1 /**************************************************************************
3 * Copyright 2011 Marek Olšák <maraeo@gmail.com>
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 * This module uploads user buffers and translates the vertex buffers which
30 * contain incompatible vertices (i.e. not supported by the driver/hardware)
31 * into compatible ones, based on the Gallium CAPs.
33 * It does not upload index buffers.
35 * The module heavily uses bitmasks to represent per-buffer and
36 * per-vertex-element flags to avoid looping over the list of buffers just
37 * to see if there's a non-zero stride, or user buffer, or unsupported format,
40 * There are 3 categories of vertex elements, which are processed separately:
41 * - per-vertex attribs (stride != 0, instance_divisor == 0)
42 * - instanced attribs (stride != 0, instance_divisor > 0)
43 * - constant attribs (stride == 0)
45 * All needed uploads and translations are performed every draw command, but
46 * only the subset of vertices needed for that draw command is uploaded or
47 * translated. (the module never translates whole buffers)
50 * The module consists of two main parts:
53 * 1) Translate (u_vbuf_translate_begin/end)
55 * This is pretty much a vertex fetch fallback. It translates vertices from
56 * one vertex buffer to another in an unused vertex buffer slot. It does
57 * whatever is needed to make the vertices readable by the hardware (changes
58 * vertex formats and aligns offsets and strides). The translate module is
61 * Each of the 3 categories is translated to a separate buffer.
62 * Only the [min_index, max_index] range is translated. For instanced attribs,
63 * the range is [start_instance, start_instance+instance_count]. For constant
64 * attribs, the range is [0, 1].
67 * 2) User buffer uploading (u_vbuf_upload_buffers)
69 * Only the [min_index, max_index] range is uploaded (just like Translate)
70 * with a single memcpy.
72 * This method works best for non-indexed draw operations or indexed draw
73 * operations where the [min_index, max_index] range is not being way bigger
74 * than the vertex count.
76 * If the range is too big (e.g. one triangle with indices {0, 1, 10000}),
77 * the per-vertex attribs are uploaded via the translate module, all packed
78 * into one vertex buffer, and the indexed draw call is turned into
79 * a non-indexed one in the process. This adds additional complexity
80 * to the translate part, but it prevents bad apps from bringing your frame
84 * If there is nothing to do, it forwards every command to the driver.
85 * The module also has its own CSO cache of vertex element states.
88 #include "util/u_vbuf.h"
90 #include "util/u_dump.h"
91 #include "util/u_format.h"
92 #include "util/u_inlines.h"
93 #include "util/u_memory.h"
94 #include "util/u_upload_mgr.h"
95 #include "translate/translate.h"
96 #include "translate/translate_cache.h"
97 #include "cso_cache/cso_cache.h"
98 #include "cso_cache/cso_hash.h"
100 struct u_vbuf_elements
{
102 struct pipe_vertex_element ve
[PIPE_MAX_ATTRIBS
];
104 unsigned src_format_size
[PIPE_MAX_ATTRIBS
];
106 /* If (velem[i].src_format != native_format[i]), the vertex buffer
107 * referenced by the vertex element cannot be used for rendering and
108 * its vertex data must be translated to native_format[i]. */
109 enum pipe_format native_format
[PIPE_MAX_ATTRIBS
];
110 unsigned native_format_size
[PIPE_MAX_ATTRIBS
];
112 /* Which buffers are used by the vertex element state. */
113 uint32_t used_vb_mask
;
114 /* This might mean two things:
115 * - src_format != native_format, as discussed above.
116 * - src_offset % 4 != 0 (if the caps don't allow such an offset). */
117 uint32_t incompatible_elem_mask
; /* each bit describes a corresp. attrib */
118 /* Which buffer has at least one vertex element referencing it
120 uint32_t incompatible_vb_mask_any
;
121 /* Which buffer has all vertex elements referencing it incompatible. */
122 uint32_t incompatible_vb_mask_all
;
123 /* Which buffer has at least one vertex element referencing it
125 uint32_t compatible_vb_mask_any
;
126 /* Which buffer has all vertex elements referencing it compatible. */
127 uint32_t compatible_vb_mask_all
;
129 /* Which buffer has at least one vertex element referencing it
131 uint32_t noninstance_vb_mask_any
;
144 struct u_vbuf_caps caps
;
146 struct pipe_context
*pipe
;
147 struct translate_cache
*translate_cache
;
148 struct cso_cache
*cso_cache
;
149 struct u_upload_mgr
*uploader
;
151 /* This is what was set in set_vertex_buffers.
152 * May contain user buffers. */
153 struct pipe_vertex_buffer vertex_buffer
[PIPE_MAX_ATTRIBS
];
154 uint32_t enabled_vb_mask
;
156 /* Saved vertex buffer. */
157 unsigned aux_vertex_buffer_slot
;
158 struct pipe_vertex_buffer aux_vertex_buffer_saved
;
160 /* Vertex buffers for the driver.
