1 /**************************************************************************
3 Copyright 2002-2008 VMware, Inc.
7 Permission is hereby granted, free of charge, to any person obtaining a
8 copy of this software and associated documentation files (the "Software"),
9 to deal in the Software without restriction, including without limitation
10 on the rights to use, copy, modify, merge, publish, distribute, sub
11 license, and/or sell copies of the Software, and to permit persons to whom
12 the Software is furnished to do so, subject to the following conditions:
14 The above copyright notice and this permission notice (including the next
15 paragraph) shall be included in all copies or substantial portions of the
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
35 /* Display list compiler attempts to store lists of vertices with the
36 * same vertex layout. Additionally it attempts to minimize the need
37 * for execute-time fixup of these vertex lists, allowing them to be
40 * There are still some circumstances where this can be thwarted, for
41 * example by building a list that consists of one very long primitive
42 * (eg Begin(Triangles), 1000 vertices, End), and calling that list
43 * from inside a different begin/end object (Begin(Lines), CallList,
46 * In that case the code will have to replay the list as individual
47 * commands through the Exec dispatch table, or fix up the copied
48 * vertices at execute-time.
50 * The other case where fixup is required is when a vertex attribute
51 * is introduced in the middle of a primitive. Eg:
53 * TexCoord1f() Vertex2f()
54 * TexCoord1f() Color3f() Vertex2f()
57 * If the current value of Color isn't known at compile-time, this
58 * primitive will require fixup.
61 * The list compiler currently doesn't attempt to compile lists
62 * containing EvalCoord or EvalPoint commands. On encountering one of
63 * these, compilation falls back to opcodes.
65 * This could be improved to fallback only when a mix of EvalCoord and
66 * Vertex commands are issued within a single primitive.
68 * The compilation process works as follows. All vertex attributes
69 * except position are copied to vbo_save_context::attrptr (see ATTR_UNION).
70 * 'attrptr' are pointers to vbo_save_context::vertex ordered according to the enabled
71 * attributes (se upgrade_vertex).
72 * When the position attribute is received, all the attributes are then
73 * copied to the vertex_store (see the end of ATTR_UNION).
74 * The vertex_store is simply an extensible float array.
75 * When the vertex list needs to be compiled (see compile_vertex_list),
76 * several transformations are performed:
77 * - some primitives are merged together (eg: two consecutive GL_TRIANGLES
78 * with 3 vertices can be merged in a single GL_TRIANGLES with 6 vertices).
79 * - an index buffer is built.
80 * - identical vertices are detected and only one is kept.
81 * At the end of this transformation, the index buffer and the vertex buffer
82 * are uploaded in vRAM in the same buffer object.
83 * This buffer object is shared between multiple display list to allow
84 * draw calls merging later.
86 * The layout of this buffer for two display lists is:
87 * V0A0|V0A1|V1A0|V1A1|P0I0|P0I1|V0A0V0A1V0A2|V1A1V1A1V1A2|...
89 * - VxAy: vertex x, attributes y
90 * - PxIy: draw x, index y
92 * To allow draw call merging, display list must use the same VAO, including
93 * the same Offset in the buffer object. To achieve this, the start values of
94 * the primitive are shifted and the indices adjusted (see offset_diff and
95 * start_offset in compile_vertex_list).
97 * Display list using the loopback code (see vbo_save_playback_vertex_list_loopback),
98 * can't be drawn with an index buffer so this transformation is disabled
103 #include "main/glheader.h"
104 #include "main/arrayobj.h"
105 #include "main/bufferobj.h"
106 #include "main/context.h"
107 #include "main/dlist.h"
108 #include "main/enums.h"
109 #include "main/eval.h"
110 #include "main/macros.h"
111 #include "main/draw_validate.h"
112 #include "main/api_arrayelt.h"
113 #include "main/dispatch.h"
114 #include "main/state.h"
115 #include "main/varray.h"
116 #include "util/bitscan.h"
117 #include "util/u_memory.h"
118 #include "util/hash_table.h"
119 #include "util/indices/u_indices.h"
120 #include "util/u_prim.h"
122 #include "gallium/include/pipe/p_state.h"
124 #include "vbo_private.h"
125 #include "api_exec_decl.h"
126 #include "api_save.h"
132 /* An interesting VBO number/name to help with debugging */
133 #define VBO_BUF_ID 12345
135 static void GLAPIENTRY
136 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params);
138 static void GLAPIENTRY
139 _save_EvalCoord1f(GLfloat u);
141 static void GLAPIENTRY
142 _save_EvalCoord2f(GLfloat u, GLfloat v);
145 * NOTE: Old 'parity' issue is gone, but copying can still be
146 * wrong-footed on replay.
