1 ;;;; the VM definition of arithmetic VOPs for the x86
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.
16 (define-vop (fast-safe-arith-op)
21 (define-vop (fixnum-unop fast-safe-arith-op)
22 (:args (x :scs (any-reg) :target res))
23 (:results (res :scs (any-reg)))
24 (:note "inline fixnum arithmetic")
25 (:arg-types tagged-num)
26 (:result-types tagged-num))
28 (define-vop (signed-unop fast-safe-arith-op)
29 (:args (x :scs (signed-reg) :target res))
30 (:results (res :scs (signed-reg)))
31 (:note "inline (signed-byte 32) arithmetic")
32 (:arg-types signed-num)
33 (:result-types signed-num))
35 (define-vop (fast-negate/fixnum fixnum-unop)
41 (define-vop (fast-negate/signed signed-unop)
47 (define-vop (fast-lognot/fixnum fixnum-unop)
51 (inst xor res (fixnumize -1))))
53 (define-vop (fast-lognot/signed signed-unop)
59 ;;;; binary fixnum operations
61 ;;; Assume that any constant operand is the second arg...
63 (define-vop (fast-fixnum-binop fast-safe-arith-op)
64 (:args (x :target r :scs (any-reg)
65 :load-if (not (and (sc-is x control-stack)
67 (sc-is r control-stack)
69 (y :scs (any-reg control-stack)))
70 (:arg-types tagged-num tagged-num)
71 (:results (r :scs (any-reg) :from (:argument 0)
72 :load-if (not (and (sc-is x control-stack)
74 (sc-is r control-stack)
76 (:result-types tagged-num)
77 (:note "inline fixnum arithmetic"))
79 (define-vop (fast-unsigned-binop fast-safe-arith-op)
80 (:args (x :target r :scs (unsigned-reg)
81 :load-if (not (and (sc-is x unsigned-stack)
82 (sc-is y unsigned-reg)
83 (sc-is r unsigned-stack)
85 (y :scs (unsigned-reg unsigned-stack)))
86 (:arg-types unsigned-num unsigned-num)
87 (:results (r :scs (unsigned-reg) :from (:argument 0)
88 :load-if (not (and (sc-is x unsigned-stack)
89 (sc-is y unsigned-reg)
90 (sc-is r unsigned-stack)
92 (:result-types unsigned-num)
93 (:note "inline (unsigned-byte 32) arithmetic"))
95 (define-vop (fast-signed-binop fast-safe-arith-op)
96 (:args (x :target r :scs (signed-reg)
97 :load-if (not (and (sc-is x signed-stack)
99 (sc-is r signed-stack)
101 (y :scs (signed-reg signed-stack)))
102 (:arg-types signed-num signed-num)
103 (:results (r :scs (signed-reg) :from (:argument 0)
104 :load-if (not (and (sc-is x signed-stack)
106 (sc-is r signed-stack)
108 (:result-types signed-num)
109 (:note "inline (signed-byte 32) arithmetic"))
111 (define-vop (fast-fixnum-binop-c fast-safe-arith-op)
112 (:args (x :target r :scs (any-reg control-stack)))
114 (:arg-types tagged-num (:constant (signed-byte 30)))
115 (:results (r :scs (any-reg)
116 :load-if (not (location= x r))))
117 (:result-types tagged-num)
118 (:note "inline fixnum arithmetic"))
120 (define-vop (fast-unsigned-binop-c fast-safe-arith-op)
121 (:args (x :target r :scs (unsigned-reg unsigned-stack)))
123 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
124 (:results (r :scs (unsigned-reg)
125 :load-if (not (location= x r))))
126 (:result-types unsigned-num)
127 (:note "inline (unsigned-byte 32) arithmetic"))
129 (define-vop (fast-signed-binop-c fast-safe-arith-op)
130 (:args (x :target r :scs (signed-reg signed-stack)))
132 (:arg-types signed-num (:constant (signed-byte 32)))
133 (:results (r :scs (signed-reg)
134 :load-if (not (location= x r))))
135 (:result-types signed-num)
136 (:note "inline (signed-byte 32) arithmetic"))
138 (macrolet ((define-binop (translate untagged-penalty op)
140 (define-vop (,(symbolicate "FAST-" translate "/FIXNUM=>FIXNUM")
142 (:translate ,translate)
146 (define-vop (,(symbolicate 'fast- translate '-c/fixnum=>fixnum)
148 (:translate ,translate)
151 (inst ,op r (fixnumize y))))
152 (define-vop (,(symbolicate "FAST-" translate "/SIGNED=>SIGNED")
154 (:translate ,translate)
155 (:generator ,(1+ untagged-penalty)
158 (define-vop (,(symbolicate 'fast- translate '-c/signed=>signed)
160 (:translate ,translate)
161 (:generator ,untagged-penalty
164 (define-vop (,(symbolicate "FAST-"
166 "/UNSIGNED=>UNSIGNED")
168 (:translate ,translate)
169 (:generator ,(1+ untagged-penalty)
172 (define-vop (,(symbolicate 'fast-
174 '-c/unsigned=>unsigned)
175 fast-unsigned-binop-c)
176 (:translate ,translate)
177 (:generator ,untagged-penalty
179 ,(if (eq translate 'logand)
180 ;; for the -C/UNSIGNED=>UNSIGNED VOP, this case
181 ;; is optimized away as an identity somewhere
182 ;; along the lines. However, this VOP is used in
183 ;; -C/SIGNED=>UNSIGNED, below, when the
184 ;; higher-level lisp code can't optimize away the
185 ;; non-trivial identity.
186 `(unless (= y #.(1- (ash 1 n-word-bits)))
188 `(inst ,op r y)))))))
189 (define-binop - 4 sub)
190 (define-binop logand 2 and)
191 (define-binop logior 2 or)
192 (define-binop logxor 2 xor))
194 ;;; Special handling of add on the x86; can use lea to avoid a
195 ;;; register load, otherwise it uses add.
