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 29)))
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 ;; 32 not 64 because it's hard work loading 64 bit constants
130 (define-vop (fast-signed-binop-c fast-safe-arith-op)
131 (:args (x :target r :scs (signed-reg signed-stack)))
133 (:arg-types signed-num (:constant (signed-byte 32)))
134 (:results (r :scs (signed-reg)
135 :load-if (not (location= x r))))
136 (:result-types signed-num)
137 (:note "inline (signed-byte 64) arithmetic"))
139 (macrolet ((define-binop (translate untagged-penalty op)
141 (define-vop (,(symbolicate "FAST-" translate "/FIXNUM=>FIXNUM")
143 (:translate ,translate)
147 (define-vop (,(symbolicate 'fast- translate '-c/fixnum=>fixnum)
149 (:translate ,translate)
152 (inst ,op r (fixnumize y))))
153 (define-vop (,(symbolicate "FAST-" translate "/SIGNED=>SIGNED")
155 (:translate ,translate)
156 (:generator ,(1+ untagged-penalty)
159 (define-vop (,(symbolicate 'fast- translate '-c/signed=>signed)
161 (:translate ,translate)
162 (:generator ,untagged-penalty
165 (define-vop (,(symbolicate "FAST-"
167 "/UNSIGNED=>UNSIGNED")
169 (:translate ,translate)
170 (:generator ,(1+ untagged-penalty)
173 (define-vop (,(symbolicate 'fast-
175 '-c/unsigned=>unsigned)
176 fast-unsigned-binop-c)
177 (:translate ,translate)
178 (:generator ,untagged-penalty
182 ;;(define-binop + 4 add)
183 (define-binop - 4 sub)
184 (define-binop logand 2 and)
185 (define-binop logior 2 or)
186 (define-binop logxor 2 xor))
188 ;;; Special handling of add on the x86; can use lea to avoid a
189 ;;; register load, otherwise it uses add.
190 (define-vop (fast-+/fixnum=>fixnum fast-safe-arith-op)
192 (:args (x :scs (any-reg) :target r
193 :load-if (not (and (sc-is x control-stack)
195 (sc-is r control-stack)
197 (y :scs (any-reg control-stack)))
198 (:arg-types tagged-num tagged-num)
199 (:results (r :scs (any-reg) :from (:argument 0)
200 :load-if (not (and (sc-is x control-stack)
202 (sc-is r control-stack)
204 (:result-types tagged-num)
205 (:note "inline fixnum arithmetic")
207 (cond ((and (sc-is x any-reg) (sc-is y any-reg) (sc-is r any-reg)
208 (not (location= x r)))
209 (inst lea r (make-ea :qword :base x :index y :scale 1)))
214 (define-vop (fast-+-c/fixnum=>fixnum fast-safe-arith-op)
216 (:args (x :target r :scs (any-reg control-stack)))
218 (:arg-types tagged-num (:constant (signed-byte 29)))
219 (:results (r :scs (any-reg)
220 :load-if (not (location= x r))))
221 (:result-types tagged-num)
222 (:note "inline fixnum arithmetic")
224 (cond ((and (sc-is x any-reg) (sc-is r any-reg) (not (location= x r)))
225 (inst lea r (make-ea :qword :base x :disp (fixnumize y))))
228 (inst add r (fixnumize y))))))
230 (define-vop (fast-+/signed=>signed fast-safe-arith-op)
232 (:args (x :scs (signed-reg) :target r
233 :load-if (not (and (sc-is x signed-stack)
235 (sc-is r signed-stack)
237 (y :scs (signed-reg signed-stack)))
238 (:arg-types signed-num signed-num)
239 (:results (r :scs (signed-reg) :from (:argument 0)
240 :load-if (not (and (sc-is x signed-stack)
243 (:result-types signed-num)
244 (:note "inline (signed-byte 32) arithmetic")
246 (cond ((and (sc-is x signed-reg) (sc-is y signed-reg) (sc-is r signed-reg)
247 (not (location= x r)))
248 (inst lea r (make-ea :qword :base x :index y :scale 1)))
254 ;;;; Special logand cases: (logand signed unsigned) => unsigned
256 (define-vop (fast-logand/signed-unsigned=>unsigned
257 fast-logand/unsigned=>unsigned)
258 (:args (x :target r :scs (signed-reg)
259 :load-if (not (and (sc-is x signed-stack)
260 (sc-is y unsigned-reg)
261 (sc-is r unsigned-stack)
263 (y :scs (unsigned-reg unsigned-stack)))
264 (:arg-types signed-num unsigned-num))
266 (define-vop (fast-logand-c/signed-unsigned=>unsigned
267 fast-logand-c/unsigned=>unsigned)
268 (:args (x :target r :scs (signed-reg signed-stack)))
269 (:arg-types signed-num (:constant (unsigned-byte 32))))
271 (define-vop (fast-logand/unsigned-signed=>unsigned
272 fast-logand/unsigned=>unsigned)
273 (:args (x :target r :scs (unsigned-reg)
274 :load-if (not (and (sc-is x unsigned-stack)
276 (sc-is r unsigned-stack)
278 (y :scs (signed-reg signed-stack)))
279 (:arg-types unsigned-num signed-num))
282 (define-vop (fast-+-c/signed=>signed fast-safe-arith-op)
284 (:args (x :target r :scs (signed-reg signed-stack)))
286 (:arg-types signed-num (:constant (signed-byte 32)))
287 (:results (r :scs (signed-reg)
288 :load-if (not (location= x r))))
289 (:result-types signed-num)
290 (:note "inline (signed-byte 32) arithmetic")
292 (cond ((and (sc-is x signed-reg) (sc-is r signed-reg)
293 (not (location= x r)))
294 (inst lea r (make-ea :qword :base x :disp y)))
301 (define-vop (fast-+/unsigned=>unsigned fast-safe-arith-op)
303 (:args (x :scs (unsigned-reg) :target r
304 :load-if (not (and (sc-is x unsigned-stack)
305 (sc-is y unsigned-reg)
306 (sc-is r unsigned-stack)
308 (y :scs (unsigned-reg unsigned-stack)))
309 (:arg-types unsigned-num unsigned-num)
310 (:results (r :scs (unsigned-reg) :from (:argument 0)
311 :load-if (not (and (sc-is x unsigned-stack)
312 (sc-is y unsigned-reg)
313 (sc-is r unsigned-stack)
315 (:result-types unsigned-num)
316 (:note "inline (unsigned-byte 32) arithmetic")
318 (cond ((and (sc-is x unsigned-reg) (sc-is y unsigned-reg)
319 (sc-is r unsigned-reg) (not (location= x r)))
320 (inst lea r (make-ea :qword :base x :index y :scale 1)))
325 (define-vop (fast-+-c/unsigned=>unsigned fast-safe-arith-op)
327 (:args (x :target r :scs (unsigned-reg unsigned-stack)))
329 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
330 (:results (r :scs (unsigned-reg)
331 :load-if (not (location= x r))))
332 (:result-types unsigned-num)
333 (:note "inline (unsigned-byte 32) arithmetic")
335 (cond ((and (sc-is x unsigned-reg) (sc-is r unsigned-reg)
336 (not (location= x r)))
337 (inst lea r (make-ea :qword :base x :disp y)))
344 ;;;; multiplication and division
346 (define-vop (fast-*/fixnum=>fixnum fast-safe-arith-op)
348 ;; We need different loading characteristics.
