ad1d38b0ee2d6592e51f4926693d7605d40ee85a
[mesa.git] / src / mesa / vbo / vbo_exec_draw.c
1 /*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2008 Brian Paul 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 "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 *
24 * Authors:
25 * Keith Whitwell <keithw@vmware.com>
26 */
27
28 #include <stdbool.h>
29 #include <stdio.h>
30 #include "main/glheader.h"
31 #include "main/bufferobj.h"
32 #include "main/compiler.h"
33 #include "main/context.h"
34 #include "main/enums.h"
35 #include "main/state.h"
36 #include "main/vtxfmt.h"
37
38 #include "vbo_context.h"
39 #include "vbo_noop.h"
40
41
42 static void
43 vbo_exec_debug_verts( struct vbo_exec_context *exec )
44 {
45 GLuint count = exec->vtx.vert_count;
46 GLuint i;
47
48 printf("%s: %u vertices %d primitives, %d vertsize\n",
49 __func__,
50 count,
51 exec->vtx.prim_count,
52 exec->vtx.vertex_size);
53
54 for (i = 0 ; i < exec->vtx.prim_count ; i++) {
55 struct _mesa_prim *prim = &exec->vtx.prim[i];
56 printf(" prim %d: %s%s %d..%d %s %s\n",
57 i,
58 _mesa_lookup_prim_by_nr(prim->mode),
59 prim->weak ? " (weak)" : "",
60 prim->start,
61 prim->start + prim->count,
62 prim->begin ? "BEGIN" : "(wrap)",
63 prim->end ? "END" : "(wrap)");
64 }
65 }
66
67
68 /**
69 * Copy zero, one or two vertices from the current vertex buffer into
70 * the temporary "copy" buffer.
71 * This is used when a single primitive overflows a vertex buffer and
72 * we need to continue the primitive in a new vertex buffer.
73 * The temporary "copy" buffer holds the vertices which need to get
74 * copied from the old buffer to the new one.
75 */
76 static GLuint
77 vbo_copy_vertices( struct vbo_exec_context *exec )
78 {
79 struct _mesa_prim *last_prim = &exec->vtx.prim[exec->vtx.prim_count - 1];
80 const GLuint nr = last_prim->count;
81 GLuint ovf, i;
82 const GLuint sz = exec->vtx.vertex_size;
83 fi_type *dst = exec->vtx.copied.buffer;
84 const fi_type *src = exec->vtx.buffer_map + last_prim->start * sz;
85
86 switch (exec->ctx->Driver.CurrentExecPrimitive) {
87 case GL_POINTS:
88 return 0;
89 case GL_LINES:
90 ovf = nr&1;
91 for (i = 0 ; i < ovf ; i++)
92 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
93 return i;
94 case GL_TRIANGLES:
95 ovf = nr%3;
96 for (i = 0 ; i < ovf ; i++)
97 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
98 return i;
99 case GL_QUADS:
100 ovf = nr&3;
101 for (i = 0 ; i < ovf ; i++)
102 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
103 return i;
104 case GL_LINE_STRIP:
105 if (nr == 0) {
106 return 0;
107 }
108 else {
109 memcpy( dst, src+(nr-1)*sz, sz * sizeof(GLfloat) );
110 return 1;
111 }
112 case GL_LINE_LOOP:
113 if (last_prim->begin == 0) {
114 /* We're dealing with the second or later section of a split/wrapped
115 * GL_LINE_LOOP. Since we're converting line loops to line strips,
116 * we've already increment the last_prim->start counter by one to
117 * skip the 0th vertex in the loop. We need to undo that (effectively
118 * subtract one from last_prim->start) so that we copy the 0th vertex
119 * to the next vertex buffer.
