Merge remote-tracking branch 'mesa-public/master' into vulkan
[mesa.git] / src / mesa / program / program.c
1 /*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2007 Brian Paul All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25 /**
26 * \file program.c
27 * Vertex and fragment program support functions.
28 * \author Brian Paul
29 */
30
31
32 #include "main/glheader.h"
33 #include "main/context.h"
34 #include "main/framebuffer.h"
35 #include "main/hash.h"
36 #include "main/macros.h"
37 #include "program.h"
38 #include "prog_cache.h"
39 #include "prog_parameter.h"
40 #include "prog_instruction.h"
41 #include "util/ralloc.h"
42
43
44 /**
45 * A pointer to this dummy program is put into the hash table when
46 * glGenPrograms is called.
47 */
48 struct gl_program _mesa_DummyProgram;
49
50
51 /**
52 * Init context's vertex/fragment program state
53 */
54 void
55 _mesa_init_program(struct gl_context *ctx)
56 {
57 /*
58 * If this assertion fails, we need to increase the field
59 * size for register indexes (see INST_INDEX_BITS).
60 */
61 assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents / 4
62 <= (1 << INST_INDEX_BITS));
63 assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents / 4
64 <= (1 << INST_INDEX_BITS));
65
66 assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps <= (1 << INST_INDEX_BITS));
67 assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxLocalParams <= (1 << INST_INDEX_BITS));
68 assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTemps <= (1 << INST_INDEX_BITS));
69 assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxLocalParams <= (1 << INST_INDEX_BITS));
70
71 assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents <= 4 * MAX_UNIFORMS);
72 assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents <= 4 * MAX_UNIFORMS);
73
74 assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxAddressOffset <= (1 << INST_INDEX_BITS));
75 assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAddressOffset <= (1 << INST_INDEX_BITS));
76
77 /* If this fails, increase prog_instruction::TexSrcUnit size */
78 STATIC_ASSERT(MAX_TEXTURE_UNITS <= (1 << 5));
79
80 /* If this fails, increase prog_instruction::TexSrcTarget size */
81 STATIC_ASSERT(NUM_TEXTURE_TARGETS <= (1 << 4));
82
83 ctx->Program.ErrorPos = -1;
84 ctx->Program.ErrorString = strdup("");
85
86 ctx->VertexProgram.Enabled = GL_FALSE;
87 ctx->VertexProgram.PointSizeEnabled =
88 (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE;
89 ctx->VertexProgram.TwoSideEnabled = GL_FALSE;
90 _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
91 ctx->Shared->DefaultVertexProgram);
92 assert(ctx->VertexProgram.Current);
93 ctx->VertexProgram.Cache = _mesa_new_program_cache();
94
95 ctx->FragmentProgram.Enabled = GL_FALSE;
96 _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
97 ctx->Shared->DefaultFragmentProgram);
98 assert(ctx->FragmentProgram.Current);
99 ctx->FragmentProgram.Cache = _mesa_new_program_cache();
100
101 /* XXX probably move this stuff */
102 ctx->ATIFragmentShader.Enabled = GL_FALSE;
103 ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader;
104 assert(ctx->ATIFragmentShader.Current);
105 ctx->ATIFragmentShader.Current->RefCount++;
106 }
107
108
109 /**
110 * Free a context's vertex/fragment program state
111 */
112 void
113 _mesa_free_program_data(struct gl_context *ctx)
114 {
115 _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, NULL);
116 _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
117 _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, NULL);
118 _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
119
120 /* XXX probably move this stuff */
121 if (ctx->ATIFragmentShader.Current) {
122 ctx->ATIFragmentShader.Current->RefCount--;
123 if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
124 free(ctx->ATIFragmentShader.Current);
125 }
126 }
127
128 free((void *) ctx->Program.ErrorString);
129 }
130
131
132 /**
133 * Update the default program objects in the given context to reference those
134 * specified in the shared state and release those referencing the old
135 * shared state.
