[Mesa-dev] [PATCH 10/12] nir: Add lowering support for unpacking opcodes.

[Matt Turner]

On Thu, Jan 28, 2016 at 12:38 AM, Iago Toral <itoral at igalia.com> wrote:
> On Mon, 2016-01-25 at 15:18 -0800, Matt Turner wrote:
>> ---
>>  src/glsl/nir/nir.h                |  4 ++++
>>  src/glsl/nir/nir_opt_algebraic.py | 28 ++++++++++++++++++++++++++++
>>  2 files changed, 32 insertions(+)
>>
>> diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
>> index bbd5b1a..3b90b51 100644
>> --- a/src/glsl/nir/nir.h
>> +++ b/src/glsl/nir/nir.h
>> @@ -1474,6 +1474,10 @@ typedef struct nir_shader_compiler_options {
>>     bool lower_pack_unorm_4x8;
>>     bool lower_pack_snorm_4x8;
>>     bool lower_unpack_half_2x16;
>> +   bool lower_unpack_unorm_2x16;
>> +   bool lower_unpack_snorm_2x16;
>> +   bool lower_unpack_unorm_4x8;
>> +   bool lower_unpack_snorm_4x8;
>>
>>     bool lower_extract_byte;
>>     bool lower_extract_word;
>> diff --git a/src/glsl/nir/nir_opt_algebraic.py b/src/glsl/nir/nir_opt_algebraic.py
>> index 56b0f5e..a0d6c07 100644
>> --- a/src/glsl/nir/nir_opt_algebraic.py
>> +++ b/src/glsl/nir/nir_opt_algebraic.py
>> @@ -278,6 +278,34 @@ optimizations = [
>>       ('pack_uvec4_to_uint',
>>          ('f2i', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 127.0)))),
>>       'options->lower_pack_snorm_4x8'),
>> +
>> +    (('unpack_unorm_2x16', 'v'),
>> +     ('fdiv', ('u2f', ('vec4', ('extract_uword', 'v', 0),
>> +                               ('extract_uword', 'v', 1), 0, 0)),
>> +              65535.0),
>> +     'options->lower_unpack_unorm_2x16'),
>
> This should return a vec2 not a vec4, right?

Yes (and I believe it does)

Is the vec4(..., ..., 0, 0) the cause of concern? The zeros are
unused, and I'm not sure how to generate a plain vec2 here -- NIR only
has a vec4 operation.