161 * There are usually no user buffers. */
162 struct pipe_vertex_buffer real_vertex_buffer
[PIPE_MAX_ATTRIBS
];
163 uint32_t dirty_real_vb_mask
; /* which buffers are dirty since the last
164 call of set_vertex_buffers */
166 /* The index buffer. */
167 struct pipe_index_buffer index_buffer
;
169 /* Vertex elements. */
170 struct u_vbuf_elements
*ve
, *ve_saved
;
172 /* Vertex elements used for the translate fallback. */
173 struct pipe_vertex_element fallback_velems
[PIPE_MAX_ATTRIBS
];
174 /* If non-NULL, this is a vertex element state used for the translate
175 * fallback and therefore used for rendering too. */
176 boolean using_translate
;
177 /* The vertex buffer slot index where translated vertices have been
179 unsigned fallback_vbs
[VB_NUM
];
181 /* Which buffer is a user buffer. */
182 uint32_t user_vb_mask
; /* each bit describes a corresp. buffer */
183 /* Which buffer is incompatible (unaligned). */
184 uint32_t incompatible_vb_mask
; /* each bit describes a corresp. buffer */
185 /* Which buffer has a non-zero stride. */
186 uint32_t nonzero_stride_vb_mask
; /* each bit describes a corresp. buffer */
190 u_vbuf_create_vertex_elements(struct u_vbuf
*mgr
, unsigned count
,
191 const struct pipe_vertex_element
*attribs
);
192 static void u_vbuf_delete_vertex_elements(struct u_vbuf
*mgr
, void *cso
);
194 static const struct {
195 enum pipe_format from
, to
;
196 } vbuf_format_fallbacks
[] = {
197 { PIPE_FORMAT_R32_FIXED
, PIPE_FORMAT_R32_FLOAT
},
198 { PIPE_FORMAT_R32G32_FIXED
, PIPE_FORMAT_R32G32_FLOAT
},
199 { PIPE_FORMAT_R32G32B32_FIXED
, PIPE_FORMAT_R32G32B32_FLOAT
},
200 { PIPE_FORMAT_R32G32B32A32_FIXED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
201 { PIPE_FORMAT_R16_FLOAT
, PIPE_FORMAT_R32_FLOAT
},
202 { PIPE_FORMAT_R16G16_FLOAT
, PIPE_FORMAT_R32G32_FLOAT
},
203 { PIPE_FORMAT_R16G16B16_FLOAT
, PIPE_FORMAT_R32G32B32_FLOAT
},
204 { PIPE_FORMAT_R16G16B16A16_FLOAT
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
205 { PIPE_FORMAT_R64_FLOAT
, PIPE_FORMAT_R32_FLOAT
},
206 { PIPE_FORMAT_R64G64_FLOAT
, PIPE_FORMAT_R32G32_FLOAT
},
207 { PIPE_FORMAT_R64G64B64_FLOAT
, PIPE_FORMAT_R32G32B32_FLOAT
},
208 { PIPE_FORMAT_R64G64B64A64_FLOAT
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
209 { PIPE_FORMAT_R32_UNORM
, PIPE_FORMAT_R32_FLOAT
},
210 { PIPE_FORMAT_R32G32_UNORM
, PIPE_FORMAT_R32G32_FLOAT
},
211 { PIPE_FORMAT_R32G32B32_UNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
212 { PIPE_FORMAT_R32G32B32A32_UNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
213 { PIPE_FORMAT_R32_SNORM
, PIPE_FORMAT_R32_FLOAT
},
214 { PIPE_FORMAT_R32G32_SNORM
, PIPE_FORMAT_R32G32_FLOAT
},
215 { PIPE_FORMAT_R32G32B32_SNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
216 { PIPE_FORMAT_R32G32B32A32_SNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
217 { PIPE_FORMAT_R32_USCALED
, PIPE_FORMAT_R32_FLOAT
},
218 { PIPE_FORMAT_R32G32_USCALED
, PIPE_FORMAT_R32G32_FLOAT
},
219 { PIPE_FORMAT_R32G32B32_USCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
220 { PIPE_FORMAT_R32G32B32A32_USCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
221 { PIPE_FORMAT_R32_SSCALED
, PIPE_FORMAT_R32_FLOAT
},
222 { PIPE_FORMAT_R32G32_SSCALED
, PIPE_FORMAT_R32G32_FLOAT
},
223 { PIPE_FORMAT_R32G32B32_SSCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
224 { PIPE_FORMAT_R32G32B32A32_SSCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
225 { PIPE_FORMAT_R16_UNORM
, PIPE_FORMAT_R32_FLOAT
},
226 { PIPE_FORMAT_R16G16_UNORM
, PIPE_FORMAT_R32G32_FLOAT
},
227 { PIPE_FORMAT_R16G16B16_UNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
228 { PIPE_FORMAT_R16G16B16A16_UNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
229 { PIPE_FORMAT_R16_SNORM
, PIPE_FORMAT_R32_FLOAT
},
230 { PIPE_FORMAT_R16G16_SNORM
, PIPE_FORMAT_R32G32_FLOAT
},
231 { PIPE_FORMAT_R16G16B16_SNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
232 { PIPE_FORMAT_R16G16B16A16_SNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
233 { PIPE_FORMAT_R16_USCALED
, PIPE_FORMAT_R32_FLOAT
},
234 { PIPE_FORMAT_R16G16_USCALED
, PIPE_FORMAT_R32G32_FLOAT
},
235 { PIPE_FORMAT_R16G16B16_USCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
236 { PIPE_FORMAT_R16G16B16A16_USCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
237 { PIPE_FORMAT_R16_SSCALED
, PIPE_FORMAT_R32_FLOAT
},
238 { PIPE_FORMAT_R16G16_SSCALED
, PIPE_FORMAT_R32G32_FLOAT
},
239 { PIPE_FORMAT_R16G16B16_SSCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
240 { PIPE_FORMAT_R16G16B16A16_SSCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
241 { PIPE_FORMAT_R8_UNORM
, PIPE_FORMAT_R32_FLOAT
},
242 { PIPE_FORMAT_R8G8_UNORM
, PIPE_FORMAT_R32G32_FLOAT
},
243 { PIPE_FORMAT_R8G8B8_UNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
244 { PIPE_FORMAT_R8G8B8A8_UNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
245 { PIPE_FORMAT_R8_SNORM
, PIPE_FORMAT_R32_FLOAT
},
246 { PIPE_FORMAT_R8G8_SNORM
, PIPE_FORMAT_R32G32_FLOAT
},
247 { PIPE_FORMAT_R8G8B8_SNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
248 { PIPE_FORMAT_R8G8B8A8_SNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
249 { PIPE_FORMAT_R8_USCALED
, PIPE_FORMAT_R32_FLOAT
},
250 { PIPE_FORMAT_R8G8_USCALED
, PIPE_FORMAT_R32G32_FLOAT
},
251 { PIPE_FORMAT_R8G8B8_USCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
252 { PIPE_FORMAT_R8G8B8A8_USCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
253 { PIPE_FORMAT_R8_SSCALED
, PIPE_FORMAT_R32_FLOAT
},
254 { PIPE_FORMAT_R8G8_SSCALED
, PIPE_FORMAT_R32G32_FLOAT
},
255 { PIPE_FORMAT_R8G8B8_SSCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
256 { PIPE_FORMAT_R8G8B8A8_SSCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
259 boolean
u_vbuf_get_caps(struct pipe_screen
*screen
, struct u_vbuf_caps
*caps
)
262 boolean fallback
= FALSE
;
264 /* I'd rather have a bitfield of which formats are supported and a static
265 * table of the translations indexed by format, but since we don't have C99
266 * we can't easily make a sparsely-populated table indexed by format. So,
267 * we construct the sparse table here.
