https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90608
--- Comment #34 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Mikael Morin <mik...@gcc.gnu.org>: https://gcc.gnu.org/g:8aa254d7a5db5189ccd944b6ca1f5443726b058d commit r15-5446-g8aa254d7a5db5189ccd944b6ca1f5443726b058d Author: Mikael Morin <mik...@gcc.gnu.org> Date: Thu Nov 16 10:00:26 2023 +0100 fortran: Add tests covering inline MINLOC/MAXLOC with DIM [PR90608] Add the tests covering the cases for which the following patches will implement inline expansion of MINLOC and MAXLOC. Those are cases where the DIM argument is a constant value, and the ARRAY argument has rank greater than 1. PR fortran/90608 gcc/testsuite/ChangeLog: * gfortran.dg/ieee/maxloc_nan_2.f90: New test. * gfortran.dg/ieee/minloc_nan_2.f90: New test. * gfortran.dg/maxloc_with_dim_1.f90: New test. * gfortran.dg/maxloc_with_dim_and_mask_1.f90: New test. * gfortran.dg/minloc_with_dim_1.f90: New test. * gfortran.dg/minloc_with_dim_and_mask_1.f90: New test. --- Comment #35 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Mikael Morin <mik...@gcc.gnu.org>: https://gcc.gnu.org/g:a3271e0613a56402ea048da7c51267c07d5962b4 commit r15-5447-ga3271e0613a56402ea048da7c51267c07d5962b4 Author: Mikael Morin <mik...@gcc.gnu.org> Date: Fri Nov 17 19:04:19 2023 +0100 fortran: Inline unmasked integral MINLOC/MAXLOC with DIM [PR90608] Enable generation of inline code for the MINLOC and MAXLOC intrinsics, if the ARRAY argument is of integral type and of rank > 1 (only the rank 1 case was previously inlined), the DIM argument is a constant value and there is no MASK argument. The restriction to integral ARRAY and absent MASK limits the scope of the change to the cases where we generate single loop inline code. This change uses the existing scalarizer suport for reductions, that is arrays used in scalarization loops, where each element uses a nested scalarization loop to calculate its value. The nested loop (and respectively the nested scalarization chain) is created while walking the MINLOC/MAXLOC expression, it's set up automatically at the time the outer loop is set up, and gfc_conv_intrinsic_minmaxloc is changed to use it as a replacement for the local loop variable (respectively ARRAY scalarization chain) used in the non-reduction case (i.e. when DIM is absent). PR fortran/90608 gcc/fortran/ChangeLog: * trans-intrinsic.cc (gfc_inline_intrinsic_function_p): Return true if DIM is constant, ARRAY is integral and MASK is absent. (walk_inline_intrinsic_minmaxloc): If DIM is present, walk ARRAY and move the dimension corresponding to DIM to a nested chain, keeping the rest of the dimensions as the returned scalarization chain. (gfc_conv_intrinsic_minmaxloc): When inside the scalarization loops, proceed with inline code generation If DIM is present. If DIM is present, skip result array creation and final initialization from individual result local variables. If DIM is present and ARRAY has rank greater than 1, use the nested loop initialized by the scalarizer instead of the local one, use 1 as scalarization dimension, and evaluate ARRAY using the inherited scalarization chain instead of creating a local one by walking the expression. gcc/testsuite/ChangeLog: * gfortran.dg/maxloc_bounds_1.f90: Also accept the error message generated by the scalarizer in case the function call is implemented through inline code. * gfortran.dg/maxloc_bounds_2.f90: Likewise. * gfortran.dg/maxloc_bounds_3.f90: Likewise. * gfortran.dg/minmaxloc_19.f90: New test.
