projects / mesa.git / blob
? search:


25a849bb1833199598421fdcbf1bb600d507b928
[mesa.git] / src / mesa / state_tracker / st_program.c
1 /**************************************************************************
2 *
3 * Copyright 2007 VMware, Inc.
4 * 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
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keithw@vmware.com>
30 * Brian Paul
31 */
32
33
34 #include "main/imports.h"
35 #include "main/hash.h"
36 #include "main/mtypes.h"
37 #include "program/prog_parameter.h"
38 #include "program/prog_print.h"
39 #include "program/programopt.h"
40
41 #include "compiler/nir/nir.h"
42
43 #include "pipe/p_context.h"
44 #include "pipe/p_defines.h"
45 #include "pipe/p_shader_tokens.h"
46 #include "draw/draw_context.h"
47 #include "tgsi/tgsi_dump.h"
48 #include "tgsi/tgsi_emulate.h"
49 #include "tgsi/tgsi_parse.h"
50 #include "tgsi/tgsi_ureg.h"
51
52 #include "st_debug.h"
53 #include "st_cb_bitmap.h"
54 #include "st_cb_drawpixels.h"
55 #include "st_context.h"
56 #include "st_tgsi_lower_yuv.h"
57 #include "st_program.h"
58 #include "st_mesa_to_tgsi.h"
59 #include "st_atifs_to_tgsi.h"
60 #include "st_nir.h"
61 #include "st_shader_cache.h"
62 #include "cso_cache/cso_context.h"
63
64
65
66 static void
67 set_affected_state_flags(uint64_t *states,
68 struct gl_program *prog,
69 uint64_t new_constants,
70 uint64_t new_sampler_views,
71 uint64_t new_samplers,
72 uint64_t new_images,
73 uint64_t new_ubos,
74 uint64_t new_ssbos,
75 uint64_t new_atomics)
76 {
77 if (prog->Parameters->NumParameters)
78 *states |= new_constants;
79
80 if (prog->info.num_textures)
81 *states |= new_sampler_views | new_samplers;
82
83 if (prog->info.num_images)
84 *states |= new_images;
85
86 if (prog->info.num_ubos)
87 *states |= new_ubos;
88
89 if (prog->info.num_ssbos)
90 *states |= new_ssbos;
91
92 if (prog->info.num_abos)
93 *states |= new_atomics;
94 }
95
96 /**
97 * This determines which states will be updated when the shader is bound.
98 */
99 void
100 st_set_prog_affected_state_flags(struct gl_program *prog)
101 {
102 uint64_t *states;
103
104 switch (prog->info.stage) {
105 case MESA_SHADER_VERTEX:
106 states = &((struct st_vertex_program*)prog)->affected_states;
107
108 *states = ST_NEW_VS_STATE |
109 ST_NEW_RASTERIZER |
110 ST_NEW_VERTEX_ARRAYS;
111
112 set_affected_state_flags(states, prog,
113 ST_NEW_VS_CONSTANTS,
114 ST_NEW_VS_SAMPLER_VIEWS,
115 ST_NEW_VS_SAMPLERS,
116 ST_NEW_VS_IMAGES,
117 ST_NEW_VS_UBOS,
118 ST_NEW_VS_SSBOS,
119 ST_NEW_VS_ATOMICS);
120 break;
121
122 case MESA_SHADER_TESS_CTRL:
123 states = &(st_common_program(prog))->affected_states;
124
125 *states = ST_NEW_TCS_STATE;
126
127 set_affected_state_flags(states, prog,
128 ST_NEW_TCS_CONSTANTS,
129 ST_NEW_TCS_SAMPLER_VIEWS,
130 ST_NEW_TCS_SAMPLERS,
131 ST_NEW_TCS_IMAGES,
132 ST_NEW_TCS_UBOS,
133 ST_NEW_TCS_SSBOS,
134 ST_NEW_TCS_ATOMICS);
135 break;
136
137 case MESA_SHADER_TESS_EVAL:
138 states = &(st_common_program(prog))->affected_states;
139
140 *states = ST_NEW_TES_STATE |
141 ST_NEW_RASTERIZER;
142
143 set_affected_state_flags(states, prog,
144 ST_NEW_TES_CONSTANTS,
145 ST_NEW_TES_SAMPLER_VIEWS,
146 ST_NEW_TES_SAMPLERS,
147 ST_NEW_TES_IMAGES,
148 ST_NEW_TES_UBOS,
149 ST_NEW_TES_SSBOS,
150 ST_NEW_TES_ATOMICS);
151 break;
152
153 case MESA_SHADER_GEOMETRY:
154 states = &(st_common_program(prog))->affected_states;
155
156 *states = ST_NEW_GS_STATE |
157 ST_NEW_RASTERIZER;
158
159 set_affected_state_flags(states, prog,
160 ST_NEW_GS_CONSTANTS,
161 ST_NEW_GS_SAMPLER_VIEWS,
162 ST_NEW_GS_SAMPLERS,
163 ST_NEW_GS_IMAGES,
164 ST_NEW_GS_UBOS,
165 ST_NEW_GS_SSBOS,
166 ST_NEW_GS_ATOMICS);
167 break;
168
169 case MESA_SHADER_FRAGMENT:
170 states = &((struct st_fragment_program*)prog)->affected_states;
171
172 /* gl_FragCoord and glDrawPixels always use constants. */
173 *states = ST_NEW_FS_STATE |
174 ST_NEW_SAMPLE_SHADING |
175 ST_NEW_FS_CONSTANTS;
176
177 set_affected_state_flags(states, prog,
178 ST_NEW_FS_CONSTANTS,
179 ST_NEW_FS_SAMPLER_VIEWS,
180 ST_NEW_FS_SAMPLERS,
181 ST_NEW_FS_IMAGES,
182 ST_NEW_FS_UBOS,
183 ST_NEW_FS_SSBOS,
184 ST_NEW_FS_ATOMICS);
185 break;
186
187 case MESA_SHADER_COMPUTE:
188 states = &((struct st_compute_program*)prog)->affected_states;
189
190 *states = ST_NEW_CS_STATE;
191
192 set_affected_state_flags(states, prog,
193 ST_NEW_CS_CONSTANTS,
194 ST_NEW_CS_SAMPLER_VIEWS,
195 ST_NEW_CS_SAMPLERS,
196 ST_NEW_CS_IMAGES,
197 ST_NEW_CS_UBOS,
198 ST_NEW_CS_SSBOS,
199 ST_NEW_CS_ATOMICS);
200 break;
201
202 default:
203 unreachable("unhandled shader stage");
204 }
205 }
206
207 /**
208 * Delete a vertex program variant. Note the caller must unlink
209 * the variant from the linked list.
210 */
211 static void
212 delete_vp_variant(struct st_context *st, struct st_vp_variant *vpv)
213 {
214 if (vpv->driver_shader)
215 cso_delete_vertex_shader(st->cso_context, vpv->driver_shader);
216
217 if (vpv->draw_shader)
218 draw_delete_vertex_shader( st->draw, vpv->draw_shader );
219
220 if (((vpv->tgsi.type == PIPE_SHADER_IR_TGSI)) && vpv->tgsi.tokens)
221 ureg_free_tokens(vpv->tgsi.tokens);
222
223 free( vpv );
224 }
225
226
227
228 /**
229 * Clean out any old compilations:
230 */
231 void
232 st_release_vp_variants( struct st_context *st,
233 struct st_vertex_program *stvp )
234 {
235 struct st_vp_variant *vpv;
236
237 for (vpv = stvp->variants; vpv; ) {
238 struct st_vp_variant *next = vpv->next;
239 delete_vp_variant(st, vpv);
240 vpv = next;
241 }
242
243 stvp->variants = NULL;
244
245 if ((stvp->tgsi.type == PIPE_SHADER_IR_TGSI) && stvp->tgsi.tokens) {
246 tgsi_free_tokens(stvp->tgsi.tokens);
247 stvp->tgsi.tokens = NULL;
248 }
249 }
250
251
252
253 /**
254 * Delete a fragment program variant. Note the caller must unlink
255 * the variant from the linked list.
256 */
257 static void
258 delete_fp_variant(struct st_context *st, struct st_fp_variant *fpv)
259 {
260 if (fpv->driver_shader)
261 cso_delete_fragment_shader(st->cso_context, fpv->driver_shader);
262 free(fpv);
263 }
264
265
266 /**
267 * Free all variants of a fragment program.
268 */
269 void
270 st_release_fp_variants(struct st_context *st, struct st_fragment_program *stfp)
271 {
272 struct st_fp_variant *fpv;
273
274 for (fpv = stfp->variants; fpv; ) {
275 struct st_fp_variant *next = fpv->next;
276 delete_fp_variant(st, fpv);
277 fpv = next;
278 }
279
280 stfp->variants = NULL;
281
282 if ((stfp->tgsi.type == PIPE_SHADER_IR_TGSI) && stfp->tgsi.tokens) {
283 ureg_free_tokens(stfp->tgsi.tokens);
284 stfp->tgsi.tokens = NULL;
285 }
286 }
287
288
289 /**
290 * Delete a basic program variant. Note the caller must unlink
291 * the variant from the linked list.
