mpir/mpn/x86w/p6/mod_1.asm

310 lines
8.3 KiB
NASM
Raw Normal View History

; Copyright 1999, 2000, 2002 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 Lesser General Public License as
; published by the Free Software Foundation; either version 2.1 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
; Lesser General Public License for more details.
;
; You should have received a copy of the GNU Lesser General Public
; License along with the GNU MP Library; see the file COPYING.LIB. If
; not, write to the Free Software Foundation, Inc., 59 Temple Place -
; Suite 330, Boston, MA 02111-1307, USA.
;
; Translation of AT&T syntax code by Brian Gladman
%include "..\x86i.inc"
global ___gmpn_preinv_mod_1
global ___gmpn_mod_1c
global ___gmpn_mod_1
%ifdef DLL
export ___gmpn_mod_1c
export ___gmpn_mod_1
%endif
%define MUL_NORM_THRESHOLD 4
%define MUL_UNNORM_THRESHOLD 5
%define MUL_NORM_DELTA MUL_NORM_THRESHOLD - MUL_UNNORM_THRESHOLD
%define PARAM_INVERSE esp+frame+16
%define PARAM_CARRY esp+frame+16
%define PARAM_DIVISOR esp+frame+12
%define PARAM_SIZE esp+frame+8
%define PARAM_SRC esp+frame+4
%define SAVE_EBX esp+frame-4
%define SAVE_ESI esp+frame-8
%define SAVE_EDI esp+frame-12
%define SAVE_EBP esp+frame-16
%define VAR_NORM esp+frame-20
%define VAR_INVERSE esp+frame-24
%define STACK_SPACE 24
section .text
align 16
%define frame 0
___gmpn_preinv_mod_1:
mov edx,[PARAM_SRC]
sub esp,STACK_SPACE
FR_sesp STACK_SPACE
mov [SAVE_EBX],ebx
mov ebx,[PARAM_SIZE]
mov [SAVE_EBP],ebp
mov ebp,[PARAM_DIVISOR]
mov [SAVE_ESI],esi
mov eax,[PARAM_INVERSE]
mov [SAVE_EDI],edi
mov edi,[-4+edx+ebx*4] ; src high limb
mov [VAR_NORM],dword 0
lea ecx,[-8+edx+ebx*4] ; &src[size-2]
mov esi,edi
sub edi,ebp ; high-divisor
cmovc edi,esi
dec ebx
jnz Lpreinv_entry
jmp Ldone_edi
align 16
%define frame 0
___gmpn_mod_1c:
mov ecx,[PARAM_SIZE]
sub esp,STACK_SPACE
FR_sesp STACK_SPACE
mov [SAVE_EBP],ebp
mov eax,[PARAM_DIVISOR]
mov [SAVE_ESI],esi
mov edx,[PARAM_CARRY]
mov esi,[PARAM_SRC]
or ecx,ecx
jz Ldone_edx ; result==carry if size==0
sar eax,31
mov ebp,[PARAM_DIVISOR]
and eax,MUL_NORM_DELTA
add eax,MUL_UNNORM_THRESHOLD
cmp ecx,eax
jb Ldivide_top
; The carry parameter pretends to be the src high limb.
mov [SAVE_EBX],ebx
lea ebx,[1+ecx] ; size+1
mov eax,edx ; carry
jmp Lmul_by_inverse_1c
align 16
%define frame 0
___gmpn_mod_1:
mov ecx,[PARAM_SIZE]
sub esp,STACK_SPACE
FR_sesp STACK_SPACE
mov edx,0 ; initial carry (if can't skip a div)
mov [SAVE_ESI],esi
mov eax,[PARAM_SRC]
mov [SAVE_EBP],ebp
mov ebp,[PARAM_DIVISOR]
mov esi,[PARAM_DIVISOR]
or ecx,ecx
jz Ldone_edx
mov eax,[-4+eax+ecx*4] ; src high limb
sar ebp,31
and ebp,MUL_NORM_DELTA
add ebp,MUL_UNNORM_THRESHOLD
cmp eax,esi ; carry flag if high<divisor
cmovc edx,eax
mov esi,[PARAM_SRC]
sbb ecx,0 ; size-1 to skip one div
jz Ldone_eax ; done if had size==1
cmp ecx,ebp
mov ebp,[PARAM_DIVISOR]
jae Lmul_by_inverse
; eax scratch (quotient)
; ebx
; ecx counter,limbs,decrementing
; edx scratch (remainder)
; esi src
; edi
; ebp divisor
Ldivide_top:
mov eax,[-4+esi+ecx*4]
div ebp
dec ecx
jnz Ldivide_top
Ldone_edx:
mov eax,edx
Ldone_eax:
mov esi,[SAVE_ESI]
mov ebp,[SAVE_EBP]
add esp,STACK_SPACE
ret
; eax src high limb
; ebx
; ecx
; edx
; esi src
; edi
; ebp divisor
Lmul_by_inverse:
mov [SAVE_EBX],ebx
mov ebx,[PARAM_SIZE]
Lmul_by_inverse_1c:
bsr ecx,ebp ; 31-l
mov [SAVE_EDI],edi
xor ecx,31 ; l
mov [VAR_NORM],ecx
shl ebp,cl ; d normalized
mov edi,eax ; src high -> n2
sub eax,ebp
cmovnc edi,eax
mov eax,-1
mov edx,-1
sub edx,ebp ; (b-d)-1 so edx:eax = b*(b-d)-1
lea ecx,[-8+esi+ebx*4] ; &src[size-2]
div ebp ; floor (b*(b-d)-1) / d
Lpreinv_entry:
mov [VAR_INVERSE],eax
; No special scheduling of loads is necessary in this loop,out of order
; execution hides the latencies already.
