)))
314         ((not (values-type-may-be-single-value-p type))
315          (inst mov esp-tn ebx-tn))
316         ((member :cmov *backend-subfeatures*)
317          (inst cmov :c esp-tn ebx-tn))
318         (t
319          (let ((single-value (gen-label)))
320            (inst jmp :nc single-value)
321            (inst mov esp-tn ebx-tn)
322            (emit-label single-value)))))
323      ((<= nvals register-arg-count)
324       (note-this-location vop :unknown-return)
325       (when (values-type-may-be-single-value-p type)
326         (let ((regs-defaulted (gen-label)))
327           (inst jmp :c regs-defaulted)
328           ;; Default the unsupplied registers.
329           (let* ((2nd-tn-ref (tn-ref-across values))
330                  (2nd-tn (tn-ref-tn 2nd-tn-ref)))
331             (inst mov 2nd-tn nil-value)
332             (when (> nvals 2)
333               (loop
334                for tn-ref = (tn-ref-across 2nd-tn-ref)
335                then (tn-ref-across tn-ref)
336                for count from 2 below register-arg-count
337                do (inst mov (tn-ref-tn tn-ref) 2nd-tn))))
338           (inst mov ebx-tn esp-tn)
339           (emit-label regs-defaulted)))
340       (when (< register-arg-count
341                (values-type-max-value-count type))
342         (inst mov esp-tn ebx-tn)))
343      ((<= nvals 7)
344       ;; The number of bytes depends on the relative jump instructions.
345       ;; Best case is 31+(n-3)*14, worst case is 35+(n-3)*18. For
346       ;; NVALS=6 that is 73/89 bytes, and for NVALS=7 that is 87/107
347       ;; bytes which is likely better than using the blt below.
348       (let ((regs-defaulted (gen-label))
349             (defaulting-done (gen-label))
350             (default-stack-slots (gen-label)))
351         (note-this-location vop :unknown-return)
352         ;; Branch off to the MV case.
353         (inst jmp :c regs-defaulted)
354         ;; Do the single value case.
355         ;; Default the register args
356         (inst mov eax-tn nil-value)
357         (do ((i 1 (1+ i))
358              (val (tn-ref-across values) (tn-ref-across val)))
359             ((= i (min nvals register-arg-count)))
360           (inst mov (tn-ref-tn val) eax-tn))
361         ;; Fake other registers so it looks like we returned with all the
362         ;; registers filled in.
363         (move ebx-tn esp-tn)
364         (inst jmp default-stack-slots)
365         (emit-label regs-defaulted)
366         (inst mov eax-tn nil-value)
367         (collect ((defaults))
368           (do ((i register-arg-count (1+ i))
369                (val (do ((i 0 (1+ i))
370                          (val values (tn-ref-across val)))
371                         ((= i register-arg-count) val))
372                     (tn-ref-across val)))
373               ((null val))
374             (let ((default-lab (gen-label))
375                   (tn (tn-ref-tn val))
376                   (first-stack-arg-p (= i register-arg-count)))
377               (defaults (cons default-lab
378                               (cons tn first-stack-arg-p)))
379               (inst cmp ecx-tn (fixnumize i))
380               (inst jmp :be default-lab)
381               (when first-stack-arg-p
382                 ;; There are stack args so the frame of the callee is
383                 ;; still there, save EDX in its first slot temporalily.
384                 (storew edx-tn ebx-tn (frame-word-offset sp->fp-offset)))
385               (loadw edx-tn ebx-tn (frame-word-offset (+ sp->fp-offset i)))
386               (inst mov tn edx-tn)))
387           (emit-label defaulting-done)
388           (loadw edx-tn ebx-tn (frame-word-offset sp->fp-offset))
389           (move esp-tn ebx-tn)
390           (let ((defaults (defaults)))
391             (when defaults
392               (assemble (*elsewhere*)
393                 (emit-label default-stack-slots)
394                 (dolist (default defaults)
395                   (emit-label (car default))
396                   (when (cddr default)
397                     ;; We are setting the first stack argument to NIL.
398                     ;; The callee's stack frame is dead, save EDX by
399                     ;; pushing it to the stack, it will end up at same
400                     ;; place as in the (STOREW EDX-TN EBX-TN -1) case
401                     ;; above.
402                     (inst push edx-tn))
403                   (inst mov (second default) eax-tn))
404                 (inst jmp defaulting-done)))))))
405      (t
406       ;; 91 bytes for this branch.
407       (let ((regs-defaulted (gen-label))
408             (restore-edi (gen-label))
409             (no-stack-args (gen-label))
410             (default-stack-vals (gen-label))
411             (count-okay (gen-label)))
412         (note-this-location vop :unknown-return)
413         ;; Branch off to the MV case.
414         (inst jmp :c regs-defaulted)
415         ;; Default the register args, and set up the stack as if we
416         ;; entered the MV return point.
417         (inst mov ebx-tn esp-tn)
418         (inst mov edi-tn nil-value)
419         (inst mov esi-tn edi-tn)
420         ;; Compute a pointer to where to put the [defaulted] stack values.
421         (emit-label no-stack-args)
422         (inst push edx-tn)
423         (inst push edi-tn)
424         (inst lea edi-tn
425               (make-ea :dword :base ebp-tn
426                        :disp (frame-byte-offset register-arg-count)))
427         ;; Load EAX with NIL so we can quickly store it, and set up
428         ;; stuff for the loop.
429         (inst mov eax-tn nil-value)
430         (inst std)
431         (inst mov ecx-tn (- nvals register-arg-count))
432         ;; Jump into the default loop.
433         (inst jmp default-stack-vals)
434         ;; The regs are defaulted. We need to copy any stack arguments,
435         ;; and then default the remaining stack arguments.
436         (emit-label regs-defaulted)
437         ;; Compute the number of stack arguments, and if it's zero or
438         ;; less, don't copy any stack arguments.
439         (inst sub ecx-tn (fixnumize register-arg-count))
440         (inst jmp :le no-stack-args)
441         ;; Save EDI.
