draw: don't try to precalculate the pipeline output primitive
[mesa.git] / src / gallium / auxiliary / draw / draw_pt_fetch_shade_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
34 #include "util/u_math.h"
35 #include "util/u_memory.h"
36 #include "draw/draw_context.h"
37 #include "draw/draw_private.h"
38 #include "draw/draw_vbuf.h"
39 #include "draw/draw_vertex.h"
40 #include "draw/draw_pt.h"
41 #include "draw/draw_vs.h"
42
43
44 struct fetch_shade_emit;
45
46
47 /* Prototype fetch, shade, emit-hw-verts all in one go.
48 */
49 struct fetch_shade_emit {
50 struct draw_pt_middle_end base;
51 struct draw_context *draw;
52
53
54 /* Temporaries:
55 */
56 const float *constants;
57 unsigned pitch[PIPE_MAX_ATTRIBS];
58 const ubyte *src[PIPE_MAX_ATTRIBS];
59 unsigned prim;
60
61 struct draw_vs_varient_key key;
62 struct draw_vs_varient *active;
63
64
65 const struct vertex_info *vinfo;
66 };
67
68
69
70 static void fse_prepare( struct draw_pt_middle_end *middle,
71 unsigned prim,
72 unsigned opt,
73 unsigned *max_vertices )
74 {
75 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
76 struct draw_context *draw = fse->draw;
77 unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
78 const struct vertex_info *vinfo;
79 unsigned i;
80 unsigned nr_vbs = 0;
81
82 /* Can't support geometry shader on this path.
83 */
84 assert(!draw->gs.geometry_shader);
85
86 if (!draw->render->set_primitive( draw->render,
87 prim )) {
88 assert(0);
89 return;
90 }
91
92 /* Must do this after set_primitive() above:
93 */
94 fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
95
96
97 fse->key.output_stride = vinfo->size * 4;
98 fse->key.nr_outputs = vinfo->num_attribs;
99 fse->key.nr_inputs = num_vs_inputs;
100
101 fse->key.nr_elements = MAX2(fse->key.nr_outputs, /* outputs - translate to hw format */
102 fse->key.nr_inputs); /* inputs - fetch from api format */
103
104 fse->key.viewport = !draw->identity_viewport;
105 fse->key.clip = !draw->bypass_clipping;
106 fse->key.const_vbuffers = 0;
107
108 memset(fse->key.element, 0,
109 fse->key.nr_elements * sizeof(fse->key.element[0]));
110
111 for (i = 0; i < num_vs_inputs; i++) {
112 const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
113 fse->key.element[i].in.format = src->src_format;
114
115 /* Consider ignoring these, ie make generated programs
116 * independent of this state:
117 */
118 fse->key.element[i].in.buffer = src->vertex_buffer_index;
119 fse->key.element[i].in.offset = src->src_offset;
120 nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
121 }
122
123 for (i = 0; i < 5 && i < nr_vbs; i++) {
124 if (draw->pt.vertex_buffer[i].stride == 0)
125 fse->key.const_vbuffers |= (1<<i);
126 }
127
128 if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers);
129
130 {
131 unsigned dst_offset = 0;
132
133 for (i = 0; i < vinfo->num_attribs; i++) {
134 unsigned emit_sz = draw_translate_vinfo_size(vinfo->attrib[i].emit);
135
136 /* doesn't handle EMIT_OMIT */
137 assert(emit_sz != 0);
138
139 /* The elements in the key correspond to vertex shader output
140 * numbers, not to positions in the hw vertex description --
141 * that's handled by the output_offset field.
142 */
143 fse->key.element[i].out.format = vinfo->attrib[i].emit;
144 fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
145 fse->key.element[i].out.offset = dst_offset;
146
147 dst_offset += emit_sz;
148 assert(fse->key.output_stride >= dst_offset);
149 }
150 }
151
152
153 fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader,
154 &fse->key );
155
156 if (!fse->active) {
157 assert(0);
158 return ;
159 }
160
161 if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__,
162 fse->active->key.const_vbuffers);
163
164 /* Now set buffer pointers:
165 */
166 for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
167 fse->active->set_buffer( fse->active,
168 i,
169 ((const ubyte *) draw->pt.user.vbuffer[i] +
170 draw->pt.vertex_buffer[i].buffer_offset),
171 draw->pt.vertex_buffer[i].stride,
172 draw->pt.vertex_buffer[i].max_index );
173 }
174
175 *max_vertices = (draw->render->max_vertex_buffer_bytes /
176 (vinfo->size * 4));
177
178 /* Return an even number of verts.
179 * This prevents "parity" errors when splitting long triangle strips which
180 * can lead to front/back culling mix-ups.
181 * Every other triangle in a strip has an alternate front/back orientation
182 * so splitting at an odd position can cause the orientation of subsequent
183 * triangles to get reversed.
