source: branches/ia32/level-0/X86/X8632/x8632-bignum.lisp @ 7963

(in-package "CCL")

;;; %BIGNUM-REF needs to access bignums as obviously as possible, and it needs
;;; to be able to return 32 bits somewhere no one looks for real objects.
;;;
;;; The easiest thing to do is to store the 32 raw bits in two fixnums
;;; and return multiple values.
;;;
;;; XXX -- figure out how we're going to handle multiple-values...
(defx8632lapfunction %bignum-ref ((bignum arg_y) (i arg_z))
  (movl (% esp) (% temp0))
  (movzwl (@ (+ 2 x8632::misc-data-offset) (% bignum) (% i)) (% imm0))
  (box-fixnum imm0 temp1)
  (push (% temp1))                      ;high
  (movzwl (@ x8632::misc-data-offset (% bignum) (% i)) (% imm0))
  (box-fixnum imm0 temp1)
  (push (% temp1))                      ;low
  (set-nargs 2)
  (jmp-subprim .SPvalues))

;;; BIGNUM[I] := DIGIT[0]
(defx8632lapfunction %set-digit ((bignum 4) #|(ra 0)|# (i arg_y) (digit arg_z))
  (movl (@ bignum (% esp)) (% temp0))
  (svref digit 0 imm0)
  (movl (% imm0) (@ x8632::misc-data-offset (% temp0) (% i)))
  (single-value-return 3))

;;; Return the sign of bignum (0 or -1) as a fixnum
(defx8632lapfunction %bignum-sign ((bignum arg_z))
  (vector-length bignum imm0)
  (movl (@ (- x8632::misc-data-offset 4) (% bignum) (% imm0)) (% imm0))
  (sarl ($ 31) (% imm0))                ;propagate sign bit
  (box-fixnum imm0 arg_z)
  (single-value-return))

;;; Count the sign bits in the most significant digit of bignum;
;;; return fixnum count.
(defx8632lapfunction %bignum-sign-bits ((bignum arg_z))
  (vector-length bignum imm0)
  (movl (@ (- x8632::misc-data-offset 4) (% bignum) (% imm0)) (% imm0))
  (mark-as-imm temp0)
  (movl (% imm0) (% temp0))
  (notl (% imm0))
  (testl (% temp0) (% temp0))
  (js @wasneg)
  (notl (% imm0))
  @wasneg
  (bsrl (% imm0) (% imm0))
  (sete (% temp0.b))
  (xorl ($ 31) (% imm0))
  (addb (% temp0.b) (% imm0.b))
  (box-fixnum imm0 arg_z)
  (mark-as-node temp0)
  (single-value-return))

(defx8632lapfunction %digit-0-or-plusp ((bignum arg_y) (idx arg_z))
  (movl (@ x8632::misc-data-offset (% bignum) (% idx)) (% imm0))
  (movl ($ x8632::nil-value) (% temp0))
  (leal (@ x8632::t-offset (% temp0)) (% arg_z))
  (testl (% imm0) (% imm0))
  (cmovll (% temp0) (% arg_z))
  (single-value-return))

;;; For oddp, evenp
(defx8632lapfunction %bignum-oddp ((bignum arg_z))
  (movl (@ x8632::misc-data-offset (% bignum)) (% imm0))
  (movl ($ x8632::nil-value) (% temp0))
  (leal (@ x8632::t-offset (% temp0)) (% arg_z))
  (testb ($ 1) (% imm0.b))
  (cmovzl (% temp0) (% arg_z))
  (single-value-return))

(defx8632lapfunction bignum-plusp ((bignum arg_z))
  (vector-length bignum imm0)
  (movl (@ (- x8632::misc-data-offset 4) (% bignum) (% imm0)) (% imm0))
  (movl ($ x8632::nil-value) (% temp0))
  (lea (@ x8632::t-offset (% temp0)) (% arg_z))
  (testl (% imm0) (% imm0))
  (cmovlel (% temp0) (% arg_z))
  (single-value-return))

(defx8632lapfunction bignum-minusp ((bignum arg_z))
  (vector-length bignum imm0)
  (movl (@ (- x8632::misc-data-offset 4) (% bignum) (% imm0)) (% imm0))
  (movl ($ x8632::nil-value) (% temp0))
  (lea (@ x8632::t-offset (% temp0)) (% arg_z))
  (testl (% imm0) (% imm0))
  (cmovgl (% temp0) (% arg_z))
  (single-value-return))