442         (storew edi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 1)))
443         ;; Throw away any unwanted args.
444         (inst cmp ecx-tn (fixnumize (- nvals register-arg-count)))
445         (inst jmp :be count-okay)
446         (inst mov ecx-tn (fixnumize (- nvals register-arg-count)))
447         (emit-label count-okay)
448         ;; Save the number of stack values.
449         (inst mov eax-tn ecx-tn)
450         ;; Compute a pointer to where the stack args go.
451         (inst lea edi-tn
452               (make-ea :dword :base ebp-tn
453                        :disp (frame-byte-offset register-arg-count)))
454         ;; Save ESI, and compute a pointer to where the args come from.
455         (storew esi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 2)))
456         (inst lea esi-tn
457               (make-ea :dword :base ebx-tn
458                        :disp (frame-byte-offset
459                               (+ sp->fp-offset register-arg-count))))
460         ;; Do the copy.
461         (inst shr ecx-tn word-shift)   ; make word count
462         (inst std)
463         (inst rep)
464         (inst movs :dword)
465         ;; Restore ESI.
466         (loadw esi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 2)))
467         ;; Now we have to default the remaining args. Find out how many.
468         (inst sub eax-tn (fixnumize (- nvals register-arg-count)))
469         (inst neg eax-tn)
470         ;; If none, then just blow out of here.
471         (inst jmp :le restore-edi)
472         (inst mov ecx-tn eax-tn)
473         (inst shr ecx-tn word-shift)   ; word count
474         ;; Load EAX with NIL for fast storing.
475         (inst mov eax-tn nil-value)
476         ;; Do the store.
477         (emit-label default-stack-vals)
478         (inst rep)
479         (inst stos eax-tn)
480         ;; Restore EDI, and reset the stack.
481         (emit-label restore-edi)
482         (loadw edi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 1)))
483         (inst mov esp-tn ebx-tn)
484         (inst cld)))))
485  (values))
486
487;;;; unknown values receiving
488
489;;; Emit code needed at the return point for an unknown-values call
490;;; for an arbitrary number of values.
491;;;
492;;; We do the single and non-single cases with no shared code: there
493;;; doesn't seem to be any potential overlap, and receiving a single
494;;; value is more important efficiency-wise.
495;;;
496;;; When there is a single value, we just push it on the stack,
497;;; returning the old SP and 1.
498;;;
499;;; When there is a variable number of values, we move all of the
500;;; argument registers onto the stack, and return ARGS and NARGS.
501;;;
502;;; ARGS and NARGS are TNs wired to the named locations. We must
503;;; explicitly allocate these TNs, since their lifetimes overlap with
504;;; the results start and count. (Also, it's nice to be able to target
505;;; them.)
506(defun receive-unknown-values (args nargs start count node)
507  (declare (type tn args nargs start count))
508  (let ((type (sb!c::basic-combination-derived-type node))
509        (variable-values (gen-label))
510        (stack-values (gen-label))
511        (done (gen-label)))
512    (when (values-type-may-be-single-value-p type)
513      (inst jmp :c variable-values)
514      (cond ((location= start (first *register-arg-tns*))
515             (inst push (first *register-arg-tns*))
516             (inst lea start (make-ea :dword :base esp-tn :disp n-word-bytes)))
517            (t (inst mov start esp-tn)
518               (inst push (first *register-arg-tns*))))
519      (inst mov count (fixnumize 1))
520      (inst jmp done)
521      (emit-label variable-values))
522    ;; The stack frame is burnt and RETurned from if there are no
523    ;; stack values. In this case quickly reallocate sufficient space.
524    (when (<= (values-type-min-value-count type)
525              register-arg-count)
526      (inst cmp nargs (fixnumize register-arg-count))
527      (inst jmp :g stack-values)
528      (inst sub esp-tn nargs)
529      (emit-label stack-values))
530    ;; dtc: this writes the registers onto the stack even if they are
531    ;; not needed, only the number specified in ecx are used and have
532    ;; stack allocated to them. No harm is done.
533    (loop
534      for arg in *register-arg-tns*
535      for i downfrom -1
536      for j below (values-type-max-value-count type)
537      do (storew arg args i))
538    (move start args)
539    (move count nargs)
540
541    (emit-label done))
542  (values))
543
544;;; VOP that can be inherited by unknown values receivers. The main thing this
545;;; handles is allocation of the result temporaries.
546(define-vop (unknown-values-receiver)
547  (:temporary (:sc descriptor-reg :offset ebx-offset
548                   :from :eval :to (:result 0))
549              values-start)
550  (:temporary (:sc any-reg :offset ecx-offset
551               :from :eval :to (:result 1))
552              nvals)
553  (:results (start :scs (any-reg control-stack))
554            (count :scs (any-reg control-stack))))
555
556;;;; local call with unknown values convention return
557
558(defun check-ocfp-and-return-pc (old-fp return-pc)
559  #+nil
560  (format t "*known-return: old-fp ~S, tn-kind ~S; ~S ~S~%"
561          old-fp (sb!c::tn-kind old-fp) (sb!c::tn-save-tn old-fp)
562          (sb!c::tn-kind (sb!c::tn-save-tn old-fp)))
563  #+nil
564  (format t "*known-return: return-pc ~S, tn-kind ~S; ~S ~S~%"
565          return-pc (sb!c::tn-kind return-pc)
566          (sb!c::tn-save-tn return-pc)
567          (sb!c::tn-kind (sb!c::tn-save-tn return-pc)))
568  (unless (and (sc-is old-fp control-stack)
569               (= (tn-offset old-fp) ocfp-save-offset))
570    (error "ocfp not on stack in standard save location?"))
571  (unless (and (sc-is return-pc sap-stack)
572               (= (tn-offset return-pc) return-pc-save-offset))
573    (error "return-pc not on stack in standard save location?")))
574
575;;; The local call convention doesn't fit that well with x86-style
576;;; calls. Emit a header for local calls to pop the return address
577;;; in the right place.