292 */
293 static void
294 delete_basic_variant(struct st_context *st, struct st_basic_variant *v,
295 GLenum target)
296 {
297 if (v->driver_shader) {
298 switch (target) {
299 case GL_TESS_CONTROL_PROGRAM_NV:
300 cso_delete_tessctrl_shader(st->cso_context, v->driver_shader);
301 break;
302 case GL_TESS_EVALUATION_PROGRAM_NV:
303 cso_delete_tesseval_shader(st->cso_context, v->driver_shader);
304 break;
305 case GL_GEOMETRY_PROGRAM_NV:
306 cso_delete_geometry_shader(st->cso_context, v->driver_shader);
307 break;
308 case GL_COMPUTE_PROGRAM_NV:
309 cso_delete_compute_shader(st->cso_context, v->driver_shader);
310 break;
311 default:
312 assert(!"this shouldn't occur");
313 }
314 }
315
316 free(v);
317 }
318
319
320 /**
321 * Free all basic program variants.
322 */
323 void
324 st_release_basic_variants(struct st_context *st, GLenum target,
325 struct st_basic_variant **variants,
326 struct pipe_shader_state *tgsi)
327 {
328 struct st_basic_variant *v;
329
330 for (v = *variants; v; ) {
331 struct st_basic_variant *next = v->next;
332 delete_basic_variant(st, v, target);
333 v = next;
334 }
335
336 *variants = NULL;
337
338 if (tgsi->tokens) {
339 ureg_free_tokens(tgsi->tokens);
340 tgsi->tokens = NULL;
341 }
342 }
343
344
345 /**
346 * Free all variants of a compute program.
347 */
348 void
349 st_release_cp_variants(struct st_context *st, struct st_compute_program *stcp)
350 {
351 struct st_basic_variant **variants = &stcp->variants;
352 struct st_basic_variant *v;
353
354 for (v = *variants; v; ) {
355 struct st_basic_variant *next = v->next;
356 delete_basic_variant(st, v, stcp->Base.Target);
357 v = next;
358 }
359
360 *variants = NULL;
361
362 if (stcp->tgsi.prog) {
363 ureg_free_tokens(stcp->tgsi.prog);
364 stcp->tgsi.prog = NULL;
365 }
366 }
367
368 /**
369 * Translate a vertex program.
370 */
371 bool
372 st_translate_vertex_program(struct st_context *st,
373 struct st_vertex_program *stvp)
374 {
375 struct ureg_program *ureg;
376 enum pipe_error error;
377 unsigned num_outputs = 0;
378 unsigned attr;
379 ubyte input_to_index[VERT_ATTRIB_MAX] = {0};
380 ubyte output_semantic_name[VARYING_SLOT_MAX] = {0};
381 ubyte output_semantic_index[VARYING_SLOT_MAX] = {0};
382
383 stvp->num_inputs = 0;
384
385 if (stvp->Base.arb.IsPositionInvariant)
386 _mesa_insert_mvp_code(st->ctx, &stvp->Base);
387
388 /*
389 * Determine number of inputs, the mappings between VERT_ATTRIB_x
390 * and TGSI generic input indexes, plus input attrib semantic info.
391 */
392 for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
393 if ((stvp->Base.info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
394 input_to_index[attr] = stvp->num_inputs;
395 stvp->index_to_input[stvp->num_inputs] = attr;
396 stvp->num_inputs++;
397 if ((stvp->Base.info.double_inputs_read &
398 BITFIELD64_BIT(attr)) != 0) {
399 /* add placeholder for second part of a double attribute */
400 stvp->index_to_input[stvp->num_inputs] = ST_DOUBLE_ATTRIB_PLACEHOLDER;
401 stvp->num_inputs++;
402 }
403 }
404 }
405 /* bit of a hack, presetup potentially unused edgeflag input */
406 input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
407 stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
408
409 /* Compute mapping of vertex program outputs to slots.
410 */
411 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
412 if ((stvp->Base.info.outputs_written & BITFIELD64_BIT(attr)) == 0) {
413 stvp->result_to_output[attr] = ~0;
414 }
415 else {
416 unsigned slot = num_outputs++;
417
418 stvp->result_to_output[attr] = slot;
419
420 unsigned semantic_name, semantic_index;
421 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
422 &semantic_name, &semantic_index);
423 output_semantic_name[slot] = semantic_name;
424 output_semantic_index[slot] = semantic_index;
425 }
426 }
427 /* similar hack to above, presetup potentially unused edgeflag output */
428 stvp->result_to_output[VARYING_SLOT_EDGE] = num_outputs;
429 output_semantic_name[num_outputs] = TGSI_SEMANTIC_EDGEFLAG;
430 output_semantic_index[num_outputs] = 0;
431
432 /* ARB_vp: */
433 if (!stvp->glsl_to_tgsi && !stvp->shader_program) {
434 _mesa_remove_output_reads(&stvp->Base, PROGRAM_OUTPUT);
435
436 /* This determines which states will be updated when the assembly
437 * shader is bound.
438 */
439 stvp->affected_states = ST_NEW_VS_STATE |
440 ST_NEW_RASTERIZER |
441 ST_NEW_VERTEX_ARRAYS;
442
443 if (stvp->Base.Parameters->NumParameters)
444 stvp->affected_states |= ST_NEW_VS_CONSTANTS;
445
446 /* No samplers are allowed in ARB_vp. */
447 }
448
449 if (stvp->shader_program) {
450 nir_shader *nir = st_glsl_to_nir(st, &stvp->Base, stvp->shader_program,
451 MESA_SHADER_VERTEX);
452
453 stvp->tgsi.type = PIPE_SHADER_IR_NIR;
454 stvp->tgsi.ir.nir = nir;
455
456 struct gl_program *prog = stvp->shader_program->last_vert_prog;
457 if (prog) {
458 st_translate_stream_output_info2(prog->sh.LinkedTransformFeedback,
459 stvp->result_to_output,
460 &stvp->tgsi.stream_output);
461 }
462
463 return true;
464 }
465
466 ureg = ureg_create_with_screen(PIPE_SHADER_VERTEX, st->pipe->screen);
467 if (ureg == NULL)
468 return false;
469
470 if (stvp->Base.info.clip_distance_array_size)
471 ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
472 stvp->Base.info.clip_distance_array_size);
473 if (stvp->Base.info.cull_distance_array_size)
474 ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
475 stvp->Base.info.cull_distance_array_size);
476
477 if (ST_DEBUG & DEBUG_MESA) {
478 _mesa_print_program(&stvp->Base);
479 _mesa_print_program_parameters(st->ctx, &stvp->Base);
480 debug_printf("\n");
481 }
482
483 if (stvp->glsl_to_tgsi) {
484 error = st_translate_program(st->ctx,
485 PIPE_SHADER_VERTEX,
486 ureg,
487 stvp->glsl_to_tgsi,
488 &stvp->Base,
489 /* inputs */
490 stvp->num_inputs,
491 input_to_index,
492 NULL, /* inputSlotToAttr */
493 NULL, /* input semantic name */
494 NULL, /* input semantic index */
495 NULL, /* interp mode */
496 /* outputs */
497 num_outputs,
498 stvp->result_to_output,
499 output_semantic_name,
500 output_semantic_index);
501
502 st_translate_stream_output_info(stvp->glsl_to_tgsi,
503 stvp->result_to_output,
504 &stvp->tgsi.stream_output);
505
506 free_glsl_to_tgsi_visitor(stvp->glsl_to_tgsi);
507 } else
508 error = st_translate_mesa_program(st->ctx,
509 PIPE_SHADER_VERTEX,
510 ureg,
511 &stvp->Base,
512 /* inputs */
513 stvp->num_inputs,
514 input_to_index,
515 NULL, /* input semantic name */
516 NULL, /* input semantic index */
517 NULL,
518 /* outputs */
519 num_outputs,
520 stvp->result_to_output,
521 output_semantic_name,
522 output_semantic_index);
523
524 if (error) {
525 debug_printf("%s: failed to translate Mesa program:\n", __func__);
526 _mesa_print_program(&stvp->Base);
527 debug_assert(0);
528 return false;
529 }
530
531 unsigned num_tokens;
532 stvp->tgsi.tokens = ureg_get_tokens(ureg, &num_tokens);
533 ureg_destroy(ureg);
534
535 if (stvp->glsl_to_tgsi) {
536 stvp->glsl_to_tgsi = NULL;
537 st_store_tgsi_in_disk_cache(st, &stvp->Base, NULL, num_tokens);
538 }
539
540 return stvp->tgsi.tokens != NULL;
541 }
542
543 static struct st_vp_variant *
544 st_create_vp_variant(struct st_context *st,
545 struct st_vertex_program *stvp,
546 const struct st_vp_variant_key *key)
547 {
548 struct st_vp_variant *vpv = CALLOC_STRUCT(st_vp_variant);
549 struct pipe_context *pipe = st->pipe;
550
551 vpv->key = *key;
552 vpv->tgsi.stream_output = stvp->tgsi.stream_output;
553 vpv->num_inputs = stvp->num_inputs;
554
555 if (stvp->tgsi.type == PIPE_SHADER_IR_NIR) {
556 vpv->tgsi.type = PIPE_SHADER_IR_NIR;
557 vpv->tgsi.ir.nir = nir_shader_clone(NULL, stvp->tgsi.ir.nir);
558 if (key->clamp_color)
559 NIR_PASS_V(vpv->tgsi.ir.nir, nir_lower_clamp_color_outputs);
560 if (key->passthrough_edgeflags) {
561 NIR_PASS_V(vpv->tgsi.ir.nir, nir_lower_passthrough_edgeflags);
562 vpv->num_inputs++;
563 }
564
565 st_finalize_nir(st, &stvp->Base, stvp->shader_program,
566 vpv->tgsi.ir.nir);
567
568 vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
569 /* driver takes ownership of IR: */
570 vpv->tgsi.ir.nir = NULL;
571 return vpv;
572 }
573
574 vpv->tgsi.tokens = tgsi_dup_tokens(stvp->tgsi.tokens);
575
576 /* Emulate features. */
577 if (key->clamp_color || key->passthrough_edgeflags) {
578 const struct tgsi_token *tokens;
579 unsigned flags =
580 (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
581 (key->passthrough_edgeflags ? TGSI_EMU_PASSTHROUGH_EDGEFLAG : 0);
582
583 tokens = tgsi_emulate(vpv->tgsi.tokens, flags);
584
585 if (tokens) {
586 tgsi_free_tokens(vpv->tgsi.tokens);
587 vpv->tgsi.tokens = tokens;
588
589 if (key->passthrough_edgeflags)
590 vpv->num_inputs++;
591 } else
592 fprintf(stderr, "mesa: cannot emulate deprecated features\n");
593 }
594
595 if (ST_DEBUG & DEBUG_TGSI) {
596 tgsi_dump(vpv->tgsi.tokens, 0);
597 debug_printf("\n");
598 }
599
600 vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
601 return vpv;
602 }
603
604
605 /**
606 * Find/create a vertex program variant.
