--- /dev/null
+ # Alpha __mpn_add_n -- Add two limb vectors of the same length > 0 and
+ # store sum in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $16
+ # s1_ptr $17
+ # s2_ptr $18
+ # size $19
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_add_n
+ .ent __mpn_add_n
+__mpn_add_n:
+ .frame $30,0,$26,0
+
+ ldq $3,0($17)
+ ldq $4,0($18)
+
+ subq $19,1,$19
+ and $19,4-1,$2 # number of limbs in first loop
+ bis $31,$31,$0
+ beq $2,.L0 # if multiple of 4 limbs, skip first loop
+
+ subq $19,$2,$19
+
+.Loop0: subq $2,1,$2
+ ldq $5,8($17)
+ addq $4,$0,$4
+ ldq $6,8($18)
+ cmpult $4,$0,$1
+ addq $3,$4,$4
+ cmpult $4,$3,$0
+ stq $4,0($16)
+ or $0,$1,$0
+
+ addq $17,8,$17
+ addq $18,8,$18
+ bis $5,$5,$3
+ bis $6,$6,$4
+ addq $16,8,$16
+ bne $2,.Loop0
+
+.L0: beq $19,.Lend
+
+ .align 3
+.Loop: subq $19,4,$19
+
+ ldq $5,8($17)
+ addq $4,$0,$4
+ ldq $6,8($18)
+ cmpult $4,$0,$1
+ addq $3,$4,$4
+ cmpult $4,$3,$0
+ stq $4,0($16)
+ or $0,$1,$0
+
+ ldq $3,16($17)
+ addq $6,$0,$6
+ ldq $4,16($18)
+ cmpult $6,$0,$1
+ addq $5,$6,$6
+ cmpult $6,$5,$0
+ stq $6,8($16)
+ or $0,$1,$0
+
+ ldq $5,24($17)
+ addq $4,$0,$4
+ ldq $6,24($18)
+ cmpult $4,$0,$1
+ addq $3,$4,$4
+ cmpult $4,$3,$0
+ stq $4,16($16)
+ or $0,$1,$0
+
+ ldq $3,32($17)
+ addq $6,$0,$6
+ ldq $4,32($18)
+ cmpult $6,$0,$1
+ addq $5,$6,$6
+ cmpult $6,$5,$0
+ stq $6,24($16)
+ or $0,$1,$0
+
+ addq $17,32,$17
+ addq $18,32,$18
+ addq $16,32,$16
+ bne $19,.Loop
+
+.Lend: addq $4,$0,$4
+ cmpult $4,$0,$1
+ addq $3,$4,$4
+ cmpult $4,$3,$0
+ stq $4,0($16)
+ or $0,$1,$0
+ ret $31,($26),1
+
+ .end __mpn_add_n
--- /dev/null
+ # Alpha 21064 __mpn_addmul_1 -- Multiply a limb vector with a limb and add
+ # the result to a second limb vector.
+
+ # Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # s2_limb r19
+
+ # This code runs at 42 cycles/limb on the 21064.
+
+ # To improve performance for long multiplications, we would use
+ # 'fetch' for S1 and 'fetch_m' for RES. It's not obvious how to use
+ # these instructions without slowing down the general code: 1. We can
+ # only have two prefetches in operation at any time in the Alpha
+ # architecture. 2. There will seldom be any special alignment
+ # between RES_PTR and S1_PTR. Maybe we can simply divide the current
+ # loop into an inner and outer loop, having the inner loop handle
+ # exactly one prefetch block?
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_addmul_1
+ .ent __mpn_addmul_1 2
+__mpn_addmul_1:
+ .frame $30,0,$26
+
+ ldq $2,0($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ subq $18,1,$18 # size--
+ mulq $2,$19,$3 # $3 = prod_low
+ ldq $5,0($16) # $5 = *res_ptr
+ umulh $2,$19,$0 # $0 = prod_high
+ beq $18,Lend1 # jump if size was == 1
+ ldq $2,0($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ subq $18,1,$18 # size--
+ addq $5,$3,$3
+ cmpult $3,$5,$4
+ stq $3,0($16)
+ addq $16,8,$16 # res_ptr++
+ beq $18,Lend2 # jump if size was == 2
+
+ .align 3
+Loop: mulq $2,$19,$3 # $3 = prod_low
+ ldq $5,0($16) # $5 = *res_ptr
+ addq $4,$0,$0 # cy_limb = cy_limb + 'cy'
+ subq $18,1,$18 # size--
+ umulh $2,$19,$4 # $4 = cy_limb
+ ldq $2,0($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ addq $3,$0,$3 # $3 = cy_limb + prod_low
+ cmpult $3,$0,$0 # $0 = carry from (cy_limb + prod_low)
+ addq $5,$3,$3
+ cmpult $3,$5,$5
+ stq $3,0($16)
+ addq $16,8,$16 # res_ptr++
+ addq $5,$0,$0 # combine carries
+ bne $18,Loop
+
+Lend2: mulq $2,$19,$3 # $3 = prod_low
+ ldq $5,0($16) # $5 = *res_ptr
+ addq $4,$0,$0 # cy_limb = cy_limb + 'cy'
+ umulh $2,$19,$4 # $4 = cy_limb
+ addq $3,$0,$3 # $3 = cy_limb + prod_low
+ cmpult $3,$0,$0 # $0 = carry from (cy_limb + prod_low)
+ addq $5,$3,$3
+ cmpult $3,$5,$5
+ stq $3,0($16)
+ addq $5,$0,$0 # combine carries
+ addq $4,$0,$0 # cy_limb = prod_high + cy
+ ret $31,($26),1
+Lend1: addq $5,$3,$3
+ cmpult $3,$5,$5
+ stq $3,0($16)
+ addq $0,$5,$0
+ ret $31,($26),1
+
+ .end __mpn_addmul_1
--- /dev/null
+ # Alpha __mpn_add_n -- Add two limb vectors of the same length > 0 and
+ # store sum in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $16
+ # s1_ptr $17
+ # s2_ptr $18
+ # size $19
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_add_n
+ .ent __mpn_add_n
+__mpn_add_n:
+ .frame $30,0,$26,0
+
+ ldq $3,0($17)
+ ldq $4,0($18)
+
+ subq $19,1,$19
+ and $19,4-1,$2 # number of limbs in first loop
+ bis $31,$31,$0
+ beq $2,.L0 # if multiple of 4 limbs, skip first loop
+
+ subq $19,$2,$19
+
+.Loop0: subq $2,1,$2
+ ldq $5,8($17)
+ addq $4,$0,$4
+ ldq $6,8($18)
+ cmpult $4,$0,$1
+ addq $3,$4,$4
+ cmpult $4,$3,$0
+ stq $4,0($16)
+ or $0,$1,$0
+
+ addq $17,8,$17
+ addq $18,8,$18
+ bis $5,$5,$3
+ bis $6,$6,$4
+ addq $16,8,$16
+ bne $2,.Loop0
+
+.L0: beq $19,.Lend
+
+ .align 4
+.Loop: subq $19,4,$19
+ unop
+
+ ldq $6,8($18)
+ addq $4,$0,$0
+ ldq $5,8($17)
+ cmpult $0,$4,$1
+ ldq $4,16($18)
+ addq $3,$0,$20
+ cmpult $20,$3,$0
+ ldq $3,16($17)
+ or $0,$1,$0
+ addq $6,$0,$0
+ cmpult $0,$6,$1
+ ldq $6,24($18)
+ addq $5,$0,$21
+ cmpult $21,$5,$0
+ ldq $5,24($17)
+ or $0,$1,$0
+ addq $4,$0,$0
+ cmpult $0,$4,$1
+ ldq $4,32($18)
+ addq $3,$0,$22
+ cmpult $22,$3,$0
+ ldq $3,32($17)
+ or $0,$1,$0
+ addq $6,$0,$0
+ cmpult $0,$6,$1
+ addq $5,$0,$23
+ cmpult $23,$5,$0
+ or $0,$1,$0
+
+ stq $20,0($16)
+ stq $21,8($16)
+ stq $22,16($16)
+ stq $23,24($16)
+
+ addq $17,32,$17
+ addq $18,32,$18
+ addq $16,32,$16
+ bne $19,.Loop
+
+.Lend: addq $4,$0,$4
+ cmpult $4,$0,$1
+ addq $3,$4,$4
+ cmpult $4,$3,$0
+ stq $4,0($16)
+ or $0,$1,$0
+ ret $31,($26),1
+
+ .end __mpn_add_n
--- /dev/null
+ # Alpha EV5 __mpn_lshift --
+
+ # Copyright (C) 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # cnt r19
+
+ # This code runs at 4.25 cycles/limb on the EV5.
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_lshift
+ .ent __mpn_lshift
+__mpn_lshift:
+ .frame $30,0,$26,0
+
+ s8addq $18,$17,$17 # make r17 point at end of s1
+ ldq $4,-8($17) # load first limb
+ subq $31,$19,$20
+ s8addq $18,$16,$16 # make r16 point at end of RES
+ subq $18,1,$18
+ and $18,4-1,$28 # number of limbs in first loop
+ srl $4,$20,$0 # compute function result
+
+ beq $28,L0
+ subq $18,$28,$18
+
+ .align 3
+Loop0: ldq $3,-16($17)
+ subq $16,8,$16
+ sll $4,$19,$5
+ subq $17,8,$17
+ subq $28,1,$28
+ srl $3,$20,$6
+ or $3,$3,$4
+ or $5,$6,$8
+ stq $8,0($16)
+ bne $28,Loop0
+
+L0: sll $4,$19,$24
+ beq $18,Lend
+ # warm up phase 1
+ ldq $1,-16($17)
+ subq $18,4,$18
+ ldq $2,-24($17)
+ ldq $3,-32($17)
+ ldq $4,-40($17)
+ beq $18,Lcool1
+ # warm up phase 2
+ srl $1,$20,$7
+ sll $1,$19,$21
+ srl $2,$20,$8
+ ldq $1,-48($17)
+ sll $2,$19,$22
+ ldq $2,-56($17)
+ srl $3,$20,$5
+ or $7,$24,$7
+ sll $3,$19,$23
+ or $8,$21,$8
+ srl $4,$20,$6
+ ldq $3,-64($17)
+ sll $4,$19,$24
+ ldq $4,-72($17)
+ subq $18,4,$18
+ beq $18,Lcool1
+ .align 4
+ # main loop
+Loop: stq $7,-8($16)
+ or $5,$22,$5
+ stq $8,-16($16)
+ or $6,$23,$6
+
+ srl $1,$20,$7
+ subq $18,4,$18
+ sll $1,$19,$21
+ unop # ldq $31,-96($17)
+
+ srl $2,$20,$8
+ ldq $1,-80($17)
+ sll $2,$19,$22
+ ldq $2,-88($17)
+
+ stq $5,-24($16)
+ or $7,$24,$7
+ stq $6,-32($16)
+ or $8,$21,$8
+
+ srl $3,$20,$5
+ unop # ldq $31,-96($17)
+ sll $3,$19,$23
+ subq $16,32,$16
+
+ srl $4,$20,$6
+ ldq $3,-96($17
+ sll $4,$19,$24
+ ldq $4,-104($17)
+
+ subq $17,32,$17
+ bne $18,Loop
+ unop
+ unop
+ # cool down phase 2/1
+Lcool1: stq $7,-8($16)
+ or $5,$22,$5
+ stq $8,-16($16)
+ or $6,$23,$6
+ srl $1,$20,$7
+ sll $1,$19,$21
+ srl $2,$20,$8
+ sll $2,$19,$22
+ stq $5,-24($16)
+ or $7,$24,$7
+ stq $6,-32($16)
+ or $8,$21,$8
+ srl $3,$20,$5
+ sll $3,$19,$23
+ srl $4,$20,$6
+ sll $4,$19,$24
+ # cool down phase 2/2
+ stq $7,-40($16)
+ or $5,$22,$5
+ stq $8,-48($16)
+ or $6,$23,$6
+ stq $5,-56($16)
+ stq $6,-64($16)
+ # cool down phase 2/3
+ stq $24,-72($16)
+ ret $31,($26),1
+
+ # cool down phase 1/1
+Lcool1: srl $1,$20,$7
+ sll $1,$19,$21
+ srl $2,$20,$8
+ sll $2,$19,$22
+ srl $3,$20,$5
+ or $7,$24,$7
+ sll $3,$19,$23
+ or $8,$21,$8
+ srl $4,$20,$6
+ sll $4,$19,$24
+ # cool down phase 1/2
+ stq $7,-8($16)
+ or $5,$22,$5
+ stq $8,-16($16)
+ or $6,$23,$6
+ stq $5,-24($16)
+ stq $6,-32($16)
+ stq $24,-40($16)
+ ret $31,($26),1
+
+Lend stq $24,-8($16)
+ ret $31,($26),1
+ .end __mpn_lshift
--- /dev/null
+ # Alpha EV5 __mpn_rshift --
+
+ # Copyright (C) 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # cnt r19
+
+ # This code runs at 4.25 cycles/limb on the EV5.
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_rshift
+ .ent __mpn_rshift
+__mpn_rshift:
+ .frame $30,0,$26,0
+
+ ldq $4,0($17) # load first limb
+ subq $31,$19,$20
+ subq $18,1,$18
+ and $18,4-1,$28 # number of limbs in first loop
+ sll $4,$20,$0 # compute function result
+
+ beq $28,L0
+ subq $18,$28,$18
+
+ .align 3
+Loop0: ldq $3,8($17)
+ addq $16,8,$16
+ srl $4,$19,$5
+ addq $17,8,$17
+ subq $28,1,$28
+ sll $3,$20,$6
+ or $3,$3,$4
+ or $5,$6,$8
+ stq $8,-8($16)
+ bne $28,Loop0
+
+L0: srl $4,$19,$24
+ beq $18,Lend
+ # warm up phase 1
+ ldq $1,8($17)
+ subq $18,4,$18
+ ldq $2,16($17)
+ ldq $3,24($17)
+ ldq $4,32($17)
+ beq $18,Lcool1
+ # warm up phase 2
+ sll $1,$20,$7
+ srl $1,$19,$21
+ sll $2,$20,$8
+ ldq $1,40($17)
+ srl $2,$19,$22
+ ldq $2,48($17)
+ sll $3,$20,$5
+ or $7,$24,$7
+ srl $3,$19,$23
+ or $8,$21,$8
+ sll $4,$20,$6
+ ldq $3,56($17)
+ srl $4,$19,$24
+ ldq $4,64($17)
+ subq $18,4,$18
+ beq $18,Lcool2
+ .align 4
+ # main loop
+Loop: stq $7,0($16)
+ or $5,$22,$5
+ stq $8,8($16)
+ or $6,$23,$6
+
+ sll $1,$20,$7
+ subq $18,4,$18
+ srl $1,$19,$21
+ unop # ldq $31,-96($17)
+
+ sll $2,$20,$8
+ ldq $1,72($17)
+ srl $2,$19,$22
+ ldq $2,80($17)
+
+ stq $5,16($16)
+ or $7,$24,$7
+ stq $6,24($16)
+ or $8,$21,$8
+
+ sll $3,$20,$5
+ unop # ldq $31,-96($17)
+ srl $3,$19,$23
+ addq $16,32,$16
+
+ sll $4,$20,$6
+ ldq $3,88($17)
+ srl $4,$19,$24
+ ldq $4,96($17)
+
+ addq $17,32,$17
+ bne $18,Loop
+ unop
+ unop
+ # cool down phase 2/1
+Lcool2: stq $7,0($16)
+ or $5,$22,$5
+ stq $8,8($16)
+ or $6,$23,$6
+ sll $1,$20,$7
+ srl $1,$19,$21
+ sll $2,$20,$8
+ srl $2,$19,$22
+ stq $5,16($16)
+ or $7,$24,$7
+ stq $6,24($16)
+ or $8,$21,$8
+ sll $3,$20,$5
+ srl $3,$19,$23
+ sll $4,$20,$6
+ srl $4,$19,$24
+ # cool down phase 2/2
+ stq $7,32($16)
+ or $5,$22,$5
+ stq $8,40($16)
+ or $6,$23,$6
+ stq $5,48($16)
+ stq $6,56($16)
+ # cool down phase 2/3
+ stq $24,64($16)
+ ret $31,($26),1
+
+ # cool down phase 1/1
+Lcool1: sll $1,$20,$7
+ srl $1,$19,$21
+ sll $2,$20,$8
+ srl $2,$19,$22
+ sll $3,$20,$5
+ or $7,$24,$7
+ srl $3,$19,$23
+ or $8,$21,$8
+ sll $4,$20,$6
+ srl $4,$19,$24
+ # cool down phase 1/2
+ stq $7,0($16)
+ or $5,$22,$5
+ stq $8,8($16)
+ or $6,$23,$6
+ stq $5,16($16)
+ stq $6,24($16)
+ stq $24,32($16)
+ ret $31,($26),1
+
+Lend: stq $24,0($16)
+ ret $31,($26),1
+ .end __mpn_rshift
--- /dev/null
+ # Alpha 21064 __mpn_lshift --
+
+ # Copyright (C) 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # cnt r19
+
+ # This code runs at 4.8 cycles/limb on the 21064. With infinite unrolling,
+ # it would take 4 cycles/limb. It should be possible to get down to 3
+ # cycles/limb since both ldq and stq can be paired with the other used
+ # instructions. But there are many restrictions in the 21064 pipeline that
+ # makes it hard, if not impossible, to get down to 3 cycles/limb:
+
+ # 1. ldq has a 3 cycle delay, srl and sll have a 2 cycle delay.