269 for (i
= 0; i
< PIPE_FORMAT_COUNT
; i
++)
270 caps
->format_translation
[i
] = i
;
272 for (i
= 0; i
< Elements(vbuf_format_fallbacks
); i
++) {
273 enum pipe_format format
= vbuf_format_fallbacks
[i
].from
;
275 if (!screen
->is_format_supported(screen
, format
, PIPE_BUFFER
, 0,
276 PIPE_BIND_VERTEX_BUFFER
)) {
277 caps
->format_translation
[format
] = vbuf_format_fallbacks
[i
].to
;
282 caps
->buffer_offset_unaligned
=
283 !screen
->get_param(screen
,
284 PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
);
285 caps
->buffer_stride_unaligned
=
286 !screen
->get_param(screen
,
287 PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
);
288 caps
->velem_src_offset_unaligned
=
289 !screen
->get_param(screen
,
290 PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
);
291 caps
->user_vertex_buffers
=
292 screen
->get_param(screen
, PIPE_CAP_USER_VERTEX_BUFFERS
);
294 if (!caps
->buffer_offset_unaligned
||
295 !caps
->buffer_stride_unaligned
||
296 !caps
->velem_src_offset_unaligned
||
297 !caps
->user_vertex_buffers
) {
305 u_vbuf_create(struct pipe_context
*pipe
,
306 struct u_vbuf_caps
*caps
, unsigned aux_vertex_buffer_index
)
308 struct u_vbuf
*mgr
= CALLOC_STRUCT(u_vbuf
);
311 mgr
->aux_vertex_buffer_slot
= aux_vertex_buffer_index
;
313 mgr
->cso_cache
= cso_cache_create();
314 mgr
->translate_cache
= translate_cache_create();
315 memset(mgr
->fallback_vbs
, ~0, sizeof(mgr
->fallback_vbs
));
317 mgr
->uploader
= u_upload_create(pipe
, 1024 * 1024, 4,
318 PIPE_BIND_VERTEX_BUFFER
);
323 /* u_vbuf uses its own caching for vertex elements, because it needs to keep
324 * its own preprocessed state per vertex element CSO. */
325 static struct u_vbuf_elements
*
326 u_vbuf_set_vertex_elements_internal(struct u_vbuf
*mgr
, unsigned count
,
327 const struct pipe_vertex_element
*states
)
329 struct pipe_context
*pipe
= mgr
->pipe
;
330 unsigned key_size
, hash_key
;
331 struct cso_hash_iter iter
;
332 struct u_vbuf_elements
*ve
;
333 struct cso_velems_state velems_state
;
335 /* need to include the count into the stored state data too. */
336 key_size
= sizeof(struct pipe_vertex_element
) * count
+ sizeof(unsigned);
337 velems_state
.count
= count
;
338 memcpy(velems_state
.velems
, states
,
339 sizeof(struct pipe_vertex_element
) * count
);
340 hash_key
= cso_construct_key((void*)&velems_state
, key_size
);
341 iter
= cso_find_state_template(mgr
->cso_cache
, hash_key
, CSO_VELEMENTS
,
342 (void*)&velems_state
, key_size
);
344 if (cso_hash_iter_is_null(iter
)) {
345 struct cso_velements
*cso
= MALLOC_STRUCT(cso_velements
);
346 memcpy(&cso
->state
, &velems_state
, key_size
);
347 cso
->data
= u_vbuf_create_vertex_elements(mgr
, count
, states
);
348 cso
->delete_state
= (cso_state_callback
)u_vbuf_delete_vertex_elements
;
349 cso
->context
= (void*)mgr
;
351 iter
= cso_insert_state(mgr
->cso_cache
, hash_key
, CSO_VELEMENTS
, cso
);
354 ve
= ((struct cso_velements
*)cso_hash_iter_data(iter
))->data
;
360 pipe
->bind_vertex_elements_state(pipe
, ve
->driver_cso
);
365 void u_vbuf_set_vertex_elements(struct u_vbuf
*mgr
, unsigned count
,
366 const struct pipe_vertex_element
*states
)
368 mgr
->ve
= u_vbuf_set_vertex_elements_internal(mgr
, count
, states
);
371 void u_vbuf_destroy(struct u_vbuf
*mgr
)
373 struct pipe_screen
*screen
= mgr
->pipe
->screen
;
375 unsigned num_vb
= screen
->get_shader_param(screen
, PIPE_SHADER_VERTEX
,
376 PIPE_SHADER_CAP_MAX_INPUTS
);
378 mgr
->pipe
->set_index_buffer(mgr
->pipe
, NULL
);
379 pipe_resource_reference(&mgr
->index_buffer
.buffer
, NULL
);
381 mgr
->pipe
->set_vertex_buffers(mgr
->pipe
, 0, num_vb
, NULL
);
383 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
384 pipe_resource_reference(&mgr
->vertex_buffer
[i
].buffer
, NULL
);
386 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
387 pipe_resource_reference(&mgr
->real_vertex_buffer
[i
].buffer
, NULL
);
389 pipe_resource_reference(&mgr
->aux_vertex_buffer_saved
.buffer
, NULL
);
391 translate_cache_destroy(mgr
->translate_cache
);
392 u_upload_destroy(mgr
->uploader
);
393 cso_cache_delete(mgr
->cso_cache
);
397 static enum pipe_error
398 u_vbuf_translate_buffers(struct u_vbuf
*mgr
, struct translate_key
*key
,
399 unsigned vb_mask
, unsigned out_vb
,
400 int start_vertex
, unsigned num_vertices
,
401 int start_index
, unsigned num_indices
, int min_index
,
402 boolean unroll_indices
)
404 struct translate
*tr
;
405 struct pipe_transfer
*vb_transfer
[PIPE_MAX_ATTRIBS
] = {0};
406 struct pipe_resource
*out_buffer
= NULL
;
408 unsigned out_offset
, mask
;
410 /* Get a translate object. */
411 tr
= translate_cache_find(mgr
->translate_cache
, key
);
413 /* Map buffers we want to translate. */
416 struct pipe_vertex_buffer
*vb
;
419 unsigned i
= u_bit_scan(&mask
);
421 vb
= &mgr
->vertex_buffer
[i
];
422 offset
= vb
->buffer_offset
+ vb
->stride
* start_vertex
;
424 if (vb
->user_buffer
) {
425 map
= (uint8_t*)vb
->user_buffer
+ offset
;
427 unsigned size
= vb
->stride
? num_vertices
* vb
->stride
430 if (offset
+size
> vb
->buffer
->width0
) {
431 size
= vb
->buffer
->width0
- offset
;
434 map
