projects / mesa.git / blob
? search:


c0e63b8a98c8cae18625c7fe25977c7085885ad9
[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 "util/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 #include "compiler/nir/nir_serialize.h"
45 #include "draw/draw_context.h"
46
47 #include "pipe/p_context.h"
48 #include "pipe/p_defines.h"
49 #include "pipe/p_shader_tokens.h"
50 #include "draw/draw_context.h"
51 #include "tgsi/tgsi_dump.h"
52 #include "tgsi/tgsi_emulate.h"
53 #include "tgsi/tgsi_parse.h"
54 #include "tgsi/tgsi_ureg.h"
55
56 #include "st_debug.h"
57 #include "st_cb_bitmap.h"
58 #include "st_cb_drawpixels.h"
59 #include "st_context.h"
60 #include "st_tgsi_lower_depth_clamp.h"
61 #include "st_tgsi_lower_yuv.h"
62 #include "st_program.h"
63 #include "st_mesa_to_tgsi.h"
64 #include "st_atifs_to_tgsi.h"
65 #include "st_nir.h"
66 #include "st_shader_cache.h"
67 #include "st_util.h"
68 #include "cso_cache/cso_context.h"
69
70
71
72 static void
73 set_affected_state_flags(uint64_t *states,
74 struct gl_program *prog,
75 uint64_t new_constants,
76 uint64_t new_sampler_views,
77 uint64_t new_samplers,
78 uint64_t new_images,
79 uint64_t new_ubos,
80 uint64_t new_ssbos,
81 uint64_t new_atomics)
82 {
83 if (prog->Parameters->NumParameters)
84 *states |= new_constants;
85
86 if (prog->info.num_textures)
87 *states |= new_sampler_views | new_samplers;
88
89 if (prog->info.num_images)
90 *states |= new_images;
91
92 if (prog->info.num_ubos)
93 *states |= new_ubos;
94
95 if (prog->info.num_ssbos)
96 *states |= new_ssbos;
97
98 if (prog->info.num_abos)
99 *states |= new_atomics;
100 }
101
102 /**
103 * This determines which states will be updated when the shader is bound.
104 */
105 void
106 st_set_prog_affected_state_flags(struct gl_program *prog)
107 {
108 uint64_t *states;
109
110 switch (prog->info.stage) {
111 case MESA_SHADER_VERTEX:
112 states = &((struct st_program*)prog)->affected_states;
113
114 *states = ST_NEW_VS_STATE |
115 ST_NEW_RASTERIZER |
116 ST_NEW_VERTEX_ARRAYS;
117
118 set_affected_state_flags(states, prog,
119 ST_NEW_VS_CONSTANTS,
120 ST_NEW_VS_SAMPLER_VIEWS,
121 ST_NEW_VS_SAMPLERS,
122 ST_NEW_VS_IMAGES,
123 ST_NEW_VS_UBOS,
124 ST_NEW_VS_SSBOS,
125 ST_NEW_VS_ATOMICS);
126 break;
127
128 case MESA_SHADER_TESS_CTRL:
129 states = &(st_program(prog))->affected_states;
130
131 *states = ST_NEW_TCS_STATE;
132
133 set_affected_state_flags(states, prog,
134 ST_NEW_TCS_CONSTANTS,
135 ST_NEW_TCS_SAMPLER_VIEWS,
136 ST_NEW_TCS_SAMPLERS,
137 ST_NEW_TCS_IMAGES,
138 ST_NEW_TCS_UBOS,
139 ST_NEW_TCS_SSBOS,
140 ST_NEW_TCS_ATOMICS);
141 break;
142
143 case MESA_SHADER_TESS_EVAL:
144 states = &(st_program(prog))->affected_states;
145
146 *states = ST_NEW_TES_STATE |
147 ST_NEW_RASTERIZER;
148
149 set_affected_state_flags(states, prog,
150 ST_NEW_TES_CONSTANTS,
151 ST_NEW_TES_SAMPLER_VIEWS,
152 ST_NEW_TES_SAMPLERS,
153 ST_NEW_TES_IMAGES,
154 ST_NEW_TES_UBOS,
155 ST_NEW_TES_SSBOS,
156 ST_NEW_TES_ATOMICS);
157 break;
158
159 case MESA_SHADER_GEOMETRY:
160 states = &(st_program(prog))->affected_states;
161
162 *states = ST_NEW_GS_STATE |
163 ST_NEW_RASTERIZER;
164
165 set_affected_state_flags(states, prog,
166 ST_NEW_GS_CONSTANTS,
167 ST_NEW_GS_SAMPLER_VIEWS,
168 ST_NEW_GS_SAMPLERS,
169 ST_NEW_GS_IMAGES,
170 ST_NEW_GS_UBOS,
171 ST_NEW_GS_SSBOS,
172 ST_NEW_GS_ATOMICS);
173 break;
174
175 case MESA_SHADER_FRAGMENT:
176 states = &((struct st_program*)prog)->affected_states;
177
178 /* gl_FragCoord and glDrawPixels always use constants. */
179 *states = ST_NEW_FS_STATE |
180 ST_NEW_SAMPLE_SHADING |
181 ST_NEW_FS_CONSTANTS;
182
183 set_affected_state_flags(states, prog,
184 ST_NEW_FS_CONSTANTS,
185 ST_NEW_FS_SAMPLER_VIEWS,
186 ST_NEW_FS_SAMPLERS,
187 ST_NEW_FS_IMAGES,
188 ST_NEW_FS_UBOS,
189 ST_NEW_FS_SSBOS,
190 ST_NEW_FS_ATOMICS);
191 break;
192
193 case MESA_SHADER_COMPUTE:
194 states = &((struct st_program*)prog)->affected_states;
195
196 *states = ST_NEW_CS_STATE;
197
198 set_affected_state_flags(states, prog,
199 ST_NEW_CS_CONSTANTS,
200 ST_NEW_CS_SAMPLER_VIEWS,
201 ST_NEW_CS_SAMPLERS,
202 ST_NEW_CS_IMAGES,
203 ST_NEW_CS_UBOS,
204 ST_NEW_CS_SSBOS,
205 ST_NEW_CS_ATOMICS);
206 break;
207
208 default:
209 unreachable("unhandled shader stage");
210 }
211 }
212
213
214 /**
215 * Delete a shader variant. Note the caller must unlink the variant from
216 * the linked list.
217 */
218 static void
219 delete_variant(struct st_context *st, struct st_variant *v, GLenum target)
220 {
221 if (v->driver_shader) {
222 if (target == GL_VERTEX_PROGRAM_ARB &&
223 ((struct st_common_variant*)v)->key.is_draw_shader) {
224 /* Draw shader. */
225 draw_delete_vertex_shader(st->draw, v->driver_shader);
226 } else if (st->has_shareable_shaders || v->st == st) {
227 /* The shader's context matches the calling context, or we
228 * don't care.
229 */
230 switch (target) {
231 case GL_VERTEX_PROGRAM_ARB:
232 st->pipe->delete_vs_state(st->pipe, v->driver_shader);
233 break;
234 case GL_TESS_CONTROL_PROGRAM_NV:
235 st->pipe->delete_tcs_state(st->pipe, v->driver_shader);
236 break;
237 case GL_TESS_EVALUATION_PROGRAM_NV:
238 st->pipe->delete_tes_state(st->pipe, v->driver_shader);
239 break;
240 case GL_GEOMETRY_PROGRAM_NV:
241 st->pipe->delete_gs_state(st->pipe, v->driver_shader);
242 break;
243 case GL_FRAGMENT_PROGRAM_ARB:
244 st->pipe->delete_fs_state(st->pipe, v->driver_shader);
245 break;
246 case GL_COMPUTE_PROGRAM_NV:
247 st->pipe->delete_compute_state(st->pipe, v->driver_shader);
248 break;
249 default:
250 unreachable("bad shader type in delete_basic_variant");
251 }
252 } else {
253 /* We can't delete a shader with a context different from the one
254 * that created it. Add it to the creating context's zombie list.
255 */
256 enum pipe_shader_type type =
257 pipe_shader_type_from_mesa(_mesa_program_enum_to_shader_stage(target));
258
259 st_save_zombie_shader(v->st, type, v->driver_shader);
260 }
261 }
262
263 free(v);
264 }
265
266 static void
267 st_unbind_program(struct st_context *st, struct st_program *p)
268 {
269 /* Unbind the shader in cso_context and re-bind in st/mesa. */
270 switch (p->Base.info.stage) {
271 case MESA_SHADER_VERTEX:
272 cso_set_vertex_shader_handle(st->cso_context, NULL);
273 st->dirty |= ST_NEW_VS_STATE;
274 break;
275 case MESA_SHADER_TESS_CTRL:
276 cso_set_tessctrl_shader_handle(st->cso_context, NULL);
277 st->dirty |= ST_NEW_TCS_STATE;
278 break;
279 case MESA_SHADER_TESS_EVAL:
280 cso_set_tesseval_shader_handle(st->cso_context, NULL);
281 st->dirty |= ST_NEW_TES_STATE;
282 break;
283 case MESA_SHADER_GEOMETRY:
284 cso_set_geometry_shader_handle(st->cso_context, NULL);
285 st->dirty |= ST_NEW_GS_STATE;
286 break;
287 case MESA_SHADER_FRAGMENT:
288 cso_set_fragment_shader_handle(st->cso_context, NULL);
289 st->dirty |= ST_NEW_FS_STATE;
290 break;
291 case MESA_SHADER_COMPUTE:
292 cso_set_compute_shader_handle(st->cso_context, NULL);
293 st->dirty |= ST_NEW_CS_STATE;
294 break;
295 default:
296 unreachable("invalid shader type");
297 }
298 }
299
300 /**
301 * Free all basic program variants.
