projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
st/mesa: call prog_to_nir sooner for ARB_fp
[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/errors.h"
35 #include "main/imports.h"
36 #include "main/hash.h"
37 #include "main/mtypes.h"
38 #include "program/prog_parameter.h"
39 #include "program/prog_print.h"
40 #include "program/prog_to_nir.h"
41 #include "program/programopt.h"
42
43 #include "compiler/nir/nir.h"
44
45 #include "pipe/p_context.h"
46 #include "pipe/p_defines.h"
47 #include "pipe/p_shader_tokens.h"
48 #include "draw/draw_context.h"
49 #include "tgsi/tgsi_dump.h"
50 #include "tgsi/tgsi_emulate.h"
51 #include "tgsi/tgsi_parse.h"
52 #include "tgsi/tgsi_ureg.h"
53
54 #include "st_debug.h"
55 #include "st_cb_bitmap.h"
56 #include "st_cb_drawpixels.h"
57 #include "st_context.h"
58 #include "st_tgsi_lower_depth_clamp.h"
59 #include "st_tgsi_lower_yuv.h"
60 #include "st_program.h"
61 #include "st_mesa_to_tgsi.h"
62 #include "st_atifs_to_tgsi.h"
63 #include "st_nir.h"
64 #include "st_shader_cache.h"
65 #include "cso_cache/cso_context.h"
66
67
68
69 static void
70 set_affected_state_flags(uint64_t *states,
71 struct gl_program *prog,
72 uint64_t new_constants,
73 uint64_t new_sampler_views,
74 uint64_t new_samplers,
75 uint64_t new_images,
76 uint64_t new_ubos,
77 uint64_t new_ssbos,
78 uint64_t new_atomics)
79 {
80 if (prog->Parameters->NumParameters)
81 *states |= new_constants;
82
83 if (prog->info.num_textures)
84 *states |= new_sampler_views | new_samplers;
85
86 if (prog->info.num_images)
87 *states |= new_images;
88
89 if (prog->info.num_ubos)
90 *states |= new_ubos;
91
92 if (prog->info.num_ssbos)
93 *states |= new_ssbos;
94
95 if (prog->info.num_abos)
96 *states |= new_atomics;
97 }
98
99 /**
100 * This determines which states will be updated when the shader is bound.
101 */
102 void
103 st_set_prog_affected_state_flags(struct gl_program *prog)
104 {
105 uint64_t *states;
106
107 switch (prog->info.stage) {
108 case MESA_SHADER_VERTEX:
109 states = &((struct st_vertex_program*)prog)->affected_states;
110
111 *states = ST_NEW_VS_STATE |
112 ST_NEW_RASTERIZER |
113 ST_NEW_VERTEX_ARRAYS;
114
115 set_affected_state_flags(states, prog,
116 ST_NEW_VS_CONSTANTS,
117 ST_NEW_VS_SAMPLER_VIEWS,
118 ST_NEW_VS_SAMPLERS,
119 ST_NEW_VS_IMAGES,
120 ST_NEW_VS_UBOS,
121 ST_NEW_VS_SSBOS,
122 ST_NEW_VS_ATOMICS);
123 break;
124
125 case MESA_SHADER_TESS_CTRL:
126 states = &(st_common_program(prog))->affected_states;
127
128 *states = ST_NEW_TCS_STATE;
129
130 set_affected_state_flags(states, prog,
131 ST_NEW_TCS_CONSTANTS,
132 ST_NEW_TCS_SAMPLER_VIEWS,
133 ST_NEW_TCS_SAMPLERS,
134 ST_NEW_TCS_IMAGES,
135 ST_NEW_TCS_UBOS,
136 ST_NEW_TCS_SSBOS,
137 ST_NEW_TCS_ATOMICS);
138 break;
139
140 case MESA_SHADER_TESS_EVAL:
141 states = &(st_common_program(prog))->affected_states;
142
143 *states = ST_NEW_TES_STATE |
144 ST_NEW_RASTERIZER;
145
146 set_affected_state_flags(states, prog,
147 ST_NEW_TES_CONSTANTS,
148 ST_NEW_TES_SAMPLER_VIEWS,
149 ST_NEW_TES_SAMPLERS,
150 ST_NEW_TES_IMAGES,
151 ST_NEW_TES_UBOS,
152 ST_NEW_TES_SSBOS,
153 ST_NEW_TES_ATOMICS);
154 break;
155
156 case MESA_SHADER_GEOMETRY:
157 states = &(st_common_program(prog))->affected_states;
158
159 *states = ST_NEW_GS_STATE |
160 ST_NEW_RASTERIZER;
161
162 set_affected_state_flags(states, prog,
163 ST_NEW_GS_CONSTANTS,
164 ST_NEW_GS_SAMPLER_VIEWS,
165 ST_NEW_GS_SAMPLERS,
166 ST_NEW_GS_IMAGES,
167 ST_NEW_GS_UBOS,
168 ST_NEW_GS_SSBOS,
169 ST_NEW_GS_ATOMICS);
170 break;
171
172 case MESA_SHADER_FRAGMENT:
173 states = &((struct st_fragment_program*)prog)->affected_states;
174
175 /* gl_FragCoord and glDrawPixels always use constants. */
176 *states = ST_NEW_FS_STATE |
177 ST_NEW_SAMPLE_SHADING |
178 ST_NEW_FS_CONSTANTS;
179
180 set_affected_state_flags(states, prog,
181 ST_NEW_FS_CONSTANTS,
182 ST_NEW_FS_SAMPLER_VIEWS,
183 ST_NEW_FS_SAMPLERS,
184 ST_NEW_FS_IMAGES,
185 ST_NEW_FS_UBOS,
186 ST_NEW_FS_SSBOS,
187 ST_NEW_FS_ATOMICS);
188 break;
189
190 case MESA_SHADER_COMPUTE:
191 states = &((struct st_common_program*)prog)->affected_states;
192
193 *states = ST_NEW_CS_STATE;
194
195 set_affected_state_flags(states, prog,
196 ST_NEW_CS_CONSTANTS,
197 ST_NEW_CS_SAMPLER_VIEWS,
198 ST_NEW_CS_SAMPLERS,
199 ST_NEW_CS_IMAGES,
200 ST_NEW_CS_UBOS,
201 ST_NEW_CS_SSBOS,
202 ST_NEW_CS_ATOMICS);
203 break;
204
205 default:
206 unreachable("unhandled shader stage");
207 }
208 }
209
210 static void
211 delete_ir(struct pipe_shader_state *ir)
212 {
213 if (ir->tokens) {
214 ureg_free_tokens(ir->tokens);
215 ir->tokens = NULL;
216 }
217
218 /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
219 * it has resulted in the driver taking ownership of the NIR. Those
220 * callers should be NULLing out the nir field in any pipe_shader_state
221 * that might have this called in order to indicate that.
222 *
223 * GLSL IR and ARB programs will have set gl_program->nir to the same
224 * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
225 */
226 }
227
228 /**
229 * Delete a vertex program variant. Note the caller must unlink
230 * the variant from the linked list.
231 */
232 static void
233 delete_vp_variant(struct st_context *st, struct st_vp_variant *vpv)
234 {
235 if (vpv->driver_shader) {
236 if (st->has_shareable_shaders || vpv->key.st == st) {
237 cso_delete_vertex_shader(st->cso_context, vpv->driver_shader);
238 } else {
239 st_save_zombie_shader(vpv->key.st, PIPE_SHADER_VERTEX,
240 vpv->driver_shader);
241 }
242 }
243
244 if (vpv->draw_shader)
245 draw_delete_vertex_shader( st->draw, vpv->draw_shader );
246
247 delete_ir(&vpv->tgsi);
248
249 free( vpv );
250 }
251
252
253
254 /**
255 * Clean out any old compilations:
256 */
257 void
258 st_release_vp_variants( struct st_context *st,
259 struct st_vertex_program *stvp )
260 {
261 struct st_vp_variant *vpv;
262
263 for (vpv = stvp->variants; vpv; ) {
264 struct st_vp_variant *next = vpv->next;
265 delete_vp_variant(st, vpv);
266 vpv = next;
267 }
268
269 stvp->variants = NULL;
270
271 delete_ir(&stvp->tgsi);
272 }
273
274
275
276 /**
277 * Delete a fragment program variant. Note the caller must unlink
278 * the variant from the linked list.
279 */
280 static void
281 delete_fp_variant(struct st_context *st, struct st_fp_variant *fpv)
282 {
283 if (fpv->driver_shader) {
284 if (st->has_shareable_shaders || fpv->key.st == st) {
285 cso_delete_fragment_shader(st->cso_context, fpv->driver_shader);
286 } else {
287 st_save_zombie_shader(fpv->key.st, PIPE_SHADER_FRAGMENT,
288 fpv->driver_shader);
289 }
290 }
291
292 free(fpv);
293 }
294
295
296 /**
297 * Free all variants of a fragment program.
298 */
299 void
300 st_release_fp_variants(struct st_context *st, struct st_fragment_program *stfp)
301 {
302 struct st_fp_variant *fpv;
303
304 for (fpv = stfp->variants; fpv; ) {
305 struct st_fp_variant *next = fpv->next;
306 delete_fp_variant(st, fpv);
307 fpv = next;
308 }
309
310 stfp->variants = NULL;
311
312 delete_ir(&stfp->tgsi);
313 }
314
315
316 /**
317 * Delete a basic program variant. Note the caller must unlink
318 * the variant from the linked list.
319 */
320 static void
321 delete_basic_variant(struct st_context *st, struct st_basic_variant *v,
322 GLenum target)
323 {
324 if (v->driver_shader) {
325 if (st->has_shareable_shaders || v->key.st == st) {
326 /* The shader's context matches the calling context, or we
327 * don't care.
328 */
329 switch (target) {
330 case GL_TESS_CONTROL_PROGRAM_NV:
331 cso_delete_tessctrl_shader(st->cso_context, v->driver_shader);
332 break;
333 case GL_TESS_EVALUATION_PROGRAM_NV:
334 cso_delete_tesseval_shader(st->cso_context, v->driver_shader);
335 break;
336 case GL_GEOMETRY_PROGRAM_NV:
337 cso_delete_geometry_shader(st->cso_context, v->driver_shader);
338 break;
339 case GL_COMPUTE_PROGRAM_NV:
340 cso_delete_compute_shader(st->cso_context, v->driver_shader);
341 break;
342 default:
343 unreachable("bad shader type in delete_basic_variant");
344 }
345 } else {
346 /* We can't delete a shader with a context different from the one
347 * that created it. Add it to the creating context's zombie list.
348 */
349 enum pipe_shader_type type;
350 switch (target) {
351 case GL_TESS_CONTROL_PROGRAM_NV:
352 type = PIPE_SHADER_TESS_CTRL;
353 break;
354 case GL_TESS_EVALUATION_PROGRAM_NV:
355 type = PIPE_SHADER_TESS_EVAL;
356 break;
357 case GL_GEOMETRY_PROGRAM_NV:
358 type = PIPE_SHADER_GEOMETRY;
359 break;
360 default:
361 unreachable("");
362 }
363 st_save_zombie_shader(v->key.st, type, v->driver_shader);
364 }
365 }
366
367 free(v);
368 }
369
370
371 /**
372 * Free all basic program variants.
373 */
374 void
375 st_release_basic_variants(struct st_context *st, struct st_common_program *p)
376 {
377 struct st_basic_variant *v;
378
379 for (v = p->variants; v; ) {
380 struct st_basic_variant *next = v->next;
381 delete_basic_variant(st, v, p->Base.Target);
382 v = next;
383 }
384
385 p->variants = NULL;
386 delete_ir(&p->tgsi);
387 }
388
389
390 /**
391 * Translate ARB (asm) program to NIR
392 */
393 static nir_shader *
394 st_translate_prog_to_nir(struct st_context *st, struct gl_program *prog,
395 gl_shader_stage stage)
396 {
397 const struct gl_shader_compiler_options *options =
398 &st->ctx->Const.ShaderCompilerOptions[stage];
399
400 /* Translate to NIR */
401 nir_shader *nir = prog_to_nir(prog, options->NirOptions);
402 NIR_PASS_V(nir, nir_lower_regs_to_ssa); /* turn registers into SSA */
403 nir_validate_shader(nir, "after st/ptn lower_regs_to_ssa");
404
405 NIR_PASS_V(nir, st_nir_lower_wpos_ytransform, prog, st->pipe->screen);
406 NIR_PASS_V(nir, nir_lower_system_values);
407
408 /* Optimise NIR */
409 NIR_PASS_V(nir, nir_opt_constant_folding);
410 st_nir_opts(nir);
411 nir_validate_shader(nir, "after st/ptn NIR opts");
412
413 return nir;
414 }
415
416 void
417 st_prepare_vertex_program(struct st_vertex_program *stvp)
418 {
419 stvp->num_inputs = 0;
420 memset(stvp->input_to_index, ~0, sizeof(stvp->input_to_index));
421 memset(stvp->result_to_output, ~0, sizeof(stvp->result_to_output));
422
423 /* Determine number of inputs, the mappings between VERT_ATTRIB_x
424 * and TGSI generic input indexes, plus input attrib semantic info.
425 */
426 for (unsigned attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
427 if ((stvp->Base.info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
428 stvp->input_to_index[attr] = stvp->num_inputs;
429 stvp->index_to_input[stvp->num_inputs] = attr;
430 stvp->num_inputs++;
431
432 if ((stvp->Base.DualSlotInputs & BITFIELD64_BIT(attr)) != 0) {
433 /* add placeholder for second part of a double attribute */
434 stvp->index_to_input[stvp->num_inputs] = ST_DOUBLE_ATTRIB_PLACEHOLDER;
435 stvp->num_inputs++;
436 }
437 }
438 }
439 /* pre-setup potentially unused edgeflag input */
440 stvp->input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
441 stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
442
443 /* Compute mapping of vertex program outputs to slots. */
444 unsigned num_outputs = 0;
445 for (unsigned attr = 0; attr < VARYING_SLOT_MAX; attr++) {
446 if (stvp->Base.info.outputs_written & BITFIELD64_BIT(attr))
447 stvp->result_to_output[attr] = num_outputs++;
448 }
449 /* pre-setup potentially unused edgeflag output */
450 stvp->result_to_output[VARYING_SLOT_EDGE] = num_outputs;
451 }
452
453 void
454 st_translate_stream_output_info(struct gl_program *prog)
455 {
456 struct gl_transform_feedback_info *info = prog->sh.LinkedTransformFeedback;
457 if (!info)
458 return;
459
460 /* Determine the (default) output register mapping for each output. */
461 unsigned num_outputs = 0;
462 ubyte output_mapping[VARYING_SLOT_TESS_MAX];
463 memset(output_mapping, 0, sizeof(output_mapping));
464
465 for (unsigned attr = 0; attr < VARYING_SLOT_MAX; attr++) {
466 if (prog->info.outputs_written & BITFIELD64_BIT(attr))
467 output_mapping[attr] = num_outputs++;
468 }
469
470 /* Translate stream output info. */
471 struct pipe_stream_output_info *so_info = NULL;
472 if (prog->info.stage == MESA_SHADER_VERTEX)
473 so_info = &((struct st_vertex_program*)prog)->tgsi.stream_output;
474 else
475 so_info = &((struct st_common_program*)prog)->tgsi.stream_output;
476
477 for (unsigned i = 0; i < info->NumOutputs; i++) {
478 so_info->output[i].register_index =
479 output_mapping[info->Outputs[i].OutputRegister];
480 so_info->output[i].start_component = info->Outputs[i].ComponentOffset;
481 so_info->output[i].num_components = info->Outputs[i].NumComponents;
482 so_info->output[i].output_buffer = info->Outputs[i].OutputBuffer;
483 so_info->output[i].dst_offset = info->Outputs[i].DstOffset;
484 so_info->output[i].stream = info->Outputs[i].StreamId;
485 }
486
487 for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
488 so_info->stride[i] = info->Buffers[i].Stride;
489 }
490 so_info->num_outputs = info->NumOutputs;
491 }
492
493 /**
494 * Translate a vertex program.
495 */
496 bool
497 st_translate_vertex_program(struct st_context *st,
498 struct st_vertex_program *stvp)
499 {
500 struct ureg_program *ureg;
501 enum pipe_error error;
502 unsigned num_outputs = 0;
503 unsigned attr;
504 ubyte output_semantic_name[VARYING_SLOT_MAX] = {0};
505 ubyte output_semantic_index[VARYING_SLOT_MAX] = {0};
506
507 if (stvp->Base.arb.IsPositionInvariant)
508 _mesa_insert_mvp_code(st->ctx, &stvp->Base);
509
510 st_prepare_vertex_program(stvp);
511
512 /* Get semantic names and indices. */
513 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
514 if (stvp->Base.info.outputs_written & BITFIELD64_BIT(attr)) {
515 unsigned slot = num_outputs++;
516 unsigned semantic_name, semantic_index;
517 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
518 &semantic_name, &semantic_index);
519 output_semantic_name[slot] = semantic_name;
520 output_semantic_index[slot] = semantic_index;
521 }
522 }
523 /* pre-setup potentially unused edgeflag output */
524 output_semantic_name[num_outputs] = TGSI_SEMANTIC_EDGEFLAG;
525 output_semantic_index[num_outputs] = 0;
526
527 /* ARB_vp: */
528 if (!stvp->glsl_to_tgsi) {
529 _mesa_remove_output_reads(&stvp->Base, PROGRAM_OUTPUT);
530
531 /* This determines which states will be updated when the assembly
532 * shader is bound.
533 */
534 stvp->affected_states = ST_NEW_VS_STATE |
535 ST_NEW_RASTERIZER |
536 ST_NEW_VERTEX_ARRAYS;
537
538 if (stvp->Base.Parameters->NumParameters)
539 stvp->affected_states |= ST_NEW_VS_CONSTANTS;
540
541 /* No samplers are allowed in ARB_vp. */
542 }
543
544 ureg = ureg_create_with_screen(PIPE_SHADER_VERTEX, st->pipe->screen);
545 if (ureg == NULL)
546 return false;
547
548 if (stvp->Base.info.clip_distance_array_size)
549 ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
550 stvp->Base.info.clip_distance_array_size);
551 if (stvp->Base.info.cull_distance_array_size)
552 ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
553 stvp->Base.info.cull_distance_array_size);
554
555 if (ST_DEBUG & DEBUG_MESA) {
556 _mesa_print_program(&stvp->Base);
557 _mesa_print_program_parameters(st->ctx, &stvp->Base);
558 debug_printf("\n");
559 }
560
561 if (stvp->glsl_to_tgsi) {
562 error = st_translate_program(st->ctx,
563 PIPE_SHADER_VERTEX,
564 ureg,
565 stvp->glsl_to_tgsi,
566 &stvp->Base,
567 /* inputs */
568 stvp->num_inputs,
569 stvp->input_to_index,
570 NULL, /* inputSlotToAttr */
571 NULL, /* input semantic name */
572 NULL, /* input semantic index */
573 NULL, /* interp mode */
574 /* outputs */
575 num_outputs,
576 stvp->result_to_output,
577 output_semantic_name,
578 output_semantic_index);
579
580 st_translate_stream_output_info(&stvp->Base);
581
582 free_glsl_to_tgsi_visitor(stvp->glsl_to_tgsi);
583 } else
584 error = st_translate_mesa_program(st->ctx,
585 PIPE_SHADER_VERTEX,
586 ureg,
587 &stvp->Base,
588 /* inputs */
589 stvp->num_inputs,
590 stvp->input_to_index,
591 NULL, /* input semantic name */
592 NULL, /* input semantic index */
593 NULL,
594 /* outputs */
595 num_outputs,
596 stvp->result_to_output,
597 output_semantic_name,
598 output_semantic_index);
599
600 if (error) {
601 debug_printf("%s: failed to translate Mesa program:\n", __func__);
602 _mesa_print_program(&stvp->Base);
603 debug_assert(0);
604 return false;
605 }
606
607 stvp->tgsi.tokens = ureg_get_tokens(ureg, &stvp->num_tgsi_tokens);
608 ureg_destroy(ureg);
609
610 if (stvp->glsl_to_tgsi) {
611 stvp->glsl_to_tgsi = NULL;
612 st_store_ir_in_disk_cache(st, &stvp->Base, false);
613 }
614
615 bool use_nir = PIPE_SHADER_IR_NIR ==
616 st->pipe->screen->get_shader_param(st->pipe->screen, PIPE_SHADER_VERTEX,
617 PIPE_SHADER_CAP_PREFERRED_IR);
618
619 if (use_nir) {
620 nir_shader *nir =
621 st_translate_prog_to_nir(st, &stvp->Base, MESA_SHADER_VERTEX);
622
623 if (stvp->tgsi.ir.nir)
624 ralloc_free(stvp->tgsi.ir.nir);
625 stvp->tgsi.type = PIPE_SHADER_IR_NIR;
626 stvp->tgsi.ir.nir = nir;
627 stvp->Base.nir = nir;
628 return true;
629 }
630
631 return stvp->tgsi.tokens != NULL;
632 }
633
634 static const gl_state_index16 depth_range_state[STATE_LENGTH] =
635 { STATE_DEPTH_RANGE };
636
637 static struct st_vp_variant *
638 st_create_vp_variant(struct st_context *st,
639 struct st_vertex_program *stvp,
640 const struct st_vp_variant_key *key)
641 {
642 struct st_vp_variant *vpv = CALLOC_STRUCT(st_vp_variant);
643 struct pipe_context *pipe = st->pipe;
644 struct gl_program_parameter_list *params = stvp->Base.Parameters;
645
646 vpv->key = *key;
647 vpv->tgsi.stream_output = stvp->tgsi.stream_output;
648 vpv->num_inputs = stvp->num_inputs;
649
650 /* When generating a NIR program, we usually don't have TGSI tokens.
651 * However, we do create them for ARB_vertex_program / fixed-function VS
652 * programs which we may need to use with the draw module for legacy
653 * feedback/select emulation. If they exist, copy them.
654 */
655 if (stvp->tgsi.tokens)
656 vpv->tgsi.tokens = tgsi_dup_tokens(stvp->tgsi.tokens);
657
658 if (stvp->tgsi.type == PIPE_SHADER_IR_NIR) {
659 vpv->tgsi.type = PIPE_SHADER_IR_NIR;
660 vpv->tgsi.ir.nir = nir_shader_clone(NULL, stvp->tgsi.ir.nir);
661 if (key->clamp_color)
662 NIR_PASS_V(vpv->tgsi.ir.nir, nir_lower_clamp_color_outputs);
663 if (key->passthrough_edgeflags) {
664 NIR_PASS_V(vpv->tgsi.ir.nir, nir_lower_passthrough_edgeflags);
665 vpv->num_inputs++;
666 }
667
668 st_finalize_nir(st, &stvp->Base, stvp->shader_program,
669 vpv->tgsi.ir.nir);
670
671 vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
672 /* driver takes ownership of IR: */
673 vpv->tgsi.ir.nir = NULL;
674 return vpv;
675 }
676
677 /* Emulate features. */
678 if (key->clamp_color || key->passthrough_edgeflags) {
679 const struct tgsi_token *tokens;
680 unsigned flags =
681 (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
682 (key->passthrough_edgeflags ? TGSI_EMU_PASSTHROUGH_EDGEFLAG : 0);
683
684 tokens = tgsi_emulate(vpv->tgsi.tokens, flags);
685
686 if (tokens) {
687 tgsi_free_tokens(vpv->tgsi.tokens);
688 vpv->tgsi.tokens = tokens;
689
690 if (key->passthrough_edgeflags)
691 vpv->num_inputs++;
692 } else
693 fprintf(stderr, "mesa: cannot emulate deprecated features\n");
694 }
695
696 if (key->lower_depth_clamp) {
697 unsigned depth_range_const =
698 _mesa_add_state_reference(params, depth_range_state);
699
700 const struct tgsi_token *tokens;
701 tokens = st_tgsi_lower_depth_clamp(vpv->tgsi.tokens, depth_range_const,
702 key->clip_negative_one_to_one);
703 if (tokens != vpv->tgsi.tokens)
704 tgsi_free_tokens(vpv->tgsi.tokens);
705 vpv->tgsi.tokens = tokens;
706 }
707
708 if (ST_DEBUG & DEBUG_TGSI) {
709 tgsi_dump(vpv->tgsi.tokens, 0);
710 debug_printf("\n");
711 }
712
713 vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
714 return vpv;
715 }
716
717
718 /**
719 * Find/create a vertex program variant.
720 */
721 struct st_vp_variant *
722 st_get_vp_variant(struct st_context *st,
723 struct st_vertex_program *stvp,
724 const struct st_vp_variant_key *key)
725 {
726 struct st_vp_variant *vpv;
727
728 /* Search for existing variant */
729 for (vpv = stvp->variants; vpv; vpv = vpv->next) {
730 if (memcmp(&vpv->key, key, sizeof(*key)) == 0) {
731 break;
732 }
733 }
734
735 if (!vpv) {
736 /* create now */
737 vpv = st_create_vp_variant(st, stvp, key);
738 if (vpv) {
739 for (unsigned index = 0; index < vpv->num_inputs; ++index) {
740 unsigned attr = stvp->index_to_input[index];
741 if (attr == ST_DOUBLE_ATTRIB_PLACEHOLDER)
742 continue;
743 vpv->vert_attrib_mask |= 1u << attr;
744 }
745
746 /* insert into list */
747 vpv->next = stvp->variants;
748 stvp->variants = vpv;
749 }
750 }
751
752 return vpv;
753 }
754
755
756 /**
757 * Translate a Mesa fragment shader into a TGSI shader.
758 */
759 bool
760 st_translate_fragment_program(struct st_context *st,
761 struct st_fragment_program *stfp)
762 {
763 /* Non-GLSL programs: */
764 if (!stfp->glsl_to_tgsi) {
765 _mesa_remove_output_reads(&stfp->Base, PROGRAM_OUTPUT);
766 if (st->ctx->Const.GLSLFragCoordIsSysVal)
767 _mesa_program_fragment_position_to_sysval(&stfp->Base);
768
769 /* This determines which states will be updated when the assembly
770 * shader is bound.
771 *
772 * fragment.position and glDrawPixels always use constants.
773 */
774 stfp->affected_states = ST_NEW_FS_STATE |
775 ST_NEW_SAMPLE_SHADING |
776 ST_NEW_FS_CONSTANTS;
777
778 if (stfp->ati_fs) {
779 /* Just set them for ATI_fs unconditionally. */
780 stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
781 ST_NEW_FS_SAMPLERS;
782 } else {
783 /* ARB_fp */
784 if (stfp->Base.SamplersUsed)
785 stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
786 ST_NEW_FS_SAMPLERS;
787 }
788
789 /* Translate to NIR. */
790 if (!stfp->ati_fs &&
791 st->pipe->screen->get_shader_param(st->pipe->screen,
792 PIPE_SHADER_FRAGMENT,
793 PIPE_SHADER_CAP_PREFERRED_IR)) {
794 nir_shader *nir =
795 st_translate_prog_to_nir(st, &stfp->Base, MESA_SHADER_FRAGMENT);
796
797 if (stfp->tgsi.ir.nir)
798 ralloc_free(stfp->tgsi.ir.nir);
799 stfp->tgsi.type = PIPE_SHADER_IR_NIR;
800 stfp->tgsi.ir.nir = nir;
801 stfp->Base.nir = nir;
802 return true;
803 }
804 }
805
806 ubyte outputMapping[2 * FRAG_RESULT_MAX];
807 ubyte inputMapping[VARYING_SLOT_MAX];
808 ubyte inputSlotToAttr[VARYING_SLOT_MAX];
809 ubyte interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
810 GLuint attr;
811 GLbitfield64 inputsRead;
812 struct ureg_program *ureg;
813
814 GLboolean write_all = GL_FALSE;
815
816 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
817 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
818 uint fs_num_inputs = 0;
819
820 ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
821 ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
822 uint fs_num_outputs = 0;
823
824 memset(inputSlotToAttr, ~0, sizeof(inputSlotToAttr));
825
826 /*
827 * Convert Mesa program inputs to TGSI input register semantics.
828 */
829 inputsRead = stfp->Base.info.inputs_read;
830 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
831 if ((inputsRead & BITFIELD64_BIT(attr)) != 0) {
832 const GLuint slot = fs_num_inputs++;
833
834 inputMapping[attr] = slot;
835 inputSlotToAttr[slot] = attr;
836
837 switch (attr) {
838 case VARYING_SLOT_POS:
839 input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
840 input_semantic_index[slot] = 0;
841 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
842 break;
843 case VARYING_SLOT_COL0:
844 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
845 input_semantic_index[slot] = 0;
846 interpMode[slot] = stfp->glsl_to_tgsi ?
847 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
848 break;
849 case VARYING_SLOT_COL1:
850 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
851 input_semantic_index[slot] = 1;
852 interpMode[slot] = stfp->glsl_to_tgsi ?
853 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
854 break;
855 case VARYING_SLOT_FOGC:
856 input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
857 input_semantic_index[slot] = 0;
858 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
859 break;
860 case VARYING_SLOT_FACE:
861 input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
862 input_semantic_index[slot] = 0;
863 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
864 break;
865 case VARYING_SLOT_PRIMITIVE_ID:
866 input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
867 input_semantic_index[slot] = 0;
868 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
869 break;
870 case VARYING_SLOT_LAYER:
871 input_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
872 input_semantic_index[slot] = 0;
873 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
874 break;
875 case VARYING_SLOT_VIEWPORT:
876 input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
877 input_semantic_index[slot] = 0;
878 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
879 break;
880 case VARYING_SLOT_CLIP_DIST0:
881 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
882 input_semantic_index[slot] = 0;
883 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
884 break;
885 case VARYING_SLOT_CLIP_DIST1:
886 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
887 input_semantic_index[slot] = 1;
888 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
889 break;
890 case VARYING_SLOT_CULL_DIST0:
891 case VARYING_SLOT_CULL_DIST1:
892 /* these should have been lowered by GLSL */
893 assert(0);
894 break;
895 /* In most cases, there is nothing special about these
896 * inputs, so adopt a convention to use the generic
897 * semantic name and the mesa VARYING_SLOT_ number as the
898 * index.
899 *
900 * All that is required is that the vertex shader labels
901 * its own outputs similarly, and that the vertex shader
902 * generates at least every output required by the
903 * fragment shader plus fixed-function hardware (such as
904 * BFC).
905 *
906 * However, some drivers may need us to identify the PNTC and TEXi
907 * varyings if, for example, their capability to replace them with
908 * sprite coordinates is limited.
909 */
910 case VARYING_SLOT_PNTC:
911 if (st->needs_texcoord_semantic) {
912 input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
913 input_semantic_index[slot] = 0;
914 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
915 break;
916 }
917 /* fall through */
918 case VARYING_SLOT_TEX0:
919 case VARYING_SLOT_TEX1:
920 case VARYING_SLOT_TEX2:
921 case VARYING_SLOT_TEX3:
922 case VARYING_SLOT_TEX4:
923 case VARYING_SLOT_TEX5:
924 case VARYING_SLOT_TEX6:
925 case VARYING_SLOT_TEX7:
926 if (st->needs_texcoord_semantic) {
927 input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
928 input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
929 interpMode[slot] = stfp->glsl_to_tgsi ?
930 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
931 break;
932 }
933 /* fall through */
934 case VARYING_SLOT_VAR0:
935 default:
936 /* Semantic indices should be zero-based because drivers may choose
937 * to assign a fixed slot determined by that index.
938 * This is useful because ARB_separate_shader_objects uses location
939 * qualifiers for linkage, and if the semantic index corresponds to
940 * these locations, linkage passes in the driver become unecessary.
941 *
942 * If needs_texcoord_semantic is true, no semantic indices will be
943 * consumed for the TEXi varyings, and we can base the locations of
944 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
945 */
946 assert(attr >= VARYING_SLOT_VAR0 || attr == VARYING_SLOT_PNTC ||
947 (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
948 input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
949 input_semantic_index[slot] = st_get_generic_varying_index(st, attr);
950 if (attr == VARYING_SLOT_PNTC)
951 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
952 else {
953 interpMode[slot] = stfp->glsl_to_tgsi ?
