1b145e903694fa932ab1e0d955e889555193ab604Eric Anholt/* 2b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * Copyright © 2010 Intel Corporation 3b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * 4b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * Permission is hereby granted, free of charge, to any person obtaining a 5b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * copy of this software and associated documentation files (the "Software"), 6b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * to deal in the Software without restriction, including without limitation 7b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * and/or sell copies of the Software, and to permit persons to whom the 9b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * Software is furnished to do so, subject to the following conditions: 10b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * 11b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * The above copyright notice and this permission notice (including the next 12b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * paragraph) shall be included in all copies or substantial portions of the 13b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * Software. 14b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * 15b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 21b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * DEALINGS IN THE SOFTWARE. 22b145e903694fa932ab1e0d955e889555193ab604Eric Anholt */ 23b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 24b145e903694fa932ab1e0d955e889555193ab604Eric Anholt/** 25df883eb1575a740bf91e01cbe2eaa4dbc1f9f154Chad Versace * \file lower_vec_index_to_swizzle.cpp 26b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * 27b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * Turns constant indexing into vector types to swizzles. This will 28b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * let other swizzle-aware optimization passes catch these constructs, 29b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * and codegen backends not have to worry about this case. 30b145e903694fa932ab1e0d955e889555193ab604Eric Anholt */ 31b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 32b145e903694fa932ab1e0d955e889555193ab604Eric Anholt#include "ir.h" 33b145e903694fa932ab1e0d955e889555193ab604Eric Anholt#include "ir_visitor.h" 34b145e903694fa932ab1e0d955e889555193ab604Eric Anholt#include "ir_optimization.h" 35b145e903694fa932ab1e0d955e889555193ab604Eric Anholt#include "glsl_types.h" 366f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick#include "main/macros.h" 37b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 38b145e903694fa932ab1e0d955e889555193ab604Eric Anholt/** 39b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * Visitor class for replacing expressions with ir_constant values. 40b145e903694fa932ab1e0d955e889555193ab604Eric Anholt */ 41b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 42b145e903694fa932ab1e0d955e889555193ab604Eric Anholtclass ir_vec_index_to_swizzle_visitor : public ir_hierarchical_visitor { 43b145e903694fa932ab1e0d955e889555193ab604Eric Anholtpublic: 44b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_vec_index_to_swizzle_visitor() 45b145e903694fa932ab1e0d955e889555193ab604Eric Anholt { 46b145e903694fa932ab1e0d955e889555193ab604Eric Anholt progress = false; 47b145e903694fa932ab1e0d955e889555193ab604Eric Anholt } 48b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 49b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_rvalue *convert_vec_index_to_swizzle(ir_rvalue *val); 50b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 51b145e903694fa932ab1e0d955e889555193ab604Eric Anholt virtual ir_visitor_status visit_enter(ir_expression *); 52b145e903694fa932ab1e0d955e889555193ab604Eric Anholt virtual ir_visitor_status visit_enter(ir_swizzle *); 53b145e903694fa932ab1e0d955e889555193ab604Eric Anholt virtual ir_visitor_status visit_enter(ir_assignment *); 54b145e903694fa932ab1e0d955e889555193ab604Eric Anholt virtual ir_visitor_status visit_enter(ir_return *); 55b145e903694fa932ab1e0d955e889555193ab604Eric Anholt virtual ir_visitor_status visit_enter(ir_call *); 56b145e903694fa932ab1e0d955e889555193ab604Eric Anholt virtual ir_visitor_status visit_enter(ir_if *); 57b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 58b145e903694fa932ab1e0d955e889555193ab604Eric Anholt bool progress; 59b145e903694fa932ab1e0d955e889555193ab604Eric Anholt}; 60b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 61b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_rvalue * 62b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::convert_vec_index_to_swizzle(ir_rvalue *ir) 63b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 64b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_dereference_array *deref = ir->as_dereference_array(); 65b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_constant *ir_constant; 66b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 67b145e903694fa932ab1e0d955e889555193ab604Eric Anholt if (!deref) 68b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return ir; 69b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 70b145e903694fa932ab1e0d955e889555193ab604Eric Anholt if (deref->array->type->is_matrix() || deref->array->type->is_array()) 71b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return ir; 72b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 73b145e903694fa932ab1e0d955e889555193ab604Eric Anholt assert(deref->array_index->type->base_type == GLSL_TYPE_INT); 74b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_constant = deref->array_index->constant_expression_value(); 75b145e903694fa932ab1e0d955e889555193ab604Eric Anholt if (!ir_constant) 76b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return ir; 77b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 78d3073f58c17d8675a2ecdd5dfa83e5520c78e1a8Kenneth Graunke void *ctx = ralloc_parent(ir); 79b145e903694fa932ab1e0d955e889555193ab604Eric Anholt this->progress = true; 806f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick 816f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick /* Page 40 of the GLSL 1.20 spec says: 826f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * 836f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * "When indexing with non-constant expressions, behavior is undefined 846f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * if the index is negative, or greater than or equal to the size of 856f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * the vector." 