test results

[Alexandre Oliva]

Jeffrey A Law writes:

> mips-sgi-irix5.2 has two regressions in the g++ tests with haifa
> enabled (eh25, eh49).  The gcc tests were the same with and without
> haifa.

I've looked with more depth into these two problems.  For may
surprise, the assembly code produced for the two examples is *exactly*
the same (unless I did some stupid mistake, which is possible since
I've slept much less than I should lately :-) I noticed that the
problem in eh49 only occurs with a high optimization level, while eh25
will core-dump either with or without optimization.  Both will
core-dump just after the throw statement (I haven't
assembly-single-stepped in the code, though)

Nevertheless, the executable code produced was different.  I tried to
run the binary produced with the Haifa scheduler with the shared
libstdc++ produced without it, and the converse too, and the results
did not change.

It seems to me that the problem is in the exception throwing function
in libgcc.

-- 
Alexandre Oliva
mailto:oliva@dcc.unicamp.br mailto:aoliva@acm.org
Universidade Estadual de Campinas, SP, Brasil