Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * Copyright © 2013 Intel Corporation
  3.  *
  4.  * Permission is hereby granted, free of charge, to any person obtaining a
  5.  * copy of this software and associated documentation files (the "Software"),
  6.  * to deal in the Software without restriction, including without limitation
  7.  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  8.  * and/or sell copies of the Software, and to permit persons to whom the
  9.  * Software is furnished to do so, subject to the following conditions:
  10.  *
  11.  * The above copyright notice and this permission notice (including the next
  12.  * paragraph) shall be included in all copies or substantial portions of the
  13.  * Software.
  14.  *
  15.  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16.  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17.  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18.  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19.  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20.  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21.  * DEALINGS IN THE SOFTWARE.
  22.  */
  23.  
  24. /**
  25.  * \file brw_lower_unnormalized_offset.cpp
  26.  *
  27.  * IR lower pass to convert a texture offset into an adjusted coordinate,
  28.  * for use with unnormalized coordinates. At least the gather4* messages
  29.  * on Ivybridge and Haswell make a mess with nonzero offsets.
  30.  *
  31.  * \author Chris Forbes <chrisf@ijw.co.nz>
  32.  */
  33.  
  34. #include "glsl/glsl_types.h"
  35. #include "glsl/ir.h"
  36. #include "glsl/ir_builder.h"
  37.  
  38. using namespace ir_builder;
  39.  
  40. class brw_lower_unnormalized_offset_visitor : public ir_hierarchical_visitor {
  41. public:
  42.    brw_lower_unnormalized_offset_visitor()
  43.    {
  44.       progress = false;
  45.    }
  46.  
  47.    ir_visitor_status visit_leave(ir_texture *ir);
  48.  
  49.    bool progress;
  50. };
  51.  
  52. ir_visitor_status
  53. brw_lower_unnormalized_offset_visitor::visit_leave(ir_texture *ir)
  54. {
  55.    if (!ir->offset)
  56.       return visit_continue;
  57.  
  58.    if (ir->op == ir_tg4 || ir->op == ir_tex) {
  59.       if (ir->sampler->type->sampler_dimensionality != GLSL_SAMPLER_DIM_RECT)
  60.          return visit_continue;
  61.    }
  62.    else if (ir->op != ir_txf) {
  63.       return visit_continue;
  64.    }
  65.  
  66.    void *mem_ctx = ralloc_parent(ir);
  67.  
  68.    if (ir->op == ir_txf) {
  69.       /* It appears that the ld instruction used for txf does its
  70.        * address bounds check before adding in the offset.  To work
  71.        * around this, just add the integer offset to the integer texel
  72.        * coordinate, and don't put the offset in the header.
  73.        */
  74.       ir_variable *var = new(mem_ctx) ir_variable(ir->coordinate->type,
  75.                                                   "coordinate",
  76.                                                   ir_var_temporary);
  77.       base_ir->insert_before(var);
  78.       base_ir->insert_before(assign(var, ir->coordinate));
  79.       base_ir->insert_before(assign(var,
  80.                add(swizzle_for_size(var, ir->offset->type->vector_elements), ir->offset),
  81.                (1 << ir->offset->type->vector_elements) - 1));
  82.  
  83.       ir->coordinate = new(mem_ctx) ir_dereference_variable(var);
  84.    } else {
  85.       ir->coordinate = add(ir->coordinate, i2f(ir->offset));
  86.    }
  87.  
  88.    ir->offset = NULL;
  89.  
  90.    progress = true;
  91.    return visit_continue;
  92. }
  93.  
  94. extern "C" {
  95.  
  96. bool
  97. brw_do_lower_unnormalized_offset(exec_list *instructions)
  98. {
  99.    brw_lower_unnormalized_offset_visitor v;
  100.  
  101.    visit_list_elements(&v, instructions);
  102.  
  103.    return v.progress;
  104. }
  105.  
  106. }
  107.