[Bug tree-optimization/107090] [aarch64] sequence logic should be combined with mul and umulh

[liuhongt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107090

--- Comment #17 from Hongtao Liu <liuhongt at gcc dot gnu.org> ---
(In reply to Hongtao Liu from comment #16)
> Created attachment 63963 [details]
> A patch.
> 
> Add gimple pattern to match and simplify pure c implementation of 64-bit to
> 128-bit widen_mult.
> 
> A pattern match in the match.pd can optimized the instruction sequence to
> widen_mult before inline, and help inline to get a more accurate number of
> instructions size of the function.

But overhead is there's extra warning.

gimple-match-9.cc: In function ‘bool
gimple_simplify_PLUS_EXPR(gimple_match_op*, gimple**, tree_node* (*)(tree),
code_helper, tree, tree, tree)’:
gimple-match-9.cc:11351:1: note: variable tracking size limit exceeded with
‘-fvar-tracking-assignments’, retrying without
11351 | gimple_simplify_PLUS_EXPR (gimple_match_op *res_op, gimple_seq *seq,