mpir/mpn/x86_64w/bulldozer/mul_basecase.asm
2017-01-18 23:22:43 +00:00

338 lines
7.2 KiB
NASM

; AMD64 mpn_mul_basecase optimised for AMD Bulldozer and Piledriver.
; Contributed to the GNU project by Torbjörn Granlund.
; Copyright 2003-2005, 2007, 2008, 2011-2013 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 either:
;
; * the GNU Lesser General Public License as published by the Free
; Software Foundation; either version 3 of the License, or (at your
; option) any later version.
;
; or
;
; * the GNU General Public License as published by the Free Software
; Foundation; either version 2 of the License, or (at your option) any
; later version.
;
; or both in parallel, as here.
;
; 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 General Public License
; for more details.
;
; You should have received copies of the GNU General Public License and the
; GNU Lesser General Public License along with the GNU MP Library. If not,
; see https://www.gnu.org/licenses/.
; cycles/limb mul_1 mul_2 mul_3 addmul_2
; AMD K8,K9
; AMD K10
; AMD bull ~4.8 ~4.55 - ~4.3
; AMD pile ~4.6 ~4.55 - ~4.55
; AMD bobcat
; AMD jaguar
; Intel P4
; Intel core
; Intel NHM
; Intel SBR
; Intel IBR
; Intel HWL
; Intel BWL
; Intel atom
; VIA nano
; The inner loops of this code are the result of running a code generation and
; optimisation tool suite written by David Harvey and Torbjorn Granlund.
; TODO
; * Merge bull-specific mul_1, if it is not slower the TOOM22 range.
; Alternatively, we could tweak the present code (which was loopmixed for a
; different CPU).
; * Merge faster mul_2, such as the one in the same directory as this file.
; * Further micro-optimise.
; mp_limb_t mpn_mul_basecase(mp_ptr, mp_ptr, mp_size_t, mp_ptr, mp_size_t)
; rax rdi rsi rdx rcx r8
; rax rcx rdx r8 r9 [rsp+40]
%include 'yasm_mac.inc'
%define reg_save_list rsi, rdi, rbx, rbp, r12, r13, r14, r15
TEXT
align 16
FRAME_PROC mpn_mul_basecase, 0, reg_save_list
mov rdi, rcx
mov rsi, rdx
mov rdx, r8
mov rcx, r9
mov r8, [rsp+stack_use+40]
mov rbx, rdx
neg rbx
mov rax, [rsi]
lea rsi, [rsi+rdx*8]
lea rdi, [rdi+rdx*8]
mov r9, [rcx]
mul r9
test r8b, 1
jz .13
.0: test bl, 1
jnz .4
.1: mov r10, rax
mov r11, rdx
mov rax, [rsi+rbx*8+8]
test bl, 2
jnz .3
.2: lea rbp, [rbx+2]
jmp .9
.3: lea rbp, [rbx]
jmp .11
.4: mov r11, rax
mov r10, rdx
test bl, 2
jz .6
.5: lea rbp, [rbx+3]
test rbp, rbp
js .8
mov [rdi-8], rax
mov [rdi], rdx
EXIT_PROC reg_save_list
.6: lea rbp, [rbx+1]
mov rax, [rsi+rbx*8+8]
jmp .10
align 16
.7: mov r10, rdx
add r11, rax
.8: mov rax, [rsi+rbp*8-16]
adc r10, 0
mul r9
add r10, rax
mov [rdi+rbp*8-24], r11
mov rax, [rsi+rbp*8-8]
mov r11, rdx
adc r11, 0
.9: mul r9
mov [rdi+rbp*8-16], r10
add r11, rax
mov r10, rdx
mov rax, [rsi+rbp*8]
adc r10, 0
.10:mul r9
mov [rdi+rbp*8-8], r11
mov r11, rdx
add r10, rax
mov rax, [rsi+rbp*8+8]
adc r11, 0
.11:mul r9
mov [rdi+rbp*8], r10
add rbp, 4
jnc .7
