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 /*
29 * Authors:
30 * Keith Whitwell <keithw@vmware.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 /* Temporaries:
54 */
55 const float *constants;
56 unsigned pitch[PIPE_MAX_ATTRIBS];
57 const ubyte *src[PIPE_MAX_ATTRIBS];
58 unsigned prim;
59
60 struct draw_vs_variant_key key;
61 struct draw_vs_variant *active;
62
63 const struct vertex_info *vinfo;
64 };
65
66
67
68 static void
fse_prepare(struct draw_pt_middle_end * middle,enum pipe_prim_type prim,unsigned opt,unsigned * max_vertices)69 fse_prepare(struct draw_pt_middle_end *middle,
70 enum pipe_prim_type prim,
71 unsigned opt,
72 unsigned *max_vertices)
73 {
74 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
75 struct draw_context *draw = fse->draw;
76 unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
77 const struct vertex_info *vinfo;
78 unsigned nr_vbs = 0;
79
80 /* Can't support geometry shader on this path.
81 */
82 assert(!draw->gs.geometry_shader);
83
84 draw->render->set_primitive(draw->render, prim);
85
86 /* Must do this after set_primitive() above:
87 */
88 fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
89
90 fse->key.output_stride = vinfo->size * 4;
91 fse->key.nr_outputs = vinfo->num_attribs;
92 fse->key.nr_inputs = num_vs_inputs;
93
94 fse->key.nr_elements = MAX2(fse->key.nr_outputs, /* outputs - translate to hw format */
95 fse->key.nr_inputs); /* inputs - fetch from api format */
96
97 fse->key.viewport = !draw->bypass_viewport;
98 fse->key.clip = draw->clip_xy || draw->clip_z || draw->clip_user;
99 fse->key.const_vbuffers = 0;
100
101 memset(fse->key.element, 0,
102 fse->key.nr_elements * sizeof(fse->key.element[0]));
103
104 for (unsigned i = 0; i < num_vs_inputs; i++) {
105 const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
106 fse->key.element[i].in.format = src->src_format;
107
108 /* Consider ignoring these, ie make generated programs
109 * independent of this state:
110 */
111 fse->key.element[i].in.buffer = src->vertex_buffer_index;
112 fse->key.element[i].in.offset = src->src_offset;
113 nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
114 }
115
116 for (unsigned i = 0; i < 5 && i < nr_vbs; i++) {
117 if (draw->pt.vertex_buffer[i].stride == 0)
118 fse->key.const_vbuffers |= (1<<i);
119 }
120
121 if (0) debug_printf("%s: lookup const_vbuffers: %x\n",
122 __FUNCTION__, fse->key.const_vbuffers);
123
124 {
125 unsigned dst_offset = 0;
126
127 for (unsigned i = 0; i < vinfo->num_attribs; i++) {
128 unsigned emit_sz = draw_translate_vinfo_size(vinfo->attrib[i].emit);
129
130 /* doesn't handle EMIT_OMIT */
131 assert(emit_sz != 0);
132
133 /* The elements in the key correspond to vertex shader output
134 * numbers, not to positions in the hw vertex description --
135 * that's handled by the output_offset field.
136 */
137 fse->key.element[i].out.format = vinfo->attrib[i].emit;
138 fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
139 fse->key.element[i].out.offset = dst_offset;
140
141 dst_offset += emit_sz;
142 assert(fse->key.output_stride >= dst_offset);
143 }
144 }
145
146 fse->active = draw_vs_lookup_variant(draw->vs.vertex_shader, &fse->key);
147
148 if (!fse->active) {
149 assert(0);
150 return ;
151 }
152
153 if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__,
154 fse->active->key.const_vbuffers);
155
156 /* Now set buffer pointers:
157 */
158 for (unsigned i = 0; i < draw->pt.nr_vertex_buffers; i++) {
159 fse->active->set_buffer(fse->active,
160 i,
161 ((const ubyte *) draw->pt.user.vbuffer[i].map +
162 draw->pt.vertex_buffer[i].buffer_offset),
163 draw->pt.vertex_buffer[i].stride,
164 draw->pt.max_index);
165 }
166
167 *max_vertices = (draw->render->max_vertex_buffer_bytes /
168 (vinfo->size * 4));
169
170 /* Probably need to do this somewhere (or fix exec shader not to
171 * need it):
172 */
173 if (1) {
174 struct draw_vertex_shader *vs = draw->vs.vertex_shader;
175 vs->prepare(vs, draw);
176 }
177 }
178
179
180 static void
fse_bind_parameters(struct draw_pt_middle_end * middle)181 fse_bind_parameters(struct draw_pt_middle_end *middle)
182 {
183 /* No-op? */
184 }
185
186
187 static void
fse_run_linear(struct draw_pt_middle_end * middle,unsigned start,unsigned count,unsigned prim_flags)188 fse_run_linear(struct draw_pt_middle_end *middle,
189 unsigned start,
190 unsigned count,
191 unsigned prim_flags)
192 {
193 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
194 struct draw_context *draw = fse->draw;
195 char *hw_verts;
196
197 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
198 */
199 draw_do_flush(draw, DRAW_FLUSH_BACKEND);
200
201 if (!draw->render->allocate_vertices(draw->render,
202 (ushort) fse->key.output_stride,
203 (ushort) count))
204 goto fail;
205
206 hw_verts = draw->render->map_vertices(draw->render);
207 if (!hw_verts)
208 goto fail;
209
210 /* Single routine to fetch vertices, run shader and emit HW verts.
