[Mesa-dev] [PATCH 04/13] i965/fs_inst: Add an is_copy_payload helper

[Jason Ekstrand]

On Tue, May 5, 2015 at 1:34 PM, Matt Turner <mattst88 at gmail.com> wrote:
> On Tue, May 5, 2015 at 1:13 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>> On Tue, Apr 14, 2015 at 1:28 AM, Kenneth Graunke <kenneth at whitecape.org> wrote:
>>> 2. CSE didn't use equals(), so it allowed things like types to differ,
>>>    or source modifiers...(which are probably bogus on LOAD_PAYLOADs
>>>    anyway, so it may be moot...)
>>>
>>>    I like using equals().  But...it might make sense to relax BAD_FILE
>>>    checks, i.e. reg_null_d and reg_null_f aren't different for practical
>>>    purposes.  (Thinking of header registers...)
>>
>> It doesn't make sense to have something that copies entirely BAD_FILE.
>
> That's not what he's saying at all.
>
> Right now, equals() called on a pair of load_payloads, one with
> reg_null_f and the other with reg_null_d as their first sources (that
> is, the header) would return false. Intuitively, that's not what we
> want.

Um... This is in is_copy_payload not CSE.  We aren't comparing two
payloads, we're comparing src[i] with src[0] for all i.  How are they
all going to be BAD_FILE?
--Jason

> Now, I don't think this is ever a problem in practice and I'd probably
> like to see it before we change fs_reg::equals.