Connor Abbott <cwabbo...@gmail.com> writes:
> On Tue, Dec 16, 2014 at 5:48 PM, Eric Anholt <e...@anholt.net> wrote:
>> Jason Ekstrand <ja...@jlekstrand.net> writes:
>>> +typedef struct {
>>> + nir_register *reg;
>>> + struct nir_src *indirect; /** < NULL for no indirect offset */
>>> + unsigned base_offset;
>>> +
>>> + /* TODO def-use chain goes here */
>>> +} nir_reg_dest;
>>
>> Aren't these TODOs resolved by the du/ud chains being in the
>> nir_register?
>
> No, those only list all the definitions and all the uses, whereas a
> hash set here would tell use which definitions each use can pick up
> and which use each definition can reach (AKA the UD and DU chains).
> That being said, the plan is to do all the optimizations in SSA, so
> unless someone has a burning need to do optimizations in non-SSA NIR
> it's pretty useless.Oh, I see. That would certainly be useful if you were doing non-SSA optimization.
signature.asc
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