196 (define-vop (fast-+/fixnum=>fixnum fast-safe-arith-op)
198 (:args (x :scs (any-reg) :target r
199 :load-if (not (and (sc-is x control-stack)
201 (sc-is r control-stack)
203 (y :scs (any-reg control-stack)))
204 (:arg-types tagged-num tagged-num)
205 (:results (r :scs (any-reg) :from (:argument 0)
206 :load-if (not (and (sc-is x control-stack)
208 (sc-is r control-stack)
210 (:result-types tagged-num)
211 (:note "inline fixnum arithmetic")
213 (cond ((and (sc-is x any-reg) (sc-is y any-reg) (sc-is r any-reg)
214 (not (location= x r)))
215 (inst lea r (make-ea :dword :base x :index y :scale 1)))
220 (define-vop (fast-+-c/fixnum=>fixnum fast-safe-arith-op)
222 (:args (x :target r :scs (any-reg control-stack)))
224 (:arg-types tagged-num (:constant (signed-byte 30)))
225 (:results (r :scs (any-reg)
226 :load-if (not (location= x r))))
227 (:result-types tagged-num)
228 (:note "inline fixnum arithmetic")
230 (cond ((and (sc-is x any-reg) (sc-is r any-reg) (not (location= x r)))
231 (inst lea r (make-ea :dword :base x :disp (fixnumize y))))
234 (inst add r (fixnumize y))))))
236 (define-vop (fast-+/signed=>signed fast-safe-arith-op)
238 (:args (x :scs (signed-reg) :target r
239 :load-if (not (and (sc-is x signed-stack)
241 (sc-is r signed-stack)
243 (y :scs (signed-reg signed-stack)))
244 (:arg-types signed-num signed-num)
245 (:results (r :scs (signed-reg) :from (:argument 0)
246 :load-if (not (and (sc-is x signed-stack)
249 (:result-types signed-num)
250 (:note "inline (signed-byte 32) arithmetic")
252 (cond ((and (sc-is x signed-reg) (sc-is y signed-reg) (sc-is r signed-reg)
253 (not (location= x r)))
254 (inst lea r (make-ea :dword :base x :index y :scale 1)))
259 ;;;; Special logand cases: (logand signed unsigned) => unsigned
261 (define-vop (fast-logand/signed-unsigned=>unsigned
262 fast-logand/unsigned=>unsigned)
263 (:args (x :target r :scs (signed-reg)
264 :load-if (not (and (sc-is x signed-stack)
265 (sc-is y unsigned-reg)
266 (sc-is r unsigned-stack)
268 (y :scs (unsigned-reg unsigned-stack)))
269 (:arg-types signed-num unsigned-num))
271 (define-vop (fast-logand-c/signed-unsigned=>unsigned
272 fast-logand-c/unsigned=>unsigned)
273 (:args (x :target r :scs (signed-reg signed-stack)))
274 (:arg-types signed-num (:constant (unsigned-byte 32))))
276 (define-vop (fast-logand/unsigned-signed=>unsigned
277 fast-logand/unsigned=>unsigned)
278 (:args (x :target r :scs (unsigned-reg)
279 :load-if (not (and (sc-is x unsigned-stack)
281 (sc-is r unsigned-stack)
283 (y :scs (signed-reg signed-stack)))
284 (:arg-types unsigned-num signed-num))
287 (define-vop (fast-+-c/signed=>signed fast-safe-arith-op)
289 (:args (x :target r :scs (signed-reg signed-stack)))
291 (:arg-types signed-num (:constant (signed-byte 32)))
292 (:results (r :scs (signed-reg)
293 :load-if (not (location= x r))))
294 (:result-types signed-num)
295 (:note "inline (signed-byte 32) arithmetic")
297 (cond ((and (sc-is x signed-reg) (sc-is r signed-reg)
298 (not (location= x r)))
299 (inst lea r (make-ea :dword :base x :disp y)))
306 (define-vop (fast-+/unsigned=>unsigned fast-safe-arith-op)
308 (:args (x :scs (unsigned-reg) :target r
309 :load-if (not (and (sc-is x unsigned-stack)
310 (sc-is y unsigned-reg)
311 (sc-is r unsigned-stack)
313 (y :scs (unsigned-reg unsigned-stack)))
314 (:arg-types unsigned-num unsigned-num)
315 (:results (r :scs (unsigned-reg) :from (:argument 0)
316 :load-if (not (and (sc-is x unsigned-stack)
317 (sc-is y unsigned-reg)
318 (sc-is r unsigned-stack)
320 (:result-types unsigned-num)
321 (:note "inline (unsigned-byte 32) arithmetic")
323 (cond ((and (sc-is x unsigned-reg) (sc-is y unsigned-reg)
324 (sc-is r unsigned-reg) (not (location= x r)))
325 (inst lea r (make-ea :dword :base x :index y :scale 1)))
330 (define-vop (fast-+-c/unsigned=>unsigned fast-safe-arith-op)
332 (:args (x :target r :scs (unsigned-reg unsigned-stack)))
334 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
335 (:results (r :scs (unsigned-reg)
336 :load-if (not (location= x r))))
337 (:result-types unsigned-num)
338 (:note "inline (unsigned-byte 32) arithmetic")
340 (cond ((and (sc-is x unsigned-reg) (sc-is r unsigned-reg)
341 (not (location= x r)))
342 (inst lea r (make-ea :dword :base x :disp y)))
349 ;;;; multiplication and division
351 (define-vop (fast-*/fixnum=>fixnum fast-safe-arith-op)
353 ;; We need different loading characteristics.
354 (:args (x :scs (any-reg) :target r)
355 (y :scs (any-reg control-stack)))
356 (:arg-types tagged-num tagged-num)
357 (:results (r :scs (any-reg) :from (:argument 0)))
358 (:result-types tagged-num)
359 (:note "inline fixnum arithmetic")
365 (define-vop (fast-*-c/fixnum=>fixnum fast-safe-arith-op)
367 ;; We need different loading characteristics.
368 (:args (x :scs (any-reg control-stack)))
370 (:arg-types tagged-num (:constant (signed-byte 30)))
371 (:results (r :scs (any-reg)))
372 (:result-types tagged-num)
373 (:note "inline fixnum arithmetic")
377 (define-vop (fast-*/signed=>signed fast-safe-arith-op)
379 ;; We need different loading characteristics.
380 (:args (x :scs (signed-reg) :target r)
381 (y :scs (signed-reg signed-stack)))
382 (:arg-types signed-num signed-num)
383 (:results (r :scs (signed-reg) :from (:argument 0)))
384 (:result-types signed-num)
385 (:note "inline (signed-byte 32) arithmetic")
390 (define-vop (fast-*-c/signed=>signed fast-safe-arith-op)
392 ;; We need different loading characteristics.
393 (:args (x :scs (signed-reg signed-stack)))
395 (:arg-types signed-num (:constant (signed-byte 32)))
396 (:results (r :scs (signed-reg)))
397 (:result-types signed-num)
398 (:note "inline (signed-byte 32) arithmetic")
402 (define-vop (fast-*/unsigned=>unsigned fast-safe-arith-op)
404 (:args (x :scs (unsigned-reg) :target eax)
405 (y :scs (unsigned-reg unsigned-stack)))
406 (:arg-types unsigned-num unsigned-num)
407 (:temporary (:sc unsigned-reg :offset eax-offset :target r
408 :from (:argument 0) :to :result) eax)
409 (:temporary (:sc unsigned-reg :offset edx-offset
410 :from :eval :to :result) edx)
412 (:results (r :scs (unsigned-reg)))
413 (:result-types unsigned-num)
414 (:note "inline (unsigned-byte 32) arithmetic")
416 (:save-p :compute-only)
423 (define-vop (fast-truncate/fixnum=>fixnum fast-safe-arith-op)
424 (:translate truncate)
425 (:args (x :scs (any-reg) :target eax)
426 (y :scs (any-reg control-stack)))
427 (:arg-types tagged-num tagged-num)
428 (:temporary (:sc signed-reg :offset eax-offset :target quo
429 :from (:argument 0) :to (:result 0)) eax)
430 (:temporary (:sc unsigned-reg :offset edx-offset :target rem
431 :from (:argument 0) :to (:result 1)) edx)
432 (:results (quo :scs (any-reg))
433 (rem :scs (any-reg)))
434 (:result-types tagged-num tagged-num)
435 (:note "inline fixnum arithmetic")
437 (:save-p :compute-only)
439 (let ((zero (generate-error-code vop division-by-zero-error x y)))
440 (if (sc-is y any-reg)
441 (inst test y y) ; smaller instruction
447 (if (location= quo eax)
449 (inst lea quo (make-ea :dword :index eax :scale 4)))
452 (define-vop (fast-truncate-c/fixnum=>fixnum fast-safe-arith-op)
453 (:translate truncate)
454 (:args (x :scs (any-reg) :target eax))
456 (:arg-types tagged-num (:constant (signed-byte 30)))
457 (:temporary (:sc signed-reg :offset eax-offset :target quo
458 :from :argument :to (:result 0)) eax)
459 (:temporary (:sc any-reg :offset edx-offset :target rem
460 :from :eval :to (:result 1)) edx)
461 (:temporary (:sc any-reg :from :eval :to :result) y-arg)
462 (:results (quo :scs (any-reg))
463 (rem :scs (any-reg)))
464 (:result-types tagged-num tagged-num)
465 (:note "inline fixnum arithmetic")
467 (:save-p :compute-only)
471 (inst mov y-arg (fixnumize y))
472 (inst idiv eax y-arg)
473 (if (location= quo eax)
475 (inst lea quo (make-ea :dword :index eax :scale 4)))
478 (define-vop (fast-truncate/unsigned=>unsigned fast-safe-arith-op)
479 (:translate truncate)
480 (:args (x :scs (unsigned-reg) :target eax)
481 (y :scs (unsigned-reg signed-stack)))
482 (:arg-types unsigned-num unsigned-num)
483 (:temporary (:sc unsigned-reg :offset eax-offset :target quo
484 :from (:argument 0) :to (:result 0)) eax)