954 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
955 }
956 break;
957 }
958 }
959 else {
960 inputMapping[attr] = -1;
961 }
962 }
963
964 /*
965 * Semantics and mapping for outputs
966 */
967 GLbitfield64 outputsWritten = stfp->Base.info.outputs_written;
968
969 /* if z is written, emit that first */
970 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
971 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
972 fs_output_semantic_index[fs_num_outputs] = 0;
973 outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
974 fs_num_outputs++;
975 outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
976 }
977
978 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
979 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
980 fs_output_semantic_index[fs_num_outputs] = 0;
981 outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
982 fs_num_outputs++;
983 outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
984 }
985
986 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) {
987 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK;
988 fs_output_semantic_index[fs_num_outputs] = 0;
989 outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs;
990 fs_num_outputs++;
991 outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK);
992 }
993
994 /* handle remaining outputs (color) */
995 for (attr = 0; attr < ARRAY_SIZE(outputMapping); attr++) {
996 const GLbitfield64 written = attr < FRAG_RESULT_MAX ? outputsWritten :
997 stfp->Base.SecondaryOutputsWritten;
998 const unsigned loc = attr % FRAG_RESULT_MAX;
999
1000 if (written & BITFIELD64_BIT(loc)) {
1001 switch (loc) {
1002 case FRAG_RESULT_DEPTH:
1003 case FRAG_RESULT_STENCIL:
1004 case FRAG_RESULT_SAMPLE_MASK:
1005 /* handled above */
1006 assert(0);
1007 break;
1008 case FRAG_RESULT_COLOR:
1009 write_all = GL_TRUE; /* fallthrough */
1010 default: {
1011 int index;
1012 assert(loc == FRAG_RESULT_COLOR ||
1013 (FRAG_RESULT_DATA0 <= loc && loc < FRAG_RESULT_MAX));
1014
1015 index = (loc == FRAG_RESULT_COLOR) ? 0 : (loc - FRAG_RESULT_DATA0);
1016
1017 if (attr >= FRAG_RESULT_MAX) {
1018 /* Secondary color for dual source blending. */
1019 assert(index == 0);
1020 index++;
1021 }
1022
1023 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
1024 fs_output_semantic_index[fs_num_outputs] = index;
1025 outputMapping[attr] = fs_num_outputs;
1026 break;
1027 }
1028 }
1029
1030 fs_num_outputs++;
1031 }
1032 }
1033
1034 ureg = ureg_create_with_screen(PIPE_SHADER_FRAGMENT, st->pipe->screen);
1035 if (ureg == NULL)
1036 return false;
1037
1038 if (ST_DEBUG & DEBUG_MESA) {
1039 _mesa_print_program(&stfp->Base);
1040 _mesa_print_program_parameters(st->ctx, &stfp->Base);
1041 debug_printf("\n");
1042 }
1043 if (write_all == GL_TRUE)
1044 ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
1045
1046 if (stfp->Base.info.fs.depth_layout != FRAG_DEPTH_LAYOUT_NONE) {
1047 switch (stfp->Base.info.fs.depth_layout) {
1048 case FRAG_DEPTH_LAYOUT_ANY:
1049 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1050 TGSI_FS_DEPTH_LAYOUT_ANY);
1051 break;
1052 case FRAG_DEPTH_LAYOUT_GREATER:
1053 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1054 TGSI_FS_DEPTH_LAYOUT_GREATER);
1055 break;
1056 case FRAG_DEPTH_LAYOUT_LESS:
1057 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1058 TGSI_FS_DEPTH_LAYOUT_LESS);
1059 break;
1060 case FRAG_DEPTH_LAYOUT_UNCHANGED:
1061 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1062 TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
1063 break;
1064 default:
1065 assert(0);
1066 }
1067 }
1068
1069 if (stfp->glsl_to_tgsi) {
1070 st_translate_program(st->ctx,
1071 PIPE_SHADER_FRAGMENT,
1072 ureg,
1073 stfp->glsl_to_tgsi,
1074 &stfp->Base,
1075 /* inputs */
1076 fs_num_inputs,
1077 inputMapping,
1078 inputSlotToAttr,
1079 input_semantic_name,
1080 input_semantic_index,
1081 interpMode,
1082 /* outputs */
1083 fs_num_outputs,
1084 outputMapping,
1085 fs_output_semantic_name,
1086 fs_output_semantic_index);
1087
1088 free_glsl_to_tgsi_visitor(stfp->glsl_to_tgsi);
1089 } else if (stfp->ati_fs)
1090 st_translate_atifs_program(ureg,
1091 stfp->ati_fs,
1092 &stfp->Base,
1093 /* inputs */
1094 fs_num_inputs,
1095 inputMapping,
1096 input_semantic_name,
1097 input_semantic_index,
1098 interpMode,
1099 /* outputs */
1100 fs_num_outputs,
1101 outputMapping,
1102 fs_output_semantic_name,
1103 fs_output_semantic_index);
1104 else
1105 st_translate_mesa_program(st->ctx,
1106 PIPE_SHADER_FRAGMENT,
1107 ureg,
1108 &stfp->Base,
1109 /* inputs */
1110 fs_num_inputs,
1111 inputMapping,
1112 input_semantic_name,
1113 input_semantic_index,
1114 interpMode,
1115 /* outputs */
1116 fs_num_outputs,
1117 outputMapping,
1118 fs_output_semantic_name,
1119 fs_output_semantic_index);
1120
1121 stfp->tgsi.tokens = ureg_get_tokens(ureg, &stfp->num_tgsi_tokens);
1122 ureg_destroy(ureg);
1123
1124 if (stfp->glsl_to_tgsi) {
1125 stfp->glsl_to_tgsi = NULL;
1126 st_store_ir_in_disk_cache(st, &stfp->Base, false);
1127 }
1128
1129 return stfp->tgsi.tokens != NULL;
1130 }
1131
1132 static struct st_fp_variant *
1133 st_create_fp_variant(struct st_context *st,
1134 struct st_fragment_program *stfp,
1135 const struct st_fp_variant_key *key)
1136 {
1137 struct pipe_context *pipe = st->pipe;
1138 struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
1139 struct pipe_shader_state tgsi = {0};
1140 struct gl_program_parameter_list *params = stfp->Base.Parameters;
1141 static const gl_state_index16 texcoord_state[STATE_LENGTH] =
1142 { STATE_INTERNAL, STATE_CURRENT_ATTRIB, VERT_ATTRIB_TEX0 };
1143 static const gl_state_index16 scale_state[STATE_LENGTH] =
1144 { STATE_INTERNAL, STATE_PT_SCALE };
1145 static const gl_state_index16 bias_state[STATE_LENGTH] =
1146 { STATE_INTERNAL, STATE_PT_BIAS };
1147
1148 if (!variant)
1149 return NULL;
1150
1151 if (stfp->tgsi.type == PIPE_SHADER_IR_NIR) {
1152 tgsi.type = PIPE_SHADER_IR_NIR;
1153 tgsi.ir.nir = nir_shader_clone(NULL, stfp->tgsi.ir.nir);
1154
1155 if (key->clamp_color)
1156 NIR_PASS_V(tgsi.ir.nir, nir_lower_clamp_color_outputs);
1157
1158 if (key->persample_shading) {
1159 nir_shader *shader = tgsi.ir.nir;
1160 nir_foreach_variable(var, &shader->inputs)
1161 var->data.sample = true;
1162 }
1163
1164 assert(!(key->bitmap && key->drawpixels));
1165
1166 /* glBitmap */
1167 if (key->bitmap) {
1168 nir_lower_bitmap_options options = {0};
1169
1170 variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1171 options.sampler = variant->bitmap_sampler;
1172 options.swizzle_xxxx = st->bitmap.tex_format == PIPE_FORMAT_R8_UNORM;
1173
1174 NIR_PASS_V(tgsi.ir.nir, nir_lower_bitmap, &options);
1175 }
1176
1177 /* glDrawPixels (color only) */
1178 if (key->drawpixels) {
1179 nir_lower_drawpixels_options options = {{0}};
1180 unsigned samplers_used = stfp->Base.SamplersUsed;
1181
1182 /* Find the first unused slot. */
1183 variant->drawpix_sampler = ffs(~samplers_used) - 1;
1184 options.drawpix_sampler = variant->drawpix_sampler;
1185 samplers_used |= (1 << variant->drawpix_sampler);
1186
1187 options.pixel_maps = key->pixelMaps;
1188 if (key->pixelMaps) {
1189 variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1190 options.pixelmap_sampler = variant->pixelmap_sampler;
1191 }
1192
1193 options.scale_and_bias = key->scaleAndBias;
1194 if (key->scaleAndBias) {
1195 _mesa_add_state_reference(params, scale_state);