302 */
303 void
304 st_release_variants(struct st_context *st, struct st_program *p)
305 {
306 struct st_variant *v;
307
308 /* If we are releasing shaders, re-bind them, because we don't
309 * know which shaders are bound in the driver.
310 */
311 if (p->variants)
312 st_unbind_program(st, p);
313
314 for (v = p->variants; v; ) {
315 struct st_variant *next = v->next;
316 delete_variant(st, v, p->Base.Target);
317 v = next;
318 }
319
320 p->variants = NULL;
321
322 if (p->state.tokens) {
323 ureg_free_tokens(p->state.tokens);
324 p->state.tokens = NULL;
325 }
326
327 /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
328 * it has resulted in the driver taking ownership of the NIR. Those
329 * callers should be NULLing out the nir field in any pipe_shader_state
330 * that might have this called in order to indicate that.
331 *
332 * GLSL IR and ARB programs will have set gl_program->nir to the same
333 * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
334 */
335 }
336
337 /**
338 * Free all basic program variants and unref program.
339 */
340 void
341 st_release_program(struct st_context *st, struct st_program **p)
342 {
343 if (!*p)
344 return;
345
346 st_release_variants(st, *p);
347 st_reference_prog(st, p, NULL);
348 }
349
350 void
351 st_finalize_nir_before_variants(struct nir_shader *nir)
352 {
353 NIR_PASS_V(nir, nir_opt_access);
354
355 NIR_PASS_V(nir, nir_split_var_copies);
356 NIR_PASS_V(nir, nir_lower_var_copies);
357 if (nir->options->lower_all_io_to_temps ||
358 nir->options->lower_all_io_to_elements ||
359 nir->info.stage == MESA_SHADER_VERTEX ||
360 nir->info.stage == MESA_SHADER_GEOMETRY) {
361 NIR_PASS_V(nir, nir_lower_io_arrays_to_elements_no_indirects, false);
362 } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
363 NIR_PASS_V(nir, nir_lower_io_arrays_to_elements_no_indirects, true);
364 }
365
366 st_nir_assign_vs_in_locations(nir);
367 }
368
369 /**
370 * Translate ARB (asm) program to NIR
371 */
372 static nir_shader *
373 st_translate_prog_to_nir(struct st_context *st, struct gl_program *prog,
374 gl_shader_stage stage)
375 {
376 struct pipe_screen *screen = st->pipe->screen;
377 const struct gl_shader_compiler_options *options =
378 &st->ctx->Const.ShaderCompilerOptions[stage];
379
380 /* Translate to NIR */
381 nir_shader *nir = prog_to_nir(prog, options->NirOptions);
382 NIR_PASS_V(nir, nir_lower_regs_to_ssa); /* turn registers into SSA */
383 nir_validate_shader(nir, "after st/ptn lower_regs_to_ssa");
384
385 NIR_PASS_V(nir, st_nir_lower_wpos_ytransform, prog, screen);
386 NIR_PASS_V(nir, nir_lower_system_values);
387
388 /* Optimise NIR */
389 NIR_PASS_V(nir, nir_opt_constant_folding);
390 st_nir_opts(nir);
391 st_finalize_nir_before_variants(nir);
392
393 if (st->allow_st_finalize_nir_twice)
394 st_finalize_nir(st, prog, NULL, nir, true);
395
396 nir_validate_shader(nir, "after st/glsl finalize_nir");
397
398 return nir;
399 }
400
401 void
402 st_prepare_vertex_program(struct st_program *stp)
403 {
404 struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
405
406 stvp->num_inputs = 0;
407 memset(stvp->input_to_index, ~0, sizeof(stvp->input_to_index));
408 memset(stvp->result_to_output, ~0, sizeof(stvp->result_to_output));
409
410 /* Determine number of inputs, the mappings between VERT_ATTRIB_x
411 * and TGSI generic input indexes, plus input attrib semantic info.
412 */
413 for (unsigned attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
414 if ((stp->Base.info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
415 stvp->input_to_index[attr] = stvp->num_inputs;
416 stvp->index_to_input[stvp->num_inputs] = attr;
417 stvp->num_inputs++;
418
419 if ((stp->Base.DualSlotInputs & BITFIELD64_BIT(attr)) != 0) {
420 /* add placeholder for second part of a double attribute */
421 stvp->index_to_input[stvp->num_inputs] = ST_DOUBLE_ATTRIB_PLACEHOLDER;
422 stvp->num_inputs++;
423 }
424 }
425 }
426 /* pre-setup potentially unused edgeflag input */
427 stvp->input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
428 stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
429
430 /* Compute mapping of vertex program outputs to slots. */
431 unsigned num_outputs = 0;
432 for (unsigned attr = 0; attr < VARYING_SLOT_MAX; attr++) {
433 if (stp->Base.info.outputs_written & BITFIELD64_BIT(attr))
434 stvp->result_to_output[attr] = num_outputs++;
435 }
436 /* pre-setup potentially unused edgeflag output */
437 stvp->result_to_output[VARYING_SLOT_EDGE] = num_outputs;
438 }
439
440 void
441 st_translate_stream_output_info(struct gl_program *prog)
442 {
443 struct gl_transform_feedback_info *info = prog->sh.LinkedTransformFeedback;
444 if (!info)
445 return;
446
447 /* Determine the (default) output register mapping for each output. */
448 unsigned num_outputs = 0;
449 ubyte output_mapping[VARYING_SLOT_TESS_MAX];
450 memset(output_mapping, 0, sizeof(output_mapping));
451
452 for (unsigned attr = 0; attr < VARYING_SLOT_MAX; attr++) {
453 if (prog->info.outputs_written & BITFIELD64_BIT(attr))
454 output_mapping[attr] = num_outputs++;
455 }
456
457 /* Translate stream output info. */
458 struct pipe_stream_output_info *so_info =
459 &((struct st_program*)prog)->state.stream_output;
460
461 for (unsigned i = 0; i < info->NumOutputs; i++) {
462 so_info->output[i].register_index =
463 output_mapping[info->Outputs[i].OutputRegister];
464 so_info->output[i].start_component = info->Outputs[i].ComponentOffset;
465 so_info->output[i].num_components = info->Outputs[i].NumComponents;
466 so_info->output[i].output_buffer = info->Outputs[i].OutputBuffer;
467 so_info->output[i].dst_offset = info->Outputs[i].DstOffset;
468 so_info->output[i].stream = info->Outputs[i].StreamId;
469 }
470
471 for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
472 so_info->stride[i] = info->Buffers[i].Stride;
473 }
474 so_info->num_outputs = info->NumOutputs;
475 }
476
477 /**
478 * Translate a vertex program.
479 */
480 bool
481 st_translate_vertex_program(struct st_context *st,
482 struct st_program *stp)
483 {
484 struct ureg_program *ureg;
485 enum pipe_error error;
486 unsigned num_outputs = 0;
487 unsigned attr;
488 ubyte output_semantic_name[VARYING_SLOT_MAX] = {0};
489 ubyte output_semantic_index[VARYING_SLOT_MAX] = {0};
490
491 if (stp->Base.arb.IsPositionInvariant)
492 _mesa_insert_mvp_code(st->ctx, &stp->Base);
493
494 st_prepare_vertex_program(stp);
495
496 /* ARB_vp: */
497 if (!stp->glsl_to_tgsi) {
498 _mesa_remove_output_reads(&stp->Base, PROGRAM_OUTPUT);
499
500 /* This determines which states will be updated when the assembly
501 * shader is bound.
502 */
503 stp->affected_states = ST_NEW_VS_STATE |
504 ST_NEW_RASTERIZER |
505 ST_NEW_VERTEX_ARRAYS;
506
507 if (stp->Base.Parameters->NumParameters)
508 stp->affected_states |= ST_NEW_VS_CONSTANTS;
509
510 /* Translate to NIR if preferred. */
511 if (st->pipe->screen->get_shader_param(st->pipe->screen,
512 PIPE_SHADER_VERTEX,
513 PIPE_SHADER_CAP_PREFERRED_IR)) {
514 assert(!stp->glsl_to_tgsi);
515
516 if (stp->Base.nir)
517 ralloc_free(stp->Base.nir);
518
519 if (stp->serialized_nir) {
520 free(stp->serialized_nir);
521 stp->serialized_nir = NULL;
522 }
523
524 stp->state.type = PIPE_SHADER_IR_NIR;
525 stp->Base.nir = st_translate_prog_to_nir(st, &stp->Base,
526 MESA_SHADER_VERTEX);
527 /* For st_draw_feedback, we need to generate TGSI too if draw doesn't
528 * use LLVM.