+ # 2. Only aligned instruction pairs can be paired.
+ # 3. The store buffer or silo might not be able to deal with the bandwidth.
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_lshift
+ .ent __mpn_lshift
+__mpn_lshift:
+ .frame $30,0,$26,0
+
+ s8addq $18,$17,$17 # make r17 point at end of s1
+ ldq $4,-8($17) # load first limb
+ subq $17,8,$17
+ subq $31,$19,$7
+ s8addq $18,$16,$16 # make r16 point at end of RES
+ subq $18,1,$18
+ and $18,4-1,$20 # number of limbs in first loop
+ srl $4,$7,$0 # compute function result
+
+ beq $20,L0
+ subq $18,$20,$18
+
+ .align 3
+Loop0:
+ ldq $3,-8($17)
+ subq $16,8,$16
+ subq $17,8,$17
+ subq $20,1,$20
+ sll $4,$19,$5
+ srl $3,$7,$6
+ bis $3,$3,$4
+ bis $5,$6,$8
+ stq $8,0($16)
+ bne $20,Loop0
+
+L0: beq $18,Lend
+
+ .align 3
+Loop: ldq $3,-8($17)
+ subq $16,32,$16
+ subq $18,4,$18
+ sll $4,$19,$5
+ srl $3,$7,$6
+
+ ldq $4,-16($17)
+ sll $3,$19,$1
+ bis $5,$6,$8
+ stq $8,24($16)
+ srl $4,$7,$2
+
+ ldq $3,-24($17)
+ sll $4,$19,$5
+ bis $1,$2,$8
+ stq $8,16($16)
+ srl $3,$7,$6
+
+ ldq $4,-32($17)
+ sll $3,$19,$1
+ bis $5,$6,$8
+ stq $8,8($16)
+ srl $4,$7,$2
+
+ subq $17,32,$17
+ bis $1,$2,$8
+ stq $8,0($16)
+
+ bgt $18,Loop
+
+Lend: sll $4,$19,$8
+ stq $8,-8($16)
+ ret $31,($26),1
+ .end __mpn_lshift
--- /dev/null
+ # Alpha 21064 __mpn_mul_1 -- Multiply a limb vector with a limb and store
+ # the result in a second limb vector.
+
+ # Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # s2_limb r19
+
+ # This code runs at 42 cycles/limb on the EV4 and 18 cycles/limb on the EV5.
+
+ # To improve performance for long multiplications, we would use
+ # 'fetch' for S1 and 'fetch_m' for RES. It's not obvious how to use
+ # these instructions without slowing down the general code: 1. We can
+ # only have two prefetches in operation at any time in the Alpha
+ # architecture. 2. There will seldom be any special alignment
+ # between RES_PTR and S1_PTR. Maybe we can simply divide the current
+ # loop into an inner and outer loop, having the inner loop handle
+ # exactly one prefetch block?
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_mul_1
+ .ent __mpn_mul_1 2
+__mpn_mul_1:
+ .frame $30,0,$26
+
+ ldq $2,0($17) # $2 = s1_limb
+ subq $18,1,$18 # size--
+ mulq $2,$19,$3 # $3 = prod_low
+ bic $31,$31,$4 # clear cy_limb
+ umulh $2,$19,$0 # $0 = prod_high
+ beq $18,Lend1 # jump if size was == 1
+ ldq $2,8($17) # $2 = s1_limb
+ subq $18,1,$18 # size--
+ stq $3,0($16)
+ beq $18,Lend2 # jump if size was == 2
+
+ .align 3
+Loop: mulq $2,$19,$3 # $3 = prod_low
+ addq $4,$0,$0 # cy_limb = cy_limb + 'cy'
+ subq $18,1,$18 # size--
+ umulh $2,$19,$4 # $4 = cy_limb
+ ldq $2,16($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ addq $3,$0,$3 # $3 = cy_limb + prod_low
+ stq $3,8($16)
+ cmpult $3,$0,$0 # $0 = carry from (cy_limb + prod_low)
+ addq $16,8,$16 # res_ptr++
+ bne $18,Loop
+
+Lend2: mulq $2,$19,$3 # $3 = prod_low
+ addq $4,$0,$0 # cy_limb = cy_limb + 'cy'
+ umulh $2,$19,$4 # $4 = cy_limb
+ addq $3,$0,$3 # $3 = cy_limb + prod_low
+ cmpult $3,$0,$0 # $0 = carry from (cy_limb + prod_low)
+ stq $3,8($16)
+ addq $4,$0,$0 # cy_limb = prod_high + cy
+ ret $31,($26),1
+Lend1: stq $3,0($16)
+ ret $31,($26),1
+
+ .end __mpn_mul_1
--- /dev/null
+ # Alpha 21064 __mpn_rshift --
+
+ # Copyright (C) 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # cnt r19
+
+ # This code runs at 4.8 cycles/limb on the 21064. With infinite unrolling,
+ # it would take 4 cycles/limb. It should be possible to get down to 3
+ # cycles/limb since both ldq and stq can be paired with the other used
+ # instructions. But there are many restrictions in the 21064 pipeline that
+ # makes it hard, if not impossible, to get down to 3 cycles/limb:
+
+ # 1. ldq has a 3 cycle delay, srl and sll have a 2 cycle delay.
+ # 2. Only aligned instruction pairs can be paired.
+ # 3. The store buffer or silo might not be able to deal with the bandwidth.
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_rshift
+ .ent __mpn_rshift
+__mpn_rshift:
+ .frame $30,0,$26,0
+
+ ldq $4,0($17) # load first limb
+ addq $17,8,$17
+ subq $31,$19,$7
+ subq $18,1,$18
+ and $18,4-1,$20 # number of limbs in first loop
+ sll $4,$7,$0 # compute function result
+
+ beq $20,L0
+ subq $18,$20,$18
+
+ .align 3
+Loop0:
+ ldq $3,0($17)
+ addq $16,8,$16
+ addq $17,8,$17
+ subq $20,1,$20
+ srl $4,$19,$5
+ sll $3,$7,$6
+ bis $3,$3,$4
+ bis $5,$6,$8
+ stq $8,-8($16)
+ bne $20,Loop0
+
+L0: beq $18,Lend
+
+ .align 3
+Loop: ldq $3,0($17)
+ addq $16,32,$16
+ subq $18,4,$18
+ srl $4,$19,$5
+ sll $3,$7,$6
+
+ ldq $4,8($17)
+ srl $3,$19,$1
+ bis $5,$6,$8
+ stq $8,-32($16)
+ sll $4,$7,$2
+
+ ldq $3,16($17)
+ srl $4,$19,$5
+ bis $1,$2,$8
+ stq $8,-24($16)
+ sll $3,$7,$6
+
+ ldq $4,24($17)
+ srl $3,$19,$1
+ bis $5,$6,$8
+ stq $8,-16($16)
+ sll $4,$7,$2
+
+ addq $17,32,$17
+ bis $1,$2,$8
+ stq $8,-8($16)
+
+ bgt $18,Loop
+
+Lend: srl $4,$19,$8
+ stq $8,0($16)
+ ret $31,($26),1
+ .end __mpn_rshift
--- /dev/null
+ # Alpha __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+ # store difference in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $16
+ # s1_ptr $17
+ # s2_ptr $18
+ # size $19
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_sub_n
+ .ent __mpn_sub_n
+__mpn_sub_n:
+ .frame $30,0,$26,0
+
+ ldq $3,0($17)
+ ldq $4,0($18)
+
+ subq $19,1,$19
+ and $19,4-1,$2 # number of limbs in first loop
+ bis $31,$31,$0
+ beq $2,.L0 # if multiple of 4 limbs, skip first loop
+
+ subq $19,$2,$19
+
+.Loop0: subq $2,1,$2
+ ldq $5,8($17)
+ addq $4,$0,$4
+ ldq $6,8($18)
+ cmpult $4,$0,$1
+ subq $3,$4,$4
+ cmpult $3,$4,$0
+ stq $4,0($16)
+ or $0,$1,$0
+
+ addq $17,8,$17
+ addq $18,8,$18
+ bis $5,$5,$3
+ bis $6,$6,$4
+ addq $16,8,$16
+ bne $2,.Loop0
+
+.L0: beq $19,.Lend
+
+ .align 3
+.Loop: subq $19,4,$19
+
+ ldq $5,8($17)
+ addq $4,$0,$4
+ ldq $6,8($18)
+ cmpult $4,$0,$1
+ subq $3,$4,$4
+ cmpult $3,$4,$0
+ stq $4,0($16)
+ or $0,$1,$0
+
+ ldq $3,16($17)
+ addq $6,$0,$6
+ ldq $4,16($18)
+ cmpult $6,$0,$1
+ subq $5,$6,$6
+ cmpult $5,$6,$0
+ stq $6,8($16)
+ or $0,$1,$0
+
+ ldq $5,24($17)
+ addq $4,$0,$4
+ ldq $6,24($18)
+ cmpult $4,$0,$1
+ subq $3,$4,$4
+ cmpult $3,$4,$0
+ stq $4,16($16)
+ or $0,$1,$0
+
+ ldq $3,32($17)
+ addq $6,$0,$6
+ ldq $4,32($18)
+ cmpult $6,$0,$1
+ subq $5,$6,$6
+ cmpult $5,$6,$0
+ stq $6,24($16)
+ or $0,$1,$0
+
+ addq $17,32,$17
+ addq $18,32,$18
+ addq $16,32,$16
+ bne $19,.Loop
+
+.Lend: addq $4,$0,$4
+ cmpult $4,$0,$1
+ subq $3,$4,$4
+ cmpult $3,$4,$0
+ stq $4,0($16)
+ or $0,$1,$0
+ ret $31,($26),1
+
+ .end __mpn_sub_n
--- /dev/null
+ # Alpha 21064 __mpn_submul_1 -- Multiply a limb vector with a limb and
+ # subtract the result from a second limb vector.
+
+ # Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr r16
+ # s1_ptr r17
+ # size r18
+ # s2_limb r19
+
+ # This code runs at 42 cycles/limb on the 21064.
+
+ # To improve performance for long multiplications, we would use
+ # 'fetch' for S1 and 'fetch_m' for RES. It's not obvious how to use
+ # these instructions without slowing down the general code: 1. We can
+ # only have two prefetches in operation at any time in the Alpha
+ # architecture. 2. There will seldom be any special alignment
+ # between RES_PTR and S1_PTR. Maybe we can simply divide the current
+ # loop into an inner and outer loop, having the inner loop handle
+ # exactly one prefetch block?
+
+ .set noreorder
+ .set noat
+.text
+ .align 3
+ .globl __mpn_submul_1
+ .ent __mpn_submul_1 2
+__mpn_submul_1:
+ .frame $30,0,$26
+
+ ldq $2,0($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ subq $18,1,$18 # size--
+ mulq $2,$19,$3 # $3 = prod_low
+ ldq $5,0($16) # $5 = *res_ptr
+ umulh $2,$19,$0 # $0 = prod_high
+ beq $18,Lend1 # jump if size was == 1
+ ldq $2,0($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ subq $18,1,$18 # size--
+ subq $5,$3,$3
+ cmpult $5,$3,$4
+ stq $3,0($16)
+ addq $16,8,$16 # res_ptr++
+ beq $18,Lend2 # jump if size was == 2
+
+ .align 3
+Loop: mulq $2,$19,$3 # $3 = prod_low
+ ldq $5,0($16) # $5 = *res_ptr
+ addq $4,$0,$0 # cy_limb = cy_limb + 'cy'
+ subq $18,1,$18 # size--
+ umulh $2,$19,$4 # $4 = cy_limb
+ ldq $2,0($17) # $2 = s1_limb
+ addq $17,8,$17 # s1_ptr++
+ addq $3,$0,$3 # $3 = cy_limb + prod_low
+ cmpult $3,$0,$0 # $0 = carry from (cy_limb + prod_low)
+ subq $5,$3,$3
+ cmpult $5,$3,$5
+ stq $3,0($16)
+ addq $16,8,$16 # res_ptr++
+ addq $5,$0,$0 # combine carries
+ bne $18,Loop
+
+Lend2: mulq $2,$19,$3 # $3 = prod_low
+ ldq $5,0($16) # $5 = *res_ptr
+ addq $4,$0,$0 # cy_limb = cy_limb + 'cy'
+ umulh $2,$19,$4 # $4 = cy_limb
+ addq $3,$0,$3 # $3 = cy_limb + prod_low
+ cmpult $3,$0,$0 # $0 = carry from (cy_limb + prod_low)
+ subq $5,$3,$3
+ cmpult $5,$3,$5
+ stq $3,0($16)
+ addq $5,$0,$0 # combine carries
+ addq $4,$0,$0 # cy_limb = prod_high + cy
+ ret $31,($26),1
+Lend1: subq $5,$3,$3
+ cmpult $5,$3,$5
+ stq $3,0($16)
+ addq $0,$5,$0
+ ret $31,($26),1
+
+ .end __mpn_submul_1
--- /dev/null
+; HP-PA __mpn_add_n -- Add two limb vectors of the same length > 0 and store
+; sum in a third limb vector.
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr gr26
+; s1_ptr gr25
+; s2_ptr gr24
+; size gr23
+
+; One might want to unroll this as for other processors, but it turns
+; out that the data cache contention after a store makes such
+; unrolling useless. We can't come under 5 cycles/limb anyway.
