nir: move to compiler/
[mesa.git] / src / glsl / ir_function.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 #include "compiler/glsl_types.h"
25 #include "ir.h"
26 #include "glsl_parser_extras.h"
27 #include "main/errors.h"
28
29 typedef enum {
30 PARAMETER_LIST_NO_MATCH,
31 PARAMETER_LIST_EXACT_MATCH,
32 PARAMETER_LIST_INEXACT_MATCH /*< Match requires implicit conversion. */
33 } parameter_list_match_t;
34
35 /**
36 * \brief Check if two parameter lists match.
37 *
38 * \param list_a Parameters of the function definition.
39 * \param list_b Actual parameters passed to the function.
40 * \see matching_signature()
41 */
42 static parameter_list_match_t
43 parameter_lists_match(_mesa_glsl_parse_state *state,
44 const exec_list *list_a, const exec_list *list_b)
45 {
46 const exec_node *node_a = list_a->head;
47 const exec_node *node_b = list_b->head;
48
49 /* This is set to true if there is an inexact match requiring an implicit
50 * conversion. */
51 bool inexact_match = false;
52
53 for (/* empty */
54 ; !node_a->is_tail_sentinel()
55 ; node_a = node_a->next, node_b = node_b->next) {
56 /* If all of the parameters from the other parameter list have been
57 * exhausted, the lists have different length and, by definition,
58 * do not match.
59 */
60 if (node_b->is_tail_sentinel())
61 return PARAMETER_LIST_NO_MATCH;
62
63
64 const ir_variable *const param = (ir_variable *) node_a;
65 const ir_rvalue *const actual = (ir_rvalue *) node_b;
66
67 if (param->type == actual->type)
68 continue;
69
70 /* Try to find an implicit conversion from actual to param. */
71 inexact_match = true;
72 switch ((enum ir_variable_mode)(param->data.mode)) {
73 case ir_var_auto:
74 case ir_var_uniform:
75 case ir_var_shader_storage:
76 case ir_var_temporary:
77 /* These are all error conditions. It is invalid for a parameter to
78 * a function to be declared as auto (not in, out, or inout) or
79 * as uniform.
80 */
81 assert(0);
82 return PARAMETER_LIST_NO_MATCH;
83
84 case ir_var_const_in:
85 case ir_var_function_in:
86 if (!actual->type->can_implicitly_convert_to(param->type, state))
87 return PARAMETER_LIST_NO_MATCH;
88 break;
89
90 case ir_var_function_out:
91 if (!param->type->can_implicitly_convert_to(actual->type, state))
92 return PARAMETER_LIST_NO_MATCH;
93 break;
94
95 case ir_var_function_inout:
96 /* Since there are no bi-directional automatic conversions (e.g.,
97 * there is int -> float but no float -> int), inout parameters must
98 * be exact matches.
99 */
100 return PARAMETER_LIST_NO_MATCH;
101
102 default:
103 assert(false);
104 return PARAMETER_LIST_NO_MATCH;
105 }
106 }
107
108 /* If all of the parameters from the other parameter list have been
109 * exhausted, the lists have different length and, by definition, do not
110 * match.
111 */
112 if (!node_b->is_tail_sentinel())
113 return PARAMETER_LIST_NO_MATCH;
114
115 if (inexact_match)
116 return PARAMETER_LIST_INEXACT_MATCH;
117 else
118 return PARAMETER_LIST_EXACT_MATCH;
119 }
120
121
122 /* Classes of parameter match, sorted (mostly) best matches first.
123 * See is_better_parameter_match() below for the exceptions.
124 * */
125 typedef enum {
126 PARAMETER_EXACT_MATCH,
127 PARAMETER_FLOAT_TO_DOUBLE,
128 PARAMETER_INT_TO_FLOAT,
129 PARAMETER_INT_TO_DOUBLE,
130 PARAMETER_OTHER_CONVERSION,
131 } parameter_match_t;
132
133
134 static parameter_match_t
135 get_parameter_match_type(const ir_variable *param,
136 const ir_rvalue *actual)
137 {
138 const glsl_type *from_type;
139 const glsl_type *to_type;
140
141 if (param->data.mode == ir_var_function_out) {
142 from_type = param->type;
143 to_type = actual->type;
144 } else {
145 from_type = actual->type;
146 to_type = param->type;
147 }
148
149 if (from_type == to_type)
150 return PARAMETER_EXACT_MATCH;
151
152 if (to_type->base_type == GLSL_TYPE_DOUBLE) {
153 if (from_type->base_type == GLSL_TYPE_FLOAT)
154 return PARAMETER_FLOAT_TO_DOUBLE;
155 return PARAMETER_INT_TO_DOUBLE;
156 }
157
158 if (to_type->base_type == GLSL_TYPE_FLOAT)
159 return PARAMETER_INT_TO_FLOAT;
160
161 /* int -> uint and any other oddball conversions */
162 return PARAMETER_OTHER_CONVERSION;
163 }
164
165
166 static bool
167 is_better_parameter_match(parameter_match_t a_match,
168 parameter_match_t b_match)
169 {
170 /* From section 6.1 of the GLSL 4.00 spec (and the ARB_gpu_shader5 spec):
171 *
172 * 1. An exact match is better than a match involving any implicit
173 * conversion.
