vbo: don't unmap persistent buffer mappings for glBegin/End
[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/arrayobj.h"
31 #include "main/glheader.h"
32 #include "main/bufferobj.h"
33 #include "main/context.h"
34 #include "main/enums.h"
35 #include "main/state.h"
36 #include "main/varray.h"
37 #include "main/vtxfmt.h"
38
39 #include "vbo_noop.h"
40 #include "vbo_private.h"
41
42
43 static void
44 vbo_exec_debug_verts(struct vbo_exec_context *exec)
45 {
46 GLuint count = exec->vtx.vert_count;
47 GLuint i;
48
49 printf("%s: %u vertices %d primitives, %d vertsize\n",
50 __func__,
51 count,
52 exec->vtx.prim_count,
53 exec->vtx.vertex_size);
54
55 for (i = 0 ; i < exec->vtx.prim_count ; i++) {
56 struct _mesa_prim *prim = &exec->vtx.prim[i];
57 printf(" prim %d: %s %d..%d %s %s\n",
58 i,
59 _mesa_lookup_prim_by_nr(prim->mode),
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 unreachable("Unexpected primitive type");
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 gl_vertex_array_object *vao = vbo->VAO;
177 struct vbo_exec_context *exec = &vbo->exec;
178
179 GLintptr buffer_offset;
180 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
181 assert(exec->vtx.bufferobj->Mappings[MAP_INTERNAL].Pointer);
182 buffer_offset = exec->vtx.bufferobj->Mappings[MAP_INTERNAL].Offset +
183 exec->vtx.buffer_offset;
184 } else {
185 /* Ptr into ordinary app memory */
186 buffer_offset = (GLbyte *)exec->vtx.buffer_map - (GLbyte *)NULL;
187 }
188
189 const gl_vertex_processing_mode mode = ctx->VertexProgram._VPMode;
190
191 /* Compute the bitmasks of vao_enabled arrays */
192 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, exec->vtx.enabled);
193
194 /* At first disable arrays no longer needed */
195 _mesa_disable_vertex_array_attribs(ctx, vao, VERT_BIT_ALL & ~vao_enabled);
196 assert((~vao_enabled & vao->Enabled) == 0);
197
198 /* Bind the buffer object */
199 const GLuint stride = exec->vtx.vertex_size*sizeof(GLfloat);
200 _mesa_bind_vertex_buffer(ctx, vao, 0, exec->vtx.bufferobj, buffer_offset,
201 stride);
202
203 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
204 * Note that the position/generic0 aliasing is done in the VAO.
205 */
206 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
207 /* Now set the enabled arrays */
208 GLbitfield mask = vao_enabled;
209 while (mask) {
210 const int vao_attr = u_bit_scan(&mask);
211 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
212
213 const GLubyte size = exec->vtx.attrsz[vbo_attr];
214 const GLenum16 type = exec->vtx.attrtype[vbo_attr];
215 const GLuint offset = (GLuint)((GLbyte *)exec->vtx.attrptr[vbo_attr] -
216 (GLbyte *)exec->vtx.vertex);
217 assert(offset <= ctx->Const.MaxVertexAttribRelativeOffset);
218
219 /* Set and enable */
220 _vbo_set_attrib_format(ctx, vao, vao_attr, buffer_offset,
221 size, type, offset);
222
223 /* The vao is initially created with all bindings set to 0. */
224 assert(vao->VertexAttrib[vao_attr].BufferBindingIndex == 0);
225 }
226 _mesa_enable_vertex_array_attribs(ctx, vao, vao_enabled);
227 assert(vao_enabled == vao->Enabled);
228 assert(!_mesa_is_bufferobj(exec->vtx.bufferobj) ||
229 (vao_enabled & ~vao->VertexAttribBufferMask) == 0);
230
231 _mesa_set_draw_vao(ctx, vao, _vbo_get_vao_filter(mode));
232 }
233
234
235 /**
236 * Unmap the VBO. This is called before drawing.