= pipe_buffer_map_range(mgr
->pipe
, vb
->buffer
, offset
, size
,
435 PIPE_TRANSFER_READ
, &vb_transfer
[i
]);
438 /* Subtract min_index so that indexing with the index buffer works. */
439 if (unroll_indices
) {
440 map
-= (ptrdiff_t)vb
->stride
* min_index
;
443 tr
->set_buffer(tr
, i
, map
, vb
->stride
, ~0);
447 if (unroll_indices
) {
448 struct pipe_index_buffer
*ib
= &mgr
->index_buffer
;
449 struct pipe_transfer
*transfer
= NULL
;
450 unsigned offset
= ib
->offset
+ start_index
* ib
->index_size
;
453 assert((ib
->buffer
|| ib
->user_buffer
) && ib
->index_size
);
455 /* Create and map the output buffer. */
456 u_upload_alloc(mgr
->uploader
, 0,
457 key
->output_stride
* num_indices
,
458 &out_offset
, &out_buffer
,
461 return PIPE_ERROR_OUT_OF_MEMORY
;
463 if (ib
->user_buffer
) {
464 map
= (uint8_t*)ib
->user_buffer
+ offset
;
466 map
= pipe_buffer_map_range(mgr
->pipe
, ib
->buffer
, offset
,
467 num_indices
* ib
->index_size
,
468 PIPE_TRANSFER_READ
, &transfer
);
471 switch (ib
->index_size
) {
473 tr
->run_elts(tr
, (unsigned*)map
, num_indices
, 0, 0, out_map
);
476 tr
->run_elts16(tr
, (uint16_t*)map
, num_indices
, 0, 0, out_map
);
479 tr
->run_elts8(tr
, map
, num_indices
, 0, 0, out_map
);
484 pipe_buffer_unmap(mgr
->pipe
, transfer
);
487 /* Create and map the output buffer. */
488 u_upload_alloc(mgr
->uploader
,
489 key
->output_stride
* start_vertex
,
490 key
->output_stride
* num_vertices
,
491 &out_offset
, &out_buffer
,
494 return PIPE_ERROR_OUT_OF_MEMORY
;
496 out_offset
-= key
->output_stride
* start_vertex
;
498 tr
->run(tr
, 0, num_vertices
, 0, 0, out_map
);
501 /* Unmap all buffers. */
504 unsigned i
= u_bit_scan(&mask
);
506 if (vb_transfer
[i
]) {
507 pipe_buffer_unmap(mgr
->pipe
, vb_transfer
[i
]);
511 /* Setup the new vertex buffer. */
512 mgr
->real_vertex_buffer
[out_vb
].buffer_offset
= out_offset
;
513 mgr
->real_vertex_buffer
[out_vb
].stride
= key
->output_stride
;
515 /* Move the buffer reference. */
516 pipe_resource_reference(
517 &mgr
->real_vertex_buffer
[out_vb
].buffer
, NULL
);
518 mgr
->real_vertex_buffer
[out_vb
].buffer
= out_buffer
;
524 u_vbuf_translate_find_free_vb_slots(struct u_vbuf
*mgr
,
525 unsigned mask
[VB_NUM
])
528 unsigned fallback_vbs
[VB_NUM
];
529 /* Set the bit for each buffer which is incompatible, or isn't set. */
530 uint32_t unused_vb_mask
=
531 mgr
->ve
->incompatible_vb_mask_all
| mgr
->incompatible_vb_mask
|
532 ~mgr
->enabled_vb_mask
;
534 memset(fallback_vbs
, ~0, sizeof(fallback_vbs
));
536 /* Find free slots for each type if needed. */
537 for (type
= 0; type
< VB_NUM
; type
++) {
541 if (!unused_vb_mask
) {
545 index
= ffs(unused_vb_mask
) - 1;
546 fallback_vbs
[type
] = index
;
547 /*printf("found slot=%i for type=%i\n", index, type);*/
551 for (type
= 0; type
< VB_NUM
; type
++) {
553 mgr
->dirty_real_vb_mask
|= 1 << fallback_vbs
[type
];
557 memcpy(mgr
->fallback_vbs
, fallback_vbs
, sizeof(fallback_vbs
));
562 u_vbuf_translate_begin(struct u_vbuf
*mgr
,
563 int start_vertex
, unsigned num_vertices
,
564 int start_instance
, unsigned num_instances
,
565 int start_index
, unsigned num_indices
, int min_index
,
566 boolean unroll_indices
)
568 unsigned mask
[VB_NUM
] = {0};
569 struct translate_key key
[VB_NUM
];
570 unsigned elem_index
[VB_NUM
][PIPE_MAX_ATTRIBS
]; /* ... into key.elements */
572 unsigned incompatible_vb_mask
= mgr
->incompatible_vb_mask
&
573 mgr
->ve
->used_vb_mask
;
575 int start
[VB_NUM
] = {
576 start_vertex
, /* VERTEX */
577 start_instance
, /* INSTANCE */
581 unsigned num
[VB_NUM
] = {
582 num_vertices
, /* VERTEX */
583 num_instances
, /* INSTANCE */
587 memset(key
, 0, sizeof(key
));
588 memset(elem_index
, ~0, sizeof(elem_index
));
590 /* See if there are vertex attribs of each type to translate and
592 for (i
= 0; i
< mgr
->ve
->count
; i
++) {
593 unsigned vb_index
= mgr
->ve
->ve
[i
].vertex_buffer_index
;
595 if (!mgr
->vertex_buffer
[vb_index
].stride
) {
596 if (!(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
597 !(incompatible_vb_mask
& (1 << vb_index
))) {
600 mask
[VB_CONST
] |= 1 << vb_index
;
601 } else if (mgr
->ve
->ve
[i
].instance_divisor
) {
602 if (!(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
603 !(incompatible_vb_mask
& (1 << vb_index
))) {
606 mask
[VB_INSTANCE
] |= 1 << vb_index
;
608 if (!unroll_indices
&&
609 !(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
610 !(incompatible_vb_mask
& (1 << vb_index
))) {
613 mask
[VB_VERTEX
] |= 1 << vb_index
;
617 assert(mask
[VB_VERTEX
] || mask
[VB_INSTANCE
] || mask
[VB_CONST
]);
619 /* Find free vertex buffer slots. */
620 if (!u_vbuf_translate_find_free_vb_slots(mgr
, mask
)) {
624 /* Initialize the translate keys. */
625 for (i
= 0; i
< mgr
->ve
->count
; i
++) {
626 struct translate_key
*k
;
627 struct translate_element
*te
;
628 unsigned bit
, vb_index
= mgr
->ve
->ve
[i
].vertex_buffer_index
;
631 if (!(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
632 !(incompatible_vb_mask
& (1 << vb_index
)) &&
633 (!unroll_indices
|| !(mask
[VB_VERTEX
] & bit
))) {
637 /* Set type to what we will translate.
