glsl2: Add optimization pass for algebraic simplifications.
[mesa.git] / src / glsl / ir_algebraic.cpp
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 ir_algebraic.cpp
26 *
27 * Takes advantage of association, commutivity, and other algebraic
28 * properties to simplify expressions.
29 */
30
31 #include "ir.h"
32 #include "ir_visitor.h"
33 #include "ir_optimization.h"
34 #include "glsl_types.h"
35
36 /**
37 * Visitor class for replacing expressions with ir_constant values.
38 */
39
40 class ir_algebraic_visitor : public ir_hierarchical_visitor {
41 public:
42 ir_algebraic_visitor()
43 {
44 this->progress = false;
45 }
46
47 virtual ~ir_algebraic_visitor()
48 {
49 }
50
51 virtual ir_visitor_status visit_leave(ir_assignment *);
52 virtual ir_visitor_status visit_leave(ir_call *);
53 virtual ir_visitor_status visit_leave(ir_dereference_array *);
54 virtual ir_visitor_status visit_leave(ir_expression *);
55 virtual ir_visitor_status visit_leave(ir_if *);
56 virtual ir_visitor_status visit_leave(ir_return *);
57 virtual ir_visitor_status visit_leave(ir_swizzle *);
58 virtual ir_visitor_status visit_leave(ir_texture *);
59
60 ir_rvalue *handle_expression(ir_rvalue *in_ir);
61
62 bool progress;
63 };
64
65 static bool
66 is_vec_zero(ir_constant *ir)
67 {
68 int c;
69
70 if (!ir)
71 return false;
72 if (!ir->type->is_scalar() &&
73 !ir->type->is_vector())
74 return false;
75
76 for (c = 0; c < ir->type->vector_elements; c++) {
77 switch (ir->type->base_type) {
78 case GLSL_TYPE_FLOAT:
79 if (ir->value.f[c] != 0.0)
80 return false;
81 break;
82 case GLSL_TYPE_INT:
83 if (ir->value.i[c] != 0)
84 return false;
85 break;
86 case GLSL_TYPE_UINT:
87 if (ir->value.u[c] != 0)
88 return false;
89 break;
90 case GLSL_TYPE_BOOL:
91 if (ir->value.b[c] != false)
92 return false;
93 break;
94 default:
95 assert(!"bad base type");
96 return false;
97 }
98 }
99
100 return true;
101 }
102
103 static bool
104 is_vec_one(ir_constant *ir)
105 {
106 int c;
107
108 if (!ir)
109 return false;
110 if (!ir->type->is_scalar() &&
111 !ir->type->is_vector())
112 return false;
113
114 for (c = 0; c < ir->type->vector_elements; c++) {
115 switch (ir->type->base_type) {
116 case GLSL_TYPE_FLOAT:
117 if (ir->value.f[c] != 1.0)
118 return false;
119 break;
120 case GLSL_TYPE_INT:
121 if (ir->value.i[c] != 1)
122 return false;
123 break;
124 case GLSL_TYPE_UINT:
125 if (ir->value.u[c] != 1)
126 return false;
127 break;
128 case GLSL_TYPE_BOOL:
129 if (ir->value.b[c] != true)
130 return false;
131 break;
132 default:
133 assert(!"bad base type");
134 return false;
135 }
136 }
137
138 return true;
139 }
140
141 ir_rvalue *
142 ir_algebraic_visitor::handle_expression(ir_rvalue *in_ir)
143 {
144 ir_expression *ir = (ir_expression *)in_ir;
145 ir_constant *op_const[2] = {NULL, NULL};
146 ir_expression *op_expr[2] = {NULL, NULL};
147 unsigned int i;
148
149 if (!in_ir)
150 return NULL;
151
152 if (in_ir->ir_type != ir_type_expression)
153 return in_ir;
154
155 for (i = 0; i < ir->get_num_operands(); i++) {
156 if (ir->operands[i]->type->is_matrix())
157 return in_ir;
158
159 op_const[i] = ir->operands[i]->constant_expression_value();
160 op_expr[i] = ir->operands[i]->as_expression();
161 }
162
163 switch (ir->operation) {
164 case ir_unop_logic_not:
165 if (op_expr[0] && op_expr[0]->operation == ir_binop_equal) {
166 this->progress = true;
167 return new(ir) ir_expression(ir_binop_nequal,
168 ir->type,
169 op_expr[0]->operands[0],
170 op_expr[0]->operands[1]);
171 }
172 if (op_expr[0] && op_expr[0]->operation == ir_binop_nequal) {
173 this->progress = true;
174 return new(ir) ir_expression(ir_binop_equal,
175 ir->type,
176 op_expr[0]->operands[0],
177 op_expr[0]->operands[1]);
178 }
179 break;
180
181 case ir_binop_add:
182 if (is_vec_zero(op_const[0])) {
183 this->progress = true;
184 return ir->operands[1];
185 }
186 if (is_vec_zero(op_const[1])) {
187 this->progress = true;
188 return ir->operands[0];
189 }
190 break;
191
192 case ir_binop_sub:
193 if (is_vec_zero(op_const[0])) {
194 this->progress = true;
195 return new(ir) ir_expression(ir_unop_neg,
196 ir->type,
