190 lines
6 KiB
ArmAsm
190 lines
6 KiB
ArmAsm
|
/* $NetBSD: modf.S,v 1.4 2003/08/07 16:42:27 agc Exp $ */
|
||
|
|
|
||
|
|
/*
|
||
|
|
* Copyright (c) 1992, 1993
|
||
|
|
* The Regents of the University of California. All rights reserved.
|
||
|
|
*
|
||
|
|
* This software was developed by the Computer Systems Engineering group
|
||
|
|
* at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
|
||
|
|
* contributed to Berkeley.
|
||
|
|
*
|
||
|
|
* Redistribution and use in source and binary forms, with or without
|
||
|
|
* modification, are permitted provided that the following conditions
|
||
|
|
* are met:
|
||
|
|
* 1. Redistributions of source code must retain the above copyright
|
||
|
|
* notice, this list of conditions and the following disclaimer.
|
||
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
||
|
|
* notice, this list of conditions and the following disclaimer in the
|
||
|
|
* documentation and/or other materials provided with the distribution.
|
||
|
|
* 3. 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 BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
|
||
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
||
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
||
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
||
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
||
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
||
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
||
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
||
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
||
|
|
* SUCH DAMAGE.
|
||
|
|
*
|
||
|
|
* from: Header: modf.s,v 1.3 92/06/20 00:00:54 torek Exp
|
||
|
|
*/
|
||
|
|
|
||
|
|
#include <machine/asm.h>
|
||
|
|
#if defined(LIBC_SCCS) && !defined(lint)
|
||
|
|
#if 0
|
||
|
|
.asciz "@(#)modf.s 8.1 (Berkeley) 6/4/93"
|
||
|
|
#else
|
||
|
|
RCSID("$NetBSD: modf.S,v 1.4 2003/08/07 16:42:27 agc Exp $")
|
||
|
|
#endif
|
||
|
|
#endif /* LIBC_SCCS and not lint */
|
||
|
|
|
||
|
|
#include <machine/fsr.h>
|
||
|
|
|
||
|
|
/*
|
||
|
|
* double modf(double val, double *iptr)
|
||
|
|
*
|
||
|
|
* Returns the fractional part of `val', storing the integer part of
|
||
|
|
* `val' in *iptr. Both *iptr and the return value have the same sign
|
||
|
|
* as `val'.
|
||
|
|
*
|
||
|
|
* Method:
|
||
|
|
*
|
||
|
|
* We use the fpu's normalization hardware to compute the integer portion
|
||
|
|
* of the double precision argument. Sun IEEE double precision numbers
|
||
|
|
* have 52 bits of mantissa, 11 bits of exponent, and one bit of sign,
|
||
|
|
* with the sign occupying bit 31 of word 0, and the exponent bits 30:20
|
||
|
|
* of word 0. Thus, values >= 2^52 are by definition integers.
|
||
|
|
*
|
||
|
|
* If we take a value that is in the range [+0..2^52) and add 2^52, all
|
||
|
|
* of the fractional bits fall out and all of the integer bits are summed
|
||
|
|
* with 2^52. If we then subtract 2^52, we get those integer bits back.
|
||
|
|
* This must be done with rounding set to `towards 0' or `towards -inf'.
|
||
|
|
* `Toward -inf' fails when the value is 0 (we get -0 back)....
|
||
|
|
*
|
||
|
|
* Note that this method will work anywhere, but is machine dependent in
|
||
|
|
* various aspects.
|
||
|
|
*
|
||
|
|
* Stack usage:
|
||
|
|
* 4@[%fp + BIAS - 4] saved %fsr
|
||
|
|
* 4@[%fp + BIAS - 8] new %fsr with rounding set to `towards 0'
|
||
|
|
* 8@[%fp + BIAS - 16] space for moving between %i and %f registers
|
||
|
|
* Register usage:
|
||
|
|
* %f0:f1 double val;
|
||
|
|
* %l0 scratch
|
||
|
|
* %l1 sign bit (0x80000000)
|
||
|
|
* %i1 double *iptr;
|
||
|
|
* %f2:f3 `magic number' 2^52, in fpu registers
|
||
|
|
* %f4:f5 double v, in fpu registers
|
||
|
|
* %f6:f7 double temp.
|
||
|
|
*/
|
||
|
|
|
||
|
|
.align 8
|
||
|
|
Lmagic:
|
||
|
|
.word 0x43300000 ! sign = 0, exponent = 52 + 1023, mantissa = 0
|
||
|
|
.word 0 ! (i.e., .double 0r4503599627370496e+00)
|
||
|
|
|
||
|
|
L0:
|
||
|
|
.word 0 ! 0.0
|
||
|
|
.word 0
|
||
|
|
|
||
|
|
ENTRY(modf)
|
||
|
|
save %sp, -CC64FSZ-16, %sp
|
||
|
|
|
||
|
|
/*
|
||
|
|
* First, compute v = abs(val)
|
||
|
|
*/
|
||
|
|
fabsd %f0, %f4 ! %f4:f5 = v
|
||
|
|
fcmped %fcc1, %f0, %f4 ! %fcc1 = (val == abs(val))
|
||
|
|
#ifdef PIC
|
||
|
|
PICCY_SET(Lmagic, %l0, %o7)
|
||
|
|
ldd [%l0], %f2
|
||
|
|
#else
|
||
|
|
sethi %hi(Lmagic), %l0
|
||
|
|
ldd [%l0 + %lo(Lmagic)], %f2
|
||
|
|
#endif
|
||
|
|
|
||
|
|
/*
|
||
|
|
* Is %f4:f5 >= %f2:f3 ? If so, it is all integer bits.
