[Mesa-dev] [RFC PATCH] nir: Transform 4*x into x << 2 during late optimizations.

[Jason Ekstrand]

On Fri, May 8, 2015 at 11:11 AM, Ian Romanick <idr at freedesktop.org> wrote:
> On 05/08/2015 03:36 AM, Kenneth Graunke wrote:
>> According to Glenn, shifts on R600 have 5x the throughput as multiplies.
>>
>> Intel GPUs have strange integer multiplication restrictions - on most
>> hardware, MUL actually only does a 32-bit x 16-bit multiply.  This
>> means the arguments aren't commutative, which can limit our constant
>> propagation options.  SHL has no such restrictions.
>>
>> Shifting is probably reasonable on most people's hardware, so let's just
>> do that.
>>
>> i965 shader-db results (using NIR for VS):
>> total instructions in shared programs: 7432587 -> 7388982 (-0.59%)
>> instructions in affected programs:     1360411 -> 1316806 (-3.21%)
>> helped:                                5772
>> HURT:                                  0
>>
>> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
>> Cc: mattst88 at gmail.com
>> Cc: jason at jlekstrand.net
>> ---
>>  src/glsl/nir/nir_opt_algebraic.py | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> So...I found a bizarre issue with this patch.
>>
>>    (('imul', 4, a), ('ishl', a, 2)),
>>
>> totally optimizes things.  However...
>>
>>    (('imul', a, 4), ('ishl', a, 2)),
>>
>> doesn't seem to do anything, even though imul is commutative, and nir_search
>> should totally handle that...
>>
>>      ▄▄      ▄▄    ▄▄     ▄▄▄▄▄▄▄▄   ▄▄▄▄▄       ▄▄
>>      ██      ██   ████    ▀▀▀██▀▀▀  █▀▀▀▀██      ██
>>      ▀█▄ ██ ▄█▀   ████       ██         ▄█▀      ██
>>       ██ ██ ██   ██  ██      ██       ▄██▀       ██
>>       ███▀▀███   ██████      ██       ██         ▀▀
>>       ███  ███  ▄██  ██▄     ██       ▄▄         ▄▄
>>       ▀▀▀  ▀▀▀  ▀▀    ▀▀     ▀▀       ▀▀         ▀▀
>>
>> If you know why, let me know, otherwise I may have to look into it when more
>> awake.
>
> I've noticed a couple other weird things that I have been unable to
> understand.  Shaders like the one below end with fmul/ffma instaed of
> flrp, for example.  I understand why that happens from GLSL IR
> opt_algebraic, but it seems like nir_opt_algebraic should handle it.

Just a guess, but it's quite possibly due to the commutative
operations bug I just sent a patch for.
--Jason

> [require]
> GLSL >= 1.30
>
> [vertex shader]
> in vec4 v;
> in vec2 tc_in;
>
> out vec2 tc;
>
> void main() {
>     gl_Position = v;
>     tc = tc_in;
> }
>
> [fragment shader]
> in vec2 tc;
>
> out vec4 color;
>
> uniform sampler2D s;
> uniform float a;
> uniform vec3 base_color;
>
> void main() {
>     vec3 tex_color = texture(s, tc).xyz;
>
>     color.xyz = (base_color * a) + (tex_color * (1.0 - a));
>     color.a = 1.0;
> }
>
>
>
>> diff --git a/src/glsl/nir/nir_opt_algebraic.py b/src/glsl/nir/nir_opt_algebraic.py
>> index 400d60e..350471f 100644
>> --- a/src/glsl/nir/nir_opt_algebraic.py
>> +++ b/src/glsl/nir/nir_opt_algebraic.py
>> @@ -247,6 +247,11 @@ late_optimizations = [
>>     (('fge', ('fadd', a, b), 0.0), ('fge', a, ('fneg', b))),
>>     (('feq', ('fadd', a, b), 0.0), ('feq', a, ('fneg', b))),
>>     (('fne', ('fadd', a, b), 0.0), ('fne', a, ('fneg', b))),
>> +
>> +   # Multiplication by 4 comes up fairly often in indirect offset calculations.
>> +   # Some GPUs have weird integer multiplication limitations, but shifts should work
>> +   # equally well everywhere.
>> +   (('imul', 4, a), ('ishl', a, 2)),
>
> This should be conditionalized on whether the platform has native integers.
>
>>  ]
>>
>>  print nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render()
>>
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev