2 * Mesa 3-D graphics library
4 * Copyright (C) 1999-2008 Brian Paul All Rights Reserved.
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:
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
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.
25 * Keith Whitwell <keithw@vmware.com>
30 #include "main/glheader.h"
31 #include "main/bufferobj.h"
32 #include "main/context.h"
33 #include "main/enums.h"
34 #include "main/state.h"
35 #include "main/vtxfmt.h"
37 #include "vbo_context.h"
42 vbo_exec_debug_verts(struct vbo_exec_context
*exec
)
44 GLuint count
= exec
->vtx
.vert_count
;
47 printf("%s: %u vertices %d primitives, %d vertsize\n",
51 exec
->vtx
.vertex_size
);
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",
57 _mesa_lookup_prim_by_nr(prim
->mode
),
58 prim
->weak
? " (weak)" : "",
60 prim
->start
+ prim
->count
,
61 prim
->begin
? "BEGIN" : "(wrap)",
62 prim
->end
? "END" : "(wrap)");
68 * Copy zero, one or two vertices from the current vertex buffer into
69 * the temporary "copy" buffer.
70 * This is used when a single primitive overflows a vertex buffer and
71 * we need to continue the primitive in a new vertex buffer.
72 * The temporary "copy" buffer holds the vertices which need to get
73 * copied from the old buffer to the new one.
76 vbo_copy_vertices(struct vbo_exec_context
*exec
)
78 struct _mesa_prim
*last_prim
= &exec
->vtx
.prim
[exec
->vtx
.prim_count
- 1];
79 const GLuint nr
= last_prim
->count
;
81 const GLuint sz
= exec
->vtx
.vertex_size
;
82 fi_type
*dst
= exec
->vtx
.copied
.buffer
;
83 const fi_type
*src
= exec
->vtx
.buffer_map
+ last_prim
->start
* sz
;
85 switch (exec
->ctx
->Driver
.CurrentExecPrimitive
) {
90 for (i
= 0 ; i
< ovf
; i
++)
91 memcpy(dst
+i
*sz
, src
+(nr
-ovf
+i
)*sz
, sz
* sizeof(GLfloat
));
95 for (i
= 0 ; i
< ovf
; i
++)
96 memcpy(dst
+i
*sz
, src
+(nr
-ovf
+i
)*sz
, sz
* sizeof(GLfloat
));
100 for (i
= 0 ; i
< ovf
; i
++)
101 memcpy(dst
+i
*sz
, src
+(nr
-ovf
+i
)*sz
, sz
* sizeof(GLfloat
));
108 memcpy(dst
, src
+(nr
-1)*sz
, sz
* sizeof(GLfloat
));
112 if (last_prim
->begin
== 0) {
113 /* We're dealing with the second or later section of a split/wrapped
114 * GL_LINE_LOOP. Since we're converting line loops to line strips,
115 * we've already increment the last_prim->start counter by one to
116 * skip the 0th vertex in the loop. We need to undo that (effectively
117 * subtract one from last_prim->start) so that we copy the 0th vertex
118 * to the next vertex buffer.