149 copy_vertices(struct gl_context *ctx,
150 const struct vbo_save_vertex_list *node,
151 const fi_type * src_buffer)
153 struct vbo_save_context *save = &vbo_context(ctx)->save;
154 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
155 GLuint sz = save->vertex_size;
157 if (prim->end || !prim->count || !sz)
160 const fi_type *src = src_buffer + prim->start * sz;
161 assert(save->copied.buffer == NULL);
162 save->copied.buffer = malloc(sizeof(fi_type) * sz * prim->count);
164 unsigned r = vbo_copy_vertices(ctx, prim->mode, prim->start, &prim->count,
165 prim->begin, sz, true, save->copied.buffer, src);
167 free(save->copied.buffer);
168 save->copied.buffer = NULL;
174 static struct vbo_save_primitive_store *
175 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
178 store = CALLOC_STRUCT(vbo_save_primitive_store);
180 uint32_t old_size = store->size;
181 store->size = prim_count;
182 assert (old_size < store->size);
183 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
184 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
191 reset_counters(struct gl_context *ctx)
193 struct vbo_save_context *save = &vbo_context(ctx)->save;
195 save->vertex_store->used = 0;
196 save->prim_store->used = 0;
197 save->dangling_attr_ref = GL_FALSE;
201 * For a list of prims, try merging prims that can just be extensions of the
205 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
209 struct _mesa_prim *prev_prim = prim_list;
211 for (i = 1; i < *prim_count; i++) {
212 struct _mesa_prim *this_prim = prim_list + i;
214 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
216 if (vbo_merge_draws(ctx, true,
217 prev_prim->mode, this_prim->mode,
218 prev_prim->start, this_prim->start,
219 &prev_prim->count, this_prim->count,
220 prev_prim->basevertex, this_prim->basevertex,
222 this_prim->begin, this_prim->end)) {
223 /* We've found a prim that just extend the previous one. Tack it
224 * onto the previous one, and let this primitive struct get dropped.
229 /* If any previous primitives have been dropped, then we need to copy
230 * this later one into the next available slot.
233 if (prev_prim != this_prim)
234 *prev_prim = *this_prim;
237 *prim_count = prev_prim - prim_list + 1;
242 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
243 * don't have to worry about handling the _mesa_prim::begin/end flags.
244 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
247 convert_line_loop_to_strip(struct vbo_save_context *save,
248 struct vbo_save_vertex_list *node)
250 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
252 assert(prim->mode == GL_LINE_LOOP);
255 /* Copy the 0th vertex to end of the buffer and extend the
256 * vertex count by one to finish the line loop.
258 const GLuint sz = save->vertex_size;
260 const fi_type *src = save->vertex_store->buffer_in_ram + prim->start * sz;
262 fi_type *dst = save->vertex_store->buffer_in_ram + (prim->start + prim->count) * sz;
264 memcpy(dst, src, sz * sizeof(float));
267 node->cold->vertex_count++;
268 save->vertex_store->used += sz;
272 /* Drawing the second or later section of a long line loop.
273 * Skip the 0th vertex.