197 ir->operands[1],
198 NULL);
199 }
200 if (is_vec_zero(op_const[1])) {
201 this->progress = true;
202 return ir->operands[0];
203 }
204 break;
205
206 case ir_binop_mul:
207 if (is_vec_one(op_const[0])) {
208 this->progress = true;
209 return ir->operands[1];
210 }
211 if (is_vec_one(op_const[1])) {
212 this->progress = true;
213 return ir->operands[0];
214 }
215
216 if (is_vec_zero(op_const[0]) ||
217 is_vec_zero(op_const[1])) {
218 ir_constant_data zero_data;
219 memset(&zero_data, 0, sizeof(zero_data));
220
221 this->progress = true;
222 return new(ir) ir_constant(ir->type, &zero_data);
223 }
224 break;
225
226 case ir_binop_div:
227 if (is_vec_one(op_const[0]) && ir->type->base_type == GLSL_TYPE_FLOAT) {
228 this->progress = true;
229 return new(ir) ir_expression(ir_unop_rcp,
230 ir->type,
231 ir->operands[1],
232 NULL);
233 }
234 if (is_vec_one(op_const[1])) {
235 this->progress = true;
236 return ir->operands[0];
237 }
238 break;
239
240 case ir_unop_rcp:
241 if (op_expr[0] && op_expr[0]->operation == ir_unop_rcp) {
242 this->progress = true;
243 return op_expr[0]->operands[0];
244 }
245
246 /* FINISHME: We should do rcp(rsq(x)) -> sqrt(x) for some
247 * backends, except that some backends will have done sqrt ->
248 * rcp(rsq(x)) and we don't want to undo it for them.
249 */
250
251 /* As far as we know, all backends are OK with rsq. */
252 if (op_expr[0] && op_expr[0]->operation == ir_unop_sqrt) {
253 this->progress = true;
254 return new(ir) ir_expression(ir_unop_rsq,
255 ir->type,
256 op_expr[0]->operands[0],
257 NULL);
258 }
259
260 break;
261
262 default:
263 break;
264 }
265
266 return in_ir;
267 }
268
269 ir_visitor_status
270 ir_algebraic_visitor::visit_leave(ir_expression *ir)
271 {
272 unsigned int operand;
273
274 for (operand = 0; operand < ir->get_num_operands(); operand++) {
275 ir->operands[operand] = handle_expression(ir->operands[operand]);
276 }
277
278 return visit_continue;
279 }
280
281 ir_visitor_status
282 ir_algebraic_visitor::visit_leave(ir_texture *ir)
283 {
284 ir->coordinate = handle_expression(ir->coordinate);
285 ir->projector = handle_expression(ir->projector);
286 ir->shadow_comparitor = handle_expression(ir->shadow_comparitor);
287
288 switch (ir->op) {
289 case ir_tex:
290 break;
291 case ir_txb:
292 ir->lod_info.bias = handle_expression(ir->lod_info.bias);
293 break;
294 case ir_txf:
295 case ir_txl:
296 ir->lod_info.lod = handle_expression(ir->lod_info.lod);
297 break;
298 case ir_txd:
299 ir->lod_info.grad.dPdx = handle_expression(ir->lod_info.grad.dPdx);
300 ir->lod_info.grad.dPdy = handle_expression(ir->lod_info.grad.dPdy);
301 break;
302 }
303
304 return visit_continue;
305 }
306
307 ir_visitor_status
308 ir_algebraic_visitor::visit_leave(ir_swizzle *ir)
309 {
310 ir->val = handle_expression(ir->val);
311 return visit_continue;
312 }
313
314 ir_visitor_status
315 ir_algebraic_visitor::visit_leave(ir_dereference_array *ir)
316 {
317 ir->array_index = handle_expression(ir->array_index);
318 return visit_continue;
319 }
320
321 ir_visitor_status
322 ir_algebraic_visitor::visit_leave(ir_assignment *ir)
323 {
324 ir->rhs = handle_expression(ir->rhs);
325 ir->condition = handle_expression(ir->condition);
326 return visit_continue;
327 }
328
329 ir_visitor_status
330 ir_algebraic_visitor::visit_leave(ir_call *ir)
331 {
332 foreach_iter(exec_list_iterator, iter, *ir) {
333 ir_rvalue *param = (ir_rvalue *)iter.get();
334 ir_rvalue *new_param = handle_expression(param);
335
336 if (new_param != param) {
337 param->replace_with(new_param);
338 }
339 }
340 return visit_continue;
341 }
342
343 ir_visitor_status
344 ir_algebraic_visitor::visit_leave(ir_return *ir)
345 {
346 ir->value = handle_expression(ir->value);;
347 return visit_continue;
348 }
349
350 ir_visitor_status
351 ir_algebraic_visitor::visit_leave(ir_if *ir)
352 {
353 ir->condition = handle_expression(ir->condition);
354 return visit_continue;
355 }
356
357
358 bool
359 do_algebraic(exec_list *instructions)
360 {
361 ir_algebraic_visitor v;
362
363 visit_list_elements(&v, instructions);
364
365 return v.progress;
366 }