607 */
608 struct st_vp_variant *
609 st_get_vp_variant(struct st_context *st,
610 struct st_vertex_program *stvp,
611 const struct st_vp_variant_key *key)
612 {
613 struct st_vp_variant *vpv;
614
615 /* Search for existing variant */
616 for (vpv = stvp->variants; vpv; vpv = vpv->next) {
617 if (memcmp(&vpv->key, key, sizeof(*key)) == 0) {
618 break;
619 }
620 }
621
622 if (!vpv) {
623 /* create now */
624 vpv = st_create_vp_variant(st, stvp, key);
625 if (vpv) {
626 /* insert into list */
627 vpv->next = stvp->variants;
628 stvp->variants = vpv;
629 }
630 }
631
632 return vpv;
633 }
634
635
636 /**
637 * Translate a Mesa fragment shader into a TGSI shader.
638 */
639 bool
640 st_translate_fragment_program(struct st_context *st,
641 struct st_fragment_program *stfp)
642 {
643 ubyte outputMapping[2 * FRAG_RESULT_MAX];
644 ubyte inputMapping[VARYING_SLOT_MAX];
645 ubyte inputSlotToAttr[VARYING_SLOT_MAX];
646 ubyte interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
647 GLuint attr;
648 GLbitfield64 inputsRead;
649 struct ureg_program *ureg;
650
651 GLboolean write_all = GL_FALSE;
652
653 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
654 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
655 uint fs_num_inputs = 0;
656
657 ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
658 ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
659 uint fs_num_outputs = 0;
660
661 memset(inputSlotToAttr, ~0, sizeof(inputSlotToAttr));
662
663 /* Non-GLSL programs: */
664 if (!stfp->glsl_to_tgsi && !stfp->shader_program) {
665 _mesa_remove_output_reads(&stfp->Base, PROGRAM_OUTPUT);
666 if (st->ctx->Const.GLSLFragCoordIsSysVal)
667 _mesa_program_fragment_position_to_sysval(&stfp->Base);
668
669 /* This determines which states will be updated when the assembly
670 * shader is bound.
671 *
672 * fragment.position and glDrawPixels always use constants.
673 */
674 stfp->affected_states = ST_NEW_FS_STATE |
675 ST_NEW_SAMPLE_SHADING |
676 ST_NEW_FS_CONSTANTS;
677
678 if (stfp->ati_fs) {
679 /* Just set them for ATI_fs unconditionally. */
680 stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
681 ST_NEW_FS_SAMPLERS;
682 } else {
683 /* ARB_fp */
684 if (stfp->Base.SamplersUsed)
685 stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
686 ST_NEW_FS_SAMPLERS;
687 }
688 }
689
690 /*
691 * Convert Mesa program inputs to TGSI input register semantics.
692 */
693 inputsRead = stfp->Base.info.inputs_read;
694 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
695 if ((inputsRead & BITFIELD64_BIT(attr)) != 0) {
696 const GLuint slot = fs_num_inputs++;
697
698 inputMapping[attr] = slot;
699 inputSlotToAttr[slot] = attr;
700
701 switch (attr) {
702 case VARYING_SLOT_POS:
703 input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
704 input_semantic_index[slot] = 0;
705 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
706 break;
707 case VARYING_SLOT_COL0:
708 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
709 input_semantic_index[slot] = 0;
710 interpMode[slot] = stfp->glsl_to_tgsi ?
711 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
712 break;
713 case VARYING_SLOT_COL1:
714 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
715 input_semantic_index[slot] = 1;
716 interpMode[slot] = stfp->glsl_to_tgsi ?
717 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
718 break;
719 case VARYING_SLOT_FOGC:
720 input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
721 input_semantic_index[slot] = 0;
722 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
723 break;
724 case VARYING_SLOT_FACE:
725 input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
726 input_semantic_index[slot] = 0;
727 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
728 break;
729 case VARYING_SLOT_PRIMITIVE_ID:
730 input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
731 input_semantic_index[slot] = 0;
732 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
733 break;
734 case VARYING_SLOT_LAYER:
735 input_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
736 input_semantic_index[slot] = 0;
737 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
738 break;
739 case VARYING_SLOT_VIEWPORT:
740 input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
741 input_semantic_index[slot] = 0;
742 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
743 break;
744 case VARYING_SLOT_CLIP_DIST0:
745 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
746 input_semantic_index[slot] = 0;
747 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
748 break;
749 case VARYING_SLOT_CLIP_DIST1:
750 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
751 input_semantic_index[slot] = 1;
752 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
753 break;
754 case VARYING_SLOT_CULL_DIST0:
755 case VARYING_SLOT_CULL_DIST1:
756 /* these should have been lowered by GLSL */
757 assert(0);
758 break;
759 /* In most cases, there is nothing special about these
760 * inputs, so adopt a convention to use the generic
761 * semantic name and the mesa VARYING_SLOT_ number as the
762 * index.
763 *
764 * All that is required is that the vertex shader labels
765 * its own outputs similarly, and that the vertex shader
766 * generates at least every output required by the
767 * fragment shader plus fixed-function hardware (such as
768 * BFC).
769 *
770 * However, some drivers may need us to identify the PNTC and TEXi
771 * varyings if, for example, their capability to replace them with
772 * sprite coordinates is limited.
773 */
774 case VARYING_SLOT_PNTC:
775 if (st->needs_texcoord_semantic) {
776 input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
777 input_semantic_index[slot] = 0;
778 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
779 break;
780 }
781 /* fall through */
782 case VARYING_SLOT_TEX0:
783 case VARYING_SLOT_TEX1:
784 case VARYING_SLOT_TEX2:
785 case VARYING_SLOT_TEX3:
786 case VARYING_SLOT_TEX4:
787 case VARYING_SLOT_TEX5:
788 case VARYING_SLOT_TEX6:
789 case VARYING_SLOT_TEX7:
790 if (st->needs_texcoord_semantic) {
791 input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
792 input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
793 interpMode[slot] = stfp->glsl_to_tgsi ?
794 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
795 break;
796 }
797 /* fall through */
798 case VARYING_SLOT_VAR0:
799 default:
800 /* Semantic indices should be zero-based because drivers may choose
801 * to assign a fixed slot determined by that index.
802 * This is useful because ARB_separate_shader_objects uses location
803 * qualifiers for linkage, and if the semantic index corresponds to
804 * these locations, linkage passes in the driver become unecessary.
805 *
806 * If needs_texcoord_semantic is true, no semantic indices will be
807 * consumed for the TEXi varyings, and we can base the locations of
808 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
809 */
810 assert(attr >= VARYING_SLOT_VAR0 || attr == VARYING_SLOT_PNTC ||
811 (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
812 input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
813 input_semantic_index[slot] = st_get_generic_varying_index(st, attr);
814 if (attr == VARYING_SLOT_PNTC)
815 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
816 else {
817 interpMode[slot] = stfp->glsl_to_tgsi ?