120 */
121 assert(last_prim->start > 0);
122 src -= sz;
123 }
124 /* fall-through */
125 case GL_TRIANGLE_FAN:
126 case GL_POLYGON:
127 if (nr == 0) {
128 return 0;
129 }
130 else if (nr == 1) {
131 memcpy( dst, src+0, sz * sizeof(GLfloat) );
132 return 1;
133 }
134 else {
135 memcpy( dst, src+0, sz * sizeof(GLfloat) );
136 memcpy( dst+sz, src+(nr-1)*sz, sz * sizeof(GLfloat) );
137 return 2;
138 }
139 case GL_TRIANGLE_STRIP:
140 /* no parity issue, but need to make sure the tri is not drawn twice */
141 if (nr & 1) {
142 last_prim->count--;
143 }
144 /* fallthrough */
145 case GL_QUAD_STRIP:
146 switch (nr) {
147 case 0:
148 ovf = 0;
149 break;
150 case 1:
151 ovf = 1;
152 break;
153 default:
154 ovf = 2 + (nr & 1);
155 break;
156 }
157 for (i = 0 ; i < ovf ; i++)
158 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
159 return i;
160 case PRIM_OUTSIDE_BEGIN_END:
161 return 0;
162 default:
163 assert(0);
164 return 0;
165 }
166 }
167
168
169
170 /* TODO: populate these as the vertex is defined:
171 */
172 static void
173 vbo_exec_bind_arrays( struct gl_context *ctx )
174 {
175 struct vbo_context *vbo = vbo_context(ctx);
176 struct vbo_exec_context *exec = &vbo->exec;
177 struct gl_vertex_array *arrays = exec->vtx.arrays;
178 const GLubyte *map;
179 GLuint attr;
180 GLbitfield varying_inputs = 0x0;
181 bool swap_pos = false;
182
183 /* Install the default (ie Current) attributes first */
184 for (attr = 0; attr < VERT_ATTRIB_FF_MAX; attr++) {
185 exec->vtx.inputs[attr] = &vbo->currval[VBO_ATTRIB_POS+attr];
186 }
187
188 /* Overlay other active attributes */
189 switch (get_program_mode(exec->ctx)) {
190 case VP_NONE:
191 for (attr = 0; attr < MAT_ATTRIB_MAX; attr++) {
192 assert(VERT_ATTRIB_GENERIC(attr) < ARRAY_SIZE(exec->vtx.inputs));
193 exec->vtx.inputs[VERT_ATTRIB_GENERIC(attr)] =
194 &vbo->currval[VBO_ATTRIB_MAT_FRONT_AMBIENT+attr];
195 }
196 map = vbo->map_vp_none;
197 break;
198 case VP_ARB:
199 for (attr = 0; attr < VERT_ATTRIB_GENERIC_MAX; attr++) {
200 assert(VERT_ATTRIB_GENERIC(attr) < ARRAY_SIZE(exec->vtx.inputs));
201 exec->vtx.inputs[VERT_ATTRIB_GENERIC(attr)] =
202 &vbo->currval[VBO_ATTRIB_GENERIC0+attr];
203 }
204 map = vbo->map_vp_arb;
205
206 /* check if VERT_ATTRIB_POS is not read but VERT_BIT_GENERIC0 is read.
207 * In that case we effectively need to route the data from
208 * glVertexAttrib(0, val) calls to feed into the GENERIC0 input.
209 * The original state gets essentially restored below.
210 */
211 const GLbitfield64 inputs_read =
212 ctx->VertexProgram._Current->info.inputs_read;
213 if ((inputs_read & VERT_BIT_POS) == 0 &&
214 (inputs_read & VERT_BIT_GENERIC0)) {
215 swap_pos = true;
216 exec->vtx.inputs[VERT_ATTRIB_GENERIC0] = exec->vtx.inputs[0];
217 exec->vtx.attrsz[VERT_ATTRIB_GENERIC0] = exec->vtx.attrsz[0];
218 exec->vtx.attrtype[VERT_ATTRIB_GENERIC0] = exec->vtx.attrtype[0];
219 exec->vtx.attrptr[VERT_ATTRIB_GENERIC0] = exec->vtx.attrptr[0];
220 exec->vtx.attrsz[0] = 0;
221 }
222 break;
223 default:
224 assert(0);
225 }
226
227 for (attr = 0; attr < VERT_ATTRIB_MAX ; attr++) {
228 const GLuint src = map[attr];
229
230 if (exec->vtx.attrsz[src]) {
231 GLsizeiptr offset = (GLbyte *)exec->vtx.attrptr[src] -
232 (GLbyte *)exec->vtx.vertex;
233
234 /* override the default array set above */
235 assert(attr < ARRAY_SIZE(exec->vtx.inputs));
236 assert(attr < ARRAY_SIZE(exec->vtx.arrays)); /* arrays[] */
237 exec->vtx.inputs[attr] = &arrays[attr];
238
239 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
240 /* a real buffer obj: Ptr is an offset, not a pointer */
241 assert(exec->vtx.bufferobj->Mappings[MAP_INTERNAL].Pointer);
242 assert(offset >= 0);
243 arrays[attr].Ptr = (GLubyte *)
244 exec->vtx.bufferobj->Mappings[MAP_INTERNAL].Offset + offset;
245 }
246 else {
247 /* Ptr into ordinary app memory */
248 arrays[attr].Ptr = (GLubyte *)exec->vtx.buffer_map + offset;
249 }
250 arrays[attr].Size = exec->vtx.attrsz[src];
251 arrays[attr].StrideB = exec->vtx.vertex_size * sizeof(GLfloat);
252 arrays[attr].Type = exec->vtx.attrtype[src];
253 arrays[attr].Integer =
254 vbo_attrtype_to_integer_flag(exec->vtx.attrtype[src]);
255 arrays[attr].Format = GL_RGBA;
256 arrays[attr]._ElementSize = arrays[attr].Size * sizeof(GLfloat);
257 _mesa_reference_buffer_object(ctx,
258 &arrays[attr].BufferObj,
259 exec->vtx.bufferobj);
260
261 varying_inputs |= VERT_BIT(attr);
262 }
263 }
264
265 /* In case we swapped the position and generic0 attribute.