485 (:temporary (:sc unsigned-reg :offset edx-offset :target rem
486 :from (:argument 0) :to (:result 1)) edx)
487 (:results (quo :scs (unsigned-reg))
488 (rem :scs (unsigned-reg)))
489 (:result-types unsigned-num unsigned-num)
490 (:note "inline (unsigned-byte 32) arithmetic")
492 (:save-p :compute-only)
494 (let ((zero (generate-error-code vop division-by-zero-error x y)))
495 (if (sc-is y unsigned-reg)
496 (inst test y y) ; smaller instruction
505 (define-vop (fast-truncate-c/unsigned=>unsigned fast-safe-arith-op)
506 (:translate truncate)
507 (:args (x :scs (unsigned-reg) :target eax))
509 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
510 (:temporary (:sc unsigned-reg :offset eax-offset :target quo
511 :from :argument :to (:result 0)) eax)
512 (:temporary (:sc unsigned-reg :offset edx-offset :target rem
513 :from :eval :to (:result 1)) edx)
514 (:temporary (:sc unsigned-reg :from :eval :to :result) y-arg)
515 (:results (quo :scs (unsigned-reg))
516 (rem :scs (unsigned-reg)))
517 (:result-types unsigned-num unsigned-num)
518 (:note "inline (unsigned-byte 32) arithmetic")
520 (:save-p :compute-only)
529 (define-vop (fast-truncate/signed=>signed fast-safe-arith-op)
530 (:translate truncate)
531 (:args (x :scs (signed-reg) :target eax)
532 (y :scs (signed-reg signed-stack)))
533 (:arg-types signed-num signed-num)
534 (:temporary (:sc signed-reg :offset eax-offset :target quo
535 :from (:argument 0) :to (:result 0)) eax)
536 (:temporary (:sc signed-reg :offset edx-offset :target rem
537 :from (:argument 0) :to (:result 1)) edx)
538 (:results (quo :scs (signed-reg))
539 (rem :scs (signed-reg)))
540 (:result-types signed-num signed-num)
541 (:note "inline (signed-byte 32) arithmetic")
543 (:save-p :compute-only)
545 (let ((zero (generate-error-code vop division-by-zero-error x y)))
546 (if (sc-is y signed-reg)
547 (inst test y y) ; smaller instruction
556 (define-vop (fast-truncate-c/signed=>signed fast-safe-arith-op)
557 (:translate truncate)
558 (:args (x :scs (signed-reg) :target eax))
560 (:arg-types signed-num (:constant (signed-byte 32)))
561 (:temporary (:sc signed-reg :offset eax-offset :target quo
562 :from :argument :to (:result 0)) eax)
563 (:temporary (:sc signed-reg :offset edx-offset :target rem
564 :from :eval :to (:result 1)) edx)
565 (:temporary (:sc signed-reg :from :eval :to :result) y-arg)
566 (:results (quo :scs (signed-reg))
567 (rem :scs (signed-reg)))
568 (:result-types signed-num signed-num)
569 (:note "inline (signed-byte 32) arithmetic")
571 (:save-p :compute-only)
576 (inst idiv eax y-arg)
583 (define-vop (fast-ash-c/fixnum=>fixnum)
586 (:args (number :scs (any-reg) :target result
587 :load-if (not (and (sc-is number any-reg control-stack)
588 (sc-is result any-reg control-stack)
589 (location= number result)))))
591 (:arg-types tagged-num (:constant integer))
592 (:results (result :scs (any-reg)
593 :load-if (not (and (sc-is number control-stack)
594 (sc-is result control-stack)
595 (location= number result)))))
596 (:result-types tagged-num)
599 (cond ((and (= amount 1) (not (location= number result)))
600 (inst lea result (make-ea :dword :index number :scale 2)))
601 ((and (= amount 2) (not (location= number result)))
602 (inst lea result (make-ea :dword :index number :scale 4)))
603 ((and (= amount 3) (not (location= number result)))
604 (inst lea result (make-ea :dword :index number :scale 8)))
607 (cond ((plusp amount)
608 ;; We don't have to worry about overflow because of the
609 ;; result type restriction.
610 (inst shl result amount))
612 ;; If the amount is greater than 31, only shift by 31. We
613 ;; have to do this because the shift instructions only look
614 ;; at the low five bits of the result.
615 (inst sar result (min 31 (- amount)))
616 ;; Fixnum correction.
617 (inst and result #xfffffffc)))))))
619 (define-vop (fast-ash-left/fixnum=>fixnum)
621 (:args (number :scs (any-reg) :target result
622 :load-if (not (and (sc-is number control-stack)
623 (sc-is result control-stack)
624 (location= number result))))
625 (amount :scs (unsigned-reg) :target ecx))
626 (:arg-types tagged-num positive-fixnum)
627 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
628 (:results (result :scs (any-reg) :from (:argument 0)
629 :load-if (not (and (sc-is number control-stack)
630 (sc-is result control-stack)
631 (location= number result)))))
632 (:result-types tagged-num)
638 ;; The result-type ensures us that this shift will not overflow.
639 (inst shl result :cl)))
641 (define-vop (fast-ash-c/signed=>signed)
644 (:args (number :scs (signed-reg) :target result
645 :load-if (not (and (sc-is number signed-stack)
646 (sc-is result signed-stack)
647 (location= number result)))))
649 (:arg-types signed-num (:constant integer))
650 (:results (result :scs (signed-reg)
651 :load-if (not (and (sc-is number signed-stack)
652 (sc-is result signed-stack)
653 (location= number result)))))
654 (:result-types signed-num)
657 (cond ((and (= amount 1) (not (location= number result)))
658 (inst lea result (make-ea :dword :index number :scale 2)))
659 ((and (= amount 2) (not (location= number result)))
660 (inst lea result (make-ea :dword :index number :scale 4)))
661 ((and (= amount 3) (not (location= number result)))
662 (inst lea result (make-ea :dword :index number :scale 8)))
665 (cond ((plusp amount) (inst shl result amount))
666 (t (inst sar result (min 31 (- amount)))))))))
668 (define-vop (fast-ash-c/unsigned=>unsigned)
671 (:args (number :scs (unsigned-reg) :target result
672 :load-if (not (and (sc-is number unsigned-stack)
673 (sc-is result unsigned-stack)
674 (location= number result)))))
676 (:arg-types unsigned-num (:constant integer))
677 (:results (result :scs (unsigned-reg)
678 :load-if (not (and (sc-is number unsigned-stack)
679 (sc-is result unsigned-stack)
680 (location= number result)))))
681 (:result-types unsigned-num)
684 (cond ((and (= amount 1) (not (location= number result)))
685 (inst lea result (make-ea :dword :index number :scale 2)))
686 ((and (= amount 2) (not (location= number result)))
687 (inst lea result (make-ea :dword :index number :scale 4)))
688 ((and (= amount 3) (not (location= number result)))
689 (inst lea result (make-ea :dword :index number :scale 8)))
692 (cond ((< -32 amount 32)
693 ;; this code is used both in ASH and ASH-MOD32, so
696 (inst shl result amount)
697 (inst shr result (- amount))))
698 (t (if (sc-is result unsigned-reg)
699 (inst xor result result)
700 (inst mov result 0))))))))
702 (define-vop (fast-ash-left/signed=>signed)
704 (:args (number :scs (signed-reg) :target result
705 :load-if (not (and (sc-is number signed-stack)
706 (sc-is result signed-stack)
707 (location= number result))))
708 (amount :scs (unsigned-reg) :target ecx))
709 (:arg-types signed-num positive-fixnum)
710 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
711 (:results (result :scs (signed-reg) :from (:argument 0)
712 :load-if (not (and (sc-is number signed-stack)
713 (sc-is result signed-stack)
714 (location= number result)))))
715 (:result-types signed-num)
721 (inst shl result :cl)))
723 (define-vop (fast-ash-left/unsigned=>unsigned)
725 (:args (number :scs (unsigned-reg) :target result
726 :load-if (not (and (sc-is number unsigned-stack)
727 (sc-is result unsigned-stack)
728 (location= number result))))
729 (amount :scs (unsigned-reg) :target ecx))
730 (:arg-types unsigned-num positive-fixnum)
731 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
732 (:results (result :scs (unsigned-reg) :from (:argument 0)
733 :load-if (not (and (sc-is number unsigned-stack)
734 (sc-is result unsigned-stack)
735 (location= number result)))))
736 (:result-types unsigned-num)
742 (inst shl result :cl)))
744 (define-vop (fast-ash/signed=>signed)
747 (:args (number :scs (signed-reg) :target result)
748 (amount :scs (signed-reg) :target ecx))
749 (:arg-types signed-num signed-num)
750 (:results (result :scs (signed-reg) :from (:argument 0)))
751 (:result-types signed-num)
752 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
758 (inst jmp :ns positive)
764 (inst sar result :cl)
768 ;; The result-type ensures us that this shift will not overflow.