349 (:args (x :scs (any-reg) :target r)
350 (y :scs (any-reg control-stack)))
351 (:arg-types tagged-num tagged-num)
352 (:results (r :scs (any-reg) :from (:argument 0)))
353 (:result-types tagged-num)
354 (:note "inline fixnum arithmetic")
360 (define-vop (fast-*-c/fixnum=>fixnum fast-safe-arith-op)
362 ;; We need different loading characteristics.
363 (:args (x :scs (any-reg control-stack)))
365 (:arg-types tagged-num (:constant (signed-byte 29)))
366 (:results (r :scs (any-reg)))
367 (:result-types tagged-num)
368 (:note "inline fixnum arithmetic")
372 (define-vop (fast-*/signed=>signed fast-safe-arith-op)
374 ;; We need different loading characteristics.
375 (:args (x :scs (signed-reg) :target r)
376 (y :scs (signed-reg signed-stack)))
377 (:arg-types signed-num signed-num)
378 (:results (r :scs (signed-reg) :from (:argument 0)))
379 (:result-types signed-num)
380 (:note "inline (signed-byte 32) arithmetic")
385 (define-vop (fast-*-c/signed=>signed fast-safe-arith-op)
387 ;; We need different loading characteristics.
388 (:args (x :scs (signed-reg signed-stack)))
390 (:arg-types signed-num (:constant (signed-byte 32)))
391 (:results (r :scs (signed-reg)))
392 (:result-types signed-num)
393 (:note "inline (signed-byte 32) arithmetic")
397 (define-vop (fast-*/unsigned=>unsigned fast-safe-arith-op)
399 (:args (x :scs (unsigned-reg) :target eax)
400 (y :scs (unsigned-reg unsigned-stack)))
401 (:arg-types unsigned-num unsigned-num)
402 (:temporary (:sc unsigned-reg :offset eax-offset :target result
403 :from (:argument 0) :to :result) eax)
404 (:temporary (:sc unsigned-reg :offset edx-offset
405 :from :eval :to :result) edx)
407 (:results (result :scs (unsigned-reg)))
408 (:result-types unsigned-num)
409 (:note "inline (unsigned-byte 32) arithmetic")
411 (:save-p :compute-only)
418 (define-vop (fast-truncate/fixnum=>fixnum fast-safe-arith-op)
419 (:translate truncate)
420 (:args (x :scs (any-reg) :target eax)
421 (y :scs (any-reg control-stack)))
422 (:arg-types tagged-num tagged-num)
423 (:temporary (:sc signed-reg :offset eax-offset :target quo
424 :from (:argument 0) :to (:result 0)) eax)
425 (:temporary (:sc unsigned-reg :offset edx-offset :target rem
426 :from (:argument 0) :to (:result 1)) edx)
427 (:results (quo :scs (any-reg))
428 (rem :scs (any-reg)))
429 (:result-types tagged-num tagged-num)
430 (:note "inline fixnum arithmetic")
432 (:save-p :compute-only)
434 (let ((zero (generate-error-code vop division-by-zero-error x y)))
435 (if (sc-is y any-reg)
436 (inst test y y) ; smaller instruction
442 (if (location= quo eax)
444 (inst lea quo (make-ea :qword :index eax :scale 8)))
447 (define-vop (fast-truncate-c/fixnum=>fixnum fast-safe-arith-op)
448 (:translate truncate)
449 (:args (x :scs (any-reg) :target eax))
451 (:arg-types tagged-num (:constant (signed-byte 29)))
452 (:temporary (:sc signed-reg :offset eax-offset :target quo
453 :from :argument :to (:result 0)) eax)
454 (:temporary (:sc any-reg :offset edx-offset :target rem
455 :from :eval :to (:result 1)) edx)
456 (:temporary (:sc any-reg :from :eval :to :result) y-arg)
457 (:results (quo :scs (any-reg))
458 (rem :scs (any-reg)))
459 (:result-types tagged-num tagged-num)
460 (:note "inline fixnum arithmetic")
462 (:save-p :compute-only)
466 (inst mov y-arg (fixnumize y))
467 (inst idiv eax y-arg)
468 (if (location= quo eax)
470 (inst lea quo (make-ea :qword :index eax :scale 8)))
473 (define-vop (fast-truncate/unsigned=>unsigned fast-safe-arith-op)
474 (:translate truncate)
475 (:args (x :scs (unsigned-reg) :target eax)
476 (y :scs (unsigned-reg signed-stack)))
477 (:arg-types unsigned-num unsigned-num)
478 (:temporary (:sc unsigned-reg :offset eax-offset :target quo
479 :from (:argument 0) :to (:result 0)) eax)
480 (:temporary (:sc unsigned-reg :offset edx-offset :target rem
481 :from (:argument 0) :to (:result 1)) edx)
482 (:results (quo :scs (unsigned-reg))
483 (rem :scs (unsigned-reg)))
484 (:result-types unsigned-num unsigned-num)
485 (:note "inline (unsigned-byte 32) arithmetic")
487 (:save-p :compute-only)
489 (let ((zero (generate-error-code vop division-by-zero-error x y)))
490 (if (sc-is y unsigned-reg)
491 (inst test y y) ; smaller instruction
500 (define-vop (fast-truncate-c/unsigned=>unsigned fast-safe-arith-op)
501 (:translate truncate)
502 (:args (x :scs (unsigned-reg) :target eax))
504 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
505 (:temporary (:sc unsigned-reg :offset eax-offset :target quo
506 :from :argument :to (:result 0)) eax)
507 (:temporary (:sc unsigned-reg :offset edx-offset :target rem
508 :from :eval :to (:result 1)) edx)
509 (:temporary (:sc unsigned-reg :from :eval :to :result) y-arg)
510 (:results (quo :scs (unsigned-reg))
511 (rem :scs (unsigned-reg)))
512 (:result-types unsigned-num unsigned-num)
513 (:note "inline (unsigned-byte 32) arithmetic")
515 (:save-p :compute-only)
524 (define-vop (fast-truncate/signed=>signed fast-safe-arith-op)
525 (:translate truncate)
526 (:args (x :scs (signed-reg) :target eax)
527 (y :scs (signed-reg signed-stack)))
528 (:arg-types signed-num signed-num)
529 (:temporary (:sc signed-reg :offset eax-offset :target quo
530 :from (:argument 0) :to (:result 0)) eax)
531 (:temporary (:sc signed-reg :offset edx-offset :target rem
532 :from (:argument 0) :to (:result 1)) edx)
533 (:results (quo :scs (signed-reg))
534 (rem :scs (signed-reg)))
535 (:result-types signed-num signed-num)
536 (:note "inline (signed-byte 32) arithmetic")
538 (:save-p :compute-only)
540 (let ((zero (generate-error-code vop division-by-zero-error x y)))
541 (if (sc-is y signed-reg)
542 (inst test y y) ; smaller instruction
551 (define-vop (fast-truncate-c/signed=>signed fast-safe-arith-op)
552 (:translate truncate)
553 (:args (x :scs (signed-reg) :target eax))
555 (:arg-types signed-num (:constant (signed-byte 32)))
556 (:temporary (:sc signed-reg :offset eax-offset :target quo