;;; Add the digits A[I] and B[J], and the incoming carry C (a fixnum,
;;; either 0 or 1).  Store the result in R[K], and return the outgoing
;;; carry.  If I is NIL, A is a fixnum.  If J is NIL, B is a fixnum.
(defx8632lapfunction %add-with-carry ((r 20) (k 16) (c 12) (a 8) (i 4) #|(ra 0)|# (b arg_y) (j arg_z))
  (mark-as-imm temp0)
  (unbox-fixnum b imm0)
  (cmpl ($ x8632::nil-value) (% j))
  ;; if j not nil, get b[j]
  (cmovnel (@ x8632::misc-data-offset (% b) (% j)) (% imm0))
  (movl (@ a (% esp)) (% arg_y))
  (unbox-fixnum arg_y temp0)
  (movl (@ i (% esp)) (% arg_z))
  (cmpl ($ x8632::nil-value) (% arg_z))
  ;; if i not nil, get a[i]
  (cmovnel (@ x8632::misc-data-offset (% arg_y) (% arg_z)) (% temp0))
  (movl ($ '1) (% arg_z))               ;for outgoing carry
  (xorl (% arg_y) (% arg_y))
  ;; I can't think of a better way to set CF at the moment.
  ;; NEG would be ideal, but we don't have a free imm reg.
  (btl ($ x8632::fixnumshift) (@ c (% esp))) ;CF = lsb of carry fixnum 
  (adc (% temp0) (% imm0))
  (mark-as-node temp0)
  (movl (@ r (% esp)) (% temp0))
  (movl (@ k (% esp)) (% temp1))
  (movl (% imm0) (@ x8632::misc-data-offset (% temp0) (% temp1)))
  (cmovnc (% arg_y) (% arg_z))          ;zero outgoing carry if CF = 0
  (single-value-return 7))

;;; Add the digits A[I] and B[J], and the incoming carry C (a fixnum).
;;; Store the result in R[K], and return the outgoing carry.  If I is
;;; NIL, A is a fixnum.  If J is NIL, B is a fixnum.
#+sse2
(defx8632lapfunction %add-with-carry ((r 20) (k 16) (c 12) (a 8) (i 4) #|(ra 0)|# (b arg_y) (j arg_z))
  (let ((aa mm2)
        (bb mm3)
        (cc mm4))
    (unbox-fixnum b imm0)               ;assume j will be nil
    (cmpl ($ x8632::nil-value) (% j))
    ;; if j not nil, get b[j]
    (cmovnel (@ x8632::misc-data-offset (% b) (% j)) (% imm0))
    (movd (% imm0) (% bb))
    (movl (@ a (% esp)) (% arg_y))
    (movl (@ i (% esp)) (% arg_z))
    (movl (@ c (% esp)) (% temp0))
    (unbox-fixnum arg_y imm0)           ;assume i will be nil
    (cmpl ($ x8632::nil-value) (% arg_z))
    ;; if i not nil, get a[i]
    (cmovnel (@ x8632::misc-data-offset (% arg_y) (% arg_z)) (% imm0))
    (movd (% imm0) (% aa))
    (unbox-fixnum temp0 imm0)
    (movd (% imm0) (% cc))
    (paddq (% xx) (% yy))
    (paddq (% cc) (% yy))
    (movl (@ r (% esp)) (% temp0))
    (movl (@ k (% esp)) (% temp1))
    (movd (% yy) (@ x8632::misc-data-offset (% temp0) (% temp1)))
    (psrlq ($ 32) (% yy))               ;carry bit
    (movd (% yy) (% imm0))
    (box-fixnum imm0 arg_z)
    (single-value-return 7)))