818 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
819 }
820 break;
821 }
822 }
823 else {
824 inputMapping[attr] = -1;
825 }
826 }
827
828 /*
829 * Semantics and mapping for outputs
830 */
831 GLbitfield64 outputsWritten = stfp->Base.info.outputs_written;
832
833 /* if z is written, emit that first */
834 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
835 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
836 fs_output_semantic_index[fs_num_outputs] = 0;
837 outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
838 fs_num_outputs++;
839 outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
840 }
841
842 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
843 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
844 fs_output_semantic_index[fs_num_outputs] = 0;
845 outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
846 fs_num_outputs++;
847 outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
848 }
849
850 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) {
851 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK;
852 fs_output_semantic_index[fs_num_outputs] = 0;
853 outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs;
854 fs_num_outputs++;
855 outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK);
856 }
857
858 /* handle remaining outputs (color) */
859 for (attr = 0; attr < ARRAY_SIZE(outputMapping); attr++) {
860 const GLbitfield64 written = attr < FRAG_RESULT_MAX ? outputsWritten :
861 stfp->Base.SecondaryOutputsWritten;
862 const unsigned loc = attr % FRAG_RESULT_MAX;
863
864 if (written & BITFIELD64_BIT(loc)) {
865 switch (loc) {
866 case FRAG_RESULT_DEPTH:
867 case FRAG_RESULT_STENCIL:
868 case FRAG_RESULT_SAMPLE_MASK:
869 /* handled above */
870 assert(0);
871 break;
872 case FRAG_RESULT_COLOR:
873 write_all = GL_TRUE; /* fallthrough */
874 default: {
875 int index;
876 assert(loc == FRAG_RESULT_COLOR ||
877 (FRAG_RESULT_DATA0 <= loc && loc < FRAG_RESULT_MAX));
878
879 index = (loc == FRAG_RESULT_COLOR) ? 0 : (loc - FRAG_RESULT_DATA0);
880
881 if (attr >= FRAG_RESULT_MAX) {
882 /* Secondary color for dual source blending. */
883 assert(index == 0);
884 index++;
885 }
886
887 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
888 fs_output_semantic_index[fs_num_outputs] = index;
889 outputMapping[attr] = fs_num_outputs;
890 break;
891 }
892 }
893
894 fs_num_outputs++;
895 }
896 }
897
898 if (stfp->shader_program) {
899 nir_shader *nir = st_glsl_to_nir(st, &stfp->Base, stfp->shader_program,
900 MESA_SHADER_FRAGMENT);
901
902 stfp->tgsi.type = PIPE_SHADER_IR_NIR;
903 stfp->tgsi.ir.nir = nir;
904
905 return true;
906 }
907
908 ureg = ureg_create_with_screen(PIPE_SHADER_FRAGMENT, st->pipe->screen);
909 if (ureg == NULL)
910 return false;
911
912 if (ST_DEBUG & DEBUG_MESA) {
913 _mesa_print_program(&stfp->Base);
914 _mesa_print_program_parameters(st->ctx, &stfp->Base);
915 debug_printf("\n");
916 }
917 if (write_all == GL_TRUE)
918 ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
919
920 if (stfp->Base.info.fs.depth_layout != FRAG_DEPTH_LAYOUT_NONE) {
921 switch (stfp->Base.info.fs.depth_layout) {
922 case FRAG_DEPTH_LAYOUT_ANY:
923 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
924 TGSI_FS_DEPTH_LAYOUT_ANY);
925 break;
926 case FRAG_DEPTH_LAYOUT_GREATER:
927 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
928 TGSI_FS_DEPTH_LAYOUT_GREATER);
929 break;
930 case FRAG_DEPTH_LAYOUT_LESS:
931 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
932 TGSI_FS_DEPTH_LAYOUT_LESS);
933 break;
934 case FRAG_DEPTH_LAYOUT_UNCHANGED:
935 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
936 TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
937 break;
938 default:
939 assert(0);
940 }
941 }
942
943 if (stfp->glsl_to_tgsi) {
944 st_translate_program(st->ctx,
945 PIPE_SHADER_FRAGMENT,
946 ureg,
947 stfp->glsl_to_tgsi,
948 &stfp->Base,
949 /* inputs */
950 fs_num_inputs,
951 inputMapping,
952 inputSlotToAttr,
953 input_semantic_name,
954 input_semantic_index,
955 interpMode,
956 /* outputs */
957 fs_num_outputs,
958 outputMapping,
959 fs_output_semantic_name,
960 fs_output_semantic_index);
961
962 free_glsl_to_tgsi_visitor(stfp->glsl_to_tgsi);
963 } else if (stfp->ati_fs)
964 st_translate_atifs_program(ureg,
965 stfp->ati_fs,
966 &stfp->Base,
967 /* inputs */
968 fs_num_inputs,
969 inputMapping,
970 input_semantic_name,
971 input_semantic_index,
972 interpMode,
973 /* outputs */
974 fs_num_outputs,
975 outputMapping,
976 fs_output_semantic_name,
977 fs_output_semantic_index);
978 else
979 st_translate_mesa_program(st->ctx,
980 PIPE_SHADER_FRAGMENT,
981 ureg,
982 &stfp->Base,
983 /* inputs */
984 fs_num_inputs,
985 inputMapping,
986 input_semantic_name,
987 input_semantic_index,
988 interpMode,
989 /* outputs */
990 fs_num_outputs,
991 outputMapping,
992 fs_output_semantic_name,
993 fs_output_semantic_index);
994
995 unsigned num_tokens;
996 stfp->tgsi.tokens = ureg_get_tokens(ureg, &num_tokens);
997 ureg_destroy(ureg);
998
999 if (stfp->glsl_to_tgsi) {
1000 stfp->glsl_to_tgsi = NULL;
1001 st_store_tgsi_in_disk_cache(st, &stfp->Base, NULL, num_tokens);
1002 }
1003
1004 return stfp->tgsi.tokens != NULL;
1005 }
1006
1007 static struct st_fp_variant *
1008 st_create_fp_variant(struct st_context *st,
1009 struct st_fragment_program *stfp,
1010 const struct st_fp_variant_key *key)
1011 {
1012 struct pipe_context *pipe = st->pipe;
1013 struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
1014 struct pipe_shader_state tgsi = {0};
1015 struct gl_program_parameter_list *params = stfp->Base.Parameters;
1016 static const gl_state_index texcoord_state[STATE_LENGTH] =
1017 { STATE_INTERNAL, STATE_CURRENT_ATTRIB, VERT_ATTRIB_TEX0 };
1018 static const gl_state_index scale_state[STATE_LENGTH] =
1019 { STATE_INTERNAL, STATE_PT_SCALE };
1020 static const gl_state_index bias_state[STATE_LENGTH] =
1021 { STATE_INTERNAL, STATE_PT_BIAS };
1022
1023 if (!variant)
1024 return NULL;
1025
1026 if (stfp->tgsi.type == PIPE_SHADER_IR_NIR) {
1027 tgsi.type = PIPE_SHADER_IR_NIR;
1028 tgsi.ir.nir = nir_shader_clone(NULL, stfp->tgsi.ir.nir);
1029
1030 if (key->clamp_color)
1031 NIR_PASS_V(tgsi.ir.nir, nir_lower_clamp_color_outputs);
1032
1033 if (key->persample_shading) {
1034 nir_shader *shader = tgsi.ir.nir;
1035 nir_foreach_variable(var, &shader->inputs)
1036 var->data.sample = true;
1037 }
1038
1039 assert(!(key->bitmap && key->drawpixels));
1040
1041 /* glBitmap */
1042 if (key->bitmap) {
1043 nir_lower_bitmap_options options = {0};
1044
1045 variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1046 options.sampler = variant->bitmap_sampler;
1047 options.swizzle_xxxx = (st->bitmap.tex_format == PIPE_FORMAT_L8_UNORM);
1048
1049 NIR_PASS_V(tgsi.ir.nir, nir_lower_bitmap, &options);
1050 }
1051
1052 /* glDrawPixels (color only) */
1053 if (key->drawpixels) {
1054 nir_lower_drawpixels_options options = {{0}};
1055 unsigned samplers_used = stfp->Base.SamplersUsed;
1056
1057 /* Find the first unused slot. */
1058 variant->drawpix_sampler = ffs(~samplers_used) - 1;
1059 options.drawpix_sampler = variant->drawpix_sampler;
1060 samplers_used |= (1 << variant->drawpix_sampler);
1061
1062 options.pixel_maps = key->pixelMaps;
1063 if (key->pixelMaps) {
1064 variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1065 options.pixelmap_sampler = variant->pixelmap_sampler;
1066 }
1067
1068 options.scale_and_bias = key->scaleAndBias;
1069 if (key->scaleAndBias) {
1070 _mesa_add_state_reference(params, scale_state);
1071 memcpy(options.scale_state_tokens, scale_state,