174 *
175 * 2. A match involving an implicit conversion from float to double
176 * is better than match involving any other implicit conversion.
177 *
178 * [XXX: Not in GLSL 4.0: Only in ARB_gpu_shader5:
179 * 3. A match involving an implicit conversion from either int or uint
180 * to float is better than a match involving an implicit conversion
181 * from either int or uint to double.]
182 *
183 * If none of the rules above apply to a particular pair of conversions,
184 * neither conversion is considered better than the other.
185 *
186 * --
187 *
188 * Notably, the int->uint conversion is *not* considered to be better
189 * or worse than int/uint->float or int/uint->double.
190 */
191
192 if (a_match >= PARAMETER_INT_TO_FLOAT && b_match == PARAMETER_OTHER_CONVERSION)
193 return false;
194
195 return a_match < b_match;
196 }
197
198
199 static bool
200 is_best_inexact_overload(const exec_list *actual_parameters,
201 ir_function_signature **matches,
202 int num_matches,
203 ir_function_signature *sig)
204 {
205 /* From section 6.1 of the GLSL 4.00 spec (and the ARB_gpu_shader5 spec):
206 *
207 * "A function definition A is considered a better
208 * match than function definition B if:
209 *
210 * * for at least one function argument, the conversion for that argument
211 * in A is better than the corresponding conversion in B; and
212 *
213 * * there is no function argument for which the conversion in B is better
214 * than the corresponding conversion in A.
215 *
216 * If a single function definition is considered a better match than every
217 * other matching function definition, it will be used. Otherwise, a
218 * semantic error occurs and the shader will fail to compile."
219 */
220 for (ir_function_signature **other = matches;
221 other < matches + num_matches; other++) {
222 if (*other == sig)
223 continue;
224
225 const exec_node *node_a = sig->parameters.head;
226 const exec_node *node_b = (*other)->parameters.head;
227 const exec_node *node_p = actual_parameters->head;
228
229 bool better_for_some_parameter = false;
230
231 for (/* empty */
232 ; !node_a->is_tail_sentinel()
233 ; node_a = node_a->next,
234 node_b = node_b->next,
235 node_p = node_p->next) {
236 parameter_match_t a_match = get_parameter_match_type(
237 (const ir_variable *)node_a,
238 (const ir_rvalue *)node_p);
239 parameter_match_t b_match = get_parameter_match_type(
240 (const ir_variable *)node_b,
241 (const ir_rvalue *)node_p);
242
243 if (is_better_parameter_match(a_match, b_match))
244 better_for_some_parameter = true;
245
246 if (is_better_parameter_match(b_match, a_match))
247 return false; /* B is better for this parameter */
248 }
249
250 if (!better_for_some_parameter)
251 return false; /* A must be better than B for some parameter */
252
253 }
254
255 return true;
256 }
257
258
259 static ir_function_signature *
260 choose_best_inexact_overload(_mesa_glsl_parse_state *state,
261 const exec_list *actual_parameters,
262 ir_function_signature **matches,
263 int num_matches)
264 {
265 if (num_matches == 0)
266 return NULL;
267
268 if (num_matches == 1)
269 return *matches;
270
271 /* Without GLSL 4.0 / ARB_gpu_shader5, there is no overload resolution
272 * among multiple inexact matches. Note that state may be NULL here if
273 * called from the linker; in that case we assume everything supported in
274 * any GLSL version is available. */
275 if (!state || state->is_version(400, 0) || state->ARB_gpu_shader5_enable) {
276 for (ir_function_signature **sig = matches; sig < matches + num_matches; sig++) {
277 if (is_best_inexact_overload(actual_parameters, matches, num_matches, *sig))
278 return *sig;
279 }
280 }
281
282 return NULL; /* no best candidate */
283 }
284
285
286 ir_function_signature *
287 ir_function::matching_signature(_mesa_glsl_parse_state *state,
288 const exec_list *actual_parameters,
289 bool allow_builtins)
290 {
291 bool is_exact;
292 return matching_signature(state, actual_parameters, allow_builtins,
293 &is_exact);
294 }
295
296 ir_function_signature *
297 ir_function::matching_signature(_mesa_glsl_parse_state *state,
298 const exec_list *actual_parameters,
299 bool allow_builtins,
300 bool *is_exact)
301 {
302 ir_function_signature **inexact_matches = NULL;
303 ir_function_signature **inexact_matches_temp;
304 ir_function_signature *match = NULL;
305 int num_inexact_matches = 0;
306
307 /* From page 42 (page 49 of the PDF) of the GLSL 1.20 spec:
308 *
309 * "If an exact match is found, the other signatures are ignored, and
310 * the exact match is used. Otherwise, if no exact match is found, then
311 * the implicit conversions in Section 4.1.10 "Implicit Conversions" will
312 * be applied to the calling arguments if this can make their types match
313 * a signature. In this case, it is a semantic error if there are
314 * multiple ways to apply these conversions to the actual arguments of a
315 * call such that the call can be made to match multiple signatures."
