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
;
150 /* This is what was set in set_vertex_buffers.
151 * May contain user buffers. */
152 struct pipe_vertex_buffer vertex_buffer
[PIPE_MAX_ATTRIBS
];
153 uint32_t enabled_vb_mask
;
155 /* Saved vertex buffer. */
156 unsigned aux_vertex_buffer_slot
;
157 struct pipe_vertex_buffer aux_vertex_buffer_saved
;
159 /* Vertex buffers for the driver.
160 * There are usually no user buffers. */
161 struct pipe_vertex_buffer real_vertex_buffer
[PIPE_MAX_ATTRIBS
];
162 uint32_t dirty_real_vb_mask
; /* which buffers are dirty since the last
163 call of set_vertex_buffers */
165 /* The index buffer. */
166 struct pipe_index_buffer index_buffer
;
168 /* Vertex elements. */
169 struct u_vbuf_elements
*ve
, *ve_saved
;
171 /* Vertex elements used for the translate fallback. */
172 struct pipe_vertex_element fallback_velems
[PIPE_MAX_ATTRIBS
];
173 /* If non-NULL, this is a vertex element state used for the translate
174 * fallback and therefore used for rendering too. */
175 boolean using_translate
;
176 /* The vertex buffer slot index where translated vertices have been
178 unsigned fallback_vbs
[VB_NUM
];
180 /* Which buffer is a user buffer. */
181 uint32_t user_vb_mask
; /* each bit describes a corresp. buffer */
182 /* Which buffer is incompatible (unaligned). */
183 uint32_t incompatible_vb_mask
; /* each bit describes a corresp. buffer */
184 /* Which buffer has a non-zero stride. */
185 uint32_t nonzero_stride_vb_mask
; /* each bit describes a corresp. buffer */
189 u_vbuf_create_vertex_elements(struct u_vbuf
*mgr
, unsigned count
,
190 const struct pipe_vertex_element
*attribs
);
191 static void u_vbuf_delete_vertex_elements(struct u_vbuf
*mgr
, void *cso
);
193 static const struct {
194 enum pipe_format from
, to
;
195 } vbuf_format_fallbacks
[] = {
196 { PIPE_FORMAT_R32_FIXED
, PIPE_FORMAT_R32_FLOAT
},
197 { PIPE_FORMAT_R32G32_FIXED
, PIPE_FORMAT_R32G32_FLOAT
},
198 { PIPE_FORMAT_R32G32B32_FIXED
, PIPE_FORMAT_R32G32B32_FLOAT
},
199 { PIPE_FORMAT_R32G32B32A32_FIXED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
200 { PIPE_FORMAT_R16_FLOAT
, PIPE_FORMAT_R32_FLOAT
},
201 { PIPE_FORMAT_R16G16_FLOAT
, PIPE_FORMAT_R32G32_FLOAT
},
202 { PIPE_FORMAT_R16G16B16_FLOAT
, PIPE_FORMAT_R32G32B32_FLOAT
},
203 { PIPE_FORMAT_R16G16B16A16_FLOAT
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
204 { PIPE_FORMAT_R64_FLOAT
, PIPE_FORMAT_R32_FLOAT
},
205 { PIPE_FORMAT_R64G64_FLOAT
, PIPE_FORMAT_R32G32_FLOAT
},
206 { PIPE_FORMAT_R64G64B64_FLOAT
, PIPE_FORMAT_R32G32B32_FLOAT
},
207 { PIPE_FORMAT_R64G64B64A64_FLOAT
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
208 { PIPE_FORMAT_R32_UNORM
, PIPE_FORMAT_R32_FLOAT
},
209 { PIPE_FORMAT_R32G32_UNORM
, PIPE_FORMAT_R32G32_FLOAT
},
210 { PIPE_FORMAT_R32G32B32_UNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
211 { PIPE_FORMAT_R32G32B32A32_UNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
212 { PIPE_FORMAT_R32_SNORM
, PIPE_FORMAT_R32_FLOAT
},
213 { PIPE_FORMAT_R32G32_SNORM
, PIPE_FORMAT_R32G32_FLOAT
},
214 { PIPE_FORMAT_R32G32B32_SNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
215 { PIPE_FORMAT_R32G32B32A32_SNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
216 { PIPE_FORMAT_R32_USCALED
, PIPE_FORMAT_R32_FLOAT
},
217 { PIPE_FORMAT_R32G32_USCALED
, PIPE_FORMAT_R32G32_FLOAT
},
218 { PIPE_FORMAT_R32G32B32_USCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
219 { PIPE_FORMAT_R32G32B32A32_USCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
220 { PIPE_FORMAT_R32_SSCALED
, PIPE_FORMAT_R32_FLOAT
},
221 { PIPE_FORMAT_R32G32_SSCALED
, PIPE_FORMAT_R32G32_FLOAT
},
222 { PIPE_FORMAT_R32G32B32_SSCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
223 { PIPE_FORMAT_R32G32B32A32_SSCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
224 { PIPE_FORMAT_R16_UNORM
, PIPE_FORMAT_R32_FLOAT
},
225 { PIPE_FORMAT_R16G16_UNORM
, PIPE_FORMAT_R32G32_FLOAT
},
226 { PIPE_FORMAT_R16G16B16_UNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
227 { PIPE_FORMAT_R16G16B16A16_UNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
228 { PIPE_FORMAT_R16_SNORM
, PIPE_FORMAT_R32_FLOAT
},
229 { PIPE_FORMAT_R16G16_SNORM
, PIPE_FORMAT_R32G32_FLOAT
},
230 { PIPE_FORMAT_R16G16B16_SNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
231 { PIPE_FORMAT_R16G16B16A16_SNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
232 { PIPE_FORMAT_R16_USCALED
, PIPE_FORMAT_R32_FLOAT