638 * Whether vertex, instance, or constant attribs. */
639 for (type
= 0; type
< VB_NUM
; type
++) {
640 if (mask
[type
] & bit
) {
644 assert(type
< VB_NUM
);
645 assert(translate_is_output_format_supported(mgr
->ve
->native_format
[i
]));
646 /*printf("velem=%i type=%i\n", i, type);*/
648 /* Add the vertex element. */
650 elem_index
[type
][i
] = k
->nr_elements
;
652 te
= &k
->element
[k
->nr_elements
];
653 te
->type
= TRANSLATE_ELEMENT_NORMAL
;
654 te
->instance_divisor
= 0;
655 te
->input_buffer
= vb_index
;
656 te
->input_format
= mgr
->ve
->ve
[i
].src_format
;
657 te
->input_offset
= mgr
->ve
->ve
[i
].src_offset
;
658 te
->output_format
= mgr
->ve
->native_format
[i
];
659 te
->output_offset
= k
->output_stride
;
661 k
->output_stride
+= mgr
->ve
->native_format_size
[i
];
665 /* Translate buffers. */
666 for (type
= 0; type
< VB_NUM
; type
++) {
667 if (key
[type
].nr_elements
) {
669 err
= u_vbuf_translate_buffers(mgr
, &key
[type
], mask
[type
],
670 mgr
->fallback_vbs
[type
],
671 start
[type
], num
[type
],
672 start_index
, num_indices
, min_index
,
673 unroll_indices
&& type
== VB_VERTEX
);
677 /* Fixup the stride for constant attribs. */
678 if (type
== VB_CONST
) {
679 mgr
->real_vertex_buffer
[mgr
->fallback_vbs
[VB_CONST
]].stride
= 0;
684 /* Setup new vertex elements. */
685 for (i
= 0; i
< mgr
->ve
->count
; i
++) {
686 for (type
= 0; type
< VB_NUM
; type
++) {
687 if (elem_index
[type
][i
] < key
[type
].nr_elements
) {
688 struct translate_element
*te
= &key
[type
].element
[elem_index
[type
][i
]];
689 mgr
->fallback_velems
[i
].instance_divisor
= mgr
->ve
->ve
[i
].instance_divisor
;
690 mgr
->fallback_velems
[i
].src_format
= te
->output_format
;
691 mgr
->fallback_velems
[i
].src_offset
= te
->output_offset
;
692 mgr
->fallback_velems
[i
].vertex_buffer_index
= mgr
->fallback_vbs
[type
];
694 /* elem_index[type][i] can only be set for one type. */
695 assert(type
> VB_INSTANCE
|| elem_index
[type
+1][i
] == ~0);
696 assert(type
> VB_VERTEX
|| elem_index
[type
+2][i
] == ~0);
700 /* No translating, just copy the original vertex element over. */
701 if (type
== VB_NUM
) {
702 memcpy(&mgr
->fallback_velems
[i
], &mgr
->ve
->ve
[i
],
703 sizeof(struct pipe_vertex_element
));
707 u_vbuf_set_vertex_elements_internal(mgr
, mgr
->ve
->count
,
708 mgr
->fallback_velems
);
709 mgr
->using_translate
= TRUE
;
713 static void u_vbuf_translate_end(struct u_vbuf
*mgr
)
717 /* Restore vertex elements. */
718 mgr
->pipe
->bind_vertex_elements_state(mgr
->pipe
, mgr
->ve
->driver_cso
);
719 mgr
->using_translate
= FALSE
;
721 /* Unreference the now-unused VBOs. */
722 for (i
= 0; i
< VB_NUM
; i
++) {
723 unsigned vb
= mgr
->fallback_vbs
[i
];
725 pipe_resource_reference(&mgr
->real_vertex_buffer
[vb
].buffer
, NULL
);
726 mgr
->fallback_vbs
[i
] = ~0;
728 /* This will cause the buffer to be unbound in the driver later. */
729 mgr
->dirty_real_vb_mask
|= 1 << vb
;
735 u_vbuf_create_vertex_elements(struct u_vbuf
*mgr
, unsigned count
,
736 const struct pipe_vertex_element
*attribs
)
738 struct pipe_context
*pipe
= mgr
->pipe
;
740 struct pipe_vertex_element driver_attribs
[PIPE_MAX_ATTRIBS
];
741 struct u_vbuf_elements
*ve
= CALLOC_STRUCT(u_vbuf_elements
);
742 uint32_t used_buffers
= 0;
746 memcpy(ve
->ve
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
747 memcpy(driver_attribs
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
749 /* Set the best native format in case the original format is not
751 for (i
= 0; i
< count
; i
++) {
752 enum pipe_format format
= ve
->ve
[i
].src_format
;
754 ve
->src_format_size
[i
] = util_format_get_blocksize(format
);
756 used_buffers
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
758 if (!ve
->ve
[i
].instance_divisor
) {
759 ve
->noninstance_vb_mask_any
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
762 format
= mgr
->caps
.format_translation
[format
];
764 driver_attribs
[i
].src_format
= format
;
765 ve
->native_format
[i
] = format
;
766 ve
->native_format_size
[i
] =
767 util_format_get_blocksize(ve
->native_format
[i
]);
769 if (ve
->ve
[i
].src_format
!= format
||
770 (!mgr
->caps
.velem_src_offset_unaligned
&&
771 ve
->ve
[i
].src_offset
% 4 != 0)) {
772 ve
->incompatible_elem_mask
|= 1 << i
;
773 ve
->incompatible_vb_mask_any
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
775 ve
->compatible_vb_mask_any
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
779 ve
->used_vb_mask
= used_buffers
;
780 ve
->compatible_vb_mask_all
= ~ve
->incompatible_vb_mask_any
& used_buffers
;
781 ve
->incompatible_vb_mask_all
= ~ve
->compatible_vb_mask_any
& used_buffers
;
783 /* Align the formats and offsets to the size of DWORD if needed. */
784 if (!mgr
->caps
.velem_src_offset_unaligned
) {
785 for (i
= 0; i
< count
; i
++) {
786 ve
->native_format_size
[i
] = align(ve
->native_format_size
[i
], 4);
787 driver_attribs
[i
].src_offset
= align(ve
->ve
[i
].src_offset
, 4);
792 pipe
->create_vertex_elements_state(pipe
, count
, driver_attribs
);
796 static void u_vbuf_delete_vertex_elements(struct u_vbuf
*mgr
, void *cso
)
798 struct pipe_context
*pipe
= mgr
->pipe
;
799 struct u_vbuf_elements
*ve
= cso
;
801 pipe
->delete_vertex_elements_state(pipe
, ve
->driver_cso
);
805 void u_vbuf_set_vertex_buffers(struct u_vbuf
*mgr
,
806 unsigned start_slot
, unsigned count
,
807 const struct pipe_vertex_buffer
*bufs
)
810 /* which buffers are enabled */
811 uint32_t enabled_vb_mask
= 0;
812 /* which buffers are in user memory */
813 uint32_t user_vb_mask
= 0;
814 /* which buffers are incompatible with the driver */
815 uint32_t incompatible_vb_mask
= 0;
816 /* which buffers have a non-zero stride */
817 uint32_t nonzero_stride_vb_mask
= 0;
818 uint32_t mask
= ~(((1ull << count
) - 1) << start_slot