;;; Store the result of A[I] - B[J] - borrow into R[K], returning the borrow.
;;; If I is NIL, A is a fixnum; likewise for J and B.
;;;
;;; (a - b) - (1 - borrow), or equivalently, (a - b) + borrow - 1
;;; 
;;; Note: borrow is 1 for no borrow and 0 for a borrow.
(defx8632lapfunction %subtract-with-borrow ((r 20) (k 16) (borrow 12) (a 8) (i 4) #|(ra 0)|# (b arg_y) (j arg_z))
  (mark-as-imm temp0)
  (unbox-fixnum b imm0)
  (cmpl ($ x8632::nil-value) (% j))
  (cmovnel (@ x8632::misc-data-offset (% b) (% j)) (% imm0))
  (movl (@ a (% esp)) (% arg_y))
  (unbox-fixnum arg_y temp0)
  (movl (@ i (% esp)) (% arg_z))
  (cmpl ($ x8632::nil-value) (% arg_z))
  (cmovnel (@ x8632::misc-data-offset (% arg_y) (% arg_z)) (% temp0))
  ;; unboxed a or a[i] in temp0, unboxed b or b[j] in imm0
  (cmpl ($ '1) (@ borrow (% esp)))      ;CF = 1 if borrow is 0 else CF = 0
  (sbb (% imm0) (% temp0))
  (movl ($ 1) (% imm0))
  (sbb ($ 0) (% imm0))
  (box-fixnum imm0 arg_z)
  (movl (% temp0) (% imm0))
  (mark-as-node temp0)
  (movl (@ r (% esp)) (% temp0))
  (movl (@ k (% esp)) (% temp1))
  (movl (% imm0) (@ x8632::misc-data-offset (% temp0) (% temp1)))
  (single-value-return 7))

#+sse2
(defx8632lapfunction %subtract-with-borrow ((r 20) (k 16) (borrow 12) (a 8) (i 4) #|(ra 0)|# (b arg_y) (j arg_z))
  (let ((aa mm2)
        (bb mm3)
        (ww mm4))
    (unbox-fixnum b imm0)
    (cmpl ($ x8632::nil-value) (% j))
    ;; if j not nil, get b[j]
    (cmovnel (@ x8632::misc-data-offset (% b) (% j)) (% imm0))
    (movd (% imm0) (% bb))
    (movl (@ a (% esp)) (% arg_y))
    (movl (@ i (% esp)) (% arg_z))
    (movl (@ borrow (% esp)) (% temp0))
    (unbox-fixnum arg_y imm0)
    (cmpl ($ x8632::nil-value) (% arg_z))
    ;; if i not nil, get a[i]
    (cmovnel (@ x8632::misc-data-offset (% arg_y) (% arg_z)) (% imm0))
    (movd (% imm0) (% aa))
    (unbox-fixnum temp0 imm0)
    (subl ($ 1) (% imm0))
    (movd (% imm0) (% ww))
    (psubq (% bb) (% aa))
    (paddq (% ww) (% aa))
    (movl (@ r (% esp)) (% temp0))
    (movl (@ k (% esp)) (% temp1))
    (movd (% aa) (@ x8632::misc-data-offset (% temp0) (% temp1)))
    (psrlq ($ 32) (% aa))               ;carry digit
    (movd (% aa) (% imm0))
    (xorl (% arg_z) (% arg_z))
    (test ($ 1) (% imm0))
    (cmovzl ($ '1) (% arg_z))
    (single-value-return 7)))

(defx8632lapfunction %subtract-one ((high arg_y) (low arg_z))
  (mark-as-imm temp0)
  (unbox-fixnum low imm0)
  (movl (% high) (% temp0))
  (shll ($ (- 16 x8632::fixnumshift)) (% temp0))
  (orl (% imm0) (% temp0))
  (subl ($ 1) (% temp0))
  (movzwl (% temp0.w) (% imm0))
  (box-fixnum imm0 low)
  (sarl ($ 16) (% temp0))
  (box-fixnum temp0 high)
  (mark-as-node temp0)
  (push (% high))
  (push (% low))
  (movl (% esp) (% temp0))
  (jmp-subprim .SPvalues))

;;; %SUBTRACT-WITH-BORROW -- Internal.
;;;
;;; This should be in assembler, and should not cons intermediate results.  It
;;; returns a 32bit digit and a borrow resulting from subtracting b from a, and
;;; subtracting a possible incoming borrow.
;;;
;;; We really do:  a - b - 1 + borrow, where borrow is either 0 or 1.
;;; 