},
233 { PIPE_FORMAT_R16G16_USCALED
, PIPE_FORMAT_R32G32_FLOAT
},
234 { PIPE_FORMAT_R16G16B16_USCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
235 { PIPE_FORMAT_R16G16B16A16_USCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
236 { PIPE_FORMAT_R16_SSCALED
, PIPE_FORMAT_R32_FLOAT
},
237 { PIPE_FORMAT_R16G16_SSCALED
, PIPE_FORMAT_R32G32_FLOAT
},
238 { PIPE_FORMAT_R16G16B16_SSCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
239 { PIPE_FORMAT_R16G16B16A16_SSCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
240 { PIPE_FORMAT_R8_UNORM
, PIPE_FORMAT_R32_FLOAT
},
241 { PIPE_FORMAT_R8G8_UNORM
, PIPE_FORMAT_R32G32_FLOAT
},
242 { PIPE_FORMAT_R8G8B8_UNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
243 { PIPE_FORMAT_R8G8B8A8_UNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
244 { PIPE_FORMAT_R8_SNORM
, PIPE_FORMAT_R32_FLOAT
},
245 { PIPE_FORMAT_R8G8_SNORM
, PIPE_FORMAT_R32G32_FLOAT
},
246 { PIPE_FORMAT_R8G8B8_SNORM
, PIPE_FORMAT_R32G32B32_FLOAT
},
247 { PIPE_FORMAT_R8G8B8A8_SNORM
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
248 { PIPE_FORMAT_R8_USCALED
, PIPE_FORMAT_R32_FLOAT
},
249 { PIPE_FORMAT_R8G8_USCALED
, PIPE_FORMAT_R32G32_FLOAT
},
250 { PIPE_FORMAT_R8G8B8_USCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
251 { PIPE_FORMAT_R8G8B8A8_USCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
252 { PIPE_FORMAT_R8_SSCALED
, PIPE_FORMAT_R32_FLOAT
},
253 { PIPE_FORMAT_R8G8_SSCALED
, PIPE_FORMAT_R32G32_FLOAT
},
254 { PIPE_FORMAT_R8G8B8_SSCALED
, PIPE_FORMAT_R32G32B32_FLOAT
},
255 { PIPE_FORMAT_R8G8B8A8_SSCALED
, PIPE_FORMAT_R32G32B32A32_FLOAT
},
258 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
< ARRAY_SIZE(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 (!(flags
& U_VBUF_FLAG_NO_USER_VBOS
) && !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
));
320 /* u_vbuf uses its own caching for vertex elements, because it needs to keep
321 * its own preprocessed state per vertex element CSO. */
322 static struct u_vbuf_elements
*
323 u_vbuf_set_vertex_elements_internal(struct u_vbuf
*mgr
, unsigned count
,
324 const struct pipe_vertex_element
*states
)
326 struct pipe_context
*pipe
= mgr
->pipe
;
327 unsigned key_size
, hash_key
;
328 struct cso_hash_iter iter
;
329 struct u_vbuf_elements
*ve
;
330 struct cso_velems_state velems_state
;
332 /* need to include the count into the stored state data too. */
333 key_size
= sizeof(struct pipe_vertex_element
) * count
+ sizeof(unsigned);
334 velems_state
.count
= count
;
335 memcpy(velems_state
.velems
, states
,
336 sizeof(struct pipe_vertex_element
) * count
);
337 hash_key
= cso_construct_key((void*)&velems_state
, key_size
);
338 iter
= cso_find_state_template(mgr
->cso_cache
, hash_key
, CSO_VELEMENTS
,
339 (void*)&velems_state
, key_size
);
341 if (cso_hash_iter_is_null(iter
)) {
342 struct cso_velements
*cso
= MALLOC_STRUCT(cso_velements
);
343 memcpy(&cso
->state
, &velems_state
, key_size
);
344 cso
->data
= u_vbuf_create_vertex_elements(mgr
, count
, states
);
345 cso
->delete_state
= (cso_state_callback
)u_vbuf_delete_vertex_elements
;
346 cso
->context
= (void*)mgr
;
348 iter
= cso_insert_state(mgr
->cso_cache
, hash_key
, CSO_VELEMENTS
, cso
);
351 ve
= ((struct cso_velements
*)cso_hash_iter_data(iter
))->data
;
357 pipe
->bind_vertex_elements_state(pipe
, ve
->driver_cso
);
362 void u_vbuf_set_vertex_elements(struct u_vbuf
*mgr
, unsigned count
,
363 const struct pipe_vertex_element
*states
)
365 mgr
->ve
= u_vbuf_set_vertex_elements_internal(mgr
, count
, states
);
368 void u_vbuf_destroy(struct u_vbuf
*mgr
)
370 struct pipe_screen
*screen
= mgr
->pipe
->screen
;
372 unsigned num_vb
= screen
->get_shader_param(screen
, PIPE_SHADER_VERTEX
,
373 PIPE_SHADER_CAP_MAX_INPUTS
);
375 mgr
->pipe
->set_index_buffer(mgr
->pipe
, NULL
);
376 pipe_resource_reference(&mgr
->index_buffer
.buffer
, NULL
);
378 mgr
->pipe
->set_vertex_buffers(mgr
->pipe
, 0, num_vb
, NULL
);
380 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
381 pipe_resource_reference(&mgr
->vertex_buffer
[i
].buffer
, NULL
);
383 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
384 pipe_resource_reference(&mgr
->real_vertex_buffer
[i
].buffer
, NULL
);
386 pipe_resource_reference(&mgr
->aux_vertex_buffer_saved
.buffer
, NULL
);
388 translate_cache_destroy(mgr
->translate_cache
);
389 cso_cache_delete(mgr
->cso_cache
);
393 static enum pipe_error
394 u_vbuf_translate_buffers(struct u_vbuf