);
820 /* Zero out the bits we are going to rewrite completely. */
821 mgr
->user_vb_mask
&= mask
;
822 mgr
->incompatible_vb_mask
&= mask
;
823 mgr
->nonzero_stride_vb_mask
&= mask
;
824 mgr
->enabled_vb_mask
&= mask
;
827 struct pipe_context
*pipe
= mgr
->pipe
;
829 mgr
->dirty_real_vb_mask
&= mask
;
831 for (i
= 0; i
< count
; i
++) {
832 unsigned dst_index
= start_slot
+ i
;
834 pipe_resource_reference(&mgr
->vertex_buffer
[dst_index
].buffer
, NULL
);
835 pipe_resource_reference(&mgr
->real_vertex_buffer
[dst_index
].buffer
,
839 pipe
->set_vertex_buffers(pipe
, start_slot
, count
, NULL
);
843 for (i
= 0; i
< count
; i
++) {
844 unsigned dst_index
= start_slot
+ i
;
845 const struct pipe_vertex_buffer
*vb
= &bufs
[i
];
846 struct pipe_vertex_buffer
*orig_vb
= &mgr
->vertex_buffer
[dst_index
];
847 struct pipe_vertex_buffer
*real_vb
= &mgr
->real_vertex_buffer
[dst_index
];
849 if (!vb
->buffer
&& !vb
->user_buffer
) {
850 pipe_resource_reference(&orig_vb
->buffer
, NULL
);
851 pipe_resource_reference(&real_vb
->buffer
, NULL
);
852 real_vb
->user_buffer
= NULL
;
856 pipe_resource_reference(&orig_vb
->buffer
, vb
->buffer
);
857 orig_vb
->user_buffer
= vb
->user_buffer
;
859 real_vb
->buffer_offset
= orig_vb
->buffer_offset
= vb
->buffer_offset
;
860 real_vb
->stride
= orig_vb
->stride
= vb
->stride
;
863 nonzero_stride_vb_mask
|= 1 << dst_index
;
865 enabled_vb_mask
|= 1 << dst_index
;
867 if ((!mgr
->caps
.buffer_offset_unaligned
&& vb
->buffer_offset
% 4 != 0) ||
868 (!mgr
->caps
.buffer_stride_unaligned
&& vb
->stride
% 4 != 0)) {
869 incompatible_vb_mask
|= 1 << dst_index
;
870 pipe_resource_reference(&real_vb
->buffer
, NULL
);
874 if (!mgr
->caps
.user_vertex_buffers
&& vb
->user_buffer
) {
875 user_vb_mask
|= 1 << dst_index
;
876 pipe_resource_reference(&real_vb
->buffer
, NULL
);
880 pipe_resource_reference(&real_vb
->buffer
, vb
->buffer
);
881 real_vb
->user_buffer
= vb
->user_buffer
;
884 mgr
->user_vb_mask
|= user_vb_mask
;
885 mgr
->incompatible_vb_mask
|= incompatible_vb_mask
;
886 mgr
->nonzero_stride_vb_mask
|= nonzero_stride_vb_mask
;
887 mgr
->enabled_vb_mask
|= enabled_vb_mask
;
889 /* All changed buffers are marked as dirty, even the NULL ones,
890 * which will cause the NULL buffers to be unbound in the driver later. */
891 mgr
->dirty_real_vb_mask
|= ~mask
;
894 void u_vbuf_set_index_buffer(struct u_vbuf
*mgr
,
895 const struct pipe_index_buffer
*ib
)
897 struct pipe_context
*pipe
= mgr
->pipe
;
900 assert(ib
->offset
% ib
->index_size
== 0);
901 pipe_resource_reference(&mgr
->index_buffer
.buffer
, ib
->buffer
);
902 memcpy(&mgr
->index_buffer
, ib
, sizeof(*ib
));
904 pipe_resource_reference(&mgr
->index_buffer
.buffer
, NULL
);
907 pipe
->set_index_buffer(pipe
, ib
);
910 static enum pipe_error
911 u_vbuf_upload_buffers(struct u_vbuf
*mgr
,
912 int start_vertex
, unsigned num_vertices
,
913 int start_instance
, unsigned num_instances
)
916 unsigned nr_velems
= mgr
->ve
->count
;
917 struct pipe_vertex_element
*velems
=
918 mgr
->using_translate
? mgr
->fallback_velems
: mgr
->ve
->ve
;
919 unsigned start_offset
[PIPE_MAX_ATTRIBS
];
920 unsigned end_offset
[PIPE_MAX_ATTRIBS
];
921 uint32_t buffer_mask
= 0;
923 /* Determine how much data needs to be uploaded. */
924 for (i
= 0; i
< nr_velems
; i
++) {
925 struct pipe_vertex_element
*velem
= &velems
[i
];
926 unsigned index
= velem
->vertex_buffer_index
;
927 struct pipe_vertex_buffer
*vb
= &mgr
->vertex_buffer
[index
];
928 unsigned instance_div
, first
, size
, index_bit
;
930 /* Skip the buffers generated by translate. */
931 if (index
== mgr
->fallback_vbs
[VB_VERTEX
] ||
932 index
== mgr
->fallback_vbs
[VB_INSTANCE
] ||
933 index
== mgr
->fallback_vbs
[VB_CONST
]) {
937 if (!vb
->user_buffer
) {
941 instance_div
= velem
->instance_divisor
;
942 first
= vb
->buffer_offset
+ velem
->src_offset
;
945 /* Constant attrib. */
946 size
= mgr
->ve
->src_format_size
[i
];
947 } else if (instance_div
) {
948 /* Per-instance attrib. */
949 unsigned count
= (num_instances
+ instance_div
- 1) / instance_div
;
950 first
+= vb
->stride
* start_instance
;
951 size
= vb
->stride
* (count
- 1) + mgr
->ve
->src_format_size
[i
];
953 /* Per-vertex attrib. */
954 first
+= vb
->stride
* start_vertex
;
955 size
= vb
->stride
* (num_vertices
- 1) + mgr
->ve
->src_format_size
[i
];
958 index_bit
= 1 << index
;
960 /* Update offsets. */
961 if (!(buffer_mask
& index_bit
)) {
962 start_offset
[index
] = first
;
963 end_offset
[index
] = first
+ size
;
965 if (first
< start_offset
[index
])
966 start_offset
[index
] = first
;
967 if (first
+ size
> end_offset
[index
])
968 end_offset
[index
] = first
+ size
;
971 buffer_mask
|= index_bit
;
974 /* Upload buffers. */
975 while (buffer_mask
) {
977 struct pipe_vertex_buffer
*real_vb
;
980 i
= u_bit_scan(&buffer_mask
);
982 start
= start_offset
[i
];
986 real_vb
= &mgr
->real_vertex_buffer
[i
];
987 ptr
= mgr
->vertex_buffer
[i
].user_buffer
;
989 u_upload_data(mgr
->uploader
, start
, end
- start
, ptr
+ start
,
990 &real_vb
->buffer_offset
, &real_vb
->buffer
);
991 if (!real_vb
->buffer
)
992 return PIPE_ERROR_OUT_OF_MEMORY
;
994 real_vb
->buffer_offset
-= start
;
1000 static boolean
u_vbuf_need_minmax_index(struct u_vbuf
*mgr
)
1002 /* See if there are any per-vertex attribs which will be uploaded or
1003 * translated. Use bitmasks to get the info instead of looping over vertex
1005 return (mgr
->ve
->used_vb_mask
&
1006 ((mgr
->user_vb_mask
| mgr
->incompatible_vb_mask
|
1007 mgr
->ve
->incompatible_vb_mask_any
) &
1008 mgr
->ve
->noninstance_vb_mask_any
& mgr
->nonzero_stride_vb_mask
)) != 0;
1011 static boolean
u_vbuf_mapping_vertex_buffer_blocks(struct u_vbuf
*mgr
)
1013 /* Return true if there are hw buffers which don't need to be translated.