237 */
238 static void
239 vbo_exec_vtx_unmap(struct vbo_exec_context *exec)
240 {
241 if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
242 struct gl_context *ctx = exec->ctx;
243
244 if (ctx->Driver.FlushMappedBufferRange &&
245 !ctx->Extensions.ARB_buffer_storage) {
246 GLintptr offset = exec->vtx.buffer_used -
247 exec->vtx.bufferobj->Mappings[MAP_INTERNAL].Offset;
248 GLsizeiptr length = (exec->vtx.buffer_ptr - exec->vtx.buffer_map) *
249 sizeof(float);
250
251 if (length)
252 ctx->Driver.FlushMappedBufferRange(ctx, offset, length,
253 exec->vtx.bufferobj,
254 MAP_INTERNAL);
255 }
256
257 exec->vtx.buffer_used += (exec->vtx.buffer_ptr -
258 exec->vtx.buffer_map) * sizeof(float);
259
260 assert(exec->vtx.buffer_used <= VBO_VERT_BUFFER_SIZE);
261 assert(exec->vtx.buffer_ptr != NULL);
262
263 ctx->Driver.UnmapBuffer(ctx, exec->vtx.bufferobj, MAP_INTERNAL);
264 exec->vtx.buffer_map = NULL;
265 exec->vtx.buffer_ptr = NULL;
266 exec->vtx.max_vert = 0;
267 }
268 }
269
270 static bool
271 vbo_exec_buffer_has_space(struct vbo_exec_context *exec)
272 {
273 return VBO_VERT_BUFFER_SIZE > exec->vtx.buffer_used + 1024;
274 }
275
276
277 /**
278 * Map the vertex buffer to begin storing glVertex, glColor, etc data.
279 */
280 void
281 vbo_exec_vtx_map(struct vbo_exec_context *exec)
282 {
283 struct gl_context *ctx = exec->ctx;
284 const GLenum usage = GL_STREAM_DRAW_ARB;
285 GLenum accessRange = GL_MAP_WRITE_BIT | /* for MapBufferRange */
286 GL_MAP_UNSYNCHRONIZED_BIT;
287
288 if (ctx->Extensions.ARB_buffer_storage) {
289 accessRange |= GL_MAP_PERSISTENT_BIT |
290 GL_MAP_COHERENT_BIT;
291 } else {
292 accessRange |= GL_MAP_INVALIDATE_RANGE_BIT |
293 GL_MAP_FLUSH_EXPLICIT_BIT |
294 MESA_MAP_NOWAIT_BIT;
295 }
296
297 if (!_mesa_is_bufferobj(exec->vtx.bufferobj))
298 return;
299
300 assert(!exec->vtx.buffer_map);
301 assert(!exec->vtx.buffer_ptr);
302
303 if (vbo_exec_buffer_has_space(exec)) {
304 /* The VBO exists and there's room for more */
305 if (exec->vtx.bufferobj->Size > 0) {
306 exec->vtx.buffer_map = (fi_type *)
307 ctx->Driver.MapBufferRange(ctx,
308 exec->vtx.buffer_used,
309 VBO_VERT_BUFFER_SIZE
310 - exec->vtx.buffer_used,
311 accessRange,
312 exec->vtx.bufferobj,
313 MAP_INTERNAL);
314 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
315 }
316 else {
317 exec->vtx.buffer_ptr = exec->vtx.buffer_map = NULL;
318 }
319 }
320
321 if (!exec->vtx.buffer_map) {
322 /* Need to allocate a new VBO */
323 exec->vtx.buffer_used = 0;
324
325 if (ctx->Driver.BufferData(ctx, GL_ARRAY_BUFFER_ARB,
326 VBO_VERT_BUFFER_SIZE,
327 NULL, usage,
328 GL_MAP_WRITE_BIT |
329 (ctx->Extensions.ARB_buffer_storage ?
