[Mesa-dev] [PATCH] gallium: remove TGSI_OPCODE_KILL

[Marek Olšák]

On Fri, Oct 6, 2017 at 1:24 AM, Roland Scheidegger <sroland at vmware.com> wrote:
> Am 06.10.2017 um 00:34 schrieb Marek Olšák:
>> On Thu, Oct 5, 2017 at 8:00 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>> On Thu, Oct 5, 2017 at 1:45 PM, Marek Olšák <maraeo at gmail.com> wrote:
>>>> On Thu, Oct 5, 2017 at 4:45 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>>>> Like Roland, I don't see what's being gained. It's more code
>>>>> everywhere and harder to read the TGSI directly.
>>>>
>>>> Harder to read the "IMM" string in the operand?
>>>
>>> Yeah. You see a KILL_IF IMM[]... and then have to look up what that
>>> immediate is among the 75 other immediates. For all you know it's not
>>> -1 but something else. For long programs, this isn't great, esp if
>>> you're matching things up to the GLSL.
>>>
>>>>
>>>> "more code everywhere"?
>>>> 21 files changed, 22 insertions(+), 200 deletions(-)
>>>>
>>>> This is the reason: 200 deletions(-)
>>>>
>>>>>
>>>>> Neither nv30 nor nv50+ do the checks you suggest.
>>>>
>>>> But they can if they want.
>>>
>>> ... thus leading to more code.
>>>
>>> Anyways, since my (recent) ability to contribute has greatly
>>> atrophied, I'm not going to explicitly NAK this. But you have my 2c.
>>
>> I don't really understand the backlash, and especially from nouveau it
>> doesn't make sense. It's just a removal of a redundant instruction
>> that simplifies driver code for me. So far nobody asked me to check
>> IMM in drivers and keep the old codepath there. Yes, KILL_IF is silly,
>> but it's a superset of both.
>>
>
> MAD is a superset of MUL and ADD too.
> (Ok maybe with precise that's not entirely true anymore, but I'm sure
> you get the point.)

Yes, that's why r300 doesn't have MUL and ADD.

Even though removing MUL and ADD might make sense in theory, I have no
interest in that.

Marek