Merge branch 'llvm-cliptest-viewport'
[mesa.git] / src / gallium / auxiliary / draw / draw_context.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
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 TUNGSTEN GRAPHICS 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 /*
29 * Authors:
30 * Keith Whitwell <keith@tungstengraphics.com>
31 */
32
33
34 #include "pipe/p_context.h"
35 #include "util/u_memory.h"
36 #include "util/u_math.h"
37 #include "util/u_cpu_detect.h"
38 #include "draw_context.h"
39 #include "draw_vs.h"
40 #include "draw_gs.h"
41
42 #if HAVE_LLVM
43 #include "gallivm/lp_bld_init.h"
44 #include "draw_llvm.h"
45
46 static boolean
47 draw_get_option_use_llvm(void)
48 {
49 static boolean first = TRUE;
50 static boolean value;
51 if (first) {
52 first = FALSE;
53 value = debug_get_bool_option("DRAW_USE_LLVM", TRUE);
54
55 #ifdef PIPE_ARCH_X86
56 util_cpu_detect();
57 /* require SSE2 due to LLVM PR6960. */
58 if (!util_cpu_caps.has_sse2)
59 value = FALSE;
60 #endif
61 }
62 return value;
63 }
64 #endif
65
66 struct draw_context *draw_create( struct pipe_context *pipe )
67 {
68 struct draw_context *draw = CALLOC_STRUCT( draw_context );
69 if (draw == NULL)
70 goto fail;
71
72 #if HAVE_LLVM
73 if(draw_get_option_use_llvm())
74 {
75 lp_build_init();
76 assert(lp_build_engine);
77 draw->engine = lp_build_engine;
78 draw->llvm = draw_llvm_create(draw);
79 }
80 #endif
81
82 if (!draw_init(draw))
83 goto fail;
84
85 draw->pipe = pipe;
86
87 return draw;
88
89 fail:
90 draw_destroy( draw );
91 return NULL;
92 }
93
94 boolean draw_init(struct draw_context *draw)
95 {
96 /*
97 * Note that several functions compute the clipmask of the predefined
98 * formats with hardcoded formulas instead of using these. So modifications
99 * here must be reflected there too.
100 */
101
102 ASSIGN_4V( draw->plane[0], -1, 0, 0, 1 );
103 ASSIGN_4V( draw->plane[1], 1, 0, 0, 1 );
104 ASSIGN_4V( draw->plane[2], 0, -1, 0, 1 );
105 ASSIGN_4V( draw->plane[3], 0, 1, 0, 1 );
106 ASSIGN_4V( draw->plane[4], 0, 0, 1, 1 ); /* yes these are correct */
107 ASSIGN_4V( draw->plane[5], 0, 0, -1, 1 ); /* mesa's a bit wonky */
108 draw->nr_planes = 6;
109 draw->clip_xy = 1;
110 draw->clip_z = 1;
111
112
113 draw->reduced_prim = ~0; /* != any of PIPE_PRIM_x */
114
115
116 if (!draw_pipeline_init( draw ))
117 return FALSE;
118
119 if (!draw_pt_init( draw ))
120 return FALSE;
121
122 if (!draw_vs_init( draw ))
123 return FALSE;
124
125 if (!draw_gs_init( draw ))
126 return FALSE;
127
128 return TRUE;
129 }
130
131
132 void draw_destroy( struct draw_context *draw )
133 {
134 struct pipe_context *pipe;
135 int i, j;
136
137 if (!draw)
138 return;
139
140 pipe = draw->pipe;
141
142 /* free any rasterizer CSOs that we may have created.
143 */
144 for (i = 0; i < 2; i++) {
145 for (j = 0; j < 2; j++) {
146 if (draw->rasterizer_no_cull[i][j]) {
147 pipe->delete_rasterizer_state(pipe, draw->rasterizer_no_cull[i][j]);
148 }
149 }
150 }
151
152 /* Not so fast -- we're just borrowing this at the moment.
153 *
154 if (draw->render)
155 draw->render->destroy( draw->render );
156 */
157
158 draw_pipeline_destroy( draw );
159 draw_pt_destroy( draw );
160 draw_vs_destroy( draw );
161 draw_gs_destroy( draw );
162 #ifdef HAVE_LLVM
163 if(draw->llvm)
164 draw_llvm_destroy( draw->llvm );
165 #endif
166
167 FREE( draw );
168 }
169
170
171
172 void draw_flush( struct draw_context *draw )
173 {
174 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
175 }
176
177
178 /**
179 * Specify the Minimum Resolvable Depth factor for polygon offset.
