| Message ID | 87o9hqsens.fsf@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | Tighten condition in vect/pr85586.c (PR 85654) | expand |
On Tue, May 8, 2018 at 11:03 AM, Richard Sandiford <richard.sandiford@linaro.org> wrote: > Another gcc.dg/vect test, another chance to play whack-a-mole > with the target selectors. In this case I think we want > { ! vect_no_align }. { { ! vect_no_align } || vect_hw_misalign } > might work too, but (a) there are other tests that use vect_no_align > on its own and (b) the point of the scan test was simply to sanity- > check that we didn't stop vectorising, rather than to test a new > vectorisation feature. > > Tested on aaarch64-linux-gnu, x86_64-linux-gnu and armeb-none-elf. > OK for trunk and GCC 8? OK. > Thanks, > Richard > > > 2018-05-08 Richard Sandiford <richard.sandiford@linaro.org> > > gcc/testsuite/ > PR testsuite/85586 > * gcc.dg/vect/pr85586.c: Restrict LOOP VECTORIZED test to > !vect_no_align. > > Index: gcc/testsuite/gcc.dg/vect/pr85586.c > =================================================================== > --- gcc/testsuite/gcc.dg/vect/pr85586.c 2018-05-02 08:39:59.942069849 +0100 > +++ gcc/testsuite/gcc.dg/vect/pr85586.c 2018-05-08 09:47:33.207979464 +0100 > @@ -40,4 +40,4 @@ main (void) > return 0; > } > > -/* { dg-final { scan-tree-dump-times "LOOP VECTORIZED" 1 "vect" { target vect_int } } } */ > +/* { dg-final { scan-tree-dump-times "LOOP VECTORIZED" 1 "vect" { target { { ! vect_no_align } && vect_int } } } } */
Index: gcc/testsuite/gcc.dg/vect/pr85586.c =================================================================== --- gcc/testsuite/gcc.dg/vect/pr85586.c 2018-05-02 08:39:59.942069849 +0100 +++ gcc/testsuite/gcc.dg/vect/pr85586.c 2018-05-08 09:47:33.207979464 +0100 @@ -40,4 +40,4 @@ main (void) return 0; } -/* { dg-final { scan-tree-dump-times "LOOP VECTORIZED" 1 "vect" { target vect_int } } } */ +/* { dg-final { scan-tree-dump-times "LOOP VECTORIZED" 1 "vect" { target { { ! vect_no_align } && vect_int } } } } */
Another gcc.dg/vect test, another chance to play whack-a-mole with the target selectors. In this case I think we want { ! vect_no_align }. { { ! vect_no_align } || vect_hw_misalign } might work too, but (a) there are other tests that use vect_no_align on its own and (b) the point of the scan test was simply to sanity- check that we didn't stop vectorising, rather than to test a new vectorisation feature. Tested on aaarch64-linux-gnu, x86_64-linux-gnu and armeb-none-elf. OK for trunk and GCC 8? Thanks, Richard 2018-05-08 Richard Sandiford <richard.sandiford@linaro.org> gcc/testsuite/ PR testsuite/85586 * gcc.dg/vect/pr85586.c: Restrict LOOP VECTORIZED test to !vect_no_align.