[Mesa-dev] [PATCH 3/3] glsl: Optimize (v.x + v.y) + (v.z + v.w) into dot(v, 1.0).

[Matt Turner]

On Tue, Jun 10, 2014 at 9:46 AM, Ian Romanick <idr at freedesktop.org> wrote:
> On 06/09/2014 02:11 PM, Matt Turner wrote:
>> Cuts five instructions out of SynMark's Gl32VSInstancing benchmark.
>> ---
>>  src/glsl/opt_algebraic.cpp | 46 ++++++++++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 46 insertions(+)
>>
>> diff --git a/src/glsl/opt_algebraic.cpp b/src/glsl/opt_algebraic.cpp
>> index d57c3e8..be65799 100644
>> --- a/src/glsl/opt_algebraic.cpp
>> +++ b/src/glsl/opt_algebraic.cpp
>> @@ -119,6 +119,44 @@ update_type(ir_expression *ir)
>>        ir->type = ir->operands[1]->type;
>>  }
>>
>> +/* Recognize (v.x + v.y) + (v.z + v.w) as dot(v, 1.0) */
>> +static ir_expression *
>> +try_replace_with_dot(ir_expression *expr0, ir_expression *expr1, void *mem_ctx)
>> +{
>> +   if (expr0 && expr0->operation == ir_binop_add &&
>> +       expr1 && expr1->operation == ir_binop_add) {
>> +      ir_swizzle *x = expr0->operands[0]->as_swizzle();
>> +      ir_swizzle *y = expr0->operands[1]->as_swizzle();
>> +      ir_swizzle *z = expr1->operands[0]->as_swizzle();
>> +      ir_swizzle *w = expr1->operands[1]->as_swizzle();
>> +
>> +      if (!x || x->mask.num_components != 1 || x->mask.has_duplicates ||
>> +          !y || y->mask.num_components != 1 || y->mask.has_duplicates ||
>> +          !z || z->mask.num_components != 1 || z->mask.has_duplicates ||
>> +          !w || w->mask.num_components != 1 || w->mask.has_duplicates) {
>
> Are the has_duplicates checks necessary?  If num_components must be 1,
> it's not obvious to me what the extra checks are doing.

Right.. I don't think so. I'll remove them.