120 lines
3.7 KiB
ArmAsm
120 lines
3.7 KiB
ArmAsm
# Copyright (c) 1985 Regents of the University of California.
|
|
# All rights reserved.
|
|
#
|
|
# Redistribution and use in source and binary forms are permitted provided
|
|
# that: (1) source distributions retain this entire copyright notice and
|
|
# comment, and (2) distributions including binaries display the following
|
|
# acknowledgement: ``This product includes software developed by the
|
|
# University of California, Berkeley and its contributors'' in the
|
|
# documentation or other materials provided with the distribution and in
|
|
# all advertising materials mentioning features or use of this software.
|
|
# Neither the name of the University nor the names of its contributors may
|
|
# be used to endorse or promote products derived from this software without
|
|
# specific prior written permission.
|
|
# THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED
|
|
# WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
|
|
# MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
|
|
#
|
|
# @(#)cabs.s 5.4 (Berkeley) 10/9/90
|
|
#
|
|
.data
|
|
.align 2
|
|
_sccsid:
|
|
.asciz "@(#)cabs.s 1.2 (Berkeley) 8/21/85; 5.4 (ucb.elefunt) 10/9/90"
|
|
|
|
# double precision complex absolute value
|
|
# CABS by W. Kahan, 9/7/80.
|
|
# Revised for reserved operands by E. LeBlanc, 8/18/82
|
|
# argument for complex absolute value by reference, *4(ap)
|
|
# argument for cabs and hypot (C fcns) by value, 4(ap)
|
|
# output is in r0:r1 (error less than 0.86 ulps)
|
|
|
|
.text
|
|
.align 1
|
|
.globl _cabs
|
|
.globl _hypot
|
|
.globl _z_abs
|
|
.globl libm$cdabs_r6
|
|
.globl libm$dsqrt_r5
|
|
|
|
# entry for c functions cabs and hypot
|
|
_cabs:
|
|
_hypot:
|
|
.word 0x807c # save r2-r6, enable floating overflow
|
|
movq 4(ap),r0 # r0:1 = x
|
|
movq 12(ap),r2 # r2:3 = y
|
|
jmp cabs2
|
|
# entry for Fortran use, call by: d = abs(z)
|
|
_z_abs:
|
|
.word 0x807c # save r2-r6, enable floating overflow
|
|
movl 4(ap),r2 # indirect addressing is necessary here
|
|
movq (r2)+,r0 # r0:1 = x
|
|
movq (r2),r2 # r2:3 = y
|
|
|
|
cabs2:
|
|
bicw3 $0x7f,r0,r4 # r4 has signed biased exp of x
|
|
cmpw $0x8000,r4
|
|
jeql return # x is a reserved operand, so return it
|
|
bicw3 $0x7f,r2,r5 # r5 has signed biased exp of y
|
|
cmpw $0x8000,r5
|
|
jneq cont # y isn't a reserved operand
|
|
movq r2,r0 # return y if it's reserved
|
|
ret
|
|
|
|
cont:
|
|
bsbb regs_set # r0:1 = dsqrt(x^2+y^2)/2^r6
|
|
addw2 r6,r0 # unscaled cdabs in r0:1
|
|
jvc return # unless it overflows
|
|
subw2 $0x80,r0 # halve r0 to get meaningful overflow
|
|
addd2 r0,r0 # overflow; r0 is half of true abs value
|
|
return:
|
|
ret
|
|
|
|
libm$cdabs_r6: # ENTRY POINT for cdsqrt
|
|
# calculates a scaled (factor in r6)
|
|
# complex absolute value
|
|
|
|
movq (r4)+,r0 # r0:r1 = x via indirect addressing
|
|
movq (r4),r2 # r2:r3 = y via indirect addressing
|
|
|
|
bicw3 $0x7f,r0,r5 # r5 has signed biased exp of x
|
|
cmpw $0x8000,r5
|
|
jeql cdreserved # x is a reserved operand
|
|
bicw3 $0x7f,r2,r5 # r5 has signed biased exp of y
|
|
cmpw $0x8000,r5
|
|
jneq regs_set # y isn't a reserved operand either?
|
|
|
|
cdreserved:
|
|
movl *4(ap),r4 # r4 -> (u,v), if x or y is reserved
|
|
movq r0,(r4)+ # copy u and v as is and return
|
|
movq r2,(r4) # (again addressing is indirect)
|
|
ret
|
|
|
|
regs_set:
|
|
bicw2 $0x8000,r0 # r0:r1 = dabs(x)
|
|
bicw2 $0x8000,r2 # r2:r3 = dabs(y)
|
|
cmpw r0,r2
|
|
jgeq ordered
|
|
movq r0,r4
|
|
movq r2,r0
|
|
movq r4,r2 # force y's exp <= x's exp
|
|
ordered:
|
|
bicw3 $0x7f,r0,r6 # r6 = exponent(x) + bias(129)
|
|
jeql retsb # if x = y = 0 then cdabs(x,y) = 0
|
|
subw2 $0x4780,r6 # r6 = exponent(x) - 14
|
|
subw2 r6,r0 # 2^14 <= scaled x < 2^15
|
|
bitw $0xff80,r2
|
|
jeql retsb # if y = 0 return dabs(x)
|
|
subw2 r6,r2
|
|
cmpw $0x3780,r2 # if scaled y < 2^-18
|
|
jgtr retsb # return dabs(x)
|
|
emodd r0,$0,r0,r4,r0 # r4 + r0:1 = scaled x^2
|
|
emodd r2,$0,r2,r5,r2 # r5 + r2:3 = scaled y^2
|
|
addd2 r2,r0
|
|
addl2 r5,r4
|
|
cvtld r4,r2
|
|
addd2 r2,r0 # r0:1 = scaled x^2 + y^2
|
|
jmp libm$dsqrt_r5 # r0:1 = dsqrt(x^2+y^2)/2^r6
|
|
retsb:
|
|
rsb # error < 0.86 ulp
|