866f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * 876f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * The quoted spec text mentions non-constant expressions, but this code 886f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * operates on constants. These constants are the result of non-constant 896f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * expressions that have been optimized to constants. The common case here 906f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * is a loop counter from an unrolled loop that is used to index a vector. 916f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * 926f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * The ir_swizzle constructor gets angry if the index is negative or too 936f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick * large. For simplicity sake, just clamp the index to [0, size-1]. 946f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick */ 956f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick const int i = MIN2(MAX2(ir_constant->value.i[0], 0), 966f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick (deref->array->type->vector_elements - 1)); 976f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick 986f5c73797087c6e7842665f84e41caedea59bb65Ian Romanick return new(ctx) ir_swizzle(deref->array, i, 0, 0, 0, 1); 99b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 100b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 101b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_visitor_status 102b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::visit_enter(ir_expression *ir) 103b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 104b145e903694fa932ab1e0d955e889555193ab604Eric Anholt unsigned int i; 105b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 106b145e903694fa932ab1e0d955e889555193ab604Eric Anholt for (i = 0; i < ir->get_num_operands(); i++) { 107b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir->operands[i] = convert_vec_index_to_swizzle(ir->operands[i]); 108b145e903694fa932ab1e0d955e889555193ab604Eric Anholt } 109b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 110b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return visit_continue; 111b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 112b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 113b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_visitor_status 114b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::visit_enter(ir_swizzle *ir) 115b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 116b145e903694fa932ab1e0d955e889555193ab604Eric Anholt /* Can't be hit from normal GLSL, since you can't swizzle a scalar (which 117b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * the result of indexing a vector is. But maybe at some point we'll end up 118b145e903694fa932ab1e0d955e889555193ab604Eric Anholt * using swizzling of scalars for vector construction. 119b145e903694fa932ab1e0d955e889555193ab604Eric Anholt */ 120b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir->val = convert_vec_index_to_swizzle(ir->val); 121b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 122b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return visit_continue; 123b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 124b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 125b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_visitor_status 126b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::visit_enter(ir_assignment *ir) 127b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 1285a7758efbe14dee026245a4f4f4fb3ccf7b2c23bIan Romanick ir->set_lhs(convert_vec_index_to_swizzle(ir->lhs)); 129b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir->rhs = convert_vec_index_to_swizzle(ir->rhs); 130b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 131b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return visit_continue; 132b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 133b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 134b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_visitor_status 135b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::visit_enter(ir_call *ir) 136b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 137b145e903694fa932ab1e0d955e889555193ab604Eric Anholt foreach_iter(exec_list_iterator, iter, *ir) { 138b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_rvalue *param = (ir_rvalue *)iter.get(); 139b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_rvalue *new_param = convert_vec_index_to_swizzle(param); 140b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 141b145e903694fa932ab1e0d955e889555193ab604Eric Anholt if (new_param != param) { 142c7a18da69022d3f9b05c21ff2473e8ea390f77f1Kenneth Graunke param->replace_with(new_param); 143b145e903694fa932ab1e0d955e889555193ab604Eric Anholt } 144b145e903694fa932ab1e0d955e889555193ab604Eric Anholt } 145b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 146b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return visit_continue; 147b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 148b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 149b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_visitor_status 150b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::visit_enter(ir_return *ir) 151b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 152b145e903694fa932ab1e0d955e889555193ab604Eric Anholt if (ir->value) { 153b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir->value = convert_vec_index_to_swizzle(ir->value); 154b145e903694fa932ab1e0d955e889555193ab604Eric Anholt } 155b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 156b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return visit_continue; 157b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 158b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 159b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_visitor_status 160b145e903694fa932ab1e0d955e889555193ab604Eric Anholtir_vec_index_to_swizzle_visitor::visit_enter(ir_if *ir) 161b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 162b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir->condition = convert_vec_index_to_swizzle(ir->condition); 163b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 164b145e903694fa932ab1e0d955e889555193ab604Eric Anholt return visit_continue; 165b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 166b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 167b145e903694fa932ab1e0d955e889555193ab604Eric Anholtbool 168b145e903694fa932ab1e0d955e889555193ab604Eric Anholtdo_vec_index_to_swizzle(exec_list *instructions) 169b145e903694fa932ab1e0d955e889555193ab604Eric Anholt{ 170b145e903694fa932ab1e0d955e889555193ab604Eric Anholt ir_vec_index_to_swizzle_visitor v; 171b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 172b145e903694fa932ab1e0d955e889555193ab604Eric Anholt v.run(instructions); 173b145e903694fa932ab1e0d955e889555193ab604Eric Anholt 174022f79e49648d465d2db0240554f58ac42754584Eric Anholt return v.progress; 175b145e903694fa932ab1e0d955e889555193ab604Eric Anholt} 176