Subject: Re: pkg/34571: TME sun3 emulator doesn't consider floating point infinities t...
To: None <skrll@NetBSD.org, gnats-admin@netbsd.org, pkgsrc-bugs@netbsd.org,>
From: None <SigmFSK@aol.com>
List: pkgsrc-bugs
Date: 09/28/2006 23:05:03
The following reply was made to PR pkg/34571; it has been noted by GNATS.
From: SigmFSK@aol.com
To: gnats-bugs@NetBSD.org
Cc:
Subject: Re: pkg/34571: TME sun3 emulator doesn't consider floating point infinities t...
Date: Thu, 28 Sep 2006 19:04:42 EDT
TME just maps the "fcmp" instruction to "fsub", which is why it doesn't
consider infinity equal to itself (since infinity minus infinity should not equal
0.0, but infinity should equal infinity).
The following changes correct this behavior, and allow the following
gcc-3.3.3-testsuite tests to pass under NetBSD 3.0 in TME (as they pass on a real
sun3):
960405-1.c
hugeval.c
inf-1.c
mzero2.c
mzero3.c
To fix, I did the following:
modified the script that maps the m6888x instructions to ieee754 routines so
that FCMP has its own routines, distinct from FSUB, then did a make in the
m68k directory.
# diff m6888x-auto.sh.orig m6888x-auto.sh
139c139
< 56) name=cmp ; name_ieee754=sub ; optype=DST_SRC ;; # fcmp pp
326
---
> 56) name=cmp ; optype=DST_SRC ;; # fcmp pp 326
#
modified the script that generates the ieee754 routines to add the cmp, then
did a make in the ieee754 directory (notice the last line of difference is
the change to fix the fneg problem of fneg of 0.0 not being -0.0):
# diff ieee754-ops-auto.sh.orig ieee754-ops-auto.sh
129c129
< for name in add sub mul div \
---
> for name in add cmp sub mul div \
147c147
< add | sub | mul | div | rem | pow | scale)
---
> add | cmp | sub | mul | div | rem | pow | scale)
245a246
> *-cmp) op_builtin='-' ;;
252c253
< partial-neg | unknown-neg) op0=0 ; op_builtin='-';
op1=src0 ;;
---
> partial-neg | unknown-neg) op0=-1.0 ; op_builtin='*';
op1=src0 ;;
#
after generating the new ieee754-ops-auto.c, modified each of the cmp
routines (which were created identically to the sub routines) to indicate that
positive infinity is equal to itself, as is negative infinity equal to itself:
51,57d50
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_SINGLE ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_SINGLE) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_SINGLE) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_SINGLE)))
< tme_ieee754_single_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_SINGLE_BUILTIN, 0.0);
< else
<
591,597d583
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_DOUBLE ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_DOUBLE) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_DOUBLE) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_DOUBLE)))
< tme_ieee754_double_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_DOUBLE_BUILTIN, 0.0);
< else
<
1135,1141d1120
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_EXTENDED80 ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_EXTENDED80) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_EXTENDED80) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_EXTENDED80)))
< tme_ieee754_extended80_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_EXTENDED80_BUILTIN, 0.0);
< else
<
1723,1729d1701
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_QUAD ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_QUAD) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_QUAD) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_QUAD)))
< tme_ieee754_quad_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_QUAD_BUILTIN, 0.0);
< else
<
2842,2848d2813
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_SINGLE ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_SINGLE) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_SINGLE) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_SINGLE)))
< tme_ieee754_single_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_SINGLE_BUILTIN, 0.0);
<
< else
3367,3373d3331
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_DOUBLE ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_DOUBLE) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_DOUBLE) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_DOUBLE)))
< tme_ieee754_double_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_DOUBLE_BUILTIN, 0.0);
< else
<
3892,3898d3849
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_EXTENDED80 ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_EXTENDED80) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_EXTENDED80) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_EXTENDED80)))
< tme_ieee754_extended80_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_EXTENDED80_BUILTIN, 0.0);
< else
<
4417,4423d4367
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_QUAD ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_QUAD) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_QUAD) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_QUAD)))
< tme_ieee754_quad_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_QUAD_BUILTIN, 0.0);
< else
<
5662,5668d5605
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_SINGLE ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_SINGLE) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_SINGLE) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_SINGLE)))
< tme_ieee754_single_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_SINGLE_BUILTIN, 0.0);
< else
<
5963,5969d5899
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_DOUBLE ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_DOUBLE) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_DOUBLE) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_DOUBLE)))
< tme_ieee754_double_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_DOUBLE_BUILTIN, 0.0);
< else
<
6264,6270d6193
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_EXTENDED80 ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_EXTENDED80) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_EXTENDED80) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_EXTENDED80)))
< tme_ieee754_extended80_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_EXTENDED80_BUILTIN, 0.0);
< else
<
6565,6571d6487
< if (tme_float_is_inf(src0, TME_FLOAT_FORMAT_IEEE754_QUAD ) &
< tme_float_is_inf(src1, TME_FLOAT_FORMAT_IEEE754_QUAD) &
< (tme_float_is_negative(src0, TME_FLOAT_FORMAT_IEEE754_QUAD) ==
< tme_float_is_negative(src1, TME_FLOAT_FORMAT_IEEE754_QUAD)))
< tme_ieee754_quad_value_builtin_set(dst,
TME_FLOAT_FORMAT_IEEE754_QUAD_BUILTIN, 0.0);
< else
<
The process of hand-editing the auto generated c file should be automated,
but I believe the above logic is sound.
/arthur