[PATCH fortran] Use fold_build everywhere

[Tobias Schlüter]

H.J. Lu wrote:
> On Sun, Feb 24, 2008 at 8:14 AM, Tobias Schlüter
>> 2008-02-24  Tobias Schlüter  <tobi@gcc.gnu.org>
>>
>>         * trans-array.c (gfc_conv_descriptor_data_get,
>>         gfc_conv_descriptor_data_set_internal,
>>         gfc_conv_descriptor_data_addr, gfc_conv_descriptor_offset,
>>         gfc_conv_descriptor_dtype, gfc_conv_descriptor_dimension,
>>         gfc_conv_descriptor_stride, gfc_conv_descriptor_lbound,
>>         gfc_conv_descriptor_ubound, gfc_trans_create_temp_array,
>>         gfc_conv_array_transpose, gfc_grow_array,
>>         gfc_trans_array_constructor_subarray,
>>         gfc_trans_array_constructor_value, gfc_trans_scalarized_loop_end,
>>         gfc_array_init_size, gfc_array_allocate, gfc_array_deallocate,
>>         gfc_conv_array_initializer, gfc_trans_array_bounds,
>>         gfc_trans_auto_array_allocation, gfc_trans_dummy_array_bias,
>>         gfc_get_dataptr_offset, gfc_conv_array_parameter,
>>         gfc_trans_dealloc_allocated, get_full_array_size,
>>         gfc_duplicate_allocatable, structure_alloc_comps): Use fold_buildN
>>         instead of buildN.
>>         * trans-expr.c (gfc_conv_expr_present, gfc_conv_missing_dummy,
>>         gfc_conv_component_ref, gfc_conv_cst_int_power,
>>         gfc_conv_function_call, gfc_trans_structur_assign): Likewise.
>>         * trans-common.c (create_common): Likewise.
>>         * trans-openmp.c (gfc_trans_omp_atomic, gfc_trans_omp_do):
>>         Likewise.
>>         * trans-const.c (gfc_conv_constant_to_tree): Likewise.
>>         * trans-stmt.c (gfc_trans_goto, gfc_trans_return, gfc_trans_do,
>>         gfc_trans_integer_select, gfc_trans_character_select,
>>         gfc_trans_forall_loop, compute_overall_iter_number,
>>         gfc_trans_forall_1, gfc_evaluate_where_mask, gfc_trans_allocate,
>>         gfc_trans_deallocate): Likewise.
>>         * trans.c (gfc_build_addr_expr, gfc_trans_runtime_check,
>>         gfc_allocate_with_status, gfc_allocate_array_with_status,
>>         gfc_deallocate_with_status): Likewise.
>>         * f95-lang.c (gfc_truthvalue_conversion): Likewise.
>>         * trans-io.c (set_parameter_const, set_parameter_value,
>>         set_parameter_ref, set_string, set_internal_unit, io_result,
>>         set_error_locus, nml_get_addr_expr, transfer_expr): Likewise.
>>         * trans-decl.c (gfc_build_qualified_array, build_entry_thunks,
>>         gfc_get_fake_result_decl, gfc_trans_auto_character_variable,
>>         gfc_generate_function_code): Likewise.
>>         * convert.c (convert): Likewise.
>>         * trans-intrinsic.c (gfc_conv_intrinsic_conversion,
>>         build_fixbound_expr, build_fix_expr, gfc_conv_intrinsic_aint,
>>         gfc_conv_intrinsic_int, gfc_conv_intrinsic_imagpart,
>>         gfc_conv_intrinsic_conjg, gfc_conv_intrinsic_abs,
>>         gfc_conv_intrinsic_cmplx, gfc_conv_intrinsic_mod,
>>         gfc_conv_intrinsic_dim, gfc_conv_intrinsic_dprod,
>>         gfc_conv_intrinsic_ctime, gfc_conv_intrinsic_fdate,
>>         gfc_conv_intrinsic_ttynam, gfc_conv_intrinsic_minmax,
>>         gfc_conv_intrinsic_minmax_char, gfc_conv_intrinsic_count,
>>         gfc_conv_intrinsic_arith, gfc_conv_intrinsic_dot_product,
>>         gfc_conv_intrinsic_minmaxloc, gfc_conv_intrinsic_minmaxval,
>>         gfc_conv_intrinsic_btest, gfc_conv_intrinsic_not,
>>         gfc_conv_intrinsic_ibits, gfc_conv_intrinsic_ishft,
>>         gfc_conv_intrinsic_ichar, gfc_conv_intrinsic_size,
>>         gfc_conv_intrinsic_array_transfer, gfc_conv_intrinsic_transfer,
>>         gfc_conv_allocated, gfc_conv_associated, gfc_conv_intrinsic_trim,
>>         gfc_conv_intrinsic_repeat): Likewise.
>>
> 
> Hi,
> 
> This patch miscompiled 172.mgrid in SPEC CPU 2K with -O2 -ffast-math on
> x86-64. We are working on a small testcase.

Hi H.J.,

yesterday I tried a follow-up patch which converted our buildN_v macros 
to use fold_buildN, and this leads to failures in a number of gomp 
tests.  Now I haven't checked if the gomp code does something weird with 
trees, but maybe this additional datapoint can help you circling the 
problem.

Cheers,
- Tobi

-- 
Tobias Schlüter
Am Coulombwall 1, Zi. 326
85748 Garching b. München
Tel.: +49/89/289-14139