i965/mesa/st/swrast: set fs shader_info directly and switch to using it
[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_program(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_program(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_program(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_FRAGMENT_PROGRAM_ARB: {
219 struct gl_fragment_program *prog = CALLOC_STRUCT(gl_fragment_program);
220 return _mesa_init_gl_program(&prog->Base, target, id);
221 }
222 case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
223 case GL_GEOMETRY_PROGRAM_NV:
224 case GL_TESS_CONTROL_PROGRAM_NV:
225 case GL_TESS_EVALUATION_PROGRAM_NV:
226 case GL_COMPUTE_PROGRAM_NV: {
227 struct gl_program *prog = CALLOC_STRUCT(gl_program);
228 return _mesa_init_gl_program(prog, target, id);
229 }
230 default:
231 _mesa_problem(ctx, "bad target in _mesa_new_program");
232 return NULL;
233 }
234 }
235
236
237 /**
238 * Delete a program and remove it from the hash table, ignoring the
239 * reference count.
240 * Called via ctx->Driver.DeleteProgram. May be wrapped (OO deriviation)
241 * by a device driver function.
242 */
243 void
244 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
245 {
246 (void) ctx;
247 assert(prog);
248 assert(prog->RefCount==0);
249
250 if (prog == &_mesa_DummyProgram)
251 return;
252
253 free(prog->String);
254 free(prog->LocalParams);
255
256 if (prog->Instructions) {
257 _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
258 }
259 if (prog->Parameters) {
260 _mesa_free_parameter_list(prog->Parameters);
261 }
262
263 if (prog->nir) {
264 ralloc_free(prog->nir);
265 }
266
267 mtx_destroy(&prog->Mutex);
268 free(prog);
269 }
270
271
272 /**
273 * Return the gl_program object for a given ID.
274 * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
275 * casts elsewhere.
276 */
277 struct gl_program *
278 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
279 {
280 if (id)
281 return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
282 else
283 return NULL;
284 }
285
286
287 /**
288 * Reference counting for vertex/fragment programs
289 * This is normally only called from the _mesa_reference_program() macro
290 * when there's a real pointer change.
291 */
292 void
293 _mesa_reference_program_(struct gl_context *ctx,
294 struct gl_program **ptr,
295 struct gl_program *prog)
296 {
297 #ifndef NDEBUG
298 assert(ptr);
299 if (*ptr && prog) {
300 /* sanity check */
301 if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
302 assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
303 else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
304 assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
305 prog->Target == GL_FRAGMENT_PROGRAM_NV);
306 else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
307 assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
308 }
309 #endif
310
311 if (*ptr) {
312 GLboolean deleteFlag;
313 struct gl_program *oldProg = *ptr;
314
315 mtx_lock(&oldProg->Mutex);
316 assert(oldProg->RefCount > 0);
317 oldProg->RefCount--;
318
319 deleteFlag = (oldProg->RefCount == 0);
320 mtx_unlock(&oldProg->Mutex);
321
322 if (deleteFlag) {
323 assert(ctx);
324 ctx->Driver.DeleteProgram(ctx, oldProg);
325 }
326
327 *ptr = NULL;
328 }
329
330 assert(!*ptr);
331 if (prog) {
332 mtx_lock(&prog->Mutex);
333 prog->RefCount++;
334 mtx_unlock(&prog->Mutex);
335 }
336
337 *ptr = prog;
338 }
339
340
341 /**
342 * Insert 'count' NOP instructions at 'start' in the given program.
343 * Adjust branch targets accordingly.
344 */
345 GLboolean
346 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
347 {
348 const GLuint origLen = prog->NumInstructions;
349 const GLuint newLen = origLen + count;
350 struct prog_instruction *newInst;
351 GLuint i;
352
353 /* adjust branches */
354 for (i = 0; i < prog->NumInstructions; i++) {
355 struct prog_instruction *inst = prog->Instructions + i;
356 if (inst->BranchTarget > 0) {
357 if ((GLuint)inst->BranchTarget >= start) {
358 inst->BranchTarget += count;
359 }
360 }
361 }
362
363 /* Alloc storage for new instructions */
364 newInst = _mesa_alloc_instructions(newLen);
365 if (!newInst) {
366 return GL_FALSE;
367 }
368
369 /* Copy 'start' instructions into new instruction buffer */
370 _mesa_copy_instructions(newInst, prog->Instructions, start);
371
372 /* init the new instructions */
373 _mesa_init_instructions(newInst + start, count);
374
375 /* Copy the remaining/tail instructions to new inst buffer */
376 _mesa_copy_instructions(newInst + start + count,
377 prog->Instructions + start,
378 origLen - start);
379
380 /* free old instructions */
381 _mesa_free_instructions(prog->Instructions, origLen);
382
383 /* install new instructions */
384 prog->Instructions = newInst;
385 prog->NumInstructions = newLen;
386
387 return GL_TRUE;
388 }
389
390 /**
391 * Delete 'count' instructions at 'start' in the given program.