120 assert(last_prim
->start
> 0);
124 case GL_TRIANGLE_FAN
:
130 memcpy(dst
, src
+0, sz
* sizeof(GLfloat
));
134 memcpy(dst
, src
+0, sz
* sizeof(GLfloat
));
135 memcpy(dst
+sz
, src
+(nr
-1)*sz
, sz
* sizeof(GLfloat
));
138 case GL_TRIANGLE_STRIP
:
139 /* no parity issue, but need to make sure the tri is not drawn twice */
156 for (i
= 0 ; i
< ovf
; i
++)
157 memcpy(dst
+i
*sz
, src
+(nr
-ovf
+i
)*sz
, sz
* sizeof(GLfloat
));
159 case PRIM_OUTSIDE_BEGIN_END
:
169 /* TODO: populate these as the vertex is defined:
172 vbo_exec_bind_arrays(struct gl_context
*ctx
)
174 struct vbo_context
*vbo
= vbo_context(ctx
);
175 struct vbo_exec_context
*exec
= &vbo
->exec
;
176 struct gl_vertex_array
*arrays
= exec
->vtx
.arrays
;
179 GLbitfield varying_inputs
= 0x0;
180 bool swap_pos
= false;
182 /* Install the default (ie Current) attributes first */
183 for (attr
= 0; attr
< VERT_ATTRIB_FF_MAX
; attr
++) {
184 exec
->vtx
.inputs
[attr
] = &vbo
->currval
[VBO_ATTRIB_POS
+attr
];
187 /* Overlay other active attributes */
188 switch (get_program_mode(exec
->ctx
)) {
190 for (attr
= 0; attr
< MAT_ATTRIB_MAX
; attr
++) {
191 assert(VERT_ATTRIB_GENERIC(attr
) < ARRAY_SIZE(exec
->vtx
.inputs
));
192 exec
->vtx
.inputs
[VERT_ATTRIB_GENERIC(attr
)] =
193 &vbo
->currval
[VBO_ATTRIB_MAT_FRONT_AMBIENT
+attr
];
195 map
= vbo
->map_vp_none
;
198 for (attr
= 0; attr
< VERT_ATTRIB_GENERIC_MAX
; attr
++) {
199 assert(VERT_ATTRIB_GENERIC(attr
) < ARRAY_SIZE(exec
->vtx
.inputs
));
200 exec
->vtx
.inputs
[VERT_ATTRIB_GENERIC(attr
)] =
201 &vbo
->currval
[VBO_ATTRIB_GENERIC0
+attr
];
203 map
= vbo
->map_vp_arb
;
205 /* check if VERT_ATTRIB_POS is not read but VERT_BIT_GENERIC0 is read.
206 * In that case we effectively need to route the data from
207 * glVertexAttrib(0, val) calls to feed into the GENERIC0 input.
208 * The original state gets essentially restored below.
210 const GLbitfield64 inputs_read
=
211 ctx
->VertexProgram
._Current
->info
.inputs_read
;
212 if ((inputs_read
& VERT_BIT_POS
) == 0 &&
213 (inputs_read
& VERT_BIT_GENERIC0
)) {
215 exec
->vtx
.inputs
[VERT_ATTRIB_GENERIC0
] = exec
->vtx
.inputs
[0];
216 exec
->vtx
.attrsz
[VERT_ATTRIB_GENERIC0
] = exec
->vtx
.attrsz
[0];
217 exec
->vtx
.attrtype
[VERT_ATTRIB_GENERIC0
] = exec
->vtx
.attrtype
[0];
218 exec
->vtx
.attrptr
[VERT_ATTRIB_GENERIC0
] = exec
->vtx
.attrptr
[0];
219 exec
->vtx
.attrsz
[0] = 0;
226 for (attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
227 const GLuint src
= map
[attr
];
229 if (exec
->vtx
.attrsz
[src
]) {
230 GLsizeiptr offset
= (GLbyte
*)exec
->vtx
.attrptr
[src
] -
231 (GLbyte
*)exec
->vtx
.vertex
;
233 /* override the default array set above */
234 assert(attr
< ARRAY_SIZE(exec
->vtx
.inputs
));
235 assert(attr
< ARRAY_SIZE(exec
->vtx
.arrays
)); /* arrays[] */
236 exec
->vtx
.inputs
[attr
] = &arrays
[attr
];
238 if (_mesa_is_bufferobj(exec
->vtx
.bufferobj
)) {
239 /* a real buffer obj: Ptr is an offset, not a pointer */
240 assert(exec
->vtx
.bufferobj
->Mappings
[MAP_INTERNAL
].Pointer
);
242 arrays
[attr
].Ptr
= (GLubyte
*)
243 exec
->vtx
.bufferobj
->Mappings
[MAP_INTERNAL
].Offset
+ offset
;
246 /* Ptr into ordinary app memory */
247 arrays
[attr
].Ptr
= (GLubyte
*)exec
->vtx
.buffer_map
+ offset
;
249 arrays
[attr
].Size
= exec
->vtx
.attrsz
[src
];
250 arrays
[attr
].StrideB
= exec
->vtx
.vertex_size
* sizeof(GLfloat
);
251 arrays
[attr
].Type
= exec
->vtx
.attrtype
[src
];
252 arrays
[attr
].Integer
=
253 vbo_attrtype_to_integer_flag(exec
->vtx
.attrtype
[src
]);
254 arrays
[attr
].Format
= GL_RGBA
;
255 arrays
[attr
]._ElementSize
= arrays
[attr
].Size
* sizeof(GLfloat
);
256 _mesa_reference_buffer_object(ctx
,
257 &arrays
[attr
].BufferObj
,
258 exec
->vtx
.bufferobj
);
260 varying_inputs
|= VERT_BIT(attr
);
264 /* In case we swapped the position and generic0 attribute.