330 GL_MAP_PERSISTENT_BIT |
331 GL_MAP_COHERENT_BIT : 0) |
332 GL_DYNAMIC_STORAGE_BIT |
333 GL_CLIENT_STORAGE_BIT,
334 exec->vtx.bufferobj)) {
335 /* buffer allocation worked, now map the buffer */
336 exec->vtx.buffer_map =
337 (fi_type *)ctx->Driver.MapBufferRange(ctx,
338 0, VBO_VERT_BUFFER_SIZE,
339 accessRange,
340 exec->vtx.bufferobj,
341 MAP_INTERNAL);
342 }
343 else {
344 _mesa_error(ctx, GL_OUT_OF_MEMORY, "VBO allocation");
345 exec->vtx.buffer_map = NULL;
346 }
347 }
348
349 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
350 exec->vtx.buffer_offset = 0;
351
352 if (!exec->vtx.buffer_map) {
353 /* out of memory */
354 _mesa_install_exec_vtxfmt(ctx, &exec->vtxfmt_noop);
355 }
356 else {
357 if (_mesa_using_noop_vtxfmt(ctx->Exec)) {
358 /* The no-op functions are installed so switch back to regular
359 * functions. We do this test just to avoid frequent and needless
360 * calls to _mesa_install_exec_vtxfmt().
361 */
362 _mesa_install_exec_vtxfmt(ctx, &exec->vtxfmt);
363 }
364 }
365
366 if (0)
367 printf("map %d..\n", exec->vtx.buffer_used);
368 }
369
370
371
372 /**
373 * Execute the buffer and save copied verts.
374 * \param unmap if true, leave the VBO unmapped when we're done.
375 */
376 void
377 vbo_exec_vtx_flush(struct vbo_exec_context *exec, GLboolean unmap)
378 {
379 /* Only unmap if persistent mappings are unsupported. */
380 bool persistent_mapping = exec->ctx->Extensions.ARB_buffer_storage &&
381 _mesa_is_bufferobj(exec->vtx.bufferobj) &&
382 exec->vtx.buffer_map;
383 unmap = unmap && !persistent_mapping;
384
385 if (0)
386 vbo_exec_debug_verts(exec);
387
388 if (exec->vtx.prim_count &&
389 exec->vtx.vert_count) {
390
391 exec->vtx.copied.nr = vbo_copy_vertices(exec);
392
393 if (exec->vtx.copied.nr != exec->vtx.vert_count) {
394 struct gl_context *ctx = exec->ctx;
395
396 /* Prepare and set the exec draws internal VAO for drawing. */
397 vbo_exec_bind_arrays(ctx);
398
399 if (ctx->NewState)
400 _mesa_update_state(ctx);
401
402 if (!persistent_mapping)
403 vbo_exec_vtx_unmap(exec);
404
405 assert(ctx->NewState == 0);
406
407 if (0)
408 printf("%s %d %d\n", __func__, exec->vtx.prim_count,
409 exec->vtx.vert_count);
410
411 ctx->Driver.Draw(ctx, exec->vtx.prim, exec->vtx.prim_count,
412 NULL, GL_TRUE, 0, exec->vtx.vert_count - 1,
413 NULL, 0, NULL);
414
415 /* Get new storage -- unless asked not to. */
416 if (!persistent_mapping && !unmap)
417 vbo_exec_vtx_map(exec);
418 }
419 }
420
421 /* May have to unmap explicitly if we didn't draw:
422 */
423 if (unmap && exec->vtx.buffer_map) {
424 vbo_exec_vtx_unmap(exec);
425 }
426
427 if (persistent_mapping) {
428 exec->vtx.buffer_used += (exec->vtx.buffer_ptr - exec->vtx.buffer_map) *
429 sizeof(float);
430 exec->vtx.buffer_map = exec->vtx.buffer_ptr;
431
432 /* Set the buffer offset for the next draw. */
433 exec->vtx.buffer_offset = exec->vtx.buffer_used;
434
435 if (!vbo_exec_buffer_has_space(exec)) {
436 /* This will allocate a new buffer. */
437 vbo_exec_vtx_unmap(exec);
438 vbo_exec_vtx_map(exec);
439 }
440 }
441
442 if (unmap || exec->vtx.vertex_size == 0)
443 exec->vtx.max_vert = 0;
444 else
445 exec->vtx.max_vert = vbo_compute_max_verts(exec);
446
447 exec->vtx.buffer_ptr = exec->vtx.buffer_map;
448 exec->vtx.prim_count = 0;
449 exec->vtx.vert_count = 0;
450 }