(defx8632lapfunction %subtract-with-borrow-1 ((a-h 12) (a-l 8) (b-h 4) #|(ra 0)|# (b-l arg_y) (borrow arg_z))
  (mark-as-imm temp0)
  (mark-as-imm temp1)
  (unbox-fixnum b-l temp0)
  (movl (@ b-h (% esp)) (% imm0))
  (unbox-fixnum imm0 imm0)
  (shll ($ 16) (% imm0))
  (orl (% imm0) (% temp0))
  (movl (@ a-l (% esp)) (% temp1))
  (unbox-fixnum temp1 temp1)
  (movl (@ a-h (% esp)) (% imm0))
  (unbox-fixnum imm0 imm0)
  (shll ($ 16) (% imm0))
  (orl (% imm0) (% temp1))
  (cmpl ($ '1) (@ borrow (% esp)))      ;CF = 1 if borrow is 0 else CF = 0
  (sbbl (% temp0) (% temp1))
  (movl ($ 1) (% imm0))
  (subb ($ 0) (% imm0))
  (box-fixnum imm0 arg_z)
  (movzwl (% temp1.w) (% imm0))
  (box-fixnum imm0 imm0)
  (push (% imm0))                       ;high
  (sarl ($ 16) (% temp1))
  (box-fixnum temp1 imm0)
  (mark-as-node temp0)
  (mark-as-node temp1)
  (push (% imm0))                       ;low
  (push (% arg_z))                      ;borrow
  (set-nargs 3)
  (popl (@ 16 (% esp)))                 ;relocate return addr
  (addl ($ '4) (% esp))                 ;discard args, part of saved frame
  (movl (% esp) (% temp0))
  (jmp-subprim .SPvalues))
  

;;; To normalize a bignum is to drop "trailing" digits which are
;;; redundant sign information.  When return-fixnum-p is non-nil, make
;;; the resultant bignum into a fixnum if it fits.
(defx8632lapfunction %normalize-bignum-2 ((return-fixnum-p arg_y) (bignum arg_z))
  (push (% return-fixnum-p))
  (mark-as-imm temp0)
  (mark-as-imm temp1)
  (let ((len arg_y)
        (sign temp0)
        (next temp1))
    (vector-length bignum len)
    (cmpl ($ '1) (% len))
    (jle @maybe-return-fixnum)
    ;; Zero trailing sign digits.
    (push (% len))
    ;; next-to-last digit
    (movl (@ (- x8632::misc-data-offset 8) (% bignum) (% len)) (% next))
    ;; last digit
    (movl (@ (- x8632::misc-data-offset 4) (% bignum) (% len)) (% sign))
    (jmp @test)
    @loop
    (subl ($ '1) (% len))
    (movl ($ 0) (@ x8632::misc-data-offset (% bignum) (% len)))
    (cmpl ($ '1) (% len))               ;any more digits?
    (je @adjust-length)
    (movl (% next) (% sign))
    ;; (bignum-ref bignum (- len 2))
    (movl (@ (- x8632::misc-data-offset 8) (% bignum) (% len)) (% next))
    @test
    (movl (% next) (% imm0))
    (sarl ($ 31) (% imm0))              ;propagate sign bit
    (xorl (% sign) (% imm0))            ;whole digit only sign?
    (jz @loop)
    ;; New length now in len.
    @adjust-length
    (pop (% imm0))                      ;original length
    (cmpl (% len) (% imm0))
    ;; If the new length is the same as the original length, we know
    ;; that the bignum is at least two digits long (because if it was
    ;; shorter, we would have branched directly to
    ;; @maybe-return-fixnum), and thus won't fit in a fixnum.
    ;; Therefore, there's no need to do either of the tests at
    ;; @maybe-return-fixnum.
    (je @done)
    (movl (% len) (% imm0))
    (shll ($ (- x8632::num-subtag-bits x8632::fixnumshift)) (% imm0))
    (movb ($ x8632::subtag-bignum) (% imm0.b))
    (movl (% imm0) (@ x8632::misc-header-offset (% bignum)))
    @maybe-return-fixnum
    ;; could use SETcc here to avoid one branch
    (cmpl ($ x8632::nil-value) (@ 0 (% esp))) ;return-fixnum-p
    (je @done)
    (cmpl ($ x8632::one-digit-bignum-header) (% bignum))
    (jne @done)
    ;; Bignum has one digit.  If it fits in a fixnum, return a fixnum.
    (movl (@ x8632::misc-data-offset (% bignum)) (% imm0))
    (box-fixnum imm0 arg_y)
    (unbox-fixnum arg_y temp0)
    (cmpl (% temp0) (% imm0))
    (cmovel (% arg_y) (% arg_z))
    @done
    (pop (% imm0))                      ;discard saved return-fixnum-p
    (mark-as-node temp0)
    (mark-as-node temp1)
    (single-value-return)))