578(defun emit-block-header (start-label trampoline-label fall-thru-p alignp)
579  (declare (ignore alignp))
580  (when trampoline-label
581    (when fall-thru-p
582      (inst jmp start-label))
583    (emit-label trampoline-label)
584    (popw ebp-tn (frame-word-offset return-pc-save-offset)))
585  (emit-label start-label))
586
587;;; Non-TR local call for a fixed number of values passed according to
588;;; the unknown values convention.
589;;;
590;;; FP is the frame pointer in install before doing the call.
591;;;
592;;; NFP would be the number-stack frame pointer if we had a separate
593;;; number stack.
594;;;
595;;; Args are the argument passing locations, which are specified only
596;;; to terminate their lifetimes in the caller.
597;;;
598;;; VALUES are the return value locations (wired to the standard
599;;; passing locations). NVALS is the number of values received.
600;;;
601;;; Save is the save info, which we can ignore since saving has been
602;;; done.
603;;;
604;;; TARGET is a continuation pointing to the start of the called
605;;; function.
606(define-vop (call-local)
607  (:args (fp)
608         (nfp)
609         (args :more t))
610  (:results (values :more t))
611  (:save-p t)
612  (:move-args :local-call)
613  (:info arg-locs callee target nvals)
614  (:vop-var vop)
615  (:ignore nfp arg-locs args callee)
616  (:node-var node)
617  (:generator 5
618    (move ebp-tn fp)
619    (note-this-location vop :call-site)
620    (inst call target)
621    (default-unknown-values vop values nvals node)))
622
623;;; Non-TR local call for a variable number of return values passed according
624;;; to the unknown values convention. The results are the start of the values
625;;; glob and the number of values received.
626(define-vop (multiple-call-local unknown-values-receiver)
627  (:args (fp)
628         (nfp)
629         (args :more t))
630  (:save-p t)
631  (:move-args :local-call)
632  (:info save callee target)
633  (:ignore args save nfp callee)
634  (:vop-var vop)
635  (:node-var node)
636  (:generator 20
637    (move ebp-tn fp)
638    (note-this-location vop :call-site)
639    (inst call target)
640    (note-this-location vop :unknown-return)
641    (receive-unknown-values values-start nvals start count node)))
642
643;;;; local call with known values return
644
645;;; Non-TR local call with known return locations. Known-value return
646;;; works just like argument passing in local call.
647;;;
648;;; Note: we can't use normal load-tn allocation for the fixed args,
649;;; since all registers may be tied up by the more operand. Instead,
650;;; we use MAYBE-LOAD-STACK-TN.
651(define-vop (known-call-local)
652  (:args (fp)
653         (nfp)
654         (args :more t))
655  (:results (res :more t))
656  (:move-args :local-call)
657  (:save-p t)
658  (:info save callee target)
659  (:ignore args res save nfp callee)
660  (:vop-var vop)
661  (:generator 5
662    (move ebp-tn fp)
663    (note-this-location vop :call-site)
664    (inst call target)
665    (note-this-location vop :known-return)))
666
667;;; From Douglas Crosher
668;;; Return from known values call. We receive the return locations as
669;;; arguments to terminate their lifetimes in the returning function. We
670;;; restore FP and CSP and jump to the Return-PC.
671(define-vop (known-return)
672  (:args (old-fp)
673         (return-pc)
674         (vals :more t))
675  (:move-args :known-return)
676  (:info val-locs)
677  (:ignore val-locs vals)
678  (:vop-var vop)
679  (:generator 6
680    (check-ocfp-and-return-pc old-fp return-pc)
681    ;; Zot all of the stack except for the old-fp and return-pc.
682    (inst mov esp-tn ebp-tn)
683    (inst pop ebp-tn)
684    (inst ret)))
685
686;;;; full call
687;;;
688;;; There is something of a cross-product effect with full calls.
689;;; Different versions are used depending on whether we know the
690;;; number of arguments or the name of the called function, and
691;;; whether we want fixed values, unknown values, or a tail call.
692;;;
693;;; In full call, the arguments are passed creating a partial frame on
694;;; the stack top and storing stack arguments into that frame. On
695;;; entry to the callee, this partial frame is pointed to by FP.
696
697;;; This macro helps in the definition of full call VOPs by avoiding
698;;; code replication in defining the cross-product VOPs.
699;;;
700;;; NAME is the name of the VOP to define.
701;;;
702;;; NAMED is true if the first argument is an fdefinition object whose
703;;; definition is to be called.
704;;;
705;;; RETURN is either :FIXED, :UNKNOWN or :TAIL:
706;;; -- If :FIXED, then the call is for a fixed number of values, returned in
707;;;    the standard passing locations (passed as result operands).
708;;; -- If :UNKNOWN, then the result values are pushed on the stack, and the
709;;;    result values are specified by the Start and Count as in the
710;;;    unknown-values continuation representation.
711;;; -- If :TAIL, then do a tail-recursive call. No values are returned.
712;;;    The Old-Fp and Return-PC are passed as the second and third arguments.
713;;;
714;;; In non-tail calls, the pointer to the stack arguments is passed as
715;;; the last fixed argument. If Variable is false, then the passing
716;;; locations are passed as a more arg. Variable is true if there are
717;;; a variable number of arguments passed on the stack. Variable
718;;; cannot be specified with :TAIL return. TR variable argument call
719;;; is implemented separately.
720;;;
721;;; In tail call with fixed arguments, the passing locations are
722;;; passed as a more arg, but there is no new-FP, since the arguments
723;;; have been set up in the current frame.