265 * Restore the original setting of the vtx.* variables.
266 * They are still needed with the original order and settings in case
267 * of a split primitive.
270 exec
->vtx
.attrsz
[0] = exec
->vtx
.attrsz
[VERT_ATTRIB_GENERIC0
];
271 exec
->vtx
.attrsz
[VERT_ATTRIB_GENERIC0
] = 0;
274 _mesa_set_varying_vp_inputs(ctx
, varying_inputs
);
275 ctx
->NewDriverState
|= ctx
->DriverFlags
.NewArray
;
280 * Unmap the VBO. This is called before drawing.
283 vbo_exec_vtx_unmap(struct vbo_exec_context
*exec
)
285 if (_mesa_is_bufferobj(exec
->vtx
.bufferobj
)) {
286 struct gl_context
*ctx
= exec
->ctx
;
288 if (ctx
->Driver
.FlushMappedBufferRange
) {
289 GLintptr offset
= exec
->vtx
.buffer_used
-
290 exec
->vtx
.bufferobj
->Mappings
[MAP_INTERNAL
].Offset
;
291 GLsizeiptr length
= (exec
->vtx
.buffer_ptr
- exec
->vtx
.buffer_map
) *
295 ctx
->Driver
.FlushMappedBufferRange(ctx
, offset
, length
,
300 exec
->vtx
.buffer_used
+= (exec
->vtx
.buffer_ptr
-
301 exec
->vtx
.buffer_map
) * sizeof(float);
303 assert(exec
->vtx
.buffer_used
<= VBO_VERT_BUFFER_SIZE
);
304 assert(exec
->vtx
.buffer_ptr
!= NULL
);
306 ctx
->Driver
.UnmapBuffer(ctx
, exec
->vtx
.bufferobj
, MAP_INTERNAL
);
307 exec
->vtx
.buffer_map
= NULL
;
308 exec
->vtx
.buffer_ptr
= NULL
;
309 exec
->vtx
.max_vert
= 0;
315 * Map the vertex buffer to begin storing glVertex, glColor, etc data.