180 * This factor potentially depends on the number of Z buffer bits,
181 * the rasterization algorithm and the arithmetic performed on Z
182 * values between vertex shading and rasterization. It will vary
183 * from one driver to another.
184 */
185 void draw_set_mrd(struct draw_context *draw, double mrd)
186 {
187 draw->mrd = mrd;
188 }
189
190
191 static void update_clip_flags( struct draw_context *draw )
192 {
193 draw->clip_xy = !draw->driver.bypass_clip_xy;
194 draw->clip_z = (!draw->driver.bypass_clip_z &&
195 !draw->depth_clamp);
196 draw->clip_user = (draw->nr_planes > 6);
197 }
198
199 /**
200 * Register new primitive rasterization/rendering state.
201 * This causes the drawing pipeline to be rebuilt.
202 */
203 void draw_set_rasterizer_state( struct draw_context *draw,
204 const struct pipe_rasterizer_state *raster,
205 void *rast_handle )
206 {
207 if (!draw->suspend_flushing) {
208 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
209
210 draw->rasterizer = raster;
211 draw->rast_handle = rast_handle;
212
213 }
214 }
215
216 /* With a little more work, llvmpipe will be able to turn this off and
217 * do its own x/y clipping.
218 *
219 * Some hardware can turn off clipping altogether - in particular any
220 * hardware with a TNL unit can do its own clipping, even if it is
221 * relying on the draw module for some other reason.
222 */
223 void draw_set_driver_clipping( struct draw_context *draw,
224 boolean bypass_clip_xy,
225 boolean bypass_clip_z )
226 {
227 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
228
229 draw->driver.bypass_clip_xy = bypass_clip_xy;
230 draw->driver.bypass_clip_z = bypass_clip_z;
231 update_clip_flags(draw);
232 }
233
234
235 /**
236 * Plug in the primitive rendering/rasterization stage (which is the last
237 * stage in the drawing pipeline).
238 * This is provided by the device driver.
239 */
240 void draw_set_rasterize_stage( struct draw_context *draw,
241 struct draw_stage *stage )
242 {
243 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
244
245 draw->pipeline.rasterize = stage;
246 }
247
248
249 /**
250 * Set the draw module's clipping state.
251 */
252 void draw_set_clip_state( struct draw_context *draw,
253 const struct pipe_clip_state *clip )
254 {
255 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
256
257 assert(clip->nr <= PIPE_MAX_CLIP_PLANES);
258 memcpy(&draw->plane[6], clip->ucp, clip->nr * sizeof(clip->ucp[0]));
259 draw->nr_planes = 6 + clip->nr;
260 draw->depth_clamp = clip->depth_clamp;
261
262 update_clip_flags(draw);
263 }
264
265
266 /**
267 * Set the draw module's viewport state.
268 */
269 void draw_set_viewport_state( struct draw_context *draw,
270 const struct pipe_viewport_state *viewport )
271 {
272 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
273 draw->viewport = *viewport; /* struct copy */
274 draw->identity_viewport = (viewport->scale[0] == 1.0f &&
275 viewport->scale[1] == 1.0f &&
276 viewport->scale[2] == 1.0f &&
277 viewport->scale[3] == 1.0f &&
278 viewport->translate[0] == 0.0f &&
279 viewport->translate[1] == 0.0f &&
280 viewport->translate[2] == 0.0f &&
281 viewport->translate[3] == 0.0f);
282
283 draw_vs_set_viewport( draw, viewport );
284 }
285
286
287
288 void
289 draw_set_vertex_buffers(struct draw_context *draw,
290 unsigned count,
291 const struct pipe_vertex_buffer *buffers)
292 {
293 assert(count <= PIPE_MAX_ATTRIBS);
294
295 memcpy(draw->pt.vertex_buffer, buffers, count * sizeof(buffers[0]));
296 draw->pt.nr_vertex_buffers = count;
297 }
298
299
300 void
301 draw_set_vertex_elements(struct draw_context *draw,
302 unsigned count,
303 const struct pipe_vertex_element *elements)
304 {
305 assert(count <= PIPE_MAX_ATTRIBS);
306
307 memcpy(draw->pt.vertex_element, elements, count * sizeof(elements[0]));
308 draw->pt.nr_vertex_elements = count;
309 }
310
311
312 /**
313 * Tell drawing context where to find mapped vertex buffers.