724(macrolet ((define-full-call (name named return variable)
725            (aver (not (and variable (eq return :tail))))
726            `(define-vop (,name
727                          ,@(when (eq return :unknown)
728                              '(unknown-values-receiver)))
729               (:args
730               ,@(unless (eq return :tail)
731                   '((new-fp :scs (any-reg) :to (:argument 1))))
732
733               (fun :scs (descriptor-reg control-stack)
734                    :target eax :to (:argument 0))
735
736               ,@(when (eq return :tail)
737                   '((old-fp)
738                     (return-pc)))
739
740               ,@(unless variable '((args :more t :scs (descriptor-reg)))))
741
742               ,@(when (eq return :fixed)
743               '((:results (values :more t))))
744
745               (:save-p ,(if (eq return :tail) :compute-only t))
746
747               ,@(unless (or (eq return :tail) variable)
748               '((:move-args :full-call)))
749
750               (:vop-var vop)
751               (:info
752               ,@(unless (or variable (eq return :tail)) '(arg-locs))
753               ,@(unless variable '(nargs))
754               ,@(when (eq return :fixed) '(nvals))
755               step-instrumenting)
756
757               (:ignore
758               ,@(unless (or variable (eq return :tail)) '(arg-locs))
759               ,@(unless variable '(args)))
760
761               ;; We pass either the fdefn object (for named call) or
762               ;; the actual function object (for unnamed call) in
763               ;; EAX. With named call, closure-tramp will replace it
764               ;; with the real function and invoke the real function
765               ;; for closures. Non-closures do not need this value,
766               ;; so don't care what shows up in it.
767               (:temporary
768               (:sc descriptor-reg
769                    :offset eax-offset
770                    :from (:argument 0)
771                    :to :eval)
772               eax)
773
774               ;; We pass the number of arguments in ECX.
775               (:temporary (:sc unsigned-reg :offset ecx-offset :to :eval) ecx)
776
777               ;; With variable call, we have to load the
778               ;; register-args out of the (new) stack frame before
779               ;; doing the call. Therefore, we have to tell the
780               ;; lifetime stuff that we need to use them.
781               ,@(when variable
782                   (mapcar (lambda (name offset)
783                             `(:temporary (:sc descriptor-reg
784                                               :offset ,offset
785                                               :from (:argument 0)
786                                               :to :eval)
787                                          ,name))
788                           *register-arg-names* *register-arg-offsets*))
789
790               ,@(when (eq return :tail)
791                   '((:temporary (:sc unsigned-reg
792                                      :from (:argument 1)
793                                      :to (:argument 2))
794                                 old-fp-tmp)))
795               ,@(unless (eq return :tail)
796                   '((:node-var node)))
797
798               (:generator ,(+ (if named 5 0)
799                               (if variable 19 1)
800                               (if (eq return :tail) 0 10)
801                               15
802                               (if (eq return :unknown) 25 0))
803
804               ;; This has to be done before the frame pointer is
805               ;; changed! EAX stores the 'lexical environment' needed
806               ;; for closures.
807               (move eax fun)
808
809
810               ,@(if variable
811                     ;; For variable call, compute the number of
812                     ;; arguments and move some of the arguments to
813                     ;; registers.
814                     (collect ((noise))
815                              ;; Compute the number of arguments.
816                              (noise '(inst mov ecx new-fp))
817                              (noise '(inst sub ecx esp-tn))
818                              ;; Move the necessary args to registers,
819                              ;; this moves them all even if they are
820                              ;; not all needed.
821                              (loop
822                               for name in *register-arg-names*
823                               for index downfrom -1
824                               do (noise `(loadw ,name new-fp ,index)))
825                              (noise))
826                   '((if (zerop nargs)
827                         (inst xor ecx ecx)
828                       (inst mov ecx (fixnumize nargs)))))
829               ,@(cond ((eq return :tail)
830                        '(;; Python has figured out what frame we should
831                          ;; return to so might as well use that clue.
832                          ;; This seems really important to the
833                          ;; implementation of things like
834                          ;; (without-interrupts ...)
835                          ;;
836                          ;; dtc; Could be doing a tail call from a
837                          ;; known-local-call etc in which the old-fp
838                          ;; or ret-pc are in regs or in non-standard
839                          ;; places. If the passing location were
840                          ;; wired to the stack in standard locations
841                          ;; then these moves will be un-necessary;
842                          ;; this is probably best for the x86.
843                          (sc-case old-fp
844                                   ((control-stack)
845                                    (unless (= ocfp-save-offset
846                                               (tn-offset old-fp))
847                                      ;; FIXME: FORMAT T for stale
848                                      ;; diagnostic output (several of
849                                      ;; them around here), ick
850                                      (error "** tail-call old-fp not S0~%")
851                                      (move old-fp-tmp old-fp)
852                                      (storew old-fp-tmp
853                                              ebp-tn
854                                              (frame-word-offset ocfp-save-offset))))
855                                   ((any-reg descriptor-reg)
856                                    (error "** tail-call old-fp in reg not S0~%")
857                                    (storew old-fp
858                                            ebp-tn
859                                            (frame-word-offset ocfp-save-offset))))
860
861                          ;; For tail call, we have to push the
862                          ;; return-pc so that it looks like we CALLed
863                          ;; despite the fact that we are going to JMP.
864                          (inst push return-pc)
865                          ))
866                       (t
867                        ;; For non-tail call, we have to save our
868                        ;; frame pointer and install the new frame
869                        ;; pointer. We can't load stack tns after this
870                        ;; point.
871                        `(;; Python doesn't seem to allocate a frame
872                          ;; here which doesn't leave room for the
873                          ;; ofp/ret stuff.
874
875                          ;; The variable args are on the stack and
876                          ;; become the frame, but there may be <3
877                          ;; args and 3 stack slots are assumed
878                          ;; allocate on the call. So need to ensure
879                          ;; there are at least 3 slots. This hack
880                          ;; just adds 3 more.
881                          ,(if variable
882                               '(inst sub esp-tn (fixnumize 3)))
883
884                          ;; Bias the new-fp for use as an fp
885                          ,(if variable
886                               '(inst sub new-fp (fixnumize sp->fp-offset)))
887
888                          ;; Save the fp
889                          (storew ebp-tn new-fp
890                                  (frame-word-offset ocfp-save-offset))
891
892                          (move ebp-tn new-fp) ; NB - now on new stack frame.