|
||
|
|
* It is probably less, though.
|
||
|
|
*/
|
||
|
|
fcmped %f4, %f2
|
||
|
|
fbuge Lbig ! if >= (or unordered), go out
|
||
|
|
nop
|
||
|
|
|
||
|
|
/*
|
||
|
|
* v < 2^52, so add 2^52, then subtract 2^52, but do it all
|
||
|
|
* with rounding set towards zero. We leave any enabled
|
||
|
|
* traps enabled, but change the rounding mode. This might
|
||
|
|
* not be so good. Oh well....
|
||
|
|
*/
|
||
|
|
st %fsr, [%fp + BIAS - 4] ! %l5 = current FSR mode
|
||
|
|
set FSR_RD, %l3 ! %l3 = rounding direction mask
|
||
|
|
ld [%fp + BIAS - 4], %l5
|
||
|
|
set FSR_RD_RZ << FSR_RD_SHIFT, %l4
|
||
|
|
andn %l5, %l3, %l6
|
||
|
|
or %l6, %l4, %l6 ! round towards zero, please
|
||
|
|
and %l5, %l3, %l5 ! save original rounding mode
|
||
|
|
st %l6, [%fp + BIAS - 8]
|
||
|
|
ld [%fp + BIAS - 8], %fsr
|
||
|
|
|
||
|
|
faddd %f4, %f2, %f4 ! %f4:f5 += 2^52
|
||
|
|
fsubd %f4, %f2, %f4 ! %f4:f5 -= 2^52
|
||
|
|
|
||
|
|
/*
|
||
|
|
* Restore %fsr, but leave exceptions accrued.
|
||
|
|
*/
|
||
|
|
st %fsr, [%fp + BIAS - 4]
|
||
|
|
ld [%fp + BIAS - 4], %l6
|
||
|
|
andn %l6, %l3, %l6 ! %l6 = %fsr & ~FSR_RD;
|
||
|
|
or %l5, %l6, %l5 ! %l5 |= %l6;
|
||
|
|
st %l5, [%fp + BIAS - 4]
|
||
|
|
ld [%fp + BIAS - 4], %fsr ! restore %fsr, leaving accrued stuff
|
||
|
|
|
||
|
|
/*
|
||
|
|
* Now insert the original sign in %f4:f5.
|
||
|
|
* %fcc1 should still have the reults of (val == abs(val))
|
||
|
|
* from above, so we use a conditional move on %fcc1 to:
|
||
|
|
*
|
||
|
|
* %f4 = (val == abs(val)) ? %f4 : -%f4
|
||
|
|
*
|
||
|
|
*/
|
||
|
|
fnegd %f4, %f6
|
||
|
|
fmovdnz %fcc1, %f6, %f4
|
||
|
|
1:
|
||
|
|
|
||
|
|
/*
|
||
|
|
* The value in %f4:f5 is now the integer portion of the original
|
||
|
|
* argument. We need to store this in *ival (%i1), subtract it
|
||
|
|
* from the original value argument (%d0), and return the result.
|
||
|
|
*/
|
||
|
|
std %f4, [%i1] ! *ival = %f4:f5;
|
||
|
|
fsubd %f0, %f4, %f0 ! %f0:f1 -= %f4:f5;
|
||
|
|
ret
|
||
|
|
restore
|
||
|
|
|
||
|
|
Lbig:
|
||
|
|
/*
|
||
|
|
* We get here if the original comparison of %f4:f5 (v) to
|
||
|
|
* %f2:f3 (2^52) came out `greater or unordered'. In this
|
||
|
|
* case the integer part is the original value, and the
|
||
|
|
* fractional part is 0.
|
||
|
|
*/
|
||
|
|
#ifdef PIC
|
||
|
|
PICCY_SET(L0, %l0, %o7)
|
||
|
|
std %f0, [%i1] ! *ival = val;
|
||
|
|
ldd [%l0], %f0 ! return 0.0;
|
||
|
|
#else
|
||
|
|
sethi %hi(L0), %l0
|
||
|
|
std %f0, [%i1] ! *ival = val;
|
||
|
|
ldd [%l0 + %lo(L0)], %f0 ! return 0.0;
|
||
|
|
#endif
|
||
|
|
ret
|
||
|
|
restore
|
||
|
|
|