557 :from :argument :to (:result 0)) eax)
558 (:temporary (:sc signed-reg :offset edx-offset :target rem
559 :from :eval :to (:result 1)) edx)
560 (:temporary (:sc signed-reg :from :eval :to :result) y-arg)
561 (:results (quo :scs (signed-reg))
562 (rem :scs (signed-reg)))
563 (:result-types signed-num signed-num)
564 (:note "inline (signed-byte 32) arithmetic")
566 (:save-p :compute-only)
571 (inst idiv eax y-arg)
578 (define-vop (fast-ash-c/fixnum=>fixnum)
581 (:args (number :scs (any-reg) :target result
582 :load-if (not (and (sc-is number any-reg control-stack)
583 (sc-is result any-reg control-stack)
584 (location= number result)))))
586 (:arg-types tagged-num (:constant integer))
587 (:results (result :scs (any-reg)
588 :load-if (not (and (sc-is number control-stack)
589 (sc-is result control-stack)
590 (location= number result)))))
591 (:result-types tagged-num)
594 (cond ((and (= amount 1) (not (location= number result)))
595 (inst lea result (make-ea :qword :index number :scale 2)))
596 ((and (= amount 2) (not (location= number result)))
597 (inst lea result (make-ea :qword :index number :scale 4)))
598 ((and (= amount 3) (not (location= number result)))
599 (inst lea result (make-ea :qword :index number :scale 8)))
602 (cond ((plusp amount)
603 ;; We don't have to worry about overflow because of the
604 ;; result type restriction.
605 (inst shl result amount))
608 (inst xor result result))
610 ;; shift too far then back again, to zero tag bits
611 (inst sar result (- 3 amount))
613 (make-ea :qword :index result :scale 8))))))))
616 (define-vop (fast-ash-left/fixnum=>fixnum)
618 (:args (number :scs (any-reg) :target result
619 :load-if (not (and (sc-is number control-stack)
620 (sc-is result control-stack)
621 (location= number result))))
622 (amount :scs (unsigned-reg) :target ecx))
623 (:arg-types tagged-num positive-fixnum)
624 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
625 (:results (result :scs (any-reg) :from (:argument 0)
626 :load-if (not (and (sc-is number control-stack)
627 (sc-is result control-stack)
628 (location= number result)))))
629 (:result-types tagged-num)
635 ;; The result-type ensures us that this shift will not overflow.
636 (inst shl result :cl)))
638 (define-vop (fast-ash-c/signed=>signed)
641 (:args (number :scs (signed-reg) :target result
642 :load-if (not (and (sc-is number signed-stack)
643 (sc-is result signed-stack)
644 (location= number result)))))
646 (:arg-types signed-num (:constant integer))
647 (:results (result :scs (signed-reg)
648 :load-if (not (and (sc-is number signed-stack)
649 (sc-is result signed-stack)
650 (location= number result)))))
651 (:result-types signed-num)
654 (cond ((and (= amount 1) (not (location= number result)))
655 (inst lea result (make-ea :qword :index number :scale 2)))
656 ((and (= amount 2) (not (location= number result)))
657 (inst lea result (make-ea :qword :index number :scale 4)))
658 ((and (= amount 3) (not (location= number result)))
659 (inst lea result (make-ea :qword :index number :scale 8)))
662 (cond ((plusp amount) (inst shl result amount))
663 (t (inst sar result (min 63 (- amount)))))))))
665 (define-vop (fast-ash-c/unsigned=>unsigned)
668 (:args (number :scs (unsigned-reg) :target result
669 :load-if (not (and (sc-is number unsigned-stack)
670 (sc-is result unsigned-stack)
671 (location= number result)))))
673 (:arg-types unsigned-num (:constant integer))
674 (:results (result :scs (unsigned-reg)
675 :load-if (not (and (sc-is number unsigned-stack)
676 (sc-is result unsigned-stack)
677 (location= number result)))))
678 (:result-types unsigned-num)
681 (cond ((and (= amount 1) (not (location= number result)))
682 (inst lea result (make-ea :qword :index number :scale 2)))
683 ((and (= amount 2) (not (location= number result)))
684 (inst lea result (make-ea :qword :index number :scale 4)))
685 ((and (= amount 3) (not (location= number result)))
686 (inst lea result (make-ea :qword :index number :scale 8)))
689 (cond ((< -64 amount 64) ;; XXXX
690 ;; this code is used both in ASH and ASH-MOD32, so
693 (inst shl result amount)
694 (inst shr result (- amount))))
695 (t (if (sc-is result unsigned-reg)
696 (inst xor result result)
697 (inst mov result 0))))))))
699 (define-vop (fast-ash-left/signed=>signed)
701 (:args (number :scs (signed-reg) :target result
702 :load-if (not (and (sc-is number signed-stack)
703 (sc-is result signed-stack)
704 (location= number result))))
705 (amount :scs (unsigned-reg) :target ecx))
706 (:arg-types signed-num positive-fixnum)
707 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
708 (:results (result :scs (signed-reg) :from (:argument 0)
709 :load-if (not (and (sc-is number signed-stack)
710 (sc-is result signed-stack)
711 (location= number result)))))
712 (:result-types signed-num)
718 (inst shl result :cl)))
720 (define-vop (fast-ash-left/unsigned=>unsigned)
722 (:args (number :scs (unsigned-reg) :target result
723 :load-if (not (and (sc-is number unsigned-stack)
724 (sc-is result unsigned-stack)
725 (location= number result))))
726 (amount :scs (unsigned-reg) :target ecx))
727 (:arg-types unsigned-num positive-fixnum)
728 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
729 (:results (result :scs (unsigned-reg) :from (:argument 0)
730 :load-if (not (and (sc-is number unsigned-stack)
731 (sc-is result unsigned-stack)
732 (location= number result)))))
733 (:result-types unsigned-num)
739 (inst shl result :cl)))
741 (define-vop (fast-ash/signed=>signed)
744 (:args (number :scs (signed-reg) :target result)
745 (amount :scs (signed-reg) :target ecx))
746 (:arg-types signed-num signed-num)
747 (:results (result :scs (signed-reg) :from (:argument 0)))
748 (:result-types signed-num)
749 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
755 (inst jmp :ns positive)
761 (inst sar result :cl)
765 ;; The result-type ensures us that this shift will not overflow.
