Implement instanced indexed draw.
[mesa.git] / src / gallium / auxiliary / draw / draw_pt_fetch_emit.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 #include "util/u_memory.h"
34 #include "draw/draw_context.h"
35 #include "draw/draw_private.h"
36 #include "draw/draw_vbuf.h"
37 #include "draw/draw_vertex.h"
38 #include "draw/draw_pt.h"
39 #include "translate/translate.h"
40 #include "translate/translate_cache.h"
41
42 /* The simplest 'middle end' in the new vertex code.
43 *
44 * The responsibilities of a middle end are to:
45 * - perform vertex fetch using
46 * - draw vertex element/buffer state
47 * - a list of fetch indices we received as an input
48 * - run the vertex shader
49 * - cliptest,
50 * - clip coord calculation
51 * - viewport transformation
52 * - if necessary, run the primitive pipeline, passing it:
53 * - a linear array of vertex_header vertices constructed here
54 * - a set of draw indices we received as an input
55 * - otherwise, drive the hw backend,
56 * - allocate space for hardware format vertices
57 * - translate the vertex-shader output vertices to hw format
58 * - calling the backend draw functions.
59 *
60 * For convenience, we provide a helper function to drive the hardware
61 * backend given similar inputs to those required to run the pipeline.
62 *
63 * In the case of passthrough mode, many of these actions are disabled
64 * or noops, so we end up doing:
65 *
66 * - perform vertex fetch
67 * - drive the hw backend
68 *
69 * IE, basically just vertex fetch to post-vs-format vertices,
70 * followed by a call to the backend helper function.
71 */
72
73
74 struct fetch_emit_middle_end {
75 struct draw_pt_middle_end base;
76 struct draw_context *draw;
77
78 struct translate *translate;
79 const struct vertex_info *vinfo;
80
81 /* Cache point size somewhere it's address won't change:
82 */
83 float point_size;
84
85 struct translate_cache *cache;
86 };
87
88
89
90
91 static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
92 unsigned prim,
93 unsigned opt,
94 unsigned *max_vertices )
95 {
96 struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
97 struct draw_context *draw = feme->draw;
98 const struct vertex_info *vinfo;
99 unsigned i, dst_offset;
100 boolean ok;
101 struct translate_key key;
102
103
104 ok = draw->render->set_primitive( draw->render,
105 prim );
106 if (!ok) {
107 assert(0);
108 return;
109 }
110
111 /* Must do this after set_primitive() above:
112 */
113 vinfo = feme->vinfo = draw->render->get_vertex_info(draw->render);
114
115
116
117 /* Transform from API vertices to HW vertices, skipping the
118 * pipeline_vertex intermediate step.
119 */
120 dst_offset = 0;
121 memset(&key, 0, sizeof(key));
122
123 for (i = 0; i < vinfo->num_attribs; i++) {
124 const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->attrib[i].src_index];
125
126 unsigned emit_sz = 0;
127 unsigned input_format = src->src_format;
128 unsigned input_buffer = src->vertex_buffer_index;
129 unsigned input_offset = src->src_offset;
130 unsigned output_format;
131
132 switch (vinfo->attrib[i].emit) {
133 case EMIT_4UB:
134 output_format = PIPE_FORMAT_R8G8B8A8_UNORM;
135 emit_sz = 4 * sizeof(unsigned char);
136 break;
137 case EMIT_4F:
138 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
139 emit_sz = 4 * sizeof(float);
140 break;
141 case EMIT_3F:
142 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
143 emit_sz = 3 * sizeof(float);
144 break;
145 case EMIT_2F:
146 output_format = PIPE_FORMAT_R32G32_FLOAT;