1015 * We could query whether each buffer is busy, but that would
1016 * be way more costly than this. */
1017 return (mgr
->ve
->used_vb_mask
&
1018 (~mgr
->user_vb_mask
& ~mgr
->incompatible_vb_mask
&
1019 mgr
->ve
->compatible_vb_mask_all
& mgr
->ve
->noninstance_vb_mask_any
&
1020 mgr
->nonzero_stride_vb_mask
)) != 0;
1023 static void u_vbuf_get_minmax_index(struct pipe_context
*pipe
,
1024 struct pipe_index_buffer
*ib
,
1025 boolean primitive_restart
,
1026 unsigned restart_index
,
1027 unsigned start
, unsigned count
,
1031 struct pipe_transfer
*transfer
= NULL
;
1032 const void *indices
;
1035 if (ib
->user_buffer
) {
1036 indices
= (uint8_t*)ib
->user_buffer
+
1037 ib
->offset
+ start
* ib
->index_size
;
1039 indices
= pipe_buffer_map_range(pipe
, ib
->buffer
,
1040 ib
->offset
+ start
* ib
->index_size
,
1041 count
* ib
->index_size
,
1042 PIPE_TRANSFER_READ
, &transfer
);
1045 switch (ib
->index_size
) {
1047 const unsigned *ui_indices
= (const unsigned*)indices
;
1048 unsigned max_ui
= 0;
1049 unsigned min_ui
= ~0U;
1050 if (primitive_restart
) {
1051 for (i
= 0; i
< count
; i
++) {
1052 if (ui_indices
[i
] != restart_index
) {
1053 if (ui_indices
[i
] > max_ui
) max_ui
= ui_indices
[i
];
1054 if (ui_indices
[i
] < min_ui
) min_ui
= ui_indices
[i
];
1059 for (i
= 0; i
< count
; i
++) {
1060 if (ui_indices
[i
] > max_ui
) max_ui
= ui_indices
[i
];
1061 if (ui_indices
[i
] < min_ui
) min_ui
= ui_indices
[i
];
1064 *out_min_index
= min_ui
;
1065 *out_max_index
= max_ui
;
1069 const unsigned short *us_indices
= (const unsigned short*)indices
;
1070 unsigned max_us
= 0;
1071 unsigned min_us
= ~0U;
1072 if (primitive_restart
) {
1073 for (i
= 0; i
< count
; i
++) {
1074 if (us_indices
[i
] != restart_index
) {
1075 if (us_indices
[i
] > max_us
) max_us
= us_indices
[i
];
1076 if (us_indices
[i
] < min_us
) min_us
= us_indices
[i
];
1081 for (i
= 0; i
< count
; i
++) {
1082 if (us_indices
[i
] > max_us
) max_us
= us_indices
[i
];
1083 if (us_indices
[i
] < min_us
) min_us
= us_indices
[i
];
1086 *out_min_index
= min_us
;
1087 *out_max_index
= max_us
;
1091 const unsigned char *ub_indices
= (const unsigned char*)indices
;
1092 unsigned max_ub
= 0;
1093 unsigned min_ub
= ~0U;
1094 if (primitive_restart
) {
1095 for (i
= 0; i
< count
; i
++) {
1096 if (ub_indices
[i
] != restart_index
) {
1097 if (ub_indices
[i
] > max_ub
) max_ub
= ub_indices
[i
];
1098 if (ub_indices
[i
] < min_ub
) min_ub
= ub_indices
[i
];
1103 for (i
= 0; i
< count
; i
++) {
1104 if (ub_indices
[i
] > max_ub
) max_ub
= ub_indices
[i
];
1105 if (ub_indices
[i
] < min_ub
) min_ub
= ub_indices
[i
];
1108 *out_min_index
= min_ub
;
1109 *out_max_index
= max_ub
;
1119 pipe_buffer_unmap(pipe
, transfer
);
1123 static void u_vbuf_set_driver_vertex_buffers(struct u_vbuf
*mgr
)
1125 struct pipe_context
*pipe
= mgr
->pipe
;
1126 unsigned start_slot
, count
;
1128 start_slot
= ffs(mgr
->dirty_real_vb_mask
) - 1;
1129 count
= util_last_bit(mgr
->dirty_real_vb_mask
>> start_slot
);
1131 pipe
->set_vertex_buffers(pipe
, start_slot
, count
,
1132 mgr
->real_vertex_buffer
+ start_slot
);
1133 mgr
->dirty_real_vb_mask
= 0;
1136 void u_vbuf_draw_vbo(struct u_vbuf
*mgr
, const struct pipe_draw_info
*info
)
1138 struct pipe_context
*pipe
= mgr
->pipe
;
1139 int start_vertex
, min_index
;
1140 unsigned num_vertices
;
1141 boolean unroll_indices
= FALSE
;
1142 uint32_t used_vb_mask
= mgr
->ve
->used_vb_mask
;
1143 uint32_t user_vb_mask
= mgr
->user_vb_mask
& used_vb_mask
;
1144 uint32_t incompatible_vb_mask
= mgr
->incompatible_vb_mask
& used_vb_mask
;
1145 struct pipe_draw_info new_info
;
1147 /* Normal draw. No fallback and no user buffers. */
1148 if (!incompatible_vb_mask
&&
1149 !mgr
->ve
->incompatible_elem_mask
&&
1152 /* Set vertex buffers if needed. */
1153 if (mgr
->dirty_real_vb_mask
& used_vb_mask
) {
1154 u_vbuf_set_driver_vertex_buffers(mgr
);
1157 pipe
->draw_vbo(pipe
, info
);
1163 /* Fallback. We need to know all the parameters. */
1164 if (new_info
.indirect
) {
1165 struct pipe_transfer
*transfer
= NULL
;
1168 if (new_info
.indexed
) {
1169 data
= pipe_buffer_map_range(pipe
, new_info
.indirect
,
1170 new_info
.indirect_offset
, 20,
1171 PIPE_TRANSFER_READ
, &transfer
);
1172 new_info
.index_bias
= data
[3];
