[MiNT] Strange FPU Stuff

[Mark Duckworth <mduckworth@atari-source.com>]

Hello guys,

I am trying to determine if my fpu problems are hardware or software
(fpu emulator of ct60) related.

Here's a post with a sample program:

http://gcc.gnu.org/ml/gcc-bugs/1999-03n/msg00897.html

Here is some interesting results similar to the bug he described:

root@washi:/root/fputest>gcc -m68020-60 -m68881 -msoft-float -o fptest
fptest.c root@washi:/root/fputest>./fptest
rounding is double, result = 0x640005, expected result = 0x640005
rounding is float, result = 0x640005, expected result = 0x640005
root@washi:/root/fputest>gcc -m68020-60 -m68881 -o fptest fptest.c
root@washi:/root/fputest>./fptest
rounding is double, result = 0x640006, expected result = 0x640005
rounding is float, result = 0x640006, expected result = 0x640005
root@washi:/root/fputest>gcc -m68060 -o fptest fptest.c
root@washi:/root/fputest>./fptest
rounding is double, result = 0x640005, expected result = 0x640005
rounding is float, result = 0x640005, expected result = 0x640005
root@washi:/root/fputest>gcc -m68060 -m68881 -o fptest fptest.c
root@washi:/root/fputest>./fptest
rounding is double, result = 0x640005, expected result = 0x640005
rounding is float, result = 0x640005, expected result = 0x640005
root@washi:/root/fputest>gcc -m68000 -m68881 -o fptest fptest.c
root@washi:/root/fputest>./fptest
rounding is double, result = 0x640006, expected result = 0x640005
rounding is float, result = 0x640006, expected result = 0x640005
root@washi:/root/fputest>

Can anyone make sense of these results?  Note the sample code and read
the post it describes a possible 060 bug so this may not be relevant as
to whether or not my 060 is okay.  But check this out.  This is an fpu
test program.

root@washi:/root/fputest>gcc -DACCURACY -DINTRIG -o fputest fputest.c
fputest.c:554: warning: `transit_surface' was declared implicitly
`extern' and later `static'
fputest.c:509: warning: previous declaration of `transit_surface'
root@washi:/root/fputest>./fputest
Beginning execution of floating point accuracy test...
Pass 100.
Pass 200.
 
root@washi:/root/fputest>

This is the only way this fpu test program works at all, the -DINTRIG
uses internal routines for all floating point functions (defined in the
.c).  

Compiling with -msoft-float gives this:

root@washi:/root/fputest>gcc -DACCURACY -m68020-60 -msoft-float -o
fputest fputest.c
fputest.c:554: warning: `transit_surface' was declared implicitly
`extern' and later `static'
fputest.c:509: warning: previous declaration of `transit_surface'
fputest.c: In function `sin':
/usr/include/bits/math-68881.h:79: inconsistent operand constraints in
an `asm'
fputest.c: In function `cos':
/usr/include/bits/math-68881.h:92: inconsistent operand constraints in
an `asm'
fputest.c: In function `tan':
/usr/include/bits/math-68881.h:105: inconsistent operand constraints in
an `asm'fputest.c: In function `asin':
/usr/include/bits/math-68881.h:118: inconsistent operand constraints in
an `asm'root@washi:/root/fputest>

Mint soft-float bug?

Compiling any other way that I can think of works but gives failed
tests, such as with -m68020-60 -m68881 or -m68020-60 (-m68881 defined
implicitly I believe).  Without -m68881 defined explicitly (68000) or
implicitly (68020 and above) the compile fails because of missing math
functions.  So I thought I'd post this here first before the mintlist
because If I can't compile the test progs then who cares ;-)

Thanks,
Mark