893                          )))
894
895               (when step-instrumenting
896                 (emit-single-step-test)
897                 (inst jmp :eq DONE)
898                 (inst break single-step-around-trap))
899               DONE
900
901               (note-this-location vop :call-site)
902
903               (inst ,(if (eq return :tail) 'jmp 'call)
904                     ,(if named
905                          '(make-ea-for-object-slot eax fdefn-raw-addr-slot
906                                                    other-pointer-lowtag)
907                          '(make-ea-for-object-slot eax closure-fun-slot
908                                                    fun-pointer-lowtag)))
909               ,@(ecase return
910                   (:fixed
911                    '((default-unknown-values vop values nvals node)))
912                   (:unknown
913                    '((note-this-location vop :unknown-return)
914                      (receive-unknown-values values-start nvals start count
915                                              node)))
916                   (:tail))))))
917
918  (define-full-call call nil :fixed nil)
919  (define-full-call call-named t :fixed nil)
920  (define-full-call multiple-call nil :unknown nil)
921  (define-full-call multiple-call-named t :unknown nil)
922  (define-full-call tail-call nil :tail nil)
923  (define-full-call tail-call-named t :tail nil)
924
925  (define-full-call call-variable nil :fixed t)
926  (define-full-call multiple-call-variable nil :unknown t))
927
928;;; This is defined separately, since it needs special code that BLT's
929;;; the arguments down. All the real work is done in the assembly
930;;; routine. We just set things up so that it can find what it needs.
931(define-vop (tail-call-variable)
932  (:args (args :scs (any-reg control-stack) :target esi)
933         (function :scs (descriptor-reg control-stack) :target eax)
934         (old-fp)
935         (return-pc))
936  (:temporary (:sc unsigned-reg :offset esi-offset :from (:argument 0)) esi)
937  (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)) eax)
938  (:generator 75
939    (check-ocfp-and-return-pc old-fp return-pc)
940    ;; Move these into the passing locations if they are not already there.
941    (move esi args)
942    (move eax function)
943    ;; And jump to the assembly routine.
944    (inst jmp (make-fixup 'tail-call-variable :assembly-routine))))
945
946;;;; unknown values return
947
948;;; Return a single-value using the Unknown-Values convention.
949;;;
950;;; pfw--get wired-tn conflicts sometimes if register sc specd for args
951;;; having problems targeting args to regs -- using temps instead.
952;;;
953;;; First off, modifying the return-pc defeats the branch-prediction
954;;; optimizations on modern CPUs quite handily. Second, we can do all
955;;; this without needing a temp register. Fixed the latter, at least.
956;;; -- AB 2006/Feb/04
957(define-vop (return-single)
958  (:args (old-fp)
959         (return-pc)
960         (value))
961  (:ignore value)
962  (:generator 6
963    (check-ocfp-and-return-pc old-fp return-pc)
964    ;; Drop stack above old-fp
965    (inst mov esp-tn ebp-tn)
966    ;; Clear the multiple-value return flag
967    (inst clc)
968    ;; Restore the old frame pointer
969    (inst pop ebp-tn)
970    ;; And return.
971    (inst ret)))
972
973;;; Do unknown-values return of a fixed (other than 1) number of
974;;; values. The VALUES are required to be set up in the standard
975;;; passing locations. NVALS is the number of values returned.
976;;;
977;;; Basically, we just load ECX with the number of values returned and
978;;; EBX with a pointer to the values, set ESP to point to the end of
979;;; the values, and jump directly to return-pc.
980(define-vop (return)
981  (:args (old-fp)
982         (return-pc :to (:eval 1))
983         (values :more t))
984  (:ignore values)
985  (:info nvals)
986  ;; In the case of other than one value, we need these registers to
987  ;; tell the caller where they are and how many there are.
988  (:temporary (:sc unsigned-reg :offset ebx-offset) ebx)
989  (:temporary (:sc unsigned-reg :offset ecx-offset) ecx)
990  ;; We need to stretch the lifetime of return-pc past the argument
991  ;; registers so that we can default the argument registers without
992  ;; trashing return-pc.
993  (:temporary (:sc unsigned-reg :offset (first *register-arg-offsets*)
994                   :from :eval) a0)
995  (:temporary (:sc unsigned-reg :offset (second *register-arg-offsets*)
996                   :from :eval) a1)
997  (:temporary (:sc unsigned-reg :offset (third *register-arg-offsets*)
998                   :from :eval) a2)
999
1000  (:generator 6
1001    (check-ocfp-and-return-pc old-fp return-pc)
1002    (when (= nvals 1)
1003      ;; This is handled in RETURN-SINGLE.
1004      (error "nvalues is 1"))
1005    ;; Establish the values pointer and values count.
1006    (inst lea ebx (make-ea :dword :base ebp-tn
1007                           :disp (* sp->fp-offset n-word-bytes)))
1008    (if (zerop nvals)
1009        (inst xor ecx ecx)              ; smaller
1010        (inst mov ecx (fixnumize nvals)))
1011    ;; Pre-default any argument register that need it.
1012    (when (< nvals register-arg-count)
1013      (let* ((arg-tns (nthcdr nvals (list a0 a1 a2)))
1014             (first (first arg-tns)))
1015        (inst mov first nil-value)
1016        (dolist (tn (cdr arg-tns))
1017          (inst mov tn first))))
1018    ;; Set the multiple value return flag.
1019    (inst stc)
1020    ;; And away we go. Except that return-pc is still on the
1021    ;; stack and we've changed the stack pointer. So we have to
1022    ;; tell it to index off of EBX instead of EBP.
1023    (cond ((<= nvals register-arg-count)
1024           (inst mov esp-tn ebp-tn)
1025           (inst pop ebp-tn)
1026           (inst ret))
1027          (t
1028           ;; Some values are on the stack after RETURN-PC and OLD-FP,
1029           ;; can't return normally and some slots of the frame will
1030           ;; be used as temporaries by the receiver.