529 */
530 if (draw_has_llvm())
531 return true;
532 }
533 }
534
535 /* Get semantic names and indices. */
536 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
537 if (stp->Base.info.outputs_written & BITFIELD64_BIT(attr)) {
538 unsigned slot = num_outputs++;
539 unsigned semantic_name, semantic_index;
540 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
541 &semantic_name, &semantic_index);
542 output_semantic_name[slot] = semantic_name;
543 output_semantic_index[slot] = semantic_index;
544 }
545 }
546 /* pre-setup potentially unused edgeflag output */
547 output_semantic_name[num_outputs] = TGSI_SEMANTIC_EDGEFLAG;
548 output_semantic_index[num_outputs] = 0;
549
550 ureg = ureg_create_with_screen(PIPE_SHADER_VERTEX, st->pipe->screen);
551 if (ureg == NULL)
552 return false;
553
554 if (stp->Base.info.clip_distance_array_size)
555 ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
556 stp->Base.info.clip_distance_array_size);
557 if (stp->Base.info.cull_distance_array_size)
558 ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
559 stp->Base.info.cull_distance_array_size);
560
561 if (ST_DEBUG & DEBUG_MESA) {
562 _mesa_print_program(&stp->Base);
563 _mesa_print_program_parameters(st->ctx, &stp->Base);
564 debug_printf("\n");
565 }
566
567 struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
568
569 if (stp->glsl_to_tgsi) {
570 error = st_translate_program(st->ctx,
571 PIPE_SHADER_VERTEX,
572 ureg,
573 stp->glsl_to_tgsi,
574 &stp->Base,
575 /* inputs */
576 stvp->num_inputs,
577 stvp->input_to_index,
578 NULL, /* inputSlotToAttr */
579 NULL, /* input semantic name */
580 NULL, /* input semantic index */
581 NULL, /* interp mode */
582 /* outputs */
583 num_outputs,
584 stvp->result_to_output,
585 output_semantic_name,
586 output_semantic_index);
587
588 st_translate_stream_output_info(&stp->Base);
589
590 free_glsl_to_tgsi_visitor(stp->glsl_to_tgsi);
591 } else
592 error = st_translate_mesa_program(st->ctx,
593 PIPE_SHADER_VERTEX,
594 ureg,
595 &stp->Base,
596 /* inputs */
597 stvp->num_inputs,
598 stvp->input_to_index,
599 NULL, /* input semantic name */
600 NULL, /* input semantic index */
601 NULL,
602 /* outputs */
603 num_outputs,
604 stvp->result_to_output,
605 output_semantic_name,
606 output_semantic_index);
607
608 if (error) {
609 debug_printf("%s: failed to translate Mesa program:\n", __func__);
610 _mesa_print_program(&stp->Base);
611 debug_assert(0);
612 return false;
613 }
614
615 stp->state.tokens = ureg_get_tokens(ureg, NULL);
616 ureg_destroy(ureg);
617
618 if (stp->glsl_to_tgsi) {
619 stp->glsl_to_tgsi = NULL;
620 st_store_ir_in_disk_cache(st, &stp->Base, false);
621 }
622
623 return stp->state.tokens != NULL;
624 }
625
626 static struct nir_shader *
627 get_nir_shader(struct st_context *st, struct st_program *stp)
628 {
629 if (stp->Base.nir) {
630 nir_shader *nir = stp->Base.nir;
631
632 /* The first shader variant takes ownership of NIR, so that there is
633 * no cloning. Additional shader variants are always generated from
634 * serialized NIR to save memory.
635 */
636 stp->Base.nir = NULL;
637 assert(stp->serialized_nir && stp->serialized_nir_size);
638 return nir;
639 }
640
641 struct blob_reader blob_reader;
642 const struct nir_shader_compiler_options *options =
643 st->ctx->Const.ShaderCompilerOptions[stp->Base.info.stage].NirOptions;
644
645 blob_reader_init(&blob_reader, stp->serialized_nir, stp->serialized_nir_size);
646 return nir_deserialize(NULL, options, &blob_reader);
647 }
648
649 static const gl_state_index16 depth_range_state[STATE_LENGTH] =
650 { STATE_DEPTH_RANGE };
651
652 static struct st_common_variant *
653 st_create_vp_variant(struct st_context *st,
654 struct st_program *stvp,
655 const struct st_common_variant_key *key)
656 {
657 struct st_common_variant *vpv = CALLOC_STRUCT(st_common_variant);
658 struct pipe_context *pipe = st->pipe;
659 struct pipe_screen *screen = pipe->screen;
660 struct pipe_shader_state state = {0};
661
662 static const gl_state_index16 point_size_state[STATE_LENGTH] =
663 { STATE_INTERNAL, STATE_POINT_SIZE_CLAMPED, 0 };
664 struct gl_program_parameter_list *params = stvp->Base.Parameters;
665
666 vpv->key = *key;
667
668 state.stream_output = stvp->state.stream_output;
669
670 if (stvp->state.type == PIPE_SHADER_IR_NIR &&
671 (!key->is_draw_shader || draw_has_llvm())) {
672 bool finalize = false;
673
674 state.type = PIPE_SHADER_IR_NIR;
675 state.ir.nir = get_nir_shader(st, stvp);
676 if (key->clamp_color) {
677 NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
678 finalize = true;
679 }
680 if (key->passthrough_edgeflags) {
681 NIR_PASS_V(state.ir.nir, nir_lower_passthrough_edgeflags);
682 finalize = true;
683 }
684
685 if (key->lower_point_size) {
686 _mesa_add_state_reference(params, point_size_state);
687 NIR_PASS_V(state.ir.nir, nir_lower_point_size_mov,
688 point_size_state);
689 finalize = true;
690 }
691
692 if (key->lower_ucp) {
693 bool can_compact = screen->get_param(screen,
694 PIPE_CAP_NIR_COMPACT_ARRAYS);
695
696 bool use_eye = st->ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX] != NULL;
697 gl_state_index16 clipplane_state[MAX_CLIP_PLANES][STATE_LENGTH];
698 for (int i = 0; i < MAX_CLIP_PLANES; ++i) {
699 if (use_eye) {
700 clipplane_state[i][0] = STATE_CLIPPLANE;
701 clipplane_state[i][1] = i;
702 } else {
703 clipplane_state[i][0] = STATE_INTERNAL;
704 clipplane_state[i][1] = STATE_CLIP_INTERNAL;
705 clipplane_state[i][2] = i;
706 }
707 _mesa_add_state_reference(params, clipplane_state[i]);
708 }
709
710 NIR_PASS_V(state.ir.nir, nir_lower_clip_vs, key->lower_ucp,
711 true, can_compact, clipplane_state);
712 NIR_PASS_V(state.ir.nir, nir_lower_io_to_temporaries,
713 nir_shader_get_entrypoint(state.ir.nir), true, false);
714 NIR_PASS_V(state.ir.nir, nir_lower_global_vars_to_local);
715 finalize = true;
716 }
717
718 if (finalize || !st->allow_st_finalize_nir_twice) {
719 st_finalize_nir(st, &stvp->Base, stvp->shader_program, state.ir.nir,
720 true);
721
722 /* Some of the lowering above may have introduced new varyings */
723 nir_shader_gather_info(state.ir.nir,
724 nir_shader_get_entrypoint(state.ir.nir));
725 }
726
727 if (ST_DEBUG & DEBUG_PRINT_IR)
728 nir_print_shader(state.ir.nir, stderr);
729
730 if (key->is_draw_shader)
731 vpv->base.driver_shader = draw_create_vertex_shader(st->draw, &state);
732 else
733 vpv->base.driver_shader = pipe->create_vs_state(pipe, &state);
734
735 return vpv;
736 }
737
738 state.type = PIPE_SHADER_IR_TGSI;
739 state.tokens = tgsi_dup_tokens(stvp->state.tokens);
740
741 /* Emulate features. */
742 if (key->clamp_color || key->passthrough_edgeflags) {
743 const struct tgsi_token *tokens;
744 unsigned flags =
745 (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
746 (key->passthrough_edgeflags ? TGSI_EMU_PASSTHROUGH_EDGEFLAG : 0);
747
748 tokens = tgsi_emulate(state.tokens, flags);
749
750 if (tokens) {
751 tgsi_free_tokens(state.tokens);
752 state.tokens = tokens;
753 } else {
754 fprintf(stderr, "mesa: cannot emulate deprecated features\n");
755 }
756 }
757
758 if (key->lower_depth_clamp) {
759 unsigned depth_range_const =
760 _mesa_add_state_reference(params, depth_range_state);
761
762 const struct tgsi_token *tokens;
763 tokens = st_tgsi_lower_depth_clamp(state.tokens, depth_range_const,
764 key->clip_negative_one_to_one);
765 if (tokens != state.tokens)
766 tgsi_free_tokens(state.tokens);
767 state.tokens = tokens;
768 }
769
770 if (ST_DEBUG & DEBUG_PRINT_IR)
771 tgsi_dump(state.tokens, 0);
772
773 if (key->is_draw_shader)
774 vpv->base.driver_shader = draw_create_vertex_shader(st->draw, &state);
775 else
776 vpv->base.driver_shader = pipe->create_vs_state(pipe, &state);
777
778 if (state.tokens) {
779 tgsi_free_tokens(state.tokens);
780 }
781
782 return vpv;
783 }
784
785
786 /**
787 * Find/create a vertex program variant.