769 (inst shl result :cl)
773 (define-vop (fast-ash/unsigned=>unsigned)
776 (:args (number :scs (unsigned-reg) :target result)
777 (amount :scs (signed-reg) :target ecx))
778 (:arg-types unsigned-num signed-num)
779 (:results (result :scs (unsigned-reg) :from (:argument 0)))
780 (:result-types unsigned-num)
781 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
787 (inst jmp :ns positive)
791 (inst xor result result)
794 (inst shr result :cl)
798 ;; The result-type ensures us that this shift will not overflow.
799 (inst shl result :cl)
805 (defknown %lea (integer integer (member 1 2 4 8) (signed-byte 32))
807 (foldable flushable movable))
809 (defoptimizer (%lea derive-type) ((base index scale disp))
810 (when (and (constant-lvar-p scale)
811 (constant-lvar-p disp))
812 (let ((scale (lvar-value scale))
813 (disp (lvar-value disp))
814 (base-type (lvar-type base))
815 (index-type (lvar-type index)))
816 (when (and (numeric-type-p base-type)
817 (numeric-type-p index-type))
818 (let ((base-lo (numeric-type-low base-type))
819 (base-hi (numeric-type-high base-type))
820 (index-lo (numeric-type-low index-type))
821 (index-hi (numeric-type-high index-type)))
822 (make-numeric-type :class 'integer
824 :low (when (and base-lo index-lo)
825 (+ base-lo (* index-lo scale) disp))
826 :high (when (and base-hi index-hi)
827 (+ base-hi (* index-hi scale) disp))))))))
829 (defun %lea (base index scale disp)
830 (+ base (* index scale) disp))
834 (define-vop (%lea/unsigned=>unsigned)
837 (:args (base :scs (unsigned-reg))
838 (index :scs (unsigned-reg)))
840 (:arg-types unsigned-num unsigned-num
841 (:constant (member 1 2 4 8))
842 (:constant (signed-byte 32)))
843 (:results (r :scs (unsigned-reg)))
844 (:result-types unsigned-num)
846 (inst lea r (make-ea :dword :base base :index index
847 :scale scale :disp disp))))
849 (define-vop (%lea/signed=>signed)
852 (:args (base :scs (signed-reg))
853 (index :scs (signed-reg)))
855 (:arg-types signed-num signed-num
856 (:constant (member 1 2 4 8))
857 (:constant (signed-byte 32)))
858 (:results (r :scs (signed-reg)))
859 (:result-types signed-num)
861 (inst lea r (make-ea :dword :base base :index index
862 :scale scale :disp disp))))
864 (define-vop (%lea/fixnum=>fixnum)
867 (:args (base :scs (any-reg))
868 (index :scs (any-reg)))
870 (:arg-types tagged-num tagged-num
871 (:constant (member 1 2 4 8))
872 (:constant (signed-byte 32)))
873 (:results (r :scs (any-reg)))
874 (:result-types tagged-num)
876 (inst lea r (make-ea :dword :base base :index index
877 :scale scale :disp disp))))
879 ;;; FIXME: before making knowledge of this too public, it needs to be
880 ;;; fixed so that it's actually _faster_ than the non-CMOV version; at
881 ;;; least on my Celeron-XXX laptop, this version is marginally slower
882 ;;; than the above version with branches. -- CSR, 2003-09-04
883 (define-vop (fast-cmov-ash/unsigned=>unsigned)
886 (:args (number :scs (unsigned-reg) :target result)
887 (amount :scs (signed-reg) :target ecx))
888 (:arg-types unsigned-num signed-num)
889 (:results (result :scs (unsigned-reg) :from (:argument 0)))
890 (:result-types unsigned-num)
891 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
892 (:temporary (:sc any-reg :from (:eval 0) :to (:eval 1)) zero)
894 (:guard (member :cmov *backend-subfeatures*))
899 (inst jmp :ns positive)
902 (inst shr result :cl)
904 (inst cmov :nbe result zero)
908 ;; The result-type ensures us that this shift will not overflow.
909 (inst shl result :cl)
913 (define-vop (signed-byte-32-len)
914 (:translate integer-length)
915 (:note "inline (signed-byte 32) integer-length")
917 (:args (arg :scs (signed-reg) :target res))
918 (:arg-types signed-num)
919 (:results (res :scs (unsigned-reg)))
920 (:result-types unsigned-num)
935 (define-vop (unsigned-byte-32-len)
936 (:translate integer-length)
937 (:note "inline (unsigned-byte 32) integer-length")
939 (:args (arg :scs (unsigned-reg)))
940 (:arg-types unsigned-num)
941 (:results (res :scs (unsigned-reg)))
942 (:result-types unsigned-num)
952 (define-vop (unsigned-byte-32-count)
953 (:translate logcount)
954 (:note "inline (unsigned-byte 32) logcount")
956 (:args (arg :scs (unsigned-reg)))
957 (:arg-types unsigned-num)
958 (:results (result :scs (unsigned-reg)))
959 (:result-types positive-fixnum)
960 (:temporary (:sc unsigned-reg :from (:argument 0)) temp)
964 (inst mov temp result)
966 (inst and result #x55555555)
967 (inst and temp #x55555555)
968 (inst add result temp)
970 (inst mov temp result)
972 (inst and result #x33333333)
973 (inst and temp #x33333333)
974 (inst add result temp)
976 (inst mov temp result)
978 (inst and result #x0f0f0f0f)
979 (inst and temp #x0f0f0f0f)
980 (inst add result temp)
982 (inst mov temp result)
984 (inst and result #x00ff00ff)
985 (inst and temp #x00ff00ff)
986 (inst add result temp)
988 (inst mov temp result)
990 (inst and result #x0000ffff)
991 (inst and temp #x0000ffff)
992 (inst add result temp)))
994 ;;;; binary conditional VOPs
996 (define-vop (fast-conditional)
1001 (:policy :fast-safe))
1003 (define-vop (fast-conditional/fixnum fast-conditional)
1004 (:args (x :scs (any-reg)
1005 :load-if (not (and (sc-is x control-stack)
1006 (sc-is y any-reg))))
1007 (y :scs (any-reg control-stack)))
1008 (:arg-types tagged-num tagged-num)
1009 (:note "inline fixnum comparison"))
1011 (define-vop (fast-conditional-c/fixnum fast-conditional/fixnum)
1012 (:args (x :scs (any-reg control-stack)))
1013 (:arg-types tagged-num (:constant (signed-byte 30)))
1014 (:info target not-p y))
1016 (define-vop (fast-conditional/signed fast-conditional)
1017 (:args (x :scs (signed-reg)
1018 :load-if (not (and (sc-is x signed-stack)
1019 (sc-is y signed-reg))))
1020 (y :scs (signed-reg signed-stack)))
1021 (:arg-types signed-num signed-num)
1022 (:note "inline (signed-byte 32) comparison"))
1024 (define-vop (fast-conditional-c/signed fast-conditional/signed)
1025 (:args (x :scs (signed-reg signed-stack)))
1026 (:arg-types signed-num (:constant (signed-byte 32)))
1027 (:info target not-p y))
1029 (define-vop (fast-conditional/unsigned fast-conditional)
1030 (:args (x :scs (unsigned-reg)
1031 :load-if (not (and (sc-is x unsigned-stack)
1032 (sc-is y unsigned-reg))))
1033 (y :scs (unsigned-reg unsigned-stack)))
1034 (:arg-types unsigned-num unsigned-num)
1035 (:note "inline (unsigned-byte 32) comparison"))
1037 (define-vop (fast-conditional-c/unsigned fast-conditional/unsigned)
1038 (:args (x :scs (unsigned-reg unsigned-stack)))
1039 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
1040 (:info target not-p y))
1043 (macrolet ((define-conditional-vop (tran cond unsigned not-cond not-unsigned)
1046 (lambda (suffix cost signed)
1047 `(define-vop (;; FIXME: These could be done more
1048 ;; cleanly with SYMBOLICATE.
