gforth
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gforth] Strange behavior of f! etc. - fixed


From: Dennis Ruffer
Subject: Re: [gforth] Strange behavior of f! etc. - fixed
Date: Mon, 13 Feb 2012 15:54:44 -0800

Still not the same output in gforth and gforth-fast:

address@hidden /usr/src/gforth
# gforth
Gforth 0.7.9-20120209, Copyright (C) 1995-2011 Free Software Foundation,
Inc.
Gforth comes with ABSOLUTELY NO WARRANTY; for details type `license'
Type `bye' to exit
1 2 3 pi cr .s f.s fsin cr .s f.s
<3> 1 2 3 <1> 3.141592653590E0
<3> 1 2 3 <1> 1.2246467991E-16  ok
bye

address@hidden /usr/src/gforth
# gforth-fast
Gforth 0.7.9-20120209, Copyright (C) 1995-2011 Free Software Foundation,
Inc.
Gforth comes with ABSOLUTELY NO WARRANTY; for details type `license'
Type `bye' to exit
1 2 3 pi cr .s f.s fsin cr .s f.s
<3> 1 2 3 <1> 3.141592653590E0
<3> 1 2 54 <1> 1.2246467991E-16  ok

DaR

-----Original Message-----
From: Bernd Paysan [mailto:address@hidden 
Sent: Monday, February 13, 2012 2:19 PM
To: address@hidden
Cc: Dennis Ruffer; 'Marcel Hendrix'
Subject: Re: [gforth] Strange behavior of f! etc. - fixed

Am Montag, 13. Februar 2012, 13:46:10 schrieb Dennis Ruffer:
> I have confirmed your results back through 0.7.0-20100915, but at
> 0.7.0-20100807 it appears to have been worse:
> 
> # gforth-fast-0.7.0-20100807.exe
> Gforth 0.7.0-20100807, Copyright (C) 1995-2009 Free Software 
> Foundation, Inc.
> Gforth comes with ABSOLUTELY NO WARRANTY; for details type `license'
> Type `bye' to exit
> 1 2 3 pi cr .s f.s fsin cr .s f.s
> <3> 1 2 3 <1> 3.141592653590E0
> <3> 1 2 2125807600 <1> 1.2246467991E-16  ok

Ok, the problem seems to be that fsin et. al globber %edx, and GCC doesn't
realize that the global register assignment won't work.  I've now moved
TOSreg to %ecx, which seems to cure that problem.  The few remaining x86
opcodes in prim that will use %ecx should be fine with that, because it's
already TOS (lshift, rshift).

Thanks for all who participated in nailing down this bug.  I have not tested
it with the original problem submission, but I'm sure someone else on the
list will do ;-).

--
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://bernd-paysan.de/




reply via email to

[Prev in Thread] Current Thread [Next in Thread]