Re: dataflow branch on the ppc.

[David Daney <ddaney at avtrex dot com>]

Kenneth Zadeck wrote:
> I have a series of gij regressions on the powerpc for the dataflow branch.
>
> I am having trouble resolving these as they appear to be deep in the
> library and it seems quite difficult to play around with these as we are
> accustomed to do with most c, c++ or fortran regressions. 
>
> FAIL: noclass execution - gij test
> FAIL: pr11951 execution - gij test
> FAIL: throwit execution - gij test
> FAIL: pr29812 execution - gij test

These four are JNI tests.

FWIW, I recently had *all* JNI tests fail on MIPS with a trunk build:

http://gcc.gnu.org/ml/gcc-testresults/2007-06/msg00305.html

But in a more recent build that I didn't do they do not occur:

http://gcc.gnu.org/ml/gcc-testresults/2007-06/msg00328.html

I am doing a new trunk build to see if the failures were transient.  I 
also have no idea if these are related to what you are seeing.


> FAIL: ExtraClassLoader execution - source compiled test
> FAIL: ExtraClassLoader -findirect-dispatch execution - source compiled test
> FAIL: ExtraClassLoader -O3 execution - source compiled test
> FAIL: ExtraClassLoader -O3 -findirect-dispatch execution - source
> compiled test

These I don't know about.

> My experience in dealing with regression on the dataflow branch in
> general is that these are most likely some hard register being misused,
> generally in a signal handler. 
> The dataflow branch is much more picky about the use of registers in
> general.  The analysis tends to be a lot more precise than the current
> trunk and the new optimizations are more aggressive because they "know "
> the dataflow information they are using is correct.
>
> The help that i need is from someone who knows that way all of this is
> put together.  I need to find out what is being expected that is not
> being delivered. 
>
> These problems only happen on the ppc.  the x86-64, x86-32, and ia-64
> are clean.
>
> Thanks,
>
> Kenny