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 <[email protected]>
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 uint8_t *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 mesa_prim prim,unsigned opt,unsigned * max_vertices)69 fse_prepare(struct draw_pt_middle_end *middle,
70 enum mesa_prim 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 if (src->src_stride == 0)
114 fse->key.const_vbuffers |= (1<<src->vertex_buffer_index);
115 nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
116 }
117
118 if (0) debug_printf("%s: lookup const_vbuffers: %x\n",
119 __func__, fse->key.const_vbuffers);
120
121 {
122 unsigned dst_offset = 0;
123
124 for (unsigned i = 0; i < vinfo->num_attribs; i++) {
125 unsigned emit_sz = draw_translate_vinfo_size(vinfo->attrib[i].emit);
126
127 /* doesn't handle EMIT_OMIT */
128 assert(emit_sz != 0);
129
130 /* The elements in the key correspond to vertex shader output
131 * numbers, not to positions in the hw vertex description --
132 * that's handled by the output_offset field.
133 */
134 fse->key.element[i].out.format = vinfo->attrib[i].emit;
135 fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
136 fse->key.element[i].out.offset = dst_offset;
137
138 dst_offset += emit_sz;
139 assert(fse->key.output_stride >= dst_offset);
140 }
141 }
142
143 fse->active = draw_vs_lookup_variant(draw->vs.vertex_shader, &fse->key);
144
145 if (!fse->active) {
146 assert(0);
147 return ;
148 }
149
150 if (0) debug_printf("%s: found const_vbuffers: %x\n", __func__,
151 fse->active->key.const_vbuffers);
152
153 /* Now set buffer pointers:
154 */
155 for (unsigned i = 0; i < draw->pt.nr_vertex_buffers; i++) {
156 fse->active->set_buffer(fse->active,
157 i,
158 ((const uint8_t *) draw->pt.user.vbuffer[i].map +
159 draw->pt.vertex_buffer[i].buffer_offset),
160 draw->pt.vertex_strides[i],
161 draw->pt.max_index);
162 }
163
164 *max_vertices = (draw->render->max_vertex_buffer_bytes /
165 (vinfo->size * 4));
166
167 /* Probably need to do this somewhere (or fix exec shader not to
168 * need it):
169 */
170 if (1) {
171 struct draw_vertex_shader *vs = draw->vs.vertex_shader;
172 vs->prepare(vs, draw);
173 }
174 }
175
176
177 static void
fse_bind_parameters(struct draw_pt_middle_end * middle)178 fse_bind_parameters(struct draw_pt_middle_end *middle)
179 {
180 /* No-op? */
181 }
182
183
184 static void
fse_run_linear(struct draw_pt_middle_end * middle,unsigned start,unsigned count,unsigned prim_flags)185 fse_run_linear(struct draw_pt_middle_end *middle,
186 unsigned start,
187 unsigned count,
188 unsigned prim_flags)
189 {
190 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
191 struct draw_context *draw = fse->draw;
192 char *hw_verts;
193
194 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
195 */
196 draw_do_flush(draw, DRAW_FLUSH_BACKEND);
197
198 if (!draw->render->allocate_vertices(draw->render,
199 (uint16_t) fse->key.output_stride,
200 (uint16_t) count))
201 goto fail;
202
203 hw_verts = draw->render->map_vertices(draw->render);
204 if (!hw_verts)
205 goto fail;
206
207 /* Single routine to fetch vertices, run shader and emit HW verts.
208 * Clipping is done elsewhere -- either by the API or on hardware,
209 * or for some other reason not required...
210 */
211 fse->active->run_linear(fse->active, start, count, hw_verts);
212
213 if (0) {
214 for (unsigned i = 0; i < count; i++) {
215 debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n",
216 __func__, i,
217 fse->key.output_stride,
218 fse->key.output_stride * i);
219
220 draw_dump_emitted_vertex(fse->vinfo,
221 (const uint8_t *) hw_verts
222 + fse->key.output_stride * i);
223 }
224 }
225
226 draw->render->unmap_vertices(draw->render, 0, (uint16_t) (count - 1));
227
228 /* Draw arrays path to avoid re-emitting index list again and
229 * again.