+
+ .code
+ .export __mpn_add_n
+__mpn_add_n
+ .proc
+ .callinfo frame=0,no_calls
+ .entry
+
+ ldws,ma 4(0,%r25),%r20
+ ldws,ma 4(0,%r24),%r19
+
+ addib,= -1,%r23,L$end ; check for (SIZE == 1)
+ add %r20,%r19,%r28 ; add first limbs ignoring cy
+
+L$loop ldws,ma 4(0,%r25),%r20
+ ldws,ma 4(0,%r24),%r19
+ stws,ma %r28,4(0,%r26)
+ addib,<> -1,%r23,L$loop
+ addc %r20,%r19,%r28
+
+L$end stws %r28,0(0,%r26)
+ bv 0(%r2)
+ addc %r0,%r0,%r28
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA-1.1 __mpn_addmul_1 -- Multiply a limb vector with a limb and
+; add the result to a second limb vector.
+
+; Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r26
+; s1_ptr r25
+; size r24
+; s2_limb r23
+
+; This runs at 11 cycles/limb on a PA7000. With the used instructions, it
+; can not become faster due to data cache contention after a store. On the
+; PA7100 it runs at 10 cycles/limb, and that can not be improved either,
+; since only the xmpyu does not need the integer pipeline, so the only
+; dual-issue we will get are addc+xmpyu. Unrolling could gain a cycle/limb
+; on the PA7100.
+
+; There are some ideas described in mul_1.s that applies to this code too.
+
+ .code
+ .export __mpn_addmul_1
+__mpn_addmul_1
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ ldo 64(%r30),%r30
+ fldws,ma 4(%r25),%fr5
+ stw %r23,-16(%r30) ; move s2_limb ...
+ addib,= -1,%r24,L$just_one_limb
+ fldws -16(%r30),%fr4 ; ... into fr4
+ add %r0,%r0,%r0 ; clear carry
+ xmpyu %fr4,%fr5,%fr6
+ fldws,ma 4(%r25),%fr7
+ fstds %fr6,-16(%r30)
+ xmpyu %fr4,%fr7,%fr8
+ ldw -12(%r30),%r19 ; least significant limb in product
+ ldw -16(%r30),%r28
+
+ fstds %fr8,-16(%r30)
+ addib,= -1,%r24,L$end
+ ldw -12(%r30),%r1
+
+; Main loop
+L$loop ldws 0(%r26),%r29
+ fldws,ma 4(%r25),%fr5
+ add %r29,%r19,%r19
+ stws,ma %r19,4(%r26)
+ addc %r28,%r1,%r19
+ xmpyu %fr4,%fr5,%fr6
+ ldw -16(%r30),%r28
+ fstds %fr6,-16(%r30)
+ addc %r0,%r28,%r28
+ addib,<> -1,%r24,L$loop
+ ldw -12(%r30),%r1
+
+L$end ldw 0(%r26),%r29
+ add %r29,%r19,%r19
+ stws,ma %r19,4(%r26)
+ addc %r28,%r1,%r19
+ ldw -16(%r30),%r28
+ ldws 0(%r26),%r29
+ addc %r0,%r28,%r28
+ add %r29,%r19,%r19
+ stws,ma %r19,4(%r26)
+ addc %r0,%r28,%r28
+ bv 0(%r2)
+ ldo -64(%r30),%r30
+
+L$just_one_limb
+ xmpyu %fr4,%fr5,%fr6
+ ldw 0(%r26),%r29
+ fstds %fr6,-16(%r30)
+ ldw -12(%r30),%r1
+ ldw -16(%r30),%r28
+ add %r29,%r1,%r19
+ stw %r19,0(%r26)
+ addc %r0,%r28,%r28
+ bv 0(%r2)
+ ldo -64(%r30),%r30
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA-1.1 __mpn_mul_1 -- Multiply a limb vector with a limb and store
+; the result in a second limb vector.
+
+; Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r26
+; s1_ptr r25
+; size r24
+; s2_limb r23
+
+; This runs at 9 cycles/limb on a PA7000. With the used instructions, it can
+; not become faster due to data cache contention after a store. On the
+; PA7100 it runs at 7 cycles/limb, and that can not be improved either, since
+; only the xmpyu does not need the integer pipeline, so the only dual-issue
+; we will get are addc+xmpyu. Unrolling would not help either CPU.
+
+; We could use fldds to read two limbs at a time from the S1 array, and that
+; could bring down the times to 8.5 and 6.5 cycles/limb for the PA7000 and
+; PA7100, respectively. We don't do that since it does not seem worth the
+; (alignment) troubles...
+
+; At least the PA7100 is rumored to be able to deal with cache-misses
+; without stalling instruction issue. If this is true, and the cache is
+; actually also lockup-free, we should use a deeper software pipeline, and
+; load from S1 very early! (The loads and stores to -12(sp) will surely be
+; in the cache.)
+
+ .code
+ .export __mpn_mul_1
+__mpn_mul_1
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ ldo 64(%r30),%r30
+ fldws,ma 4(%r25),%fr5
+ stw %r23,-16(%r30) ; move s2_limb ...
+ addib,= -1,%r24,L$just_one_limb
+ fldws -16(%r30),%fr4 ; ... into fr4
+ add %r0,%r0,%r0 ; clear carry
+ xmpyu %fr4,%fr5,%fr6
+ fldws,ma 4(%r25),%fr7
+ fstds %fr6,-16(%r30)
+ xmpyu %fr4,%fr7,%fr8
+ ldw -12(%r30),%r19 ; least significant limb in product
+ ldw -16(%r30),%r28
+
+ fstds %fr8,-16(%r30)
+ addib,= -1,%r24,L$end
+ ldw -12(%r30),%r1
+
+; Main loop
+L$loop fldws,ma 4(%r25),%fr5
+ stws,ma %r19,4(%r26)
+ addc %r28,%r1,%r19
+ xmpyu %fr4,%fr5,%fr6
+ ldw -16(%r30),%r28
+ fstds %fr6,-16(%r30)
+ addib,<> -1,%r24,L$loop
+ ldw -12(%r30),%r1
+
+L$end stws,ma %r19,4(%r26)
+ addc %r28,%r1,%r19
+ ldw -16(%r30),%r28
+ stws,ma %r19,4(%r26)
+ addc %r0,%r28,%r28
+ bv 0(%r2)
+ ldo -64(%r30),%r30
+
+L$just_one_limb
+ xmpyu %fr4,%fr5,%fr6
+ fstds %fr6,-16(%r30)
+ ldw -16(%r30),%r28
+ ldo -64(%r30),%r30
+ bv 0(%r2)
+ fstws %fr6R,0(%r26)
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA-1.1 __mpn_submul_1 -- Multiply a limb vector with a limb and
+; subtract the result from a second limb vector.
+
+; Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r26
+; s1_ptr r25
+; size r24
+; s2_limb r23
+
+; This runs at 12 cycles/limb on a PA7000. With the used instructions, it
+; can not become faster due to data cache contention after a store. On the
+; PA7100 it runs at 11 cycles/limb, and that can not be improved either,
+; since only the xmpyu does not need the integer pipeline, so the only
+; dual-issue we will get are addc+xmpyu. Unrolling could gain a cycle/limb
+; on the PA7100.
+
+; There are some ideas described in mul_1.s that applies to this code too.
+
+; It seems possible to make this run as fast as __mpn_addmul_1, if we use
+; sub,>>= %r29,%r19,%r22
+; addi 1,%r28,%r28
+; but that requires reworking the hairy software pipeline...
+
+ .code
+ .export __mpn_submul_1
+__mpn_submul_1
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ ldo 64(%r30),%r30
+ fldws,ma 4(%r25),%fr5
+ stw %r23,-16(%r30) ; move s2_limb ...
+ addib,= -1,%r24,L$just_one_limb
+ fldws -16(%r30),%fr4 ; ... into fr4
+ add %r0,%r0,%r0 ; clear carry
+ xmpyu %fr4,%fr5,%fr6
+ fldws,ma 4(%r25),%fr7
+ fstds %fr6,-16(%r30)
+ xmpyu %fr4,%fr7,%fr8
+ ldw -12(%r30),%r19 ; least significant limb in product
+ ldw -16(%r30),%r28
+
+ fstds %fr8,-16(%r30)
+ addib,= -1,%r24,L$end
+ ldw -12(%r30),%r1
+
+; Main loop
+L$loop ldws 0(%r26),%r29
+ fldws,ma 4(%r25),%fr5
+ sub %r29,%r19,%r22
+ add %r22,%r19,%r0
+ stws,ma %r22,4(%r26)
+ addc %r28,%r1,%r19
+ xmpyu %fr4,%fr5,%fr6
+ ldw -16(%r30),%r28
+ fstds %fr6,-16(%r30)
+ addc %r0,%r28,%r28
+ addib,<> -1,%r24,L$loop
+ ldw -12(%r30),%r1
+
+L$end ldw 0(%r26),%r29
+ sub %r29,%r19,%r22
+ add %r22,%r19,%r0
+ stws,ma %r22,4(%r26)
+ addc %r28,%r1,%r19
+ ldw -16(%r30),%r28
+ ldws 0(%r26),%r29
+ addc %r0,%r28,%r28
+ sub %r29,%r19,%r22
+ add %r22,%r19,%r0
+ stws,ma %r22,4(%r26)
+ addc %r0,%r28,%r28
+ bv 0(%r2)
+ ldo -64(%r30),%r30
+
+L$just_one_limb
+ xmpyu %fr4,%fr5,%fr6
+ ldw 0(%r26),%r29
+ fstds %fr6,-16(%r30)
+ ldw -12(%r30),%r1
+ ldw -16(%r30),%r28
+ sub %r29,%r1,%r22
+ add %r22,%r1,%r0
+ stw %r22,0(%r26)
+ addc %r0,%r28,%r28
+ bv 0(%r2)
+ ldo -64(%r30),%r30
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA __udiv_qrnnd division support, used from longlong.h.
+; This version runs fast on PA 7000 and later.
+
+; Copyright (C) 1993, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; rem_ptr gr26
+; n1 gr25
+; n0 gr24
+; d gr23
+
+ .code
+L$0000 .word 0x43f00000
+ .word 0x0
+ .export __udiv_qrnnd
+__udiv_qrnnd
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+ ldo 64(%r30),%r30
+
+ stws %r25,-16(0,%r30) ; n_hi
+ stws %r24,-12(0,%r30) ; n_lo
+ ldil L'L$0000,%r19
+ ldo R'L$0000(%r19),%r19
+ fldds -16(0,%r30),%fr5
+ stws %r23,-12(0,%r30)
+ comib,<= 0,%r25,L$1
+ fcnvxf,dbl,dbl %fr5,%fr5
+ fldds 0(0,%r19),%fr4
+ fadd,dbl %fr4,%fr5,%fr5
+L$1
+ fcpy,sgl %fr0,%fr6L
+ fldws -12(0,%r30),%fr6R
+ fcnvxf,dbl,dbl %fr6,%fr4
+
+ fdiv,dbl %fr5,%fr4,%fr5
+
+ fcnvfx,dbl,dbl %fr5,%fr4
+ fstws %fr4R,-16(%r30)
+ xmpyu %fr4R,%fr6R,%fr6
+ ldws -16(%r30),%r28
+ fstds %fr6,-16(0,%r30)
+ ldws -12(0,%r30),%r21
+ ldws -16(0,%r30),%r20
+ sub %r24,%r21,%r22
+ subb %r25,%r20,%r19
+ comib,= 0,%r19,L$2
+ ldo -64(%r30),%r30
+
+ add %r22,%r23,%r22
+ ldo -1(%r28),%r28
+L$2 bv 0(%r2)
+ stws %r22,0(0,%r26)
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA __mpn_lshift --
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr gr26
+; s_ptr gr25
+; size gr24
+; cnt gr23
+
+ .code
+ .export __mpn_lshift
+__mpn_lshift
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ sh2add %r24,%r25,%r25
+ sh2add %r24,%r26,%r26
+ ldws,mb -4(0,%r25),%r22
+ subi 32,%r23,%r1
+ mtsar %r1
+ addib,= -1,%r24,L$0004
+ vshd %r0,%r22,%r28 ; compute carry out limb
+ ldws,mb -4(0,%r25),%r29
+ addib,= -1,%r24,L$0002
+ vshd %r22,%r29,%r20
+
+L$loop ldws,mb -4(0,%r25),%r22
+ stws,mb %r20,-4(0,%r26)
+ addib,= -1,%r24,L$0003
+ vshd %r29,%r22,%r20
+ ldws,mb -4(0,%r25),%r29
+ stws,mb %r20,-4(0,%r26)
+ addib,<> -1,%r24,L$loop
+ vshd %r22,%r29,%r20
+
+L$0002 stws,mb %r20,-4(0,%r26)
+ vshd %r29,%r0,%r20
+ bv 0(%r2)
+ stw %r20,-4(0,%r26)
+L$0003 stws,mb %r20,-4(0,%r26)
+L$0004 vshd %r22,%r0,%r20
+ bv 0(%r2)
+ stw %r20,-4(0,%r26)
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA __mpn_rshift --
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr gr26
+; s_ptr gr25
+; size gr24
+; cnt gr23
+
+ .code
+ .export __mpn_rshift
+__mpn_rshift
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ ldws,ma 4(0,%r25),%r22
+ mtsar %r23
+ addib,= -1,%r24,L$0004
+ vshd %r22,%r0,%r28 ; compute carry out limb
+ ldws,ma 4(0,%r25),%r29
+ addib,= -1,%r24,L$0002
+ vshd %r29,%r22,%r20
+
+L$loop ldws,ma 4(0,%r25),%r22
+ stws,ma %r20,4(0,%r26)
+ addib,= -1,%r24,L$0003
+ vshd %r22,%r29,%r20
+ ldws,ma 4(0,%r25),%r29
+ stws,ma %r20,4(0,%r26)
+ addib,<> -1,%r24,L$loop
+ vshd %r29,%r22,%r20
+
+L$0002 stws,ma %r20,4(0,%r26)
+ vshd %r0,%r29,%r20
+ bv 0(%r2)
+ stw %r20,0(0,%r26)
+L$0003 stws,ma %r20,4(0,%r26)
+L$0004 vshd %r0,%r22,%r20
+ bv 0(%r2)
+ stw %r20,0(0,%r26)
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+; store difference in a third limb vector.
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr gr26
+; s1_ptr gr25
+; s2_ptr gr24
+; size gr23
+
+; One might want to unroll this as for other processors, but it turns
+; out that the data cache contention after a store makes such
+; unrolling useless. We can't come under 5 cycles/limb anyway.
+
+ .code
+ .export __mpn_sub_n
+__mpn_sub_n
+ .proc
+ .callinfo frame=0,no_calls
+ .entry
+
+ ldws,ma 4(0,%r25),%r20
+ ldws,ma 4(0,%r24),%r19
+
+ addib,= -1,%r23,L$end ; check for (SIZE == 1)
+ sub %r20,%r19,%r28 ; subtract first limbs ignoring cy
+
+L$loop ldws,ma 4(0,%r25),%r20
+ ldws,ma 4(0,%r24),%r19
+ stws,ma %r28,4(0,%r26)
+ addib,<> -1,%r23,L$loop
+ subb %r20,%r19,%r28
+
+L$end stws %r28,0(0,%r26)
+ addc %r0,%r0,%r28
+ bv 0(%r2)
+ subi 1,%r28,%r28
+
+ .exit
+ .procend
--- /dev/null
+; HP-PA __udiv_qrnnd division support, used from longlong.h.
+; This version runs fast on pre-PA7000 CPUs.
+
+; Copyright (C) 1993, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; rem_ptr gr26
+; n1 gr25
+; n0 gr24
+; d gr23
+
+; The code size is a bit excessive. We could merge the last two ds;addc
+; sequences by simply moving the "bb,< Odd" instruction down. The only
+; trouble is the FFFFFFFF code that would need some hacking.
