我想为计算汇编代码sin(x)
使用“泰勒展开”)在Linux中。
Answer 1:
将这篇文章能帮你吗?
http://www.coranac.com/2009/07/sines/
它用于计算近似的sin(x)的值,与C和汇编版本几个算法。 诚然,这是ARM汇编,但它的精神应该很容易转换为x86或相似。
Answer 2:
所以我假设的x86你没有说明哪些CPU架构。
的simplist(以及可能的最无效率)的方法是在写RPN式中,其可以几乎直接映射到FPU的指令。
例,
代数公式:X - (!X ^ 3/3)+(!X ^ 5/5)
RPN:XXX * X * 3 2 * / - XX * X * X * X * 5 4 * 3 * 2 * / +
成为:
fld x
fld x
fld x
fmul
fld x
fmul
fild [const_3]
fild [const_2]
fmul
fdiv
fsub
fld x
fld x
fmul
fld x
fmul
fld x
fmul
fld x
fmul
fild [const_5]
fild [const_4]
fmul
fild [const_3]
fmul
fild [const_2]
fmul
fdiv
fadd
有一些明显的优化策略 -
- 而不是计算X,X X X,X X X X X等每个术语,储存“运行的产品”,只是乘以X * X每次
- 而不是计算阶乘每个术语,做同样的“运行产品”
下面是针对x86 FPU一些注释代码,每个FPU指令之后的言论表明后指令被执行,与栈顶(ST0)在左边,如栈状态:
fldz ; 0
fld1 ; 1, 0
--snip--
bits 32
section .text
extern printf
extern atof
extern atoi
extern puts
global main
taylor_sin:
push eax
push ecx
; input :
; st(0) = x, value to approximate sin(x) of
; [esp+12] = number of taylor series terms
; variables we'll use :
; s = sum of all terms (final result)
; x = value we want to take the sin of
; fi = factorial index (1, 3, 5, 7, ...)
; fc = factorial current (1, 6, 120, 5040, ...)
; n = numerator of term (x, x^3, x^5, x^7, ...)
; setup state for each iteration (term)
fldz ; s x
fxch st1 ; x s
fld1 ; fi x s
fld1 ; fc fi x s
fld st2 ; n fc fi x s
; first term
fld st1 ; fc n fc fi x s
fdivr st0,st1 ; r n fc fi x s
faddp st5,st0 ; n fc fi x s
; loop through each term
mov ecx,[esp+12] ; number of terms
xor eax,eax ; zero add/sub counter
loop_term:
; calculate next odd factorial
fld1 ; 1 n fc fi x s
faddp st3 ; n fc fi x s
fld st2 ; fi n fc fi x s
fmulp st2,st0
fld1 ; 1 n fc fi x s
faddp st3 ; n fc fi x s
fld st2 ; fi n fc fi x s
fmulp st2,st0 ; n fc fi x s
; calculate next odd power of x
fmul st0,st3 ; n*x fc fi x s
fmul st0,st3 ; n*x*x fc fi x s
; divide power by factorial
fld st1 ; fc n fc fi x s
fdivr st0,st1 ; r n fc fi x s
; check if we need to add or subtract this term
test eax,1
jnz odd_term
fsubp st5,st0 ; n fc fi x s
jmp skip
odd_term:
; accumulate result
faddp st5,st0 ; n fc fi x s
skip:
inc eax ; increment add/sub counter
loop loop_term
; unstack work variables
fstp st0
fstp st0
fstp st0
fstp st0
; result is in st(0)
pop ecx
pop eax
ret
main:
; check if we have 2 command-line args
mov eax, [esp+4]
cmp eax, 3
jnz error
; get arg 1 - value to calc sin of
mov ebx, [esp+8]
push dword [ebx+4]
call atof
add esp, 4
; get arg 2 - number of taylor series terms
mov ebx, [esp+8]
push dword [ebx+8]
call atoi
add esp, 4
; do the taylor series approximation
push eax
call taylor_sin
add esp, 4
; output result
sub esp, 8
fstp qword [esp]
push format
call printf
add esp,12
; return to libc
xor eax,eax
ret
error:
push error_message
call puts
add esp,4
mov eax,1
ret
section .data
error_message: db "syntax: <x> <terms>",0
format: db "%0.10f",10,0
运行程序:
$ ./taylor-sine 0.5 1
0.4791666667
$ ./taylor-sine 0.5 5
0.4794255386
$ echo "s(0.5)"|bc -l
.47942553860420300027
Answer 3:
为什么? 已经有自80387(大约1987)处理器FCOS和FSIN操作码
资源:
http://ref.x86asm.net/coder32.html
维基百科
从演示现场私交
文章来源: Assembly code for sin(x)