Merge branch 'draw-instanced'
[mesa.git] / src / mesa / vbo / vbo_exec_draw.c
1 /*
2 * Mesa 3-D graphics library
3 * Version: 7.2
4 *
5 * Copyright (C) 1999-2008 Brian Paul All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 *
24 * Authors:
25 * Keith Whitwell <keith@tungstengraphics.com>
26 */
27
28 #include "main/glheader.h"
29 #include "main/bufferobj.h"
30 #include "main/compiler.h"
31 #include "main/enums.h"
32 #include "main/mfeatures.h"
33 #include "main/state.h"
34
35 #include "vbo_context.h"
36
37
38 #if FEATURE_beginend
39
40
41 static void
42 vbo_exec_debug_verts( struct vbo_exec_context *exec )
43 {
44 GLuint count = exec->vtx.vert_count;
45 GLuint i;
46
47 printf("%s: %u vertices %d primitives, %d vertsize\n",
48 __FUNCTION__,
49 count,
50 exec->vtx.prim_count,
51 exec->vtx.vertex_size);
52
53 for (i = 0 ; i < exec->vtx.prim_count ; i++) {
54 struct _mesa_prim *prim = &exec->vtx.prim[i];
55 printf(" prim %d: %s%s %d..%d %s %s\n",
56 i,
57 _mesa_lookup_prim_by_nr(prim->mode),
58 prim->weak ? " (weak)" : "",
59 prim->start,
60 prim->start + prim->count,
61 prim->begin ? "BEGIN" : "(wrap)",
62 prim->end ? "END" : "(wrap)");
63 }
64 }
65
66
67 /*
68 * NOTE: Need to have calculated primitives by this point -- do it on the fly.
69 * NOTE: Old 'parity' issue is gone.
70 */
71 static GLuint
72 vbo_copy_vertices( struct vbo_exec_context *exec )
73 {
74 GLuint nr = exec->vtx.prim[exec->vtx.prim_count-1].count;
75 GLuint ovf, i;
76 GLuint sz = exec->vtx.vertex_size;
77 GLfloat *dst = exec->vtx.copied.buffer;
78 const GLfloat *src = (exec->vtx.buffer_map +
79 exec->vtx.prim[exec->vtx.prim_count-1].start *
80 exec->vtx.vertex_size);
81
82
83 switch (exec->ctx->Driver.CurrentExecPrimitive) {
84 case GL_POINTS:
85 return 0;
86 case GL_LINES:
87 ovf = nr&1;
88 for (i = 0 ; i < ovf ; i++)
89 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
90 return i;
91 case GL_TRIANGLES:
92 ovf = nr%3;
93 for (i = 0 ; i < ovf ; i++)
94 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
95 return i;
96 case GL_QUADS:
97 ovf = nr&3;
98 for (i = 0 ; i < ovf ; i++)
99 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
100 return i;
101 case GL_LINE_STRIP:
102 if (nr == 0) {
103 return 0;
104 }
105 else {
106 memcpy( dst, src+(nr-1)*sz, sz * sizeof(GLfloat) );
107 return 1;
108 }
109 case GL_LINE_LOOP:
110 case GL_TRIANGLE_FAN:
111 case GL_POLYGON:
112 if (nr == 0) {
113 return 0;
114 }
115 else if (nr == 1) {
116 memcpy( dst, src+0, sz * sizeof(GLfloat) );
117 return 1;
118 }
119 else {
120 memcpy( dst, src+0, sz * sizeof(GLfloat) );
121 memcpy( dst+sz, src+(nr-1)*sz, sz * sizeof(GLfloat) );
122 return 2;
123 }
124 case GL_TRIANGLE_STRIP:
125 /* no parity issue, but need to make sure the tri is not drawn twice */
126 if (nr & 1) {
127 exec->vtx.prim[exec->vtx.prim_count-1].count--;
128 }
129 /* fallthrough */
130 case GL_QUAD_STRIP:
131 switch (nr) {
132 case 0:
133 ovf = 0;
134 break;
135 case 1:
136 ovf = 1;
137 break;
138 default:
139 ovf = 2 + (nr & 1);
140 break;
141 }
142 for (i = 0 ; i < ovf ; i++)
143 memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
144 return i;
145 case PRIM_OUTSIDE_BEGIN_END:
146 return 0;
147 default:
148 assert(0);
149 return 0;
150 }
151 }
152
153
154
155 /* TODO: populate these as the vertex is defined:
156 */
157 static void
158 vbo_exec_bind_arrays( struct gl_context *ctx )
159 {
160 struct vbo_context *vbo = vbo_context(ctx);
161 struct vbo_exec_context *exec = &vbo->exec;
162 struct gl_client_array *arrays = exec->vtx.arrays;
163 const GLuint count = exec->vtx.vert_count;
164 const GLuint *map;
165 GLuint attr;
166 GLbitfield varying_inputs = 0x0;
167
168 /* Install the default (ie Current) attributes first, then overlay
169 * all active ones.