392 * Adjust branch targets accordingly.
393 */
394 GLboolean
395 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
396 {
397 const GLuint origLen = prog->NumInstructions;
398 const GLuint newLen = origLen - count;
399 struct prog_instruction *newInst;
400 GLuint i;
401
402 /* adjust branches */
403 for (i = 0; i < prog->NumInstructions; i++) {
404 struct prog_instruction *inst = prog->Instructions + i;
405 if (inst->BranchTarget > 0) {
406 if (inst->BranchTarget > (GLint) start) {
407 inst->BranchTarget -= count;
408 }
409 }
410 }
411
412 /* Alloc storage for new instructions */
413 newInst = _mesa_alloc_instructions(newLen);
414 if (!newInst) {
415 return GL_FALSE;
416 }
417
418 /* Copy 'start' instructions into new instruction buffer */
419 _mesa_copy_instructions(newInst, prog->Instructions, start);
420
421 /* Copy the remaining/tail instructions to new inst buffer */
422 _mesa_copy_instructions(newInst + start,
423 prog->Instructions + start + count,
424 newLen - start);
425
426 /* free old instructions */
427 _mesa_free_instructions(prog->Instructions, origLen);
428
429 /* install new instructions */
430 prog->Instructions = newInst;
431 prog->NumInstructions = newLen;
432
433 return GL_TRUE;
434 }
435
436
437 /**
438 * Populate the 'used' array with flags indicating which registers (TEMPs,
439 * INPUTs, OUTPUTs, etc, are used by the given program.
440 * \param file type of register to scan for
441 * \param used returns true/false flags for in use / free
442 * \param usedSize size of the 'used' array
443 */
444 void
445 _mesa_find_used_registers(const struct gl_program *prog,
446 gl_register_file file,
447 GLboolean used[], GLuint usedSize)
448 {
449 GLuint i, j;
450
451 memset(used, 0, usedSize);
452
453 for (i = 0; i < prog->NumInstructions; i++) {
454 const struct prog_instruction *inst = prog->Instructions + i;
455 const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
456
457 if (inst->DstReg.File == file) {
458 assert(inst->DstReg.Index < usedSize);
459 if(inst->DstReg.Index < usedSize)
460 used[inst->DstReg.Index] = GL_TRUE;
461 }
462
463 for (j = 0; j < n; j++) {
464 if (inst->SrcReg[j].File == file) {
465 assert(inst->SrcReg[j].Index < (GLint) usedSize);
466 if (inst->SrcReg[j].Index < (GLint) usedSize)
467 used[inst->SrcReg[j].Index] = GL_TRUE;
468 }
469 }
470 }
471 }
472
473
474 /**
475 * Scan the given 'used' register flag array for the first entry
476 * that's >= firstReg.
477 * \param used vector of flags indicating registers in use (as returned
478 * by _mesa_find_used_registers())
479 * \param usedSize size of the 'used' array
480 * \param firstReg first register to start searching at
481 * \return index of unused register, or -1 if none.
482 */
483 GLint
484 _mesa_find_free_register(const GLboolean used[],
485 GLuint usedSize, GLuint firstReg)
486 {
487 GLuint i;
488
489 assert(firstReg < usedSize);
490
491 for (i = firstReg; i < usedSize; i++)
492 if (!used[i])
493 return i;
494
495 return -1;
496 }
497
498
499 /* Gets the minimum number of shader invocations per fragment.
500 * This function is useful to determine if we need to do per
501 * sample shading or per fragment shading.
502 */
503 GLint
504 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
505 const struct gl_fragment_program *prog,
506 bool ignore_sample_qualifier)
507 {
508 /* From ARB_sample_shading specification:
509 * "Using gl_SampleID in a fragment shader causes the entire shader
510 * to be evaluated per-sample."
511 *
512 * "Using gl_SamplePosition in a fragment shader causes the entire
513 * shader to be evaluated per-sample."
514 *
515 * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
516 * has no effect."
517 */
518 if (ctx->Multisample.Enabled) {
519 /* The ARB_gpu_shader5 specification says:
520 *
521 * "Use of the "sample" qualifier on a fragment shader input
522 * forces per-sample shading"
523 */
524 if (prog->Base.info.fs.uses_sample_qualifier && !ignore_sample_qualifier)
525 return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
526
527 if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
528 SYSTEM_BIT_SAMPLE_POS))
529 return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
530 else if (ctx->Multisample.SampleShading)
531 return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
532 _mesa_geometric_samples(ctx->DrawBuffer)), 1);
533 else
534 return 1;
535 }
536 return 1;
537 }