gcc 1.28 on Sequent Symmetry float problem
Ken Wellsch
(same problem with 1.26 and 1.27):
config.h -> config-i386.h
tm.h -> tm-seq386.h
md -> i386.md
aux-output.c -> output-i386.c
Offending event extracted from tail of Make output:
| ...
| stage2/gcc -g -O -Bstage2/ -c tree.c
| tm-bsd386.h:202: warning: ASM_OUTPUT_DOUBLE redefined
| Assembler:
| "tree.c", line 1796: "den0x00000000000000001000" is not followed by a ':'
| for a label definition
What this bizarre double constant (should be ".double 0d1.00000e+00")
means (from the local printf man page):
| PRINTF(3S) DYNIX Programmer's Manual PRINTF(3S)
|
| NAME
| printf, fprintf, sprintf - formatted output conversion
|
| SYNOPSIS
| #include <stdio.h>
|
| printf(format [, arg ] ... )
| char *format;
|
| ...
|
| If the exponent is zero and the mantissa is not equal to
| zero, then the output is
|
| [-]den0xdddddddd
|
| where 0xdddddddd is the hexadecimal representation of the
| leftmost 32 bits of the mantissa.
|
| ...
Richard Stallman
someone to send me a convincing fix. Chances are only a person with
one of those machines can find one.
That info about printf explains the last stages of the problem, but
someone still needs use a debugger to get to the root of the matter.