1196 memcpy(options.scale_state_tokens, scale_state,
1197 sizeof(options.scale_state_tokens));
1198 _mesa_add_state_reference(params, bias_state);
1199 memcpy(options.bias_state_tokens, bias_state,
1200 sizeof(options.bias_state_tokens));
1201 }
1202
1203 _mesa_add_state_reference(params, texcoord_state);
1204 memcpy(options.texcoord_state_tokens, texcoord_state,
1205 sizeof(options.texcoord_state_tokens));
1206
1207 NIR_PASS_V(tgsi.ir.nir, nir_lower_drawpixels, &options);
1208 }
1209
1210 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1211 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv ||
1212 key->external.lower_ayuv || key->external.lower_xyuv)) {
1213 nir_lower_tex_options options = {0};
1214 options.lower_y_uv_external = key->external.lower_nv12;
1215 options.lower_y_u_v_external = key->external.lower_iyuv;
1216 options.lower_xy_uxvx_external = key->external.lower_xy_uxvx;
1217 options.lower_yx_xuxv_external = key->external.lower_yx_xuxv;
1218 options.lower_ayuv_external = key->external.lower_ayuv;
1219 options.lower_xyuv_external = key->external.lower_xyuv;
1220 NIR_PASS_V(tgsi.ir.nir, nir_lower_tex, &options);
1221 }
1222
1223 st_finalize_nir(st, &stfp->Base, stfp->shader_program, tgsi.ir.nir);
1224
1225 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1226 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1227 /* This pass needs to happen *after* nir_lower_sampler */
1228 NIR_PASS_V(tgsi.ir.nir, st_nir_lower_tex_src_plane,
1229 ~stfp->Base.SamplersUsed,
1230 key->external.lower_nv12 || key->external.lower_xy_uxvx ||
1231 key->external.lower_yx_xuxv,
1232 key->external.lower_iyuv);
1233 }
1234
1235 /* Some of the lowering above may have introduced new varyings */
1236 nir_shader_gather_info(tgsi.ir.nir,
1237 nir_shader_get_entrypoint(tgsi.ir.nir));
1238
1239 variant->driver_shader = pipe->create_fs_state(pipe, &tgsi);
1240 variant->key = *key;
1241
1242 return variant;
1243 }
1244
1245 tgsi.tokens = stfp->tgsi.tokens;
1246
1247 assert(!(key->bitmap && key->drawpixels));
1248
1249 /* Fix texture targets and add fog for ATI_fs */
1250 if (stfp->ati_fs) {
1251 const struct tgsi_token *tokens = st_fixup_atifs(tgsi.tokens, key);
1252
1253 if (tokens)
1254 tgsi.tokens = tokens;
1255 else
1256 fprintf(stderr, "mesa: cannot post-process ATI_fs\n");
1257 }
1258
1259 /* Emulate features. */
1260 if (key->clamp_color || key->persample_shading) {
1261 const struct tgsi_token *tokens;
1262 unsigned flags =
1263 (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
1264 (key->persample_shading ? TGSI_EMU_FORCE_PERSAMPLE_INTERP : 0);
1265
1266 tokens = tgsi_emulate(tgsi.tokens, flags);
1267
1268 if (tokens) {
1269 if (tgsi.tokens != stfp->tgsi.tokens)
1270 tgsi_free_tokens(tgsi.tokens);
1271 tgsi.tokens = tokens;
1272 } else
1273 fprintf(stderr, "mesa: cannot emulate deprecated features\n");
1274 }
1275
1276 /* glBitmap */
1277 if (key->bitmap) {
1278 const struct tgsi_token *tokens;
1279
1280 variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1281
1282 tokens = st_get_bitmap_shader(tgsi.tokens,
1283 st->internal_target,
1284 variant->bitmap_sampler,
1285 st->needs_texcoord_semantic,
1286 st->bitmap.tex_format ==
1287 PIPE_FORMAT_R8_UNORM);
1288
1289 if (tokens) {
1290 if (tgsi.tokens != stfp->tgsi.tokens)
1291 tgsi_free_tokens(tgsi.tokens);
1292 tgsi.tokens = tokens;
1293 } else
1294 fprintf(stderr, "mesa: cannot create a shader for glBitmap\n");
1295 }
1296
1297 /* glDrawPixels (color only) */
1298 if (key->drawpixels) {
1299 const struct tgsi_token *tokens;
1300 unsigned scale_const = 0, bias_const = 0, texcoord_const = 0;
1301
1302 /* Find the first unused slot. */
1303 variant->drawpix_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1304
1305 if (key->pixelMaps) {
1306 unsigned samplers_used = stfp->Base.SamplersUsed |
1307 (1 << variant->drawpix_sampler);
1308
1309 variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1310 }
1311
1312 if (key->scaleAndBias) {
1313 scale_const = _mesa_add_state_reference(params, scale_state);
1314 bias_const = _mesa_add_state_reference(params, bias_state);
1315 }
1316
1317 texcoord_const = _mesa_add_state_reference(params, texcoord_state);
1318
1319 tokens = st_get_drawpix_shader(tgsi.tokens,
1320 st->needs_texcoord_semantic,
1321 key->scaleAndBias, scale_const,
1322 bias_const, key->pixelMaps,
1323 variant->drawpix_sampler,
1324 variant->pixelmap_sampler,
1325 texcoord_const, st->internal_target);
1326
1327 if (tokens) {
1328 if (tgsi.tokens != stfp->tgsi.tokens)
1329 tgsi_free_tokens(tgsi.tokens);
1330 tgsi.tokens = tokens;
1331 } else
1332 fprintf(stderr, "mesa: cannot create a shader for glDrawPixels\n");
1333 }
1334
1335 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1336 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1337 const struct tgsi_token *tokens;
1338
1339 /* samplers inserted would conflict, but this should be unpossible: */
1340 assert(!(key->bitmap || key->drawpixels));
1341
1342 tokens = st_tgsi_lower_yuv(tgsi.tokens,
1343 ~stfp->Base.SamplersUsed,
1344 key->external.lower_nv12 ||
1345 key->external.lower_xy_uxvx ||
1346 key->external.lower_yx_xuxv,
1347 key->external.lower_iyuv);
1348 if (tokens) {
1349 if (tgsi.tokens != stfp->tgsi.tokens)
1350 tgsi_free_tokens(tgsi.tokens);
1351 tgsi.tokens = tokens;
1352 } else {
1353 fprintf(stderr, "mesa: cannot create a shader for samplerExternalOES\n");
1354 }
1355 }
1356
1357 if (key->lower_depth_clamp) {
1358 unsigned depth_range_const = _mesa_add_state_reference(params, depth_range_state);
1359
1360 const struct tgsi_token *tokens;
1361 tokens = st_tgsi_lower_depth_clamp_fs(tgsi.tokens, depth_range_const);
1362 if (tgsi.tokens != stfp->tgsi.tokens)
1363 tgsi_free_tokens(tgsi.tokens);
1364 tgsi.tokens = tokens;
1365 }
1366
1367 if (ST_DEBUG & DEBUG_TGSI) {
1368 tgsi_dump(tgsi.tokens, 0);
1369 debug_printf("\n");
1370 }
1371
1372 /* fill in variant */
1373 variant->driver_shader = pipe->create_fs_state(pipe, &tgsi);
1374 variant->key = *key;
1375
1376 if (tgsi.tokens != stfp->tgsi.tokens)
1377 tgsi_free_tokens(tgsi.tokens);
1378 return variant;
1379 }
1380
1381 /**
1382 * Translate fragment program if needed.
1383 */
1384 struct st_fp_variant *
1385 st_get_fp_variant(struct st_context *st,
1386 struct st_fragment_program *stfp,
1387 const struct st_fp_variant_key *key)
1388 {
1389 struct st_fp_variant *fpv;
1390
1391 /* Search for existing variant */
1392 for (fpv = stfp->variants; fpv; fpv = fpv->next) {
1393 if (memcmp(&fpv->key, key, sizeof(*key)) == 0) {
1394 break;
1395 }
1396 }
1397
1398 if (!fpv) {
1399 /* create new */
1400 fpv = st_create_fp_variant(st, stfp, key);
1401 if (fpv) {
1402 if (key->bitmap || key->drawpixels) {
1403 /* Regular variants should always come before the
1404 * bitmap & drawpixels variants, (unless there
1405 * are no regular variants) so that
1406 * st_update_fp can take a fast path when
1407 * shader_has_one_variant is set.
1408 */
1409 if (!stfp->variants) {
1410 stfp->variants = fpv;
1411 } else {
1412 /* insert into list after the first one */
1413 fpv->next = stfp->variants->next;
1414 stfp->variants->next = fpv;
1415 }
1416 } else {
1417 /* insert into list */
1418 fpv->next = stfp->variants;
1419 stfp->variants = fpv;
1420 }
1421 }
1422 }
1423
1424 return fpv;
1425 }
1426
1427 /**
1428 * Translate a program. This is common code for geometry and tessellation
1429 * shaders.