136 */
137 void
138 _mesa_update_default_objects_program(struct gl_context *ctx)
139 {
140 _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
141 ctx->Shared->DefaultVertexProgram);
142 assert(ctx->VertexProgram.Current);
143
144 _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
145 ctx->Shared->DefaultFragmentProgram);
146 assert(ctx->FragmentProgram.Current);
147
148 /* XXX probably move this stuff */
149 if (ctx->ATIFragmentShader.Current) {
150 ctx->ATIFragmentShader.Current->RefCount--;
151 if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
152 free(ctx->ATIFragmentShader.Current);
153 }
154 }
155 ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
156 assert(ctx->ATIFragmentShader.Current);
157 ctx->ATIFragmentShader.Current->RefCount++;
158 }
159
160
161 /**
162 * Set the vertex/fragment program error state (position and error string).
163 * This is generally called from within the parsers.
164 */
165 void
166 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
167 {
168 ctx->Program.ErrorPos = pos;
169 free((void *) ctx->Program.ErrorString);
170 if (!string)
171 string = "";
172 ctx->Program.ErrorString = strdup(string);
173 }
174
175
176 /**
177 * Initialize a new gl_program object.
178 */
179 struct gl_program *
180 _mesa_init_gl_program(struct gl_program *prog, GLenum target, GLuint id)
181 {
182 GLuint i;
183
184 if (!prog)
185 return NULL;
186
187 memset(prog, 0, sizeof(*prog));
188 mtx_init(&prog->Mutex, mtx_plain);
189 prog->Id = id;
190 prog->Target = target;
191 prog->RefCount = 1;
192 prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
193
194 /* default mapping from samplers to texture units */
195 for (i = 0; i < MAX_SAMPLERS; i++)
196 prog->SamplerUnits[i] = i;
197
198 return prog;
199 }
200
201
202 /**
203 * Allocate and initialize a new fragment/vertex program object but
204 * don't put it into the program hash table. Called via
205 * ctx->Driver.NewProgram. May be overridden (ie. replaced) by a
206 * device driver function to implement OO deriviation with additional
207 * types not understood by this function.
208 *
209 * \param ctx context
210 * \param id program id/number
211 * \param target program target/type
212 * \return pointer to new program object
213 */
214 struct gl_program *
215 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id)
216 {
217 switch (target) {
218 case GL_VERTEX_PROGRAM_ARB: { /* == GL_VERTEX_PROGRAM_NV */
219 struct gl_vertex_program *prog = CALLOC_STRUCT(gl_vertex_program);
220 return _mesa_init_gl_program(&prog->Base, target, id);
221 }
222 case GL_FRAGMENT_PROGRAM_NV:
223 case GL_FRAGMENT_PROGRAM_ARB: {
224 struct gl_fragment_program *prog = CALLOC_STRUCT(gl_fragment_program);
225 return _mesa_init_gl_program(&prog->Base, target, id);
226 }
227 case GL_GEOMETRY_PROGRAM_NV: {
228 struct gl_geometry_program *prog = CALLOC_STRUCT(gl_geometry_program);
229 return _mesa_init_gl_program(&prog->Base, target, id);
230 }
231 case GL_TESS_CONTROL_PROGRAM_NV: {
232 struct gl_tess_ctrl_program *prog = CALLOC_STRUCT(gl_tess_ctrl_program);
233 return _mesa_init_gl_program(&prog->Base, target, id);
234 }
235 case GL_TESS_EVALUATION_PROGRAM_NV: {
236 struct gl_tess_eval_program *prog = CALLOC_STRUCT(gl_tess_eval_program);
237 return _mesa_init_gl_program(&prog->Base, target, id);
238 }
239 case GL_COMPUTE_PROGRAM_NV: {
240 struct gl_compute_program *prog = CALLOC_STRUCT(gl_compute_program);
241 return _mesa_init_gl_program(&prog->Base, target, id);
242 }
243 default:
244 _mesa_problem(ctx, "bad target in _mesa_new_program");
245 return NULL;
246 }
247 }
248
249
250 /**
251 * Delete a program and remove it from the hash table, ignoring the
252 * reference count.
253 * Called via ctx->Driver.DeleteProgram. May be wrapped (OO deriviation)
254 * by a device driver function.
