[PATCHES] R600/SI: VI fixes

Marek Olšák maraeo at gmail.com
Mon Jan 12 05:24:30 PST 2015


Ping. These are critical VI fixes.

Marek

On Sat, Jan 3, 2015 at 2:11 PM, Marek Olšák <maraeo at gmail.com> wrote:
> Ping.
>
> On Mon, Dec 15, 2014 at 1:53 PM, Marek Olšák <maraeo at gmail.com> wrote:
>> An updated version is attached. I fixed the shrinking pass for opcodes
>> that don't have e32 encoding (patch 3). I'm not sure if patch 2 is a
>> good idea, I'd appreciate some opinion on that.
>>
>> The MIN3/MAX3 patch is not included here. Still don't know what aspect
>> of those opcodes should be tested. Also, I can't re-use the current
>> tests, because the store opcode that the tests use isn't implemented
>> for VI.
>>
>> Marek
>>
>> On Sun, Dec 14, 2014 at 11:02 PM, Marek Olšák <maraeo at gmail.com> wrote:
>>> The current pseudos define everything except for the encoding.
>>>
>>> The 5 attached patches replace patch 1 in the previous series.
>>> This new series unifies other VOP2 opcodes which are only available as
>>> VOP3 on VI.
>>>
>>> Marek
>>>
>>> On Sat, Dec 13, 2014 at 7:20 PM, Matt Arsenault <arsenm2 at gmail.com> wrote:
>>>>
>>>>> On Dec 13, 2014, at 5:12 AM, Marek Olšák <maraeo at gmail.com> wrote:
>>>>>
>>>>> On Fri, Dec 12, 2014 at 11:57 PM, Matt Arsenault <arsenm2 at gmail.com> wrote:
>>>>>>
>>>>>>> On Dec 12, 2014, at 4:41 PM, Marek Olšák <maraeo at gmail.com> wrote:
>>>>>>>
>>>>>>> Please review.
>>>>>>>
>>>>>>> Marek
>>>>>>
>>>>>> #1 - Why can’t this use the same encoding mapping trick the other instructions do so everything except encoding doesn’t need to see separate opcodes?
>>>>>
>>>>> The instructions are VOP2 on SI and VOP3 on VI. I can't define common
>>>>> pseudo instructions, because it's not clear whether I should set the
>>>>> VOP2 bit or the VOP3 bit or both or none.
>>>>>
>>>>> Marek
>>>>
>>>> I would expect pseudos to have neither set, but I’m not sure what the current pseudos do




More information about the llvm-commits mailing list