1173 new_info
.start_instance
= data
[4];
1176 data
= pipe_buffer_map_range(pipe
, new_info
.indirect
,
1177 new_info
.indirect_offset
, 16,
1178 PIPE_TRANSFER_READ
, &transfer
);
1179 new_info
.start_instance
= data
[3];
1182 new_info
.count
= data
[0];
1183 new_info
.instance_count
= data
[1];
1184 new_info
.start
= data
[2];
1185 pipe_buffer_unmap(pipe
, transfer
);
1186 new_info
.indirect
= NULL
;
1189 if (new_info
.indexed
) {
1190 /* See if anything needs to be done for per-vertex attribs. */
1191 if (u_vbuf_need_minmax_index(mgr
)) {
1194 if (new_info
.max_index
!= ~0) {
1195 min_index
= new_info
.min_index
;
1196 max_index
= new_info
.max_index
;
1198 u_vbuf_get_minmax_index(mgr
->pipe
, &mgr
->index_buffer
,
1199 new_info
.primitive_restart
,
1200 new_info
.restart_index
, new_info
.start
,
1201 new_info
.count
, &min_index
, &max_index
);
1204 assert(min_index
<= max_index
);
1206 start_vertex
= min_index
+ new_info
.index_bias
;
1207 num_vertices
= max_index
+ 1 - min_index
;
1209 /* Primitive restart doesn't work when unrolling indices.
1210 * We would have to break this drawing operation into several ones. */
1211 /* Use some heuristic to see if unrolling indices improves
1213 if (!new_info
.primitive_restart
&&
1214 num_vertices
> new_info
.count
*2 &&
1215 num_vertices
- new_info
.count
> 32 &&
1216 !u_vbuf_mapping_vertex_buffer_blocks(mgr
)) {
1217 unroll_indices
= TRUE
;
1218 user_vb_mask
&= ~(mgr
->nonzero_stride_vb_mask
&
1219 mgr
->ve
->noninstance_vb_mask_any
);
1222 /* Nothing to do for per-vertex attribs. */
1228 start_vertex
= new_info
.start
;
1229 num_vertices
= new_info
.count
;
1233 /* Translate vertices with non-native layouts or formats. */
1234 if (unroll_indices
||
1235 incompatible_vb_mask
||
1236 mgr
->ve
->incompatible_elem_mask
) {
1237 if (!u_vbuf_translate_begin(mgr
, start_vertex
, num_vertices
,
1238 new_info
.start_instance
,
1239 new_info
.instance_count
, new_info
.start
,
1240 new_info
.count
, min_index
, unroll_indices
)) {
1241 debug_warn_once("u_vbuf_translate_begin() failed");
1245 if (unroll_indices
) {
1246 new_info
.indexed
= FALSE
;
1247 new_info
.index_bias
= 0;
1248 new_info
.min_index
= 0;
1249 new_info
.max_index
= new_info
.count
- 1;
1253 user_vb_mask
&= ~(incompatible_vb_mask
|
1254 mgr
->ve
->incompatible_vb_mask_all
);
1257 /* Upload user buffers. */
1259 if (u_vbuf_upload_buffers(mgr
, start_vertex
, num_vertices
,
1260 new_info
.start_instance
,
1261 new_info
.instance_count
) != PIPE_OK
) {
1262 debug_warn_once("u_vbuf_upload_buffers() failed");
1266 mgr
->dirty_real_vb_mask
|= user_vb_mask
;
1270 if (unroll_indices) {
1271 printf("unrolling indices: start_vertex = %i, num_vertices = %i\n",
1272 start_vertex, num_vertices);
1273 util_dump_draw_info(stdout, info);
1278 for (i = 0; i < mgr->nr_vertex_buffers; i++) {
1279 printf("input %i: ", i);
1280 util_dump_vertex_buffer(stdout, mgr->vertex_buffer+i);
1283 for (i = 0; i < mgr->nr_real_vertex_buffers; i++) {
1284 printf("real %i: ", i);
1285 util_dump_vertex_buffer(stdout, mgr->real_vertex_buffer+i);
1290 u_upload_unmap(mgr
->uploader
);
1291 u_vbuf_set_driver_vertex_buffers(mgr
);
1293 pipe
->draw_vbo(pipe
, &new_info
);
1295 if (mgr
->using_translate
) {
1296 u_vbuf_translate_end(mgr
);
1300 void u_vbuf_save_vertex_elements(struct u_vbuf
*mgr
)
1302 assert(!mgr
->ve_saved
);
1303 mgr
->ve_saved
= mgr
->ve
;
1306 void u_vbuf_restore_vertex_elements(struct u_vbuf
*mgr
)
1308 if (mgr
->ve
!= mgr
->ve_saved
) {
1309 struct pipe_context
*pipe
= mgr
->pipe
;
1311 mgr
->ve
= mgr
->ve_saved
;
1312 pipe
->bind_vertex_elements_state(pipe
,
1313 mgr
->ve
? mgr
->ve
->driver_cso
: NULL
);
1315 mgr
->ve_saved
= NULL
;
1318 void u_vbuf_save_aux_vertex_buffer_slot(struct u_vbuf
*mgr
)
1320 struct pipe_vertex_buffer
*vb
=
1321 &mgr
->vertex_buffer
[mgr
->aux_vertex_buffer_slot
];
1323 pipe_resource_reference(&mgr
->aux_vertex_buffer_saved
.buffer
, vb
->buffer
);
1324 memcpy(&mgr
->aux_vertex_buffer_saved
, vb
, sizeof(*vb
));
1327 void u_vbuf_restore_aux_vertex_buffer_slot(struct u_vbuf
*mgr
)
1329 u_vbuf_set_vertex_buffers(mgr
, mgr
->aux_vertex_buffer_slot
, 1,
1330 &mgr
->aux_vertex_buffer_saved
);
1331 pipe_resource_reference(&mgr
->aux_vertex_buffer_saved
.buffer
, NULL
);