;
; The way q1+1 is generated in %ebx and d is moved to %eax for the multiply
; seems fastest. The obvious change to generate q1+1 in %eax and then just
; multiply by %ebp (as per mpn/x86/pentium/mod_1.asm in fact) runs 1 cycle
; slower,for no obvious reason.
; eax n10 (then scratch)
; ebx scratch (nadj,q1)
; ecx src pointer,decrementing
; edx scratch
; esi n10
; edi n2
; ebp divisor
align 16
Linverse_top:
mov eax,[ecx] ; next src limb
mov esi,eax
sar eax,31 ; -n1
mov ebx,ebp
and ebx,eax ; -n1 & d
neg eax ; n1
add eax,edi ; n2+n1
mul dword [VAR_INVERSE] ; m*(n2+n1)
add ebx,esi ; nadj = n10 + (-n1 & d),ignoring overflow
sub ecx,4
add eax,ebx ; m*(n2+n1) + nadj,low giving carry flag
lea ebx,[1+edi] ; n2+1
mov eax,ebp ; d
adc ebx,edx ; 1 + high(n2<<32 + m*(n2+n1) + nadj) = q1+1
jz Lq1_ff
mul ebx ; (q1+1)*d
sub esi,eax ; low n - (q1+1)*d
sbb edi,edx ; high n - (q1+1)*d,0 or -1
and edi,ebp ; d if underflow
add edi,esi ; remainder with addback if necessary
cmp ecx,[PARAM_SRC]
jae Linverse_top
; %edi is the remainder modulo d*2^n and now must be reduced to
; 0<=r<d by calculating r*2^n mod d*2^n and then right shifting by
; n. If d was already normalized on entry so that n==0 then nothing
; is needed here. The chance of n==0 is low,but it's true of say
; PP from gmp-impl.h.
;
; eax
; ebx
; ecx
; edx
; esi
; edi remainder
; ebp divisor (normalized)
Linverse_loop_done:
mov ecx,[VAR_NORM]
mov esi,0
or ecx,ecx
jz Ldone_edi
; Here use %edi=n10 and %esi=n2,opposite to the loop above.
;
; The q1=0xFFFFFFFF case is handled with an sbbl to adjust q1+1
; back,rather than q1_ff special case code. This is simpler and
; costs only 2 uops.
shld esi,edi,cl
shl edi,cl
mov eax,edi ; n10
mov ebx,ebp ; d
sar eax,31 ; -n1
and ebx,eax ; -n1 & d
neg eax ; n1
add ebx,edi ; nadj = n10 + (-n1 & d),ignoring overflow
add eax,esi ; n2+n1
mul dword [VAR_INVERSE] ; m*(n2+n1)
add eax,ebx ; m*(n2+n1) + nadj,low giving carry flag
lea ebx,[1+esi] ; n2+1
adc ebx,edx ; 1 + high(n2<<32 + m*(n2+n1) + nadj) = q1+1
sbb ebx,0
mov eax,ebp ; d
mul ebx ; (q1+1)*d
mov ebx,[SAVE_EBX]
sub edi,eax ; low n - (q1+1)*d is remainder
sbb esi,edx ; high n - (q1+1)*d,0 or -1
and esi,ebp
mov ebp,[SAVE_EBP]
lea eax,[esi+edi] ; remainder
mov esi,[SAVE_ESI]
shr eax,cl ; denorm remainder
mov edi,[SAVE_EDI]
add esp,STACK_SPACE
ret
Ldone_edi:
mov ebx,[SAVE_EBX]
mov eax,edi
mov esi,[SAVE_ESI]
mov edi,[SAVE_EDI]
mov ebp,[SAVE_EBP]
add esp,STACK_SPACE
ret
; Special case for q1=0xFFFFFFFF,giving q=0xFFFFFFFF meaning the low dword
; of q*d is simply -d and the remainder n-q*d = n10+d.
;
; This is reached only very rarely.
;
; eax (divisor)
; ebx (q1+1 == 0)
; ecx src pointer
; edx
; esi n10
; edi (n2)
; ebp divisor
Lq1_ff:
lea edi,[ebp+esi] ; n-q*d remainder -> next n2
cmp ecx,[PARAM_SRC]
jae Linverse_top
jmp Linverse_loop_done
end