1 /*
2 * Copyright © 2010 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 lower_vector.cpp
26 * IR lowering pass to remove some types of ir_quadop_vector
27 *
28 * \author Ian Romanick <ian.d.romanick@intel.com>
29 */
30
31 #include "ir.h"
32 #include "ir_rvalue_visitor.h"
33
34 class lower_vector_visitor : public ir_rvalue_visitor {
35 public:
lower_vector_visitor()36 lower_vector_visitor() : progress(false)
37 {
38 /* empty */
39 }
40
41 void handle_rvalue(ir_rvalue **rvalue);
42
43 /**
44 * Should SWZ-like expressions be lowered?
45 */
46 bool dont_lower_swz;
47
48 bool progress;
49 };
50
51 /**
52 * Determine if an IR expression tree looks like an extended swizzle
53 *
54 * Extended swizzles consist of access of a single vector source (with possible
55 * per component negation) and the constants -1, 0, or 1.
56 */
57 bool
is_extended_swizzle(ir_expression * ir)58 is_extended_swizzle(ir_expression *ir)
59 {
60 /* Track any variables that are accessed by this expression.
61 */
62 ir_variable *var = NULL;
63
64 assert(ir->operation == ir_quadop_vector);
65
66 for (unsigned i = 0; i < ir->type->vector_elements; i++) {
67 ir_rvalue *op = ir->operands[i];
68
69 while (op != NULL) {
70 switch (op->ir_type) {
71 case ir_type_constant: {
72 const ir_constant *const c = op->as_constant();
73
74 if (!c->is_one() && !c->is_zero() && !c->is_negative_one())
75 return false;
76
77 op = NULL;
78 break;
79 }
80
81 case ir_type_dereference_variable: {
82 ir_dereference_variable *const d = (ir_dereference_variable *) op;
83
84 if ((var != NULL) && (var != d->var))
85 return false;
86
87 var = d->var;
88 op = NULL;
89 break;
90 }
91
92 case ir_type_expression: {
93 ir_expression *const ex = (ir_expression *) op;
94
95 if (ex->operation != ir_unop_neg)
96 return false;
97
98 op = ex->operands[0];
99 break;
100 }
101
102 case ir_type_swizzle:
103 op = ((ir_swizzle *) op)->val;
104 break;
105
106 default:
107 return false;
108 }
109 }
110 }
111
112 return true;
113 }
114
115 void
handle_rvalue(ir_rvalue ** rvalue)116 lower_vector_visitor::handle_rvalue(ir_rvalue **rvalue)
117 {
118 if (!*rvalue)
119 return;
120
121 ir_expression *expr = (*rvalue)->as_expression();
122 if ((expr == NULL) || (expr->operation != ir_quadop_vector))
123 return;
124
125 if (this->dont_lower_swz && is_extended_swizzle(expr))
126 return;
127
128 /* FINISHME: Is this the right thing to use for the hieralloc context?
129 */
130 void *const mem_ctx = expr;
131
132 assert(expr->type->vector_elements == expr->get_num_operands());
133
134 /* Generate a temporary with the same type as the ir_quadop_operation.
135 */
136 ir_variable *const temp =
137 new(mem_ctx) ir_variable(expr->type, "vecop_tmp", ir_var_temporary);
138
139 this->base_ir->insert_before(temp);
140
141 /* Counter of the number of components collected so far.
142 */
143 unsigned assigned;
144
145 /* Write-mask in the destination that receives counted by 'assigned'.
146 */
147 unsigned write_mask;
148
149
150 /* Generate upto four assignments to that variable. Try to group component
151 * assignments together:
152 *
153 * - All constant components can be assigned at once.
154 * - All assigments of components from a single variable with the same
155 * unary operator can be assigned at once.
156 */
157 ir_constant_data d = { { 0 } };
158
159 assigned = 0;
160 write_mask = 0;
161 for (unsigned i = 0; i < expr->type->vector_elements; i++) {
162 const ir_constant *const c = expr->operands[i]->as_constant();
163
164 if (c == NULL)
165 continue;
166
167 switch (expr->type->base_type) {
168 case GLSL_TYPE_UINT: d.u[assigned] = c->value.u[0]; break;
169 case GLSL_TYPE_INT: d.i[assigned] = c->value.i[0]; break;
170 case GLSL_TYPE_FLOAT: d.f[assigned] = c->value.f[0]; break;
171 case GLSL_TYPE_BOOL: d.b[assigned] = c->value.b[0]; break;
172 default: assert(!"Should not get here."); break;
173 }
174
175 write_mask |= (1U << i);
176 assigned++;
177 }
178
179 assert((write_mask == 0) == (assigned == 0));
180
181 /* If there were constant values, generate an assignment.
182 */
183 if (assigned > 0) {
184 ir_constant *const c =
185 new(mem_ctx) ir_constant(glsl_type::get_instance(expr->type->base_type,
186 assigned, 0),
187 &d);
188 ir_dereference *const lhs = new(mem_ctx) ir_dereference_variable(temp);
189 ir_assignment *const assign =
190 new(mem_ctx) ir_assignment(lhs, c, NULL, write_mask);
191
192 this->base_ir->insert_before(assign);
193 }
194
195 /* FINISHME: This should try to coalesce assignments.
196 */
197 for (unsigned i = 0; i < expr->type->vector_elements; i++) {
198 if (expr->operands[i]->ir_type == ir_type_constant)
199 continue;
200
201 ir_dereference *const lhs = new(mem_ctx) ir_dereference_variable(temp);
202 ir_assignment *const assign =
203 new(mem_ctx) ir_assignment(lhs, expr->operands[i], NULL, (1U << i));
204
205 this->base_ir->insert_before(assign);
206 assigned++;
207 }
208
209 assert(assigned == expr->type->vector_elements);
210
211 *rvalue = new(mem_ctx) ir_dereference_variable(temp);
212 this->progress = true;
213 }
214
215 bool
lower_quadop_vector(exec_list * instructions,bool dont_lower_swz)216 lower_quadop_vector(exec_list *instructions, bool dont_lower_swz)
217 {
218 lower_vector_visitor v;
219
220 v.dont_lower_swz = dont_lower_swz;
221 visit_list_elements(&v, instructions);
222
223 return v.progress;
224 }
225