314 */
315 void
316 draw_set_mapped_vertex_buffer(struct draw_context *draw,
317 unsigned attr, const void *buffer)
318 {
319 draw->pt.user.vbuffer[attr] = buffer;
320 }
321
322
323 void
324 draw_set_mapped_constant_buffer(struct draw_context *draw,
325 unsigned shader_type,
326 unsigned slot,
327 const void *buffer,
328 unsigned size )
329 {
330 debug_assert(shader_type == PIPE_SHADER_VERTEX ||
331 shader_type == PIPE_SHADER_GEOMETRY);
332 debug_assert(slot < PIPE_MAX_CONSTANT_BUFFERS);
333
334 switch (shader_type) {
335 case PIPE_SHADER_VERTEX:
336 draw->pt.user.vs_constants[slot] = buffer;
337 draw->pt.user.vs_constants_size[slot] = size;
338 draw->pt.user.planes = (float (*) [12][4]) &(draw->plane[0]);
339 draw_vs_set_constants(draw, slot, buffer, size);
340 break;
341 case PIPE_SHADER_GEOMETRY:
342 draw->pt.user.gs_constants[slot] = buffer;
343 draw->pt.user.gs_constants_size[slot] = size;
344 draw_gs_set_constants(draw, slot, buffer, size);
345 break;
346 default:
347 assert(0 && "invalid shader type in draw_set_mapped_constant_buffer");
348 }
349 }
350
351
352 /**
353 * Tells the draw module to draw points with triangles if their size
354 * is greater than this threshold.
355 */
356 void
357 draw_wide_point_threshold(struct draw_context *draw, float threshold)
358 {
359 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
360 draw->pipeline.wide_point_threshold = threshold;
361 }
362
363
364 /**
365 * Should the draw module handle point->quad conversion for drawing sprites?
366 */
367 void
368 draw_wide_point_sprites(struct draw_context *draw, boolean draw_sprite)
369 {
370 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
371 draw->pipeline.wide_point_sprites = draw_sprite;
372 }
373
374
375 /**
376 * Tells the draw module to draw lines with triangles if their width
377 * is greater than this threshold.
378 */
379 void
380 draw_wide_line_threshold(struct draw_context *draw, float threshold)
381 {
382 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
383 draw->pipeline.wide_line_threshold = threshold;
384 }
385
386
387 /**
388 * Tells the draw module whether or not to implement line stipple.
389 */
390 void
391 draw_enable_line_stipple(struct draw_context *draw, boolean enable)
392 {
393 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
394 draw->pipeline.line_stipple = enable;
395 }
396
397
398 /**
399 * Tells draw module whether to convert points to quads for sprite mode.
400 */
401 void
402 draw_enable_point_sprites(struct draw_context *draw, boolean enable)
403 {
404 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
405 draw->pipeline.point_sprite = enable;
406 }
407
408
409 void
410 draw_set_force_passthrough( struct draw_context *draw, boolean enable )
411 {
412 draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
413 draw->force_passthrough = enable;
414 }
415
416
417
418 /**
419 * Allocate an extra vertex/geometry shader vertex attribute.
420 * This is used by some of the optional draw module stages such
421 * as wide_point which may need to allocate additional generic/texcoord
422 * attributes.
423 */
424 int
425 draw_alloc_extra_vertex_attrib(struct draw_context *draw,
426 uint semantic_name, uint semantic_index)
427 {
428 const int num_outputs = draw_current_shader_outputs(draw);
429 const int n = draw->extra_shader_outputs.num;
430
431 assert(n < Elements(draw->extra_shader_outputs.semantic_name));
432
433 draw->extra_shader_outputs.semantic_name[n] = semantic_name;
434 draw->extra_shader_outputs.semantic_index[n] = semantic_index;
435 draw->extra_shader_outputs.slot[n] = num_outputs + n;
436 draw->extra_shader_outputs.num++;
437
438 return draw->extra_shader_outputs.slot[n];
439 }
440
441
442 /**
443 * Remove all extra vertex attributes that were allocated with
444 * draw_alloc_extra_vertex_attrib().
