mpir/mpn/x86_64/nehalem/redc_basecase.as

420 lines
7.8 KiB
ActionScript

; core2 mpn_redc_basecase
; Copyright 2009 Jason Moxham
; This file is part of the MPIR Library.
; The MPIR 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 MPIR 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 MPIR Library; see the file COPYING.LIB. If not, write
; to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
; Boston, MA 02110-1301, USA.
%include 'yasm_mac.inc'
; (rdi, rcx) = (rsi, rcx) + (rdx, rcx) with the carry flag set for the carry
; this is the usual mpn_add_n with the final dec rax;adc rax,rax;ret removed
; and a jump where we have two rets
%macro mpn_add 0
mov rax, rcx
and rax, 3
shr rcx, 2
cmp rcx, 0
; carry flag is clear here
jnz %%1
mov r11, [rsi]
add r11, [rdx]
mov [rdi], r11
dec rax
jz %%2
mov r11, [rsi+8]
adc r11, [rdx+8]
mov [rdi+8], r11
dec rax
jz %%2
mov r11, [rsi+16]
adc r11, [rdx+16]
mov [rdi+16], r11
jmp %%2
align 16
%%1:
mov r11, [rsi]
mov r8, [rsi+8]
lea rsi, [rsi+32]
adc r11, [rdx]
adc r8, [rdx+8]
lea rdx, [rdx+32]
mov [rdi], r11
mov [rdi+8], r8
lea rdi, [rdi+32]
mov r9, [rsi-16]
mov r10, [rsi-8]
adc r9, [rdx-16]
adc r10, [rdx-8]
mov [rdi-16], r9
dec rcx
mov [rdi-8], r10
jnz %%1
inc rax
dec rax
jz %%2
mov r11, [rsi]
adc r11, [rdx]
mov [rdi], r11
dec rax
jz %%2
mov r11, [rsi+8]
adc r11, [rdx+8]
mov [rdi+8], r11
dec rax
jz %%2
mov r11, [rsi+16]
adc r11, [rdx+16]
mov [rdi+16], r11
%%2:
%endmacro
; (rbx, rbp) = (rsi, rbp) - (rdx, rbp)
%macro mpn_sub 0
mov rax, rbp
and rax, 3
shr rbp, 2
cmp rbp, 0
; carry flag is clear here
jnz %%1
mov r11, [rsi]
sub r11, [rdx]
mov [rbx], r11
dec rax
jz %%2
mov r11, [rsi+8]
sbb r11, [rdx+8]
mov [rbx+8], r11
dec rax
jz %%2
mov r11, [rsi+16]
sbb r11, [rdx+16]
mov [rbx+16], r11
jmp %%2
align 16
%%1:
mov r11, [rsi]
mov r8, [rsi+8]
lea rsi, [rsi+32]
sbb r11, [rdx]
sbb r8, [rdx+8]
lea rdx, [rdx+32]
mov [rbx], r11
mov [rbx+8], r8
lea rbx, [rbx+32]
mov r9, [rsi-16]
mov r10, [rsi-8]
sbb r9, [rdx-16]
sbb r10, [rdx-8]
mov [rbx-16], r9
dec rbp
mov [rbx-8], r10
jnz %%1
inc rax
dec rax
jz %%2
mov r11, [rsi]
sbb r11, [rdx]
mov [rbx], r11
dec rax
jz %%2
mov r11, [rsi+8]
sbb r11, [rdx+8]
mov [rbx+8], r11
dec rax
jz %%2
mov r11, [rsi+16]
sbb r11, [rdx+16]
mov [rbx+16], r11
%%2:
%endmacro
; changes from standard addmul
; change r8 to r12 and rcx to r13 and rdi to r8
; reemove ret and write last limb but to beginning
%macro addmulloop 1
align 16
%%1:
mov r10, 0
mul r13
add [r8+r11*8], r12
adc r9, rax
db 0x26
adc r10, rdx
mov rax, [rsi+r11*8+16]
mul r13
add [r8+r11*8+8], r9
adc r10, rax
mov ebx, 0
adc rbx, rdx
mov rax, [rsi+r11*8+24]
mov r12, 0
mov r9, 0
mul r13
add [r8+r11*8+16], r10
db 0x26
adc rbx, rax
db 0x26
adc r12, rdx
mov rax, [rsi+r11*8+32]
mul r13
add [r8+r11*8+24], rbx
db 0x26
adc r12, rax
db 0x26
adc r9, rdx
add r11, 4
mov rax, [rsi+r11*8+8]
jnc %%1
%endmacro
%macro addmulpropro0 0
imul r13, rcx
lea r8, [r8-8]
%endmacro
%macro addmulpro0 0
mov r11, r14
lea r8, [r8+8]
mov rax, [rsi+r14*8]
mul r13
mov r12, rax
mov rax, [rsi+r14*8+8]