;;; Multiply X[I] by the unboxed value of the (non-negative) fixnum Y;
;;; add the incoming carry from CARRY[0] to the 64-bit product.  Store
;;; the low word of the 64-bit sum in R[0] and the high word in
;;; CARRY[0].
(defx8632lapfunction %multiply-and-add ((r 12) (carry 8) (x 4) #|(ra 0)|# (i arg_y) (y arg_z))
  (let ((xx mm2)
        (yy mm3)
        (cc mm4))
    (movl (@ x (% esp)) (% imm0))
    (movd (@ x8632::misc-data-offset (% imm0) (% i)) (% xx))
    (unbox-fixnum y imm0)
    (movd (% imm0) (% yy))
    (pmuludq (% xx) (% yy))             ;64 bit product
    (movl (@ carry (% esp)) (% arg_y))
    (movd (@ x8632::misc-data-offset (% arg_y)) (% cc))
    (paddq (% cc) (% yy))               ;add in 32 bit carry digit
    (movl (@ r (% esp)) (% arg_z))
    (movd (% yy) (@ x8632::misc-data-offset (% arg_z)))
    (psrlq ($ 32) (% yy))
    (movd (% yy) (@ x8632::misc-data-offset (% arg_y)))
    (single-value-return 5)))

;; multiply x[i] by y and add to result starting at digit i
(defx8632lapfunction %multiply-and-add-harder-loop-2
    ((x 12) (y 8) (r 4) #|(ra 0)|# (i arg_y) (ylen arg_z))
  (let ((cc mm2)
        (xx mm3)
        (yy mm4)
        (j imm0))
    (movl (@ x (% esp)) (% temp0))
    (movd (@ x8632::misc-data-offset (% temp0) (% i)) (% xx)) ;x[i]
    (movl (@ y (% esp)) (% temp0))
    (movl (@ r (% esp)) (% temp1))
    (pxor (% cc) (% cc))
    (xorl (% j) (% j))
    @loop
    (movd (@ x8632::misc-data-offset (% temp0) (% j)) (% yy)) ;y[j]
    (pmuludq (% xx) (% yy))
    (paddq (% cc) (% yy))
    (movd (% yy) (@ x8632::misc-data-offset (% temp1) (% i))) ;store r[i]
    (movq (% yy) (% cc))
    (psrlq ($ 32) (% cc))               ;carry high digit
    (addl ($ '1) (% i))
    (addl ($ '1) (% j))
    (subl ($ '1) (% ylen))
    (jg @loop)
    (movd (% cc) (@ x8632::misc-data-offset (% temp1) (% i)))
    (single-value-return 5)))

;; this is silly  
(defx8632lapfunction %add-the-carry ((high 4) #|(ra 0)|# (low arg_y) (c arg_z))
  (mark-as-imm temp0)
  (unbox-fixnum low imm0)
  (movl (@ high (% esp)) (% temp0))
  (shll ($ (- 16 x8632::fixnumshift)) (% temp0))
  (orl (% imm0) (% temp0))
  (unbox-fixnum c imm0)
  (addl (% imm0) (% temp0))
  (movzwl (% temp0.w) (% imm0))
  (box-fixnum imm0 temp1)
  (sarl ($ 16) (% temp0))
  (box-fixnum temp0 temp0)
  (push (% temp0))                      ;high
  (push (% temp1))                      ;low
  (set-nargs 2)
  (mark-as-node temp0)
  (movl (% esp) (% temp0))
  (jmp-subprim .SPvalues))