147 emit_sz = 2 * sizeof(float);
148 break;
149 case EMIT_1F:
150 output_format = PIPE_FORMAT_R32_FLOAT;
151 emit_sz = 1 * sizeof(float);
152 break;
153 case EMIT_1F_PSIZE:
154 input_format = PIPE_FORMAT_R32_FLOAT;
155 input_buffer = draw->pt.nr_vertex_buffers;
156 input_offset = 0;
157 output_format = PIPE_FORMAT_R32_FLOAT;
158 emit_sz = 1 * sizeof(float);
159 break;
160 case EMIT_OMIT:
161 continue;
162 default:
163 assert(0);
164 output_format = PIPE_FORMAT_NONE;
165 emit_sz = 0;
166 continue;
167 }
168
169 key.element[i].input_format = input_format;
170 key.element[i].input_buffer = input_buffer;
171 key.element[i].input_offset = input_offset;
172 key.element[i].instance_divisor = src->instance_divisor;
173 key.element[i].output_format = output_format;
174 key.element[i].output_offset = dst_offset;
175
176 dst_offset += emit_sz;
177 }
178
179 key.nr_elements = vinfo->num_attribs;
180 key.output_stride = vinfo->size * 4;
181
182 /* Don't bother with caching at this stage:
183 */
184 if (!feme->translate ||
185 translate_key_compare(&feme->translate->key, &key) != 0)
186 {
187 translate_key_sanitize(&key);
188 feme->translate = translate_cache_find(feme->cache,
189 &key);
190
191
192 feme->translate->set_buffer(feme->translate,
193 draw->pt.nr_vertex_buffers,
194 &feme->point_size,
195 0);
196 }
197
198 feme->point_size = draw->rasterizer->point_size;
199
200 for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
201 feme->translate->set_buffer(feme->translate,
202 i,
203 ((char *)draw->pt.user.vbuffer[i] +
204 draw->pt.vertex_buffer[i].buffer_offset),
205 draw->pt.vertex_buffer[i].stride );
206 }
207
208 *max_vertices = (draw->render->max_vertex_buffer_bytes /
209 (vinfo->size * 4));
210
211 /* Return an even number of verts.
212 * This prevents "parity" errors when splitting long triangle strips which
213 * can lead to front/back culling mix-ups.
214 * Every other triangle in a strip has an alternate front/back orientation
215 * so splitting at an odd position can cause the orientation of subsequent
216 * triangles to get reversed.
217 */
218 *max_vertices = *max_vertices & ~1;
219 }
220
221
222
223
224
225 static void fetch_emit_run( struct draw_pt_middle_end *middle,
226 const unsigned *fetch_elts,
227 unsigned fetch_count,
228 const ushort *draw_elts,
229 unsigned draw_count )
230 {
231 struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
232 struct draw_context *draw = feme->draw;
233 void *hw_verts;
234
235 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
236 */
237 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
238
239 if (fetch_count >= UNDEFINED_VERTEX_ID) {
240 assert(0);
241 return;
242 }
243
244 draw->render->allocate_vertices( draw->render,
245 (ushort)feme->translate->key.output_stride,
246 (ushort)fetch_count );
247
248 hw_verts = draw->render->map_vertices( draw->render );
249 if (!hw_verts) {
250 assert(0);
251 return;
252 }
253
254
255 /* Single routine to fetch vertices and emit HW verts.
256 */
257 feme->translate->run_elts( feme->translate,
258 fetch_elts,
259 fetch_count,
260 draw->instance_id,
261 hw_verts );
262
263 if (0) {
264 unsigned i;
265 for (i = 0; i < fetch_count; i++) {
266 debug_printf("\n\nvertex %d:\n", i);
267 draw_dump_emitted_vertex( feme->vinfo,
268 (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i );
269 }
270 }
271
272 draw->render->unmap_vertices( draw->render,
273 0,
274 (ushort)(fetch_count - 1) );
275
276 /* XXX: Draw arrays path to avoid re-emitting index list again and
277 * again.
