draw_pt_fetch_shade_emit.c revision cdca3c58aa2d9549f5188910e2a77b438516714f 
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
44struct fetch_shade_emit;
45
46
47/* Prototype fetch, shade, emit-hw-verts all in one go.
48 */
49struct 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_variant_key key;
62   struct draw_vs_variant *active;
63
64
65   const struct vertex_info *vinfo;
66};
67
68
69
70static 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->clip_xy || draw->clip_z || draw->clip_user;
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_variant( 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.user.max_index );
173   }
174
175   *max_vertices = (draw->render->max_vertex_buffer_bytes /
176                    (vinfo->size * 4));
177
178   /* Probably need to do this somewhere (or fix exec shader not to
179    * need it):
180    */
181   if (1) {
182      struct draw_vertex_shader *vs = draw->vs.vertex_shader;
183      vs->prepare(vs, draw);
184   }
185}
186
187
188
189static void fse_run_linear( struct draw_pt_middle_end *middle,
190                            unsigned start,
191                            unsigned count,
192                            unsigned prim_flags )
193{
194   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
195   struct draw_context *draw = fse->draw;
196   char *hw_verts;
197
198   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
199    */
200   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
201
202   if (!draw->render->allocate_vertices( draw->render,
203                                         (ushort)fse->key.output_stride,
204                                         (ushort)count ))
205      goto fail;
206
207   hw_verts = draw->render->map_vertices( draw->render );
208   if (!hw_verts)
209      goto fail;
210
211   /* Single routine to fetch vertices, run shader and emit HW verts.
212    * Clipping is done elsewhere -- either by the API or on hardware,
213    * or for some other reason not required...
214    */
215   fse->active->run_linear( fse->active,
216                            start, count,
217                            hw_verts );
218
219
220   if (0) {
221      unsigned i;
222      for (i = 0; i < count; i++) {
223         debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i,
224                      fse->key.output_stride,
225                      fse->key.output_stride * i);
226
227         draw_dump_emitted_vertex( fse->vinfo,
228                                   (const uint8_t *)hw_verts + fse->key.output_stride * i );
229      }
230   }
231
232   draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
233
234   /* Draw arrays path to avoid re-emitting index list again and
235    * again.
236    */
237   draw->render->draw_arrays( draw->render,
238                              0,
239                              count );
240
241
242   draw->render->release_vertices( draw->render );
243
244   return;
245
246fail:
247   assert(0);
248   return;
249}
250
251
252static void
253fse_run(struct draw_pt_middle_end *middle,
254        const unsigned *fetch_elts,
255        unsigned fetch_count,
256        const ushort *draw_elts,
257        unsigned draw_count,
258        unsigned prim_flags )
259{
260   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
261   struct draw_context *draw = fse->draw;
262   void *hw_verts;
263
264   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
265    */
266   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
267
268   if (!draw->render->allocate_vertices( draw->render,
269                                         (ushort)fse->key.output_stride,
270                                         (ushort)fetch_count ))
271      goto fail;
272
273   hw_verts = draw->render->map_vertices( draw->render );
274   if (!hw_verts)
275      goto fail;
276
277
278   /* Single routine to fetch vertices, run shader and emit HW verts.
279    */
280   fse->active->run_elts( fse->active,
281                          fetch_elts,
282                          fetch_count,
283                          hw_verts );
284
285
286   if (0) {
287      unsigned i;
288      for (i = 0; i < fetch_count; i++) {
289         debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
290         draw_dump_emitted_vertex( fse->vinfo,
291                                   (const uint8_t *)hw_verts +
292                                   fse->key.output_stride * i );
293      }
294   }
295
296   draw->render->unmap_vertices( draw->render, 0, (ushort)(fetch_count - 1) );
297
298   draw->render->draw_elements( draw->render,
299                                draw_elts,
300                                draw_count );
301
302
303   draw->render->release_vertices( draw->render );
304   return;
305
306fail:
307   assert(0);
308   return;
309}
310
311
312
313static boolean fse_run_linear_elts( struct draw_pt_middle_end *middle,
314                                 unsigned start,
315                                 unsigned count,
316                                 const ushort *draw_elts,
317                                 unsigned draw_count,
318                                 unsigned prim_flags )
319{
320   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
321   struct draw_context *draw = fse->draw;
322   char *hw_verts;
323
324   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
325    */
326   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
327
328   if (!draw->render->allocate_vertices( draw->render,
329                                         (ushort)fse->key.output_stride,
330                                         (ushort)count ))
331      return FALSE;
332
333   hw_verts = draw->render->map_vertices( draw->render );
334   if (!hw_verts)
335      return FALSE;
336
337   /* Single routine to fetch vertices, run shader and emit HW verts.
338    * Clipping is done elsewhere -- either by the API or on hardware,
339    * or for some other reason not required...
340    */
341   fse->active->run_linear( fse->active,
342                            start, count,
343                            hw_verts );
344
345
346   draw->render->draw_elements( draw->render,
347                                draw_elts,
348                                draw_count );
349
350
351   draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
352
353   draw->render->release_vertices( draw->render );
354
355   return TRUE;
356}
357
358
359
360static void fse_finish( struct draw_pt_middle_end *middle )
361{
362}
363
364
365static void
366fse_destroy( struct draw_pt_middle_end *middle )
367{
368   FREE(middle);
369}
370
371struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
372{
373   struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
374   if (!fse)
375      return NULL;
376
377   fse->base.prepare = fse_prepare;
378   fse->base.run = fse_run;
379   fse->base.run_linear = fse_run_linear;
380   fse->base.run_linear_elts = fse_run_linear_elts;
381   fse->base.finish = fse_finish;
382   fse->base.destroy = fse_destroy;
383   fse->draw = draw;
384
385   return &fse->base;
386}
387