*mgr
, struct translate_key
*key
,
395 unsigned vb_mask
, unsigned out_vb
,
396 int start_vertex
, unsigned num_vertices
,
397 int start_index
, unsigned num_indices
, int min_index
,
398 boolean unroll_indices
)
400 struct translate
*tr
;
401 struct pipe_transfer
*vb_transfer
[PIPE_MAX_ATTRIBS
] = {0};
402 struct pipe_resource
*out_buffer
= NULL
;
404 unsigned out_offset
, mask
;
406 /* Get a translate object. */
407 tr
= translate_cache_find(mgr
->translate_cache
, key
);
409 /* Map buffers we want to translate. */
412 struct pipe_vertex_buffer
*vb
;
415 unsigned i
= u_bit_scan(&mask
);
417 vb
= &mgr
->vertex_buffer
[i
];
418 offset
= vb
->buffer_offset
+ vb
->stride
* start_vertex
;
420 if (vb
->user_buffer
) {
421 map
= (uint8_t*)vb
->user_buffer
+ offset
;
423 unsigned size
= vb
->stride
? num_vertices
* vb
->stride
426 if (offset
+size
> vb
->buffer
->width0
) {
427 size
= vb
->buffer
->width0
- offset
;
430 map
= pipe_buffer_map_range(mgr
->pipe
, vb
->buffer
, offset
, size
,
431 PIPE_TRANSFER_READ
, &vb_transfer
[i
]);
434 /* Subtract min_index so that indexing with the index buffer works. */
435 if (unroll_indices
) {
436 map
-= (ptrdiff_t)vb
->stride
* min_index
;
439 tr
->set_buffer(tr
, i
, map
, vb
->stride
, ~0);
443 if (unroll_indices
) {
444 struct pipe_index_buffer
*ib
= &mgr
->index_buffer
;
445 struct pipe_transfer
*transfer
= NULL
;
446 unsigned offset
= ib
->offset
+ start_index
* ib
->index_size
;
449 assert((ib
->buffer
|| ib
->user_buffer
) && ib
->index_size
);
451 /* Create and map the output buffer. */
452 u_upload_alloc(mgr
->pipe
->stream_uploader
, 0,
453 key
->output_stride
* num_indices
, 4,
454 &out_offset
, &out_buffer
,
457 return PIPE_ERROR_OUT_OF_MEMORY
;
459 if (ib
->user_buffer
) {
460 map
= (uint8_t*)ib
->user_buffer
+ offset
;
462 map
= pipe_buffer_map_range(mgr
->pipe
, ib
->buffer
, offset
,
463 num_indices
* ib
->index_size
,
464 PIPE_TRANSFER_READ
, &transfer
);
467 switch (ib
->index_size
) {
469 tr
->run_elts(tr
, (unsigned*)map
, num_indices
, 0, 0, out_map
);
472 tr
->run_elts16(tr
, (uint16_t*)map
, num_indices
, 0, 0, out_map
);
475 tr
->run_elts8(tr
, map
, num_indices
, 0, 0, out_map
);
480 pipe_buffer_unmap(mgr
->pipe
, transfer
);
483 /* Create and map the output buffer. */
484 u_upload_alloc(mgr
->pipe
->stream_uploader
,
485 key
->output_stride
* start_vertex
,
486 key
->output_stride
* num_vertices
, 4,
487 &out_offset
, &out_buffer
,
490 return PIPE_ERROR_OUT_OF_MEMORY
;
492 out_offset
-= key
->output_stride
* start_vertex
;
494 tr
->run(tr
, 0, num_vertices
, 0, 0, out_map
);
497 /* Unmap all buffers. */
500 unsigned i
= u_bit_scan(&mask
);
502 if (vb_transfer
[i
]) {
503 pipe_buffer_unmap(mgr
->pipe
, vb_transfer
[i
]);
507 /* Setup the new vertex buffer. */
508 mgr
->real_vertex_buffer
[out_vb
].buffer_offset
= out_offset
;
509 mgr
->real_vertex_buffer
[out_vb
].stride
= key
->output_stride
;
511 /* Move the buffer reference. */
512 pipe_resource_reference(
513 &mgr
->real_vertex_buffer
[out_vb
].buffer
, NULL
);
514 mgr
->real_vertex_buffer
[out_vb
].buffer
= out_buffer
;
520 u_vbuf_translate_find_free_vb_slots(struct u_vbuf
*mgr
,
521 unsigned mask
[VB_NUM
])
524 unsigned fallback_vbs
[VB_NUM
];
525 /* Set the bit for each buffer which is incompatible, or isn't set. */
526 uint32_t unused_vb_mask
=
527 mgr
->ve
->incompatible_vb_mask_all
| mgr
->incompatible_vb_mask
|
528 ~mgr
->enabled_vb_mask
;
530 memset(fallback_vbs
, ~0, sizeof(fallback_vbs
));
532 /* Find free slots for each type if needed. */
533 for (type
= 0; type
< VB_NUM
; type
++) {
537 if (!unused_vb_mask
) {
541 index
= ffs(unused_vb_mask
) - 1;
542 fallback_vbs
[type
] = index
;
543 unused_vb_mask
&= ~(1 << index
);
544 /*printf("found slot=%i for type=%i\n", index, type);*/
548 for (type
= 0; type
< VB_NUM
; type
++) {
550 mgr
->dirty_real_vb_mask
|= 1 << fallback_vbs
[type
];
554 memcpy(mgr
->fallback_vbs
, fallback_vbs
, sizeof(fallback_vbs
));
559 u_vbuf_translate_begin(struct u_vbuf
*mgr
,
560 int start_vertex
, unsigned num_vertices
,
561 int start_instance
, unsigned num_instances
,
562 int start_index
, unsigned num_indices
, int min_index
,
563 boolean unroll_indices
)
565 unsigned mask
[VB_NUM
] = {0};
566 struct translate_key key
[VB_NUM
];
567 unsigned elem_index
[VB_NUM
][PIPE_MAX_ATTRIBS
]; /* ... into key.elements */
569 unsigned incompatible_vb_mask
= mgr
->incompatible_vb_mask
&
570 mgr
->ve
->used_vb_mask
;