+
+ .code
+ .export __udiv_qrnnd
+__udiv_qrnnd
+ .proc
+ .callinfo frame=0,no_calls
+ .entry
+
+ comb,< %r23,0,L$largedivisor
+ sub %r0,%r23,%r1 ; clear cy as side-effect
+ ds %r0,%r1,%r0
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r23,%r25
+ addc %r24,%r24,%r28
+ ds %r25,%r23,%r25
+ comclr,>= %r25,%r0,%r0
+ addl %r25,%r23,%r25
+ stws %r25,0(0,%r26)
+ bv 0(%r2)
+ addc %r28,%r28,%r28
+
+L$largedivisor
+ extru %r24,31,1,%r19 ; r19 = n0 & 1
+ bb,< %r23,31,L$odd
+ extru %r23,30,31,%r22 ; r22 = d >> 1
+ shd %r25,%r24,1,%r24 ; r24 = new n0
+ extru %r25,30,31,%r25 ; r25 = new n1
+ sub %r0,%r22,%r21
+ ds %r0,%r21,%r0
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ comclr,>= %r25,%r0,%r0
+ addl %r25,%r22,%r25
+ sh1addl %r25,%r19,%r25
+ stws %r25,0(0,%r26)
+ bv 0(%r2)
+ addc %r24,%r24,%r28
+
+L$odd addib,sv,n 1,%r22,L$FF.. ; r22 = (d / 2 + 1)
+ shd %r25,%r24,1,%r24 ; r24 = new n0
+ extru %r25,30,31,%r25 ; r25 = new n1
+ sub %r0,%r22,%r21
+ ds %r0,%r21,%r0
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r24
+ ds %r25,%r22,%r25
+ addc %r24,%r24,%r28
+ comclr,>= %r25,%r0,%r0
+ addl %r25,%r22,%r25
+ sh1addl %r25,%r19,%r25
+; We have computed (n1,,n0) / (d + 1), q' = r28, r' = r25
+ add,nuv %r28,%r25,%r25
+ addl %r25,%r1,%r25
+ addc %r0,%r28,%r28
+ sub,<< %r25,%r23,%r0
+ addl %r25,%r1,%r25
+ stws %r25,0(0,%r26)
+ bv 0(%r2)
+ addc %r0,%r28,%r28
+
+; This is just a special case of the code above.
+; We come here when d == 0xFFFFFFFF
+L$FF.. add,uv %r25,%r24,%r24
+ sub,<< %r24,%r23,%r0
+ ldo 1(%r24),%r24
+ stws %r24,0(0,%r26)
+ bv 0(%r2)
+ addc %r0,%r25,%r28
+
+ .exit
+ .procend
--- /dev/null
+.text
+ .align 4
+ .globl ___mpn_add_n
+___mpn_add_n:
+ mov 0,g6 # clear carry-save register
+ cmpo 1,0 # clear cy
+
+Loop: subo 1,g3,g3 # update loop counter
+ ld (g1),g5 # load from s1_ptr
+ addo 4,g1,g1 # s1_ptr++
+ ld (g2),g4 # load from s2_ptr
+ addo 4,g2,g2 # s2_ptr++
+ cmpo g6,1 # restore cy from g6, relies on cy being 0
+ addc g4,g5,g4 # main add
+ subc 0,0,g6 # save cy in g6
+ st g4,(g0) # store result to res_ptr
+ addo 4,g0,g0 # res_ptr++
+ cmpobne 0,g3,Loop # when branch is taken, clears C bit
+
+ mov g6,g0
+ ret
--- /dev/null
+.text
+ .align 4
+ .globl ___mpn_mul_1
+___mpn_mul_1:
+ subo g2,0,g2
+ shlo 2,g2,g4
+ subo g4,g1,g1
+ subo g4,g0,g13
+ mov 0,g0
+
+ cmpo 1,0 # clear C bit on AC.cc
+
+Loop: ld (g1)[g2*4],g5
+ emul g3,g5,g6
+ ld (g13)[g2*4],g5
+
+ addc g0,g6,g6 # relies on that C bit is clear
+ addc 0,g7,g7
+ addc g5,g6,g6 # relies on that C bit is clear
+ st g6,(g13)[g2*4]
+ addc 0,g7,g0
+
+ addo g2,1,g2
+ cmpobne 0,g2,Loop # when branch is taken, clears C bit
+
+ ret
--- /dev/null
+.text
+ .align 4
+ .globl ___mpn_mul_1
+___mpn_mul_1:
+ subo g2,0,g2
+ shlo 2,g2,g4
+ subo g4,g1,g1
+ subo g4,g0,g13
+ mov 0,g0
+
+ cmpo 1,0 # clear C bit on AC.cc
+
+Loop: ld (g1)[g2*4],g5
+ emul g3,g5,g6
+
+ addc g0,g6,g6 # relies on that C bit is clear
+ st g6,(g13)[g2*4]
+ addc 0,g7,g0
+
+ addo g2,1,g2
+ cmpobne 0,g2,Loop # when branch is taken, clears C bit
+
+ ret
--- /dev/null
+.text
+ .align 4
+ .globl ___mpn_sub_n
+___mpn_sub_n:
+ mov 1,g6 # set carry-save register
+ cmpo 1,0 # clear cy
+
+Loop: subo 1,g3,g3 # update loop counter
+ ld (g1),g5 # load from s1_ptr
+ addo 4,g1,g1 # s1_ptr++
+ ld (g2),g4 # load from s2_ptr
+ addo 4,g2,g2 # s2_ptr++
+ cmpo g6,1 # restore cy from g6, relies on cy being 0
+ subc g4,g5,g4 # main subtract
+ subc 0,0,g6 # save cy in g6
+ st g4,(g0) # store result to res_ptr
+ addo 4,g0,g0 # res_ptr++
+ cmpobne 0,g3,Loop # when branch is taken, cy will be 0
+
+ mov g6,g0
+ ret
--- /dev/null
+; mc88100 __mpn_add -- Add two limb vectors of the same length > 0 and store
+; sum in a third limb vector.
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r2
+; s1_ptr r3
+; s2_ptr r4
+; size r5
+
+; This code has been optimized to run one instruction per clock, avoiding
+; load stalls and writeback contention. As a result, the instruction
+; order is not always natural.
+
+; The speed is about 4.6 clocks/limb + 18 clocks/limb-vector on an 88100,
+; but on the 88110, it seems to run much slower, 6.6 clocks/limb.
+
+ text
+ align 16
+ global ___mpn_add_n
+___mpn_add_n:
+ ld r6,r3,0 ; read first limb from s1_ptr
+ extu r10,r5,3
+ ld r7,r4,0 ; read first limb from s2_ptr
+
+ subu.co r5,r0,r5 ; (clear carry as side effect)
+ mak r5,r5,3<4>
+ bcnd eq0,r5,Lzero
+
+ or r12,r0,lo16(Lbase)
+ or.u r12,r12,hi16(Lbase)
+ addu r12,r12,r5 ; r12 is address for entering in loop
+
+ extu r5,r5,2 ; divide by 4
+ subu r2,r2,r5 ; adjust res_ptr
+ subu r3,r3,r5 ; adjust s1_ptr
+ subu r4,r4,r5 ; adjust s2_ptr
+
+ or r8,r6,r0
+
+ jmp.n r12
+ or r9,r7,r0
+
+Loop: addu r3,r3,32
+ st r8,r2,28
+ addu r4,r4,32
+ ld r6,r3,0
+ addu r2,r2,32
+ ld r7,r4,0
+Lzero: subu r10,r10,1 ; add 0 + 8r limbs (adj loop cnt)
+Lbase: ld r8,r3,4
+ addu.cio r6,r6,r7
+ ld r9,r4,4
+ st r6,r2,0
+ ld r6,r3,8 ; add 7 + 8r limbs
+ addu.cio r8,r8,r9
+ ld r7,r4,8
+ st r8,r2,4
+ ld r8,r3,12 ; add 6 + 8r limbs
+ addu.cio r6,r6,r7
+ ld r9,r4,12
+ st r6,r2,8
+ ld r6,r3,16 ; add 5 + 8r limbs
+ addu.cio r8,r8,r9
+ ld r7,r4,16
+ st r8,r2,12
+ ld r8,r3,20 ; add 4 + 8r limbs
+ addu.cio r6,r6,r7
+ ld r9,r4,20
+ st r6,r2,16
+ ld r6,r3,24 ; add 3 + 8r limbs
+ addu.cio r8,r8,r9
+ ld r7,r4,24
+ st r8,r2,20
+ ld r8,r3,28 ; add 2 + 8r limbs
+ addu.cio r6,r6,r7
+ ld r9,r4,28
+ st r6,r2,24
+ bcnd.n ne0,r10,Loop ; add 1 + 8r limbs
+ addu.cio r8,r8,r9
+
+ st r8,r2,28 ; store most significant limb
+
+ jmp.n r1
+ addu.ci r2,r0,r0 ; return carry-out from most sign. limb
--- /dev/null
+; mc88100 __mpn_mul_1 -- Multiply a limb vector with a single limb and
+; store the product in a second limb vector.
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r2
+; s1_ptr r3
+; size r4
+; s2_limb r5
+
+; Common overhead is about 11 cycles/invocation.
+
+; The speed for S2_LIMB >= 0x10000 is approximately 21 cycles/limb. (The
+; pipeline stalls 2 cycles due to WB contention.)
+
+; The speed for S2_LIMB < 0x10000 is approximately 16 cycles/limb. (The
+; pipeline stalls 2 cycles due to WB contention and 1 cycle due to latency.)
+
+; To enhance speed:
+; 1. Unroll main loop 4-8 times.
+; 2. Schedule code to avoid WB contention. It might be tempting to move the
+; ld instruction in the loops down to save 2 cycles (less WB contention),
+; but that looses because the ultimate value will be read from outside
+; the allocated space. But if we handle the ultimate multiplication in
+; the tail, we can do this.
+; 3. Make the multiplication with less instructions. I think the code for
+; (S2_LIMB >= 0x10000) is not minimal.
+; With these techniques the (S2_LIMB >= 0x10000) case would run in 17 or
+; less cycles/limb; the (S2_LIMB < 0x10000) case would run in 11
+; cycles/limb. (Assuming infinite unrolling.)
+
+ text
+ align 16
+ global ___mpn_mul_1
+___mpn_mul_1:
+
+ ; Make S1_PTR and RES_PTR point at the end of their blocks
+ ; and negate SIZE.
+ lda r3,r3[r4]
+ lda r6,r2[r4] ; RES_PTR in r6 since r2 is retval
+ subu r4,r0,r4
+
+ addu.co r2,r0,r0 ; r2 = cy = 0
+ ld r9,r3[r4]
+ mask r7,r5,0xffff ; r7 = lo(S2_LIMB)
+ extu r8,r5,16 ; r8 = hi(S2_LIMB)
+ bcnd.n eq0,r8,Lsmall ; jump if (hi(S2_LIMB) == 0)
+ subu r6,r6,4
+
+; General code for any value of S2_LIMB.
+
+ ; Make a stack frame and save r25 and r26
+ subu r31,r31,16
+ st.d r25,r31,8
+
+ ; Enter the loop in the middle
+ br.n L1
+ addu r4,r4,1
+
+Loop:
+ ld r9,r3[r4]
+ st r26,r6[r4]
+; bcnd ne0,r0,0 ; bubble
+ addu r4,r4,1
+L1: mul r26,r9,r5 ; low word of product mul_1 WB ld
+ mask r12,r9,0xffff ; r12 = lo(s1_limb) mask_1
+ mul r11,r12,r7 ; r11 = prod_0 mul_2 WB mask_1
+ mul r10,r12,r8 ; r10 = prod_1a mul_3
+ extu r13,r9,16 ; r13 = hi(s1_limb) extu_1 WB mul_1
+ mul r12,r13,r7 ; r12 = prod_1b mul_4 WB extu_1
+ mul r25,r13,r8 ; r25 = prod_2 mul_5 WB mul_2
+ extu r11,r11,16 ; r11 = hi(prod_0) extu_2 WB mul_3
+ addu r10,r10,r11 ; addu_1 WB extu_2
+; bcnd ne0,r0,0 ; bubble WB addu_1
+ addu.co r10,r10,r12 ; WB mul_4
+ mask.u r10,r10,0xffff ; move the 16 most significant bits...
+ addu.ci r10,r10,r0 ; ...to the low half of the word...
+ rot r10,r10,16 ; ...and put carry in pos 16.
+ addu.co r26,r26,r2 ; add old carry limb
+ bcnd.n ne0,r4,Loop
+ addu.ci r2,r25,r10 ; compute new carry limb
+
+ st r26,r6[r4]
+ ld.d r25,r31,8
+ jmp.n r1
+ addu r31,r31,16
+
+; Fast code for S2_LIMB < 0x10000
+Lsmall:
+ ; Enter the loop in the middle
+ br.n SL1
+ addu r4,r4,1
+
+SLoop:
+ ld r9,r3[r4] ;
+ st r8,r6[r4] ;
+ addu r4,r4,1 ;
+SL1: mul r8,r9,r5 ; low word of product
+ mask r12,r9,0xffff ; r12 = lo(s1_limb)
+ extu r13,r9,16 ; r13 = hi(s1_limb)
+ mul r11,r12,r7 ; r11 = prod_0
+ mul r12,r13,r7 ; r12 = prod_1b
+ addu.cio r8,r8,r2 ; add old carry limb
+ extu r10,r11,16 ; r11 = hi(prod_0)
+ addu r10,r10,r12 ;
+ bcnd.n ne0,r4,SLoop
+ extu r2,r10,16 ; r2 = new carry limb
+
+ jmp.n r1
+ st r8,r6[r4]
--- /dev/null
+; mc88100 __mpn_sub -- Subtract two limb vectors of the same length > 0 and
+; store difference in a third limb vector.
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r2
+; s1_ptr r3
+; s2_ptr r4
+; size r5
+
+; This code has been optimized to run one instruction per clock, avoiding
+; load stalls and writeback contention. As a result, the instruction
+; order is not always natural.
+
+; The speed is about 4.6 clocks/limb + 18 clocks/limb-vector on an 88100,
+; but on the 88110, it seems to run much slower, 6.6 clocks/limb.
+
+ text
+ align 16
+ global ___mpn_sub_n
+___mpn_sub_n:
+ ld r6,r3,0 ; read first limb from s1_ptr
+ extu r10,r5,3
+ ld r7,r4,0 ; read first limb from s2_ptr
+
+ subu.co r5,r0,r5 ; (clear carry as side effect)
+ mak r5,r5,3<4>
+ bcnd eq0,r5,Lzero
+
+ or r12,r0,lo16(Lbase)
+ or.u r12,r12,hi16(Lbase)
+ addu r12,r12,r5 ; r12 is address for entering in loop
+
+ extu r5,r5,2 ; divide by 4
+ subu r2,r2,r5 ; adjust res_ptr
+ subu r3,r3,r5 ; adjust s1_ptr
+ subu r4,r4,r5 ; adjust s2_ptr
+
+ or r8,r6,r0
+
+ jmp.n r12
+ or r9,r7,r0
+
+Loop: addu r3,r3,32
+ st r8,r2,28
+ addu r4,r4,32
+ ld r6,r3,0
+ addu r2,r2,32
+ ld r7,r4,0
+Lzero: subu r10,r10,1 ; subtract 0 + 8r limbs (adj loop cnt)
+Lbase: ld r8,r3,4
+ subu.cio r6,r6,r7
+ ld r9,r4,4
+ st r6,r2,0
+ ld r6,r3,8 ; subtract 7 + 8r limbs
+ subu.cio r8,r8,r9
+ ld r7,r4,8
+ st r8,r2,4
+ ld r8,r3,12 ; subtract 6 + 8r limbs
+ subu.cio r6,r6,r7
+ ld r9,r4,12
+ st r6,r2,8
+ ld r6,r3,16 ; subtract 5 + 8r limbs
+ subu.cio r8,r8,r9
+ ld r7,r4,16
+ st r8,r2,12
+ ld r8,r3,20 ; subtract 4 + 8r limbs
+ subu.cio r6,r6,r7
+ ld r9,r4,20
+ st r6,r2,16
+ ld r6,r3,24 ; subtract 3 + 8r limbs
+ subu.cio r8,r8,r9
+ ld r7,r4,24
+ st r8,r2,20
+ ld r8,r3,28 ; subtract 2 + 8r limbs
+ subu.cio r6,r6,r7
+ ld r9,r4,28
+ st r6,r2,24
+ bcnd.n ne0,r10,Loop ; subtract 1 + 8r limbs
+ subu.cio r8,r8,r9
+
+ st r8,r2,28 ; store most significant limb
+
+ addu.ci r2,r0,r0 ; return carry-out from most sign. limb
+ jmp.n r1
+ xor r2,r2,1
--- /dev/null
+; mc88110 __mpn_mul_1 -- Multiply a limb vector with a single limb and
+; store the product in a second limb vector.
