This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

IPA-SRA misc trivia observations


Hi,

0) Please run CSiBE so we can see if or how much IPA-SRA regresses for -Os

1)
[PATCH, pretty-ipa merge 1/4] Function parameter manipulation infrastructer through param notes
s|/* Structure do describe|/* Structure to describe|
s|functions modyfying|functions modifying|

If ipa_parm_note.base_index is zero-based, why not make it unsigned int?
Same for the retval of count_formal_params_1() etc.

docs for count_formal_params_1():
/* Return number of formal parameters of FNDECL.  */
sounds a little bit like type_num_arguments() but is of course a different thing.
It seems to be used quite often in the tree, perhaps
just add a FOR_EACH_DECL_ARGUMENT macro and use that everywhere?

dump_aggregate_1(): 2 surplus empty lines
Perhaps make dump_aggregate_1 and dump_aggregate's "indent" param unsigned?


get_vector_of_formal_parm_types(): Reimplements
type_num_arguments(). Just int count = type_num_arguments (fntype) ?

ipa_modify_formal_parameters(): Sounds odd to have a final return in that void fn.

ipa_modify_call_arguments(): Why don't you just VEC_iterate over the notes?
I didn look close, but the last part of the function looks similar to
gimple_call_copy_skip_args(), no?

/* Return true iff BASE_INDEX is in  NOTES twice or more times.  */
just
/* Return true iff BASE_INDEX is in NOTES more than once.  */

ipa_combine_notes(): VEC_iterate ? A bitmap would make that one look prettier i suppose ;)

s|assuming they are menat to be|assuming they are ment to be|

ipa_dump_param_notes(): Yet another odd final return in a void function.

cheers,


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]