766 (inst shl result :cl)
770 (define-vop (fast-ash/unsigned=>unsigned)
773 (:args (number :scs (unsigned-reg) :target result)
774 (amount :scs (signed-reg) :target ecx))
775 (:arg-types unsigned-num signed-num)
776 (:results (result :scs (unsigned-reg) :from (:argument 0)))
777 (:result-types unsigned-num)
778 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
784 (inst jmp :ns positive)
788 (inst xor result result)
791 (inst shr result :cl)
795 ;; The result-type ensures us that this shift will not overflow.
796 (inst shl result :cl)
802 (defknown %lea (integer integer (member 1 2 4 8 16) (signed-byte 64))
804 (foldable flushable movable))
806 (defoptimizer (%lea derive-type) ((base index scale disp))
807 (when (and (constant-lvar-p scale)
808 (constant-lvar-p disp))
809 (let ((scale (lvar-value scale))
810 (disp (lvar-value disp))
811 (base-type (lvar-type base))
812 (index-type (lvar-type index)))
813 (when (and (numeric-type-p base-type)
814 (numeric-type-p index-type))
815 (let ((base-lo (numeric-type-low base-type))
816 (base-hi (numeric-type-high base-type))
817 (index-lo (numeric-type-low index-type))
818 (index-hi (numeric-type-high index-type)))
819 (make-numeric-type :class 'integer
821 :low (when (and base-lo index-lo)
822 (+ base-lo (* index-lo scale) disp))
823 :high (when (and base-hi index-hi)
824 (+ base-hi (* index-hi scale) disp))))))))
826 (defun %lea (base index scale disp)
827 (+ base (* index scale) disp))
831 (define-vop (%lea/unsigned=>unsigned)
834 (:args (base :scs (unsigned-reg))
835 (index :scs (unsigned-reg)))
837 (:arg-types unsigned-num unsigned-num
838 (:constant (member 1 2 4 8))
839 (:constant (signed-byte 64)))
840 (:results (r :scs (unsigned-reg)))
841 (:result-types unsigned-num)
843 (inst lea r (make-ea :qword :base base :index index
844 :scale scale :disp disp))))
846 (define-vop (%lea/signed=>signed)
849 (:args (base :scs (signed-reg))
850 (index :scs (signed-reg)))
852 (:arg-types signed-num signed-num
853 (:constant (member 1 2 4 8))
854 (:constant (signed-byte 64)))
855 (:results (r :scs (signed-reg)))
856 (:result-types signed-num)
858 (inst lea r (make-ea :qword :base base :index index
859 :scale scale :disp disp))))
861 (define-vop (%lea/fixnum=>fixnum)
864 (:args (base :scs (any-reg))
865 (index :scs (any-reg)))
867 (:arg-types tagged-num tagged-num
868 (:constant (member 1 2 4 8))
869 (:constant (signed-byte 64)))
870 (:results (r :scs (any-reg)))
871 (:result-types tagged-num)
873 (inst lea r (make-ea :qword :base base :index index
874 :scale scale :disp disp))))
876 ;;; FIXME: before making knowledge of this too public, it needs to be
877 ;;; fixed so that it's actually _faster_ than the non-CMOV version; at
878 ;;; least on my Celeron-XXX laptop, this version is marginally slower
879 ;;; than the above version with branches. -- CSR, 2003-09-04
880 (define-vop (fast-cmov-ash/unsigned=>unsigned)
883 (:args (number :scs (unsigned-reg) :target result)
884 (amount :scs (signed-reg) :target ecx))
885 (:arg-types unsigned-num signed-num)
886 (:results (result :scs (unsigned-reg) :from (:argument 0)))
887 (:result-types unsigned-num)
888 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
889 (:temporary (:sc any-reg :from (:eval 0) :to (:eval 1)) zero)
891 (:guard (member :cmov *backend-subfeatures*))
896 (inst jmp :ns positive)
899 (inst shr result :cl)
901 (inst cmov :nbe result zero)
905 ;; The result-type ensures us that this shift will not overflow.
