[Mesa-dev] [PATCH 2/5] nvc0/ir: use levelZero flag when the lod is set to 0

[Ilia Mirkin]

Right, so arg+1, -1 is the right thing. However, this patch is bogus.
It relies on getImmediate which in turn relies on getUniqueInsn, which
in turn doesn't work pre-ssa (except in very contrived scenarios). So
I'm going to try to put together a series that resolves many of these
things, like making getImmediate safe pre-ssa, and making it more
likely to succeed pre-ssa as well (by doing a semi-kinda-sorta SSA
pass as we're going from TGSI -> nv50 ir - effectively undoing the
stupid renumbering thing st/mesa does).

On Fri, Oct 21, 2016 at 9:59 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> Yeah, it might actually need to be arg+1, -1. I didn't test this patch
> too thoroughly...
>
> On Fri, Oct 21, 2016 at 9:53 AM, Samuel Pitoiset
> <samuel.pitoiset at gmail.com> wrote:
>> This patch breaks a bunch of piglit tests, see a short list below:
>>
>> bin/arb_texture_barrier-blending-in-shader 512 42 1 128 7 -auto -fbo
>> bin/arb_texture_buffer_object-formats vs core -auto -fbo
>> bin/texelFetch 140 vs sampler2DRect -auto -fbo
>> bin/mesa_pack_invert-readpixels -auto -fbo
>> ...
>>
>> Around 150 regressions.
>>
>> I suspect the moveSources() to be wrong just because texture arguments are
>> crazy. :-)
>>
>> On 10/21/2016 08:30 AM, Ilia Mirkin wrote:
>>>
>>> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
>>> ---
>>>  src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 9
>>> +++++++++
>>>  1 file changed, 9 insertions(+)
>>>
>>> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>>> b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>>> index 68f2b15..4181422 100644
>>> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>>> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>>> @@ -662,6 +662,15 @@ NVC0LoweringPass::handleTEX(TexInstruction *i)
>>>        }
>>>     }
>>>
>>> +   ImmediateValue lod;
>>> +   if (!i->tex.target.isMS() && (i->op == OP_TXL || i->op == OP_TXF) &&
>>> +       i->src(arg).getImmediate(lod) && lod.isInteger(0)) {
>>> +      if (i->op == OP_TXL)
>>> +         i->op = OP_TEX;
>>> +      i->tex.levelZero = true;
>>> +      i->moveSources(arg, -1);
>>> +   }
>>> +
>>>     // Arguments to the TEX instruction are a little insane. Even though
>>> the
>>>     // encoding is identical between SM20 and SM30, the arguments mean
>>>     // different things between Fermi and Kepler+. A lot of arguments are
>>>
>>
>> --
>> -Samuel