170 */
171 switch (get_program_mode(exec->ctx)) {
172 case VP_NONE:
173 for (attr = 0; attr < 16; attr++) {
174 exec->vtx.inputs[attr] = &vbo->legacy_currval[attr];
175 }
176 for (attr = 0; attr < MAT_ATTRIB_MAX; attr++) {
177 ASSERT(attr + 16 < Elements(exec->vtx.inputs));
178 exec->vtx.inputs[attr + 16] = &vbo->mat_currval[attr];
179 }
180 map = vbo->map_vp_none;
181 break;
182 case VP_NV:
183 case VP_ARB:
184 /* The aliasing of attributes for NV vertex programs has already
185 * occurred. NV vertex programs cannot access material values,
186 * nor attributes greater than VERT_ATTRIB_TEX7.
187 */
188 for (attr = 0; attr < 16; attr++) {
189 exec->vtx.inputs[attr] = &vbo->legacy_currval[attr];
190 ASSERT(attr + 16 < Elements(exec->vtx.inputs));
191 exec->vtx.inputs[attr + 16] = &vbo->generic_currval[attr];
192 }
193 map = vbo->map_vp_arb;
194
195 /* check if VERT_ATTRIB_POS is not read but VERT_BIT_GENERIC0 is read.
196 * In that case we effectively need to route the data from
197 * glVertexAttrib(0, val) calls to feed into the GENERIC0 input.
198 */
199 if ((ctx->VertexProgram._Current->Base.InputsRead & VERT_BIT_POS) == 0 &&
200 (ctx->VertexProgram._Current->Base.InputsRead & VERT_BIT_GENERIC0)) {
201 exec->vtx.inputs[16] = exec->vtx.inputs[0];
202 exec->vtx.attrsz[16] = exec->vtx.attrsz[0];
203 exec->vtx.attrptr[16] = exec->vtx.attrptr[0];
204 exec->vtx.attrsz[0] = 0;
205 }
206 break;
207 default:
208 assert(0);
209 }
210
211 /* Make all active attributes (including edgeflag) available as
212 * arrays of floats.
213 */
214 for (attr = 0; attr < VERT_ATTRIB_MAX ; attr++) {
215 const GLuint src = map[attr];
216
217 if (exec->vtx.attrsz[src]) {
218 GLsizeiptr offset = (GLbyte *)exec->vtx.attrptr[src] -
219 (GLbyte *)exec->vtx.vertex;
220
221 /* override the default array set above */
222 ASSERT(attr < Elements(exec->vtx.inputs));
223 ASSERT(attr < Elements(exec->vtx.arrays)); /* arrays[] */
224 exec->vtx.inputs[attr] = &arrays[attr];
225
226 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
227 /* a real buffer obj: Ptr is an offset, not a pointer*/
228 assert(exec->vtx.bufferobj->Pointer); /* buf should be mapped */
229 assert(offset >= 0);
230 arrays[attr].Ptr = (GLubyte *)exec->vtx.bufferobj->Offset + offset;
231 }
232 else {
233 /* Ptr into ordinary app memory */
234 arrays[attr].Ptr = (GLubyte *)exec->vtx.buffer_map + offset;
235 }
236 arrays[attr].Size = exec->vtx.attrsz[src];
237 arrays[attr].StrideB = exec->vtx.vertex_size * sizeof(GLfloat);
238 arrays[attr].Stride = exec->vtx.vertex_size * sizeof(GLfloat);
239 arrays[attr].Type = GL_FLOAT;
240 arrays[attr].Format = GL_RGBA;
241 arrays[attr].Enabled = 1;
242 _mesa_reference_buffer_object(ctx,
243 &arrays[attr].BufferObj,
244 exec->vtx.bufferobj);
245 arrays[attr]._MaxElement = count; /* ??? */
246
247 varying_inputs |= 1 << attr;
248 }
249 }
250
251 _mesa_set_varying_vp_inputs( ctx, varying_inputs );
252 }
253
254
255 static void
256 vbo_exec_vtx_unmap( struct vbo_exec_context *exec )
257 {
258 GLenum target = GL_ARRAY_BUFFER_ARB;
259
260 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
261 struct gl_context *ctx = exec->ctx;
262
263 if (ctx->Driver.FlushMappedBufferRange) {
264 GLintptr offset = exec->vtx.buffer_used - exec->vtx.bufferobj->Offset;
265 GLsizeiptr length = (exec->vtx.buffer_ptr - exec->vtx.buffer_map) * sizeof(float);
266
267 if (length)
268 ctx->Driver.FlushMappedBufferRange(ctx, target,
269 offset, length,
270 exec->vtx.bufferobj);
271 }
272
273 exec->vtx.buffer_used += (exec->vtx.buffer_ptr -
274 exec->vtx.buffer_map) * sizeof(float);
275
276 assert(exec->vtx.buffer_used <= VBO_VERT_BUFFER_SIZE);
277 assert(exec->vtx.buffer_ptr != NULL);