906 (inst shl result :cl)
910 ;;; Note: documentation for this function is wrong - rtfm
911 (define-vop (signed-byte-64-len)
912 (:translate integer-length)
913 (:note "inline (signed-byte 32) integer-length")
915 (:args (arg :scs (signed-reg) :target res))
916 (:arg-types signed-num)
917 (:results (res :scs (unsigned-reg)))
918 (:result-types unsigned-num)
933 (define-vop (unsigned-byte-64-len)
934 (:translate integer-length)
935 (:note "inline (unsigned-byte 32) integer-length")
937 (:args (arg :scs (unsigned-reg)))
938 (:arg-types unsigned-num)
939 (:results (res :scs (unsigned-reg)))
940 (:result-types unsigned-num)
951 (define-vop (unsigned-byte-64-count)
952 (:translate logcount)
953 (:note "inline (unsigned-byte 64) logcount")
955 (:args (arg :scs (unsigned-reg)))
956 (:arg-types unsigned-num)
957 (:results (result :scs (unsigned-reg)))
958 (:result-types positive-fixnum)
959 (:temporary (:sc unsigned-reg :from (:argument 0)) temp)
960 (:temporary (:sc unsigned-reg :from (:argument 0)) t1)
964 (inst mov temp result)
966 (inst and result #x55555555) ; note these masks will restrict the
967 (inst and temp #x55555555) ; count to the lower half of arg
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 ;;; now do the upper half
1000 (inst and t1 #x55555555)
1001 (inst and temp #x55555555)
1006 (inst and t1 #x33333333)
1007 (inst and temp #x33333333)
1012 (inst and t1 #x0f0f0f0f)
1013 (inst and temp #x0f0f0f0f)
1018 (inst and t1 #x00ff00ff)
1019 (inst and temp #x00ff00ff)
1024 (inst and t1 #x0000ffff)
1025 (inst and temp #x0000ffff)
1027 (inst add result t1)))
1031 ;;;; binary conditional VOPs
1033 (define-vop (fast-conditional)
1035 (:info target not-p)
1038 (:policy :fast-safe))
1040 ;;; constant variants are declared for 32 bits not 64 bits, because
1041 ;;; loading a 64 bit constant is silly
1043 (define-vop (fast-conditional/fixnum fast-conditional)
1044 (:args (x :scs (any-reg)
1045 :load-if (not (and (sc-is x control-stack)
1046 (sc-is y any-reg))))
1047 (y :scs (any-reg control-stack)))
1048 (:arg-types tagged-num tagged-num)
1049 (:note "inline fixnum comparison"))
1051 (define-vop (fast-conditional-c/fixnum fast-conditional/fixnum)
1052 (:args (x :scs (any-reg control-stack)))
1053 (:arg-types tagged-num (:constant (signed-byte 29)))
1054 (:info target not-p y))
1056 (define-vop (fast-conditional/signed fast-conditional)
1057 (:args (x :scs (signed-reg)
1058 :load-if (not (and (sc-is x signed-stack)
1059 (sc-is y signed-reg))))
1060 (y :scs (signed-reg signed-stack)))
1061 (:arg-types signed-num signed-num)
1062 (:note "inline (signed-byte 32) comparison"))
1064 (define-vop (fast-conditional-c/signed fast-conditional/signed)
1065 (:args (x :scs (signed-reg signed-stack)))
1066 (:arg-types signed-num (:constant (signed-byte 32)))
1067 (:info target not-p y))
1069 (define-vop (fast-conditional/unsigned fast-conditional)
1070 (:args (x :scs (unsigned-reg)
1071 :load-if (not (and (sc-is x unsigned-stack)
1072 (sc-is y unsigned-reg))))
1073 (y :scs (unsigned-reg unsigned-stack)))
1074 (:arg-types unsigned-num unsigned-num)
1075 (:note "inline (unsigned-byte 32) comparison"))
1077 (define-vop (fast-conditional-c/unsigned fast-conditional/unsigned)
1078 (:args (x :scs (unsigned-reg unsigned-stack)))
1079 (:arg-types unsigned-num (:constant (unsigned-byte 32)))
1080 (:info target not-p y))
1083 (macrolet ((define-conditional-vop (tran cond unsigned not-cond not-unsigned)
1086 (lambda (suffix cost signed)
1087 `(define-vop (;; FIXME: These could be done more
1088 ;; cleanly with SYMBOLICATE.
1089 ,(intern (format nil "~:@(FAST-IF-~A~A~)"
1092 (format nil "~:@(FAST-CONDITIONAL~A~)"
1097 ,(if (eq suffix '-c/fixnum)
1108 '(/fixnum -c/fixnum /signed -c/signed /unsigned -c/unsigned)
1109 ; '(/fixnum /signed /unsigned)
1111 '(t t t t nil nil)))))
1113 (define-conditional-vop < :l :b :ge :ae)
1114 (define-conditional-vop > :g :a :le :be))
1116 (define-vop (fast-if-eql/signed fast-conditional/signed)
1120 (inst jmp (if not-p :ne :e) target)))
1122 (define-vop (fast-if-eql-c/signed fast-conditional-c/signed)
1125 (cond ((and (sc-is x signed-reg) (zerop y))
1126 (inst test x x)) ; smaller instruction
1129 (inst jmp (if not-p :ne :e) target)))
1131 (define-vop (fast-if-eql/unsigned fast-conditional/unsigned)
1135 (inst jmp (if not-p :ne :e) target)))
1137 (define-vop (fast-if-eql-c/unsigned fast-conditional-c/unsigned)
1140 (cond ((and (sc-is x unsigned-reg) (zerop y))
1141 (inst test x x)) ; smaller instruction
1144 (inst jmp (if not-p :ne :e) target)))
1146 ;;; EQL/FIXNUM is funny because the first arg can be of any type, not just a
1149 ;;; These versions specify a fixnum restriction on their first arg. We have
1150 ;;; also generic-eql/fixnum VOPs which are the same, but have no restriction on
1151 ;;; the first arg and a higher cost. The reason for doing this is to prevent
1152 ;;; fixnum specific operations from being used on word integers, spuriously
1153 ;;; consing the argument.