(defx8632lapfunction %bignum-count-trailing-zero-bits ((bignum arg_z))
  (let ((i arg_y)
        (len temp0)
        (zeros temp1))
    (vector-length bignum temp0)
    (xorl (% i) (% i))
    (movl ($ '32) (% zeros))
    @loop
    (movl (@ x8632::misc-data-offset (% bignum) (% i)) (% imm0))
    (addl ($ '1) (% i))
    (addl ($ '32) (% zeros))
    (testl (% imm0) (% imm0))
    (jz @loop)
    ;; now count zero bits in digit
    (bsrl (% imm0) (% imm0))
    (box-fixnum imm0 imm0)
    (addl (% imm0) (% zeros))
    (movl (% zeros) (% arg_z))
    (single-value-return)))

;;; dest[i] = (logand x[i] y[i])
(defx8632lapfunction %bignum-logand ((idx 8) (x 4) #|(ra 0)|# (y arg_y) (dest arg_z))
  (let ((i temp0)
        (xx temp1)
        (yy arg_y))
    (movl (@ idx (% esp)) (% i))
    (movl (@ x (% esp)) (% xx))
    (movl (@ x8632::misc-data-offset (% xx) (% i)) (% imm0))
    (andl (@ x8632::misc-data-offset (% yy) (% i)) (% imm0))
    (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% i)))
    (single-value-return 4)))

;;; dest[i] = (logandc1 x[i] y[i])
(defx8632lapfunction %bignum-logandc1 ((idx 8) (x 4) #|(ra 0)|# (y arg_y) (dest arg_z))
  (let ((i temp0)
        (xx temp1)
        (yy arg_y))
    (movl (@ idx (% esp)) (% i))
    (movl (@ x (% esp)) (% xx))
    (movl (@ x8632::misc-data-offset (% xx) (% i)) (% imm0))
    (not (% imm0))
    (andl (@ x8632::misc-data-offset (% yy) (% i)) (% imm0))
    (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% i)))
    (single-value-return 4)))

;;; dest[i] = (logandc2 x[i] y[i])
(defx8632lapfunction %bignum-logandc2 ((idx 8) (x 4) #|(ra 0)|# (y arg_y) (dest arg_z))
  (let ((i temp0)
        (xx temp1)
        (yy arg_y))
    (movl (@ idx (% esp)) (% i))
    (movl (@ x (% esp)) (% xx))
    (movl (@ x8632::misc-data-offset (% yy) (% i)) (% imm0))
    (not (% imm0))
    (andl (@ x8632::misc-data-offset (% xx) (% i)) (% imm0))
    (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% i)))
    (single-value-return 4)))

;;; dest[i] = (logior x[i] y[i])
(defx8632lapfunction %bignum-logior ((idx 8) (x 4) #|(ra 0)|# (y arg_y) (dest arg_z))
  (let ((i temp0)
        (xx temp1)
        (yy arg_y))
    (movl (@ idx (% esp)) (% i))
    (movl (@ x (% esp)) (% xx))
    (movl (@ x8632::misc-data-offset (% xx) (% i)) (% imm0))
    (orl (@ x8632::misc-data-offset (% yy) (% i)) (% imm0))
    (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% i)))
    (single-value-return 4)))

;;; dest[i] = (lognot x[i])
(defx8632lapfunction %bignum-logior ((idx 4) #|(ra 0)|# (x arg_y) (dest arg_z))
  (let ((i temp0))
    (movl (@ idx (% esp)) (% i))
    (movl (@ x8632::misc-data-offset (% x) (% i)) (% imm0))
    (not (% imm0))
    (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% i)))
    (single-value-return 3)))

;;; dest[i] = (logxor x[i] y[i])
(defx8632lapfunction %bignum-logxor ((idx 8) (x 4) #|(ra 0)|# (y arg_y) (dest arg_z))
  (let ((i temp0)
        (xx temp1)
        (yy arg_y))
    (movl (@ idx (% esp)) (% i))
    (movl (@ x (% esp)) (% xx))
    (movl (@ x8632::misc-data-offset (% xx) (% i)) (% imm0))
    (xorl (@ x8632::misc-data-offset (% yy) (% i)) (% imm0))
    (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% i)))
    (single-value-return 4)))