+
+; Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Library General Public License as published by
+; the Free Software Foundation; either version 2 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+; License for more details.
+
+; You should have received a copy of the GNU Library General Public License
+; along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+; the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr r2
+; s1_ptr r3
+; size r4
+; s2_limb r5
+
+ text
+ align 16
+ global ___mpn_mul_1
+___mpn_mul_1:
+ ; Make S1_PTR and RES_PTR point at the end of their blocks
+ ; and negate SIZE.
+ lda r3,r3[r4]
+ lda r8,r2[r4] ; RES_PTR in r8 since r2 is retval
+ subu r4,r0,r4
+
+ addu.co r2,r0,r0 ; r2 = cy = 0
+
+ ld r6,r3[r4]
+ addu r4,r4,1
+ mulu.d r10,r6,r5
+ bcnd.n eq0,r4,Lend
+ subu r8,r8,8
+
+Loop: ld r6,r3[r4]
+ addu.cio r9,r11,r2
+ or r2,r10,r0 ; could be avoided if unrolled
+ addu r4,r4,1
+ mulu.d r10,r6,r5
+ bcnd.n ne0,r4,Loop
+ st r9,r8[r4]
+
+Lend: addu.cio r9,r11,r2
+ st r9,r8,4
+ jmp.n r1
+ addu.ci r2,r10,r0
+
+; This is the Right Way to do this on '110. 4 cycles / 64-bit limb.
+; ld.d r10,
+; mulu.d
+; addu.cio
+; addu.cio
+; st.d
+; mulu.d ,r11,r5
+; ld.d r12,
+; mulu.d ,r10,r5
+; addu.cio
+; addu.cio
+; st.d
+; mulu.d
+; ld.d r10,
+; mulu.d
+; addu.cio
+; addu.cio
+; st.d
+; mulu.d
+; ld.d r10,
+; mulu.d
+; addu.cio
+; addu.cio
+; st.d
+; mulu.d
--- /dev/null
+ # MIPS2 __mpn_add_n -- Add two limb vectors of the same length > 0 and
+ # store sum in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # s2_ptr $6
+ # size $7
+
+ .text
+ .align 2
+ .globl __mpn_add_n
+ .ent __mpn_add_n
+__mpn_add_n:
+ .set noreorder
+ .set nomacro
+
+ lw $10,0($5)
+ lw $11,0($6)
+
+ addiu $7,$7,-1
+ and $9,$7,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ move $2,$0
+
+ subu $7,$7,$9
+
+.Loop0: addiu $9,$9,-1
+ lw $12,4($5)
+ addu $11,$11,$2
+ lw $13,4($6)
+ sltu $8,$11,$2
+ addu $11,$10,$11
+ sltu $2,$11,$10
+ sw $11,0($4)
+ or $2,$2,$8
+
+ addiu $5,$5,4
+ addiu $6,$6,4
+ move $10,$12
+ move $11,$13
+ bne $9,$0,.Loop0
+ addiu $4,$4,4
+
+.L0: beq $7,$0,.Lend
+ nop
+
+.Loop: addiu $7,$7,-4
+
+ lw $12,4($5)
+ addu $11,$11,$2
+ lw $13,4($6)
+ sltu $8,$11,$2
+ addu $11,$10,$11
+ sltu $2,$11,$10
+ sw $11,0($4)
+ or $2,$2,$8
+
+ lw $10,8($5)
+ addu $13,$13,$2
+ lw $11,8($6)
+ sltu $8,$13,$2
+ addu $13,$12,$13
+ sltu $2,$13,$12
+ sw $13,4($4)
+ or $2,$2,$8
+
+ lw $12,12($5)
+ addu $11,$11,$2
+ lw $13,12($6)
+ sltu $8,$11,$2
+ addu $11,$10,$11
+ sltu $2,$11,$10
+ sw $11,8($4)
+ or $2,$2,$8
+
+ lw $10,16($5)
+ addu $13,$13,$2
+ lw $11,16($6)
+ sltu $8,$13,$2
+ addu $13,$12,$13
+ sltu $2,$13,$12
+ sw $13,12($4)
+ or $2,$2,$8
+
+ addiu $5,$5,16
+ addiu $6,$6,16
+
+ bne $7,$0,.Loop
+ addiu $4,$4,16
+
+.Lend: addu $11,$11,$2
+ sltu $8,$11,$2
+ addu $11,$10,$11
+ sltu $2,$11,$10
+ sw $11,0($4)
+ j $31
+ or $2,$2,$8
+
+ .end __mpn_add_n
--- /dev/null
+ # MIPS __mpn_addmul_1 -- Multiply a limb vector with a single limb and
+ # add the product to a second limb vector.
+
+ # Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # size $6
+ # s2_limb $7
+
+ .text
+ .align 4
+ .globl __mpn_addmul_1
+ .ent __mpn_addmul_1
+__mpn_addmul_1:
+ .set noreorder
+ .set nomacro
+
+ # warm up phase 0
+ lw $8,0($5)
+
+ # warm up phase 1
+ addiu $5,$5,4
+ multu $8,$7
+
+ addiu $6,$6,-1
+ beq $6,$0,$LC0
+ move $2,$0 # zero cy2
+
+ addiu $6,$6,-1
+ beq $6,$0,$LC1
+ lw $8,0($5) # load new s1 limb as early as possible
+
+Loop: lw $10,0($4)
+ mflo $3
+ mfhi $9
+ addiu $5,$5,4
+ addu $3,$3,$2 # add old carry limb to low product limb
+ multu $8,$7
+ lw $8,0($5) # load new s1 limb as early as possible
+ addiu $6,$6,-1 # decrement loop counter
+ sltu $2,$3,$2 # carry from previous addition -> $2
+ addu $3,$10,$3
+ sltu $10,$3,$10
+ addu $2,$2,$10
+ sw $3,0($4)
+ addiu $4,$4,4
+ bne $6,$0,Loop # should be "bnel"
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 1
+$LC1: lw $10,0($4)
+ mflo $3
+ mfhi $9
+ addu $3,$3,$2
+ sltu $2,$3,$2
+ multu $8,$7
+ addu $3,$10,$3
+ sltu $10,$3,$10
+ addu $2,$2,$10
+ sw $3,0($4)
+ addiu $4,$4,4
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 0
+$LC0: lw $10,0($4)
+ mflo $3
+ mfhi $9
+ addu $3,$3,$2
+ sltu $2,$3,$2
+ addu $3,$10,$3
+ sltu $10,$3,$10
+ addu $2,$2,$10
+ sw $3,0($4)
+ j $31
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ .end __mpn_addmul_1
--- /dev/null
+ # MIPS2 __mpn_lshift --
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # src_ptr $5
+ # size $6
+ # cnt $7
+
+ .text
+ .align 2
+ .globl __mpn_lshift
+ .ent __mpn_lshift
+__mpn_lshift:
+ .set noreorder
+ .set nomacro
+
+ sll $2,$6,2
+ addu $5,$5,$2 # make r5 point at end of src
+ lw $10,-4($5) # load first limb
+ subu $13,$0,$7
+ addu $4,$4,$2 # make r4 point at end of res
+ addiu $6,$6,-1
+ and $9,$6,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ srl $2,$10,$13 # compute function result
+
+ subu $6,$6,$9
+
+.Loop0: lw $3,-8($5)
+ addiu $4,$4,-4
+ addiu $5,$5,-4
+ addiu $9,$9,-1
+ sll $11,$10,$7
+ srl $12,$3,$13
+ move $10,$3
+ or $8,$11,$12
+ bne $9,$0,.Loop0
+ sw $8,0($4)
+
+.L0: beq $6,$0,.Lend
+ nop
+
+.Loop: lw $3,-8($5)
+ addiu $4,$4,-16
+ addiu $6,$6,-4
+ sll $11,$10,$7
+ srl $12,$3,$13
+
+ lw $10,-12($5)
+ sll $14,$3,$7
+ or $8,$11,$12
+ sw $8,12($4)
+ srl $9,$10,$13
+
+ lw $3,-16($5)
+ sll $11,$10,$7
+ or $8,$14,$9
+ sw $8,8($4)
+ srl $12,$3,$13
+
+ lw $10,-20($5)
+ sll $14,$3,$7
+ or $8,$11,$12
+ sw $8,4($4)
+ srl $9,$10,$13
+
+ addiu $5,$5,-16
+ or $8,$14,$9
+ bgtz $6,.Loop
+ sw $8,0($4)
+
+.Lend: sll $8,$10,$7
+ j $31
+ sw $8,-4($4)
+ .end __mpn_lshift
--- /dev/null
+ # MIPS3 __mpn_add_n -- Add two limb vectors of the same length > 0 and
+ # store sum in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # s2_ptr $6
+ # size $7
+
+ .text
+ .align 2
+ .globl __mpn_add_n
+ .ent __mpn_add_n
+__mpn_add_n:
+ .set noreorder
+ .set nomacro
+
+ ld $10,0($5)
+ ld $11,0($6)
+
+ daddiu $7,$7,-1
+ and $9,$7,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ move $2,$0
+
+ dsubu $7,$7,$9
+
+.Loop0: daddiu $9,$9,-1
+ ld $12,8($5)
+ daddu $11,$11,$2
+ ld $13,8($6)
+ sltu $8,$11,$2
+ daddu $11,$10,$11
+ sltu $2,$11,$10
+ sd $11,0($4)
+ or $2,$2,$8
+
+ daddiu $5,$5,8
+ daddiu $6,$6,8
+ move $10,$12
+ move $11,$13
+ bne $9,$0,.Loop0
+ daddiu $4,$4,8
+
+.L0: beq $7,$0,.Lend
+ nop
+
+.Loop: daddiu $7,$7,-4
+
+ ld $12,8($5)
+ daddu $11,$11,$2
+ ld $13,8($6)
+ sltu $8,$11,$2
+ daddu $11,$10,$11
+ sltu $2,$11,$10
+ sd $11,0($4)
+ or $2,$2,$8
+
+ ld $10,16($5)
+ daddu $13,$13,$2
+ ld $11,16($6)
+ sltu $8,$13,$2
+ daddu $13,$12,$13
+ sltu $2,$13,$12
+ sd $13,8($4)
+ or $2,$2,$8
+
+ ld $12,24($5)
+ daddu $11,$11,$2
+ ld $13,24($6)
+ sltu $8,$11,$2
+ daddu $11,$10,$11
+ sltu $2,$11,$10
+ sd $11,16($4)
+ or $2,$2,$8
+
+ ld $10,32($5)
+ daddu $13,$13,$2
+ ld $11,32($6)
+ sltu $8,$13,$2
+ daddu $13,$12,$13
+ sltu $2,$13,$12
+ sd $13,24($4)
+ or $2,$2,$8
+
+ daddiu $5,$5,32
+ daddiu $6,$6,32
+
+ bne $7,$0,.Loop
+ daddiu $4,$4,32
+
+.Lend: daddu $11,$11,$2
+ sltu $8,$11,$2
+ daddu $11,$10,$11
+ sltu $2,$11,$10
+ sd $11,0($4)
+ j $31
+ or $2,$2,$8
+
+ .end __mpn_add_n
--- /dev/null
+ # MIPS3 __mpn_addmul_1 -- Multiply a limb vector with a single limb and
+ # add the product to a second limb vector.
+
+ # Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # size $6
+ # s2_limb $7
+
+ .text
+ .align 4
+ .globl __mpn_addmul_1
+ .ent __mpn_addmul_1
+__mpn_addmul_1:
+ .set noreorder
+ .set nomacro
+
+ # warm up phase 0
+ ld $8,0($5)
+
+ # warm up phase 1
+ daddiu $5,$5,8
+ dmultu $8,$7
+
+ daddiu $6,$6,-1
+ beq $6,$0,$LC0
+ move $2,$0 # zero cy2
+
+ daddiu $6,$6,-1
+ beq $6,$0,$LC1
+ ld $8,0($5) # load new s1 limb as early as possible
+
+Loop: ld $10,0($4)
+ mflo $3
+ mfhi $9
+ daddiu $5,$5,8
+ daddu $3,$3,$2 # add old carry limb to low product limb
+ dmultu $8,$7
+ ld $8,0($5) # load new s1 limb as early as possible
+ daddiu $6,$6,-1 # decrement loop counter
+ sltu $2,$3,$2 # carry from previous addition -> $2
+ daddu $3,$10,$3
+ sltu $10,$3,$10
+ daddu $2,$2,$10
+ sd $3,0($4)
+ daddiu $4,$4,8
+ bne $6,$0,Loop # should be "bnel"
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 1
+$LC1: ld $10,0($4)
+ mflo $3
+ mfhi $9
+ daddu $3,$3,$2
+ sltu $2,$3,$2
+ dmultu $8,$7
+ daddu $3,$10,$3
+ sltu $10,$3,$10
+ daddu $2,$2,$10
+ sd $3,0($4)
+ daddiu $4,$4,8
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 0
+$LC0: ld $10,0($4)
+ mflo $3
+ mfhi $9
+ daddu $3,$3,$2
+ sltu $2,$3,$2
+ daddu $3,$10,$3
+ sltu $10,$3,$10
+ daddu $2,$2,$10
+ sd $3,0($4)
+ j $31
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ .end __mpn_addmul_1
--- /dev/null
+ # MIPS3 __mpn_lshift --
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # src_ptr $5
+ # size $6
+ # cnt $7
+
+ .text
+ .align 2
+ .globl __mpn_lshift
+ .ent __mpn_lshift
+__mpn_lshift:
+ .set noreorder
+ .set nomacro
+
+ dsll $2,$6,3
+ daddu $5,$5,$2 # make r5 point at end of src
+ ld $10,-8($5) # load first limb
+ dsubu $13,$0,$7
+ daddu $4,$4,$2 # make r4 point at end of res
+ daddiu $6,$6,-1
+ and $9,$6,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ dsrl $2,$10,$13 # compute function result
+
+ dsubu $6,$6,$9
+
+.Loop0: ld $3,-16($5)
+ daddiu $4,$4,-8
+ daddiu $5,$5,-8
+ daddiu $9,$9,-1
+ dsll $11,$10,$7
+ dsrl $12,$3,$13
+ move $10,$3
+ or $8,$11,$12
+ bne $9,$0,.Loop0
+ sd $8,0($4)
+
+.L0: beq $6,$0,.Lend
+ nop
+
+.Loop: ld $3,-16($5)
+ daddiu $4,$4,-32
+ daddiu $6,$6,-4
+ dsll $11,$10,$7
+ dsrl $12,$3,$13
+
+ ld $10,-24($5)
+ dsll $14,$3,$7
+ or $8,$11,$12
+ sd $8,24($4)
+ dsrl $9,$10,$13
+
+ ld $3,-32($5)
+ dsll $11,$10,$7
+ or $8,$14,$9
+ sd $8,16($4)
+ dsrl $12,$3,$13
+
+ ld $10,-40($5)
+ dsll $14,$3,$7
+ or $8,$11,$12
+ sd $8,8($4)
+ dsrl $9,$10,$13
+
+ daddiu $5,$5,-32
+ or $8,$14,$9
+ bgtz $6,.Loop
+ sd $8,0($4)
+
+.Lend: dsll $8,$10,$7
+ j $31
+ sd $8,-8($4)
+ .end __mpn_lshift
--- /dev/null
+ # MIPS3 __mpn_mul_1 -- Multiply a limb vector with a single limb and
+ # store the product in a second limb vector.