1155 (define-vop (fast-eql/fixnum fast-conditional)
1156 (:args (x :scs (any-reg)
1157 :load-if (not (and (sc-is x control-stack)
1158 (sc-is y any-reg))))
1159 (y :scs (any-reg control-stack)))
1160 (:arg-types tagged-num tagged-num)
1161 (:note "inline fixnum comparison")
1165 (inst jmp (if not-p :ne :e) target)))
1166 (define-vop (generic-eql/fixnum fast-eql/fixnum)
1167 (:args (x :scs (any-reg descriptor-reg)
1168 :load-if (not (and (sc-is x control-stack)
1169 (sc-is y any-reg))))
1170 (y :scs (any-reg control-stack)))
1171 (:arg-types * tagged-num)
1175 (define-vop (fast-eql-c/fixnum fast-conditional/fixnum)
1176 (:args (x :scs (any-reg control-stack)))
1177 (:arg-types tagged-num (:constant (signed-byte 29)))
1178 (:info target not-p y)
1181 (cond ((and (sc-is x any-reg) (zerop y))
1182 (inst test x x)) ; smaller instruction
1184 (inst cmp x (fixnumize y))))
1185 (inst jmp (if not-p :ne :e) target)))
1187 (define-vop (generic-eql-c/fixnum fast-eql-c/fixnum)
1188 (:args (x :scs (any-reg descriptor-reg control-stack)))
1189 (:arg-types * (:constant (signed-byte 29)))
1192 ;;;; 32-bit logical operations
1194 (define-vop (merge-bits)
1195 (:translate merge-bits)
1196 (:args (shift :scs (signed-reg unsigned-reg) :target ecx)
1197 (prev :scs (unsigned-reg) :target result)
1198 (next :scs (unsigned-reg)))
1199 (:arg-types tagged-num unsigned-num unsigned-num)
1200 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 0)) ecx)
1201 (:results (result :scs (unsigned-reg) :from (:argument 1)))
1202 (:result-types unsigned-num)
1203 (:policy :fast-safe)
1207 (inst shrd result next :cl)))
1209 ;;; Only the lower 6 bits of the shift amount are significant.
1210 (define-vop (shift-towards-someplace)
1211 (:policy :fast-safe)
1212 (:args (num :scs (unsigned-reg) :target r)
1213 (amount :scs (signed-reg) :target ecx))
1214 (:arg-types unsigned-num tagged-num)
1215 (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
1216 (:results (r :scs (unsigned-reg) :from (:argument 0)))
1217 (:result-types unsigned-num))
1219 (define-vop (shift-towards-start shift-towards-someplace)
1220 (:translate shift-towards-start)
1221 (:note "SHIFT-TOWARDS-START")
1227 (define-vop (shift-towards-end shift-towards-someplace)
1228 (:translate shift-towards-end)
1229 (:note "SHIFT-TOWARDS-END")
1235 ;;;; Modular functions
1237 (define-modular-fun +-mod64 (x y) + 64)
1238 (define-vop (fast-+-mod64/unsigned=>unsigned fast-+/unsigned=>unsigned)
1239 (:translate +-mod64))
1240 (define-vop (fast-+-mod64-c/unsigned=>unsigned fast-+-c/unsigned=>unsigned)
1241 (:translate +-mod64))
1242 (define-modular-fun --mod64 (x y) - 64)
1243 (define-vop (fast---mod64/unsigned=>unsigned fast--/unsigned=>unsigned)
1244 (:translate --mod64))
1245 (define-vop (fast---mod64-c/unsigned=>unsigned fast---c/unsigned=>unsigned)
1246 (:translate --mod64))
1248 (define-modular-fun *-mod64 (x y) * 64)
1249 (define-vop (fast-*-mod64/unsigned=>unsigned fast-*/unsigned=>unsigned)
1250 (:translate *-mod64))
1251 ;;; (no -C variant as x86 MUL instruction doesn't take an immediate)
1253 (define-vop (fast-ash-left-mod64-c/unsigned=>unsigned
1254 fast-ash-c/unsigned=>unsigned)
1255 (:translate ash-left-mod64))
1259 (defknown sb!vm::%lea-mod64 (integer integer (member 1 2 4 8) (signed-byte 64))
1261 (foldable flushable movable))
1263 (define-modular-fun-optimizer %lea ((base index scale disp) :width width)
1264 (when (and (<= width 64)
1265 (constant-lvar-p scale)
1266 (constant-lvar-p disp))
1267 (cut-to-width base width)
1268 (cut-to-width index width)
1269 'sb!vm::%lea-mod64))
1272 (defun sb!vm::%lea-mod64 (base index scale disp)
1273 (ldb (byte 64 0) (%lea base index scale disp)))
1275 (defun sb!vm::%lea-mod64 (base index scale disp)
1276 (let ((base (logand base #xffffffffffffffff))
1277 (index (logand index #xffffffffffffffff)))
1278 ;; can't use modular version of %LEA, as we only have VOPs for
1279 ;; constant SCALE and DISP.
1280 (ldb (byte 64 0) (+ base (* index scale) disp))))
1282 (in-package "SB!VM")
1284 (define-vop (%lea-mod64/unsigned=>unsigned
1285 %lea/unsigned=>unsigned)
1286 (:translate %lea-mod64))
1288 ;;; logical operations
1289 (define-modular-fun lognot-mod64 (x) lognot 64)
1290 (define-vop (lognot-mod64/unsigned=>unsigned)
1291 (:translate lognot-mod64)
1292 (:args (x :scs (unsigned-reg unsigned-stack) :target r
1293 :load-if (not (and (sc-is x unsigned-stack)
1294 (sc-is r unsigned-stack)
1296 (:arg-types unsigned-num)
1297 (:results (r :scs (unsigned-reg)
1298 :load-if (not (and (sc-is x unsigned-stack)
1299 (sc-is r unsigned-stack)
1301 (:result-types unsigned-num)
1302 (:policy :fast-safe)
1307 (define-modular-fun logxor-mod64 (x y) logxor 64)
1308 (define-vop (fast-logxor-mod64/unsigned=>unsigned
1309 fast-logxor/unsigned=>unsigned)
1310 (:translate logxor-mod64))
1311 (define-vop (fast-logxor-mod64-c/unsigned=>unsigned
1312 fast-logxor-c/unsigned=>unsigned)
1313 (:translate logxor-mod64))
1315 (define-source-transform logeqv (&rest args)
1316 (if (oddp (length args))
1318 `(lognot (logxor ,@args))))
1319 (define-source-transform logandc1 (x y)
1320 `(logand (lognot ,x) ,y))
1321 (define-source-transform logandc2 (x y)
1322 `(logand ,x (lognot ,y)))
