[Phew, why do we allow empty reports? Just mishit "enter" and then this.] The vect_* target selectors are evaluated with the options in DEFAULT_VECTCFLAGS in effect, but these options are not automatically passed to tests out of the vect directories. This causes test failures to show up here and there. To me using vect_* target selectors out of vect directories is simply wrong. Instead of forcing the maintainers to fix these failures one by one, target by target, we should just move the tests into vect directories, or adjust them not to rely on vect_* target selectors. And we should make a policy to disallow using { target vect_* } in tests out of vect directories.
Confirmed. I ran into this 4 years ago and I fixed some. My original attempt of fixing this, ran into this on MIPS even: https://gcc.gnu.org/pipermail/gcc-patches/2020-January/538822.html See https://gcc.gnu.org/pipermail/gcc-patches/2020-January/538956.html for the move suggestion. Patch which moved the testcases that was found at the time: https://gcc.gnu.org/pipermail/gcc-patches/2020-January/538958.html > we should just move the tests into vect directories Yes like I did 4 years ago :). Do you have a list of the ones that still use `{ target vect_*`; I think they are all have been added in the last 4 years even since I did an audit back then too?
In gcc.dg: align-2.c analyzer/torture/pr93350.c analyzer/torture/uninit-bit-field-ref.c bic-bitmask-13.c bic-bitmask-14.c bic-bitmask-15.c bic-bitmask-16.c bic-bitmask-17.c bic-bitmask-18.c bic-bitmask-19.c bic-bitmask-20.c bic-bitmask-21.c bic-bitmask-22.c bic-bitmask-7.c gomp/_Atomic-4.c gomp/pr87887-1.c gomp/pr87887-2.c gomp/pr88107.c gomp/pr95108.c graphite/run-id-5.c graphite/run-id-6.c graphite/vect-pr43423.c lto/simd-function_0.c pr104992.c signbit-2.c tree-ssa/gen-vect-11a.c tree-ssa/gen-vect-11b.c tree-ssa/gen-vect-11.c tree-ssa/gen-vect-11c.c tree-ssa/gen-vect-25.c tree-ssa/gen-vect-26.c tree-ssa/gen-vect-28.c tree-ssa/gen-vect-2.c tree-ssa/gen-vect-32.c tree-ssa/gen-vect-34.c tree-ssa/scev-16.c In c-c++-common: gomp/declare-variant-13.c gomp/declare-variant-14.c gomp/pr60823-2.c gomp/pr60823-4.c In gfortran.dg: gomp/declare-variant-13.f90 gomp/declare-variant-14.f90 graphite/vect-pr40979.f90 I just found them with a simple grep command so there might be false positives or false negatives. There are also a dozen matches in gcc.target but I consider them fine as the target maintainers should know exactly what they are doing.
Thanks for filing this, I just realized that it's unexpected to use vect_* effective target checks outside */vect/ in generic test suites. > > I just found them with a simple grep command so there might be false > positives or false negatives. There are also a dozen matches in gcc.target > but I consider them fine as the target maintainers should know exactly what > they are doing. Yes, I think those in target should be fine, although they can be replaced with some corresponding target specific check(s), sometimes the vect_* is more readable.
For pr104992.c: https://gcc.gnu.org/pipermail/gcc-patches/2024-January/643697.html
The master branch has been updated by Xi Ruoyao <xry111@gcc.gnu.org>: https://gcc.gnu.org/g:889fbc9454e2d4e2b9a11a9e02b3b7e698edcd1c commit r14-9290-g889fbc9454e2d4e2b9a11a9e02b3b7e698edcd1c Author: Xi Ruoyao <xry111@xry111.site> Date: Tue Jan 23 19:58:21 2024 +0800 testsuite: Make pr104992.c irrelated to target vector feature [PR113418] The vect_int_mod target selector is evaluated with the options in DEFAULT_VECTCFLAGS in effect, but these options are not automatically passed to tests out of the vect directories. So this test fails on targets where integer vector modulo operation is supported but requiring an option to enable, for example LoongArch. In this test case, the only expected optimization not happened in original is in corge because it needs forward propogation. So we can scan the forwprop2 dump (where the vector operation is not expanded to scalars yet) instead of optimized, then we don't need to consider vect_int_mod or not. gcc/testsuite/ChangeLog: PR testsuite/113418 * gcc.dg/pr104992.c (dg-options): Use -fdump-tree-forwprop2 instead of -fdump-tree-optimized. (dg-final): Scan forwprop2 dump instead of optimized, and remove the use of vect_int_mod. * lib/target-supports.exp (check_effective_target_vect_int_mod): Remove because it's not used anymore.