dnl PowerPC-32 mpn_modexact_1_odd -- mpn by limb exact remainder.
dnl Copyright 2002, 2003, 2005, 2006 Free Software Foundation, Inc.
dnl This file is part of the GNU MP Library.
dnl
dnl The GNU MP Library is free software; you can redistribute it and/or modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
C cycles/limb
C 603e: ?
C 604e: 6.0
C 75x (G3): 6.0-13.0, depending on divisor
C 7400,7410 (G4): 6.0-13.0, depending on divisor
C 744x,745x (G4+): 8.0-10.0, depending on divisor
C power4/ppc970: 12.0
C power5: 12.0
C mp_limb_t mpn_modexact_1_odd (mp_srcptr src, mp_size_t size,
C mp_limb_t divisor);
C mp_limb_t mpn_modexact_1c_odd (mp_srcptr src, mp_size_t size,
C mp_limb_t divisor, mp_limb_t carry);
C
C For PIC, the inverse is established arithmetically since it measures about
C 5 cycles faster than the nonsense needed to access binvert_limb_table in
C SVR4 or Darwin style PIC. AIX might be better, since it avoids bl/mflr to
C get at the GOT/TOC/whatever.
C
C Using divwu for size==1 measured about 10 cycles slower on 604e, or about
C 3-5 cycles faster on 750. For now it doesn't seem worth bothering with.
C
C The loop allows an early-out on mullw for the inverse, and on mulhwu for
C the divisor. So the fastest is for instance divisor==1 (inverse==-1), and
C the slowest is anything giving a full 32-bits in both, such as
C divisor==0xDEADBEEF (inverse==0x904B300F). These establish the stated
C range above for 750 and 7400.
ASM_START()
EXTERN(binvert_limb_table)
PROLOGUE(mpn_modexact_1_odd)
li r6, 0
PROLOGUE(mpn_modexact_1c_odd)
mtctr r4 C size
ifdef(`PIC_SLOW',`
C Load from our table with PIC is so slow on Linux and Darwin that we avoid it
rlwinm r7, r5, 1,28,28 C (divisor << 1) & 8
rlwinm r8, r5, 2,28,28 C (divisor << 2) & 8
xor r7, r7, r8 C ((divisor << 1) ^ (divisor << 2)) & 8
rlwinm r4, r5, 0,28,31 C divisor low 4 bits, speedup mullw
xor r4, r4, r7 C inverse, 4 bits
mullw r7, r4, r4 C i*i
slwi r4, r4, 1 C 2*i
rlwinm r8, r5, 0,24,31 C divisor low 8 bits, speedup mullw
mullw r7, r7, r8 C i*i*d
sub r4, r4, r7 C inverse, 8 bits
',`
LEA( r7, binvert_limb_table)
rlwinm r4, r5, 31,25,31 C (divisor/2) & 0x7F
lbzx r4, r4,r7 C inverse, 8 bits
')
mullw r7, r4, r4 C i*i
slwi r4, r4, 1 C 2*i
mullw r7, r5, r7 C i*i*d [i*i is 16 bits, so second operand]
sub r4, r4, r7 C inverse, 16 bits
mullw r7, r4, r4 C i*i
slwi r4, r4, 1 C 2*i
mullw r7, r7, r5 C i*i*d
lwz r0, 0(r3) C src[0]
sub r4, r4, r7 C inverse, 32 bits
subfc r7, r6, r0 C l = src[0] - carry
mullw r7, r7, r4 C q = l * inverse
bdz L(one)
lwzu r0, 4(r3) C src[1]
mulhwu r6, r7, r5 C carry = high(q*divisor)
subfe r7, r6, r0 C l = src[1] - carry
bdz L(two)
L(top):
mullw r7, r7, r4 C q = l * inverse
lwzu r0, 4(r3) C src[i]
mulhwu r6, r7, r5 C carry = high(q*divisor)
subfe r7, r6, r0 C l = src[i] - carry
bdnz L(top)
L(two): mullw r7, r7, r4 C q = l * inverse
L(one): subfe r3, r3, r3 C ca 0 or -1
mulhwu r6, r7, r5 C carry = high(q*divisor)
subf r3, r3, r6 C carry + ca
blr
EPILOGUE(mpn_modexact_1c_odd)
EPILOGUE(mpn_modexact_1_odd)
ASM_END()