1323 (define-source-transform logorc1 (x y)
1324 `(logior (lognot ,x) ,y))
1325 (define-source-transform logorc2 (x y)
1326 `(logior ,x (lognot ,y)))
1327 (define-source-transform lognor (x y)
1328 `(lognot (logior ,x ,y)))
1329 (define-source-transform lognand (x y)
1330 `(lognot (logand ,x ,y)))
1334 (define-vop (bignum-length get-header-data)
1335 (:translate sb!bignum:%bignum-length)
1336 (:policy :fast-safe))
1338 (define-vop (bignum-set-length set-header-data)
1339 (:translate sb!bignum:%bignum-set-length)
1340 (:policy :fast-safe))
1342 (define-full-reffer bignum-ref * bignum-digits-offset other-pointer-lowtag
1343 (unsigned-reg) unsigned-num sb!bignum:%bignum-ref)
1345 (define-full-setter bignum-set * bignum-digits-offset other-pointer-lowtag
1346 (unsigned-reg) unsigned-num sb!bignum:%bignum-set)
1348 (define-vop (digit-0-or-plus)
1349 (:translate sb!bignum:%digit-0-or-plusp)
1350 (:policy :fast-safe)
1351 (:args (digit :scs (unsigned-reg)))
1352 (:arg-types unsigned-num)
1354 (:info target not-p)
1356 (inst or digit digit)
1357 (inst jmp (if not-p :s :ns) target)))
1360 ;;; For add and sub with carry the sc of carry argument is any-reg so
1361 ;;; the it may be passed as a fixnum or word and thus may be 0, 1, or
1362 ;;; 4. This is easy to deal with and may save a fixnum-word
1364 (define-vop (add-w/carry)
1365 (:translate sb!bignum:%add-with-carry)
1366 (:policy :fast-safe)
1367 (:args (a :scs (unsigned-reg) :target result)
1368 (b :scs (unsigned-reg unsigned-stack) :to :eval)
1369 (c :scs (any-reg) :target temp))
1370 (:arg-types unsigned-num unsigned-num positive-fixnum)
1371 (:temporary (:sc any-reg :from (:argument 2) :to :eval) temp)
1372 (:results (result :scs (unsigned-reg) :from (:argument 0))
1373 (carry :scs (unsigned-reg)))
1374 (:result-types unsigned-num positive-fixnum)
1378 (inst neg temp) ; Set the carry flag to 0 if c=0 else to 1
1381 (inst adc carry carry)))
1383 ;;; Note: the borrow is the oppostite of the x86 convention - 1 for no
1384 ;;; borrow and 0 for a borrow.
1385 (define-vop (sub-w/borrow)
1386 (:translate sb!bignum:%subtract-with-borrow)
1387 (:policy :fast-safe)
1388 (:args (a :scs (unsigned-reg) :to :eval :target result)
1389 (b :scs (unsigned-reg unsigned-stack) :to :result)
1390 (c :scs (any-reg control-stack)))
1391 (:arg-types unsigned-num unsigned-num positive-fixnum)
1392 (:results (result :scs (unsigned-reg) :from :eval)
1393 (borrow :scs (unsigned-reg)))
1394 (:result-types unsigned-num positive-fixnum)
1396 (inst cmp c 1) ; Set the carry flag to 1 if c=0 else to 0
1400 (inst adc borrow borrow)
1401 (inst xor borrow 1)))
1404 (define-vop (bignum-mult-and-add-3-arg)
1405 (:translate sb!bignum:%multiply-and-add)
1406 (:policy :fast-safe)
1407 (:args (x :scs (unsigned-reg) :target eax)
1408 (y :scs (unsigned-reg unsigned-stack))
1409 (carry-in :scs (unsigned-reg unsigned-stack)))
1410 (:arg-types unsigned-num unsigned-num unsigned-num)
1411 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
1412 :to (:result 1) :target lo) eax)
1413 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
1414 :to (:result 0) :target hi) edx)
1415 (:results (hi :scs (unsigned-reg))
1416 (lo :scs (unsigned-reg)))
1417 (:result-types unsigned-num unsigned-num)
1421 (inst add eax carry-in)
1426 (define-vop (bignum-mult-and-add-4-arg)
1427 (:translate sb!bignum:%multiply-and-add)
1428 (:policy :fast-safe)
1429 (:args (x :scs (unsigned-reg) :target eax)
1430 (y :scs (unsigned-reg unsigned-stack))
1431 (prev :scs (unsigned-reg unsigned-stack))
1432 (carry-in :scs (unsigned-reg unsigned-stack)))
1433 (:arg-types unsigned-num unsigned-num unsigned-num unsigned-num)
1434 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
1435 :to (:result 1) :target lo) eax)
1436 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
1437 :to (:result 0) :target hi) edx)
1438 (:results (hi :scs (unsigned-reg))
1439 (lo :scs (unsigned-reg)))
1440 (:result-types unsigned-num unsigned-num)
1446 (inst add eax carry-in)
1452 (define-vop (bignum-mult)
1453 (:translate sb!bignum:%multiply)
1454 (:policy :fast-safe)
1455 (:args (x :scs (unsigned-reg) :target eax)
1456 (y :scs (unsigned-reg unsigned-stack)))
1457 (:arg-types unsigned-num unsigned-num)
1458 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
1459 :to (:result 1) :target lo) eax)
1460 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
1461 :to (:result 0) :target hi) edx)
1462 (:results (hi :scs (unsigned-reg))
1463 (lo :scs (unsigned-reg)))
1464 (:result-types unsigned-num unsigned-num)
1471 (define-vop (bignum-lognot lognot-mod64/unsigned=>unsigned)
1472 (:translate sb!bignum:%lognot))
1474 (define-vop (fixnum-to-digit)
1475 (:translate sb!bignum:%fixnum-to-digit)
1476 (:policy :fast-safe)
1477 (:args (fixnum :scs (any-reg control-stack) :target digit))
1478 (:arg-types tagged-num)
1479 (:results (digit :scs (unsigned-reg)
1480 :load-if (not (and (sc-is fixnum control-stack)
1481 (sc-is digit unsigned-stack)
1482 (location= fixnum digit)))))
1483 (:result-types unsigned-num)
1486 (inst sar digit 3)))
1488 (define-vop (bignum-floor)
1489 (:translate sb!bignum:%floor)
1490 (:policy :fast-safe)
1491 (:args (div-high :scs (unsigned-reg) :target edx)
