mesa/src/compiler/glsl/lower_vec_index_to_cond_assign.cpp

/*
 * Copyright © 2010 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

/**
 * \file lower_vec_index_to_cond_assign.cpp
 *
 * Turns indexing into vector types to a series of conditional moves
 * of each channel's swizzle into a temporary.
 *
 * Most GPUs don't have a native way to do this operation, and this
 * works around that.  For drivers using both this pass and
 * ir_vec_index_to_swizzle, there's a risk that this pass will happen
 * before sufficient constant folding to find that the array index is
 * constant.  However, we hope that other optimization passes,
 * particularly constant folding of assignment conditions and copy
 * propagation, will result in the same code in the end.
 */

#include "ir.h"
#include "ir_visitor.h"
#include "ir_optimization.h"
#include "compiler/glsl_types.h"
#include "ir_builder.h"

using namespace ir_builder;

namespace {

/**
 * Visitor class for replacing expressions with ir_constant values.
 */

class ir_vec_index_to_cond_assign_visitor : public ir_hierarchical_visitor {
public:
   ir_vec_index_to_cond_assign_visitor()
      : progress(false)
   {
      /* empty */
   }

   ir_rvalue *convert_vec_index_to_cond_assign(void *mem_ctx,
                                               ir_rvalue *orig_vector,
                                               ir_rvalue *orig_index,
                                               const glsl_type *type);

   ir_rvalue *convert_vector_extract_to_cond_assign(ir_rvalue *ir);

   virtual ir_visitor_status visit_enter(ir_expression *);
   virtual ir_visitor_status visit_enter(ir_swizzle *);
   virtual ir_visitor_status visit_leave(ir_assignment *);
   virtual ir_visitor_status visit_enter(ir_return *);
   virtual ir_visitor_status visit_enter(ir_call *);
   virtual ir_visitor_status visit_enter(ir_if *);

   bool progress;
};

} /* anonymous namespace */

ir_rvalue *
ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(void *mem_ctx,
                                                                      ir_rvalue *orig_vector,
                                                                      ir_rvalue *orig_index,
                                                                      const glsl_type *type)
{
   exec_list list;
   ir_factory body(&list, base_ir);

   /* Store the index to a temporary to avoid reusing its tree. */
   assert(orig_index->type == glsl_type::int_type ||
          orig_index->type == glsl_type::uint_type);
   ir_variable *const index =
      body.make_temp(orig_index->type, "vec_index_tmp_i");

   body.emit(assign(index, orig_index));

   /* Store the value inside a temp, thus avoiding matrixes duplication */
   ir_variable *const value =
      body.make_temp(orig_vector->type, "vec_value_tmp");

   body.emit(assign(value, orig_vector));


   /* Temporary where we store whichever value we swizzle out. */
   ir_variable *const var = body.make_temp(type, "vec_index_tmp_v");

   /* Generate a single comparison condition "mask" for all of the components
    * in the vector.
    */
   ir_variable *const cond =
      compare_index_block(body, index, 0, orig_vector->type->vector_elements);

   /* Generate a conditional move of each vector element to the temp. */
   for (unsigned i = 0; i < orig_vector->type->vector_elements; i++)
      body.emit(assign(var, swizzle(value, i, 1), swizzle(cond, i, 1)));

   /* Put all of the new instructions in the IR stream before the old
    * instruction.
    */
   base_ir->insert_before(&list);

   this->progress = true;
   return deref(var).val;
}

ir_rvalue *
ir_vec_index_to_cond_assign_visitor::convert_vector_extract_to_cond_assign(ir_rvalue *ir)
{
   ir_expression *const expr = ir->as_expression();

   if (expr == NULL)
      return ir;

   if (expr->operation == ir_unop_interpolate_at_centroid ||
       expr->operation == ir_binop_interpolate_at_offset ||
       expr->operation == ir_binop_interpolate_at_sample) {
      /* Lower interpolateAtXxx(some_vec[idx], ...) to
       * interpolateAtXxx(some_vec, ...)[idx] before lowering to conditional
       * assignments, to maintain the rule that the interpolant is an l-value
       * referring to a (part of a) shader input.
       *
       * This is required when idx is dynamic (otherwise it gets lowered to
       * a swizzle).
       */
      ir_expression *const interpolant = expr->operands[0]->as_expression();
      if (!interpolant || interpolant->operation != ir_binop_vector_extract)
         return ir;

      ir_rvalue *vec_input = interpolant->operands[0];
      ir_expression *const vec_interpolate =
         new(base_ir) ir_expression(expr->operation, vec_input->type,
                                    vec_input, expr->operands[1]);

      return convert_vec_index_to_cond_assign(ralloc_parent(ir),
                                              vec_interpolate,
                                              interpolant->operands[1],
                                              ir->type);
   }

   if (expr->operation != ir_binop_vector_extract)
      return ir;

   return convert_vec_index_to_cond_assign(ralloc_parent(ir),
                                           expr->operands[0],
                                           expr->operands[1],
                                           ir->type);
}

ir_visitor_status
ir_vec_index_to_cond_assign_visitor::visit_enter(ir_expression *ir)
{
   for (unsigned i = 0; i < ir->num_operands; i++)
      ir->operands[i] = convert_vector_extract_to_cond_assign(ir->operands[i]);

   return visit_continue;
}

ir_visitor_status
ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir)
{
   /* Can't be hit from normal GLSL, since you can't swizzle a scalar (which
    * the result of indexing a vector is.  But maybe at some point we'll end up
    * using swizzling of scalars for vector construction.
    */
   ir->val = convert_vector_extract_to_cond_assign(ir->val);

   return visit_continue;
}

ir_visitor_status
ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
{
   ir->rhs = convert_vector_extract_to_cond_assign(ir->rhs);

   if (ir->condition)
      ir->condition = convert_vector_extract_to_cond_assign(ir->condition);

   return visit_continue;
}

ir_visitor_status
ir_vec_index_to_cond_assign_visitor::visit_enter(ir_call *ir)
{
   foreach_in_list_safe(ir_rvalue, param, &ir->actual_parameters) {
      ir_rvalue *new_param = convert_vector_extract_to_cond_assign(param);

      if (new_param != param) {
         param->replace_with(new_param);
      }
   }

   return visit_continue;
}

ir_visitor_status
ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir)
{
   if (ir->value)
      ir->value = convert_vector_extract_to_cond_assign(ir->value);

   return visit_continue;
}

ir_visitor_status
ir_vec_index_to_cond_assign_visitor::visit_enter(ir_if *ir)
{
   ir->condition = convert_vector_extract_to_cond_assign(ir->condition);

   return visit_continue;
}

bool
do_vec_index_to_cond_assign(exec_list *instructions)
{
   ir_vec_index_to_cond_assign_visitor v;

   visit_list_elements(&v, instructions);

   return v.progress;
}