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]

Re: [patch, nios2] testsuite cleanup


On 08/21/2014 03:50 PM, Mike Stump wrote:

Sorry, I meant 4…  Your patch has four instances of this change:

-/* { dg-do run { target { ! "m68k*-*-* mmix*-*-* mep*-*-* bfin*-*-* v850*-*-* picochip*-*-* moxie*-*-* cris*-*-* m32c*-*-* fr30*-*-* mcore*-*-* powerpc*-*-* xtensa*-*-* hppa*-*-*"} } } */
+/* { dg-do run { target { ! "m68k*-*-* mmix*-*-* mep*-*-* bfin*-*-* v850*-*-* picochip*-*-* moxie*-*-* cris*-*-* m32c*-*-* fr30*-*-* mcore*-*-* powerpc*-*-* xtensa*-*-* hppa*-*-* nios2*-*-*"} } } */

see your patch for them.  Can you change the patch effectively to:

-/* { dg-do run { target { ! "m68k*-*-* mmix*-*-* mep*-*-* bfin*-*-* v850*-*-* picochip*-*-* moxie*-*-* cris*-*-* m32c*-*-* fr30*-*-* mcore*-*-* powerpc*-*-* xtensa*-*-* hppa*-*-*"} } } */
+/* non default branch cost */
+/* { dg-do run { target { ! "m68k*-*-* mmix*-*-* mep*-*-* bfin*-*-* v850*-*-* picochip*-*-* moxie*-*-* cris*-*-* m32c*-*-* fr30*-*-* mcore*-*-* powerpc*-*-* xtensa*-*-* hppa*-*-* nios2*-*-*"} } } */

instead?  The comment serves as documentation as to what all the listed targets
have in common.  A person doing a new port, can then read the comment, and say
I am non-default branch cost, so add me, or alternatively, I am the default,
this failure is a bug I need to investigate and fix.

Hmmm, I think this is backwards. At least nios2 does not override BRANCH_COST. The default value (in defaults.h) is 1; when I was investigating these tree-ssa failures I saw that the test being used to enable the optimizations is >= 2. It's also quite possible for back ends to override BRANCH_COST but in a way that still causes the tests to fail, at least for some multilibs or processor options.

I'd really like the maintainers of these tree-ssa tests to figure out what target they're supposed to work for or come up with a suitable test for feature support, rather than me trying to guess the failure mode for all these other back ends I can't test.

-Sandra


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