mpir/mpn/x86_64w/sandybridge/karasub.asm
2012-07-08 13:56:43 +00:00

346 lines
10 KiB
NASM

; PROLOGUE(karasub)
; mpn_karasub
;
; Copyright 2011 The Code Cavern
;
; Copyright 2012 Brian Gladman
;
; 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.
;
; void mpn_karasub(mp_ptr, mp_ptr, mp_size_t)
; rax rdi rsi rdx
; rax rcx rdx r8
;
; Karasuba Multiplication - split x and y into two equal length halves so
; that x = xh.B + xl and y = yh.B + yl. Then their product is:
;
; x.y = xh.yh.B^2 + (xh.yl + xl.yh).B + xl.yl
; = xh.yh.B^2 + (xh.yh + xl.yl - {xh - xl}.{yh - yl}).B + xl.yl
;
; If the length of the elements is m (about n / 2), the output length is 4 * m
; as illustrated below. The middle two blocks involve three additions and one
; subtraction:
;
; -------------------- rp
; | |-->
; | A:xl.yl[lo] | |
; | | | (xh - xl).(yh - yl)
; -------------------- | -------------------- tp
; <-- | |<--< <-- | |
; | | B:xl.yl[hi] | | | E:[lo] |
; | | | | | |
; | -------------------- | --------------------
; >--> | |--> <-- | |
; |\___ | C:xh.yh[lo] | ____/ | F:[hi] |
; | | | | |
; | -------------------- --------------------
; <-- | |
; | D:xh.yh[hi] |
; | |
; --------------------
;
; To avoid overwriting B before it is used, we need to do two operations
; in parallel:
;
; (1) B = B + C + A - E = (B + C) + A - E
; (2) C = C + B + D - F = (B + C) + D - F
;
; The final carry from (1) has to be propagated into C and D, and the final
; carry from (2) has to be propagated into D. When the number of input limbs
; is odd, some extra operations have to be undertaken.
%include "yasm_mac.inc"
%define reg_save_list rbx, rbp, rsi, rdi, r12, r13, r14, r15
%macro add_one 1
inc %1
%endmacro
BITS 64
TEXT
; requires n >= 8
FRAME_PROC mpn_karasub, 2, reg_save_list
mov rdi, rcx
mov rsi, rdx
mov rdx, r8
mov [rsp], rdx
mov [rsp+8], rdi
; rp is rdi, tp is rsi, L is rdi, H is rbp, tp is rsi
; carries/borrows in rax, rbx
shr rdx, 1
lea rcx, [rdx+rdx*1]
lea rbp, [rdi+rcx*8]
xor rax, rax
xor rbx, rbx
lea rdi, [rdi+rdx*8-24]
lea rsi, [rsi+rdx*8-24]
lea rbp, [rbp+rdx*8-24]
mov ecx, 3
sub rcx, rdx
mov edx, 3
align 16
.1: bt rbx, 2
mov r8, [rdi+rdx*8]
adc r8, [rbp+rcx*8]
mov r12, r8
mov r9, [rdi+rdx*8+8]
adc r9, [rbp+rcx*8+8]
mov r10, [rdi+rdx*8+16]
adc r10, [rbp+rcx*8+16]
mov r11, [rdi+rdx*8+24]
adc r11, [rbp+rcx*8+24]
rcl rbx, 1
bt rax, 1
mov r15, r11
adc r8, [rdi+rcx*8]
mov r13, r9
adc r9, [rdi+rcx*8+8]
mov r14, r10
adc r10, [rdi+rcx*8+16]
adc r11, [rdi+rcx*8+24]
rcl rax, 1
bt rbx, 2
adc r12, [rbp+rdx*8]
adc r13, [rbp+rdx*8+8]
adc r14, [rbp+rdx*8+16]
adc r15, [rbp+rdx*8+24]
rcl rbx, 1
bt rax, 1
sbb r8, [rsi+rcx*8]
sbb r9, [rsi+rcx*8+8]
sbb r10, [rsi+rcx*8+16]
sbb r11, [rsi+rcx*8+24]
mov [rdi+rdx*8+16], r10
mov [rdi+rdx*8+24], r11
rcl rax, 1
bt rbx, 2
mov [rdi+rdx*8], r8
mov [rdi+rdx*8+8], r9
sbb r12, [rsi+rdx*8]
sbb r13, [rsi+rdx*8+8]
sbb r14, [rsi+rdx*8+16]
sbb r15, [rsi+rdx*8+24]
rcl rbx, 1
add rdx, 4
mov [rbp+rcx*8], r12
mov [rbp+rcx*8+8], r13
mov [rbp+rcx*8+16], r14
mov [rbp+rcx*8+24], r15
add rcx, 4
jnc .1
cmp rcx, 2
jg .5
jz .4
jp .3
.2: bt rbx, 2
mov r8, [rdi+rdx*8]
adc r8, [rbp]
mov r12, r8