572 int start
[VB_NUM
] = {
573 start_vertex
, /* VERTEX */
574 start_instance
, /* INSTANCE */
578 unsigned num
[VB_NUM
] = {
579 num_vertices
, /* VERTEX */
580 num_instances
, /* INSTANCE */
584 memset(key
, 0, sizeof(key
));
585 memset(elem_index
, ~0, sizeof(elem_index
));
587 /* See if there are vertex attribs of each type to translate and
589 for (i
= 0; i
< mgr
->ve
->count
; i
++) {
590 unsigned vb_index
= mgr
->ve
->ve
[i
].vertex_buffer_index
;
592 if (!mgr
->vertex_buffer
[vb_index
].stride
) {
593 if (!(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
594 !(incompatible_vb_mask
& (1 << vb_index
))) {
597 mask
[VB_CONST
] |= 1 << vb_index
;
598 } else if (mgr
->ve
->ve
[i
].instance_divisor
) {
599 if (!(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
600 !(incompatible_vb_mask
& (1 << vb_index
))) {
603 mask
[VB_INSTANCE
] |= 1 << vb_index
;
605 if (!unroll_indices
&&
606 !(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
607 !(incompatible_vb_mask
& (1 << vb_index
))) {
610 mask
[VB_VERTEX
] |= 1 << vb_index
;
614 assert(mask
[VB_VERTEX
] || mask
[VB_INSTANCE
] || mask
[VB_CONST
]);
616 /* Find free vertex buffer slots. */
617 if (!u_vbuf_translate_find_free_vb_slots(mgr
, mask
)) {
621 /* Initialize the translate keys. */
622 for (i
= 0; i
< mgr
->ve
->count
; i
++) {
623 struct translate_key
*k
;
624 struct translate_element
*te
;
625 enum pipe_format output_format
= mgr
->ve
->native_format
[i
];
626 unsigned bit
, vb_index
= mgr
->ve
->ve
[i
].vertex_buffer_index
;
629 if (!(mgr
->ve
->incompatible_elem_mask
& (1 << i
)) &&
630 !(incompatible_vb_mask
& (1 << vb_index
)) &&
631 (!unroll_indices
|| !(mask
[VB_VERTEX
] & bit
))) {
635 /* Set type to what we will translate.
636 * Whether vertex, instance, or constant attribs. */
637 for (type
= 0; type
< VB_NUM
; type
++) {
638 if (mask
[type
] & bit
) {
642 assert(type
< VB_NUM
);
643 if (mgr
->ve
->ve
[i
].src_format
!= output_format
)
644 assert(translate_is_output_format_supported(output_format
));
645 /*printf("velem=%i type=%i\n", i, type);*/
647 /* Add the vertex element. */
649 elem_index
[type
][i
] = k
->nr_elements
;
651 te
= &k
->element
[k
->nr_elements
];
652 te
->type
= TRANSLATE_ELEMENT_NORMAL
;
653 te
->instance_divisor
= 0;
654 te
->input_buffer
= vb_index
;
655 te
->input_format
= mgr
->ve
->ve
[i
].src_format
;
656 te
->input_offset
= mgr
->ve
->ve
[i
].src_offset
;
657 te
->output_format
= output_format
;
658 te
->output_offset
= k
->output_stride
;
660 k
->output_stride
+= mgr
->ve
->native_format_size
[i
];
664 /* Translate buffers. */
665 for (type
= 0; type
< VB_NUM
; type
++) {
666 if (key
[type
].nr_elements
) {
668 err
= u_vbuf_translate_buffers(mgr
, &key
[type
], mask
[type
],
669 mgr
->fallback_vbs
[type
],
670 start
[type
], num
[type
],
671 start_index
, num_indices
, min_index
,
672 unroll_indices
&& type
== VB_VERTEX
);
676 /* Fixup the stride for constant attribs. */
677 if (type
== VB_CONST
) {
678 mgr
->real_vertex_buffer
[mgr
->fallback_vbs
[VB_CONST
]].stride
= 0;
683 /* Setup new vertex elements. */
684 for (i
= 0; i
< mgr
->ve
->count
; i
++) {
685 for (type
= 0; type
< VB_NUM
; type
++) {
686 if (elem_index
[type
][i
] < key
[type
].nr_elements
) {
687 struct translate_element
*te
= &key
[type
].element
[elem_index
[type
][i
]];
688 mgr
->fallback_velems
[i
].instance_divisor
= mgr
->ve
->ve
[i
].instance_divisor
;
689 mgr
->fallback_velems
[i
].src_format
= te
->output_format
;
690 mgr
->fallback_velems
[i
].src_offset
= te
->output_offset
;
691 mgr
->fallback_velems
[i
].vertex_buffer_index
= mgr
->fallback_vbs
[type
];
693 /* elem_index[type][i] can only be set for one type. */
694 assert(type
> VB_INSTANCE
|| elem_index
[type
+1][i
] == ~0u);
695 assert(type
> VB_VERTEX
|| elem_index
[type
+2][i
] == ~0u);
699 /* No translating, just copy the original vertex element over. */
700 if (type
== VB_NUM
) {
701 memcpy(&mgr
->fallback_velems
[i
], &mgr
->ve
->ve
[i
],
702 sizeof(struct pipe_vertex_element
));
706 u_vbuf_set_vertex_elements_internal(mgr
, mgr
->ve
->count
,
707 mgr
->fallback_velems
);
708 mgr
->using_translate
= TRUE
;
712 static void u_vbuf_translate_end(struct u_vbuf
*mgr
)
716 /* Restore vertex elements. */
717 mgr
->pipe
->bind_vertex_elements_state(mgr
->pipe
, mgr
->ve
->driver_cso
);
718 mgr
->using_translate
= FALSE
;
720 /* Unreference the now-unused VBOs. */
721 for (i
= 0; i
< VB_NUM
; i
++) {
722 unsigned vb