255 */
256 void
257 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
258 {
259 (void) ctx;
260 assert(prog);
261 assert(prog->RefCount==0);
262
263 if (prog == &_mesa_DummyProgram)
264 return;
265
266 free(prog->String);
267 free(prog->LocalParams);
268
269 if (prog->Instructions) {
270 _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
271 }
272 if (prog->Parameters) {
273 _mesa_free_parameter_list(prog->Parameters);
274 }
275
276 if (prog->nir) {
277 ralloc_free(prog->nir);
278 }
279
280 mtx_destroy(&prog->Mutex);
281 free(prog);
282 }
283
284
285 /**
286 * Return the gl_program object for a given ID.
287 * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
288 * casts elsewhere.
289 */
290 struct gl_program *
291 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
292 {
293 if (id)
294 return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
295 else
296 return NULL;
297 }
298
299
300 /**
301 * Reference counting for vertex/fragment programs
302 * This is normally only called from the _mesa_reference_program() macro
303 * when there's a real pointer change.
304 */
305 void
306 _mesa_reference_program_(struct gl_context *ctx,
307 struct gl_program **ptr,
308 struct gl_program *prog)
309 {
310 #ifndef NDEBUG
311 assert(ptr);
312 if (*ptr && prog) {
313 /* sanity check */
314 if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
315 assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
316 else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
317 assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
318 prog->Target == GL_FRAGMENT_PROGRAM_NV);
319 else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
320 assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
321 }
322 #endif
323
324 if (*ptr) {
325 GLboolean deleteFlag;
326 struct gl_program *oldProg = *ptr;
327
328 mtx_lock(&oldProg->Mutex);
329 assert(oldProg->RefCount > 0);
330 oldProg->RefCount--;
331
332 deleteFlag = (oldProg->RefCount == 0);
333 mtx_unlock(&oldProg->Mutex);
334
335 if (deleteFlag) {
336 assert(ctx);
337 ctx->Driver.DeleteProgram(ctx, oldProg);
338 }
339
340 *ptr = NULL;
341 }
342
343 assert(!*ptr);
344 if (prog) {
345 mtx_lock(&prog->Mutex);
346 prog->RefCount++;
347 mtx_unlock(&prog->Mutex);
348 }
349
350 *ptr = prog;
351 }
352
353
354 /**
355 * Insert 'count' NOP instructions at 'start' in the given program.
356 * Adjust branch targets accordingly.
357 */
358 GLboolean
359 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
360 {
361 const GLuint origLen = prog->NumInstructions;
362 const GLuint newLen = origLen + count;
363 struct prog_instruction *newInst;
364 GLuint i;
365
366 /* adjust branches */
367 for (i = 0; i < prog->NumInstructions; i++) {
368 struct prog_instruction *inst = prog->Instructions + i;
369 if (inst->BranchTarget > 0) {
370 if ((GLuint)inst->BranchTarget >= start) {
371 inst->BranchTarget += count;
372 }
373 }
374 }
375
376 /* Alloc storage for new instructions */
377 newInst = _mesa_alloc_instructions(newLen);
378 if (!newInst) {
379 return GL_FALSE;
380 }
381
382 /* Copy 'start' instructions into new instruction buffer */
383 _mesa_copy_instructions(newInst, prog->Instructions, start);
384
385 /* init the new instructions */
386 _mesa_init_instructions(newInst + start, count);
387
388 /* Copy the remaining/tail instructions to new inst buffer */
389 _mesa_copy_instructions(newInst + start + count,
390 prog->Instructions + start,
391 origLen - start);
392
393 /* free old instructions */
394 _mesa_free_instructions(prog->Instructions, origLen);
395
396 /* install new instructions */
397 prog->Instructions = newInst;
398 prog->NumInstructions = newLen;
399
400 return GL_TRUE;
401 }
402
403 /**
404 * Delete 'count' instructions at 'start' in the given program.
405 * Adjust branch targets accordingly.