1072 sizeof(options.scale_state_tokens));
1073 _mesa_add_state_reference(params, bias_state);
1074 memcpy(options.bias_state_tokens, bias_state,
1075 sizeof(options.bias_state_tokens));
1076 }
1077
1078 _mesa_add_state_reference(params, texcoord_state);
1079 memcpy(options.texcoord_state_tokens, texcoord_state,
1080 sizeof(options.texcoord_state_tokens));
1081
1082 NIR_PASS_V(tgsi.ir.nir, nir_lower_drawpixels, &options);
1083 }
1084
1085 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv)) {
1086 nir_lower_tex_options options = {0};
1087 options.lower_y_uv_external = key->external.lower_nv12;
1088 options.lower_y_u_v_external = key->external.lower_iyuv;
1089 NIR_PASS_V(tgsi.ir.nir, nir_lower_tex, &options);
1090 }
1091
1092 st_finalize_nir(st, &stfp->Base, stfp->shader_program, tgsi.ir.nir);
1093
1094 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv)) {
1095 /* This pass needs to happen *after* nir_lower_sampler */
1096 NIR_PASS_V(tgsi.ir.nir, st_nir_lower_tex_src_plane,
1097 ~stfp->Base.SamplersUsed,
1098 key->external.lower_nv12,
1099 key->external.lower_iyuv);
1100 }
1101
1102 variant->driver_shader = pipe->create_fs_state(pipe, &tgsi);
1103 variant->key = *key;
1104
1105 return variant;
1106 }
1107
1108 tgsi.tokens = stfp->tgsi.tokens;
1109
1110 assert(!(key->bitmap && key->drawpixels));
1111
1112 /* Fix texture targets and add fog for ATI_fs */
1113 if (stfp->ati_fs) {
1114 const struct tgsi_token *tokens = st_fixup_atifs(tgsi.tokens, key);
1115
1116 if (tokens)
1117 tgsi.tokens = tokens;
1118 else
1119 fprintf(stderr, "mesa: cannot post-process ATI_fs\n");
1120 }
1121
1122 /* Emulate features. */
1123 if (key->clamp_color || key->persample_shading) {
1124 const struct tgsi_token *tokens;
1125 unsigned flags =
1126 (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
1127 (key->persample_shading ? TGSI_EMU_FORCE_PERSAMPLE_INTERP : 0);
1128
1129 tokens = tgsi_emulate(tgsi.tokens, flags);
1130
1131 if (tokens) {
1132 if (tgsi.tokens != stfp->tgsi.tokens)
1133 tgsi_free_tokens(tgsi.tokens);
1134 tgsi.tokens = tokens;
1135 } else
1136 fprintf(stderr, "mesa: cannot emulate deprecated features\n");
1137 }
1138
1139 /* glBitmap */
1140 if (key->bitmap) {
1141 const struct tgsi_token *tokens;
1142
1143 variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1144
1145 tokens = st_get_bitmap_shader(tgsi.tokens,
1146 st->internal_target,
1147 variant->bitmap_sampler,
1148 st->needs_texcoord_semantic,
1149 st->bitmap.tex_format ==
1150 PIPE_FORMAT_L8_UNORM);
1151
1152 if (tokens) {
1153 if (tgsi.tokens != stfp->tgsi.tokens)
1154 tgsi_free_tokens(tgsi.tokens);
1155 tgsi.tokens = tokens;
1156 } else
1157 fprintf(stderr, "mesa: cannot create a shader for glBitmap\n");
1158 }
1159
1160 /* glDrawPixels (color only) */
1161 if (key->drawpixels) {
1162 const struct tgsi_token *tokens;
1163 unsigned scale_const = 0, bias_const = 0, texcoord_const = 0;
1164
1165 /* Find the first unused slot. */
1166 variant->drawpix_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1167
1168 if (key->pixelMaps) {
1169 unsigned samplers_used = stfp->Base.SamplersUsed |
1170 (1 << variant->drawpix_sampler);
1171
1172 variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1173 }
1174
1175 if (key->scaleAndBias) {
1176 scale_const = _mesa_add_state_reference(params, scale_state);
1177 bias_const = _mesa_add_state_reference(params, bias_state);
1178 }
1179
1180 texcoord_const = _mesa_add_state_reference(params, texcoord_state);
1181
1182 tokens = st_get_drawpix_shader(tgsi.tokens,
1183 st->needs_texcoord_semantic,
1184 key->scaleAndBias, scale_const,
1185 bias_const, key->pixelMaps,
1186 variant->drawpix_sampler,
1187 variant->pixelmap_sampler,
1188 texcoord_const, st->internal_target);
1189
1190 if (tokens) {
1191 if (tgsi.tokens != stfp->tgsi.tokens)
1192 tgsi_free_tokens(tgsi.tokens);
1193 tgsi.tokens = tokens;
1194 } else
1195 fprintf(stderr, "mesa: cannot create a shader for glDrawPixels\n");
1196 }
1197
1198 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv)) {
1199 const struct tgsi_token *tokens;
1200
1201 /* samplers inserted would conflict, but this should be unpossible: */
1202 assert(!(key->bitmap || key->drawpixels));
1203
1204 tokens = st_tgsi_lower_yuv(tgsi.tokens,
1205 ~stfp->Base.SamplersUsed,
1206 key->external.lower_nv12,
1207 key->external.lower_iyuv);
1208 if (tokens) {
1209 if (tgsi.tokens != stfp->tgsi.tokens)
1210 tgsi_free_tokens(tgsi.tokens);
1211 tgsi.tokens = tokens;
1212 } else {
1213 fprintf(stderr, "mesa: cannot create a shader for samplerExternalOES\n");
1214 }
1215 }
1216
1217 if (ST_DEBUG & DEBUG_TGSI) {
1218 tgsi_dump(tgsi.tokens, 0);
1219 debug_printf("\n");
1220 }
1221
1222 /* fill in variant */
1223 variant->driver_shader = pipe->create_fs_state(pipe, &tgsi);
1224 variant->key = *key;
1225
1226 if (tgsi.tokens != stfp->tgsi.tokens)
1227 tgsi_free_tokens(tgsi.tokens);
1228 return variant;
1229 }
1230
1231 /**
1232 * Translate fragment program if needed.
1233 */
1234 struct st_fp_variant *
1235 st_get_fp_variant(struct st_context *st,
1236 struct st_fragment_program *stfp,
1237 const struct st_fp_variant_key *key)
1238 {
1239 struct st_fp_variant *fpv;
1240
1241 /* Search for existing variant */
1242 for (fpv = stfp->variants; fpv; fpv = fpv->next) {
1243 if (memcmp(&fpv->key, key, sizeof(*key)) == 0) {
1244 break;
1245 }
1246 }
1247
1248 if (!fpv) {
1249 /* create new */
1250 fpv = st_create_fp_variant(st, stfp, key);
1251 if (fpv) {
1252 if (key->bitmap || key->drawpixels) {
1253 /* Regular variants should always come before the
1254 * bitmap & drawpixels variants, (unless there
1255 * are no regular variants) so that
1256 * st_update_fp can take a fast path when
1257 * shader_has_one_variant is set.
1258 */
1259 if (!stfp->variants) {
1260 stfp->variants = fpv;
1261 } else {
1262 /* insert into list after the first one */
1263 fpv->next = stfp->variants->next;
1264 stfp->variants->next = fpv;
1265 }
1266 } else {
1267 /* insert into list */
1268 fpv->next = stfp->variants;
1269 stfp->variants = fpv;
1270 }
1271 }
1272 }
1273
1274 return fpv;
1275 }
1276
1277
1278 /**
1279 * Translate a program. This is common code for geometry and tessellation
1280 * shaders.
1281 */
1282 static void
1283 st_translate_program_common(struct st_context *st,
1284 struct gl_program *prog,
1285 struct glsl_to_tgsi_visitor *glsl_to_tgsi,
1286 struct ureg_program *ureg,
1287 unsigned tgsi_processor,
1288 struct pipe_shader_state *out_state)
1289 {
1290 ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
1291 ubyte inputMapping[VARYING_SLOT_TESS_MAX];
1292 ubyte outputMapping[VARYING_SLOT_TESS_MAX];
1293 GLuint attr;
1294
1295 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
1296 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
1297 uint num_inputs = 0;
1298
1299 ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
1300 ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
1301 uint num_outputs = 0;
1302
1303 GLint i;
1304
1305 memset(inputSlotToAttr, 0, sizeof(inputSlotToAttr));
1306 memset(inputMapping, 0, sizeof(inputMapping));
1307 memset(outputMapping, 0, sizeof(outputMapping));
1308 memset(out_state, 0, sizeof(*out_state));
1309
1310 if (prog->info.clip_distance_array_size)
1311 ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
1312 prog->info.clip_distance_array_size);
1313 if (prog->info.cull_distance_array_size)
1314 ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
1315 prog->info.cull_distance_array_size);
1316
1317 /*
1318 * Convert Mesa program inputs to TGSI input register semantics.
