[ was: Re: [RFC] ldist: Recognize rawmemchr loop patterns ]

On 1/31/22 16:00, Richard Biener wrote:
>> I'm running into PR56888 (
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56888 ) on nvptx due to
>> this, f.i. in gcc/testsuite/gcc.c-torture/execute/builtins/strlen.c,
>> where gcc/testsuite/gcc.c-torture/execute/builtins/lib/strlen.c contains
>> a strlen function, with a strlen loop, which is transformed by
>> pass_loop_distribution into a __builtin_strlen, which is then expanded
>> into a strlen call, creating a self-recursive function. [ And on nvptx,
>> that happens to result in a compilation failure, which is how I found
>> this. ]
>>
>> According to this (
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56888#c21 ) comment:
>> ...
>> -fno-tree-loop-distribute-patterns is the reliable way to not
>> transform loops into library calls.
>> ...
>>
>> Then should we have something along the lines of:
>> ...
>> $ git diff
>> diff --git a/gcc/tree-loop-distribution.c b/gcc/tree-loop-distribution.c
>> index 6fe59cd56855..9a211d30cd7e 100644
>> --- a/gcc/tree-loop-distribution.c
>> +++ b/gcc/tree-loop-distribution.c
>> @@ -3683,7 +3683,11 @@ loop_distribution::transform_reduction_loop
>>                  && TYPE_PRECISION (ptr_type_node) >= 32)
>>                 || (TYPE_OVERFLOW_UNDEFINED (reduction_var_type)
>>                     && TYPE_PRECISION (reduction_var_type) <=
>> TYPE_PRECISION (sizetype)))
>> -         && builtin_decl_implicit (BUILT_IN_STRLEN))
>> +         && builtin_decl_implicit (BUILT_IN_STRLEN)
>> +         && flag_tree_loop_distribute_patterns)
>>           generate_strlen_builtin (loop, reduction_var, load_iv.base,
>>                                    reduction_iv.base, loc);
>>          else if (direct_optab_handler (rawmemchr_optab, TYPE_MODE
>> (load_type))
>> ...
>> ?
>>
>> Or is the comment no longer valid?
> 
> It is still valid - and yes, I think we need to guard it with this flag
> but please do it in the caller to transform_reduction_loop.

Done.

Ok for trunk?

Thanks,
- Tom