279 prim->mode = GL_LINE_STRIP;
283 /* Compare the present vao if it has the same setup. */
285 compare_vao(gl_vertex_processing_mode mode,
286 const struct gl_vertex_array_object *vao,
287 const struct gl_buffer_object *bo, GLintptr buffer_offset,
288 GLuint stride, GLbitfield64 vao_enabled,
289 const GLubyte size[VBO_ATTRIB_MAX],
290 const GLenum16 type[VBO_ATTRIB_MAX],
291 const GLuint offset[VBO_ATTRIB_MAX])
296 /* If the enabled arrays are not the same we are not equal. */
297 if (vao_enabled != vao->Enabled)
300 /* Check the buffer binding at 0 */
301 if (vao->BufferBinding[0].BufferObj != bo)
303 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
304 if (vao->BufferBinding[0].Stride != stride)
306 assert(vao->BufferBinding[0].InstanceDivisor == 0);
308 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
309 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
311 /* Now check the enabled arrays */
312 GLbitfield mask = vao_enabled;
314 const int attr = u_bit_scan(&mask);
315 const unsigned char vbo_attr = vao_to_vbo_map[attr];
316 const GLenum16 tp = type[vbo_attr];
317 const GLintptr off = offset[vbo_attr] + buffer_offset;
318 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
319 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
321 if (attrib->Format.Type != tp)
323 if (attrib->Format.Size != size[vbo_attr])
325 assert(attrib->Format.Format == GL_RGBA);
326 assert(attrib->Format.Normalized == GL_FALSE);
327 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
328 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
329 assert(attrib->BufferBindingIndex == 0);
336 /* Create or reuse the vao for the vertex processing mode. */
338 update_vao(struct gl_context *ctx,
339 gl_vertex_processing_mode mode,
340 struct gl_vertex_array_object **vao,
341 struct gl_buffer_object *bo, GLintptr buffer_offset,
342 GLuint stride, GLbitfield64 vbo_enabled,
343 const GLubyte size[VBO_ATTRIB_MAX],
344 const GLenum16 type[VBO_ATTRIB_MAX],
345 const GLuint offset[VBO_ATTRIB_MAX])
347 /* Compute the bitmasks of vao_enabled arrays */
348 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
351 * Check if we can possibly reuse the exisiting one.
352 * In the long term we should reset them when something changes.
354 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
355 vao_enabled, size, type, offset))
358 /* The initial refcount is 1 */
359 _mesa_reference_vao(ctx, vao, NULL);
360 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
363 * assert(stride <= ctx->Const.MaxVertexAttribStride);
364 * MaxVertexAttribStride is not set for drivers that does not
365 * expose GL 44 or GLES 31.
368 /* Bind the buffer object at binding point 0 */
369 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
372 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
373 * Note that the position/generic0 aliasing is done in the VAO.
375 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
376 /* Now set the enable arrays */
377 GLbitfield mask = vao_enabled;
379 const int vao_attr = u_bit_scan(&mask);
380 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
381 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
383 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
384 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
385 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
387 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
388 assert(vao_enabled == (*vao)->Enabled);
389 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
391 /* Finalize and freeze the VAO */
392 _mesa_set_vao_immutable(ctx, *vao);
395 static void wrap_filled_vertex(struct gl_context *ctx);
397 /* Grow the vertex storage to accomodate for vertex_count new vertices */
399 grow_vertex_storage(struct gl_context *ctx, int vertex_count)
401 struct vbo_save_context *save = &vbo_context(ctx)->save;
402 assert (save->vertex_store);
404 int new_size = (save->vertex_store->used +
405 vertex_count * save->vertex_size) * sizeof(GLfloat);
407 /* Limit how much memory we allocate. */
408 if (save->prim_store->used > 0 &&
410 new_size > VBO_SAVE_BUFFER_SIZE) {
411 wrap_filled_vertex(ctx);
412 new_size = VBO_SAVE_BUFFER_SIZE;
415 if (new_size > save->vertex_store->buffer_in_ram_size) {
416 save->vertex_store->buffer_in_ram_size = new_size;
417 save->vertex_store->buffer_in_ram = realloc(save->vertex_store->buffer_in_ram,
418 save->vertex_store->buffer_in_ram_size);
419 if (save->vertex_store->buffer_in_ram == NULL)
420 save->out_of_memory = true;
425 unsigned vertex_size;
426 fi_type *vertex_attributes;
429 static uint32_t _hash_vertex_key(const void *key)
431 struct vertex_key *k = (struct vertex_key*)key;
432 unsigned sz = k->vertex_size;
434 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
437 static bool _compare_vertex_key(const void *key1, const void *key2)
439 struct vertex_key *k1 = (struct vertex_key*)key1;
440 struct vertex_key *k2 = (struct vertex_key*)key2;
441 /* All the compared vertices are going to be drawn with the same VAO,
442 * so we can compare the attributes. */
443 assert (k1->vertex_size == k2->vertex_size);
444 return memcmp(k1->vertex_attributes,
445 k2->vertex_attributes,
446 k1->vertex_size * sizeof(float)) == 0;
449 static void _free_entry(struct hash_entry *entry)
451 free((void*)entry->key);
454 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
455 * of an existing vertex, return the original index instead.
