[Mesa-dev] [PATCH v2 25/28] glsl/lower_instructions: add double lowering passes

[Ilia Mirkin]

On Fri, Feb 6, 2015 at 4:21 AM, Ian Romanick <idr at freedesktop.org> wrote:
> On 02/06/2015 06:56 AM, Ilia Mirkin wrote:
>> From: Dave Airlie <airlied at gmail.com>
>>
>> This lowers double dot product and lrp to fma.
>>
>> Signed-off-by: Dave Airlie <airlied at redhat.com>
>> ---
>>  src/glsl/lower_instructions.cpp | 83 +++++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 83 insertions(+)
>>
>> diff --git a/src/glsl/lower_instructions.cpp b/src/glsl/lower_instructions.cpp
>> index 5c1d6aa..140b6d4 100644
>> --- a/src/glsl/lower_instructions.cpp
>> +++ b/src/glsl/lower_instructions.cpp
>> @@ -115,6 +115,7 @@
>>   */
>>
>>  #include "main/core.h" /* for M_LOG2E */
>> +#include "program/prog_instruction.h" /* for swizzle */
>>  #include "glsl_types.h"
>>  #include "ir.h"
>>  #include "ir_builder.h"
>> @@ -148,6 +149,8 @@ private:
>>     void carry_to_arith(ir_expression *);
>>     void borrow_to_arith(ir_expression *);
>>     void sat_to_clamp(ir_expression *);
>> +   void double_dot_to_fma(ir_expression *);
>> +   void double_lrp(ir_expression *);
>>  };
>>
>>  } /* anonymous namespace */
>> @@ -521,10 +524,90 @@ lower_instructions_visitor::sat_to_clamp(ir_expression *ir)
>>     this->progress = true;
>>  }
>>
>> +void
>> +lower_instructions_visitor::double_dot_to_fma(ir_expression *ir)
>> +{
>> +   ir_variable *temp = new(ir) ir_variable(ir->operands[0]->type->get_base_type(), "dot_res",
>> +                                        ir_var_temporary);
>> +   this->base_ir->insert_before(temp);
>> +
>> +   int nc = ir->operands[0]->type->components();
>> +   for (int i = nc - 1; i >= 1; i--) {
>> +      ir_assignment *assig;
>> +      if (i == (nc - 1)) {
>> +         assig = assign(temp, mul(swizzle(ir->operands[0]->clone(ir, NULL), i, 1),
>> +                                  swizzle(ir->operands[1]->clone(ir, NULL), i, 1)));
>> +      } else {
>> +         assig = assign(temp, fma(swizzle(ir->operands[0]->clone(ir, NULL), i, 1),
>> +                                  swizzle(ir->operands[1]->clone(ir, NULL), i, 1),
>> +                                  temp));
>> +      }
>> +      this->base_ir->insert_before(assig);
>> +   }
>> +
>> +   ir->operation = ir_triop_fma;
>> +   ir->operands[0] = swizzle(ir->operands[0], 0, 1);
>> +   ir->operands[1] = swizzle(ir->operands[1], 0, 1);
>> +   ir->operands[2] = new(ir) ir_dereference_variable(temp);
>> +
>> +   this->progress = true;
>> +
>> +}
>> +
>> +void
>> +lower_instructions_visitor::double_lrp(ir_expression *ir)
>> +{
>> +   ir_assignment *assig;
>> +   ir_constant *one = new(ir) ir_constant(1.0, ir->operands[2]->type->vector_elements);
>> +   ir_variable *temp = new(ir) ir_variable(ir->operands[0]->type, "lrp_res",
>> +                                        ir_var_temporary);
>> +   ir_variable *t2 = new(ir) ir_variable(ir->operands[0]->type, "aval",
>> +                                        ir_var_temporary);
>> +   int swizval;
>> +   this->base_ir->insert_before(temp);
>> +   this->base_ir->insert_before(t2);
>> +
>> +   assig = assign(temp, mul(sub(one, ir->operands[2]), ir->operands[0]));
>> +   this->base_ir->insert_before(assig);
>> +
>> +   switch (ir->operands[2]->type->vector_elements) {
>> +   case 1:
>> +      swizval = SWIZZLE_XXXX;
>> +      break;
>> +   case 2:
>> +      swizval = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_X, SWIZZLE_X);
>> +      break;
>> +   case 3:
>> +      swizval = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X);
>> +      break;
>> +   case 4:
>> +   default:
>> +      swizval = SWIZZLE_XYZW;
>> +      break;
>> +   }
>
> Isn't this just swizzle_for_size(ir->operands[2], 4) from ir_builder?

Not quite. swizzle_for_size has

   if (a.val->type->vector_elements < components)
      components = a.val->type->vector_elements;

Which messes things up here. We want to always have the same number of
components... there are 3 mix variants:

      genDType mix(genDType x, genDType y, genDType a);
      genDType mix(genDType x, genDType y, double a);
      genDType mix(genDType x, genDType y, genBType a);

and ir->operands[2] will only have 1 component in the double case.


>
>> +   assig = assign(t2, swizzle(ir->operands[2], swizval, ir->operands[0]->type->vector_elements));
>> +   this->base_ir->insert_before(assig);
>> +
>> +   ir->operation = ir_triop_fma;
>> +   ir->operands[0] = new(ir) ir_dereference_variable(t2);
>> +   ir->operands[1] = ir->operands[1];
>> +   ir->operands[2] = new(ir) ir_dereference_variable(temp);
>> +
>> +   this->progress = true;
>> +}
>> +
>>  ir_visitor_status
>>  lower_instructions_visitor::visit_leave(ir_expression *ir)
>>  {
>>     switch (ir->operation) {
>> +   case ir_binop_dot:
>> +      if (ir->operands[0]->type->is_double())
>> +         double_dot_to_fma(ir);
>> +      break;
>> +   case ir_triop_lrp:
>> +      if (ir->operands[0]->type->is_double())
>> +         double_lrp(ir);
>> +      break;
>>     case ir_binop_sub:
>>        if (lowering(SUB_TO_ADD_NEG))
>>        sub_to_add_neg(ir);
>>
>