= mgr
->fallback_vbs
[i
];
724 pipe_resource_reference(&mgr
->real_vertex_buffer
[vb
].buffer
, NULL
);
725 mgr
->fallback_vbs
[i
] = ~0;
727 /* This will cause the buffer to be unbound in the driver later. */
728 mgr
->dirty_real_vb_mask
|= 1 << vb
;
734 u_vbuf_create_vertex_elements(struct u_vbuf
*mgr
, unsigned count
,
735 const struct pipe_vertex_element
*attribs
)
737 struct pipe_context
*pipe
= mgr
->pipe
;
739 struct pipe_vertex_element driver_attribs
[PIPE_MAX_ATTRIBS
];
740 struct u_vbuf_elements
*ve
= CALLOC_STRUCT(u_vbuf_elements
);
741 uint32_t used_buffers
= 0;
745 memcpy(ve
->ve
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
746 memcpy(driver_attribs
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
748 /* Set the best native format in case the original format is not
750 for (i
= 0; i
< count
; i
++) {
751 enum pipe_format format
= ve
->ve
[i
].src_format
;
753 ve
->src_format_size
[i
] = util_format_get_blocksize(format
);
755 used_buffers
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
757 if (!ve
->ve
[i
].instance_divisor
) {
758 ve
->noninstance_vb_mask_any
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
761 format
= mgr
->caps
.format_translation
[format
];
763 driver_attribs
[i
].src_format
= format
;
764 ve
->native_format
[i
] = format
;
765 ve
->native_format_size
[i
] =
766 util_format_get_blocksize(ve
->native_format
[i
]);
768 if (ve
->ve
[i
].src_format
!= format
||
769 (!mgr
->caps
.velem_src_offset_unaligned
&&
770 ve
->ve
[i
].src_offset
% 4 != 0)) {
771 ve
->incompatible_elem_mask
|= 1 << i
;
772 ve
->incompatible_vb_mask_any
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
774 ve
->compatible_vb_mask_any
|= 1 << ve
->ve
[i
].vertex_buffer_index
;
778 ve
->used_vb_mask
= used_buffers
;
779 ve
->compatible_vb_mask_all
= ~ve
->incompatible_vb_mask_any
& used_buffers
;
780 ve
->incompatible_vb_mask_all
= ~ve
->compatible_vb_mask_any
& used_buffers
;
782 /* Align the formats and offsets to the size of DWORD if needed. */
783 if (!mgr
->caps
.velem_src_offset_unaligned
) {
784 for (i
= 0; i
< count
; i
++) {
785 ve
->native_format_size
[i
] = align(ve
->native_format_size
[i
], 4);
786 driver_attribs
[i
].src_offset
= align(ve
->ve
[i
].src_offset
, 4);
791 pipe
->create_vertex_elements_state(pipe
, count
, driver_attribs
);
795 static void u_vbuf_delete_vertex_elements(struct u_vbuf
*mgr
, void *cso
)
797 struct pipe_context
*pipe
= mgr
->pipe
;
798 struct u_vbuf_elements
*ve
= cso
;
800 pipe
->delete_vertex_elements_state(pipe
, ve
->driver_cso
);
804 void u_vbuf_set_vertex_buffers(struct u_vbuf
*mgr
,
805 unsigned start_slot
, unsigned count
,
806 const struct pipe_vertex_buffer
*bufs
)
809 /* which buffers are enabled */
810 uint32_t enabled_vb_mask
= 0;
811 /* which buffers are in user memory */
812 uint32_t user_vb_mask
= 0;
813 /* which buffers are incompatible with the driver */
814 uint32_t incompatible_vb_mask
= 0;
815 /* which buffers have a non-zero stride */
816 uint32_t nonzero_stride_vb_mask
= 0;
817 uint32_t mask
= ~(((1ull << count
) - 1) << start_slot
);
819 /* Zero out the bits we are going to rewrite completely. */
820 mgr
->user_vb_mask
&= mask
;
821 mgr
->incompatible_vb_mask
&= mask
;
822 mgr
->nonzero_stride_vb_mask
&= mask
;
823 mgr
->enabled_vb_mask
&= mask
;
826 struct pipe_context
*pipe
= mgr
->pipe
;
828 mgr
->dirty_real_vb_mask
&= mask
;
830 for (i
= 0; i
< count
; i
++) {
831 unsigned dst_index
= start_slot
+ i
;
833 pipe_resource_reference(&mgr
->vertex_buffer
[dst_index
].buffer
, NULL
);
834 pipe_resource_reference(&mgr
->real_vertex_buffer
[dst_index
].buffer
,
838 pipe
->set_vertex_buffers(pipe
, start_slot
, count
, NULL
);
842 for (i
= 0; i
< count
; i
++) {
843 unsigned dst_index
= start_slot
+ i
;
844 const struct pipe_vertex_buffer
*vb
= &bufs
[i
];
845 struct pipe_vertex_buffer
*orig_vb
= &mgr
->vertex_buffer
[dst_index
];
846 struct pipe_vertex_buffer
*real_vb
= &mgr
->real_vertex_buffer
[dst_index
];
848 if (!vb
->buffer
&& !vb
->user_buffer
) {
849 pipe_resource_reference(&orig_vb
->buffer
, NULL
);
850 pipe_resource_reference(&real_vb
->buffer
, NULL
);
851 real_vb
->user_buffer
= NULL
;
855 pipe_resource_reference(&orig_vb
->buffer
, vb
->buffer
);
856 orig_vb
->user_buffer
= vb
->user_buffer
;
858 real_vb
->buffer_offset
= orig_vb
->buffer_offset
= vb
->buffer_offset
;
859 real_vb
->stride
= orig_vb
->stride
= vb
->stride
;
862 nonzero_stride_vb_mask
