nir: Invalidate live SSA def information when making new SSA defs.
[mesa.git] / src / compiler / glsl / linker_util.cpp
1 /*
2 * Copyright © 2018 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 DEALINGS
21 * IN THE SOFTWARE.
22 *
23 */
24 #include "main/mtypes.h"
25 #include "glsl_types.h"
26 #include "linker_util.h"
27 #include "util/bitscan.h"
28 #include "util/set.h"
29 #include "ir_uniform.h" /* for gl_uniform_storage */
30
31 /* Utility methods shared between the GLSL IR and the NIR */
32
33 /* From the OpenGL 4.6 specification, 7.3.1.1 Naming Active Resources:
34 *
35 * "For an active shader storage block member declared as an array of an
36 * aggregate type, an entry will be generated only for the first array
37 * element, regardless of its type. Such block members are referred to as
38 * top-level arrays. If the block member is an aggregate type, the
39 * enumeration rules are then applied recursively."
40 */
41 bool
42 link_util_should_add_buffer_variable(struct gl_shader_program *prog,
43 struct gl_uniform_storage *uniform,
44 int top_level_array_base_offset,
45 int top_level_array_size_in_bytes,
46 int second_element_offset,
47 int block_index)
48 {
49 /* If the uniform is not a shader storage buffer or is not an array return
50 * true.
51 */
52 if (!uniform->is_shader_storage || top_level_array_size_in_bytes == 0)
53 return true;
54
55 int after_top_level_array = top_level_array_base_offset +
56 top_level_array_size_in_bytes;
57
58 /* Check for a new block, or that we are not dealing with array elements of
59 * a top member array other than the first element.
60 */
61 if (block_index != uniform->block_index ||
62 uniform->offset >= after_top_level_array ||
63 uniform->offset < second_element_offset) {
64 return true;
65 }
66
67 return false;
68 }
69
70 bool
71 link_util_add_program_resource(struct gl_shader_program *prog,
72 struct set *resource_set,
73 GLenum type, const void *data, uint8_t stages)
74 {
75 assert(data);
76
77 /* If resource already exists, do not add it again. */
78 if (_mesa_set_search(resource_set, data))
79 return true;
80
81 prog->data->ProgramResourceList =
82 reralloc(prog->data,
83 prog->data->ProgramResourceList,
84 gl_program_resource,
85 prog->data->NumProgramResourceList + 1);
86
87 if (!prog->data->ProgramResourceList) {
88 linker_error(prog, "Out of memory during linking.\n");
89 return false;
90 }
91
92 struct gl_program_resource *res =
93 &prog->data->ProgramResourceList[prog->data->NumProgramResourceList];
94
95 res->Type = type;
96 res->Data = data;
97 res->StageReferences = stages;
98
99 prog->data->NumProgramResourceList++;
100
101 _mesa_set_add(resource_set, data);
102
103 return true;
104 }
105
106 /**
107 * Search through the list of empty blocks to find one that fits the current
108 * uniform.
109 */
110 int
111 link_util_find_empty_block(struct gl_shader_program *prog,
112 struct gl_uniform_storage *uniform)
113 {
114 const unsigned entries = MAX2(1, uniform->array_elements);
115
116 foreach_list_typed(struct empty_uniform_block, block, link,
117 &prog->EmptyUniformLocations) {
118 /* Found a block with enough slots to fit the uniform */
119 if (block->slots == entries) {
120 unsigned start = block->start;
121 exec_node_remove(&block->link);
122 ralloc_free(block);
123
124 return start;
125 /* Found a block with more slots than needed. It can still be used. */
126 } else if (block->slots > entries) {
127 unsigned start = block->start;
128 block->start += entries;
129 block->slots -= entries;
130
131 return start;
132 }
133 }
134
135 return -1;
136 }
137
138 void
139 link_util_update_empty_uniform_locations(struct gl_shader_program *prog)
140 {
141 struct empty_uniform_block *current_block = NULL;
142
143 for (unsigned i = 0; i < prog->NumUniformRemapTable; i++) {
144 /* We found empty space in UniformRemapTable. */
145 if (prog->UniformRemapTable[i] == NULL) {
