Patch: PR testsuite/21865 (really a middle-end vector bug)

[Steve Ellcey]

> > Hm, not sure how I missed that, I can reproduce it on ToT sources too.
> > Here is the patch again with this test skipped on PA.  This test works
> > on 64 bit PA and only fails on 32 bit PA (where BIGGEST_ALIGNMENT is 8)
> > but I don't see an easy way to specify that with xfail or dg-skip-if so
> > I just skip it on all PA platforms.  This test will probably start
> > failing on other platforms with 8 byte maximum alignment too.
> 
> Actually, there are quite a few targets with BIGGEST_ALIGNMENT less
> than 128.
> 
> Dave

You are right, I was thinking the cutoff would be 64 bits, but if floats
are 4 bytes, then we have 4*4*8 which is 128.

Maybe Richard will offer an opinion since both the code I am changing
and the testcase were added by him.

Steve Ellcey
sje@cup.hp.com