mov r9, [rdi+rdx*8+8]
adc r9, [rbp+8]
mov r10, [rdi+rdx*8+16]
adc r10, [rbp+16]
rcl rbx, 1
bt rax, 1
adc r8, [rdi]
mov r13, r9
adc r9, [rdi+8]
mov r14, r10
adc r10, [rdi+16]
rcl rax, 1
bt rbx, 2
adc r12, [rbp+rdx*8]
adc r13, [rbp+rdx*8+8]
adc r14, [rbp+rdx*8+16]
rcl rbx, 1
bt rax, 1
sbb r8, [rsi]
sbb r9, [rsi+8]
sbb r10, [rsi+16]
mov [rdi+rdx*8+16], r10
rcl rax, 1
bt rbx, 2
mov [rdi+rdx*8], r8
mov [rdi+rdx*8+8], r9
sbb r12, [rsi+rdx*8]
sbb r13, [rsi+rdx*8+8]
sbb r14, [rsi+rdx*8+16]
rcl rbx, 1
add rdx, 3
mov [rbp], r12
mov [rbp+8], r13
mov [rbp+16], r14
jmp .5
.3: bt rbx, 2
mov r8, [rdi+rdx*8]
adc r8, [rbp+8]
mov r12, r8
mov r9, [rdi+rdx*8+8]
adc r9, [rbp+16]
rcl rbx, 1
bt rax, 1
adc r8, [rdi+8]
mov r13, r9
adc r9, [rdi+16]
rcl rax, 1
bt rbx, 2
adc r12, [rbp+rdx*8]
adc r13, [rbp+rdx*8+8]
rcl rbx, 1
bt rax, 1
sbb r8, [rsi+8]
sbb r9, [rsi+16]
rcl rax, 1
bt rbx, 2
mov [rdi+rdx*8], r8
mov [rdi+rdx*8+8], r9
sbb r12, [rsi+rdx*8]
sbb r13, [rsi+rdx*8+8]
rcl rbx, 1
add rdx, 2
mov [rbp+8], r12
mov [rbp+16], r13
jmp .5
.4: bt rbx, 2
mov r8, [rdi+rdx*8]
adc r8, [rbp+16]
mov r12, r8
rcl rbx, 1
bt rax, 1
adc r8, [rdi+16]
rcl rax, 1
bt rbx, 2
adc r12, [rbp+rdx*8]
rcl rbx, 1
bt rax, 1
sbb r8, [rsi+16]
rcl rax, 1
bt rbx, 2
mov [rdi+rdx*8], r8
sbb r12, [rsi+rdx*8]
rcl rbx, 1
add_one rdx
mov [rbp+rcx*8], r12
; move low half rbx carry into rax
.5: rcr rax, 3
bt rbx, 2
rcl rax, 3
mov r8, [rsp]
mov rcx, rsi
mov rsi,[rsp+8]
lea r9, [r8+r8]
lea rsi, [rsi+r9*8]
lea r11, [rbp+24]
sub r11, rsi
sar r11, 3
bt r8, 0
jnc .9
; if odd the do next two
add r11, 2
mov r8, [rbp+rdx*8]
mov r9, [rbp+rdx*8+8]
rcr rbx, 2
adc r8,0
adc r9, 0
rcl rbx, 1
sbb r8, [rcx+rdx*8]
sbb r9, [rcx+rdx*8+8]
rcr rbx, 2
adc [rbp+24], r8
adc [rbp+32], r9
rcl rbx, 3
; Now add in any accummulated carries and/or borrows
;
; NOTE: We can't propagate individual borrows or carries from the second
; and third quarter blocks into the fourth quater block by simply waiting
; for carry (or borrow) propagation to end. This is because a carry into
; the fourth quarter block when it contains only maximum integers or a
; borrow when it contains all zero integers will incorrectly propagate
; beyond the end of the top quarter block.
.9: lea rdx, [rdi+rdx*8]
sub rdx, rsi
sar rdx, 3
; carries/borrrow from second to third quarter quarter block
; rax{2} is the carry in (B + C)
; rax{1} is the carry in (B + C) + A
; rax{0} is the borrow in (B + C + A) - E
mov rcx, rdx
bt rax, 0
.10: sbb qword[rsi+rcx*8], 0
add_one rcx
jrcxz .11
jc .10
.11 mov rcx, rdx
bt rax, 1
.12: adc qword[rsi+rcx*8], 0
add_one rcx
jrcxz .13
jc .12
.13 mov rcx, rdx
bt rax, 2
.14: adc qword[rsi+rcx*8], 0
add_one rcx
jrcxz .15
jc .14
; carries/borrrow from third to fourth quarter quarter block
; rbx{2} is the carry in (B + C)
; rbx{1} is the carry in (B + C) + D
; rbx{0} is the borrow in (B + C + D) - F
.15: mov rcx, r11
bt rbx, 0
.16: sbb qword[rsi+rcx*8], 0
add_one rcx
jrcxz .17
jc .16
.17: mov rcx, r11
bt rbx, 1
.18: adc qword[rsi+rcx*8], 0
add_one rcx
jrcxz .19
jc .18
.19: mov rcx, r11
bt rbx, 2
.20: adc qword[rsi+rcx*8], 0
add_one rcx
jrcxz .21
jc .20
.21:
END_PROC reg_save_list
end