146 /* We've found the beginning of a new continous block of empty slots */
147 if (!current_block || current_block->start + current_block->slots != i) {
148 current_block = rzalloc(prog, struct empty_uniform_block);
149 current_block->start = i;
150 exec_list_push_tail(&prog->EmptyUniformLocations,
151 &current_block->link);
152 }
153
154 /* The current block continues, so we simply increment its slots */
155 current_block->slots++;
156 }
157 }
158 }
159
160 void
161 link_util_check_subroutine_resources(struct gl_shader_program *prog)
162 {
163 unsigned mask = prog->data->linked_stages;
164 while (mask) {
165 const int i = u_bit_scan(&mask);
166 struct gl_program *p = prog->_LinkedShaders[i]->Program;
167
168 if (p->sh.NumSubroutineUniformRemapTable > MAX_SUBROUTINE_UNIFORM_LOCATIONS) {
169 linker_error(prog, "Too many %s shader subroutine uniforms\n",
170 _mesa_shader_stage_to_string(i));
171 }
172 }
173 }
174
175 /**
176 * Validate uniform resources used by a program versus the implementation limits
177 */
178 void
179 link_util_check_uniform_resources(struct gl_context *ctx,
180 struct gl_shader_program *prog)
181 {
182 unsigned total_uniform_blocks = 0;
183 unsigned total_shader_storage_blocks = 0;
184
185 for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
186 struct gl_linked_shader *sh = prog->_LinkedShaders[i];
187
188 if (sh == NULL)
189 continue;
190
191 if (sh->num_uniform_components >
192 ctx->Const.Program[i].MaxUniformComponents) {
193 if (ctx->Const.GLSLSkipStrictMaxUniformLimitCheck) {
194 linker_warning(prog, "Too many %s shader default uniform block "
195 "components, but the driver will try to optimize "
196 "them out; this is non-portable out-of-spec "
197 "behavior\n",
198 _mesa_shader_stage_to_string(i));
199 } else {
200 linker_error(prog, "Too many %s shader default uniform block "
201 "components\n",
202 _mesa_shader_stage_to_string(i));
203 }
204 }
205
206 if (sh->num_combined_uniform_components >
207 ctx->Const.Program[i].MaxCombinedUniformComponents) {
208 if (ctx->Const.GLSLSkipStrictMaxUniformLimitCheck) {
209 linker_warning(prog, "Too many %s shader uniform components, "
210 "but the driver will try to optimize them out; "
211 "this is non-portable out-of-spec behavior\n",
212 _mesa_shader_stage_to_string(i));
213 } else {
214 linker_error(prog, "Too many %s shader uniform components\n",
215 _mesa_shader_stage_to_string(i));
216 }
217 }
218
219 total_shader_storage_blocks += sh->Program->info.num_ssbos;
220 total_uniform_blocks += sh->Program->info.num_ubos;
221 }
222
223 if (total_uniform_blocks > ctx->Const.MaxCombinedUniformBlocks) {
224 linker_error(prog, "Too many combined uniform blocks (%d/%d)\n",
225 total_uniform_blocks, ctx->Const.MaxCombinedUniformBlocks);
226 }
227
228 if (total_shader_storage_blocks > ctx->Const.MaxCombinedShaderStorageBlocks) {
229 linker_error(prog, "Too many combined shader storage blocks (%d/%d)\n",
230 total_shader_storage_blocks,
231 ctx->Const.MaxCombinedShaderStorageBlocks);
232 }
233
234 for (unsigned i = 0; i < prog->data->NumUniformBlocks; i++) {
235 if (prog->data->UniformBlocks[i].UniformBufferSize >
236 ctx->Const.MaxUniformBlockSize) {
237 linker_error(prog, "Uniform block %s too big (%d/%d)\n",
238 prog->data->UniformBlocks[i].Name,
239 prog->data->UniformBlocks[i].UniformBufferSize,
240 ctx->Const.MaxUniformBlockSize);
241 }
242 }
243
244 for (unsigned i = 0; i < prog->data->NumShaderStorageBlocks; i++) {
245 if (prog->data->ShaderStorageBlocks[i].UniformBufferSize >
246 ctx->Const.MaxShaderStorageBlockSize) {
247 linker_error(prog, "Shader storage block %s too big (%d/%d)\n",
248 prog->data->ShaderStorageBlocks[i].Name,
249 prog->data->ShaderStorageBlocks[i].UniformBufferSize,
250 ctx->Const.MaxShaderStorageBlockSize);
251 }
252 }
253 }
254
255 void
256 link_util_calculate_subroutine_compat(struct gl_shader_program *prog)