+
+ # Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # size $6
+ # s2_limb $7
+
+ .text
+ .align 4
+ .globl __mpn_mul_1
+ .ent __mpn_mul_1
+__mpn_mul_1:
+ .set noreorder
+ .set nomacro
+
+ # warm up phase 0
+ ld $8,0($5)
+
+ # warm up phase 1
+ daddiu $5,$5,8
+ dmultu $8,$7
+
+ daddiu $6,$6,-1
+ beq $6,$0,$LC0
+ move $2,$0 # zero cy2
+
+ daddiu $6,$6,-1
+ beq $6,$0,$LC1
+ ld $8,0($5) # load new s1 limb as early as possible
+
+Loop: mflo $10
+ mfhi $9
+ daddiu $5,$5,8
+ daddu $10,$10,$2 # add old carry limb to low product limb
+ dmultu $8,$7
+ ld $8,0($5) # load new s1 limb as early as possible
+ daddiu $6,$6,-1 # decrement loop counter
+ sltu $2,$10,$2 # carry from previous addition -> $2
+ sd $10,0($4)
+ daddiu $4,$4,8
+ bne $6,$0,Loop # should be "bnel"
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 1
+$LC1: mflo $10
+ mfhi $9
+ daddu $10,$10,$2
+ sltu $2,$10,$2
+ dmultu $8,$7
+ sd $10,0($4)
+ daddiu $4,$4,8
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 0
+$LC0: mflo $10
+ mfhi $9
+ daddu $10,$10,$2
+ sltu $2,$10,$2
+ sd $10,0($4)
+ j $31
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ .end __mpn_mul_1
--- /dev/null
+ # MIPS3 __mpn_rshift --
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # src_ptr $5
+ # size $6
+ # cnt $7
+
+ .text
+ .align 2
+ .globl __mpn_rshift
+ .ent __mpn_rshift
+__mpn_rshift:
+ .set noreorder
+ .set nomacro
+
+ ld $10,0($5) # load first limb
+ dsubu $13,$0,$7
+ daddiu $6,$6,-1
+ and $9,$6,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ dsll $2,$10,$13 # compute function result
+
+ dsubu $6,$6,$9
+
+.Loop0: ld $3,8($5)
+ daddiu $4,$4,8
+ daddiu $5,$5,8
+ daddiu $9,$9,-1
+ dsrl $11,$10,$7
+ dsll $12,$3,$13
+ move $10,$3
+ or $8,$11,$12
+ bne $9,$0,.Loop0
+ sd $8,-8($4)
+
+.L0: beq $6,$0,.Lend
+ nop
+
+.Loop: ld $3,8($5)
+ daddiu $4,$4,32
+ daddiu $6,$6,-4
+ dsrl $11,$10,$7
+ dsll $12,$3,$13
+
+ ld $10,16($5)
+ dsrl $14,$3,$7
+ or $8,$11,$12
+ sd $8,-32($4)
+ dsll $9,$10,$13
+
+ ld $3,24($5)
+ dsrl $11,$10,$7
+ or $8,$14,$9
+ sd $8,-24($4)
+ dsll $12,$3,$13
+
+ ld $10,32($5)
+ dsrl $14,$3,$7
+ or $8,$11,$12
+ sd $8,-16($4)
+ dsll $9,$10,$13
+
+ daddiu $5,$5,32
+ or $8,$14,$9
+ bgtz $6,.Loop
+ sd $8,-8($4)
+
+.Lend: dsrl $8,$10,$7
+ j $31
+ sd $8,0($4)
+ .end __mpn_rshift
--- /dev/null
+ # MIPS3 __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+ # store difference in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # s2_ptr $6
+ # size $7
+
+ .text
+ .align 2
+ .globl __mpn_sub_n
+ .ent __mpn_sub_n
+__mpn_sub_n:
+ .set noreorder
+ .set nomacro
+
+ ld $10,0($5)
+ ld $11,0($6)
+
+ daddiu $7,$7,-1
+ and $9,$7,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ move $2,$0
+
+ dsubu $7,$7,$9
+
+.Loop0: daddiu $9,$9,-1
+ ld $12,8($5)
+ daddu $11,$11,$2
+ ld $13,8($6)
+ sltu $8,$11,$2
+ dsubu $11,$10,$11
+ sltu $2,$10,$11
+ sd $11,0($4)
+ or $2,$2,$8
+
+ daddiu $5,$5,8
+ daddiu $6,$6,8
+ move $10,$12
+ move $11,$13
+ bne $9,$0,.Loop0
+ daddiu $4,$4,8
+
+.L0: beq $7,$0,.Lend
+ nop
+
+.Loop: daddiu $7,$7,-4
+
+ ld $12,8($5)
+ daddu $11,$11,$2
+ ld $13,8($6)
+ sltu $8,$11,$2
+ dsubu $11,$10,$11
+ sltu $2,$10,$11
+ sd $11,0($4)
+ or $2,$2,$8
+
+ ld $10,16($5)
+ daddu $13,$13,$2
+ ld $11,16($6)
+ sltu $8,$13,$2
+ dsubu $13,$12,$13
+ sltu $2,$12,$13
+ sd $13,8($4)
+ or $2,$2,$8
+
+ ld $12,24($5)
+ daddu $11,$11,$2
+ ld $13,24($6)
+ sltu $8,$11,$2
+ dsubu $11,$10,$11
+ sltu $2,$10,$11
+ sd $11,16($4)
+ or $2,$2,$8
+
+ ld $10,32($5)
+ daddu $13,$13,$2
+ ld $11,32($6)
+ sltu $8,$13,$2
+ dsubu $13,$12,$13
+ sltu $2,$12,$13
+ sd $13,24($4)
+ or $2,$2,$8
+
+ daddiu $5,$5,32
+ daddiu $6,$6,32
+
+ bne $7,$0,.Loop
+ daddiu $4,$4,32
+
+.Lend: daddu $11,$11,$2
+ sltu $8,$11,$2
+ dsubu $11,$10,$11
+ sltu $2,$10,$11
+ sd $11,0($4)
+ j $31
+ or $2,$2,$8
+
+ .end __mpn_sub_n
--- /dev/null
+ # MIPS3 __mpn_submul_1 -- Multiply a limb vector with a single limb and
+ # subtract the product from a second limb vector.
+
+ # Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # size $6
+ # s2_limb $7
+
+ .text
+ .align 4
+ .globl __mpn_submul_1
+ .ent __mpn_submul_1
+__mpn_submul_1:
+ .set noreorder
+ .set nomacro
+
+ # warm up phase 0
+ ld $8,0($5)
+
+ # warm up phase 1
+ daddiu $5,$5,8
+ dmultu $8,$7
+
+ daddiu $6,$6,-1
+ beq $6,$0,$LC0
+ move $2,$0 # zero cy2
+
+ daddiu $6,$6,-1
+ beq $6,$0,$LC1
+ ld $8,0($5) # load new s1 limb as early as possible
+
+Loop: ld $10,0($4)
+ mflo $3
+ mfhi $9
+ daddiu $5,$5,8
+ daddu $3,$3,$2 # add old carry limb to low product limb
+ dmultu $8,$7
+ ld $8,0($5) # load new s1 limb as early as possible
+ daddiu $6,$6,-1 # decrement loop counter
+ sltu $2,$3,$2 # carry from previous addition -> $2
+ dsubu $3,$10,$3
+ sgtu $10,$3,$10
+ daddu $2,$2,$10
+ sd $3,0($4)
+ daddiu $4,$4,8
+ bne $6,$0,Loop # should be "bnel"
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 1
+$LC1: ld $10,0($4)
+ mflo $3
+ mfhi $9
+ daddu $3,$3,$2
+ sltu $2,$3,$2
+ dmultu $8,$7
+ dsubu $3,$10,$3
+ sgtu $10,$3,$10
+ daddu $2,$2,$10
+ sd $3,0($4)
+ daddiu $4,$4,8
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 0
+$LC0: ld $10,0($4)
+ mflo $3
+ mfhi $9
+ daddu $3,$3,$2
+ sltu $2,$3,$2
+ dsubu $3,$10,$3
+ sgtu $10,$3,$10
+ daddu $2,$2,$10
+ sd $3,0($4)
+ j $31
+ daddu $2,$9,$2 # add high product limb and carry from addition
+
+ .end __mpn_submul_1
--- /dev/null
+ # MIPS __mpn_mul_1 -- Multiply a limb vector with a single limb and
+ # store the product in a second limb vector.
+
+ # Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # size $6
+ # s2_limb $7
+
+ .text
+ .align 4
+ .globl __mpn_mul_1
+ .ent __mpn_mul_1
+__mpn_mul_1:
+ .set noreorder
+ .set nomacro
+
+ # warm up phase 0
+ lw $8,0($5)
+
+ # warm up phase 1
+ addiu $5,$5,4
+ multu $8,$7
+
+ addiu $6,$6,-1
+ beq $6,$0,$LC0
+ move $2,$0 # zero cy2
+
+ addiu $6,$6,-1
+ beq $6,$0,$LC1
+ lw $8,0($5) # load new s1 limb as early as possible
+
+Loop: mflo $10
+ mfhi $9
+ addiu $5,$5,4
+ addu $10,$10,$2 # add old carry limb to low product limb
+ multu $8,$7
+ lw $8,0($5) # load new s1 limb as early as possible
+ addiu $6,$6,-1 # decrement loop counter
+ sltu $2,$10,$2 # carry from previous addition -> $2
+ sw $10,0($4)
+ addiu $4,$4,4
+ bne $6,$0,Loop # should be "bnel"
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 1
+$LC1: mflo $10
+ mfhi $9
+ addu $10,$10,$2
+ sltu $2,$10,$2
+ multu $8,$7
+ sw $10,0($4)
+ addiu $4,$4,4
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 0
+$LC0: mflo $10
+ mfhi $9
+ addu $10,$10,$2
+ sltu $2,$10,$2
+ sw $10,0($4)
+ j $31
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ .end __mpn_mul_1
--- /dev/null
+ # MIPS2 __mpn_rshift --
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # src_ptr $5
+ # size $6
+ # cnt $7
+
+ .text
+ .align 2
+ .globl __mpn_rshift
+ .ent __mpn_rshift
+__mpn_rshift:
+ .set noreorder
+ .set nomacro
+
+ lw $10,0($5) # load first limb
+ subu $13,$0,$7
+ addiu $6,$6,-1
+ and $9,$6,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ sll $2,$10,$13 # compute function result
+
+ subu $6,$6,$9
+
+.Loop0: lw $3,4($5)
+ addiu $4,$4,4
+ addiu $5,$5,4
+ addiu $9,$9,-1
+ srl $11,$10,$7
+ sll $12,$3,$13
+ move $10,$3
+ or $8,$11,$12
+ bne $9,$0,.Loop0
+ sw $8,-4($4)
+
+.L0: beq $6,$0,.Lend
+ nop
+
+.Loop: lw $3,4($5)
+ addiu $4,$4,16
+ addiu $6,$6,-4
+ srl $11,$10,$7
+ sll $12,$3,$13
+
+ lw $10,8($5)
+ srl $14,$3,$7
+ or $8,$11,$12
+ sw $8,-16($4)
+ sll $9,$10,$13
+
+ lw $3,12($5)
+ srl $11,$10,$7
+ or $8,$14,$9
+ sw $8,-12($4)
+ sll $12,$3,$13
+
+ lw $10,16($5)
+ srl $14,$3,$7
+ or $8,$11,$12
+ sw $8,-8($4)
+ sll $9,$10,$13
+
+ addiu $5,$5,16
+ or $8,$14,$9
+ bgtz $6,.Loop
+ sw $8,-4($4)
+
+.Lend: srl $8,$10,$7
+ j $31
+ sw $8,0($4)
+ .end __mpn_rshift
--- /dev/null
+ # MIPS2 __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+ # store difference in a third limb vector.
+
+ # Copyright (C) 1995 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # s2_ptr $6
+ # size $7
+
+ .text
+ .align 2
+ .globl __mpn_sub_n
+ .ent __mpn_sub_n
+__mpn_sub_n:
+ .set noreorder
+ .set nomacro
+
+ lw $10,0($5)
+ lw $11,0($6)
+
+ addiu $7,$7,-1
+ and $9,$7,4-1 # number of limbs in first loop
+ beq $9,$0,.L0 # if multiple of 4 limbs, skip first loop
+ move $2,$0
+
+ subu $7,$7,$9
+
+.Loop0: addiu $9,$9,-1
+ lw $12,4($5)
+ addu $11,$11,$2
+ lw $13,4($6)
+ sltu $8,$11,$2
+ subu $11,$10,$11
+ sltu $2,$10,$11
+ sw $11,0($4)
+ or $2,$2,$8
+
+ addiu $5,$5,4
+ addiu $6,$6,4
+ move $10,$12
+ move $11,$13
+ bne $9,$0,.Loop0
+ addiu $4,$4,4
+
+.L0: beq $7,$0,.Lend
+ nop
+
+.Loop: addiu $7,$7,-4
+
+ lw $12,4($5)
+ addu $11,$11,$2
+ lw $13,4($6)
+ sltu $8,$11,$2
+ subu $11,$10,$11
+ sltu $2,$10,$11
+ sw $11,0($4)
+ or $2,$2,$8
+
+ lw $10,8($5)
+ addu $13,$13,$2
+ lw $11,8($6)
+ sltu $8,$13,$2
+ subu $13,$12,$13
+ sltu $2,$12,$13
+ sw $13,4($4)
+ or $2,$2,$8
+
+ lw $12,12($5)
+ addu $11,$11,$2
+ lw $13,12($6)
+ sltu $8,$11,$2
+ subu $11,$10,$11
+ sltu $2,$10,$11
+ sw $11,8($4)
+ or $2,$2,$8
+
+ lw $10,16($5)
+ addu $13,$13,$2
+ lw $11,16($6)
+ sltu $8,$13,$2
+ subu $13,$12,$13
+ sltu $2,$12,$13
+ sw $13,12($4)
+ or $2,$2,$8
+
+ addiu $5,$5,16
+ addiu $6,$6,16
+
+ bne $7,$0,.Loop
+ addiu $4,$4,16
+
+.Lend: addu $11,$11,$2
+ sltu $8,$11,$2
+ subu $11,$10,$11
+ sltu $2,$10,$11
+ sw $11,0($4)
+ j $31
+ or $2,$2,$8
+
+ .end __mpn_sub_n
--- /dev/null
+ # MIPS __mpn_submul_1 -- Multiply a limb vector with a single limb and
+ # subtract the product from a second limb vector.