1430 */
1431 bool
1432 st_translate_common_program(struct st_context *st,
1433 struct st_common_program *stcp)
1434 {
1435 struct gl_program *prog = &stcp->Base;
1436 enum pipe_shader_type stage =
1437 pipe_shader_type_from_mesa(stcp->Base.info.stage);
1438 struct ureg_program *ureg = ureg_create_with_screen(stage, st->pipe->screen);
1439
1440 if (ureg == NULL)
1441 return false;
1442
1443 switch (stage) {
1444 case PIPE_SHADER_TESS_CTRL:
1445 ureg_property(ureg, TGSI_PROPERTY_TCS_VERTICES_OUT,
1446 stcp->Base.info.tess.tcs_vertices_out);
1447 break;
1448
1449 case PIPE_SHADER_TESS_EVAL:
1450 if (stcp->Base.info.tess.primitive_mode == GL_ISOLINES)
1451 ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE, GL_LINES);
1452 else
1453 ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE,
1454 stcp->Base.info.tess.primitive_mode);
1455
1456 STATIC_ASSERT((TESS_SPACING_EQUAL + 1) % 3 == PIPE_TESS_SPACING_EQUAL);
1457 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD + 1) % 3 ==
1458 PIPE_TESS_SPACING_FRACTIONAL_ODD);
1459 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN + 1) % 3 ==
1460 PIPE_TESS_SPACING_FRACTIONAL_EVEN);
1461
1462 ureg_property(ureg, TGSI_PROPERTY_TES_SPACING,
1463 (stcp->Base.info.tess.spacing + 1) % 3);
1464
1465 ureg_property(ureg, TGSI_PROPERTY_TES_VERTEX_ORDER_CW,
1466 !stcp->Base.info.tess.ccw);
1467 ureg_property(ureg, TGSI_PROPERTY_TES_POINT_MODE,
1468 stcp->Base.info.tess.point_mode);
1469 break;
1470
1471 case PIPE_SHADER_GEOMETRY:
1472 ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM,
1473 stcp->Base.info.gs.input_primitive);
1474 ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM,
1475 stcp->Base.info.gs.output_primitive);
1476 ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
1477 stcp->Base.info.gs.vertices_out);
1478 ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS,
1479 stcp->Base.info.gs.invocations);
1480 break;
1481
1482 default:
1483 break;
1484 }
1485
1486 ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
1487 ubyte inputMapping[VARYING_SLOT_TESS_MAX];
1488 ubyte outputMapping[VARYING_SLOT_TESS_MAX];
1489 GLuint attr;
1490
1491 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
1492 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
1493 uint num_inputs = 0;
1494
1495 ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
1496 ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
1497 uint num_outputs = 0;
1498
1499 GLint i;
1500
1501 memset(inputSlotToAttr, 0, sizeof(inputSlotToAttr));
1502 memset(inputMapping, 0, sizeof(inputMapping));
1503 memset(outputMapping, 0, sizeof(outputMapping));
1504 memset(&stcp->tgsi, 0, sizeof(stcp->tgsi));
1505
1506 if (prog->info.clip_distance_array_size)
1507 ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
1508 prog->info.clip_distance_array_size);
1509 if (prog->info.cull_distance_array_size)
1510 ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
1511 prog->info.cull_distance_array_size);
1512
1513 /*
1514 * Convert Mesa program inputs to TGSI input register semantics.
1515 */
1516 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1517 if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) == 0)
1518 continue;
1519
1520 unsigned slot = num_inputs++;
1521
1522 inputMapping[attr] = slot;
1523 inputSlotToAttr[slot] = attr;
1524
1525 unsigned semantic_name, semantic_index;
1526 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1527 &semantic_name, &semantic_index);
1528 input_semantic_name[slot] = semantic_name;
1529 input_semantic_index[slot] = semantic_index;
1530 }
1531
1532 /* Also add patch inputs. */
1533 for (attr = 0; attr < 32; attr++) {
1534 if (prog->info.patch_inputs_read & (1u << attr)) {
1535 GLuint slot = num_inputs++;
1536 GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1537
1538 inputMapping[patch_attr] = slot;
1539 inputSlotToAttr[slot] = patch_attr;
1540 input_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1541 input_semantic_index[slot] = attr;
1542 }
1543 }
1544
1545 /* initialize output semantics to defaults */
1546 for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
1547 output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
1548 output_semantic_index[i] = 0;
1549 }
1550
1551 /*
1552 * Determine number of outputs, the (default) output register
1553 * mapping and the semantic information for each output.
1554 */
1555 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1556 if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
1557 GLuint slot = num_outputs++;
1558
1559 outputMapping[attr] = slot;
1560
1561 unsigned semantic_name, semantic_index;
1562 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1563 &semantic_name, &semantic_index);
1564 output_semantic_name[slot] = semantic_name;
1565 output_semantic_index[slot] = semantic_index;
1566 }
1567 }
1568
1569 /* Also add patch outputs. */
1570 for (attr = 0; attr < 32; attr++) {
1571 if (prog->info.patch_outputs_written & (1u << attr)) {
1572 GLuint slot = num_outputs++;
1573 GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1574
1575 outputMapping[patch_attr] = slot;
1576 output_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1577 output_semantic_index[slot] = attr;
1578 }
1579 }
1580
1581 st_translate_program(st->ctx,
1582 stage,
1583 ureg,
1584 stcp->glsl_to_tgsi,
1585 prog,
1586 /* inputs */
1587 num_inputs,
1588 inputMapping,
1589 inputSlotToAttr,
1590 input_semantic_name,
1591 input_semantic_index,
1592 NULL,
1593 /* outputs */
1594 num_outputs,
1595 outputMapping,
1596 output_semantic_name,
1597 output_semantic_index);
1598
1599 stcp->tgsi.tokens = ureg_get_tokens(ureg, &stcp->num_tgsi_tokens);
1600
1601 ureg_destroy(ureg);
1602
1603 st_translate_stream_output_info(prog);
1604
1605 st_store_ir_in_disk_cache(st, prog, false);
1606
1607 if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
1608 _mesa_print_program(prog);
1609 debug_printf("\n");
1610 }
1611
1612 if (ST_DEBUG & DEBUG_TGSI) {
1613 tgsi_dump(stcp->tgsi.tokens, 0);
1614 debug_printf("\n");
1615 }
1616
1617 free_glsl_to_tgsi_visitor(stcp->glsl_to_tgsi);
1618 stcp->glsl_to_tgsi = NULL;
1619 return true;
1620 }
1621
1622
1623 /**
1624 * Get/create a basic program variant.
1625 */
1626 struct st_basic_variant *
1627 st_get_basic_variant(struct st_context *st,
1628 struct st_common_program *prog,
1629 const struct st_basic_variant_key *key)
1630 {
1631 struct pipe_context *pipe = st->pipe;
1632 struct st_basic_variant *v;
1633 struct pipe_shader_state tgsi = {0};
1634
1635 /* Search for existing variant */
1636 for (v = prog->variants; v; v = v->next) {
1637 if (memcmp(&v->key, key, sizeof(*key)) == 0) {
1638 break;
1639 }
1640 }
1641
1642 if (!v) {
1643 /* create new */
1644 v = CALLOC_STRUCT(st_basic_variant);
1645 if (v) {
1646
1647 if (prog->tgsi.type == PIPE_SHADER_IR_NIR) {
1648 tgsi.type = PIPE_SHADER_IR_NIR;
1649 tgsi.ir.nir = nir_shader_clone(NULL, prog->tgsi.ir.nir);
1650
1651 if (key->clamp_color)
1652 NIR_PASS_V(tgsi.ir.nir, nir_lower_clamp_color_outputs);
1653
1654 tgsi.stream_output = prog->tgsi.stream_output;
1655
1656 st_finalize_nir(st, &prog->Base, prog->shader_program,
1657 tgsi.ir.nir);
1658 } else {
1659 if (key->lower_depth_clamp) {
1660 struct gl_program_parameter_list *params = prog->Base.Parameters;
1661
1662 unsigned depth_range_const =
1663 _mesa_add_state_reference(params, depth_range_state);
1664
1665 const struct tgsi_token *tokens;
1666 tokens =
1667 st_tgsi_lower_depth_clamp(prog->tgsi.tokens,
1668 depth_range_const,
1669 key->clip_negative_one_to_one);
1670
1671 if (tokens != prog->tgsi.tokens)
1672 tgsi_free_tokens(prog->tgsi.tokens);
1673
1674 prog->tgsi.tokens = tokens;
1675 prog->num_tgsi_tokens = tgsi_num_tokens(tokens);
1676 }
1677 tgsi = prog->tgsi;
1678 }
1679 /* fill in new variant */
1680 switch (prog->Base.info.stage) {
1681 case MESA_SHADER_TESS_CTRL:
1682 v->driver_shader = pipe->create_tcs_state(pipe, &tgsi);
1683 break;
1684 case MESA_SHADER_TESS_EVAL:
1685 v->driver_shader = pipe->create_tes_state(pipe, &tgsi);
1686 break;
1687 case MESA_SHADER_GEOMETRY:
1688 v->driver_shader = pipe->create_gs_state(pipe, &tgsi);
1689 break;
1690 case MESA_SHADER_COMPUTE: {
1691 struct pipe_compute_state cs = {0};
1692 cs.ir_type = tgsi.type;
1693 cs.req_local_mem = prog->Base.info.cs.shared_size;
1694
1695 if (tgsi.type == PIPE_SHADER_IR_NIR)
1696 cs.prog = tgsi.ir.nir;
1697 else
1698 cs.prog = tgsi.tokens;
1699
1700 v->driver_shader = pipe->create_compute_state(pipe, &cs);
1701 break;
1702 }
1703 default:
1704 assert(!"unhandled shader type");
1705 free(v);
1706 return NULL;
1707 }
1708
1709 v->key = *key;
1710
1711 /* insert into list */
1712 v->next = prog->variants;
1713 prog->variants = v;
1714 }
1715 }
1716
1717 return v;
1718 }
1719
1720
1721 /**
1722 * Vert/Geom/Frag programs have per-context variants. Free all the
1723 * variants attached to the given program which match the given context.