1319 */
1320 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1321 if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
1322 const GLuint slot = num_inputs++;
1323
1324 inputMapping[attr] = slot;
1325 inputSlotToAttr[slot] = attr;
1326
1327 switch (attr) {
1328 case VARYING_SLOT_PRIMITIVE_ID:
1329 assert(tgsi_processor == PIPE_SHADER_GEOMETRY);
1330 input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
1331 input_semantic_index[slot] = 0;
1332 break;
1333 case VARYING_SLOT_POS:
1334 input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
1335 input_semantic_index[slot] = 0;
1336 break;
1337 case VARYING_SLOT_COL0:
1338 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
1339 input_semantic_index[slot] = 0;
1340 break;
1341 case VARYING_SLOT_COL1:
1342 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
1343 input_semantic_index[slot] = 1;
1344 break;
1345 case VARYING_SLOT_FOGC:
1346 input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
1347 input_semantic_index[slot] = 0;
1348 break;
1349 case VARYING_SLOT_CLIP_VERTEX:
1350 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
1351 input_semantic_index[slot] = 0;
1352 break;
1353 case VARYING_SLOT_CLIP_DIST0:
1354 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
1355 input_semantic_index[slot] = 0;
1356 break;
1357 case VARYING_SLOT_CLIP_DIST1:
1358 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
1359 input_semantic_index[slot] = 1;
1360 break;
1361 case VARYING_SLOT_CULL_DIST0:
1362 case VARYING_SLOT_CULL_DIST1:
1363 /* these should have been lowered by GLSL */
1364 assert(0);
1365 break;
1366 case VARYING_SLOT_PSIZ:
1367 input_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
1368 input_semantic_index[slot] = 0;
1369 break;
1370 case VARYING_SLOT_TEX0:
1371 case VARYING_SLOT_TEX1:
1372 case VARYING_SLOT_TEX2:
1373 case VARYING_SLOT_TEX3:
1374 case VARYING_SLOT_TEX4:
1375 case VARYING_SLOT_TEX5:
1376 case VARYING_SLOT_TEX6:
1377 case VARYING_SLOT_TEX7:
1378 if (st->needs_texcoord_semantic) {
1379 input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
1380 input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
1381 break;
1382 }
1383 /* fall through */
1384 case VARYING_SLOT_VAR0:
1385 default:
1386 assert(attr >= VARYING_SLOT_VAR0 ||
1387 (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
1388 input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
1389 input_semantic_index[slot] =
1390 st_get_generic_varying_index(st, attr);
1391 break;
1392 }
1393 }
1394 }
1395
1396 /* Also add patch inputs. */
1397 for (attr = 0; attr < 32; attr++) {
1398 if (prog->info.patch_inputs_read & (1u << attr)) {
1399 GLuint slot = num_inputs++;
1400 GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1401
1402 inputMapping[patch_attr] = slot;
1403 inputSlotToAttr[slot] = patch_attr;
1404 input_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1405 input_semantic_index[slot] = attr;
1406 }
1407 }
1408
1409 /* initialize output semantics to defaults */
1410 for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
1411 output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
1412 output_semantic_index[i] = 0;
1413 }
1414
1415 /*
1416 * Determine number of outputs, the (default) output register
1417 * mapping and the semantic information for each output.
1418 */
1419 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1420 if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
1421 GLuint slot = num_outputs++;
1422
1423 outputMapping[attr] = slot;
1424
1425 switch (attr) {
1426 case VARYING_SLOT_POS:
1427 assert(slot == 0);
1428 output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
1429 output_semantic_index[slot] = 0;
1430 break;
1431 case VARYING_SLOT_COL0:
1432 output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
1433 output_semantic_index[slot] = 0;
1434 break;
1435 case VARYING_SLOT_COL1:
1436 output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
1437 output_semantic_index[slot] = 1;
1438 break;
1439 case VARYING_SLOT_BFC0:
1440 output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
1441 output_semantic_index[slot] = 0;
1442 break;
1443 case VARYING_SLOT_BFC1:
1444 output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
1445 output_semantic_index[slot] = 1;
1446 break;
1447 case VARYING_SLOT_FOGC:
1448 output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
1449 output_semantic_index[slot] = 0;
1450 break;
1451 case VARYING_SLOT_PSIZ:
1452 output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
1453 output_semantic_index[slot] = 0;
1454 break;
1455 case VARYING_SLOT_CLIP_VERTEX:
1456 output_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
1457 output_semantic_index[slot] = 0;
1458 break;
1459 case VARYING_SLOT_CLIP_DIST0:
1460 output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
1461 output_semantic_index[slot] = 0;
1462 break;
1463 case VARYING_SLOT_CLIP_DIST1:
1464 output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
1465 output_semantic_index[slot] = 1;
1466 break;
1467 case VARYING_SLOT_CULL_DIST0:
1468 case VARYING_SLOT_CULL_DIST1:
1469 /* these should have been lowered by GLSL */
1470 assert(0);
1471 break;
1472 case VARYING_SLOT_LAYER:
1473 output_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
1474 output_semantic_index[slot] = 0;
1475 break;
1476 case VARYING_SLOT_PRIMITIVE_ID:
1477 output_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
1478 output_semantic_index[slot] = 0;
1479 break;
1480 case VARYING_SLOT_VIEWPORT:
1481 output_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
1482 output_semantic_index[slot] = 0;
1483 break;
1484 case VARYING_SLOT_TESS_LEVEL_OUTER:
1485 output_semantic_name[slot] = TGSI_SEMANTIC_TESSOUTER;
1486 output_semantic_index[slot] = 0;
1487 break;
1488 case VARYING_SLOT_TESS_LEVEL_INNER:
1489 output_semantic_name[slot] = TGSI_SEMANTIC_TESSINNER;
1490 output_semantic_index[slot] = 0;
1491 break;
1492 case VARYING_SLOT_TEX0:
1493 case VARYING_SLOT_TEX1:
1494 case VARYING_SLOT_TEX2:
1495 case VARYING_SLOT_TEX3:
1496 case VARYING_SLOT_TEX4:
1497 case VARYING_SLOT_TEX5:
1498 case VARYING_SLOT_TEX6:
1499 case VARYING_SLOT_TEX7:
1500 if (st->needs_texcoord_semantic) {
1501 output_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
1502 output_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
1503 break;
1504 }
1505 /* fall through */
1506 case VARYING_SLOT_VAR0:
1507 default:
1508 assert(slot < ARRAY_SIZE(output_semantic_name));
1509 assert(attr >= VARYING_SLOT_VAR0 ||
1510 (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
1511 output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
1512 output_semantic_index[slot] =
1513 st_get_generic_varying_index(st, attr);
1514 break;
1515 }
1516 }
1517 }
1518
1519 /* Also add patch outputs. */
1520 for (attr = 0; attr < 32; attr++) {
1521 if (prog->info.patch_outputs_written & (1u << attr)) {
1522 GLuint slot = num_outputs++;
1523 GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1524
1525 outputMapping[patch_attr] = slot;
1526 output_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1527 output_semantic_index[slot] = attr;
1528 }
1529 }
1530
1531 st_translate_program(st->ctx,
1532 tgsi_processor,
1533 ureg,
1534 glsl_to_tgsi,
1535 prog,
1536 /* inputs */
1537 num_inputs,
1538 inputMapping,
1539 inputSlotToAttr,
1540 input_semantic_name,
1541 input_semantic_index,
1542 NULL,
1543 /* outputs */
1544 num_outputs,
1545 outputMapping,
1546 output_semantic_name,
1547 output_semantic_index);
1548
1549 unsigned num_tokens;
1550 out_state->tokens = ureg_get_tokens(ureg, &num_tokens);
1551 ureg_destroy(ureg);
1552
1553 st_translate_stream_output_info(glsl_to_tgsi,
1554 outputMapping,
1555 &out_state->stream_output);
1556
1557 st_store_tgsi_in_disk_cache(st, prog, out_state, num_tokens);
1558
1559 if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
1560 _mesa_print_program(prog);
1561 debug_printf("\n");
1562 }
1563
1564 if (ST_DEBUG & DEBUG_TGSI) {
1565 tgsi_dump(out_state->tokens, 0);
1566 debug_printf("\n");
1567 }
1568 }
1569
1570
1571 /**
1572 * Translate a geometry program to create a new variant.