211 * Clipping is done elsewhere -- either by the API or on hardware,
212 * or for some other reason not required...
213 */
214 fse->active->run_linear(fse->active, start, count, hw_verts);
215
216 if (0) {
217 for (unsigned i = 0; i < count; i++) {
218 debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n",
219 __FUNCTION__, i,
220 fse->key.output_stride,
221 fse->key.output_stride * i);
222
223 draw_dump_emitted_vertex(fse->vinfo,
224 (const uint8_t *) hw_verts
225 + fse->key.output_stride * i);
226 }
227 }
228
229 draw->render->unmap_vertices(draw->render, 0, (ushort) (count - 1));
230
231 /* Draw arrays path to avoid re-emitting index list again and
232 * again.
233 */
234 draw->render->draw_arrays(draw->render, 0, count);
235
236 draw->render->release_vertices(draw->render);
237
238 return;
239
240 fail:
241 debug_warn_once("allocate or map of vertex buffer failed (out of memory?)");
242 return;
243 }
244
245
246 static void
fse_run(struct draw_pt_middle_end * middle,const unsigned * fetch_elts,unsigned fetch_count,const ushort * draw_elts,unsigned draw_count,unsigned prim_flags)247 fse_run(struct draw_pt_middle_end *middle,
248 const unsigned *fetch_elts,
249 unsigned fetch_count,
250 const ushort *draw_elts,
251 unsigned draw_count,
252 unsigned prim_flags)
253 {
254 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
255 struct draw_context *draw = fse->draw;
256 void *hw_verts;
257
258 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
259 */
260 draw_do_flush(draw, DRAW_FLUSH_BACKEND);
261
262 if (!draw->render->allocate_vertices(draw->render,
263 (ushort) fse->key.output_stride,
264 (ushort) fetch_count))
265 goto fail;
266
267 hw_verts = draw->render->map_vertices(draw->render);
268 if (!hw_verts)
269 goto fail;
270
271 /* Single routine to fetch vertices, run shader and emit HW verts.
272 */
273 fse->active->run_elts(fse->active, fetch_elts, fetch_count, hw_verts);
274
275 if (0) {
276 for (unsigned i = 0; i < fetch_count; i++) {
277 debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
278 draw_dump_emitted_vertex(fse->vinfo,
279 (const uint8_t *)hw_verts +
280 fse->key.output_stride * i);
281 }
282 }
283
284 draw->render->unmap_vertices(draw->render, 0, (ushort)(fetch_count - 1));
285
286 draw->render->draw_elements(draw->render, draw_elts, draw_count);
287
288 draw->render->release_vertices(draw->render);
289 return;
290
291 fail:
292 debug_warn_once("allocate or map of vertex buffer failed (out of memory?)");
293 return;
294 }
295
296
297 static boolean
fse_run_linear_elts(struct draw_pt_middle_end * middle,unsigned start,unsigned count,const ushort * draw_elts,unsigned draw_count,unsigned prim_flags)298 fse_run_linear_elts(struct draw_pt_middle_end *middle,
299 unsigned start,
300 unsigned count,
301 const ushort *draw_elts,
302 unsigned draw_count,
303 unsigned prim_flags)
304 {
305 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
306 struct draw_context *draw = fse->draw;
307 char *hw_verts;
308
309 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
310 */
311 draw_do_flush(draw, DRAW_FLUSH_BACKEND);
312
313 if (!draw->render->allocate_vertices(draw->render,
314 (ushort) fse->key.output_stride,
315 (ushort) count))
316 return FALSE;
317
318 hw_verts = draw->render->map_vertices(draw->render);
319 if (!hw_verts)
320 return FALSE;
321
322 /* Single routine to fetch vertices, run shader and emit HW verts.
323 * Clipping is done elsewhere -- either by the API or on hardware,
324 * or for some other reason not required...
325 */
326 fse->active->run_linear(fse->active, start, count, hw_verts);
327
328 draw->render->draw_elements(draw->render, draw_elts, draw_count);
329
330 draw->render->unmap_vertices(draw->render, 0, (ushort)(count - 1));
331
332 draw->render->release_vertices(draw->render);
333
334 return TRUE;
335 }
336
337
338 static void
fse_finish(struct draw_pt_middle_end * middle)339 fse_finish(struct draw_pt_middle_end *middle)
340 {
341 }
342
343
344 static void
fse_destroy(struct draw_pt_middle_end * middle)345 fse_destroy(struct draw_pt_middle_end *middle)
346 {
347 FREE(middle);
348 }
349
350
351 struct draw_pt_middle_end *
draw_pt_middle_fse(struct draw_context * draw)352 draw_pt_middle_fse(struct draw_context *draw)
353 {
354 struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
355 if (!fse)
356 return NULL;
357
358 fse->base.prepare = fse_prepare;
359 fse->base.bind_parameters = fse_bind_parameters;
360 fse->base.run = fse_run;
361 fse->base.run_linear = fse_run_linear;
362 fse->base.run_linear_elts = fse_run_linear_elts;
363 fse->base.finish = fse_finish;
364 fse->base.destroy = fse_destroy;
365 fse->draw = draw;
366
367 return &fse->base;
368 }
369