(defx8632lapfunction %compare-digits ((a 4) #|(ra 0)|# (b arg_y) (i arg_z))
  (movl (@ a (% esp)) (% temp0))
  (movl (@ x8632::misc-data-offset (% temp0) (% i)) (% imm0))
  (xorl (% temp0) (% temp0))
  (movl ($ '1) (% temp1))
  (movl ($ '-1) (% arg_y))
  (cmpl (@ x8632::misc-data-offset (% b) (% i)) (% imm0))
  (cmovzl (% temp0) (% arg_z))
  (cmovgl (% temp1) (% arg_z))
  (cmovll (% arg_y) (% arg_z))
  (single-value-return 3))

;; returns number of bits in digit-hi,digit-lo that are sign bits
;; 32 - digits-sign-bits is integer-length
(defx8632lapfunction %digits-sign-bits ((hi arg_y) (lo arg_z))
  (mark-as-imm temp0)
  (shll ($ (- 16 x8632::fixnumshift)) (% hi))
  (unbox-fixnum lo imm0)
  (orl (% hi) (% imm0))
  (movl (% imm0) (% temp0))
  (not (% imm0))
  (testl (% temp0) (% temp0))
  (js @wasneg)
  (not (% imm0))
  @wasneg
  (bsrl (% imm0) (% imm0))
  (sete (% temp0.b))
  (xorl ($ 31) (% imm0))
  (addb (% temp0.b) (% imm0.b))
  (box-fixnum imm0 arg_z)
  (mark-as-node temp0)
  (single-value-return))

; if dest not nil store unboxed result in dest(0), else return a fixnum
(defx8632lapfunction fix-digit-logandc2 ((fix 4) #|(ra 0)|# (big arg_y) (dest arg_z)) ; index 0
  (mark-as-imm temp0)
  (movl (@ fix (% esp)) (% temp0))
  (unbox-fixnum temp0 temp0)
  (movl (@ x8632::misc-data-offset (% big)) (% imm0))
  (not (% imm0))
  (andl (% temp0) (% imm0))
  (mark-as-node temp0)
  (cmpl ($ x8632::nil-value) (% dest))
  (jne @store)
  (box-fixnum imm0 arg_z)
  (single-value-return 3)
  @store
  (movl (% imm0) (@ x8632::misc-data-offset (% dest)))
  (single-value-return 3))

(defx8632lapfunction digit-lognot-move ((index 4) #|(ra 0)|# (source arg_y) (dest arg_z))
  (movl (@ index (% esp)) (% temp0))
  (movl (@ x8632::misc-data-offset (% source) (% temp0)) (% imm0))
  (not (% imm0))
  (movl (% imm0) (@ x8632::misc-data-offset (% dest) (% temp0)))
  (single-value-return 3))

;; Add b to a starting at a[i]
;; might want not to use SSE2 for this.  use lea to update loop counter
;; variables so that the flags don't get set.
(defx8632lapfunction bignum-add-loop-+ ((i 8) (a 4) #|(ra 0)|# (b arg_y) (blen arg_z))
  (let ((aa mm2)
        (bb mm3)
        (cc mm4))
    (movl (@ a (% esp)) (% temp0))
    (movl (@ i (% esp)) (% temp1))
    (xorl (% imm0) (% imm0))
    (pxor (% cc) (% cc))
    @loop
    (movd (@ x8632::misc-data-offset (% temp0) (% temp1)) (% aa))
    (movd (@ x8632::misc-data-offset (% b) (% imm0)) (% bb))
    (paddq (% bb) (% aa))
    (paddq (% cc) (% aa))
    (movd (% aa) (@ x8632::misc-data-offset (% temp0) (% temp1)))
    (psrlq ($ 32) (% aa))
    (movq (% aa) (% cc))
    (addl ($ '1) (% temp1))
    (addl ($ '1) (% imm0))
    (subl ($ '1) (% blen))
    (jg @loop)
    ;; add in final carry
    (movd (% cc) (% imm0))
    (addl (% imm0) (@ x8632::misc-data-offset (% temp0) (% temp1)))
    (single-value-return 4)))

(defx8632lapfunction bignum-logtest-loop ((count 4) #|(ra 0)|# (s1 arg_y) (s2 arg_z))
  (let ((i temp1)
        (c temp0))
    (movl (@ count (% esp)) (% c))
    (xorl (% i) (% i))
    @loop
    (movl (@ x8632::misc-data-offset (% s1) (% i)) (% imm0))
    (test (@ x8632::misc-data-offset (% s2) (% i)) (% imm0))
    (jnz @true)
    (addl ($ '1) (% i))
    (cmpl (% i) (% c))
    (jg @loop)
    (movl ($ x8632::nil-value) (% arg_z))
    (single-value-return 3)
    @true
    (movl ($ x8632::t-value) (% arg_z))
    (single-value-return 3)))