1573 */
1574 bool
1575 st_translate_geometry_program(struct st_context *st,
1576 struct st_common_program *stgp)
1577 {
1578 struct ureg_program *ureg;
1579
1580 if (stgp->shader_program) {
1581 nir_shader *nir = st_glsl_to_nir(st, &stgp->Base, stgp->shader_program,
1582 MESA_SHADER_GEOMETRY);
1583
1584 stgp->tgsi.type = PIPE_SHADER_IR_NIR;
1585 stgp->tgsi.ir.nir = nir;
1586
1587 return true;
1588 }
1589
1590 ureg = ureg_create_with_screen(PIPE_SHADER_GEOMETRY, st->pipe->screen);
1591 if (ureg == NULL)
1592 return false;
1593
1594 ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM,
1595 stgp->Base.info.gs.input_primitive);
1596 ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM,
1597 stgp->Base.info.gs.output_primitive);
1598 ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
1599 stgp->Base.info.gs.vertices_out);
1600 ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS,
1601 stgp->Base.info.gs.invocations);
1602
1603 st_translate_program_common(st, &stgp->Base, stgp->glsl_to_tgsi, ureg,
1604 PIPE_SHADER_GEOMETRY, &stgp->tgsi);
1605
1606 free_glsl_to_tgsi_visitor(stgp->glsl_to_tgsi);
1607 stgp->glsl_to_tgsi = NULL;
1608 return true;
1609 }
1610
1611
1612 /**
1613 * Get/create a basic program variant.
1614 */
1615 struct st_basic_variant *
1616 st_get_basic_variant(struct st_context *st,
1617 unsigned pipe_shader,
1618 struct st_common_program *prog)
1619 {
1620 struct pipe_context *pipe = st->pipe;
1621 struct st_basic_variant *v;
1622 struct st_basic_variant_key key;
1623 struct pipe_shader_state tgsi = {0};
1624 memset(&key, 0, sizeof(key));
1625 key.st = st->has_shareable_shaders ? NULL : st;
1626
1627 /* Search for existing variant */
1628 for (v = prog->variants; v; v = v->next) {
1629 if (memcmp(&v->key, &key, sizeof(key)) == 0) {
1630 break;
1631 }
1632 }
1633
1634 if (!v) {
1635 /* create new */
1636 v = CALLOC_STRUCT(st_basic_variant);
1637 if (v) {
1638
1639 if (prog->tgsi.type == PIPE_SHADER_IR_NIR) {
1640 tgsi.type = PIPE_SHADER_IR_NIR;
1641 tgsi.ir.nir = nir_shader_clone(NULL, prog->tgsi.ir.nir);
1642 st_finalize_nir(st, &prog->Base, prog->shader_program,
1643 tgsi.ir.nir);
1644 } else
1645 tgsi = prog->tgsi;
1646 /* fill in new variant */
1647 switch (pipe_shader) {
1648 case PIPE_SHADER_TESS_CTRL:
1649 v->driver_shader = pipe->create_tcs_state(pipe, &tgsi);
1650 break;
1651 case PIPE_SHADER_TESS_EVAL:
1652 v->driver_shader = pipe->create_tes_state(pipe, &tgsi);
1653 break;
1654 case PIPE_SHADER_GEOMETRY:
1655 v->driver_shader = pipe->create_gs_state(pipe, &tgsi);
1656 break;
1657 default:
1658 assert(!"unhandled shader type");
1659 free(v);
1660 return NULL;
1661 }
1662
1663 v->key = key;
1664
1665 /* insert into list */
1666 v->next = prog->variants;
1667 prog->variants = v;
1668 }
1669 }
1670
1671 return v;
1672 }
1673
1674
1675 /**
1676 * Translate a tessellation control program to create a new variant.
1677 */
1678 bool
1679 st_translate_tessctrl_program(struct st_context *st,
1680 struct st_common_program *sttcp)
1681 {
1682 struct ureg_program *ureg;
1683
1684 if (sttcp->shader_program) {
1685 nir_shader *nir = st_glsl_to_nir(st, &sttcp->Base, sttcp->shader_program,
1686 MESA_SHADER_TESS_EVAL);
1687
1688 sttcp->tgsi.type = PIPE_SHADER_IR_NIR;
1689 sttcp->tgsi.ir.nir = nir;
1690
1691 return true;
1692 }
1693
1694 ureg = ureg_create_with_screen(PIPE_SHADER_TESS_CTRL, st->pipe->screen);
1695 if (ureg == NULL)
1696 return false;
1697
1698 ureg_property(ureg, TGSI_PROPERTY_TCS_VERTICES_OUT,
1699 sttcp->Base.info.tess.tcs_vertices_out);
1700
1701 st_translate_program_common(st, &sttcp->Base, sttcp->glsl_to_tgsi, ureg,
1702 PIPE_SHADER_TESS_CTRL, &sttcp->tgsi);
1703
1704 free_glsl_to_tgsi_visitor(sttcp->glsl_to_tgsi);
1705 sttcp->glsl_to_tgsi = NULL;
1706 return true;
1707 }
1708
1709
1710 /**
1711 * Translate a tessellation evaluation program to create a new variant.
1712 */
1713 bool
1714 st_translate_tesseval_program(struct st_context *st,
1715 struct st_common_program *sttep)
1716 {
1717 struct ureg_program *ureg;
1718
1719 if (sttep->shader_program) {
1720 nir_shader *nir = st_glsl_to_nir(st, &sttep->Base, sttep->shader_program,
1721 MESA_SHADER_TESS_EVAL);
1722
1723 sttep->tgsi.type = PIPE_SHADER_IR_NIR;
1724 sttep->tgsi.ir.nir = nir;
1725
1726 return true;
1727 }
1728
1729 ureg = ureg_create_with_screen(PIPE_SHADER_TESS_EVAL, st->pipe->screen);
1730 if (ureg == NULL)
1731 return false;
1732
1733 if (sttep->Base.info.tess.primitive_mode == GL_ISOLINES)
1734 ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE, GL_LINES);
1735 else
1736 ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE,
1737 sttep->Base.info.tess.primitive_mode);
1738
1739 STATIC_ASSERT((TESS_SPACING_EQUAL + 1) % 3 == PIPE_TESS_SPACING_EQUAL);
1740 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD + 1) % 3 ==
1741 PIPE_TESS_SPACING_FRACTIONAL_ODD);
1742 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN + 1) % 3 ==
1743 PIPE_TESS_SPACING_FRACTIONAL_EVEN);
1744
1745 ureg_property(ureg, TGSI_PROPERTY_TES_SPACING,
1746 (sttep->Base.info.tess.spacing + 1) % 3);
1747
1748 ureg_property(ureg, TGSI_PROPERTY_TES_VERTEX_ORDER_CW,
1749 !sttep->Base.info.tess.ccw);
1750 ureg_property(ureg, TGSI_PROPERTY_TES_POINT_MODE,
1751 sttep->Base.info.tess.point_mode);
1752
1753 st_translate_program_common(st, &sttep->Base, sttep->glsl_to_tgsi,
1754 ureg, PIPE_SHADER_TESS_EVAL, &sttep->tgsi);
1755
1756 free_glsl_to_tgsi_visitor(sttep->glsl_to_tgsi);
1757 sttep->glsl_to_tgsi = NULL;
1758 return true;
1759 }
1760
1761
1762 /**
1763 * Translate a compute program to create a new variant.
1764 */
1765 bool
1766 st_translate_compute_program(struct st_context *st,
1767 struct st_compute_program *stcp)
1768 {
1769 struct ureg_program *ureg;
1770 struct pipe_shader_state prog;
1771
1772 if (stcp->shader_program) {
1773 nir_shader *nir = st_glsl_to_nir(st, &stcp->Base, stcp->shader_program,
1774 MESA_SHADER_COMPUTE);
1775
1776 /* no compute variants: */
1777 st_finalize_nir(st, &stcp->Base, stcp->shader_program, nir);
1778
1779 stcp->tgsi.ir_type = PIPE_SHADER_IR_NIR;
1780 stcp->tgsi.prog = nir;
1781
1782 return true;
1783 }
1784
1785 ureg = ureg_create_with_screen(PIPE_SHADER_COMPUTE, st->pipe->screen);
1786 if (ureg == NULL)
1787 return false;
1788
1789 st_translate_program_common(st, &stcp->Base, stcp->glsl_to_tgsi, ureg,
1790 PIPE_SHADER_COMPUTE, &prog);
1791
1792 stcp->tgsi.ir_type = PIPE_SHADER_IR_TGSI;
1793 stcp->tgsi.prog = prog.tokens;
1794 stcp->tgsi.req_local_mem = stcp->Base.info.cs.shared_size;
1795 stcp->tgsi.req_private_mem = 0;
1796 stcp->tgsi.req_input_mem = 0;
1797
1798 free_glsl_to_tgsi_visitor(stcp->glsl_to_tgsi);
1799 stcp->glsl_to_tgsi = NULL;
1800 return true;
1801 }
1802
1803
1804 /**
1805 * Get/create compute program variant.