1031           ;;
1032           ;; Clear as much of the stack as possible, but not past the
1033           ;; old frame address.
1034           (inst lea esp-tn
1035                 (make-ea :dword :base ebp-tn
1036                          :disp (frame-byte-offset (1- nvals))))
1037           (move ebp-tn old-fp)
1038           (inst push (make-ea :dword :base ebx
1039                               :disp (frame-byte-offset
1040                                      (+ sp->fp-offset
1041                                         (tn-offset return-pc)))))
1042           (inst ret)))))
1043
1044;;; Do unknown-values return of an arbitrary number of values (passed
1045;;; on the stack.) We check for the common case of a single return
1046;;; value, and do that inline using the normal single value return
1047;;; convention. Otherwise, we branch off to code that calls an
1048;;; assembly-routine.
1049;;;
1050;;; The assembly routine takes the following args:
1051;;;  ECX -- number of values to find there.
1052;;;  ESI -- pointer to where to find the values.
1053(define-vop (return-multiple)
1054  (:args (old-fp)
1055         (return-pc)
1056         (vals :scs (any-reg) :target esi)
1057         (nvals :scs (any-reg) :target ecx))
1058  (:temporary (:sc unsigned-reg :offset esi-offset :from (:argument 2)) esi)
1059  (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 3)) ecx)
1060  (:temporary (:sc descriptor-reg :offset (first *register-arg-offsets*)
1061                   :from (:eval 0)) a0)
1062  (:node-var node)
1063  (:generator 13
1064    (check-ocfp-and-return-pc old-fp return-pc)
1065    (unless (policy node (> space speed))
1066      ;; Check for the single case.
1067      (let ((not-single (gen-label)))
1068        (inst cmp nvals (fixnumize 1))
1069        (inst jmp :ne not-single)
1070        ;; Return with one value.
1071        (loadw a0 vals -1)
1072        ;; Clear the stack until ocfp.
1073        (inst mov esp-tn ebp-tn)
1074        ;; clear the multiple-value return flag
1075        (inst clc)
1076        ;; Out of here.
1077        (inst pop ebp-tn)
1078        (inst ret)
1079        ;; Nope, not the single case. Jump to the assembly routine.
1080        (emit-label not-single)))
1081    (move esi vals)
1082    (move ecx nvals)
1083    (inst jmp (make-fixup 'return-multiple :assembly-routine))))
1084
1085;;;; XEP hackery
1086
1087;;; Get the lexical environment from its passing location.
1088(define-vop (setup-closure-environment)
1089  (:results (closure :scs (descriptor-reg)))
1090  (:info label)
1091  (:ignore label)
1092  (:generator 6
1093    ;; Get result.
1094    (move closure eax-tn)))
1095
1096;;; Copy a &MORE arg from the argument area to the end of the current
1097;;; frame. FIXED is the number of non-&MORE arguments.
1098;;;
1099;;; The tricky part is doing this without trashing any of the calling
1100;;; convention registers that are still needed. This vop is emitted
1101;;; directly after the xep-allocate frame. That means the registers
1102;;; are in use as follows:
1103;;;
1104;;;  EAX -- The lexenv.
1105;;;  EBX -- Available.
1106;;;  ECX -- The total number of arguments * N-WORD-BYTES.
1107;;;  EDX -- The first arg.
1108;;;  EDI -- The second arg.
1109;;;  ESI -- The third arg.
1110;;;
1111;;; So basically, we have one register available for our use: EBX.
1112;;;
1113;;; What we can do is push the other regs onto the stack, and then
1114;;; restore their values by looking directly below where we put the
1115;;; more-args.
1116(define-vop (copy-more-arg)
1117  (:info fixed)
1118  (:generator 20
1119    ;; Avoid the copy if there are no more args.
1120    (cond ((zerop fixed)
1121           (inst jecxz JUST-ALLOC-FRAME))
1122          (t
1123           (inst cmp ecx-tn (fixnumize fixed))
1124           (inst jmp :be JUST-ALLOC-FRAME)))
1125
1126    ;; Allocate the space on the stack.
1127    ;; stack = ebp + sp->fp-offset - (max 3 frame-size) - (nargs - fixed)
1128    ;;
1129    ;; Problem: this might leave some &more args outside esp, so
1130    ;; clamp the movement for now.  If fixed > frame-size, reset
1131    ;; esp to the end of the current &more args (which *should*
1132    ;; be a noop?), and only set esp to its final value after the
1133    ;; stack-stack memmove loop.  Otherwise, an unlucky signal
1134    ;; could end up overwriting the &more arguments before they're
1135    ;; moved in their final place.
1136    (inst lea ebx-tn
1137          (make-ea :dword :base ebp-tn
1138                          :disp (* n-word-bytes
1139                                   (- sp->fp-offset
1140                                      (max 0
1141                                           (- (max 3 (sb-allocated-size 'stack))
1142                                              fixed))))))
1143    (inst sub ebx-tn ecx-tn)          ; Got the new stack in ebx
1144    (inst mov esp-tn ebx-tn)
1145
1146    ;; Now: nargs>=1 && nargs>fixed
1147
1148    ;; Save the original count of args.
1149    (inst mov ebx-tn ecx-tn)
1150
1151    (cond ((< fixed register-arg-count)
1152           ;; We must stop when we run out of stack args, not when we
1153           ;; run out of more args.
1154           ;; Number to copy = nargs-3
1155           (inst sub ecx-tn (fixnumize register-arg-count))
1156           ;; Everything of interest in registers.
1157           (inst jmp :be DO-REGS))
1158          (t
1159           ;; Number to copy = nargs-fixed
1160           (inst sub ecx-tn (fixnumize fixed))))
1161
1162    (let ((delta (* n-word-bytes
1163                    (- (max 3 (sb-allocated-size 'stack))
1164                       fixed)))
1165          (LOOP (gen-label)))
1166      (cond ((zerop delta)
1167             ;; nothing to move!