316 */
317 foreach_in_list(ir_function_signature, sig, &this->signatures) {
318 /* Skip over any built-ins that aren't available in this shader. */
319 if (sig->is_builtin() && (!allow_builtins ||
320 !sig->is_builtin_available(state)))
321 continue;
322
323 switch (parameter_lists_match(state, & sig->parameters, actual_parameters)) {
324 case PARAMETER_LIST_EXACT_MATCH:
325 *is_exact = true;
326 free(inexact_matches);
327 return sig;
328 case PARAMETER_LIST_INEXACT_MATCH:
329 inexact_matches_temp = (ir_function_signature **)
330 realloc(inexact_matches,
331 sizeof(*inexact_matches) *
332 (num_inexact_matches + 1));
333 if (inexact_matches_temp == NULL) {
334 _mesa_error_no_memory(__func__);
335 free(inexact_matches);
336 return NULL;
337 }
338 inexact_matches = inexact_matches_temp;
339 inexact_matches[num_inexact_matches++] = sig;
340 continue;
341 case PARAMETER_LIST_NO_MATCH:
342 continue;
343 default:
344 assert(false);
345 return NULL;
346 }
347 }
348
349 /* There is no exact match (we would have returned it by now). If there
350 * are multiple inexact matches, the call is ambiguous, which is an error.
351 *
352 * FINISHME: Report a decent error. Returning NULL will likely result in
353 * FINISHME: a "no matching signature" error; it should report that the
354 * FINISHME: call is ambiguous. But reporting errors from here is hard.
355 */
356 *is_exact = false;
357
358 match = choose_best_inexact_overload(state, actual_parameters,
359 inexact_matches, num_inexact_matches);
360
361 free(inexact_matches);
362 return match;
363 }
364
365
366 static bool
367 parameter_lists_match_exact(const exec_list *list_a, const exec_list *list_b)
368 {
369 const exec_node *node_a = list_a->head;
370 const exec_node *node_b = list_b->head;
371
372 for (/* empty */
373 ; !node_a->is_tail_sentinel() && !node_b->is_tail_sentinel()
374 ; node_a = node_a->next, node_b = node_b->next) {
375 ir_variable *a = (ir_variable *) node_a;
376 ir_variable *b = (ir_variable *) node_b;
377
378 /* If the types of the parameters do not match, the parameters lists
379 * are different.
380 */
381 if (a->type != b->type)
382 return false;
383 }
384
385 /* Unless both lists are exhausted, they differ in length and, by
386 * definition, do not match.
387 */
388 return (node_a->is_tail_sentinel() == node_b->is_tail_sentinel());
389 }
390
391 ir_function_signature *
392 ir_function::exact_matching_signature(_mesa_glsl_parse_state *state,
393 const exec_list *actual_parameters)
394 {
395 foreach_in_list(ir_function_signature, sig, &this->signatures) {
396 /* Skip over any built-ins that aren't available in this shader. */
397 if (sig->is_builtin() && !sig->is_builtin_available(state))
398 continue;
399
400 if (parameter_lists_match_exact(&sig->parameters, actual_parameters))
401 return sig;
402 }
403 return NULL;
404 }