445 */
446 void
447 draw_remove_extra_vertex_attribs(struct draw_context *draw)
448 {
449 draw->extra_shader_outputs.num = 0;
450 }
451
452
453 /**
454 * Ask the draw module for the location/slot of the given vertex attribute in
455 * a post-transformed vertex.
456 *
457 * With this function, drivers that use the draw module should have no reason
458 * to track the current vertex/geometry shader.
459 *
460 * Note that the draw module may sometimes generate vertices with extra
461 * attributes (such as texcoords for AA lines). The driver can call this
462 * function to find those attributes.
463 *
464 * Zero is returned if the attribute is not found since this is
465 * a don't care / undefined situtation. Returning -1 would be a bit more
466 * work for the drivers.
467 */
468 int
469 draw_find_shader_output(const struct draw_context *draw,
470 uint semantic_name, uint semantic_index)
471 {
472 const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
473 const struct draw_geometry_shader *gs = draw->gs.geometry_shader;
474 uint i;
475 const struct tgsi_shader_info *info = &vs->info;
476
477 if (gs)
478 info = &gs->info;
479
480 for (i = 0; i < info->num_outputs; i++) {
481 if (info->output_semantic_name[i] == semantic_name &&
482 info->output_semantic_index[i] == semantic_index)
483 return i;
484 }
485
486 /* Search the extra vertex attributes */
487 for (i = 0; i < draw->extra_shader_outputs.num; i++) {
488 if (draw->extra_shader_outputs.semantic_name[i] == semantic_name &&
489 draw->extra_shader_outputs.semantic_index[i] == semantic_index) {
490 return draw->extra_shader_outputs.slot[i];
491 }
492 }
493
494 return 0;
495 }
496
497
498 /**
499 * Return total number of the shader outputs. This function is similar to
500 * draw_current_shader_outputs() but this function also counts any extra
501 * vertex/geometry output attributes that may be filled in by some draw
502 * stages (such as AA point, AA line).
503 *
504 * If geometry shader is present, its output will be returned,
505 * if not vertex shader is used.
506 */
507 uint
508 draw_num_shader_outputs(const struct draw_context *draw)
509 {
510 uint count;
511
512 /* If a geometry shader is present, its outputs go to the
513 * driver, else the vertex shader's outputs.
514 */
515 if (draw->gs.geometry_shader)
516 count = draw->gs.geometry_shader->info.num_outputs;
517 else
518 count = draw->vs.vertex_shader->info.num_outputs;
519
520 count += draw->extra_shader_outputs.num;
521
522 return count;
523 }
524
525
526 /**
527 * Provide TGSI sampler objects for vertex/geometry shaders that use
528 * texture fetches.
529 * This might only be used by software drivers for the time being.
530 */
531 void
532 draw_texture_samplers(struct draw_context *draw,
533 uint shader,
534 uint num_samplers,
535 struct tgsi_sampler **samplers)
536 {
537 if (shader == PIPE_SHADER_VERTEX) {
538 draw->vs.num_samplers = num_samplers;
539 draw->vs.samplers = samplers;
540 } else {
541 debug_assert(shader == PIPE_SHADER_GEOMETRY);
542 draw->gs.num_samplers = num_samplers;
543 draw->gs.samplers = samplers;
544 }
545 }
546
547
548
549
550 void draw_set_render( struct draw_context *draw,
551 struct vbuf_render *render )
552 {
553 draw->render = render;
554 }
555
556
557 void
558 draw_set_index_buffer(struct draw_context *draw,
559 const struct pipe_index_buffer *ib)
560 {
561 if (ib)
562 memcpy(&draw->pt.index_buffer, ib, sizeof(draw->pt.index_buffer));
563 else
564 memset(&draw->pt.index_buffer, 0, sizeof(draw->pt.index_buffer));
565 }
566
567
568 /**
569 * Tell drawing context where to find mapped index/element buffer.
570 */
571 void
572 draw_set_mapped_index_buffer(struct draw_context *draw,
573 const void *elements)
574 {
575 draw->pt.user.elts = elements;
576 }
577
578
579 /* Revamp me please:
580 */
581 void draw_do_flush( struct draw_context *draw, unsigned flags )
582 {
583 if (!draw->suspend_flushing)
584 {
585 assert(!draw->flushing); /* catch inadvertant recursion */
586
587 draw->flushing = TRUE;
588
589 draw_pipeline_flush( draw, flags );
590
591 draw->reduced_prim = ~0; /* is reduced_prim needed any more? */
592
593 draw->flushing = FALSE;
594 }
595 }
596
597
598 /**
599 * Return the number of output attributes produced by the geometry
600 * shader, if present. If no geometry shader, return the number of
601 * outputs from the vertex shader.