266 * Restore the original setting of the vtx.* variables.
267 * They are still needed with the original order and settings in case
268 * of a split primitive.
269 */
270 if (swap_pos) {
271 exec->vtx.attrsz[0] = exec->vtx.attrsz[VERT_ATTRIB_GENERIC0];
272 exec->vtx.attrsz[VERT_ATTRIB_GENERIC0] = 0;
273 }
274
275 _mesa_set_varying_vp_inputs( ctx, varying_inputs );
276 ctx->NewDriverState |= ctx->DriverFlags.NewArray;
277 }
278
279
280 /**
281 * Unmap the VBO. This is called before drawing.
282 */
283 static void
284 vbo_exec_vtx_unmap( struct vbo_exec_context *exec )
285 {
286 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
287 struct gl_context *ctx = exec->ctx;
288
289 if (ctx->Driver.FlushMappedBufferRange) {
290 GLintptr offset = exec->vtx.buffer_used -
291 exec->vtx.bufferobj->Mappings[MAP_INTERNAL].Offset;
292 GLsizeiptr length = (exec->vtx.buffer_ptr - exec->vtx.buffer_map) *
293 sizeof(float);
294
295 if (length)
296 ctx->Driver.FlushMappedBufferRange(ctx, offset, length,
297 exec->vtx.bufferobj,
298 MAP_INTERNAL);
299 }
300
301 exec->vtx.buffer_used += (exec->vtx.buffer_ptr -
302 exec->vtx.buffer_map) * sizeof(float);
303
304 assert(exec->vtx.buffer_used <= VBO_VERT_BUFFER_SIZE);
305 assert(exec->vtx.buffer_ptr != NULL);
306
307 ctx->Driver.UnmapBuffer(ctx, exec->vtx.bufferobj, MAP_INTERNAL);
308 exec->vtx.buffer_map = NULL;
309 exec->vtx.buffer_ptr = NULL;
310 exec->vtx.max_vert = 0;
311 }
312 }
313
314
315 /**
316 * Map the vertex buffer to begin storing glVertex, glColor, etc data.