278
279 ctx->Driver.UnmapBuffer(ctx, target, exec->vtx.bufferobj);
280 exec->vtx.buffer_map = NULL;
281 exec->vtx.buffer_ptr = NULL;
282 exec->vtx.max_vert = 0;
283 }
284 }
285
286
287 void
288 vbo_exec_vtx_map( struct vbo_exec_context *exec )
289 {
290 struct gl_context *ctx = exec->ctx;
291 const GLenum target = GL_ARRAY_BUFFER_ARB;
292 const GLenum access = GL_READ_WRITE_ARB; /* for MapBuffer */
293 const GLenum accessRange = GL_MAP_WRITE_BIT | /* for MapBufferRange */
294 GL_MAP_INVALIDATE_RANGE_BIT |
295 GL_MAP_UNSYNCHRONIZED_BIT |
296 GL_MAP_FLUSH_EXPLICIT_BIT |
297 MESA_MAP_NOWAIT_BIT;
298 const GLenum usage = GL_STREAM_DRAW_ARB;
299
300 if (!_mesa_is_bufferobj(exec->vtx.bufferobj))
301 return;
302
303 if (exec->vtx.buffer_map != NULL) {
304 assert(0);
305 exec->vtx.buffer_map = NULL;
306 exec->vtx.buffer_ptr = NULL;
307 }
308
309 if (VBO_VERT_BUFFER_SIZE > exec->vtx.buffer_used + 1024 &&
310 ctx->Driver.MapBufferRange) {
311 exec->vtx.buffer_map =
312 (GLfloat *)ctx->Driver.MapBufferRange(ctx,
313 target,
314 exec->vtx.buffer_used,
315 (VBO_VERT_BUFFER_SIZE -
316 exec->vtx.buffer_used),
317 accessRange,
318 exec->vtx.bufferobj);
319 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
320 }
321
322 if (!exec->vtx.buffer_map) {
323 exec->vtx.buffer_used = 0;
324
325 ctx->Driver.BufferData(ctx, target,
326 VBO_VERT_BUFFER_SIZE,
327 NULL, usage, exec->vtx.bufferobj);
328
329
330 if (ctx->Driver.MapBufferRange)
331 exec->vtx.buffer_map =
332 (GLfloat *)ctx->Driver.MapBufferRange(ctx, target,
333 0, VBO_VERT_BUFFER_SIZE,
334 accessRange,
335 exec->vtx.bufferobj);
336 if (!exec->vtx.buffer_map)
337 exec->vtx.buffer_map =
338 (GLfloat *)ctx->Driver.MapBuffer(ctx, target, access, exec->vtx.bufferobj);
339 assert(exec->vtx.buffer_map);
340 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
341 }
342
343 if (0)
344 printf("map %d..\n", exec->vtx.buffer_used);
345 }
346
347
348
349 /**
350 * Execute the buffer and save copied verts.
351 */
352 void
353 vbo_exec_vtx_flush( struct vbo_exec_context *exec, GLboolean unmap )
354 {
355 if (0)
356 vbo_exec_debug_verts( exec );
357
358 if (exec->vtx.prim_count &&
359 exec->vtx.vert_count) {
360
361 exec->vtx.copied.nr = vbo_copy_vertices( exec );
362
363 if (exec->vtx.copied.nr != exec->vtx.vert_count) {
364 struct gl_context *ctx = exec->ctx;
365
366 /* Before the update_state() as this may raise _NEW_ARRAY
367 * from _mesa_set_varying_vp_inputs().
368 */
369 vbo_exec_bind_arrays( ctx );
370
371 if (ctx->NewState)
372 _mesa_update_state( ctx );
373
374 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
375 vbo_exec_vtx_unmap( exec );
376 }
377
378 if (0)
379 printf("%s %d %d\n", __FUNCTION__, exec->vtx.prim_count,
380 exec->vtx.vert_count);
381
382 vbo_context(ctx)->draw_prims( ctx,
383 exec->vtx.inputs,
384 exec->vtx.prim,
385 exec->vtx.prim_count,
386 NULL,
387 GL_TRUE,
388 0,
389 exec->vtx.vert_count - 1);
390
391 /* If using a real VBO, get new storage -- unless asked not to.
392 */
393 if (_mesa_is_bufferobj(exec->vtx.bufferobj) && !unmap) {
394 vbo_exec_vtx_map( exec );
395 }
396 }
397 }
398
399 /* May have to unmap explicitly if we didn't draw:
400 */
401 if (unmap &&
402 _mesa_is_bufferobj(exec->vtx.bufferobj) &&
403 exec->vtx.buffer_map) {
404 vbo_exec_vtx_unmap( exec );
405 }
406
407
408 if (unmap || exec->vtx.vertex_size == 0)
409 exec->vtx.max_vert = 0;
410 else
411 exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) /
412 (exec->vtx.vertex_size * sizeof(GLfloat)));
413
414 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
415 exec->vtx.prim_count = 0;
416 exec->vtx.vert_count = 0;
417 }
418
419
420 #endif /* FEATURE_beginend */