.12:add r11, rax
adc rdx, 0
mov [rdi-8], r11
mov [rdi], rdx
dec r8d
jz .35
lea rcx, [rcx+8]
lea rdi, [rdi+8]
jmp .20
.13:mov r14, [rcx+8]
test bl, 1
jnz .15
.14:lea rbp, [rbx]
mov r12, rax
mov rax, [rsi+rbx*8]
mov r11, rdx
mul r14
mov r10, rax
mov [rdi+rbx*8], r12
mov rax, [rsi+rbx*8+8]
mov r12, rdx
jmp .18
.15:lea rbp, [rbx+1]
mov r10, rax
mov r13, rdx
mov rax, [rsi+rbx*8]
mul r14
mov [rdi+rbx*8], r10
mov r10, rdx
mov r12, rax
mov rax, [rsi+rbx*8+8]
jmp .17
align 32
.16:add r12, rax
mov rax, [rsi+rbp*8]
adc r10, 0
.17:mul r9
add r12, rax
mov rax, [rsi+rbp*8]
mov r11, rdx
adc r11, 0
mul r14
add r12, r13
adc r11, 0
add r10, rax
mov [rdi+rbp*8], r12
mov rax, [rsi+rbp*8+8]
mov r12, rdx
adc r12, 0
.18:mul r9
add r10, rax
mov r13, rdx
adc r13, 0
add r10, r11
adc r13, 0
mov rax, [rsi+rbp*8+8]
mul r14
add rbp, 2
mov [rdi+rbp*8-8], r10
mov r10, rdx
jnc .16
.19:add r12, rax
adc rdx, 0
add r12, r13
adc rdx, 0
mov [rdi], r12
mov [rdi+8], rdx
add r8d, -2
jz .34
lea rcx, [rcx+16]
lea rdi, [rdi+16]
.20:
mov [rsp+stack_use+8], r8
.21:mov r9, [rcx]
mov r8, [rcx+8]
mov rax, [rsi+rbx*8]
mul r9
test bl, 1
jnz .25
.22:mov r15, rax
mov rax, [rsi+rbx*8]
mov r14, rdx
mul r8
test bl, 2
jnz .24
.23:lea rbp, [rbx]
mov r13, [rdi+rbx*8]
mov r10, rax
mov rax, [rsi+rbx*8+8]
mov r11, rdx
jmp .31
.24:lea rbp, [rbx+2]
mov r11, [rdi+rbx*8]
mov r13, rdx
mov r12, rax
mov rax, [rsi+rbx*8+8]
jmp .29
.25:mov r14, rax
mov rax, [rsi+rbx*8]
mov r15, rdx
mul r8
test bl, 2
jz .27
.26:lea rbp, [rbx+1]
mov r12, [rdi+rbx*8]
mov r10, rdx
mov r13, rax
jmp .30
.27:lea rbp, [rbx-1]
mov r10, [rdi+rbx*8]
mov r11, rax
mov rax, [rsi+rbx*8+8]
mov r12, rdx
jmp .32
align 32
.28:
.29:mul r9
add r15, r11
mov [rdi+rbp*8-16], r15
mov r15, rdx
adc r14, rax
adc r15, 0
mov rax, [rsi+rbp*8-8]
mul r8
mov r11, [rdi+rbp*8-8]
mov r10, rdx
add r12, r11
adc r13, rax
adc r10, 0
.30:mov rax, [rsi+rbp*8]
mul r9
add r14, r12
mov [rdi+rbp*8-8], r14
mov r14, rdx
adc r15, rax
mov rax, [rsi+rbp*8]
adc r14, 0
mov r12, [rdi+rbp*8]
mul r8
add r13, r12
adc r10, rax
mov rax, [rsi+rbp*8+8]
mov r11, rdx
adc r11, 0
.31:mul r9
add r15, r13
mov [rdi+rbp*8], r15
adc r14, rax
mov rax, [rsi+rbp*8+8]
mov r15, rdx
adc r15, 0
mov r13, [rdi+rbp*8+8]
mul r8
add r10, r13
adc r11, rax
mov rax, [rsi+rbp*8+16]
mov r12, rdx
adc r12, 0
.32:mul r9
add r14, r10
mov [rdi+rbp*8+8], r14
mov r14, rdx
adc r15, rax
adc r14, 0
mov rax, [rsi+rbp*8+16]
mov r10, [rdi+rbp*8+16]
mul r8
mov r13, rdx
add r11, r10
adc r12, rax
adc r13, 0
mov rax, [rsi+rbp*8+24]
add rbp, 4
jnc .28
.33:mul r9
add r15, r11
mov [rdi-16], r15
mov r15, rdx
adc r14, rax
adc r15, 0
mov rax, [rsi-8]
mul r8
mov r11, [rdi-8]
add r12, r11
adc r13, rax
adc rdx, 0
add r14, r12
adc r15, 0
mov [rdi-8], r14
add r15, r13
mov [rdi], r15
adc rdx, 0
mov [rdi+8], rdx
add qword [rsp+stack_use+8], -2
lea rcx, [rcx+16]
lea rdi, [rdi+16]
jnz .21
mov rax, [rsp+stack_use+8]
.34:
.35:
END_PROC reg_save_list