|= 1 << dst_index
;
864 enabled_vb_mask
|= 1 << dst_index
;
866 if ((!mgr
->caps
.buffer_offset_unaligned
&& vb
->buffer_offset
% 4 != 0) ||
867 (!mgr
->caps
.buffer_stride_unaligned
&& vb
->stride
% 4 != 0)) {
868 incompatible_vb_mask
|= 1 << dst_index
;
869 pipe_resource_reference(&real_vb
->buffer
, NULL
);
873 if (!mgr
->caps
.user_vertex_buffers
&& vb
->user_buffer
) {
874 user_vb_mask
|= 1 << dst_index
;
875 pipe_resource_reference(&real_vb
->buffer
, NULL
);
879 pipe_resource_reference(&real_vb
->buffer
, vb
->buffer
);
880 real_vb
->user_buffer
= vb
->user_buffer
;
883 mgr
->user_vb_mask
|= user_vb_mask
;
884 mgr
->incompatible_vb_mask
|= incompatible_vb_mask
;
885 mgr
->nonzero_stride_vb_mask
|= nonzero_stride_vb_mask
;
886 mgr
->enabled_vb_mask
|= enabled_vb_mask
;
888 /* All changed buffers are marked as dirty, even the NULL ones,
889 * which will cause the NULL buffers to be unbound in the driver later. */
890 mgr
->dirty_real_vb_mask
|= ~mask
;
893 void u_vbuf_set_index_buffer(struct u_vbuf
*mgr
,
894 const struct pipe_index_buffer
*ib
)
896 struct pipe_context
*pipe
= mgr
->pipe
;
899 assert(ib
->offset
% ib
->index_size
== 0);
900 pipe_resource_reference(&mgr
->index_buffer
.buffer
, ib
->buffer
);
901 memcpy(&mgr
->index_buffer
, ib
, sizeof(*ib
));
903 pipe_resource_reference(&mgr
->index_buffer
.buffer
, NULL
);
906 pipe
->set_index_buffer(pipe
, ib
);
909 static enum pipe_error
910 u_vbuf_upload_buffers(struct u_vbuf
*mgr
,
911 int start_vertex
, unsigned num_vertices
,
912 int start_instance
, unsigned num_instances
)
915 unsigned nr_velems
= mgr
->ve
->count
;
916 struct pipe_vertex_element
*velems
=
917 mgr
->using_translate
? mgr
->fallback_velems
: mgr
->ve
->ve
;
918 unsigned start_offset
[PIPE_MAX_ATTRIBS
];
919 unsigned end_offset
[PIPE_MAX_ATTRIBS
];
920 uint32_t buffer_mask
= 0;
922 /* Determine how much data needs to be uploaded. */
923 for (i
= 0; i
< nr_velems
; i
++) {
924 struct pipe_vertex_element
*velem
= &velems
[i
];
925 unsigned index
= velem
->vertex_buffer_index
;
926 struct pipe_vertex_buffer
*vb
= &mgr
->vertex_buffer
[index
];
927 unsigned instance_div
, first
, size
, index_bit
;
929 /* Skip the buffers generated by translate. */
930 if (index
== mgr
->fallback_vbs
[VB_VERTEX
] ||
931 index
== mgr
->fallback_vbs
[VB_INSTANCE
] ||
932 index
== mgr
->fallback_vbs
[VB_CONST
]) {
936 if (!vb
->user_buffer
) {
940 instance_div
= velem
->instance_divisor
;
941 first
= vb
->buffer_offset
+ velem
->src_offset
;
944 /* Constant attrib. */
945 size
= mgr
->ve
->src_format_size
[i
];
946 } else if (instance_div
) {
947 /* Per-instance attrib. */
948 unsigned count
= (num_instances
+ instance_div
- 1) / instance_div
;
949 first
+= vb
->stride
* start_instance
;
950 size
= vb
->stride
* (count
- 1) + mgr
->ve
->src_format_size
[i
];
952 /* Per-vertex attrib. */
953 first
+= vb
->stride
* start_vertex
;
954 size
= vb
->stride
* (num_vertices
- 1) + mgr
->ve
->src_format_size
[i
];
957 index_bit
= 1 << index
;
959 /* Update offsets. */
960 if (!(buffer_mask
& index_bit
)) {
961 start_offset
[index
] = first
;
962 end_offset
[index
] = first
+ size
;
964 if (first
< start_offset
[index
])
965 start_offset
[index
] = first
;
966 if (first
+ size
> end_offset
[index
])
967 end_offset
[index
] = first
+ size
;
970 buffer_mask
|= index_bit
;
973 /* Upload buffers. */
974 while (buffer_mask
) {
976 struct pipe_vertex_buffer
*real_vb
;
979 i
= u_bit_scan(&buffer_mask
);
981 start
= start_offset
[i
];
985 real_vb
= &mgr
->real_vertex_buffer
[i
];
986 ptr
= mgr
->vertex_buffer
[i
].user_buffer
;
988 u_upload_data(mgr
->pipe
->stream_uploader
, start
, end
- start
, 4, ptr
+ start
,
989 &real_vb
->buffer_offset
, &real_vb
->buffer
);
990 if (!real_vb
->buffer
)
991 return PIPE_ERROR_OUT_OF_MEMORY
;
993 real_vb
->buffer_offset
-= start
;
999 static boolean
u_vbuf_need_minmax_index(const struct u_vbuf
*mgr
)
1001 /* See if there are any per-vertex attribs which will be uploaded or
1002 * translated. Use bitmasks to get the info instead of looping over vertex
1004 return (mgr
->ve
->used_vb_mask
&
1005 ((mgr
->user_vb_mask
|
1006 mgr
->incompatible_vb_mask
|
1007 mgr
->ve
->incompatible_vb_mask_any
) &
1008 mgr
->ve
->noninstance_vb_mask_any
&
1009 mgr
->nonzero_stride_vb_mask
)) != 0;
1012 static boolean
u_vbuf_mapping_vertex_buffer_blocks(const struct u_vbuf
*mgr
)
1014 /* Return true if there are hw buffers which don't need to be translated.