+
+ # Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+ # This file is part of the GNU MP Library.
+
+ # The GNU MP Library is free software; you can redistribute it and/or modify
+ # it under the terms of the GNU Library General Public License as published by
+ # the Free Software Foundation; either version 2 of the License, or (at your
+ # option) any later version.
+
+ # The GNU MP Library is distributed in the hope that it will be useful, but
+ # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+ # License for more details.
+
+ # You should have received a copy of the GNU Library General Public License
+ # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+ # the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+ # INPUT PARAMETERS
+ # res_ptr $4
+ # s1_ptr $5
+ # size $6
+ # s2_limb $7
+
+ .text
+ .align 4
+ .globl __mpn_submul_1
+ .ent __mpn_submul_1
+__mpn_submul_1:
+ .set noreorder
+ .set nomacro
+
+ # warm up phase 0
+ lw $8,0($5)
+
+ # warm up phase 1
+ addiu $5,$5,4
+ multu $8,$7
+
+ addiu $6,$6,-1
+ beq $6,$0,$LC0
+ move $2,$0 # zero cy2
+
+ addiu $6,$6,-1
+ beq $6,$0,$LC1
+ lw $8,0($5) # load new s1 limb as early as possible
+
+Loop: lw $10,0($4)
+ mflo $3
+ mfhi $9
+ addiu $5,$5,4
+ addu $3,$3,$2 # add old carry limb to low product limb
+ multu $8,$7
+ lw $8,0($5) # load new s1 limb as early as possible
+ addiu $6,$6,-1 # decrement loop counter
+ sltu $2,$3,$2 # carry from previous addition -> $2
+ subu $3,$10,$3
+ sgtu $10,$3,$10
+ addu $2,$2,$10
+ sw $3,0($4)
+ addiu $4,$4,4
+ bne $6,$0,Loop # should be "bnel"
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 1
+$LC1: lw $10,0($4)
+ mflo $3
+ mfhi $9
+ addu $3,$3,$2
+ sltu $2,$3,$2
+ multu $8,$7
+ subu $3,$10,$3
+ sgtu $10,$3,$10
+ addu $2,$2,$10
+ sw $3,0($4)
+ addiu $4,$4,4
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ # cool down phase 0
+$LC0: lw $10,0($4)
+ mflo $3
+ mfhi $9
+ addu $3,$3,$2
+ sltu $2,$3,$2
+ subu $3,$10,$3
+ sgtu $10,$3,$10
+ addu $2,$2,$10
+ sw $3,0($4)
+ j $31
+ addu $2,$9,$2 # add high product limb and carry from addition
+
+ .end __mpn_submul_1
--- /dev/null
+# IBM POWER __mpn_add_n -- Add two limb vectors of equal, non-zero length.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s1_ptr r4
+# s2_ptr r5
+# size r6
+
+ .toc
+ .extern __mpn_add_n[DS]
+ .extern .__mpn_add_n
+.csect [PR]
+ .align 2
+ .globl __mpn_add_n
+ .globl .__mpn_add_n
+ .csect __mpn_add_n[DS]
+__mpn_add_n:
+ .long .__mpn_add_n, TOC[tc0], 0
+ .csect [PR]
+.__mpn_add_n:
+ mtctr 6 # copy size into CTR
+ l 8,0(4) # load least significant s1 limb
+ l 0,0(5) # load least significant s2 limb
+ cal 3,-4(3) # offset res_ptr, it's updated before used
+ a 7,0,8 # add least significant limbs, set cy
+ bdz Lend # If done, skip loop
+Loop: lu 8,4(4) # load s1 limb and update s1_ptr
+ lu 0,4(5) # load s2 limb and update s2_ptr
+ stu 7,4(3) # store previous limb in load latecny slot
+ ae 7,0,8 # add new limbs with cy, set cy
+ bdn Loop # decrement CTR and loop back
+Lend: st 7,4(3) # store ultimate result limb
+ lil 3,0 # load cy into ...
+ aze 3,3 # ... return value register
+ br
--- /dev/null
+# IBM POWER __mpn_addmul_1 -- Multiply a limb vector with a limb and add
+# the result to a second limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s1_ptr r4
+# size r5
+# s2_limb r6
+
+# The RS/6000 has no unsigned 32x32->64 bit multiplication instruction. To
+# obtain that operation, we have to use the 32x32->64 signed multiplication
+# instruction, and add the appropriate compensation to the high limb of the
+# result. We add the multiplicand if the multiplier has its most significant
+# bit set, and we add the multiplier if the multiplicand has its most
+# significant bit set. We need to preserve the carry flag between each
+# iteration, so we have to compute the compensation carefully (the natural,
+# srai+and doesn't work). Since the POWER architecture has a branch unit
+# we can branch in zero cycles, so that's how we perform the additions.
+
+ .toc
+ .csect .__mpn_addmul_1[PR]
+ .align 2
+ .globl __mpn_addmul_1
+ .globl .__mpn_addmul_1
+ .csect __mpn_addmul_1[DS]
+__mpn_addmul_1:
+ .long .__mpn_addmul_1[PR], TOC[tc0], 0
+ .csect .__mpn_addmul_1[PR]
+.__mpn_addmul_1:
+
+ cal 3,-4(3)
+ l 0,0(4)
+ cmpi 0,6,0
+ mtctr 5
+ mul 9,0,6
+ srai 7,0,31
+ and 7,7,6
+ mfmq 8
+ cax 9,9,7
+ l 7,4(3)
+ a 8,8,7 # add res_limb
+ blt Lneg
+Lpos: bdz Lend
+
+Lploop: lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 10,0,6
+ mfmq 0
+ ae 8,0,9 # low limb + old_cy_limb + old cy
+ l 7,4(3)
+ aze 10,10 # propagate cy to new cy_limb
+ a 8,8,7 # add res_limb
+ bge Lp0
+ cax 10,10,6 # adjust high limb for negative limb from s1
+Lp0: bdz Lend0
+ lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 9,0,6
+ mfmq 0
+ ae 8,0,10
+ l 7,4(3)
+ aze 9,9
+ a 8,8,7
+ bge Lp1
+ cax 9,9,6 # adjust high limb for negative limb from s1
+Lp1: bdn Lploop
+
+ b Lend
+
+Lneg: cax 9,9,0
+ bdz Lend
+Lnloop: lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 10,0,6
+ mfmq 7
+ ae 8,7,9
+ l 7,4(3)
+ ae 10,10,0 # propagate cy to new cy_limb
+ a 8,8,7 # add res_limb
+ bge Ln0
+ cax 10,10,6 # adjust high limb for negative limb from s1
+Ln0: bdz Lend0
+ lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 9,0,6
+ mfmq 7
+ ae 8,7,10
+ l 7,4(3)
+ ae 9,9,0 # propagate cy to new cy_limb
+ a 8,8,7 # add res_limb
+ bge Ln1
+ cax 9,9,6 # adjust high limb for negative limb from s1
+Ln1: bdn Lnloop
+ b Lend
+
+Lend0: cal 9,0(10)
+Lend: st 8,4(3)
+ aze 3,9
+ br
--- /dev/null
+# IBM POWER __mpn_lshift --
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s_ptr r4
+# size r5
+# cnt r6
+
+ .toc
+ .extern __mpn_lshift[DS]
+ .extern .__mpn_lshift
+.csect [PR]
+ .align 2
+ .globl __mpn_lshift
+ .globl .__mpn_lshift
+ .csect __mpn_lshift[DS]
+__mpn_lshift:
+ .long .__mpn_lshift, TOC[tc0], 0
+ .csect [PR]
+.__mpn_lshift:
+ sli 0,5,2
+ cax 9,3,0
+ cax 4,4,0
+ sfi 8,6,32
+ mtctr 5 # put limb count in CTR loop register
+ lu 0,-4(4) # read most significant limb
+ sre 3,0,8 # compute carry out limb, and init MQ register
+ bdz Lend2 # if just one limb, skip loop
+ lu 0,-4(4) # read 2:nd most significant limb
+ sreq 7,0,8 # compute most significant limb of result
+ bdz Lend # if just two limb, skip loop
+Loop: lu 0,-4(4) # load next lower limb
+ stu 7,-4(9) # store previous result during read latency
+ sreq 7,0,8 # compute result limb
+ bdn Loop # loop back until CTR is zero
+Lend: stu 7,-4(9) # store 2:nd least significant limb
+Lend2: sle 7,0,6 # compute least significant limb
+ st 7,-4(9) # store it" \
+ br
--- /dev/null
+# IBM POWER __mpn_mul_1 -- Multiply a limb vector with a limb and store
+# the result in a second limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s1_ptr r4
+# size r5
+# s2_limb r6
+
+# The RS/6000 has no unsigned 32x32->64 bit multiplication instruction. To
+# obtain that operation, we have to use the 32x32->64 signed multiplication
+# instruction, and add the appropriate compensation to the high limb of the
+# result. We add the multiplicand if the multiplier has its most significant
+# bit set, and we add the multiplier if the multiplicand has its most
+# significant bit set. We need to preserve the carry flag between each
+# iteration, so we have to compute the compensation carefully (the natural,
+# srai+and doesn't work). Since the POWER architecture has a branch unit
+# we can branch in zero cycles, so that's how we perform the additions.
+
+ .toc
+ .csect .__mpn_mul_1[PR]
+ .align 2
+ .globl __mpn_mul_1
+ .globl .__mpn_mul_1
+ .csect __mpn_mul_1[DS]
+__mpn_mul_1:
+ .long .__mpn_mul_1[PR], TOC[tc0], 0
+ .csect .__mpn_mul_1[PR]
+.__mpn_mul_1:
+
+ cal 3,-4(3)
+ l 0,0(4)
+ cmpi 0,6,0
+ mtctr 5
+ mul 9,0,6
+ srai 7,0,31
+ and 7,7,6
+ mfmq 8
+ ai 0,0,0 # reset carry
+ cax 9,9,7
+ blt Lneg
+Lpos: bdz Lend
+Lploop: lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 10,0,6
+ mfmq 0
+ ae 8,0,9
+ bge Lp0
+ cax 10,10,6 # adjust high limb for negative limb from s1
+Lp0: bdz Lend0
+ lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 9,0,6
+ mfmq 0
+ ae 8,0,10
+ bge Lp1
+ cax 9,9,6 # adjust high limb for negative limb from s1
+Lp1: bdn Lploop
+ b Lend
+
+Lneg: cax 9,9,0
+ bdz Lend
+Lnloop: lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 10,0,6
+ cax 10,10,0 # adjust high limb for negative s2_limb
+ mfmq 0
+ ae 8,0,9
+ bge Ln0
+ cax 10,10,6 # adjust high limb for negative limb from s1
+Ln0: bdz Lend0
+ lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 9,0,6
+ cax 9,9,0 # adjust high limb for negative s2_limb
+ mfmq 0
+ ae 8,0,10
+ bge Ln1
+ cax 9,9,6 # adjust high limb for negative limb from s1
+Ln1: bdn Lnloop
+ b Lend
+
+Lend0: cal 9,0(10)
+Lend: st 8,4(3)
+ aze 3,9
+ br
--- /dev/null
+# IBM POWER __mpn_rshift --
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s_ptr r4
+# size r5
+# cnt r6
+
+ .toc
+ .extern __mpn_rshift[DS]
+ .extern .__mpn_rshift
+.csect [PR]
+ .align 2
+ .globl __mpn_rshift
+ .globl .__mpn_rshift
+ .csect __mpn_rshift[DS]
+__mpn_rshift:
+ .long .__mpn_rshift, TOC[tc0], 0
+ .csect [PR]
+.__mpn_rshift:
+ sfi 8,6,32
+ mtctr 5 # put limb count in CTR loop register
+ l 0,0(4) # read least significant limb
+ ai 9,3,-4 # adjust res_ptr since it's offset in the stu:s
+ sle 3,0,8 # compute carry limb, and init MQ register
+ bdz Lend2 # if just one limb, skip loop
+ lu 0,4(4) # read 2:nd least significant limb
+ sleq 7,0,8 # compute least significant limb of result
+ bdz Lend # if just two limb, skip loop
+Loop: lu 0,4(4) # load next higher limb
+ stu 7,4(9) # store previous result during read latency
+ sleq 7,0,8 # compute result limb
+ bdn Loop # loop back until CTR is zero
+Lend: stu 7,4(9) # store 2:nd most significant limb
+Lend2: sre 7,0,6 # compute most significant limb
+ st 7,4(9) # store it" \
+ br
--- /dev/null
+# IBM POWER __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+# store difference in a third limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s1_ptr r4
+# s2_ptr r5
+# size r6
+
+ .toc
+ .extern __mpn_sub_n[DS]
+ .extern .__mpn_sub_n
+.csect [PR]
+ .align 2
+ .globl __mpn_sub_n
+ .globl .__mpn_sub_n
+ .csect __mpn_sub_n[DS]
+__mpn_sub_n:
+ .long .__mpn_sub_n, TOC[tc0], 0
+ .csect [PR]
+.__mpn_sub_n:
+ mtctr 6 # copy size into CTR
+ l 8,0(4) # load least significant s1 limb
+ l 0,0(5) # load least significant s2 limb
+ cal 3,-4(3) # offset res_ptr, it's updated before used
+ sf 7,0,8 # add least significant limbs, set cy
+ bdz Lend # If done, skip loop
+Loop: lu 8,4(4) # load s1 limb and update s1_ptr
+ lu 0,4(5) # load s2 limb and update s2_ptr
+ stu 7,4(3) # store previous limb in load latecny slot
+ sfe 7,0,8 # add new limbs with cy, set cy
+ bdn Loop # decrement CTR and loop back
+Lend: st 7,4(3) # store ultimate result limb
+ sfe 3,0,0 # load !cy into ...
+ sfi 3,3,0 # ... return value register
+ br
--- /dev/null
+# IBM POWER __mpn_submul_1 -- Multiply a limb vector with a limb and subtract
+# the result from a second limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr r3
+# s1_ptr r4
+# size r5
+# s2_limb r6
+
+# The RS/6000 has no unsigned 32x32->64 bit multiplication instruction. To
+# obtain that operation, we have to use the 32x32->64 signed multiplication
+# instruction, and add the appropriate compensation to the high limb of the
+# result. We add the multiplicand if the multiplier has its most significant
+# bit set, and we add the multiplier if the multiplicand has its most
+# significant bit set. We need to preserve the carry flag between each
+# iteration, so we have to compute the compensation carefully (the natural,
+# srai+and doesn't work). Since the POWER architecture has a branch unit
+# we can branch in zero cycles, so that's how we perform the additions.