458 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
459 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
461 /* If vertex deduplication is disabled return the original index. */
465 fi_type *vert = save->vertex_store->buffer_in_ram + save->vertex_size * index;
467 struct vertex_key *key = malloc(sizeof(struct vertex_key));
468 key->vertex_size = save->vertex_size;
469 key->vertex_attributes = vert;
471 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
474 /* We found an existing vertex with the same hash, return its index. */
475 return (uintptr_t) entry->data;
477 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
478 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
479 * starting at index 0.
481 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
482 *max_index = MAX2(n, *max_index);
484 memcpy(&new_buffer[save->vertex_size * n],
486 save->vertex_size * sizeof(fi_type));
488 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
490 /* The index buffer is shared between list compilations, so add the base index to get
499 get_vertex_count(struct vbo_save_context *save)
501 if (!save->vertex_size)
503 return save->vertex_store->used / save->vertex_size;
508 * Insert the active immediate struct onto the display list currently
512 compile_vertex_list(struct gl_context *ctx)
514 struct vbo_save_context *save = &vbo_context(ctx)->save;
515 struct vbo_save_vertex_list *node;
517 /* Allocate space for this structure in the display list currently
520 node = (struct vbo_save_vertex_list *)
521 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
526 node->cold = calloc(1, sizeof(*node->cold));
528 /* Make sure the pointer is aligned to the size of a pointer */
529 assert((GLintptr) node % sizeof(void *) == 0);
531 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
533 node->cold->vertex_count = get_vertex_count(save);
534 node->cold->wrap_count = save->copied.nr;
535 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
536 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
537 node->cold->ib.obj = NULL;
538 node->cold->prim_count = save->prim_store->used;
540 if (save->no_current_update) {
541 node->cold->current_data = NULL;
544 GLuint current_size = save->vertex_size - save->attrsz[0];
545 node->cold->current_data = NULL;
548 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
549 if (node->cold->current_data) {
550 const char *buffer = (const char *)save->vertex_store->buffer_in_ram;
551 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
552 unsigned vertex_offset = 0;
554 if (node->cold->vertex_count)
555 vertex_offset = (node->cold->vertex_count - 1) * stride;
557 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
558 current_size * sizeof(GLfloat));
560 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
561 save->out_of_memory = true;
566 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
568 if (save->dangling_attr_ref)
569 ctx->ListState.Current.UseLoopback = true;
571 /* Copy duplicated vertices
573 save->copied.nr = copy_vertices(ctx, node, save->vertex_store->buffer_in_ram);
575 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
576 convert_line_loop_to_strip(save, node);
579 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
581 GLintptr buffer_offset = 0;
582 GLuint start_offset = 0;
584 /* Create an index buffer. */
585 node->cold->min_index = node->cold->max_index = 0;
586 if (node->cold->vertex_count == 0 || node->cold->prim_count == 0)
589 /* We won't modify node->prims, so use a const alias to avoid unintended
591 const struct _mesa_prim *original_prims = node->cold->prims;
593 int end = original_prims[node->cold->prim_count - 1].start +
594 original_prims[node->cold->prim_count - 1].count;
595 int total_vert_count = end - original_prims[0].start;
597 node->cold->min_index = node->cold->prims[0].start;
598 node->cold->max_index = end - 1;
600 /* converting primitive types may result in many more indices */
601 bool all_prims_supported = (ctx->Const.DriverSupportedPrimMask & BITFIELD_MASK(PIPE_PRIM_MAX)) == BITFIELD_MASK(PIPE_PRIM_MAX);
602 int max_index_count = total_vert_count * (all_prims_supported ? 2 : 3);
603 uint32_t* indices = (uint32_t*) malloc(max_index_count * sizeof(uint32_t));
604 void *tmp_indices = all_prims_supported ? NULL : malloc(max_index_count * sizeof(uint32_t));
605 struct _mesa_prim *merged_prims = NULL;
608 struct hash_table *vertex_to_index = NULL;
609 fi_type *temp_vertices_buffer = NULL;
611 /* The loopback replay code doesn't use the index buffer, so we can't
612 * dedup vertices in this case.