1016 * We could query whether each buffer is busy, but that would
1017 * be way more costly than this. */
1018 return (mgr
->ve
->used_vb_mask
&
1019 (~mgr
->user_vb_mask
&
1020 ~mgr
->incompatible_vb_mask
&
1021 mgr
->ve
->compatible_vb_mask_all
&
1022 mgr
->ve
->noninstance_vb_mask_any
&
1023 mgr
->nonzero_stride_vb_mask
)) != 0;
1026 static void u_vbuf_get_minmax_index(struct pipe_context
*pipe
,
1027 struct pipe_index_buffer
*ib
,
1028 boolean primitive_restart
,
1029 unsigned restart_index
,
1030 unsigned start
, unsigned count
,
1034 struct pipe_transfer
*transfer
= NULL
;
1035 const void *indices
;
1038 if (ib
->user_buffer
) {
1039 indices
= (uint8_t*)ib
->user_buffer
+
1040 ib
->offset
+ start
* ib
->index_size
;
1042 indices
= pipe_buffer_map_range(pipe
, ib
->buffer
,
1043 ib
->offset
+ start
* ib
->index_size
,
1044 count
* ib
->index_size
,
1045 PIPE_TRANSFER_READ
, &transfer
);
1048 switch (ib
->index_size
) {
1050 const unsigned *ui_indices
= (const unsigned*)indices
;
1051 unsigned max_ui
= 0;
1052 unsigned min_ui
= ~0U;
1053 if (primitive_restart
) {
1054 for (i
= 0; i
< count
; i
++) {
1055 if (ui_indices
[i
] != restart_index
) {
1056 if (ui_indices
[i
] > max_ui
) max_ui
= ui_indices
[i
];
1057 if (ui_indices
[i
] < min_ui
) min_ui
= ui_indices
[i
];
1062 for (i
= 0; i
< count
; i
++) {
1063 if (ui_indices
[i
] > max_ui
) max_ui
= ui_indices
[i
];
1064 if (ui_indices
[i
] < min_ui
) min_ui
= ui_indices
[i
];
1067 *out_min_index
= min_ui
;
1068 *out_max_index
= max_ui
;
1072 const unsigned short *us_indices
= (const unsigned short*)indices
;
1073 unsigned max_us
= 0;
1074 unsigned min_us
= ~0U;
1075 if (primitive_restart
) {
1076 for (i
= 0; i
< count
; i
++) {
1077 if (us_indices
[i
] != restart_index
) {
1078 if (us_indices
[i
] > max_us
) max_us
= us_indices
[i
];
1079 if (us_indices
[i
] < min_us
) min_us
= us_indices
[i
];
1084 for (i
= 0; i
< count
; i
++) {
1085 if (us_indices
[i
] > max_us
) max_us
= us_indices
[i
];
1086 if (us_indices
[i
] < min_us
) min_us
= us_indices
[i
];
1089 *out_min_index
= min_us
;
1090 *out_max_index
= max_us
;
1094 const unsigned char *ub_indices
= (const unsigned char*)indices
;
1095 unsigned max_ub
= 0;
1096 unsigned min_ub
= ~0U;
1097 if (primitive_restart
) {
1098 for (i
= 0; i
< count
; i
++) {
1099 if (ub_indices
[i
] != restart_index
) {
1100 if (ub_indices
[i
] > max_ub
) max_ub
= ub_indices
[i
];
1101 if (ub_indices
[i
] < min_ub
) min_ub
= ub_indices
[i
];
1106 for (i
= 0; i
< count
; i
++) {
1107 if (ub_indices
[i
] > max_ub
) max_ub
= ub_indices
[i
];
1108 if (ub_indices
[i
] < min_ub
) min_ub
= ub_indices
[i
];
1111 *out_min_index
= min_ub
;
1112 *out_max_index
= max_ub
;
1122 pipe_buffer_unmap(pipe
, transfer
);
1126 static void u_vbuf_set_driver_vertex_buffers(struct u_vbuf
*mgr
)
1128 struct pipe_context
*pipe
= mgr
->pipe
;
1129 unsigned start_slot
, count
;
1131 start_slot
= ffs(mgr
->dirty_real_vb_mask
) - 1;
1132 count
= util_last_bit(mgr
->dirty_real_vb_mask
>> start_slot
);
1134 pipe
->set_vertex_buffers(pipe
, start_slot
, count
,
1135 mgr
->real_vertex_buffer
+ start_slot
);
1136 mgr
->dirty_real_vb_mask
= 0;
1139 void u_vbuf_draw_vbo(struct u_vbuf
*mgr
, const struct pipe_draw_info
*info
)
1141 struct pipe_context
*pipe
= mgr
->pipe
;
1142 int start_vertex
, min_index
;
1143 unsigned num_vertices
;
1144 boolean unroll_indices
= FALSE
;
1145 uint32_t used_vb_mask
= mgr
->ve
->used_vb_mask
;
1146 uint32_t user_vb_mask
= mgr
->user_vb_mask
& used_vb_mask
;
1147 uint32_t incompatible_vb_mask
= mgr
->incompatible_vb_mask
& used_vb_mask
;
1148 struct pipe_draw_info new_info
;
1150 /* Normal draw. No fallback and no user buffers. */
1151 if (!incompatible_vb_mask
&&
1152 !mgr
->ve
->incompatible_elem_mask
&&
1155 /* Set vertex buffers if needed. */
1156 if (mgr
->dirty_real_vb_mask
& used_vb_mask
) {
1157 u_vbuf_set_driver_vertex_buffers(mgr
);
1160 pipe
->draw_vbo(pipe
, info
);
1166 /* Fallback. We need to know all the parameters. */
1167 if (new_info
.indirect
) {
1168 struct pipe_transfer
*transfer
= NULL
;
1171 if (new_info
.indexed
) {
1172 data
= pipe_buffer_map_range(pipe
, new_info
.indirect
,
1173 new_info
.indirect_offset
, 20,
1174 PIPE_TRANSFER_READ
, &transfer
);
1175 new_info
.index_bias
= data
[3];
1176 new_info
.start_instance
= data
[4];
1179 data
= pipe_buffer_map_range(pipe
, new_info
.indirect
,
1180 new_info
.indirect_offset
, 16,
1181 PIPE_TRANSFER_READ
, &transfer
);
1182 new_info
.start_instance
= data
[3];
1185 new_info
.count
= data
[0];
1186 new_info
.instance_count
= data
[1];
1187 new_info
.start
= data
[2];
1188 pipe_buffer_unmap(pipe
, transfer
);
1189 new_info
.indirect
= NULL
;
1192 if (new_info
.indexed
) {
1193 /* See if anything needs to be done for per-vertex attribs. */
1194 if (u_vbuf_need_minmax_index(mgr
)) {
1197 if (new_info
.max_index
!= ~0u) {
1198 min_index
= new_info
.min_index
;
1199 max_index
= new_info
.max_index
;
1201 u_vbuf_get_minmax_index(mgr
->pipe
, &mgr
->index_buffer
,
1202 new_info
.primitive_restart
,
1203 new_info
.restart_index
, new_info
.start
,
1204 new_info
.count
, &min_index
, &max_index
);
1207 assert(min_index
<= max_index
);
1209 start_vertex
= min_index
+ new_info
.index_bias
;
1210 num_vertices
= max_index
+ 1 - min_index
;
1212 /* Primitive restart doesn't work when unrolling indices.