1049 ,(intern (format nil "~:@(FAST-IF-~A~A~)"
1052 (format nil "~:@(FAST-CONDITIONAL~A~)"
1057 ,(if (eq suffix '-c/fixnum)
1068 '(/fixnum -c/fixnum /signed -c/signed /unsigned -c/unsigned)
1070 '(t t t t nil nil)))))
1072 (define-conditional-vop < :l :b :ge :ae)
1073 (define-conditional-vop > :g :a :le :be))
1075 (define-vop (fast-if-eql/signed fast-conditional/signed)
1079 (inst jmp (if not-p :ne :e) target)))
1081 (define-vop (fast-if-eql-c/signed fast-conditional-c/signed)
1084 (cond ((and (sc-is x signed-reg) (zerop y))
1085 (inst test x x)) ; smaller instruction
1088 (inst jmp (if not-p :ne :e) target)))
1090 (define-vop (fast-if-eql/unsigned fast-conditional/unsigned)
1094 (inst jmp (if not-p :ne :e) target)))
1096 (define-vop (fast-if-eql-c/unsigned fast-conditional-c/unsigned)
1099 (cond ((and (sc-is x unsigned-reg) (zerop y))
1100 (inst test x x)) ; smaller instruction
1103 (inst jmp (if not-p :ne :e) target)))
1105 ;;; EQL/FIXNUM is funny because the first arg can be of any type, not just a
1108 ;;; These versions specify a fixnum restriction on their first arg. We have
1109 ;;; also generic-eql/fixnum VOPs which are the same, but have no restriction on
1110 ;;; the first arg and a higher cost. The reason for doing this is to prevent
1111 ;;; fixnum specific operations from being used on word integers, spuriously
1112 ;;; consing the argument.
1114 (define-vop (fast-eql/fixnum fast-conditional)
1115 (:args (x :scs (any-reg)
1116 :load-if (not (and (sc-is x control-stack)
1117 (sc-is y any-reg))))
1118 (y :scs (any-reg control-stack)))
1119 (:arg-types tagged-num tagged-num)
1120 (:note "inline fixnum comparison")
1124 (inst jmp (if not-p :ne :e) target)))
1125 (define-vop (generic-eql/fixnum fast-eql/fixnum)
1126 (:args (x :scs (any-reg descriptor-reg)
1127 :load-if (not (and (sc-is x control-stack)
1128 (sc-is y any-reg))))
1129 (y :scs (any-reg control-stack)))
1130 (:arg-types * tagged-num)
1133 (define-vop (fast-eql-c/fixnum fast-conditional/fixnum)
1134 (:args (x :scs (any-reg control-stack)))
1135 (:arg-types tagged-num (:constant (signed-byte 30)))
1136 (:info target not-p y)
1139 (cond ((and (sc-is x any-reg) (zerop y))
1140 (inst test x x)) ; smaller instruction
1142 (inst cmp x (fixnumize y))))
1143 (inst jmp (if not-p :ne :e) target)))
1144 (define-vop (generic-eql-c/fixnum fast-eql-c/fixnum)
1145 (:args (x :scs (any-reg descriptor-reg control-stack)))
1146 (:arg-types * (:constant (signed-byte 30)))
1149 ;;;; 32-bit logical operations
1151 (define-vop (merge-bits)
1152 (:translate merge-bits)
1153 (:args (shift :scs (signed-reg unsigned-reg) :target ecx)
1154 (prev :scs (unsigned-reg) :target result)
1155 (next :scs (unsigned-reg)))
1156 (:arg-types tagged-num unsigned-num unsigned-num)
1157 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 0)) ecx)
1158 (:results (result :scs (unsigned-reg) :from (:argument 1)))
1159 (:result-types unsigned-num)
1160 (:policy :fast-safe)
1164 (inst shrd result next :cl)))
1166 ;;; Only the lower 5 bits of the shift amount are significant.
1167 (define-vop (shift-towards-someplace)
1168 (:policy :fast-safe)
1169 (:args (num :scs (unsigned-reg) :target r)
1170 (amount :scs (signed-reg) :target ecx))
1171 (:arg-types unsigned-num tagged-num)
1172 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
1173 (:results (r :scs (unsigned-reg) :from (:argument 0)))
1174 (:result-types unsigned-num))
1176 (define-vop (shift-towards-start shift-towards-someplace)
1177 (:translate shift-towards-start)
1178 (:note "SHIFT-TOWARDS-START")
1184 (define-vop (shift-towards-end shift-towards-someplace)
1185 (:translate shift-towards-end)
1186 (:note "SHIFT-TOWARDS-END")
1192 ;;;; Modular functions
1193 (defmacro define-mod-binop ((name prototype) function)
1194 `(define-vop (,name ,prototype)
1195 (:args (x :target r :scs (unsigned-reg signed-reg)
1196 :load-if (not (and (or (sc-is x unsigned-stack)
1197 (sc-is x signed-stack))
1198 (or (sc-is y unsigned-reg)
1199 (sc-is y signed-reg))
1200 (or (sc-is r unsigned-stack)
1201 (sc-is r signed-stack))
1203 (y :scs (unsigned-reg signed-reg unsigned-stack signed-stack)))
1204 (:arg-types untagged-num untagged-num)
1205 (:results (r :scs (unsigned-reg signed-reg) :from (:argument 0)
1206 :load-if (not (and (or (sc-is x unsigned-stack)
1207 (sc-is x signed-stack))
1208 (or (sc-is y unsigned-reg)
1209 (sc-is y unsigned-reg))
1210 (or (sc-is r unsigned-stack)
1211 (sc-is r unsigned-stack))
1213 (:result-types unsigned-num)
1214 (:translate ,function)))
1215 (defmacro define-mod-binop-c ((name prototype) function)
1216 `(define-vop (,name ,prototype)
1217 (:args (x :target r :scs (unsigned-reg signed-reg)
1218 :load-if (not (and (or (sc-is x unsigned-stack)
1219 (sc-is x signed-stack))
1220 (or (sc-is r unsigned-stack)
1221 (sc-is r signed-stack))
1224 (:arg-types untagged-num (:constant (or (unsigned-byte 32) (signed-byte 32))))
1225 (:results (r :scs (unsigned-reg signed-reg) :from (:argument 0)
1226 :load-if (not (and (or (sc-is x unsigned-stack)
1227 (sc-is x signed-stack))
1228 (or (sc-is r unsigned-stack)
1229 (sc-is r unsigned-stack))
1231 (:result-types unsigned-num)
1232 (:translate ,function)))
1234 (macrolet ((def (name -c-p)
1235 (let ((fun32 (intern (format nil "~S-MOD32" name)))
1236 (vopu (intern (format nil "FAST-~S/UNSIGNED=>UNSIGNED" name)))
1237 (vopcu (intern (format nil "FAST-~S-C/UNSIGNED=>UNSIGNED" name)))
1238 (vopf (intern (format nil "FAST-~S/FIXNUM=>FIXNUM" name)))
1239 (vopcf (intern (format nil "FAST-~S-C/FIXNUM=>FIXNUM" name)))
1240 (vop32u (intern (format nil "FAST-~S-MOD32/WORD=>UNSIGNED" name)))
1241 (vop32f (intern (format nil "FAST-~S-MOD32/FIXNUM=>FIXNUM" name)))
1242 (vop32cu (intern (format nil "FAST-~S-MOD32-C/WORD=>UNSIGNED" name)))
1243 (vop32cf (intern (format nil "FAST-~S-MOD32-C/FIXNUM=>FIXNUM" name)))
1244 (sfun30 (intern (format nil "~S-SMOD30" name)))
1245 (svop30f (intern (format nil "FAST-~S-SMOD30/FIXNUM=>FIXNUM" name)))
1246 (svop30cf (intern (format nil "FAST-~S-SMOD30-C/FIXNUM=>FIXNUM" name))))
1248 (define-modular-fun ,fun32 (x y) ,name :unsigned 32)
1249 (define-modular-fun ,sfun30 (x y) ,name :signed 30)
1250 (define-mod-binop (,vop32u ,vopu) ,fun32)
1251 (define-vop (,vop32f ,vopf) (:translate ,fun32))
1252 (define-vop (,svop30f ,vopf) (:translate ,sfun30))
1254 `((define-mod-binop-c (,vop32cu ,vopcu) ,fun32)