788 */
789 struct st_common_variant *
790 st_get_vp_variant(struct st_context *st,
791 struct st_program *stp,
792 const struct st_common_variant_key *key)
793 {
794 struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
795 struct st_common_variant *vpv;
796
797 /* Search for existing variant */
798 for (vpv = st_common_variant(stp->variants); vpv;
799 vpv = st_common_variant(vpv->base.next)) {
800 if (memcmp(&vpv->key, key, sizeof(*key)) == 0) {
801 break;
802 }
803 }
804
805 if (!vpv) {
806 /* create now */
807 vpv = st_create_vp_variant(st, stp, key);
808 if (vpv) {
809 vpv->base.st = key->st;
810
811 unsigned num_inputs = stvp->num_inputs + key->passthrough_edgeflags;
812 for (unsigned index = 0; index < num_inputs; ++index) {
813 unsigned attr = stvp->index_to_input[index];
814 if (attr == ST_DOUBLE_ATTRIB_PLACEHOLDER)
815 continue;
816 vpv->vert_attrib_mask |= 1u << attr;
817 }
818
819 /* insert into list */
820 vpv->base.next = stp->variants;
821 stp->variants = &vpv->base;
822 }
823 }
824
825 return vpv;
826 }
827
828
829 /**
830 * Translate a Mesa fragment shader into a TGSI shader.
831 */
832 bool
833 st_translate_fragment_program(struct st_context *st,
834 struct st_program *stfp)
835 {
836 /* Non-GLSL programs: */
837 if (!stfp->glsl_to_tgsi) {
838 _mesa_remove_output_reads(&stfp->Base, PROGRAM_OUTPUT);
839 if (st->ctx->Const.GLSLFragCoordIsSysVal)
840 _mesa_program_fragment_position_to_sysval(&stfp->Base);
841
842 /* This determines which states will be updated when the assembly
843 * shader is bound.
844 *
845 * fragment.position and glDrawPixels always use constants.
846 */
847 stfp->affected_states = ST_NEW_FS_STATE |
848 ST_NEW_SAMPLE_SHADING |
849 ST_NEW_FS_CONSTANTS;
850
851 if (stfp->ati_fs) {
852 /* Just set them for ATI_fs unconditionally. */
853 stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
854 ST_NEW_FS_SAMPLERS;
855 } else {
856 /* ARB_fp */
857 if (stfp->Base.SamplersUsed)
858 stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
859 ST_NEW_FS_SAMPLERS;
860 }
861
862 /* Translate to NIR. */
863 if (!stfp->ati_fs &&
864 st->pipe->screen->get_shader_param(st->pipe->screen,
865 PIPE_SHADER_FRAGMENT,
866 PIPE_SHADER_CAP_PREFERRED_IR)) {
867 nir_shader *nir =
868 st_translate_prog_to_nir(st, &stfp->Base, MESA_SHADER_FRAGMENT);
869
870 if (stfp->Base.nir)
871 ralloc_free(stfp->Base.nir);
872 if (stfp->serialized_nir) {
873 free(stfp->serialized_nir);
874 stfp->serialized_nir = NULL;
875 }
876 stfp->state.type = PIPE_SHADER_IR_NIR;
877 stfp->Base.nir = nir;
878 return true;
879 }
880 }
881
882 ubyte outputMapping[2 * FRAG_RESULT_MAX];
883 ubyte inputMapping[VARYING_SLOT_MAX];
884 ubyte inputSlotToAttr[VARYING_SLOT_MAX];
885 ubyte interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
886 GLuint attr;
887 GLbitfield64 inputsRead;
888 struct ureg_program *ureg;
889
890 GLboolean write_all = GL_FALSE;
891
892 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
893 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
894 uint fs_num_inputs = 0;
895
896 ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
897 ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
898 uint fs_num_outputs = 0;
899
900 memset(inputSlotToAttr, ~0, sizeof(inputSlotToAttr));
901
902 /*
903 * Convert Mesa program inputs to TGSI input register semantics.
904 */
905 inputsRead = stfp->Base.info.inputs_read;
906 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
907 if ((inputsRead & BITFIELD64_BIT(attr)) != 0) {
908 const GLuint slot = fs_num_inputs++;
909
910 inputMapping[attr] = slot;
911 inputSlotToAttr[slot] = attr;
912
913 switch (attr) {
914 case VARYING_SLOT_POS:
915 input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
916 input_semantic_index[slot] = 0;
917 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
918 break;
919 case VARYING_SLOT_COL0:
920 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
921 input_semantic_index[slot] = 0;
922 interpMode[slot] = stfp->glsl_to_tgsi ?
923 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
924 break;
925 case VARYING_SLOT_COL1:
926 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
927 input_semantic_index[slot] = 1;
928 interpMode[slot] = stfp->glsl_to_tgsi ?
929 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
930 break;
931 case VARYING_SLOT_FOGC:
932 input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
933 input_semantic_index[slot] = 0;
934 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
935 break;
936 case VARYING_SLOT_FACE:
937 input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
938 input_semantic_index[slot] = 0;
939 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
940 break;
941 case VARYING_SLOT_PRIMITIVE_ID:
942 input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
943 input_semantic_index[slot] = 0;
944 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
945 break;
946 case VARYING_SLOT_LAYER:
947 input_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
948 input_semantic_index[slot] = 0;
949 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
950 break;
951 case VARYING_SLOT_VIEWPORT:
952 input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
953 input_semantic_index[slot] = 0;
954 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
955 break;
956 case VARYING_SLOT_CLIP_DIST0:
957 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
958 input_semantic_index[slot] = 0;
959 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
960 break;
961 case VARYING_SLOT_CLIP_DIST1:
962 input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
963 input_semantic_index[slot] = 1;
964 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
965 break;
966 case VARYING_SLOT_CULL_DIST0:
967 case VARYING_SLOT_CULL_DIST1:
968 /* these should have been lowered by GLSL */
969 assert(0);
970 break;
971 /* In most cases, there is nothing special about these
972 * inputs, so adopt a convention to use the generic
973 * semantic name and the mesa VARYING_SLOT_ number as the
974 * index.
975 *
976 * All that is required is that the vertex shader labels
977 * its own outputs similarly, and that the vertex shader
978 * generates at least every output required by the
979 * fragment shader plus fixed-function hardware (such as
980 * BFC).
981 *
982 * However, some drivers may need us to identify the PNTC and TEXi
983 * varyings if, for example, their capability to replace them with
984 * sprite coordinates is limited.
985 */
986 case VARYING_SLOT_PNTC:
987 if (st->needs_texcoord_semantic) {
988 input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
989 input_semantic_index[slot] = 0;
990 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
991 break;
992 }
993 /* fall through */
994 case VARYING_SLOT_TEX0:
995 case VARYING_SLOT_TEX1:
996 case VARYING_SLOT_TEX2:
997 case VARYING_SLOT_TEX3:
998 case VARYING_SLOT_TEX4:
999 case VARYING_SLOT_TEX5:
1000 case VARYING_SLOT_TEX6:
1001 case VARYING_SLOT_TEX7:
1002 if (st->needs_texcoord_semantic) {
1003 input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
1004 input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
1005 interpMode[slot] = stfp->glsl_to_tgsi ?
1006 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
1007 break;
1008 }
1009 /* fall through */
1010 case VARYING_SLOT_VAR0:
1011 default:
1012 /* Semantic indices should be zero-based because drivers may choose
1013 * to assign a fixed slot determined by that index.
1014 * This is useful because ARB_separate_shader_objects uses location
1015 * qualifiers for linkage, and if the semantic index corresponds to
1016 * these locations, linkage passes in the driver become unecessary.
1017 *
1018 * If needs_texcoord_semantic is true, no semantic indices will be
1019 * consumed for the TEXi varyings, and we can base the locations of
1020 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
1021 */
1022 assert(attr >= VARYING_SLOT_VAR0 || attr == VARYING_SLOT_PNTC ||
1023 (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
1024 input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
1025 input_semantic_index[slot] = st_get_generic_varying_index(st, attr);
1026 if (attr == VARYING_SLOT_PNTC)
1027 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
1028 else {
1029 interpMode[slot] = stfp->glsl_to_tgsi ?