;;; called from bignum-ashift-left-unaligned
(defx8632lapfunction bignum-shift-left-loop ((nbits 12) (result 8) (bignum 4) #|(ra 0)|# (res-len-1 arg_y) (j arg_z))
  (let ((r temp0)
        (b temp1)
        (bb mm2)
        (bits mm3)
        (tt mm4)
        (remaining-bits mm5))
    (movl (% j) (% imm0))
    (subl ($ '1) (% imm0))
    (pushl (% imm0))                    ;digits
    (movl (@ nbits (% esp)) (% imm0))
    (unbox-fixnum imm0 imm0)
    (movd (% imm0) (% bits))            ;shift count
    (negl (% imm0))
    (addl ($ 32) (% imm0))
    (movd (% imm0) (% remaining-bits))
    (movl (@ result (% esp)) (% r))
    (movl (@ bignum (% esp)) (% b))
    (xorl (% imm0) (% imm0))
    @loop
    ;; bignum[i]
    (movd (@ x8632::misc-data-offset (% b) (% imm0)) (% bb))
    (psrlq (% remaining-bits) (% bb))
    ;; bignum[i+1]
    (movd (@ (+ x8632::misc-data-offset 4) (% b) (% imm0)) (% tt))
    (psllq (% bits) (% tt))
    (por (% tt) (% bb))
    (movd (% bb) (@ x8632::misc-data-offset (% r) (% j)))
    (addl ($ '1) (% imm0))
    (addl ($ '1) (% j))
    (cmpl (% j) (% res-len-1))
    (jne @loop)
    (movd (@ x8632::misc-data-offset (% bignum)) (% bb)) ;bignum[0]
    (psllq (% bits) (% bb))
    (popl (% arg_y))                    ;digits
    (movd (% bb) (@ x8632::misc-data-offset (% r) (% arg_y)))
    (movd (@ x8632::misc-data-offset (% bignum) (% imm0)) (% bb)) ;bignum[i]
    (psrlq (% remaining-bits) (% bb))
    (movd (% bb) (@ x8632::misc-data-offset (% r) (% j)))
    (single-value-return 5)))

(defx8632lapfunction %logcount-complement ((bignum arg_y) (i arg_z))
  (mark-as-imm temp0)
  (let ((rshift imm0)
        (temp temp0))
    (movl (@ x8632::misc-data-offset (% bignum) (% i)) (% rshift))
    (notl (% rshift))
    (xorl (% arg_z) (% arg_z))
    (testl (% rshift) (% rshift))
    (jmp @test)
    @next
    (lea (@ -1 (% rshift)) (% temp))
    (and (% temp) (% rshift))           ;sets flags
    (lea (@ '1 (% arg_z)) (% arg_z))    ;doesn't set flags
    @test
    (jne @next)
    (mark-as-node temp0)
    (single-value-return)))

(defx8632lapfunction %logcount ((bignum arg_y) (i arg_z))
  (mark-as-imm temp0)
  (let ((rshift imm0)
        (temp temp0))
    (movl (@ x8632::misc-data-offset (% bignum) (% i)) (% rshift))
    (xorl (% arg_z) (% arg_z))
    (testl (% rshift) (% rshift))
    (jmp @test)
    @next
    (lea (@ -1 (% rshift)) (% temp))
    (and (% temp) (% rshift))           ;sets flags
    (lea (@ '1 (% arg_z)) (% arg_z))    ;doesn't set flags
    @test
    (jne @next)
    (mark-as-node temp0)
    (single-value-return)))


;;; floor: given x and y, producing q and r, q * y + r = x.

;;; Divide bignum x by single digit y (passed as two halves).
;;; The quotient in stored in q, and the remainder is returned
;;; in two halves.
(defx8632lapfunction %floor-loop-quo ((x 8) (q 4) #|(ra 0)|# (yhi arg_y) (ylo arg_z))
  )