Re: Request for code review - (ZEE patch : Redundant Zero extension elimination)

[Sriraman Tallam <tmsriram at google dot com>]

On Wed, May 12, 2010 at 11:56 AM, Andrew Pinski <pinskia@gmail.com> wrote:
> On Tue, May 11, 2010 at 6:21 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>> Hi Diego,
>>
>> ? I have made all the changes and attached the new patch.
>
> I know this most likely has been hashed before but why is this pass
> have SImode and DImode hardcoded?
> Can't you use the modes and then try a wider mode? ?Or does that need
> a total rewrite of the pass?

Yes, that requires modifying a large portion.
I have made the changes you mentioned and submiited the patch.

2010-05-12  Sriraman Tallam  <tmsriram@google.com>

	* implicit-zee.c: New file.
	* tree-pass.h (pass_implicit_zee): Declare.
	* passes.c (init_optimization_passes): Add zee pass.
	* common.opt (fzee): New flag.
	* timevar.def (TV_ZEE): Define.
	* config/i386/i386.c (optimization_options): Turn on ZEE for level 2
	and beyond.
	* Makefile.in (implicit-zee.o): Add new build file.

2010-05-12  Sriraman Tallam  <tmsriram@google.com>

	* gcc.target/i386/zee.c: New file.

>
> Also: "32-bit (SImode)" is not always true. ?SImode is 4 times QImode
> and QImode does not have to be 8bit (though GCC support for non 8bit
> has been bit-rotten).
>
> +static unsigned int
> +rest_of_handle_zee (void)
>
> Needs a comment.
>
> +static bool
> +gate_handle_zee (void)
>
> Likewise.
>
> +/* Get and Set methods for is_insn_merge_attempted */
> +
> +static enum insn_merge_code
> +get_insn_status (rtx insn)
>
>
> This comment does not help me understand what these functions are doing. ?Maybe:
> Returns the merge code status for INSN.
>
> Thanks,
> Andrew Pinski
>

Attachment: gcc_zee_patch.txt
Description: Text document