1030 TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
1031 }
1032 break;
1033 }
1034 }
1035 else {
1036 inputMapping[attr] = -1;
1037 }
1038 }
1039
1040 /*
1041 * Semantics and mapping for outputs
1042 */
1043 GLbitfield64 outputsWritten = stfp->Base.info.outputs_written;
1044
1045 /* if z is written, emit that first */
1046 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
1047 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
1048 fs_output_semantic_index[fs_num_outputs] = 0;
1049 outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
1050 fs_num_outputs++;
1051 outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
1052 }
1053
1054 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
1055 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
1056 fs_output_semantic_index[fs_num_outputs] = 0;
1057 outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
1058 fs_num_outputs++;
1059 outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
1060 }
1061
1062 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) {
1063 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK;
1064 fs_output_semantic_index[fs_num_outputs] = 0;
1065 outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs;
1066 fs_num_outputs++;
1067 outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK);
1068 }
1069
1070 /* handle remaining outputs (color) */
1071 for (attr = 0; attr < ARRAY_SIZE(outputMapping); attr++) {
1072 const GLbitfield64 written = attr < FRAG_RESULT_MAX ? outputsWritten :
1073 stfp->Base.SecondaryOutputsWritten;
1074 const unsigned loc = attr % FRAG_RESULT_MAX;
1075
1076 if (written & BITFIELD64_BIT(loc)) {
1077 switch (loc) {
1078 case FRAG_RESULT_DEPTH:
1079 case FRAG_RESULT_STENCIL:
1080 case FRAG_RESULT_SAMPLE_MASK:
1081 /* handled above */
1082 assert(0);
1083 break;
1084 case FRAG_RESULT_COLOR:
1085 write_all = GL_TRUE; /* fallthrough */
1086 default: {
1087 int index;
1088 assert(loc == FRAG_RESULT_COLOR ||
1089 (FRAG_RESULT_DATA0 <= loc && loc < FRAG_RESULT_MAX));
1090
1091 index = (loc == FRAG_RESULT_COLOR) ? 0 : (loc - FRAG_RESULT_DATA0);
1092
1093 if (attr >= FRAG_RESULT_MAX) {
1094 /* Secondary color for dual source blending. */
1095 assert(index == 0);
1096 index++;
1097 }
1098
1099 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
1100 fs_output_semantic_index[fs_num_outputs] = index;
1101 outputMapping[attr] = fs_num_outputs;
1102 break;
1103 }
1104 }
1105
1106 fs_num_outputs++;
1107 }
1108 }
1109
1110 ureg = ureg_create_with_screen(PIPE_SHADER_FRAGMENT, st->pipe->screen);
1111 if (ureg == NULL)
1112 return false;
1113
1114 if (ST_DEBUG & DEBUG_MESA) {
1115 _mesa_print_program(&stfp->Base);
1116 _mesa_print_program_parameters(st->ctx, &stfp->Base);
1117 debug_printf("\n");
1118 }
1119 if (write_all == GL_TRUE)
1120 ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
1121
1122 if (stfp->Base.info.fs.depth_layout != FRAG_DEPTH_LAYOUT_NONE) {
1123 switch (stfp->Base.info.fs.depth_layout) {
1124 case FRAG_DEPTH_LAYOUT_ANY:
1125 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1126 TGSI_FS_DEPTH_LAYOUT_ANY);
1127 break;
1128 case FRAG_DEPTH_LAYOUT_GREATER:
1129 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1130 TGSI_FS_DEPTH_LAYOUT_GREATER);
1131 break;
1132 case FRAG_DEPTH_LAYOUT_LESS:
1133 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1134 TGSI_FS_DEPTH_LAYOUT_LESS);
1135 break;
1136 case FRAG_DEPTH_LAYOUT_UNCHANGED:
1137 ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1138 TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
1139 break;
1140 default:
1141 assert(0);
1142 }
1143 }
1144
1145 if (stfp->glsl_to_tgsi) {
1146 st_translate_program(st->ctx,
1147 PIPE_SHADER_FRAGMENT,
1148 ureg,
1149 stfp->glsl_to_tgsi,
1150 &stfp->Base,
1151 /* inputs */
1152 fs_num_inputs,
1153 inputMapping,
1154 inputSlotToAttr,
1155 input_semantic_name,
1156 input_semantic_index,
1157 interpMode,
1158 /* outputs */
1159 fs_num_outputs,
1160 outputMapping,
1161 fs_output_semantic_name,
1162 fs_output_semantic_index);
1163
1164 free_glsl_to_tgsi_visitor(stfp->glsl_to_tgsi);
1165 } else if (stfp->ati_fs)
1166 st_translate_atifs_program(ureg,
1167 stfp->ati_fs,
1168 &stfp->Base,
1169 /* inputs */
1170 fs_num_inputs,
1171 inputMapping,
1172 input_semantic_name,
1173 input_semantic_index,
1174 interpMode,
1175 /* outputs */
1176 fs_num_outputs,
1177 outputMapping,
1178 fs_output_semantic_name,
1179 fs_output_semantic_index);
1180 else
1181 st_translate_mesa_program(st->ctx,
1182 PIPE_SHADER_FRAGMENT,
1183 ureg,
1184 &stfp->Base,
1185 /* inputs */
1186 fs_num_inputs,
1187 inputMapping,
1188 input_semantic_name,
1189 input_semantic_index,
1190 interpMode,
1191 /* outputs */
1192 fs_num_outputs,
1193 outputMapping,
1194 fs_output_semantic_name,
1195 fs_output_semantic_index);
1196
1197 stfp->state.tokens = ureg_get_tokens(ureg, NULL);
1198 ureg_destroy(ureg);
1199
1200 if (stfp->glsl_to_tgsi) {
1201 stfp->glsl_to_tgsi = NULL;
1202 st_store_ir_in_disk_cache(st, &stfp->Base, false);
1203 }
1204
1205 return stfp->state.tokens != NULL;
1206 }
1207
1208 static struct st_fp_variant *
1209 st_create_fp_variant(struct st_context *st,
1210 struct st_program *stfp,
1211 const struct st_fp_variant_key *key)
1212 {
1213 struct pipe_context *pipe = st->pipe;
1214 struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
1215 struct pipe_shader_state state = {0};
1216 struct gl_program_parameter_list *params = stfp->Base.Parameters;
1217 static const gl_state_index16 texcoord_state[STATE_LENGTH] =
1218 { STATE_INTERNAL, STATE_CURRENT_ATTRIB, VERT_ATTRIB_TEX0 };
1219 static const gl_state_index16 scale_state[STATE_LENGTH] =
1220 { STATE_INTERNAL, STATE_PT_SCALE };
1221 static const gl_state_index16 bias_state[STATE_LENGTH] =
1222 { STATE_INTERNAL, STATE_PT_BIAS };
1223 static const gl_state_index16 alpha_ref_state[STATE_LENGTH] =
1224 { STATE_INTERNAL, STATE_ALPHA_REF };
1225
1226 if (!variant)
1227 return NULL;
1228
1229 if (stfp->state.type == PIPE_SHADER_IR_NIR) {
1230 bool finalize = false;
1231
1232 state.type = PIPE_SHADER_IR_NIR;
1233 state.ir.nir = get_nir_shader(st, stfp);
1234
1235 if (key->clamp_color) {
1236 NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
1237 finalize = true;
1238 }
1239
1240 if (key->lower_flatshade) {
1241 NIR_PASS_V(state.ir.nir, nir_lower_flatshade);
1242 finalize = true;
1243 }
1244
1245 if (key->lower_alpha_func != COMPARE_FUNC_NEVER) {
1246 _mesa_add_state_reference(params, alpha_ref_state);
1247 NIR_PASS_V(state.ir.nir, nir_lower_alpha_test, key->lower_alpha_func,
1248 false, alpha_ref_state);
1249 finalize = true;
1250 }
1251
1252 if (key->lower_two_sided_color) {
1253 NIR_PASS_V(state.ir.nir, nir_lower_two_sided_color);
1254 finalize = true;
1255 }
1256
1257 if (key->persample_shading) {
1258 nir_shader *shader = state.ir.nir;
1259 nir_foreach_variable(var, &shader->inputs)
1260 var->data.sample = true;
1261 finalize = true;
1262 }
1263
1264 assert(!(key->bitmap && key->drawpixels));
1265
1266 /* glBitmap */
1267 if (key->bitmap) {
1268 nir_lower_bitmap_options options = {0};
1269
1270 variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1271 options.sampler = variant->bitmap_sampler;
1272 options.swizzle_xxxx = st->bitmap.tex_format == PIPE_FORMAT_R8_UNORM;
1273
1274 NIR_PASS_V(state.ir.nir, nir_lower_bitmap, &options);
1275 finalize = true;
1276 }
1277
1278 /* glDrawPixels (color only) */
1279 if (key->drawpixels) {