318 vbo_exec_vtx_map(struct vbo_exec_context
*exec
)
320 struct gl_context
*ctx
= exec
->ctx
;
321 const GLenum accessRange
= GL_MAP_WRITE_BIT
| /* for MapBufferRange */
322 GL_MAP_INVALIDATE_RANGE_BIT
|
323 GL_MAP_UNSYNCHRONIZED_BIT
|
324 GL_MAP_FLUSH_EXPLICIT_BIT
|
326 const GLenum usage
= GL_STREAM_DRAW_ARB
;
328 if (!_mesa_is_bufferobj(exec
->vtx
.bufferobj
))
331 assert(!exec
->vtx
.buffer_map
);
332 assert(!exec
->vtx
.buffer_ptr
);
334 if (VBO_VERT_BUFFER_SIZE
> exec
->vtx
.buffer_used
+ 1024) {
335 /* The VBO exists and there's room for more */
336 if (exec
->vtx
.bufferobj
->Size
> 0) {
337 exec
->vtx
.buffer_map
= (fi_type
*)
338 ctx
->Driver
.MapBufferRange(ctx
,
339 exec
->vtx
.buffer_used
,
341 - exec
->vtx
.buffer_used
,
345 exec
->vtx
.buffer_ptr
= exec
->vtx
.buffer_map
;
348 exec
->vtx
.buffer_ptr
= exec
->vtx
.buffer_map
= NULL
;
352 if (!exec
->vtx
.buffer_map
) {
353 /* Need to allocate a new VBO */
354 exec
->vtx
.buffer_used
= 0;
356 if (ctx
->Driver
.BufferData(ctx
, GL_ARRAY_BUFFER_ARB
,
357 VBO_VERT_BUFFER_SIZE
,
360 GL_DYNAMIC_STORAGE_BIT
|
361 GL_CLIENT_STORAGE_BIT
,
362 exec
->vtx
.bufferobj
)) {
363 /* buffer allocation worked, now map the buffer */
364 exec
->vtx
.buffer_map
=
365 (fi_type
*)ctx
->Driver
.MapBufferRange(ctx
,
366 0, VBO_VERT_BUFFER_SIZE
,
372 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "VBO allocation");
373 exec
->vtx
.buffer_map
= NULL
;
377 exec
->vtx
.buffer_ptr
= exec
->vtx
.buffer_map
;
379 if (!exec
->vtx
.buffer_map
) {
381 _mesa_install_exec_vtxfmt(ctx
, &exec
->vtxfmt_noop
);
384 if (_mesa_using_noop_vtxfmt(ctx
->Exec
)) {
385 /* The no-op functions are installed so switch back to regular
386 * functions. We do this test just to avoid frequent and needless
387 * calls to _mesa_install_exec_vtxfmt().
389 _mesa_install_exec_vtxfmt(ctx
, &exec
->vtxfmt
);
394 printf("map %d..\n", exec
->vtx
.buffer_used
);
400 * Execute the buffer and save copied verts.
401 * \param keep_unmapped if true, leave the VBO unmapped when we're done.
404 vbo_exec_vtx_flush(struct vbo_exec_context
*exec
, GLboolean keepUnmapped
)
407 vbo_exec_debug_verts(exec
);
409 if (exec
->vtx
.prim_count
&&
410 exec
->vtx
.vert_count
) {
412 exec
->vtx
.copied
.nr
= vbo_copy_vertices(exec
);
414 if (exec
->vtx
.copied
.nr
!= exec
->vtx
.vert_count
) {
415 struct gl_context
*ctx
= exec
->ctx
;
417 /* Before the update_state() as this may raise _NEW_VARYING_VP_INPUTS
418 * from _mesa_set_varying_vp_inputs().
420 vbo_exec_bind_arrays(ctx
);
423 _mesa_update_state(ctx
);
425 vbo_exec_vtx_unmap(exec
);
428 printf("%s %d %d\n", __func__
, exec
->vtx
.prim_count
,
429 exec
->vtx
.vert_count
);
431 vbo_context(ctx
)->draw_prims(ctx
,
433 exec
->vtx
.prim_count
,
437 exec
->vtx
.vert_count
- 1,
440 /* Get new storage -- unless asked not to. */
442 vbo_exec_vtx_map(exec
);
446 /* May have to unmap explicitly if we didn't draw:
448 if (keepUnmapped
&& exec
->vtx
.buffer_map
) {
449 vbo_exec_vtx_unmap(exec
);
452 if (keepUnmapped
|| exec
->vtx
.vertex_size
== 0)
453 exec
->vtx
.max_vert
= 0;
455 exec
->vtx
.max_vert
= vbo_compute_max_verts(exec
);
457 exec
->vtx
.buffer_ptr
= exec
->vtx
.buffer_map
;
458 exec
->vtx
.prim_count
= 0;
459 exec
->vtx
.vert_count
= 0;