1806 */
1807 struct st_basic_variant *
1808 st_get_cp_variant(struct st_context *st,
1809 struct pipe_compute_state *tgsi,
1810 struct st_basic_variant **variants)
1811 {
1812 struct pipe_context *pipe = st->pipe;
1813 struct st_basic_variant *v;
1814 struct st_basic_variant_key key;
1815
1816 memset(&key, 0, sizeof(key));
1817 key.st = st->has_shareable_shaders ? NULL : st;
1818
1819 /* Search for existing variant */
1820 for (v = *variants; v; v = v->next) {
1821 if (memcmp(&v->key, &key, sizeof(key)) == 0) {
1822 break;
1823 }
1824 }
1825
1826 if (!v) {
1827 /* create new */
1828 v = CALLOC_STRUCT(st_basic_variant);
1829 if (v) {
1830 /* fill in new variant */
1831 v->driver_shader = pipe->create_compute_state(pipe, tgsi);
1832 v->key = key;
1833
1834 /* insert into list */
1835 v->next = *variants;
1836 *variants = v;
1837 }
1838 }
1839
1840 return v;
1841 }
1842
1843
1844 /**
1845 * Vert/Geom/Frag programs have per-context variants. Free all the
1846 * variants attached to the given program which match the given context.
1847 */
1848 static void
1849 destroy_program_variants(struct st_context *st, struct gl_program *target)
1850 {
1851 if (!target || target == &_mesa_DummyProgram)
1852 return;
1853
1854 switch (target->Target) {
1855 case GL_VERTEX_PROGRAM_ARB:
1856 {
1857 struct st_vertex_program *stvp = (struct st_vertex_program *) target;
1858 struct st_vp_variant *vpv, **prevPtr = &stvp->variants;
1859
1860 for (vpv = stvp->variants; vpv; ) {
1861 struct st_vp_variant *next = vpv->next;
1862 if (vpv->key.st == st) {
1863 /* unlink from list */
1864 *prevPtr = next;
1865 /* destroy this variant */
1866 delete_vp_variant(st, vpv);
1867 }
1868 else {
1869 prevPtr = &vpv->next;
1870 }
1871 vpv = next;
1872 }
1873 }
1874 break;
1875 case GL_FRAGMENT_PROGRAM_ARB:
1876 {
1877 struct st_fragment_program *stfp =
1878 (struct st_fragment_program *) target;
1879 struct st_fp_variant *fpv, **prevPtr = &stfp->variants;
1880
1881 for (fpv = stfp->variants; fpv; ) {
1882 struct st_fp_variant *next = fpv->next;
1883 if (fpv->key.st == st) {
1884 /* unlink from list */
1885 *prevPtr = next;
1886 /* destroy this variant */
1887 delete_fp_variant(st, fpv);
1888 }
1889 else {
1890 prevPtr = &fpv->next;
1891 }
1892 fpv = next;
1893 }
1894 }
1895 break;
1896 case GL_GEOMETRY_PROGRAM_NV:
1897 case GL_TESS_CONTROL_PROGRAM_NV:
1898 case GL_TESS_EVALUATION_PROGRAM_NV:
1899 case GL_COMPUTE_PROGRAM_NV:
1900 {
1901 struct st_common_program *p = st_common_program(target);
1902 struct st_compute_program *cp = (struct st_compute_program*)target;
1903 struct st_basic_variant **variants =
1904 target->Target == GL_COMPUTE_PROGRAM_NV ? &cp->variants :
1905 &p->variants;
1906 struct st_basic_variant *v, **prevPtr = variants;
1907
1908 for (v = *variants; v; ) {
1909 struct st_basic_variant *next = v->next;
1910 if (v->key.st == st) {
1911 /* unlink from list */
1912 *prevPtr = next;
1913 /* destroy this variant */
1914 delete_basic_variant(st, v, target->Target);
1915 }
1916 else {
1917 prevPtr = &v->next;
1918 }
1919 v = next;
1920 }
1921 }
1922 break;
1923 default:
1924 _mesa_problem(NULL, "Unexpected program target 0x%x in "
1925 "destroy_program_variants_cb()", target->Target);
1926 }
1927 }
1928
1929
1930 /**
1931 * Callback for _mesa_HashWalk. Free all the shader's program variants
1932 * which match the given context.
1933 */
1934 static void
1935 destroy_shader_program_variants_cb(GLuint key, void *data, void *userData)
1936 {
1937 struct st_context *st = (struct st_context *) userData;
1938 struct gl_shader *shader = (struct gl_shader *) data;
1939
1940 switch (shader->Type) {
1941 case GL_SHADER_PROGRAM_MESA:
1942 {
1943 struct gl_shader_program *shProg = (struct gl_shader_program *) data;
1944 GLuint i;
1945
1946 for (i = 0; i < ARRAY_SIZE(shProg->_LinkedShaders); i++) {
1947 if (shProg->_LinkedShaders[i])
1948 destroy_program_variants(st, shProg->_LinkedShaders[i]->Program);
1949 }
1950 }
1951 break;
1952 case GL_VERTEX_SHADER:
1953 case GL_FRAGMENT_SHADER:
1954 case GL_GEOMETRY_SHADER:
1955 case GL_TESS_CONTROL_SHADER:
1956 case GL_TESS_EVALUATION_SHADER:
1957 case GL_COMPUTE_SHADER:
1958 break;
1959 default:
1960 assert(0);
1961 }
1962 }
1963
1964
1965 /**
1966 * Callback for _mesa_HashWalk. Free all the program variants which match
1967 * the given context.
1968 */
1969 static void
1970 destroy_program_variants_cb(GLuint key, void *data, void *userData)
1971 {
1972 struct st_context *st = (struct st_context *) userData;
1973 struct gl_program *program = (struct gl_program *) data;
1974 destroy_program_variants(st, program);
1975 }
1976
1977
1978 /**
1979 * Walk over all shaders and programs to delete any variants which
1980 * belong to the given context.
1981 * This is called during context tear-down.
1982 */
1983 void
1984 st_destroy_program_variants(struct st_context *st)
1985 {
1986 /* If shaders can be shared with other contexts, the last context will
1987 * call DeleteProgram on all shaders, releasing everything.
1988 */
1989 if (st->has_shareable_shaders)
1990 return;
1991
1992 /* ARB vert/frag program */
1993 _mesa_HashWalk(st->ctx->Shared->Programs,
1994 destroy_program_variants_cb, st);
1995
1996 /* GLSL vert/frag/geom shaders */
1997 _mesa_HashWalk(st->ctx->Shared->ShaderObjects,
1998 destroy_shader_program_variants_cb, st);
1999 }
2000
2001
2002 /**
2003 * For debugging, print/dump the current vertex program.
2004 */
2005 void
2006 st_print_current_vertex_program(void)
2007 {
2008 GET_CURRENT_CONTEXT(ctx);
2009
2010 if (ctx->VertexProgram._Current) {
2011 struct st_vertex_program *stvp =
2012 (struct st_vertex_program *) ctx->VertexProgram._Current;
2013 struct st_vp_variant *stv;
2014
2015 debug_printf("Vertex program %u\n", stvp->Base.Id);
2016
2017 for (stv = stvp->variants; stv; stv = stv->next) {
2018 debug_printf("variant %p\n", stv);
2019 tgsi_dump(stv->tgsi.tokens, 0);
2020 }
2021 }
2022 }
2023
2024
2025 /**
2026 * Compile one shader variant.
2027 */
2028 void
2029 st_precompile_shader_variant(struct st_context *st,
2030 struct gl_program *prog)
2031 {
2032 switch (prog->Target) {
2033 case GL_VERTEX_PROGRAM_ARB: {
2034 struct st_vertex_program *p = (struct st_vertex_program *)prog;
2035 struct st_vp_variant_key key;
2036
2037 memset(&key, 0, sizeof(key));
2038 key.st = st->has_shareable_shaders ? NULL : st;
2039 st_get_vp_variant(st, p, &key);
2040 break;
2041 }
2042
2043 case GL_TESS_CONTROL_PROGRAM_NV: {
2044 struct st_common_program *p = st_common_program(prog);
2045 st_get_basic_variant(st, PIPE_SHADER_TESS_CTRL, p);
2046 break;
2047 }
2048
2049 case GL_TESS_EVALUATION_PROGRAM_NV: {
2050 struct st_common_program *p = st_common_program(prog);
2051 st_get_basic_variant(st, PIPE_SHADER_TESS_EVAL, p);
2052 break;
2053 }
2054
2055 case GL_GEOMETRY_PROGRAM_NV: {
2056 struct st_common_program *p = st_common_program(prog);
2057 st_get_basic_variant(st, PIPE_SHADER_GEOMETRY, p);
2058 break;
2059 }
2060
2061 case GL_FRAGMENT_PROGRAM_ARB: {
2062 struct st_fragment_program *p = (struct st_fragment_program *)prog;
2063 struct st_fp_variant_key key;
2064
2065 memset(&key, 0, sizeof(key));
2066 key.st = st->has_shareable_shaders ? NULL : st;
2067 st_get_fp_variant(st, p, &key);
2068 break;
2069 }
2070
2071 case GL_COMPUTE_PROGRAM_NV: {
2072 struct st_compute_program *p = (struct st_compute_program *)prog;
2073 st_get_cp_variant(st, &p->tgsi, &p->variants);
2074 break;
2075 }
2076
2077 default:
2078 assert(0);
2079 }
2080 }