1280 nir_lower_drawpixels_options options = {{0}};
1281 unsigned samplers_used = stfp->Base.SamplersUsed;
1282
1283 /* Find the first unused slot. */
1284 variant->drawpix_sampler = ffs(~samplers_used) - 1;
1285 options.drawpix_sampler = variant->drawpix_sampler;
1286 samplers_used |= (1 << variant->drawpix_sampler);
1287
1288 options.pixel_maps = key->pixelMaps;
1289 if (key->pixelMaps) {
1290 variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1291 options.pixelmap_sampler = variant->pixelmap_sampler;
1292 }
1293
1294 options.scale_and_bias = key->scaleAndBias;
1295 if (key->scaleAndBias) {
1296 _mesa_add_state_reference(params, scale_state);
1297 memcpy(options.scale_state_tokens, scale_state,
1298 sizeof(options.scale_state_tokens));
1299 _mesa_add_state_reference(params, bias_state);
1300 memcpy(options.bias_state_tokens, bias_state,
1301 sizeof(options.bias_state_tokens));
1302 }
1303
1304 _mesa_add_state_reference(params, texcoord_state);
1305 memcpy(options.texcoord_state_tokens, texcoord_state,
1306 sizeof(options.texcoord_state_tokens));
1307
1308 NIR_PASS_V(state.ir.nir, nir_lower_drawpixels, &options);
1309 finalize = true;
1310 }
1311
1312 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1313 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv ||
1314 key->external.lower_ayuv || key->external.lower_xyuv)) {
1315
1316 st_nir_lower_samplers(pipe->screen, state.ir.nir,
1317 stfp->shader_program, &stfp->Base);
1318
1319 nir_lower_tex_options options = {0};
1320 options.lower_y_uv_external = key->external.lower_nv12;
1321 options.lower_y_u_v_external = key->external.lower_iyuv;
1322 options.lower_xy_uxvx_external = key->external.lower_xy_uxvx;
1323 options.lower_yx_xuxv_external = key->external.lower_yx_xuxv;
1324 options.lower_ayuv_external = key->external.lower_ayuv;
1325 options.lower_xyuv_external = key->external.lower_xyuv;
1326 NIR_PASS_V(state.ir.nir, nir_lower_tex, &options);
1327 finalize = true;
1328 }
1329
1330 if (finalize || !st->allow_st_finalize_nir_twice) {
1331 st_finalize_nir(st, &stfp->Base, stfp->shader_program, state.ir.nir,
1332 false);
1333 }
1334
1335 /* This pass needs to happen *after* nir_lower_sampler */
1336 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1337 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv ||
1338 key->external.lower_ayuv || key->external.lower_xyuv)) {
1339 NIR_PASS_V(state.ir.nir, st_nir_lower_tex_src_plane,
1340 ~stfp->Base.SamplersUsed,
1341 key->external.lower_nv12 || key->external.lower_xy_uxvx ||
1342 key->external.lower_yx_xuxv,
1343 key->external.lower_iyuv);
1344 finalize = true;
1345 }
1346
1347 if (finalize || !st->allow_st_finalize_nir_twice) {
1348 /* Some of the lowering above may have introduced new varyings */
1349 nir_shader_gather_info(state.ir.nir,
1350 nir_shader_get_entrypoint(state.ir.nir));
1351
1352 struct pipe_screen *screen = pipe->screen;
1353 if (screen->finalize_nir)
1354 screen->finalize_nir(screen, state.ir.nir, false);
1355 }
1356
1357 if (ST_DEBUG & DEBUG_PRINT_IR)
1358 nir_print_shader(state.ir.nir, stderr);
1359
1360 variant->base.driver_shader = pipe->create_fs_state(pipe, &state);
1361 variant->key = *key;
1362
1363 return variant;
1364 }
1365
1366 state.tokens = stfp->state.tokens;
1367
1368 assert(!(key->bitmap && key->drawpixels));
1369
1370 /* Fix texture targets and add fog for ATI_fs */
1371 if (stfp->ati_fs) {
1372 const struct tgsi_token *tokens = st_fixup_atifs(state.tokens, key);
1373
1374 if (tokens)
1375 state.tokens = tokens;
1376 else
1377 fprintf(stderr, "mesa: cannot post-process ATI_fs\n");
1378 }
1379
1380 /* Emulate features. */
1381 if (key->clamp_color || key->persample_shading) {
1382 const struct tgsi_token *tokens;
1383 unsigned flags =
1384 (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
1385 (key->persample_shading ? TGSI_EMU_FORCE_PERSAMPLE_INTERP : 0);
1386
1387 tokens = tgsi_emulate(state.tokens, flags);
1388
1389 if (tokens) {
1390 if (state.tokens != stfp->state.tokens)
1391 tgsi_free_tokens(state.tokens);
1392 state.tokens = tokens;
1393 } else
1394 fprintf(stderr, "mesa: cannot emulate deprecated features\n");
1395 }
1396
1397 /* glBitmap */
1398 if (key->bitmap) {
1399 const struct tgsi_token *tokens;
1400
1401 variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1402
1403 tokens = st_get_bitmap_shader(state.tokens,
1404 st->internal_target,
1405 variant->bitmap_sampler,
1406 st->needs_texcoord_semantic,
1407 st->bitmap.tex_format ==
1408 PIPE_FORMAT_R8_UNORM);
1409
1410 if (tokens) {
1411 if (state.tokens != stfp->state.tokens)
1412 tgsi_free_tokens(state.tokens);
1413 state.tokens = tokens;
1414 } else
1415 fprintf(stderr, "mesa: cannot create a shader for glBitmap\n");
1416 }
1417
1418 /* glDrawPixels (color only) */
1419 if (key->drawpixels) {
1420 const struct tgsi_token *tokens;
1421 unsigned scale_const = 0, bias_const = 0, texcoord_const = 0;
1422
1423 /* Find the first unused slot. */
1424 variant->drawpix_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1425
1426 if (key->pixelMaps) {
1427 unsigned samplers_used = stfp->Base.SamplersUsed |
1428 (1 << variant->drawpix_sampler);
1429
1430 variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1431 }
1432
1433 if (key->scaleAndBias) {
1434 scale_const = _mesa_add_state_reference(params, scale_state);
1435 bias_const = _mesa_add_state_reference(params, bias_state);
1436 }
1437
1438 texcoord_const = _mesa_add_state_reference(params, texcoord_state);
1439
1440 tokens = st_get_drawpix_shader(state.tokens,
1441 st->needs_texcoord_semantic,
1442 key->scaleAndBias, scale_const,
1443 bias_const, key->pixelMaps,
1444 variant->drawpix_sampler,
1445 variant->pixelmap_sampler,
1446 texcoord_const, st->internal_target);
1447
1448 if (tokens) {
1449 if (state.tokens != stfp->state.tokens)
1450 tgsi_free_tokens(state.tokens);
1451 state.tokens = tokens;
1452 } else
1453 fprintf(stderr, "mesa: cannot create a shader for glDrawPixels\n");
1454 }
1455
1456 if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1457 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1458 const struct tgsi_token *tokens;
1459
1460 /* samplers inserted would conflict, but this should be unpossible: */
1461 assert(!(key->bitmap || key->drawpixels));
1462
1463 tokens = st_tgsi_lower_yuv(state.tokens,
1464 ~stfp->Base.SamplersUsed,
1465 key->external.lower_nv12 ||
1466 key->external.lower_xy_uxvx ||
1467 key->external.lower_yx_xuxv,
1468 key->external.lower_iyuv);
1469 if (tokens) {
1470 if (state.tokens != stfp->state.tokens)
1471 tgsi_free_tokens(state.tokens);
1472 state.tokens = tokens;
1473 } else {
1474 fprintf(stderr, "mesa: cannot create a shader for samplerExternalOES\n");
1475 }
1476 }
1477
1478 if (key->lower_depth_clamp) {
1479 unsigned depth_range_const = _mesa_add_state_reference(params, depth_range_state);
1480
1481 const struct tgsi_token *tokens;
1482 tokens = st_tgsi_lower_depth_clamp_fs(state.tokens, depth_range_const);
1483 if (state.tokens != stfp->state.tokens)
1484 tgsi_free_tokens(state.tokens);
1485 state.tokens = tokens;
1486 }
1487
1488 if (ST_DEBUG & DEBUG_PRINT_IR)
1489 tgsi_dump(state.tokens, 0);
1490
1491 /* fill in variant */
1492 variant->base.driver_shader = pipe->create_fs_state(pipe, &state);
1493 variant->key = *key;
1494
1495 if (state.tokens != stfp->state.tokens)
1496 tgsi_free_tokens(state.tokens);
1497 return variant;
1498 }
1499
1500 /**
1501 * Translate fragment program if needed.