1168             )
1169            ((minusp delta)
1170             ;; stack frame smaller than fixed; moving args to higher
1171             ;; addresses (stack grows downard), so copy from the
1172             ;; end.  Moreover, because we'd have to shrink the frame,
1173             ;; esp currently points at the end of the source args.
1174             (inst push ebx-tn)
1175
1176             (emit-label LOOP)
1177             (inst sub ecx-tn n-word-bytes)
1178             (inst mov ebx-tn (make-ea :dword
1179                                       :base esp-tn :index ecx-tn
1180                                       ;; compensate for PUSH above
1181                                       :disp n-word-bytes))
1182             (inst mov (make-ea :dword
1183                                :base esp-tn :index ecx-tn
1184                                ;; compensate for PUSH, and
1185                                ;; add (abs delta)
1186                                :disp (- n-word-bytes delta))
1187                   ebx-tn)
1188             (inst jmp :nz LOOP)
1189
1190             (inst pop ebx-tn))
1191            ((plusp delta)
1192             ;; stack frame larger than fixed. Moving args to lower
1193             ;; addresses, so copy from the lowest address.  esp
1194             ;; already points to the lowest address of the destination.
1195             (inst push ebx-tn)
1196             (inst push esi-tn)
1197
1198             (inst xor ebx-tn ebx-tn)
1199             (emit-label LOOP)
1200             (inst mov esi-tn (make-ea :dword
1201                                       :base esp-tn :index ebx-tn
1202                                       ;; PUSHed 2 words
1203                                       :disp (+ (* 2 n-word-bytes)
1204                                                delta)))
1205             (inst mov (make-ea :dword
1206                                :base esp-tn :index ebx-tn
1207                                :disp (* 2 n-word-bytes))
1208                   esi-tn)
1209             (inst add ebx-tn n-word-bytes)
1210             (inst sub ecx-tn n-word-bytes)
1211             (inst jmp :nz LOOP)
1212
1213             (inst pop esi-tn)
1214             (inst pop ebx-tn))))
1215    DO-REGS
1216    ;; stack can now be set to its final size
1217    (when (< (max 3 (sb-allocated-size 'stack)) fixed)
1218      (inst add esp-tn (* n-word-bytes
1219                          (- fixed
1220                             (max 3 (sb-allocated-size 'stack))))))
1221
1222    ;; Restore ECX
1223    (inst mov ecx-tn ebx-tn)
1224
1225    ;; Here: nargs>=1 && nargs>fixed
1226    (when (< fixed register-arg-count)
1227      ;; Now we have to deposit any more args that showed up in
1228      ;; registers.
1229      (do ((i fixed))
1230          ( nil )
1231        ;; Store it relative to ebp
1232        (inst mov (make-ea :dword :base ebp-tn
1233                           :disp (* n-word-bytes
1234                                    (- sp->fp-offset
1235                                       (+ 1
1236                                          (- i fixed)
1237                                          (max 3 (sb-allocated-size
1238                                                  'stack))))))
1239              (nth i *register-arg-tns*))
1240
1241        (incf i)
1242        (when (>= i register-arg-count)
1243          (return))
1244
1245        ;; Don't deposit any more than there are.
1246        (if (zerop i)
1247            (inst test ecx-tn ecx-tn)
1248            (inst cmp ecx-tn (fixnumize i)))
1249        (inst jmp :eq DONE)))
1250
1251    (inst jmp DONE)
1252
1253    JUST-ALLOC-FRAME
1254    (inst lea esp-tn
1255          (make-ea :dword :base ebp-tn
1256                   :disp (* n-word-bytes
1257                            (- sp->fp-offset
1258                               (max 3 (sb-allocated-size 'stack))))))
1259
1260    DONE))
1261
1262(define-vop (more-kw-arg)
1263  (:translate sb!c::%more-kw-arg)
1264  (:policy :fast-safe)
1265  (:args (object :scs (descriptor-reg) :to (:result 1))
1266         (index :scs (any-reg immediate) :to (:result 1) :target keyword))
1267  (:arg-types * tagged-num)
1268  (:results (value :scs (descriptor-reg any-reg))
1269            (keyword :scs (descriptor-reg any-reg)))
1270  (:result-types * *)
1271  (:generator 4
1272    (sc-case index
1273      (immediate
1274       (inst mov value (make-ea :dword :base object :disp (tn-value index)))
1275       (inst mov keyword (make-ea :dword :base object
1276                                  :disp (+ (tn-value index) n-word-bytes))))
1277      (t
1278       (inst mov value (make-ea :dword :base object :index index))
1279       (inst mov keyword (make-ea :dword :base object :index index
1280                                  :disp n-word-bytes))))))
1281
1282(define-vop (more-arg/c)
1283  (:translate sb!c::%more-arg)
1284  (:policy :fast-safe)
1285  (:args (object :scs (descriptor-reg) :to (:result 1)))
1286  (:info index)
1287  (:arg-types * (:constant (signed-byte 32)))
1288  (:results (value :scs (descriptor-reg any-reg)))
1289  (:result-types *)
1290  (:generator 3
1291    (inst mov value (make-ea :dword :base object
1292                                    :disp (- (* index n-word-bytes))))))
1293
1294(define-vop (more-arg)
1295    (:translate sb!c::%more-arg)
1296  (:policy :fast-safe)
1297  (:args (object :scs (descriptor-reg) :to (:result 1))
1298         (index :scs (any-reg) :to (:result 1) :target value))
1299  (:arg-types * tagged-num)
1300  (:results (value :scs (descriptor-reg any-reg)))
1301  (:result-types *)
1302  (:generator 4
1303    (move value index)
1304    (inst neg value)
1305    (inst mov value (make-ea :dword :base object :index value))))
1306
1307;;; Turn more arg (context, count) into a list.