184 */
185 *max_vertices = *max_vertices & ~1;
186
187 /* Probably need to do this somewhere (or fix exec shader not to
188 * need it):
189 */
190 if (1) {
191 struct draw_vertex_shader *vs = draw->vs.vertex_shader;
192 vs->prepare(vs, draw);
193 }
194 }
195
196
197
198 static void fse_run_linear( struct draw_pt_middle_end *middle,
199 unsigned start,
200 unsigned count )
201 {
202 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
203 struct draw_context *draw = fse->draw;
204 char *hw_verts;
205
206 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
207 */
208 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
209
210 if (count >= UNDEFINED_VERTEX_ID)
211 goto fail;
212
213 if (!draw->render->allocate_vertices( draw->render,
214 (ushort)fse->key.output_stride,
215 (ushort)count ))
216 goto fail;
217
218 hw_verts = draw->render->map_vertices( draw->render );
219 if (!hw_verts)
220 goto fail;
221
222 /* Single routine to fetch vertices, run shader and emit HW verts.
223 * Clipping is done elsewhere -- either by the API or on hardware,
224 * or for some other reason not required...
225 */
226 fse->active->run_linear( fse->active,
227 start, count,
228 hw_verts );
229
230
231 if (0) {
232 unsigned i;
233 for (i = 0; i < count; i++) {
234 debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i,
235 fse->key.output_stride,
236 fse->key.output_stride * i);
237
238 draw_dump_emitted_vertex( fse->vinfo,
239 (const uint8_t *)hw_verts + fse->key.output_stride * i );
240 }
241 }
242
243 draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
244
245 /* Draw arrays path to avoid re-emitting index list again and
246 * again.
247 */
248 draw->render->draw_arrays( draw->render,
249 0,
250 count );
251
252
253 draw->render->release_vertices( draw->render );
254
255 return;
256
257 fail:
258 assert(0);
259 return;
260 }
261
262
263 static void
264 fse_run(struct draw_pt_middle_end *middle,
265 const unsigned *fetch_elts,
266 unsigned fetch_count,
267 const ushort *draw_elts,
268 unsigned draw_count )
269 {
270 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
271 struct draw_context *draw = fse->draw;
272 void *hw_verts;
273
274 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
275 */
276 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
277
278 if (fetch_count >= UNDEFINED_VERTEX_ID)
279 goto fail;
280
281 if (!draw->render->allocate_vertices( draw->render,
282 (ushort)fse->key.output_stride,
283 (ushort)fetch_count ))
284 goto fail;
285
286 hw_verts = draw->render->map_vertices( draw->render );
287 if (!hw_verts)
288 goto fail;
289
290
291 /* Single routine to fetch vertices, run shader and emit HW verts.
292 */
293 fse->active->run_elts( fse->active,
294 fetch_elts,
295 fetch_count,
296 hw_verts );
297
298
299 if (0) {
300 unsigned i;
301 for (i = 0; i < fetch_count; i++) {
302 debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
303 draw_dump_emitted_vertex( fse->vinfo,
304 (const uint8_t *)hw_verts +
305 fse->key.output_stride * i );
306 }
307 }
308
309 draw->render->unmap_vertices( draw->render, 0, (ushort)(fetch_count - 1) );
310
311 draw->render->draw_elements( draw->render,
312 draw_elts,
313 draw_count );
314
315
316 draw->render->release_vertices( draw->render );
317 return;
318
319 fail:
320 assert(0);
321 return;
322 }
323
324
325
326 static boolean fse_run_linear_elts( struct draw_pt_middle_end *middle,
327 unsigned start,
328 unsigned count,
329 const ushort *draw_elts,
330 unsigned draw_count )
331 {
332 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
333 struct draw_context *draw = fse->draw;
334 char *hw_verts;
335
336 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
337 */
338 draw_do_flush( draw, DRAW_FLUSH_BACKEND );
339
340 if (count >= UNDEFINED_VERTEX_ID)
341 return FALSE;
342
343 if (!draw->render->allocate_vertices( draw->render,
344 (ushort)fse->key.output_stride,
345 (ushort)count ))
346 return FALSE;
347
348 hw_verts = draw->render->map_vertices( draw->render );
349 if (!hw_verts)
350 return FALSE;
351
352 /* Single routine to fetch vertices, run shader and emit HW verts.
353 * Clipping is done elsewhere -- either by the API or on hardware,
354 * or for some other reason not required...
355 */
356 fse->active->run_linear( fse->active,
357 start, count,
358 hw_verts );
359
360
361 draw->render->draw_elements( draw->render,
362 draw_elts,
363 draw_count );
364
365
366 draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
367
368 draw->render->release_vertices( draw->render );
369
370 return TRUE;
371 }
372
373
374
375 static void fse_finish( struct draw_pt_middle_end *middle )
376 {
377 }
378
379
380 static void
381 fse_destroy( struct draw_pt_middle_end *middle )
382 {
383 FREE(middle);
384 }
385
386 struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
387 {
388 struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
389 if (!fse)
390 return NULL;
391
392 fse->base.prepare = fse_prepare;
393 fse->base.run = fse_run;
394 fse->base.run_linear = fse_run_linear;
395 fse->base.run_linear_elts = fse_run_linear_elts;
396 fse->base.finish = fse_finish;
397 fse->base.destroy = fse_destroy;
398 fse->draw = draw;
399
400 return &fse->base;
401 }