This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

New test cases - "VMX" (Altivec) ABI and API


This patch adds a large set of test cases for the Altivec ABI and
API.  These originally were written at Motorola, but were passed on to
me by people at Wind River.  I have had to redact them heavily to make
them acceptable to the present compiler -- not to mention,
comprehensible by humans; I suspect the originals were machine
generated, but the generator has been lost.

These test cases expose a few compiler bugs, which I will address in
forthcoming patches.  They also expose the disappointing fact that the
PPC simulator (sim/ppc) doesn't support Altivec instructions.  Thus,
if you run this test set against powerpc-eabisimaltivec, almost all
the execution tests will fail.  I've got another simulator here which
I may be able to get to work; or, can someone recommend an
alternative?

There are also outstanding issues with the code in ops.c and
yousufi-1.c; specifically, ops.c wants to use the nonexistent
"vec_unpack2{uh,ul,sh,sl}" primitives (ignore my previous message on
the subject -- the call signature is different from vec_unpack{h,l}),
and yousufi-1.c wants to give vec_ctu just one argument.  I would
appreciate input on these.

The test case gcc.c-torture/execute/va-arg-23.c used to be
gcc.dg/vmx/varargs-8.c, but since it has nothing to do with Altivec I
moved it to a generic location.

zw

        * gcc.dg/vmx: New test suite.
        * gcc.c-torture/execute/va-arg-23.c: New test case.

Attachment: vmx-testcases.diff.gz
Description: Patch adding Altivec testcases.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]