1308(define-vop (listify-rest-args)
1309  (:translate %listify-rest-args)
1310  (:policy :safe)
1311  (:args (context :scs (descriptor-reg) :target src)
1312         (count :scs (any-reg) :target ecx))
1313  (:arg-types * tagged-num)
1314  (:temporary (:sc unsigned-reg :offset esi-offset :from (:argument 0)) src)
1315  (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
1316  (:temporary (:sc unsigned-reg :offset eax-offset) eax)
1317  (:temporary (:sc unsigned-reg) dst)
1318  (:results (result :scs (descriptor-reg)))
1319  (:node-var node)
1320  (:generator 20
1321    (let ((enter (gen-label))
1322          (loop (gen-label))
1323          (done (gen-label))
1324          (stack-allocate-p (node-stack-allocate-p node)))
1325      (move src context)
1326      (move ecx count)
1327      ;; Check to see whether there are no args, and just return NIL if so.
1328      (inst mov result nil-value)
1329      (inst jecxz done)
1330      (inst lea dst (make-ea :dword :base ecx :index ecx))
1331      (maybe-pseudo-atomic stack-allocate-p
1332       (allocation dst dst node stack-allocate-p list-pointer-lowtag)
1333       ;; Set decrement mode (successive args at lower addresses)
1334       (inst std)
1335       ;; Set up the result.
1336       (move result dst)
1337       ;; Jump into the middle of the loop, 'cause that's where we want
1338       ;; to start.
1339       (inst jmp enter)
1340       (emit-label loop)
1341       ;; Compute a pointer to the next cons.
1342       (inst add dst (* cons-size n-word-bytes))
1343       ;; Store a pointer to this cons in the CDR of the previous cons.
1344       (storew dst dst -1 list-pointer-lowtag)
1345       (emit-label enter)
1346       ;; Grab one value and stash it in the car of this cons.
1347       (inst lods eax)
1348       (storew eax dst 0 list-pointer-lowtag)
1349       ;; Go back for more.
1350       (inst sub ecx n-word-bytes)
1351       (inst jmp :nz loop)
1352       ;; NIL out the last cons.
1353       (storew nil-value dst 1 list-pointer-lowtag)
1354       (inst cld))
1355      (emit-label done))))
1356
1357;;; Return the location and size of the &MORE arg glob created by
1358;;; COPY-MORE-ARG. SUPPLIED is the total number of arguments supplied
1359;;; (originally passed in ECX). FIXED is the number of non-rest
1360;;; arguments.
1361;;;
1362;;; We must duplicate some of the work done by COPY-MORE-ARG, since at
1363;;; that time the environment is in a pretty brain-damaged state,
1364;;; preventing this info from being returned as values. What we do is
1365;;; compute supplied - fixed, and return a pointer that many words
1366;;; below the current stack top.
1367(define-vop (more-arg-context)
1368  (:policy :fast-safe)
1369  (:translate sb!c::%more-arg-context)
1370  (:args (supplied :scs (any-reg) :target count))
1371  (:arg-types positive-fixnum (:constant fixnum))
1372  (:info fixed)
1373  (:results (context :scs (descriptor-reg))
1374            (count :scs (any-reg)))
1375  (:result-types t tagged-num)
1376  (:note "more-arg-context")
1377  (:generator 5
1378    (move count supplied)
1379    ;; SP at this point points at the last arg pushed.
1380    ;; Point to the first more-arg, not above it.
1381    (inst lea context (make-ea :dword :base esp-tn
1382                               :index count :scale 1
1383                               :disp (- (+ (fixnumize fixed) n-word-bytes))))
1384    (unless (zerop fixed)
1385      (inst sub count (fixnumize fixed)))))
1386
1387;;; Signal wrong argument count error if NARGS isn't equal to COUNT.
1388(define-vop (verify-arg-count)
1389  (:policy :fast-safe)
1390  (:args (nargs :scs (any-reg)))
1391  (:arg-types positive-fixnum (:constant t) (:constant t))
1392  (:info min max)
1393  (:vop-var vop)
1394  (:save-p :compute-only)
1395  (:generator 3
1396    (let ((err-lab
1397            (generate-error-code vop 'invalid-arg-count-error nargs)))
1398      (flet ((check-min ()
1399               (cond ((= min 1)
1400                      (inst test nargs nargs)
1401                      (inst jmp :e err-lab))
1402                     ((plusp min)
1403                      (inst cmp nargs (fixnumize min))
1404                      (inst jmp :b err-lab)))))
1405        (cond ((not min)
1406               (if (zerop max)
1407                   (inst test nargs nargs)
1408                   (inst cmp nargs (fixnumize max)))
1409               (inst jmp :ne err-lab))
1410              (max
1411               (check-min)
1412               (inst cmp nargs (fixnumize max))
1413               (inst jmp :a err-lab))
1414              (t
1415               (check-min)))))))
1416
1417;;; Single-stepping
1418(defun emit-single-step-test ()
1419  ;; We use different ways of representing whether stepping is on on
1420  ;; +SB-THREAD / -SB-THREAD: on +SB-THREAD, we use a slot in the
1421  ;; thread structure. On -SB-THREAD we use the value of a static
1422  ;; symbol. Things are done this way, since reading a thread-local
1423  ;; slot from a symbol would require an extra register on +SB-THREAD,
1424  ;; and reading a slot from a thread structure would require an extra
1425  ;; register on -SB-THREAD.
1426  #!+sb-thread
1427  (progn
1428    #!+win32 (inst push eax-tn)
1429    (with-tls-ea (EA :base #!+win32 eax-tn #!-win32 :unused
1430                     :disp-type :constant
1431                     :disp (* thread-stepping-slot n-word-bytes))
1432      (inst cmp EA 0 :maybe-fs))
1433    #!+win32 (inst pop eax-tn))
1434  #!-sb-thread
1435  (inst cmp (make-ea-for-symbol-value sb!impl::*stepping*) 0))
1436
1437(define-vop (step-instrument-before-vop)
1438  (:policy :fast-safe)
1439  (:vop-var vop)
1440  (:generator 3
1441     (emit-single-step-test)
1442     (inst jmp :eq DONE)
1443     (inst break single-step-before-trap)
1444     DONE
1445     (note-this-location vop :step-before-vop)))
1446