1502 */
1503 struct st_fp_variant *
1504 st_get_fp_variant(struct st_context *st,
1505 struct st_program *stfp,
1506 const struct st_fp_variant_key *key)
1507 {
1508 struct st_fp_variant *fpv;
1509
1510 /* Search for existing variant */
1511 for (fpv = st_fp_variant(stfp->variants); fpv;
1512 fpv = st_fp_variant(fpv->base.next)) {
1513 if (memcmp(&fpv->key, key, sizeof(*key)) == 0) {
1514 break;
1515 }
1516 }
1517
1518 if (!fpv) {
1519 /* create new */
1520 fpv = st_create_fp_variant(st, stfp, key);
1521 if (fpv) {
1522 fpv->base.st = key->st;
1523
1524 if (key->bitmap || key->drawpixels) {
1525 /* Regular variants should always come before the
1526 * bitmap & drawpixels variants, (unless there
1527 * are no regular variants) so that
1528 * st_update_fp can take a fast path when
1529 * shader_has_one_variant is set.
1530 */
1531 if (!stfp->variants) {
1532 stfp->variants = &fpv->base;
1533 } else {
1534 /* insert into list after the first one */
1535 fpv->base.next = stfp->variants->next;
1536 stfp->variants->next = &fpv->base;
1537 }
1538 } else {
1539 /* insert into list */
1540 fpv->base.next = stfp->variants;
1541 stfp->variants = &fpv->base;
1542 }
1543 }
1544 }
1545
1546 return fpv;
1547 }
1548
1549 /**
1550 * Translate a program. This is common code for geometry and tessellation
1551 * shaders.
1552 */
1553 bool
1554 st_translate_common_program(struct st_context *st,
1555 struct st_program *stp)
1556 {
1557 struct gl_program *prog = &stp->Base;
1558 enum pipe_shader_type stage =
1559 pipe_shader_type_from_mesa(stp->Base.info.stage);
1560 struct ureg_program *ureg = ureg_create_with_screen(stage, st->pipe->screen);
1561
1562 if (ureg == NULL)
1563 return false;
1564
1565 switch (stage) {
1566 case PIPE_SHADER_TESS_CTRL:
1567 ureg_property(ureg, TGSI_PROPERTY_TCS_VERTICES_OUT,
1568 stp->Base.info.tess.tcs_vertices_out);
1569 break;
1570
1571 case PIPE_SHADER_TESS_EVAL:
1572 if (stp->Base.info.tess.primitive_mode == GL_ISOLINES)
1573 ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE, GL_LINES);
1574 else
1575 ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE,
1576 stp->Base.info.tess.primitive_mode);
1577
1578 STATIC_ASSERT((TESS_SPACING_EQUAL + 1) % 3 == PIPE_TESS_SPACING_EQUAL);
1579 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD + 1) % 3 ==
1580 PIPE_TESS_SPACING_FRACTIONAL_ODD);
1581 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN + 1) % 3 ==
1582 PIPE_TESS_SPACING_FRACTIONAL_EVEN);
1583
1584 ureg_property(ureg, TGSI_PROPERTY_TES_SPACING,
1585 (stp->Base.info.tess.spacing + 1) % 3);
1586
1587 ureg_property(ureg, TGSI_PROPERTY_TES_VERTEX_ORDER_CW,
1588 !stp->Base.info.tess.ccw);
1589 ureg_property(ureg, TGSI_PROPERTY_TES_POINT_MODE,
1590 stp->Base.info.tess.point_mode);
1591 break;
1592
1593 case PIPE_SHADER_GEOMETRY:
1594 ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM,
1595 stp->Base.info.gs.input_primitive);
1596 ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM,
1597 stp->Base.info.gs.output_primitive);
1598 ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
1599 stp->Base.info.gs.vertices_out);
1600 ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS,
1601 stp->Base.info.gs.invocations);
1602 break;
1603
1604 default:
1605 break;
1606 }
1607
1608 ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
1609 ubyte inputMapping[VARYING_SLOT_TESS_MAX];
1610 ubyte outputMapping[VARYING_SLOT_TESS_MAX];
1611 GLuint attr;
1612
1613 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
1614 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
1615 uint num_inputs = 0;
1616
1617 ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
1618 ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
1619 uint num_outputs = 0;
1620
1621 GLint i;
1622
1623 memset(inputSlotToAttr, 0, sizeof(inputSlotToAttr));
1624 memset(inputMapping, 0, sizeof(inputMapping));
1625 memset(outputMapping, 0, sizeof(outputMapping));
1626 memset(&stp->state, 0, sizeof(stp->state));
1627
1628 if (prog->info.clip_distance_array_size)
1629 ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
1630 prog->info.clip_distance_array_size);
1631 if (prog->info.cull_distance_array_size)
1632 ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
1633 prog->info.cull_distance_array_size);
1634
1635 /*
1636 * Convert Mesa program inputs to TGSI input register semantics.
1637 */
1638 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1639 if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) == 0)
1640 continue;
1641
1642 unsigned slot = num_inputs++;
1643
1644 inputMapping[attr] = slot;
1645 inputSlotToAttr[slot] = attr;
1646
1647 unsigned semantic_name, semantic_index;
1648 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1649 &semantic_name, &semantic_index);
1650 input_semantic_name[slot] = semantic_name;
1651 input_semantic_index[slot] = semantic_index;
1652 }
1653
1654 /* Also add patch inputs. */
1655 for (attr = 0; attr < 32; attr++) {
1656 if (prog->info.patch_inputs_read & (1u << attr)) {
1657 GLuint slot = num_inputs++;
1658 GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1659
1660 inputMapping[patch_attr] = slot;
1661 inputSlotToAttr[slot] = patch_attr;
1662 input_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1663 input_semantic_index[slot] = attr;
1664 }
1665 }
1666
1667 /* initialize output semantics to defaults */
1668 for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
1669 output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
1670 output_semantic_index[i] = 0;
1671 }
1672
1673 /*
1674 * Determine number of outputs, the (default) output register
1675 * mapping and the semantic information for each output.
1676 */
1677 for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1678 if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
1679 GLuint slot = num_outputs++;
1680
1681 outputMapping[attr] = slot;
1682
1683 unsigned semantic_name, semantic_index;
1684 tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1685 &semantic_name, &semantic_index);
1686 output_semantic_name[slot] = semantic_name;
1687 output_semantic_index[slot] = semantic_index;
1688 }
1689 }
1690
1691 /* Also add patch outputs. */
1692 for (attr = 0; attr < 32; attr++) {
1693 if (prog->info.patch_outputs_written & (1u << attr)) {
1694 GLuint slot = num_outputs++;
1695 GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1696
1697 outputMapping[patch_attr] = slot;
1698 output_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1699 output_semantic_index[slot] = attr;
1700 }
1701 }
1702
1703 st_translate_program(st->ctx,
1704 stage,
1705 ureg,
1706 stp->glsl_to_tgsi,
1707 prog,
1708 /* inputs */
1709 num_inputs,
1710 inputMapping,
1711 inputSlotToAttr,
1712 input_semantic_name,
1713 input_semantic_index,
1714 NULL,
1715 /* outputs */
1716 num_outputs,
1717 outputMapping,
1718 output_semantic_name,
1719 output_semantic_index);
1720
1721 stp->state.tokens = ureg_get_tokens(ureg, NULL);
1722
1723 ureg_destroy(ureg);
1724
1725 st_translate_stream_output_info(prog);
1726
1727 st_store_ir_in_disk_cache(st, prog, false);
1728
1729 if (ST_DEBUG & DEBUG_PRINT_IR && ST_DEBUG & DEBUG_MESA)
1730 _mesa_print_program(prog);
1731
1732 free_glsl_to_tgsi_visitor(stp->glsl_to_tgsi);
1733 stp->glsl_to_tgsi = NULL;
1734 return true;
1735 }
1736
1737
1738 /**
1739 * Get/create a basic program variant.