+
+ .toc
+ .csect .__mpn_submul_1[PR]
+ .align 2
+ .globl __mpn_submul_1
+ .globl .__mpn_submul_1
+ .csect __mpn_submul_1[DS]
+__mpn_submul_1:
+ .long .__mpn_submul_1[PR], TOC[tc0], 0
+ .csect .__mpn_submul_1[PR]
+.__mpn_submul_1:
+
+ cal 3,-4(3)
+ l 0,0(4)
+ cmpi 0,6,0
+ mtctr 5
+ mul 9,0,6
+ srai 7,0,31
+ and 7,7,6
+ mfmq 11
+ cax 9,9,7
+ l 7,4(3)
+ sf 8,11,7 # add res_limb
+ a 11,8,11 # invert cy (r11 is junk)
+ blt Lneg
+Lpos: bdz Lend
+
+Lploop: lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 10,0,6
+ mfmq 0
+ ae 11,0,9 # low limb + old_cy_limb + old cy
+ l 7,4(3)
+ aze 10,10 # propagate cy to new cy_limb
+ sf 8,11,7 # add res_limb
+ a 11,8,11 # invert cy (r11 is junk)
+ bge Lp0
+ cax 10,10,6 # adjust high limb for negative limb from s1
+Lp0: bdz Lend0
+ lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 9,0,6
+ mfmq 0
+ ae 11,0,10
+ l 7,4(3)
+ aze 9,9
+ sf 8,11,7
+ a 11,8,11 # invert cy (r11 is junk)
+ bge Lp1
+ cax 9,9,6 # adjust high limb for negative limb from s1
+Lp1: bdn Lploop
+
+ b Lend
+
+Lneg: cax 9,9,0
+ bdz Lend
+Lnloop: lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 10,0,6
+ mfmq 7
+ ae 11,7,9
+ l 7,4(3)
+ ae 10,10,0 # propagate cy to new cy_limb
+ sf 8,11,7 # add res_limb
+ a 11,8,11 # invert cy (r11 is junk)
+ bge Ln0
+ cax 10,10,6 # adjust high limb for negative limb from s1
+Ln0: bdz Lend0
+ lu 0,4(4)
+ stu 8,4(3)
+ cmpi 0,0,0
+ mul 9,0,6
+ mfmq 7
+ ae 11,7,10
+ l 7,4(3)
+ ae 9,9,0 # propagate cy to new cy_limb
+ sf 8,11,7 # add res_limb
+ a 11,8,11 # invert cy (r11 is junk)
+ bge Ln1
+ cax 9,9,6 # adjust high limb for negative limb from s1
+Ln1: bdn Lnloop
+ b Lend
+
+Lend0: cal 9,0(10)
+Lend: st 8,4(3)
+ aze 3,9
+ br
--- /dev/null
+# VAX __mpn_add_n -- Add two limb vectors of the same length > 0 and store
+# sum in a third limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr (sp + 4)
+# s1_ptr (sp + 8)
+# s2_ptr (sp + 12)
+# size (sp + 16)
+
+.text
+ .align 1
+.globl ___mpn_add_n
+___mpn_add_n:
+ .word 0x0
+ movl 16(ap),r0
+ movl 12(ap),r1
+ movl 8(ap),r2
+ movl 4(ap),r3
+ subl2 r4,r4
+
+Loop:
+ movl (r2)+,r4
+ adwc (r1)+,r4
+ movl r4,(r3)+
+ jsobgtr r0,Loop
+
+ adwc r0,r0
+ ret
--- /dev/null
+# VAX __mpn_addmul_1 -- Multiply a limb vector with a limb and add
+# the result to a second limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr (sp + 4)
+# s1_ptr (sp + 8)
+# size (sp + 12)
+# s2_limb (sp + 16)
+
+.text
+ .align 1
+.globl ___mpn_addmul_1
+___mpn_addmul_1:
+ .word 0xfc0
+ movl 12(ap),r4
+ movl 8(ap),r8
+ movl 4(ap),r9
+ movl 16(ap),r6
+ jlss s2_big
+
+ clrl r3
+ incl r4
+ ashl $-1,r4,r7
+ jlbc r4,L1
+ clrl r11
+
+# Loop for S2_LIMB < 0x80000000
+Loop1: movl (r8)+,r1
+ jlss L1n0
+ emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc $0,r3
+ addl2 r2,(r9)+
+ adwc $0,r3
+L1: movl (r8)+,r1
+ jlss L1n1
+L1p1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc $0,r11
+ addl2 r10,(r9)+
+ adwc $0,r11
+
+ jsobgtr r7,Loop1
+ movl r11,r0
+ ret
+
+L1n0: emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r6,r3
+ addl2 r2,(r9)+
+ adwc $0,r3
+ movl (r8)+,r1
+ jgeq L1p1
+L1n1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r6,r11
+ addl2 r10,(r9)+
+ adwc $0,r11
+
+ jsobgtr r7,Loop1
+ movl r11,r0
+ ret
+
+
+s2_big: clrl r3
+ incl r4
+ ashl $-1,r4,r7
+ jlbc r4,L2
+ clrl r11
+
+# Loop for S2_LIMB >= 0x80000000
+Loop2: movl (r8)+,r1
+ jlss L2n0
+ emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r1,r3
+ addl2 r2,(r9)+
+ adwc $0,r3
+L2: movl (r8)+,r1
+ jlss L2n1
+L2p1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r1,r11
+ addl2 r10,(r9)+
+ adwc $0,r11
+
+ jsobgtr r7,Loop2
+ movl r11,r0
+ ret
+
+L2n0: emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r6,r3
+ addl2 r2,(r9)+
+ adwc r1,r3
+ movl (r8)+,r1
+ jgeq L2p1
+L2n1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r6,r11
+ addl2 r10,(r9)+
+ adwc r1,r11
+
+ jsobgtr r7,Loop2
+ movl r11,r0
+ ret
--- /dev/null
+# VAX __mpn_mul_1 -- Multiply a limb vector with a limb and store
+# the result in a second limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr (sp + 4)
+# s1_ptr (sp + 8)
+# size (sp + 12)
+# s2_limb (sp + 16)
+
+.text
+ .align 1
+.globl ___mpn_mul_1
+___mpn_mul_1:
+ .word 0xfc0
+ movl 12(ap),r4
+ movl 8(ap),r8
+ movl 4(ap),r9
+ movl 16(ap),r6
+ jlss s2_big
+
+# One might want to combine the addl2 and the store below, but that
+# is actually just slower according to my timing tests. (VAX 3600)
+
+ clrl r3
+ incl r4
+ ashl $-1,r4,r7
+ jlbc r4,L1
+ clrl r11
+
+# Loop for S2_LIMB < 0x80000000
+Loop1: movl (r8)+,r1
+ jlss L1n0
+ emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc $0,r3
+ movl r2,(r9)+
+L1: movl (r8)+,r1
+ jlss L1n1
+L1p1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc $0,r11
+ movl r10,(r9)+
+
+ jsobgtr r7,Loop1
+ movl r11,r0
+ ret
+
+L1n0: emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r6,r3
+ movl r2,(r9)+
+ movl (r8)+,r1
+ jgeq L1p1
+L1n1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r6,r11
+ movl r10,(r9)+
+
+ jsobgtr r7,Loop1
+ movl r11,r0
+ ret
+
+
+s2_big: clrl r3
+ incl r4
+ ashl $-1,r4,r7
+ jlbc r4,L2
+ clrl r11
+
+# Loop for S2_LIMB >= 0x80000000
+Loop2: movl (r8)+,r1
+ jlss L2n0
+ emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r1,r3
+ movl r2,(r9)+
+L2: movl (r8)+,r1
+ jlss L2n1
+L2p1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r1,r11
+ movl r10,(r9)+
+
+ jsobgtr r7,Loop2
+ movl r11,r0
+ ret
+
+L2n0: emul r1,r6,$0,r2
+ addl2 r1,r3
+ addl2 r11,r2
+ adwc r6,r3
+ movl r2,(r9)+
+ movl (r8)+,r1
+ jgeq L2p1
+L2n1: emul r1,r6,$0,r10
+ addl2 r1,r11
+ addl2 r3,r10
+ adwc r6,r11
+ movl r10,(r9)+
+
+ jsobgtr r7,Loop2
+ movl r11,r0
+ ret
--- /dev/null
+# VAX __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and store
+# difference in a third limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr (sp + 4)
+# s1_ptr (sp + 8)
+# s2_ptr (sp + 12)
+# size (sp + 16)
+
+.text
+ .align 1
+.globl ___mpn_sub_n
+___mpn_sub_n:
+ .word 0x0
+ movl 16(ap),r0
+ movl 12(ap),r1
+ movl 8(ap),r2
+ movl 4(ap),r3
+ subl2 r4,r4
+
+Loop:
+ movl (r2)+,r4
+ sbwc (r1)+,r4
+ movl r4,(r3)+
+ jsobgtr r0,Loop
+
+ adwc r0,r0
+ ret
--- /dev/null
+# VAX __mpn_submul_1 -- Multiply a limb vector with a limb and subtract
+# the result from a second limb vector.
+
+# Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+
+# This file is part of the GNU MP Library.
+
+# The GNU MP Library is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Library General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or (at your
+# option) any later version.
+
+# The GNU MP Library is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+# License for more details.
+
+# You should have received a copy of the GNU Library General Public License
+# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+# INPUT PARAMETERS
+# res_ptr (sp + 4)
+# s1_ptr (sp + 8)
+# size (sp + 12)
+# s2_limb (sp + 16)
+
+.text
+ .align 1
+.globl ___mpn_submul_1
+___mpn_submul_1:
+ .word 0xfc0
+ movl 12(ap),r4
+ movl 8(ap),r8
+ movl 4(ap),r9
+ movl 16(ap),r6
+ jlss s2_big
+
+ clrl r3
+ incl r4
+ ashl $-1,r4,r7
+ jlbc r4,L1
+ clrl r11
+
+# Loop for S2_LIMB < 0x80000000
+Loop1: movl (r8)+,r1
+ jlss L1n0
+ emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc $0,r3
+ subl2 r2,(r9)+
+ adwc $0,r3
+L1: movl (r8)+,r1
+ jlss L1n1
+L1p1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc $0,r11
+ subl2 r10,(r9)+
+ adwc $0,r11
+
+ jsobgtr r7,Loop1
+ movl r11,r0
+ ret
+
+L1n0: emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r6,r3
+ subl2 r2,(r9)+
+ adwc $0,r3
+ movl (r8)+,r1
+ jgeq L1p1
+L1n1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r6,r11
+ subl2 r10,(r9)+
+ adwc $0,r11
+
+ jsobgtr r7,Loop1
+ movl r11,r0
+ ret
+
+
+s2_big: clrl r3
+ incl r4
+ ashl $-1,r4,r7
+ jlbc r4,L2
+ clrl r11
+
+# Loop for S2_LIMB >= 0x80000000
+Loop2: movl (r8)+,r1
+ jlss L2n0
+ emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r1,r3
+ subl2 r2,(r9)+
+ adwc $0,r3
+L2: movl (r8)+,r1
+ jlss L2n1
+L2p1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r1,r11
+ subl2 r10,(r9)+
+ adwc $0,r11
+
+ jsobgtr r7,Loop2
+ movl r11,r0
+ ret
+
+L2n0: emul r1,r6,$0,r2
+ addl2 r11,r2
+ adwc r6,r3
+ subl2 r2,(r9)+
+ adwc r1,r3
+ movl (r8)+,r1
+ jgeq L2p1
+L2n1: emul r1,r6,$0,r10
+ addl2 r3,r10
+ adwc r6,r11
+ subl2 r10,(r9)+
+ adwc r1,r11
+
+ jsobgtr r7,Loop2
+ movl r11,r0
+ ret
--- /dev/null
+! Z8000 __mpn_add_n -- Add two limb vectors of equal, non-zero length.
+
+! Copyright (C) 1993, 1994 Free Software Foundation, Inc.
+
+! This file is part of the GNU MP Library.
+
+! The GNU MP Library is free software; you can redistribute it and/or modify
+! it under the terms of the GNU Library General Public License as published by
+! the Free Software Foundation; either version 2 of the License, or (at your
+! option) any later version.
+
+! The GNU MP Library is distributed in the hope that it will be useful, but
+! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+! License for more details.
+
+! You should have received a copy of the GNU Library General Public License
+! along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+! INPUT PARAMETERS
+! res_ptr r7
+! s1_ptr r6
+! s2_ptr r5
+! size r4
+
+! If we are really crazy, we can use push to write a few result words
+! backwards, using push just because it is faster than reg+disp. We'd
+! then add 2x the number of words written to r7...
+
+ unseg
+ .text
+ even
+ global ___mpn_add_n
+___mpn_add_n:
+ pop r0,@r6
+ pop r1,@r5
+ add r0,r1
+ ld @r7,r0
+ dec r4
+ jr eq,Lend
+Loop: pop r0,@r6
+ pop r1,@r5
+ adc r0,r1
+ inc r7,#2
+ ld @r7,r0
+ dec r4
+ jr ne,Loop
+Lend: ld r2,r4 ! use 0 already in r4
+ adc r2,r2
+ ret t
--- /dev/null
+! Z8000 __mpn_mul_1 -- Multiply a limb vector with a limb and store
+! the result in a second limb vector.
+
+! Copyright (C) 1993, 1994 Free Software Foundation, Inc.
+
+! This file is part of the GNU MP Library.
+
+! The GNU MP Library is free software; you can redistribute it and/or modify
+! it under the terms of the GNU Library General Public License as published by
+! the Free Software Foundation; either version 2 of the License, or (at your
+! option) any later version.
+
+! The GNU MP Library is distributed in the hope that it will be useful, but
+! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+! License for more details.
+
+! You should have received a copy of the GNU Library General Public License
+! along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+! INPUT PARAMETERS
+! res_ptr r7
+! s1_ptr r6
+! size r5
+! s2_limb r4
+
+ unseg
+ .text
+ even
+ global ___mpn_mul_1
+___mpn_mul_1:
+ sub r2,r2 ! zero carry limb
+ and r4,r4
+ jr mi,Lneg
+
+Lpos: pop r1,@r6
+ ld r9,r1
+ mult rr8,r4
+ and r1,r1 ! shift msb of loaded limb into cy
+ jr mi,Lp ! branch if loaded limb's msb is set
+ add r8,r4 ! hi_limb += sign_comp2
+Lp: add r9,r2 ! lo_limb += cy_limb
+ xor r2,r2
+ adc r2,r8
+ ld @r7,r9
+ inc r7,#2
+ dec r5
+ jr ne,Lpos
+ ret t
+
+Lneg: pop r1,@r6
+ ld r9,r1
+ mult rr8,r4
+ add r8,r1 ! hi_limb += sign_comp1
+ and r1,r1
+ jr mi,Ln
+ add r8,r4 ! hi_limb += sign_comp2
+Ln: add r9,r2 ! lo_limb += cy_limb
+ xor r2,r2
+ adc r2,r8
+ ld @r7,r9
+ inc r7,#2
+ dec r5
+ jr ne,Lneg
+ ret t
--- /dev/null
+! Z8000 __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+! store difference in a third limb vector.
+
+! Copyright (C) 1993, 1994 Free Software Foundation, Inc.
+
+! This file is part of the GNU MP Library.
+
+! The GNU MP Library is free software; you can redistribute it and/or modify
+! it under the terms of the GNU Library General Public License as published by
+! the Free Software Foundation; either version 2 of the License, or (at your
+! option) any later version.
+
+! The GNU MP Library is distributed in the hope that it will be useful, but
+! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
+! License for more details.
+
+! You should have received a copy of the GNU Library General Public License
+! along with the GNU MP Library; see the file COPYING.LIB. If not, write to
+! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+
+! INPUT PARAMETERS
+! res_ptr r7
+! s1_ptr r6
+! s2_ptr r5
+! size r4
+
+! If we are really crazy, we can use push to write a few result words
+! backwards, using push just because it is faster than reg+disp. We'd
+! then add 2x the number of words written to r7...
+
+ unseg
+ .text
+ even
+ global ___mpn_sub_n
+___mpn_sub_n:
+ pop r0,@r6
+ pop r1,@r5
+ sub r0,r1
+ ld @r7,r0
+ dec r4
+ jr eq,Lend
+Loop: pop r0,@r6
+ pop r1,@r5
+ sbc r0,r1
+ inc r7,#2
+ ld @r7,r0
+ dec r4
+ jr ne,Loop
+Lend: ld r2,r4 ! use 0 already in r4
+ adc r2,r2
+ ret t