1255 (define-vop (,svop30cf ,vopcf) (:translate ,sfun30))))))))
1258 ;; (no -C variant as x86 MUL instruction doesn't take an immediate)
1262 (define-vop (fast-ash-left-mod32-c/unsigned=>unsigned
1263 fast-ash-c/unsigned=>unsigned)
1264 (:translate ash-left-mod32))
1266 (define-vop (fast-ash-left-mod32/unsigned=>unsigned
1267 fast-ash-left/unsigned=>unsigned))
1268 (deftransform ash-left-mod32 ((integer count)
1269 ((unsigned-byte 32) (unsigned-byte 5)))
1270 (when (sb!c::constant-lvar-p count)
1271 (sb!c::give-up-ir1-transform))
1272 '(%primitive fast-ash-left-mod32/unsigned=>unsigned integer count))
1274 (define-vop (fast-ash-left-smod30-c/fixnum=>fixnum
1275 fast-ash-c/fixnum=>fixnum)
1276 (:translate ash-left-smod30))
1278 (define-vop (fast-ash-left-smod30/fixnum=>fixnum
1279 fast-ash-left/fixnum=>fixnum))
1280 (deftransform ash-left-smod30 ((integer count)
1281 ((signed-byte 30) (unsigned-byte 5)))
1282 (when (sb!c::constant-lvar-p count)
1283 (sb!c::give-up-ir1-transform))
1284 '(%primitive fast-ash-left-smod30/fixnum=>fixnum integer count))
1288 (defknown sb!vm::%lea-mod32 (integer integer (member 1 2 4 8) (signed-byte 32))
1290 (foldable flushable movable))
1291 (defknown sb!vm::%lea-smod30 (integer integer (member 1 2 4 8) (signed-byte 32))
1293 (foldable flushable movable))
1295 (define-modular-fun-optimizer %lea ((base index scale disp) :unsigned :width width)
1296 (when (and (<= width 32)
1297 (constant-lvar-p scale)
1298 (constant-lvar-p disp))
1299 (cut-to-width base :unsigned width)
1300 (cut-to-width index :unsigned width)
1301 'sb!vm::%lea-mod32))
1302 (define-modular-fun-optimizer %lea ((base index scale disp) :signed :width width)
1303 (when (and (<= width 30)
1304 (constant-lvar-p scale)
1305 (constant-lvar-p disp))
1306 (cut-to-width base :signed width)
1307 (cut-to-width index :signed width)
1308 'sb!vm::%lea-smod30))
1312 (defun sb!vm::%lea-mod32 (base index scale disp)
1313 (ldb (byte 32 0) (%lea base index scale disp)))
1314 (defun sb!vm::%lea-smod30 (base index scale disp)
1315 (mask-signed-field 30 (%lea base index scale disp))))
1318 (defun sb!vm::%lea-mod32 (base index scale disp)
1319 (let ((base (logand base #xffffffff))
1320 (index (logand index #xffffffff)))
1321 ;; can't use modular version of %LEA, as we only have VOPs for
1322 ;; constant SCALE and DISP.
1323 (ldb (byte 32 0) (+ base (* index scale) disp))))
1324 (defun sb!vm::%lea-smod30 (base index scale disp)
1325 (let ((base (mask-signed-field 30 base))
1326 (index (mask-signed-field 30 index)))
1327 ;; can't use modular version of %LEA, as we only have VOPs for
1328 ;; constant SCALE and DISP.
1329 (mask-signed-field 30 (+ base (* index scale) disp)))))
1331 (in-package "SB!VM")
1333 (define-vop (%lea-mod32/unsigned=>unsigned
1334 %lea/unsigned=>unsigned)
1335 (:translate %lea-mod32))
1336 (define-vop (%lea-smod30/fixnum=>fixnum
1337 %lea/fixnum=>fixnum)
1338 (:translate %lea-smod30))
1340 ;;; logical operations
1341 (define-modular-fun lognot-mod32 (x) lognot :unsigned 32)
1342 (define-vop (lognot-mod32/word=>unsigned)
1343 (:translate lognot-mod32)
1344 (:args (x :scs (unsigned-reg signed-reg unsigned-stack signed-stack) :target r
1345 :load-if (not (and (or (sc-is x unsigned-stack)
1346 (sc-is x signed-stack))
1347 (or (sc-is r unsigned-stack)
1348 (sc-is r signed-stack))
1350 (:arg-types unsigned-num)
1351 (:results (r :scs (unsigned-reg)
1352 :load-if (not (and (or (sc-is x unsigned-stack)
1353 (sc-is x signed-stack))
1354 (or (sc-is r unsigned-stack)
1355 (sc-is r signed-stack))
1356 (sc-is r unsigned-stack)
1358 (:result-types unsigned-num)
1359 (:policy :fast-safe)
1364 (define-modular-fun logxor-mod32 (x y) logxor :unsigned 32)
1365 (define-mod-binop (fast-logxor-mod32/word=>unsigned
1366 fast-logxor/unsigned=>unsigned)
1368 (define-mod-binop-c (fast-logxor-mod32-c/word=>unsigned
1369 fast-logxor-c/unsigned=>unsigned)
1371 (define-vop (fast-logxor-mod32/fixnum=>fixnum
1372 fast-logxor/fixnum=>fixnum)
1373 (:translate logxor-mod32))
1374 (define-vop (fast-logxor-mod32-c/fixnum=>fixnum
1375 fast-logxor-c/fixnum=>fixnum)
1376 (:translate logxor-mod32))
1378 (define-source-transform logeqv (&rest args)
1379 (if (oddp (length args))
1381 `(lognot (logxor ,@args))))
1382 (define-source-transform logandc1 (x y)
1383 `(logand (lognot ,x) ,y))
1384 (define-source-transform logandc2 (x y)
1385 `(logand ,x (lognot ,y)))
1386 (define-source-transform logorc1 (x y)
1387 `(logior (lognot ,x) ,y))
1388 (define-source-transform logorc2 (x y)
1389 `(logior ,x (lognot ,y)))
1390 (define-source-transform lognor (x y)
1391 `(lognot (logior ,x ,y)))
1392 (define-source-transform lognand (x y)
1393 `(lognot (logand ,x ,y)))
1397 (define-vop (bignum-length get-header-data)
1398 (:translate sb!bignum:%bignum-length)
1399 (:policy :fast-safe))
1401 (define-vop (bignum-set-length set-header-data)
1402 (:translate sb!bignum:%bignum-set-length)
1403 (:policy :fast-safe))
1405 (define-full-reffer bignum-ref * bignum-digits-offset other-pointer-lowtag
1406 (unsigned-reg) unsigned-num sb!bignum:%bignum-ref)
1408 (define-full-setter bignum-set * bignum-digits-offset other-pointer-lowtag
1409 (unsigned-reg) unsigned-num sb!bignum:%bignum-set)
1411 (define-vop (digit-0-or-plus)
1412 (:translate sb!bignum:%digit-0-or-plusp)
1413 (:policy :fast-safe)
1414 (:args (digit :scs (unsigned-reg)))
1415 (:arg-types unsigned-num)
1417 (:info target not-p)
1419 (inst or digit digit)
1420 (inst jmp (if not-p :s :ns) target)))
1423 ;;; For add and sub with carry the sc of carry argument is any-reg so
1424 ;;; the it may be passed as a fixnum or word and thus may be 0, 1, or
1425 ;;; 4. This is easy to deal with and may save a fixnum-word
1427 (define-vop (add-w/carry)
1428 (:translate sb!bignum:%add-with-carry)
1429 (:policy :fast-safe)
1430 (:args (a :scs (unsigned-reg) :target result)
1431 (b :scs (unsigned-reg unsigned-stack) :to :eval)
1432 (c :scs (any-reg) :target temp))
1433 (:arg-types unsigned-num unsigned-num positive-fixnum)
1434 (:temporary (:sc any-reg :from (:argument 2) :to :eval) temp)
1435 (:results (result :scs (unsigned-reg) :from (:argument 0))
1436 (carry :scs (unsigned-reg)))
1437 (:result-types unsigned-num positive-fixnum)
1441 (inst neg temp) ; Set the carry flag to 0 if c=0 else to 1
1444 (inst adc carry carry)))
1446 ;;; Note: the borrow is the oppostite of the x86 convention - 1 for no
1447 ;;; borrow and 0 for a borrow.