602 * \sa draw_num_shader_outputs
603 */
604 uint
605 draw_current_shader_outputs(const struct draw_context *draw)
606 {
607 if (draw->gs.geometry_shader)
608 return draw->gs.num_gs_outputs;
609 return draw->vs.num_vs_outputs;
610 }
611
612
613 /**
614 * Return the index of the shader output which will contain the
615 * vertex position.
616 */
617 uint
618 draw_current_shader_position_output(const struct draw_context *draw)
619 {
620 if (draw->gs.geometry_shader)
621 return draw->gs.position_output;
622 return draw->vs.position_output;
623 }
624
625
626 /**
627 * Return a pointer/handle for a driver/CSO rasterizer object which
628 * disabled culling, stippling, unfilled tris, etc.
629 * This is used by some pipeline stages (such as wide_point, aa_line
630 * and aa_point) which convert points/lines into triangles. In those
631 * cases we don't want to accidentally cull the triangles.
632 *
633 * \param scissor should the rasterizer state enable scissoring?
634 * \param flatshade should the rasterizer state use flat shading?
635 * \return rasterizer CSO handle
636 */
637 void *
638 draw_get_rasterizer_no_cull( struct draw_context *draw,
639 boolean scissor,
640 boolean flatshade )
641 {
642 if (!draw->rasterizer_no_cull[scissor][flatshade]) {
643 /* create now */
644 struct pipe_context *pipe = draw->pipe;
645 struct pipe_rasterizer_state rast;
646
647 memset(&rast, 0, sizeof(rast));
648 rast.scissor = scissor;
649 rast.flatshade = flatshade;
650 rast.front_ccw = 1;
651 rast.gl_rasterization_rules = draw->rasterizer->gl_rasterization_rules;
652
653 draw->rasterizer_no_cull[scissor][flatshade] =
654 pipe->create_rasterizer_state(pipe, &rast);
655 }
656 return draw->rasterizer_no_cull[scissor][flatshade];
657 }
658
659 void
660 draw_set_mapped_so_buffers(struct draw_context *draw,
661 void *buffers[PIPE_MAX_SO_BUFFERS],
662 unsigned num_buffers)
663 {
664 int i;
665
666 for (i = 0; i < num_buffers; ++i) {
667 draw->so.buffers[i] = buffers[i];
668 }
669 draw->so.num_buffers = num_buffers;
670 }
671
672 void
673 draw_set_so_state(struct draw_context *draw,
674 struct pipe_stream_output_state *state)
675 {
676 memcpy(&draw->so.state,
677 state,
678 sizeof(struct pipe_stream_output_state));
679 }
680
681 void
682 draw_set_sampler_views(struct draw_context *draw,
683 struct pipe_sampler_view **views,
684 unsigned num)
685 {
686 unsigned i;
687
688 debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS);
689
690 for (i = 0; i < num; ++i)
691 draw->sampler_views[i] = views[i];
692 for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i)
693 draw->sampler_views[i] = NULL;
694
695 draw->num_sampler_views = num;
696 }
697
698 void
699 draw_set_samplers(struct draw_context *draw,
700 struct pipe_sampler_state **samplers,
701 unsigned num)
702 {
703 unsigned i;
704
705 debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS);
706
707 for (i = 0; i < num; ++i)
708 draw->samplers[i] = samplers[i];
709 for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i)
710 draw->samplers[i] = NULL;
711
712 draw->num_samplers = num;
713
714 #ifdef HAVE_LLVM
715 if (draw->llvm)
716 draw_llvm_set_sampler_state(draw);
717 #endif
718 }
719
720 void
721 draw_set_mapped_texture(struct draw_context *draw,
722 unsigned sampler_idx,
723 uint32_t width, uint32_t height, uint32_t depth,
724 uint32_t last_level,
725 uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
726 uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
727 const void *data[PIPE_MAX_TEXTURE_LEVELS])
728 {
729 #ifdef HAVE_LLVM
730 if(draw->llvm)
731 draw_llvm_set_mapped_texture(draw,
732 sampler_idx,
733 width, height, depth, last_level,
734 row_stride, img_stride, data);
735 #endif
736 }