257 {
258 unsigned mask = prog->data->linked_stages;
259 while (mask) {
260 const int i = u_bit_scan(&mask);
261 struct gl_program *p = prog->_LinkedShaders[i]->Program;
262
263 for (unsigned j = 0; j < p->sh.NumSubroutineUniformRemapTable; j++) {
264 if (p->sh.SubroutineUniformRemapTable[j] == INACTIVE_UNIFORM_EXPLICIT_LOCATION)
265 continue;
266
267 struct gl_uniform_storage *uni = p->sh.SubroutineUniformRemapTable[j];
268
269 if (!uni)
270 continue;
271
272 int count = 0;
273 if (p->sh.NumSubroutineFunctions == 0) {
274 linker_error(prog, "subroutine uniform %s defined but no valid functions found\n", uni->type->name);
275 continue;
276 }
277 for (unsigned f = 0; f < p->sh.NumSubroutineFunctions; f++) {
278 struct gl_subroutine_function *fn = &p->sh.SubroutineFunctions[f];
279 for (int k = 0; k < fn->num_compat_types; k++) {
280 if (fn->types[k] == uni->type) {
281 count++;
282 break;
283 }
284 }
285 }
286 uni->num_compatible_subroutines = count;
287 }
288 }
289 }
290
291 /**
292 * Recursive part of the public mark_array_elements_referenced function.
293 *
294 * The recursion occurs when an entire array-of- is accessed. See the
295 * implementation for more details.
296 *
297 * \param dr List of array_deref_range elements to be
298 * processed.
299 * \param count Number of array_deref_range elements to be
300 * processed.
301 * \param scale Current offset scale.
302 * \param linearized_index Current accumulated linearized array index.
303 */
304 void
305 _mark_array_elements_referenced(const struct array_deref_range *dr,
306 unsigned count, unsigned scale,
307 unsigned linearized_index,
308 BITSET_WORD *bits)
309 {
310 /* Walk through the list of array dereferences in least- to
311 * most-significant order. Along the way, accumulate the current
312 * linearized offset and the scale factor for each array-of-.
313 */
314 for (unsigned i = 0; i < count; i++) {
315 if (dr[i].index < dr[i].size) {
316 linearized_index += dr[i].index * scale;
317 scale *= dr[i].size;
318 } else {
319 /* For each element in the current array, update the count and
320 * offset, then recurse to process the remaining arrays.
321 *
322 * There is some inefficency here if the last eBITSET_WORD *bitslement in the
323 * array_deref_range list specifies the entire array. In that case,
324 * the loop will make recursive calls with count == 0. In the call,
325 * all that will happen is the bit will be set.
326 */
327 for (unsigned j = 0; j < dr[i].size; j++) {
328 _mark_array_elements_referenced(&dr[i + 1],
329 count - (i + 1),
330 scale * dr[i].size,
331 linearized_index + (j * scale),
332 bits);
333 }
334
335 return;
336 }
337 }
338
339 BITSET_SET(bits, linearized_index);
340 }
341
342 /**
343 * Mark a set of array elements as accessed.
344 *
345 * If every \c array_deref_range is for a single index, only a single
346 * element will be marked. If any \c array_deref_range is for an entire
347 * array-of-, then multiple elements will be marked.
348 *
349 * Items in the \c array_deref_range list appear in least- to
350 * most-significant order. This is the \b opposite order the indices
351 * appear in the GLSL shader text. An array access like
352 *
353 * x = y[1][i][3];
354 *
355 * would appear as
356 *
357 * { { 3, n }, { m, m }, { 1, p } }
358 *
359 * where n, m, and p are the sizes of the arrays-of-arrays.
360 *
361 * The set of marked array elements can later be queried by
362 * \c ::is_linearized_index_referenced.
363 *
364 * \param dr List of array_deref_range elements to be processed.
365 * \param count Number of array_deref_range elements to be processed.
366 */
367 void
368 link_util_mark_array_elements_referenced(const struct array_deref_range *dr,
369 unsigned count, unsigned array_depth,
370 BITSET_WORD *bits)
371 {
372 if (count != array_depth)
373 return;
374
375 _mark_array_elements_referenced(dr, count, 1, 0, bits);
376 }