1448 (define-vop (sub-w/borrow)
1449 (:translate sb!bignum:%subtract-with-borrow)
1450 (:policy :fast-safe)
1451 (:args (a :scs (unsigned-reg) :to :eval :target result)
1452 (b :scs (unsigned-reg unsigned-stack) :to :result)
1453 (c :scs (any-reg control-stack)))
1454 (:arg-types unsigned-num unsigned-num positive-fixnum)
1455 (:results (result :scs (unsigned-reg) :from :eval)
1456 (borrow :scs (unsigned-reg)))
1457 (:result-types unsigned-num positive-fixnum)
1459 (inst cmp c 1) ; Set the carry flag to 1 if c=0 else to 0
1463 (inst adc borrow borrow)
1464 (inst xor borrow 1)))
1467 (define-vop (bignum-mult-and-add-3-arg)
1468 (:translate sb!bignum:%multiply-and-add)
1469 (:policy :fast-safe)
1470 (:args (x :scs (unsigned-reg) :target eax)
1471 (y :scs (unsigned-reg unsigned-stack))
1472 (carry-in :scs (unsigned-reg unsigned-stack)))
1473 (:arg-types unsigned-num unsigned-num unsigned-num)
1474 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
1475 :to (:result 1) :target lo) eax)
1476 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
1477 :to (:result 0) :target hi) edx)
1478 (:results (hi :scs (unsigned-reg))
1479 (lo :scs (unsigned-reg)))
1480 (:result-types unsigned-num unsigned-num)
1484 (inst add eax carry-in)
1489 (define-vop (bignum-mult-and-add-4-arg)
1490 (:translate sb!bignum:%multiply-and-add)
1491 (:policy :fast-safe)
1492 (:args (x :scs (unsigned-reg) :target eax)
1493 (y :scs (unsigned-reg unsigned-stack))
1494 (prev :scs (unsigned-reg unsigned-stack))
1495 (carry-in :scs (unsigned-reg unsigned-stack)))
1496 (:arg-types unsigned-num unsigned-num unsigned-num unsigned-num)
1497 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
1498 :to (:result 1) :target lo) eax)
1499 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
1500 :to (:result 0) :target hi) edx)
1501 (:results (hi :scs (unsigned-reg))
1502 (lo :scs (unsigned-reg)))
1503 (:result-types unsigned-num unsigned-num)
1509 (inst add eax carry-in)
1515 (define-vop (bignum-mult)
1516 (:translate sb!bignum:%multiply)
1517 (:policy :fast-safe)
1518 (:args (x :scs (unsigned-reg) :target eax)
1519 (y :scs (unsigned-reg unsigned-stack)))
1520 (:arg-types unsigned-num unsigned-num)
1521 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
1522 :to (:result 1) :target lo) eax)
1523 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
1524 :to (:result 0) :target hi) edx)
1525 (:results (hi :scs (unsigned-reg))
1526 (lo :scs (unsigned-reg)))
1527 (:result-types unsigned-num unsigned-num)
1534 (define-vop (bignum-lognot lognot-mod32/word=>unsigned)
1535 (:translate sb!bignum:%lognot))
1537 (define-vop (fixnum-to-digit)
1538 (:translate sb!bignum:%fixnum-to-digit)
1539 (:policy :fast-safe)
1540 (:args (fixnum :scs (any-reg control-stack) :target digit))
1541 (:arg-types tagged-num)
1542 (:results (digit :scs (unsigned-reg)
1543 :load-if (not (and (sc-is fixnum control-stack)
1544 (sc-is digit unsigned-stack)
1545 (location= fixnum digit)))))
1546 (:result-types unsigned-num)
1549 (inst sar digit 2)))
1551 (define-vop (bignum-floor)
1552 (:translate sb!bignum:%floor)
1553 (:policy :fast-safe)
1554 (:args (div-high :scs (unsigned-reg) :target edx)
1555 (div-low :scs (unsigned-reg) :target eax)
1556 (divisor :scs (unsigned-reg unsigned-stack)))
1557 (:arg-types unsigned-num unsigned-num unsigned-num)
1558 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)
1559 :to (:result 0) :target quo) eax)
1560 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 0)
1561 :to (:result 1) :target rem) edx)
1562 (:results (quo :scs (unsigned-reg))
1563 (rem :scs (unsigned-reg)))
1564 (:result-types unsigned-num unsigned-num)
1568 (inst div eax divisor)
1572 (define-vop (signify-digit)
1573 (:translate sb!bignum:%fixnum-digit-with-correct-sign)
1574 (:policy :fast-safe)
1575 (:args (digit :scs (unsigned-reg unsigned-stack) :target res))
1576 (:arg-types unsigned-num)
1577 (:results (res :scs (any-reg signed-reg)
1578 :load-if (not (and (sc-is digit unsigned-stack)
1579 (sc-is res control-stack signed-stack)
1580 (location= digit res)))))
1581 (:result-types signed-num)
1584 (when (sc-is res any-reg control-stack)
1587 (define-vop (digit-ashr)
1588 (:translate sb!bignum:%ashr)
1589 (:policy :fast-safe)
1590 (:args (digit :scs (unsigned-reg unsigned-stack) :target result)
1591 (count :scs (unsigned-reg) :target ecx))
1592 (:arg-types unsigned-num positive-fixnum)
1593 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
1594 (:results (result :scs (unsigned-reg) :from (:argument 0)
1595 :load-if (not (and (sc-is result unsigned-stack)
1596 (location= digit result)))))
1597 (:result-types unsigned-num)
1601 (inst sar result :cl)))
1603 (define-vop (digit-lshr digit-ashr)
1604 (:translate sb!bignum:%digit-logical-shift-right)
1608 (inst shr result :cl)))
1610 (define-vop (digit-ashl digit-ashr)
1611 (:translate sb!bignum:%ashl)
1615 (inst shl result :cl)))
1617 ;;;; static functions
1619 (define-static-fun two-arg-/ (x y) :translate /)
1621 (define-static-fun two-arg-gcd (x y) :translate gcd)
1622 (define-static-fun two-arg-lcm (x y) :translate lcm)
1624 (define-static-fun two-arg-and (x y) :translate logand)
1625 (define-static-fun two-arg-ior (x y) :translate logior)
1626 (define-static-fun two-arg-xor (x y) :translate logxor)
1629 ;;; Support for the Mersenne Twister, MT19937, random number generator
1630 ;;; due to Matsumoto and Nishimura.