1213 * We would have to break this drawing operation into several ones. */
1214 /* Use some heuristic to see if unrolling indices improves
1216 if (!new_info
.primitive_restart
&&
1217 num_vertices
> new_info
.count
*2 &&
1218 num_vertices
- new_info
.count
> 32 &&
1219 !u_vbuf_mapping_vertex_buffer_blocks(mgr
)) {
1220 unroll_indices
= TRUE
;
1221 user_vb_mask
&= ~(mgr
->nonzero_stride_vb_mask
&
1222 mgr
->ve
->noninstance_vb_mask_any
);
1225 /* Nothing to do for per-vertex attribs. */
1231 start_vertex
= new_info
.start
;
1232 num_vertices
= new_info
.count
;
1236 /* Translate vertices with non-native layouts or formats. */
1237 if (unroll_indices
||
1238 incompatible_vb_mask
||
1239 mgr
->ve
->incompatible_elem_mask
) {
1240 if (!u_vbuf_translate_begin(mgr
, start_vertex
, num_vertices
,
1241 new_info
.start_instance
,
1242 new_info
.instance_count
, new_info
.start
,
1243 new_info
.count
, min_index
, unroll_indices
)) {
1244 debug_warn_once("u_vbuf_translate_begin() failed");
1248 if (unroll_indices
) {
1249 new_info
.indexed
= FALSE
;
1250 new_info
.index_bias
= 0;
1251 new_info
.min_index
= 0;
1252 new_info
.max_index
= new_info
.count
- 1;
1256 user_vb_mask
&= ~(incompatible_vb_mask
|
1257 mgr
->ve
->incompatible_vb_mask_all
);
1260 /* Upload user buffers. */
1262 if (u_vbuf_upload_buffers(mgr
, start_vertex
, num_vertices
,
1263 new_info
.start_instance
,
1264 new_info
.instance_count
) != PIPE_OK
) {
1265 debug_warn_once("u_vbuf_upload_buffers() failed");
1269 mgr
->dirty_real_vb_mask
|= user_vb_mask
;
1273 if (unroll_indices) {
1274 printf("unrolling indices: start_vertex = %i, num_vertices = %i\n",
1275 start_vertex, num_vertices);
1276 util_dump_draw_info(stdout, info);
1281 for (i = 0; i < mgr->nr_vertex_buffers; i++) {
1282 printf("input %i: ", i);
1283 util_dump_vertex_buffer(stdout, mgr->vertex_buffer+i);
1286 for (i = 0; i < mgr->nr_real_vertex_buffers; i++) {
1287 printf("real %i: ", i);
1288 util_dump_vertex_buffer(stdout, mgr->real_vertex_buffer+i);
1293 u_upload_unmap(pipe
->stream_uploader
);
1294 u_vbuf_set_driver_vertex_buffers(mgr
);
1296 pipe
->draw_vbo(pipe
, &new_info
);
1298 if (mgr
->using_translate
) {
1299 u_vbuf_translate_end(mgr
);
1303 void u_vbuf_save_vertex_elements(struct u_vbuf
*mgr
)
1305 assert(!mgr
->ve_saved
);
1306 mgr
->ve_saved
= mgr
->ve
;
1309 void u_vbuf_restore_vertex_elements(struct u_vbuf
*mgr
)
1311 if (mgr
->ve
!= mgr
->ve_saved
) {
1312 struct pipe_context
*pipe
= mgr
->pipe
;
1314 mgr
->ve
= mgr
->ve_saved
;
1315 pipe
->bind_vertex_elements_state(pipe
,
1316 mgr
->ve
? mgr
->ve
->driver_cso
: NULL
);
1318 mgr
->ve_saved
= NULL
;
1321 void u_vbuf_save_aux_vertex_buffer_slot(struct u_vbuf
*mgr
)
1323 struct pipe_vertex_buffer
*vb
=
1324 &mgr
->vertex_buffer
[mgr
->aux_vertex_buffer_slot
];
1326 pipe_resource_reference(&mgr
->aux_vertex_buffer_saved
.buffer
, vb
->buffer
);
1327 memcpy(&mgr
->aux_vertex_buffer_saved
, vb
, sizeof(*vb
));
1330 void u_vbuf_restore_aux_vertex_buffer_slot(struct u_vbuf
*mgr
)
1332 u_vbuf_set_vertex_buffers(mgr
, mgr
->aux_vertex_buffer_slot
, 1,
1333 &mgr
->aux_vertex_buffer_saved
);
1334 pipe_resource_reference(&mgr
->aux_vertex_buffer_saved
.buffer
, NULL
);