1740 */
1741 struct st_variant *
1742 st_get_common_variant(struct st_context *st,
1743 struct st_program *prog,
1744 const struct st_common_variant_key *key)
1745 {
1746 struct pipe_context *pipe = st->pipe;
1747 struct st_variant *v;
1748 struct pipe_shader_state state = {0};
1749
1750 /* Search for existing variant */
1751 for (v = prog->variants; v; v = v->next) {
1752 if (memcmp(&st_common_variant(v)->key, key, sizeof(*key)) == 0)
1753 break;
1754 }
1755
1756 if (!v) {
1757 /* create new */
1758 v = (struct st_variant*)CALLOC_STRUCT(st_common_variant);
1759 if (v) {
1760 if (prog->state.type == PIPE_SHADER_IR_NIR) {
1761 bool finalize = false;
1762
1763 state.type = PIPE_SHADER_IR_NIR;
1764 state.ir.nir = get_nir_shader(st, prog);
1765
1766 if (key->clamp_color) {
1767 NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
1768 finalize = true;
1769 }
1770
1771 state.stream_output = prog->state.stream_output;
1772
1773 if (finalize || !st->allow_st_finalize_nir_twice) {
1774 st_finalize_nir(st, &prog->Base, prog->shader_program,
1775 state.ir.nir, true);
1776 }
1777
1778 if (ST_DEBUG & DEBUG_PRINT_IR)
1779 nir_print_shader(state.ir.nir, stderr);
1780 } else {
1781 if (key->lower_depth_clamp) {
1782 struct gl_program_parameter_list *params = prog->Base.Parameters;
1783
1784 unsigned depth_range_const =
1785 _mesa_add_state_reference(params, depth_range_state);
1786
1787 const struct tgsi_token *tokens;
1788 tokens =
1789 st_tgsi_lower_depth_clamp(prog->state.tokens,
1790 depth_range_const,
1791 key->clip_negative_one_to_one);
1792
1793 if (tokens != prog->state.tokens)
1794 tgsi_free_tokens(prog->state.tokens);
1795
1796 prog->state.tokens = tokens;
1797 }
1798 state = prog->state;
1799
1800 if (ST_DEBUG & DEBUG_PRINT_IR)
1801 tgsi_dump(state.tokens, 0);
1802 }
1803 /* fill in new variant */
1804 switch (prog->Base.info.stage) {
1805 case MESA_SHADER_TESS_CTRL:
1806 v->driver_shader = pipe->create_tcs_state(pipe, &state);
1807 break;
1808 case MESA_SHADER_TESS_EVAL:
1809 v->driver_shader = pipe->create_tes_state(pipe, &state);
1810 break;
1811 case MESA_SHADER_GEOMETRY:
1812 v->driver_shader = pipe->create_gs_state(pipe, &state);
1813 break;
1814 case MESA_SHADER_COMPUTE: {
1815 struct pipe_compute_state cs = {0};
1816 cs.ir_type = state.type;
1817 cs.req_local_mem = prog->Base.info.cs.shared_size;
1818
1819 if (state.type == PIPE_SHADER_IR_NIR)
1820 cs.prog = state.ir.nir;
1821 else
1822 cs.prog = state.tokens;
1823
1824 v->driver_shader = pipe->create_compute_state(pipe, &cs);
1825 break;
1826 }
1827 default:
1828 assert(!"unhandled shader type");
1829 free(v);
1830 return NULL;
1831 }
1832
1833 st_common_variant(v)->key = *key;
1834 v->st = key->st;
1835
1836 /* insert into list */
1837 v->next = prog->variants;
1838 prog->variants = v;
1839 }
1840 }
1841
1842 return v;
1843 }
1844
1845
1846 /**
1847 * Vert/Geom/Frag programs have per-context variants. Free all the
1848 * variants attached to the given program which match the given context.
1849 */
1850 static void
1851 destroy_program_variants(struct st_context *st, struct gl_program *target)
1852 {
1853 if (!target || target == &_mesa_DummyProgram)
1854 return;
1855
1856 struct st_program *p = st_program(target);
1857 struct st_variant *v, **prevPtr = &p->variants;
1858 bool unbound = false;
1859
1860 for (v = p->variants; v; ) {
1861 struct st_variant *next = v->next;
1862 if (v->st == st) {
1863 if (!unbound) {
1864 st_unbind_program(st, p);
1865 unbound = true;
1866 }
1867
1868 /* unlink from list */
1869 *prevPtr = next;
1870 /* destroy this variant */
1871 delete_variant(st, v, target->Target);
1872 }
1873 else {
1874 prevPtr = &v->next;
1875 }
1876 v = next;
1877 }
1878 }
1879
1880
1881 /**
1882 * Callback for _mesa_HashWalk. Free all the shader's program variants
1883 * which match the given context.
1884 */
1885 static void
1886 destroy_shader_program_variants_cb(GLuint key, void *data, void *userData)
1887 {
1888 struct st_context *st = (struct st_context *) userData;
1889 struct gl_shader *shader = (struct gl_shader *) data;
1890
1891 switch (shader->Type) {
1892 case GL_SHADER_PROGRAM_MESA:
1893 {
1894 struct gl_shader_program *shProg = (struct gl_shader_program *) data;
1895 GLuint i;
1896
1897 for (i = 0; i < ARRAY_SIZE(shProg->_LinkedShaders); i++) {
1898 if (shProg->_LinkedShaders[i])
1899 destroy_program_variants(st, shProg->_LinkedShaders[i]->Program);
1900 }
1901 }
1902 break;
1903 case GL_VERTEX_SHADER:
1904 case GL_FRAGMENT_SHADER:
1905 case GL_GEOMETRY_SHADER:
1906 case GL_TESS_CONTROL_SHADER:
1907 case GL_TESS_EVALUATION_SHADER:
1908 case GL_COMPUTE_SHADER:
1909 break;
1910 default:
1911 assert(0);
1912 }
1913 }
1914
1915
1916 /**
1917 * Callback for _mesa_HashWalk. Free all the program variants which match
1918 * the given context.
1919 */
1920 static void
1921 destroy_program_variants_cb(GLuint key, void *data, void *userData)
1922 {
1923 struct st_context *st = (struct st_context *) userData;
1924 struct gl_program *program = (struct gl_program *) data;
1925 destroy_program_variants(st, program);
1926 }
1927
1928
1929 /**
1930 * Walk over all shaders and programs to delete any variants which
1931 * belong to the given context.
1932 * This is called during context tear-down.
1933 */
1934 void
1935 st_destroy_program_variants(struct st_context *st)
1936 {
1937 /* If shaders can be shared with other contexts, the last context will
1938 * call DeleteProgram on all shaders, releasing everything.
1939 */
1940 if (st->has_shareable_shaders)
1941 return;
1942
1943 /* ARB vert/frag program */
1944 _mesa_HashWalk(st->ctx->Shared->Programs,
1945 destroy_program_variants_cb, st);
1946
1947 /* GLSL vert/frag/geom shaders */
1948 _mesa_HashWalk(st->ctx->Shared->ShaderObjects,
1949 destroy_shader_program_variants_cb, st);
1950 }
1951
1952
1953 /**
1954 * Compile one shader variant.
1955 */
1956 static void
1957 st_precompile_shader_variant(struct st_context *st,
1958 struct gl_program *prog)
1959 {
1960 switch (prog->Target) {
1961 case GL_VERTEX_PROGRAM_ARB: {
1962 struct st_program *p = (struct st_program *)prog;
1963 struct st_common_variant_key key;
1964
1965 memset(&key, 0, sizeof(key));
1966
1967 key.st = st->has_shareable_shaders ? NULL : st;
1968 st_get_vp_variant(st, p, &key);
1969 break;
1970 }
1971
1972 case GL_FRAGMENT_PROGRAM_ARB: {
1973 struct st_program *p = (struct st_program *)prog;
1974 struct st_fp_variant_key key;
1975
1976 memset(&key, 0, sizeof(key));
1977
1978 key.st = st->has_shareable_shaders ? NULL : st;
1979 st_get_fp_variant(st, p, &key);
1980 break;
1981 }
1982
1983 case GL_TESS_CONTROL_PROGRAM_NV:
1984 case GL_TESS_EVALUATION_PROGRAM_NV:
1985 case GL_GEOMETRY_PROGRAM_NV:
1986 case GL_COMPUTE_PROGRAM_NV: {
1987 struct st_program *p = st_program(prog);
1988 struct st_common_variant_key key;
1989
1990 memset(&key, 0, sizeof(key));
1991
1992 key.st = st->has_shareable_shaders ? NULL : st;
1993 st_get_common_variant(st, p, &key);
1994 break;
1995 }
1996
1997 default:
1998 assert(0);
1999 }
2000 }
2001
2002 void
2003 st_serialize_nir(struct st_program *stp)
2004 {
2005 if (!stp->serialized_nir) {
2006 struct blob blob;
2007 size_t size;
2008
2009 blob_init(&blob);
2010 nir_serialize(&blob, stp->Base.nir, false);
2011 blob_finish_get_buffer(&blob, &stp->serialized_nir, &size);
2012 stp->serialized_nir_size = size;
2013 }
2014 }
2015
2016 void
2017 st_finalize_program(struct st_context *st, struct gl_program *prog)
2018 {
2019 if (st->current_program[prog->info.stage] == prog) {
2020 if (prog->info.stage == MESA_SHADER_VERTEX)
2021 st->dirty |= ST_NEW_VERTEX_PROGRAM(st, (struct st_program *)prog);
2022 else
2023 st->dirty |= ((struct st_program *)prog)->affected_states;
2024 }
2025
2026 if (prog->nir) {
2027 nir_sweep(prog->nir);
2028
2029 /* This is only needed for ARB_vp/fp programs and when the disk cache
2030 * is disabled. If the disk cache is enabled, GLSL programs are
2031 * serialized in write_nir_to_cache.
2032 */
2033 st_serialize_nir(st_program(prog));
2034 }
2035
2036 /* Create Gallium shaders now instead of on demand. */
2037 if (ST_DEBUG & DEBUG_PRECOMPILE ||
2038 st->shader_has_one_variant[prog->info.stage])
2039 st_precompile_shader_variant(st, prog);
2040 }