1492 (div-low :scs (unsigned-reg) :target eax)
1493 (divisor :scs (unsigned-reg unsigned-stack)))
1494 (:arg-types unsigned-num unsigned-num unsigned-num)
1495 (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)
1496 :to (:result 0) :target quo) eax)
1497 (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 0)
1498 :to (:result 1) :target rem) edx)
1499 (:results (quo :scs (unsigned-reg))
1500 (rem :scs (unsigned-reg)))
1501 (:result-types unsigned-num unsigned-num)
1505 (inst div eax divisor)
1509 (define-vop (signify-digit)
1510 (:translate sb!bignum:%fixnum-digit-with-correct-sign)
1511 (:policy :fast-safe)
1512 (:args (digit :scs (unsigned-reg unsigned-stack) :target res))
1513 (:arg-types unsigned-num)
1514 (:results (res :scs (any-reg signed-reg)
1515 :load-if (not (and (sc-is digit unsigned-stack)
1516 (sc-is res control-stack signed-stack)
1517 (location= digit res)))))
1518 (:result-types signed-num)
1521 (when (sc-is res any-reg control-stack)
1524 (define-vop (digit-ashr)
1525 (:translate sb!bignum:%ashr)
1526 (:policy :fast-safe)
1527 (:args (digit :scs (unsigned-reg unsigned-stack) :target result)
1528 (count :scs (unsigned-reg) :target ecx))
1529 (:arg-types unsigned-num positive-fixnum)
1530 (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
1531 (:results (result :scs (unsigned-reg) :from (:argument 0)
1532 :load-if (not (and (sc-is result unsigned-stack)
1533 (location= digit result)))))
1534 (:result-types unsigned-num)
1538 (inst sar result :cl)))
1540 (define-vop (digit-lshr digit-ashr)
1541 (:translate sb!bignum:%digit-logical-shift-right)
1545 (inst shr result :cl)))
1547 (define-vop (digit-ashl digit-ashr)
1548 (:translate sb!bignum:%ashl)
1552 (inst shl result :cl)))
1554 ;;;; static functions
1556 (define-static-fun two-arg-/ (x y) :translate /)
1558 (define-static-fun two-arg-gcd (x y) :translate gcd)
1559 (define-static-fun two-arg-lcm (x y) :translate lcm)
1561 (define-static-fun two-arg-and (x y) :translate logand)
1562 (define-static-fun two-arg-ior (x y) :translate logior)
1563 (define-static-fun two-arg-xor (x y) :translate logxor)
1568 ;;; This is essentially a straight implementation of the algorithm in
1569 ;;; "Strength Reduction of Multiplications by Integer Constants",
1570 ;;; Youfeng Wu, ACM SIGPLAN Notices, Vol. 30, No.2, February 1995.
1571 (defun basic-decompose-multiplication (arg num n-bits condensed)
1572 (case (aref condensed 0)
1574 (let ((tmp (min 3 (aref condensed 1))))
1575 (decf (aref condensed 1) tmp)
1578 ,(decompose-multiplication
1579 arg (ash (1- num) (- tmp)) (1- n-bits) (subseq condensed 1))
1582 (let ((r0 (aref condensed 0)))
1583 (incf (aref condensed 1) r0)
1585 (%lea ,(decompose-multiplication
1586 arg (- num (ash 1 r0)) (1- n-bits) (subseq condensed 1))
1589 (t (let ((r0 (aref condensed 0)))
1590 (setf (aref condensed 0) 0)
1592 (ash ,(decompose-multiplication
1593 arg (ash num (- r0)) n-bits condensed)
1596 (defun decompose-multiplication (arg num n-bits condensed)
1601 `(logand #xffffffff (ash ,arg ,(1- (integer-length num)))))
1602 ((let ((max 0) (end 0))
1603 (loop for i from 2 to (length condensed)
1604 for j = (reduce #'+ (subseq condensed 0 i))
1605 when (and (> (- (* 2 i) 3 j) max)
1606 (< (+ (ash 1 (1+ j))
1607 (ash (ldb (byte (- 64 (1+ j)) (1+ j)) num)
1610 do (setq max (- (* 2 i) 3 j)
1613 (let ((j (reduce #'+ (subseq condensed 0 end))))
1614 (let ((n2 (+ (ash 1 (1+ j))
1615 (ash (ldb (byte (- 64 (1+ j)) (1+ j)) num) (1+ j))))
1616 (n1 (1+ (ldb (byte (1+ j) 0) (lognot num)))))
1618 (- ,(optimize-multiply arg n2) ,(optimize-multiply arg n1))))))))
1619 ((dolist (i '(9 5 3))
1620 (when (integerp (/ num i))
1621 (when (< (logcount (/ num i)) (logcount num))
1623 (return `(let ((,x ,(optimize-multiply arg (/ num i))))
1625 (%lea ,x ,x (1- ,i) 0)))))))))
1626 (t (basic-decompose-multiplication arg num n-bits condensed))))
1628 (defun optimize-multiply (arg x)
1629 (let* ((n-bits (logcount x))
1630 (condensed (make-array n-bits)))
1631 (let ((count 0) (bit 0))
1633 (cond ((logbitp i x)
1634 (setf (aref condensed bit) count)
1638 (decompose-multiplication arg x n-bits condensed)))
1640 (defun *-transformer (y)
1642 (t (give-up-ir1-transform))
1643 ((= y (ash 1 (integer-length y)))
1644 ;; there's a generic transform for y = 2^k
1645 (give-up-ir1-transform))
1646 ((member y '(3 5 9))
1647 ;; we can do these multiplications directly using LEA
1648 `(%lea x x ,(1- y) 0))
1649 ((member :pentium4 *backend-subfeatures*)
1650 ;; the pentium4's multiply unit is reportedly very good
1651 (give-up-ir1-transform))
1652 ;; FIXME: should make this more fine-grained. If nothing else,
1653 ;; there should probably be a cutoff of about 9 instructions on
1654 ;; pentium-class machines.
1655 (t (optimize-multiply 'x y))))
1657 (deftransform * ((x y)
1658 ((unsigned-byte 64) (constant-arg (unsigned-byte 64)))
1660 "recode as leas, shifts and adds"
1661 (let ((y (lvar-value y)))
1664 (deftransform sb!vm::*-mod64
1665 ((x y) ((unsigned-byte 64) (constant-arg (unsigned-byte 64)))
1667 "recode as leas, shifts and adds"
1668 (let ((y (lvar-value y)))
1671 ;;; FIXME: we should also be able to write an optimizer or two to
1672 ;;; convert (+ (* x 2) 17), (- (* x 9) 5) to a %LEA.