317 */
318 void
319 vbo_exec_vtx_map( struct vbo_exec_context *exec )
320 {
321 struct gl_context *ctx = exec->ctx;
322 const GLenum accessRange = GL_MAP_WRITE_BIT | /* for MapBufferRange */
323 GL_MAP_INVALIDATE_RANGE_BIT |
324 GL_MAP_UNSYNCHRONIZED_BIT |
325 GL_MAP_FLUSH_EXPLICIT_BIT |
326 MESA_MAP_NOWAIT_BIT;
327 const GLenum usage = GL_STREAM_DRAW_ARB;
328
329 if (!_mesa_is_bufferobj(exec->vtx.bufferobj))
330 return;
331
332 assert(!exec->vtx.buffer_map);
333 assert(!exec->vtx.buffer_ptr);
334
335 if (VBO_VERT_BUFFER_SIZE > exec->vtx.buffer_used + 1024) {
336 /* The VBO exists and there's room for more */
337 if (exec->vtx.bufferobj->Size > 0) {
338 exec->vtx.buffer_map =
339 (fi_type *)ctx->Driver.MapBufferRange(ctx,
340 exec->vtx.buffer_used,
341 (VBO_VERT_BUFFER_SIZE -
342 exec->vtx.buffer_used),
343 accessRange,
344 exec->vtx.bufferobj,
345 MAP_INTERNAL);
346 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
347 }
348 else {
349 exec->vtx.buffer_ptr = exec->vtx.buffer_map = NULL;
350 }
351 }
352
353 if (!exec->vtx.buffer_map) {
354 /* Need to allocate a new VBO */
355 exec->vtx.buffer_used = 0;
356
357 if (ctx->Driver.BufferData(ctx, GL_ARRAY_BUFFER_ARB,
358 VBO_VERT_BUFFER_SIZE,
359 NULL, usage,
360 GL_MAP_WRITE_BIT |
361 GL_DYNAMIC_STORAGE_BIT |
362 GL_CLIENT_STORAGE_BIT,
363 exec->vtx.bufferobj)) {
364 /* buffer allocation worked, now map the buffer */
365 exec->vtx.buffer_map =
366 (fi_type *)ctx->Driver.MapBufferRange(ctx,
367 0, VBO_VERT_BUFFER_SIZE,
368 accessRange,
369 exec->vtx.bufferobj,
370 MAP_INTERNAL);
371 }
372 else {
373 _mesa_error(ctx, GL_OUT_OF_MEMORY, "VBO allocation");
374 exec->vtx.buffer_map = NULL;
375 }
376 }
377
378 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
379
380 if (!exec->vtx.buffer_map) {
381 /* out of memory */
382 _mesa_install_exec_vtxfmt( ctx, &exec->vtxfmt_noop );
383 }
384 else {
385 if (_mesa_using_noop_vtxfmt(ctx->Exec)) {
386 /* The no-op functions are installed so switch back to regular
387 * functions. We do this test just to avoid frequent and needless
388 * calls to _mesa_install_exec_vtxfmt().
389 */
390 _mesa_install_exec_vtxfmt(ctx, &exec->vtxfmt);
391 }
392 }
393
394 if (0)
395 printf("map %d..\n", exec->vtx.buffer_used);
396 }
397
398
399
400 /**
401 * Execute the buffer and save copied verts.
402 * \param keep_unmapped if true, leave the VBO unmapped when we're done.
403 */
404 void
405 vbo_exec_vtx_flush(struct vbo_exec_context *exec, GLboolean keepUnmapped)
406 {
407 if (0)
408 vbo_exec_debug_verts( exec );
409
410 if (exec->vtx.prim_count &&
411 exec->vtx.vert_count) {
412
413 exec->vtx.copied.nr = vbo_copy_vertices( exec );
414
415 if (exec->vtx.copied.nr != exec->vtx.vert_count) {
416 struct gl_context *ctx = exec->ctx;
417
418 /* Before the update_state() as this may raise _NEW_VARYING_VP_INPUTS
419 * from _mesa_set_varying_vp_inputs().
420 */
421 vbo_exec_bind_arrays( ctx );
422
423 if (ctx->NewState)
424 _mesa_update_state( ctx );
425
426 vbo_exec_vtx_unmap(exec);
427
428 if (0)
429 printf("%s %d %d\n", __func__, exec->vtx.prim_count,
430 exec->vtx.vert_count);
431
432 vbo_context(ctx)->draw_prims( ctx,
433 exec->vtx.prim,
434 exec->vtx.prim_count,
435 NULL,
436 GL_TRUE,
437 0,
438 exec->vtx.vert_count - 1,
439 NULL, 0, NULL);
440
441 /* Get new storage -- unless asked not to. */
442 if (!keepUnmapped)
443 vbo_exec_vtx_map( exec );
444 }
445 }
446
447 /* May have to unmap explicitly if we didn't draw:
448 */
449 if (keepUnmapped && exec->vtx.buffer_map) {
450 vbo_exec_vtx_unmap( exec );
451 }
452
453 if (keepUnmapped || exec->vtx.vertex_size == 0)
454 exec->vtx.max_vert = 0;
455 else
456 exec->vtx.max_vert = vbo_compute_max_verts(exec);
457
458 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
459 exec->vtx.prim_count = 0;
460 exec->vtx.vert_count = 0;
461 }