278 */
279 draw->render->draw( draw->render,
280 draw_elts,
281 draw_count );
282
283 /* Done -- that was easy, wasn't it:
284 */
285 draw->render->release_vertices( draw->render );
286
287 }
288
289
290 static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
291 unsigned start,
292 unsigned count )
293 {
294 struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
295 struct draw_context *draw = feme->draw;
296 void *hw_verts;
297
298 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
299 */
300 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
301
302 if (count >= UNDEFINED_VERTEX_ID)
303 goto fail;
304
305 if (!draw->render->allocate_vertices( draw->render,
306 (ushort)feme->translate->key.output_stride,
307 (ushort)count ))
308 goto fail;
309
310 hw_verts = draw->render->map_vertices( draw->render );
311 if (!hw_verts)
312 goto fail;
313
314 /* Single routine to fetch vertices and emit HW verts.
315 */
316 feme->translate->run( feme->translate,
317 start,
318 count,
319 draw->instance_id,
320 hw_verts );
321
322 if (0) {
323 unsigned i;
324 for (i = 0; i < count; i++) {
325 debug_printf("\n\nvertex %d:\n", i);
326 draw_dump_emitted_vertex( feme->vinfo,
327 (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i );
328 }
329 }
330
331 draw->render->unmap_vertices( draw->render, 0, count - 1 );
332
333 /* XXX: Draw arrays path to avoid re-emitting index list again and
334 * again.
335 */
336 draw->render->draw_arrays( draw->render, 0, count );
337
338 /* Done -- that was easy, wasn't it:
339 */
340 draw->render->release_vertices( draw->render );
341 return;
342
343 fail:
344 assert(0);
345 return;
346 }
347
348
349 static boolean fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
350 unsigned start,
351 unsigned count,
352 const ushort *draw_elts,
353 unsigned draw_count )
354 {
355 struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
356 struct draw_context *draw = feme->draw;
357 void *hw_verts;
358
359 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
360 */
361 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
362
363 if (count >= UNDEFINED_VERTEX_ID)
364 return FALSE;
365
366 if (!draw->render->allocate_vertices( draw->render,
367 (ushort)feme->translate->key.output_stride,
368 (ushort)count ))
369 return FALSE;
370
371 hw_verts = draw->render->map_vertices( draw->render );
372 if (!hw_verts)
373 return FALSE;
374
375 /* Single routine to fetch vertices and emit HW verts.
376 */
377 feme->translate->run( feme->translate,
378 start,
379 count,
380 draw->instance_id,
381 hw_verts );
382
383 draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
384
385 /* XXX: Draw arrays path to avoid re-emitting index list again and
386 * again.
387 */
388 draw->render->draw( draw->render,
389 draw_elts,
390 draw_count );
391
392 /* Done -- that was easy, wasn't it:
393 */
394 draw->render->release_vertices( draw->render );
395
396 return TRUE;
397 }
398
399
400
401
402 static void fetch_emit_finish( struct draw_pt_middle_end *middle )
403 {
404 /* nothing to do */
405 }
406
407 static void fetch_emit_destroy( struct draw_pt_middle_end *middle )
408 {
409 struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
410
411 if (feme->cache)
412 translate_cache_destroy(feme->cache);
413
414 FREE(middle);
415 }
416
417
418 struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw )
419 {
420 struct fetch_emit_middle_end *fetch_emit = CALLOC_STRUCT( fetch_emit_middle_end );
421 if (fetch_emit == NULL)
422 return NULL;
423
424 fetch_emit->cache = translate_cache_create();
425 if (!fetch_emit->cache) {
426 FREE(fetch_emit);
427 return NULL;
428 }
429
430 fetch_emit->base.prepare = fetch_emit_prepare;
431 fetch_emit->base.run = fetch_emit_run;
432 fetch_emit->base.run_linear = fetch_emit_run_linear;
433 fetch_emit->base.run_linear_elts = fetch_emit_run_linear_elts;
434 fetch_emit->base.finish = fetch_emit_finish;
435 fetch_emit->base.destroy = fetch_emit_destroy;
436
437 fetch_emit->draw = draw;
438
439 return &fetch_emit->base;
440 }
441