bug in System V c compiler for 3b2

[Dennis D. Jurgensen]

*** REPLACE THIS LINE WITH YOUR MESSAGE ***

I compiled a 68000 cross-compiler on a 3b2 successfully I thought.
However when I tried to use it to compile some floating point
libraries the compiler core-dumped due to an illegal instruction.
Here is the sdb output:

Source path: .:Mcc:mip
No core image
*makety:1341 b
*Breakpoint at
makety:1341: if( ISUNSIGNED(pt) ){
*0x808088a6 (makety:1341+3): BGUB 0x11 <808088b7> [0x11]
*0x808088a8 (makety:1341+5): CMPW &0xc,0(%fp) [0xc]
*0x808088ab (makety:1341+8): BLUB 0xc <808088b7> [0xc]
*0x808088b7 (makety:1345): MOVWD 0x2c(%r8),%r0 [p,%r0]
* r0/ 7 r1/ 0xfffffff2 r2/ 0x29
r3/ 0x80896bc0 r4/ 0x80896b8c r5/ 4
r6/ 7 r7/ 0x72 r8/ 0x80896bc0
ap/ 0xc00205c0 fp/ 0xc00205f4 sp/ 0xc00205f8
psw/ 0x241f0b pc/ 0x808088b7
0x808088b7 (makety:1345): MOVWD 0x2c(%r8),%r0 [p,%r0]
*0x80896bc0
*Illegal Instruction (4) (sig 4)
- core dumped
Process terminated
*

The c statement that is involved is:
p->fpn.dval = p->tn.lval;

where fpn and tn are structures in an union and dval is a double
and lval is a long. Here is the disassembly of the c statement.

[37] 2111: 85 c8 2c 40 MOVWD 0x2c(%r8),%r0
2115: a5 40 c8 14 MOVDD %r0,0x14(%r8)

if somebody could tell me what is wrong with the instruction,
i.e. is the opcode invalid, is the addressing mode invalid, etc.
and if so is there a fix for this problem please send mail to
{ihnp4!}[mtuxo, mtx5a, mtx5c]!ddj
Thank you in advance.
Dennis D. Jurgensen

[Ron Thompson]

> I compiled a 68000 cross-compiler on a 3b2 successfully I thought.
> However when I tried to use it to compile some floating point
> libraries the compiler core-dumped due to an illegal instruction.

Try using the -f option to include floating point software - then
be prepared for snail-speed execution.
--
Ron Thompson AT&T Information Systems Customer Programming
(404) 982-4217 Atlanta, Georgia Services Center
..{ihnp4,akgua}!cpsc53!rt (Opinions expressed are mine alone.)

[Robert Viduya]

>d...@mtx5a.UUCP (Dennis D. Jurgensen) (d...@mtx5a.UUCP, <13...@mtx5a.UUCP>):

>
> I compiled a 68000 cross-compiler on a 3b2 successfully I thought.
> However when I tried to use it to compile some floating point
> libraries the compiler core-dumped due to an illegal instruction.
>

My guess as to what the problem is is that a 3B2/300 (don't know about 3B2/310s
or 3B2/400s) doesn't support hardware floating point instructions, but the
C compiler will generate them. In order to compile programs that use floats
or doubles, the C compiler has to be called with the '-f' option (this is, of
course, undocumented, at least in all the 3B2 manuals I have). Note
that the compiler will still generate floating point instructions; the -f
option merely loads up a different runtime object file (/lib/fcrt0.o instead
of /lib/crt0.o). I don't know for sure, but I think fcrt0.o enables an
illegal instruction trap to catch all illegal instructions and simulate the
floating point ones in software. Slow, but effective.

robert
--
Robert Viduya 01111000
Office of Computing Services
Georgia Institute of Technology

UUCP: ..gatech!gitpyr!robert
BITNET: CC100RV @ GITVM1