230 */
231 draw->render->draw_arrays(draw->render, 0, count);
232
233 draw->render->release_vertices(draw->render);
234
235 return;
236
237 fail:
238 debug_warn_once("allocate or map of vertex buffer failed (out of memory?)");
239 return;
240 }
241
242
243 static void
fse_run(struct draw_pt_middle_end * middle,const unsigned * fetch_elts,unsigned fetch_count,const uint16_t * draw_elts,unsigned draw_count,unsigned prim_flags)244 fse_run(struct draw_pt_middle_end *middle,
245 const unsigned *fetch_elts,
246 unsigned fetch_count,
247 const uint16_t *draw_elts,
248 unsigned draw_count,
249 unsigned prim_flags)
250 {
251 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
252 struct draw_context *draw = fse->draw;
253 void *hw_verts;
254
255 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
256 */
257 draw_do_flush(draw, DRAW_FLUSH_BACKEND);
258
259 if (!draw->render->allocate_vertices(draw->render,
260 (uint16_t) fse->key.output_stride,
261 (uint16_t) fetch_count))
262 goto fail;
263
264 hw_verts = draw->render->map_vertices(draw->render);
265 if (!hw_verts)
266 goto fail;
267
268 /* Single routine to fetch vertices, run shader and emit HW verts.
269 */
270 fse->active->run_elts(fse->active, fetch_elts, fetch_count, hw_verts);
271
272 if (0) {
273 for (unsigned i = 0; i < fetch_count; i++) {
274 debug_printf("\n\n%s vertex %d:\n", __func__, i);
275 draw_dump_emitted_vertex(fse->vinfo,
276 (const uint8_t *)hw_verts +
277 fse->key.output_stride * i);
278 }
279 }
280
281 draw->render->unmap_vertices(draw->render, 0, (uint16_t)(fetch_count - 1));
282
283 draw->render->draw_elements(draw->render, draw_elts, draw_count);
284
285 draw->render->release_vertices(draw->render);
286 return;
287
288 fail:
289 debug_warn_once("allocate or map of vertex buffer failed (out of memory?)");
290 return;
291 }
292
293
294 static bool
fse_run_linear_elts(struct draw_pt_middle_end * middle,unsigned start,unsigned count,const uint16_t * draw_elts,unsigned draw_count,unsigned prim_flags)295 fse_run_linear_elts(struct draw_pt_middle_end *middle,
296 unsigned start,
297 unsigned count,
298 const uint16_t *draw_elts,
299 unsigned draw_count,
300 unsigned prim_flags)
301 {
302 struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
303 struct draw_context *draw = fse->draw;
304 char *hw_verts;
305
306 /* XXX: need to flush to get prim_vbuf.c to release its allocation??
307 */
308 draw_do_flush(draw, DRAW_FLUSH_BACKEND);
309
310 if (!draw->render->allocate_vertices(draw->render,
311 (uint16_t) fse->key.output_stride,
312 (uint16_t) count))
313 return false;
314
315 hw_verts = draw->render->map_vertices(draw->render);
316 if (!hw_verts)
317 return false;
318
319 /* Single routine to fetch vertices, run shader and emit HW verts.
320 * Clipping is done elsewhere -- either by the API or on hardware,
321 * or for some other reason not required...
322 */
323 fse->active->run_linear(fse->active, start, count, hw_verts);
324
325 draw->render->draw_elements(draw->render, draw_elts, draw_count);
326
327 draw->render->unmap_vertices(draw->render, 0, (uint16_t)(count - 1));
328
329 draw->render->release_vertices(draw->render);
330
331 return true;
332 }
333
334
335 static void
fse_finish(struct draw_pt_middle_end * middle)336 fse_finish(struct draw_pt_middle_end *middle)
337 {
338 }
339
340
341 static void
fse_destroy(struct draw_pt_middle_end * middle)342 fse_destroy(struct draw_pt_middle_end *middle)
343 {
344 FREE(middle);
345 }
346
347
348 struct draw_pt_middle_end *
draw_pt_middle_fse(struct draw_context * draw)349 draw_pt_middle_fse(struct draw_context *draw)
350 {
351 struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
352 if (!fse)
353 return NULL;
354
355 fse->base.prepare = fse_prepare;
356 fse->base.bind_parameters = fse_bind_parameters;
357 fse->base.run = fse_run;
358 fse->base.run_linear = fse_run_linear;
359 fse->base.run_linear_elts = fse_run_linear_elts;
360 fse->base.finish = fse_finish;
361 fse->base.destroy = fse_destroy;
362 fse->draw = draw;
363
364 return &fse->base;
365 }
366