This is the mail archive of the 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]

Re: [gomp4] rewrite simd clone argument adjustment to avoid regimplification

On 11/07/13 10:58, Martin Jambor wrote:

Sorry for the delay.  I'd just like to re-iterate one comment from my
previous email which is that I do not think tree-sra.c should export
any function to the outside world apart from the entry points of the
passes (yes, there is already build_ref_for_offset which I admit is
entirely my creation but that should be moved elswhere too).

So please put sra_ipa_get_adjustment_candidate to ipa-prop.c.  I see
that it requires get_ssa_base_param to be moved there as well but
since IPA-SRA uses it in different places, it would need exporting
too, which would be weird because it does not really do anything with
parameters.  Since the function is so trivial, I would even suggest
introducing another private copy to ipa-prop.c (and leaving the
original without the new parameter).  Alternatively, you can move the
function to tree.c but for that it looks too specialized.


Note that I didn't have to move
ipa_sra_modify_function_body, since it wasn't used outside of tree-sra.c, and omp-low.c has its own version. Instead I just removed it from ipa-prop.h where I had inadvertently placed it.

OK for branch?

Attachment: curr
Description: Text document

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