mov r9, rdx
cmp r14, 0
%endmacro
%macro addmulnext0 0
mov r10d, 0
mul r13
add [r8+r11*8], r12
adc r9, rax
adc r10, rdx
mov rax, [rsi+r11*8+16]
mul r13
add [r8+r11*8+8], r9
adc r10, rax
mov ebx, 0
adc rbx, rdx
mov rax, [rsi+r11*8+24]
mov r12d, 0
mov r9d, 0
mul r13
add [r8+r11*8+16], r10
adc rbx, rax
adc r12, rdx
mov rax, [rsi+r11*8+32]
mul r13
add [r8+r11*8+24], rbx
mov r13, [r8+r14*8+8]
adc r12, rax
adc r9, rdx
imul r13, rcx
add [r8+r11*8+32], r12
adc r9, 0
sub r15, 1
mov [r8+r14*8], r9
%endmacro
%macro addmulpropro1 0
%endmacro
%macro addmulpro1 0
imul r13, rcx
mov rax, [rsi+r14*8]
mov r11, r14
mul r13
mov r12, rax
mov rax, [rsi+r14*8+8]
mov r9, rdx
cmp r14, 0
%endmacro
%macro addmulnext1 0
mov r10d, 0
mul r13
add [r8+r11*8], r12
adc r9, rax
adc r10, rdx
mov rax, [rsi+r11*8+16]
mul r13
add [r8+r11*8+8], r9
adc r10, rax
mov ebx, 0
adc rbx, rdx
mov rax, [rsi+r11*8+24]
mov r12d, 0
mul r13
add [r8+r11*8+16], r10
adc rbx, rax
adc r12, rdx
add [r8+r11*8+24], rbx
mov r13, [r8+r14*8+8]
adc r12, 0
sub r15, 1
mov [r8+r14*8], r12
lea r8, [r8+8]
%endmacro
%macro addmulpropro2 0
%endmacro
%macro addmulpro2 0
imul r13, rcx
mov rax, [rsi+r14*8]
mov r11, r14
mul r13
mov r12, rax
mov rax, [rsi+r14*8+8]
mov r9, rdx
cmp r14, 0
%endmacro
%macro addmulnext2 0
mul r13
add [r8+r11*8], r12
adc r9, rax
mov r10d, 0
adc r10, rdx
mov rax, [rsi+r11*8+16]
mul r13
add [r8+r11*8+8], r9
adc r10, rax
mov ebx, 0
adc rbx, rdx
mov r13, [r8+r14*8+8]
add [r8+r11*8+16], r10
adc rbx, 0
mov [r8+r14*8], rbx
sub r15, 1
lea r8, [r8+8]
%endmacro
%macro addmulpropro3 0
%endmacro
%macro addmulpro3 0
imul r13, rcx
mov rax, [rsi+r14*8]
mov r11, r14
mul r13
mov r12, rax
mov rax, [rsi+r14*8+8]
mov r9, rdx
cmp r14, 0
%endmacro
%macro addmulnext3 0
mul r13
add [r8+r11*8], r12
adc r9, rax
mov r10d, 0
adc r10, rdx
add [r8+r11*8+8], r9
adc r10, 0
mov r13, [r8+r14*8+8]
mov [r8+r14*8], r10
lea r8, [r8+8]
sub r15, 1
%endmacro
; change r8 to r12
; write top limb ax straight to mem dont return (NOTE we WRITE NOT ADD)
%macro mpn_addmul_1_int 1
addmulpropro%1
align 16
%%1:
addmulpro%1
jge %%2
addmulloop %1
%%2:
addmulnext%1
jnz %%1
jmp end
%endmacro
BITS 64
GLOBAL_FUNC mpn_redc_basecase
cmp rdx, 1
je one
push r13
push r14
push rbx
push r12
push r15
push rbp
mov r14, 5
sub r14, rdx
; store copys
push rsi
push r8
lea r8, [r8+rdx*8-40]
lea rsi, [rsi+rdx*8-40]
mov rbp, rdx
mov r15, rdx
mov rax, r14
and rax, 3
mov r13, [r8+r14*8]
je case0
jp case3
cmp rax, 1
je case1
case2:
mpn_addmul_1_int 2
align 16
case0:
mpn_addmul_1_int 0
align 16
case1:
mpn_addmul_1_int 1
align 16
case3:
mpn_addmul_1_int 3
align 16
end:
mov rcx, rbp
pop rdx
lea rsi, [rdx+rbp*8]
mov rbx, rdi
mpn_add
; mpnadd(rdi,rsi,rdx,rcx)
pop rdx
jnc skip
mov rsi, rbx
mpn_sub
; mpn_sub_n(rbx,rsi,rdx,rbp) we can certainly improve this sub
skip:
pop rbp
pop r15
pop r12
pop rbx
pop r14
pop r13
ret
align 16
one:
mov r9, [r8]
mov r11, [rsi]
imul rcx, r9
mov rax, rcx
mul r11
add rax, r9
; rax is zero here
adc rdx, [r8+8]
cmovnc r11, rax
sub rdx, r11
mov [rdi], rdx
ret