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