1724 */
1725 static void
1726 destroy_program_variants(struct st_context *st, struct gl_program *target)
1727 {
1728 if (!target || target == &_mesa_DummyProgram)
1729 return;
1730
1731 switch (target->Target) {
1732 case GL_VERTEX_PROGRAM_ARB:
1733 {
1734 struct st_vertex_program *stvp = (struct st_vertex_program *) target;
1735 struct st_vp_variant *vpv, **prevPtr = &stvp->variants;
1736
1737 for (vpv = stvp->variants; vpv; ) {
1738 struct st_vp_variant *next = vpv->next;
1739 if (vpv->key.st == st) {
1740 /* unlink from list */
1741 *prevPtr = next;
1742 /* destroy this variant */
1743 delete_vp_variant(st, vpv);
1744 }
1745 else {
1746 prevPtr = &vpv->next;
1747 }
1748 vpv = next;
1749 }
1750 }
1751 break;
1752 case GL_FRAGMENT_PROGRAM_ARB:
1753 {
1754 struct st_fragment_program *stfp =
1755 (struct st_fragment_program *) target;
1756 struct st_fp_variant *fpv, **prevPtr = &stfp->variants;
1757
1758 for (fpv = stfp->variants; fpv; ) {
1759 struct st_fp_variant *next = fpv->next;
1760 if (fpv->key.st == st) {
1761 /* unlink from list */
1762 *prevPtr = next;
1763 /* destroy this variant */
1764 delete_fp_variant(st, fpv);
1765 }
1766 else {
1767 prevPtr = &fpv->next;
1768 }
1769 fpv = next;
1770 }
1771 }
1772 break;
1773 case GL_GEOMETRY_PROGRAM_NV:
1774 case GL_TESS_CONTROL_PROGRAM_NV:
1775 case GL_TESS_EVALUATION_PROGRAM_NV:
1776 case GL_COMPUTE_PROGRAM_NV:
1777 {
1778 struct st_common_program *p = st_common_program(target);
1779 struct st_basic_variant *v, **prevPtr = &p->variants;
1780
1781 for (v = p->variants; v; ) {
1782 struct st_basic_variant *next = v->next;
1783 if (v->key.st == st) {
1784 /* unlink from list */
1785 *prevPtr = next;
1786 /* destroy this variant */
1787 delete_basic_variant(st, v, target->Target);
1788 }
1789 else {
1790 prevPtr = &v->next;
1791 }
1792 v = next;
1793 }
1794 }
1795 break;
1796 default:
1797 _mesa_problem(NULL, "Unexpected program target 0x%x in "
1798 "destroy_program_variants_cb()", target->Target);
1799 }
1800 }
1801
1802
1803 /**
1804 * Callback for _mesa_HashWalk. Free all the shader's program variants
1805 * which match the given context.
1806 */
1807 static void
1808 destroy_shader_program_variants_cb(GLuint key, void *data, void *userData)
1809 {
1810 struct st_context *st = (struct st_context *) userData;
1811 struct gl_shader *shader = (struct gl_shader *) data;
1812
1813 switch (shader->Type) {
1814 case GL_SHADER_PROGRAM_MESA:
1815 {
1816 struct gl_shader_program *shProg = (struct gl_shader_program *) data;
1817 GLuint i;
1818
1819 for (i = 0; i < ARRAY_SIZE(shProg->_LinkedShaders); i++) {
1820 if (shProg->_LinkedShaders[i])
1821 destroy_program_variants(st, shProg->_LinkedShaders[i]->Program);
1822 }
1823 }
1824 break;
1825 case GL_VERTEX_SHADER:
1826 case GL_FRAGMENT_SHADER:
1827 case GL_GEOMETRY_SHADER:
1828 case GL_TESS_CONTROL_SHADER:
1829 case GL_TESS_EVALUATION_SHADER:
1830 case GL_COMPUTE_SHADER:
1831 break;
1832 default:
1833 assert(0);
1834 }
1835 }
1836
1837
1838 /**
1839 * Callback for _mesa_HashWalk. Free all the program variants which match
1840 * the given context.
1841 */
1842 static void
1843 destroy_program_variants_cb(GLuint key, void *data, void *userData)
1844 {
1845 struct st_context *st = (struct st_context *) userData;
1846 struct gl_program *program = (struct gl_program *) data;
1847 destroy_program_variants(st, program);
1848 }
1849
1850
1851 /**
1852 * Walk over all shaders and programs to delete any variants which
1853 * belong to the given context.
1854 * This is called during context tear-down.
1855 */
1856 void
1857 st_destroy_program_variants(struct st_context *st)
1858 {
1859 /* If shaders can be shared with other contexts, the last context will
1860 * call DeleteProgram on all shaders, releasing everything.
1861 */
1862 if (st->has_shareable_shaders)
1863 return;
1864
1865 /* ARB vert/frag program */
1866 _mesa_HashWalk(st->ctx->Shared->Programs,
1867 destroy_program_variants_cb, st);
1868
1869 /* GLSL vert/frag/geom shaders */
1870 _mesa_HashWalk(st->ctx->Shared->ShaderObjects,
1871 destroy_shader_program_variants_cb, st);
1872 }
1873
1874
1875 /**
1876 * For debugging, print/dump the current vertex program.
1877 */
1878 void
1879 st_print_current_vertex_program(void)
1880 {
1881 GET_CURRENT_CONTEXT(ctx);
1882
1883 if (ctx->VertexProgram._Current) {
1884 struct st_vertex_program *stvp =
1885 (struct st_vertex_program *) ctx->VertexProgram._Current;
1886 struct st_vp_variant *stv;
1887
1888 debug_printf("Vertex program %u\n", stvp->Base.Id);
1889
1890 for (stv = stvp->variants; stv; stv = stv->next) {
1891 debug_printf("variant %p\n", stv);
1892 tgsi_dump(stv->tgsi.tokens, 0);
1893 }
1894 }
1895 }
1896
1897
1898 /**
1899 * Compile one shader variant.
1900 */
1901 void
1902 st_precompile_shader_variant(struct st_context *st,
1903 struct gl_program *prog)
1904 {
1905 switch (prog->Target) {
1906 case GL_VERTEX_PROGRAM_ARB: {
1907 struct st_vertex_program *p = (struct st_vertex_program *)prog;
1908 struct st_vp_variant_key key;
1909
1910 memset(&key, 0, sizeof(key));
1911
1912 key.st = st->has_shareable_shaders ? NULL : st;
1913 st_get_vp_variant(st, p, &key);
1914 break;
1915 }
1916
1917 case GL_FRAGMENT_PROGRAM_ARB: {
1918 struct st_fragment_program *p = (struct st_fragment_program *)prog;
1919 struct st_fp_variant_key key;
1920
1921 memset(&key, 0, sizeof(key));
1922
1923 key.st = st->has_shareable_shaders ? NULL : st;
1924 st_get_fp_variant(st, p, &key);
1925 break;
1926 }
1927
1928 case GL_TESS_CONTROL_PROGRAM_NV:
1929 case GL_TESS_EVALUATION_PROGRAM_NV:
1930 case GL_GEOMETRY_PROGRAM_NV:
1931 case GL_COMPUTE_PROGRAM_NV: {
1932 struct st_common_program *p = st_common_program(prog);
1933 struct st_basic_variant_key key;
1934
1935 memset(&key, 0, sizeof(key));
1936
1937 key.st = st->has_shareable_shaders ? NULL : st;
1938 st_get_basic_variant(st, p, &key);
1939 break;
1940 }
1941
1942 default:
1943 assert(0);
1944 }
1945 }