406 */
407 GLboolean
408 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
409 {
410 const GLuint origLen = prog->NumInstructions;
411 const GLuint newLen = origLen - count;
412 struct prog_instruction *newInst;
413 GLuint i;
414
415 /* adjust branches */
416 for (i = 0; i < prog->NumInstructions; i++) {
417 struct prog_instruction *inst = prog->Instructions + i;
418 if (inst->BranchTarget > 0) {
419 if (inst->BranchTarget > (GLint) start) {
420 inst->BranchTarget -= count;
421 }
422 }
423 }
424
425 /* Alloc storage for new instructions */
426 newInst = _mesa_alloc_instructions(newLen);
427 if (!newInst) {
428 return GL_FALSE;
429 }
430
431 /* Copy 'start' instructions into new instruction buffer */
432 _mesa_copy_instructions(newInst, prog->Instructions, start);
433
434 /* Copy the remaining/tail instructions to new inst buffer */
435 _mesa_copy_instructions(newInst + start,
436 prog->Instructions + start + count,
437 newLen - start);
438
439 /* free old instructions */
440 _mesa_free_instructions(prog->Instructions, origLen);
441
442 /* install new instructions */
443 prog->Instructions = newInst;
444 prog->NumInstructions = newLen;
445
446 return GL_TRUE;
447 }
448
449
450 /**
451 * Populate the 'used' array with flags indicating which registers (TEMPs,
452 * INPUTs, OUTPUTs, etc, are used by the given program.
453 * \param file type of register to scan for
454 * \param used returns true/false flags for in use / free
455 * \param usedSize size of the 'used' array
456 */
457 void
458 _mesa_find_used_registers(const struct gl_program *prog,
459 gl_register_file file,
460 GLboolean used[], GLuint usedSize)
461 {
462 GLuint i, j;
463
464 memset(used, 0, usedSize);
465
466 for (i = 0; i < prog->NumInstructions; i++) {
467 const struct prog_instruction *inst = prog->Instructions + i;
468 const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
469
470 if (inst->DstReg.File == file) {
471 assert(inst->DstReg.Index < usedSize);
472 if(inst->DstReg.Index < usedSize)
473 used[inst->DstReg.Index] = GL_TRUE;
474 }
475
476 for (j = 0; j < n; j++) {
477 if (inst->SrcReg[j].File == file) {
478 assert(inst->SrcReg[j].Index < (GLint) usedSize);
479 if (inst->SrcReg[j].Index < (GLint) usedSize)
480 used[inst->SrcReg[j].Index] = GL_TRUE;
481 }
482 }
483 }
484 }
485
486
487 /**
488 * Scan the given 'used' register flag array for the first entry
489 * that's >= firstReg.
490 * \param used vector of flags indicating registers in use (as returned
491 * by _mesa_find_used_registers())
492 * \param usedSize size of the 'used' array
493 * \param firstReg first register to start searching at
494 * \return index of unused register, or -1 if none.
495 */
496 GLint
497 _mesa_find_free_register(const GLboolean used[],
498 GLuint usedSize, GLuint firstReg)
499 {
500 GLuint i;
501
502 assert(firstReg < usedSize);
503
504 for (i = firstReg; i < usedSize; i++)
505 if (!used[i])
506 return i;
507
508 return -1;
509 }
510
511
512 /* Gets the minimum number of shader invocations per fragment.
513 * This function is useful to determine if we need to do per
514 * sample shading or per fragment shading.
515 */
516 GLint
517 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
518 const struct gl_fragment_program *prog,
519 bool ignore_sample_qualifier)
520 {
521 /* From ARB_sample_shading specification:
522 * "Using gl_SampleID in a fragment shader causes the entire shader
523 * to be evaluated per-sample."
524 *
525 * "Using gl_SamplePosition in a fragment shader causes the entire
526 * shader to be evaluated per-sample."
527 *
528 * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
529 * has no effect."
530 */
531 if (ctx->Multisample.Enabled) {
532 /* The ARB_gpu_shader5 specification says:
533 *
534 * "Use of the "sample" qualifier on a fragment shader input
535 * forces per-sample shading"
536 */
537 if (prog->IsSample && !ignore_sample_qualifier)
538 return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
539
540 if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
541 SYSTEM_BIT_SAMPLE_POS))
542 return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
543 else if (ctx->Multisample.SampleShading)
544 return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
545 _mesa_geometric_samples(ctx->DrawBuffer)), 1);
546 else
547 return 1;
548 }
549 return 1;
550 }