1632 ;;; Makoto Matsumoto and T. Nishimura, "Mersenne twister: A
1633 ;;; 623-dimensionally equidistributed uniform pseudorandom number
1634 ;;; generator.", ACM Transactions on Modeling and Computer Simulation,
1635 ;;; 1997, to appear.
1638 ;;; 0-1: Constant matrix A. [0, #x9908b0df] (not used here)
1639 ;;; 2: Index; init. to 1.
1641 (defknown random-mt19937 ((simple-array (unsigned-byte 32) (*)))
1642 (unsigned-byte 32) ())
1643 (define-vop (random-mt19937)
1644 (:policy :fast-safe)
1645 (:translate random-mt19937)
1646 (:args (state :scs (descriptor-reg) :to :result))
1647 (:arg-types simple-array-unsigned-byte-32)
1648 (:temporary (:sc unsigned-reg :from (:eval 0) :to :result) k)
1649 (:temporary (:sc unsigned-reg :offset eax-offset
1650 :from (:eval 0) :to :result) tmp)
1651 (:results (y :scs (unsigned-reg) :from (:eval 0)))
1652 (:result-types unsigned-num)
1654 (inst mov k (make-ea :dword :base state
1655 :disp (- (* (+ 2 vector-data-offset)
1657 other-pointer-lowtag)))
1659 (inst jmp :ne no-update)
1660 (inst mov tmp state) ; The state is passed in EAX.
1661 (inst call (make-fixup 'random-mt19937-update :assembly-routine))
1662 ;; Restore k, and set to 0.
1666 (inst mov y (make-ea :dword :base state :index k :scale 4
1667 :disp (- (* (+ 3 vector-data-offset)
1669 other-pointer-lowtag)))
1672 (inst xor y (make-ea :dword :base state :index k :scale 4
1673 :disp (- (* (+ 3 vector-data-offset)
1675 other-pointer-lowtag)))
1676 ;; y ^= (y << 7) & #x9d2c5680
1680 (inst mov (make-ea :dword :base state
1681 :disp (- (* (+ 2 vector-data-offset)
1683 other-pointer-lowtag))
1685 (inst and tmp #x9d2c5680)
1687 ;; y ^= (y << 15) & #xefc60000
1690 (inst and tmp #xefc60000)
1699 (defun mask-result (class width result)
1702 `(logand ,result ,(1- (ash 1 width))))
1704 `(mask-signed-field ,width ,result))))
1706 ;;; This is essentially a straight implementation of the algorithm in
1707 ;;; "Strength Reduction of Multiplications by Integer Constants",
1708 ;;; Youfeng Wu, ACM SIGPLAN Notices, Vol. 30, No.2, February 1995.
1709 (defun basic-decompose-multiplication (class width arg num n-bits condensed)
1710 (case (aref condensed 0)
1712 (let ((tmp (min 3 (aref condensed 1))))
1713 (decf (aref condensed 1) tmp)
1714 (mask-result class width
1716 ,(decompose-multiplication class width
1717 arg (ash (1- num) (- tmp)) (1- n-bits) (subseq condensed 1))
1720 (let ((r0 (aref condensed 0)))
1721 (incf (aref condensed 1) r0)
1722 (mask-result class width
1723 `(%lea ,(decompose-multiplication class width
1724 arg (- num (ash 1 r0)) (1- n-bits) (subseq condensed 1))
1727 (t (let ((r0 (aref condensed 0)))
1728 (setf (aref condensed 0) 0)
1729 (mask-result class width
1730 `(ash ,(decompose-multiplication class width
1731 arg (ash num (- r0)) n-bits condensed)
1734 (defun decompose-multiplication (class width arg num n-bits condensed)
1739 (mask-result class width `(ash ,arg ,(1- (integer-length num)))))
1740 ((let ((max 0) (end 0))
1741 (loop for i from 2 to (length condensed)
1742 for j = (reduce #'+ (subseq condensed 0 i))
1743 when (and (> (- (* 2 i) 3 j) max)
1744 (< (+ (ash 1 (1+ j))
1745 (ash (ldb (byte (- 32 (1+ j)) (1+ j)) num)
1748 do (setq max (- (* 2 i) 3 j)
1751 (let ((j (reduce #'+ (subseq condensed 0 end))))
1752 (let ((n2 (+ (ash 1 (1+ j))
1753 (ash (ldb (byte (- 32 (1+ j)) (1+ j)) num) (1+ j))))
1754 (n1 (1+ (ldb (byte (1+ j) 0) (lognot num)))))
1755 (mask-result class width
1756 `(- ,(optimize-multiply class width arg n2)
1757 ,(optimize-multiply class width arg n1))))))))
1758 ((dolist (i '(9 5 3))
1759 (when (integerp (/ num i))
1760 (when (< (logcount (/ num i)) (logcount num))
1762 (return `(let ((,x ,(optimize-multiply class width arg (/ num i))))
1763 ,(mask-result class width
1764 `(%lea ,x ,x (1- ,i) 0)))))))))
1765 (t (basic-decompose-multiplication class width arg num n-bits condensed))))
1767 (defun optimize-multiply (class width arg x)
1768 (let* ((n-bits (logcount x))
1769 (condensed (make-array n-bits)))
1770 (let ((count 0) (bit 0))
1772 (cond ((logbitp i x)
1773 (setf (aref condensed bit) count)
1777 (decompose-multiplication class width arg x n-bits condensed)))
1779 (defun *-transformer (class width y)
1781 ((= y (ash 1 (integer-length y)))
1782 ;; there's a generic transform for y = 2^k
1783 (give-up-ir1-transform))
1784 ((member y '(3 5 9))
1785 ;; we can do these multiplications directly using LEA
1786 `(%lea x x ,(1- y) 0))
1787 ((member :pentium4 *backend-subfeatures*)
1788 ;; the pentium4's multiply unit is reportedly very good
1789 (give-up-ir1-transform))
1790 ;; FIXME: should make this more fine-grained. If nothing else,
1791 ;; there should probably be a cutoff of about 9 instructions on
1792 ;; pentium-class machines.
1793 (t (optimize-multiply class width 'x y))))
1795 (deftransform * ((x y)
1796 ((unsigned-byte 32) (constant-arg (unsigned-byte 32)))
1798 "recode as leas, shifts and adds"
1799 (let ((y (lvar-value y)))
1800 (*-transformer :unsigned 32 y)))
1801 (deftransform sb!vm::*-mod32
1802 ((x y) ((unsigned-byte 32) (constant-arg (unsigned-byte 32)))
1804 "recode as leas, shifts and adds"
1805 (let ((y (lvar-value y)))
1806 (*-transformer :unsigned 32 y)))
1808 (deftransform * ((x y)
1809 ((signed-byte 30) (constant-arg (unsigned-byte 32)))
1811 "recode as leas, shifts and adds"
1812 (let ((y (lvar-value y)))
1813 (*-transformer :signed 30 y)))
1814 (deftransform sb!vm::*-smod30
1815 ((x y) ((signed-byte 30) (constant-arg (unsigned-byte 32)))
1817 "recode as leas, shifts and adds"
1818 (let ((y (lvar-value y)))
1819 (*-transformer :signed 30 y)))
1821 ;;; FIXME: we should also be able to write an optimizer or two to
1822 ;;; convert (+ (* x 2) 17), (- (* x 9) 5) to a %LEA.