614 if (!ctx->ListState.Current.UseLoopback) {
615 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
616 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
619 uint32_t max_index = 0;
621 int last_valid_prim = -1;
622 /* Construct indices array. */
623 for (unsigned i = 0; i < node->cold->prim_count; i++) {
624 assert(original_prims[i].basevertex == 0);
625 GLubyte mode = original_prims[i].mode;
626 bool converted_prim = false;
629 int vertex_count = original_prims[i].count;
634 /* Increase indices storage if the original estimation was too small. */
635 if (idx + 3 * vertex_count > max_index_count) {
636 max_index_count = max_index_count + 3 * vertex_count;
637 indices = (uint32_t*) realloc(indices, max_index_count * sizeof(uint32_t));
638 tmp_indices = all_prims_supported ? NULL : realloc(tmp_indices, max_index_count * sizeof(uint32_t));
641 /* Line strips may get converted to lines */
642 if (mode == GL_LINE_STRIP)
645 if (!(ctx->Const.DriverSupportedPrimMask & BITFIELD_BIT(mode))) {
647 u_generate_func trans_func;
648 enum pipe_prim_type pmode = (enum pipe_prim_type)mode;
649 u_index_generator(ctx->Const.DriverSupportedPrimMask,
650 pmode, original_prims[i].start, vertex_count,
652 &pmode, &index_size, &new_count,
655 trans_func(original_prims[i].start, new_count, tmp_indices);
656 vertex_count = new_count;
657 mode = (GLubyte)pmode;
658 converted_prim = true;
661 /* If 2 consecutive prims use the same mode => merge them. */
662 bool merge_prims = last_valid_prim >= 0 &&
663 mode == merged_prims[last_valid_prim].mode &&
664 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
665 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
668 /* index generation uses uint16_t if the index count is small enough */
669 #define CAST_INDEX(BASE, SIZE, IDX) ((SIZE == 2 ? (uint32_t)(((uint16_t*)BASE)[IDX]) : ((uint32_t*)BASE)[IDX]))
670 /* To be able to merge consecutive triangle strips we need to insert
671 * a degenerate triangle.
674 mode == GL_TRIANGLE_STRIP) {
675 /* Insert a degenerate triangle */
676 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
677 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
679 indices[idx] = indices[idx - 1];
680 indices[idx + 1] = add_vertex(save, vertex_to_index,
681 converted_prim ? CAST_INDEX(tmp_indices, index_size, 0) : original_prims[i].start,
682 temp_vertices_buffer, &max_index);
684 merged_prims[last_valid_prim].count += 2;
687 /* Add another index to preserve winding order */
688 indices[idx++] = add_vertex(save, vertex_to_index,
689 converted_prim ? CAST_INDEX(tmp_indices, index_size, 0) : original_prims[i].start,
690 temp_vertices_buffer, &max_index);
691 merged_prims[last_valid_prim].count++;
697 /* Convert line strips to lines if it'll allow if the previous
698 * prim mode is GL_LINES (so merge_prims is true) or if the next
699 * primitive mode is GL_LINES or GL_LINE_LOOP.
701 if (original_prims[i].mode == GL_LINE_STRIP &&
703 (i < node->cold->prim_count - 1 &&
704 (original_prims[i + 1].mode == GL_LINE_STRIP ||
705 original_prims[i + 1].mode == GL_LINES)))) {
706 for (unsigned j = 0; j < vertex_count; j++) {
707 indices[idx++] = add_vertex(save, vertex_to_index,
708 converted_prim ? CAST_INDEX(tmp_indices, index_size, j) : original_prims[i].start + j,
709 temp_vertices_buffer, &max_index);
710 /* Repeat all but the first/last indices. */
711 if (j && j != vertex_count - 1) {
712 indices[idx++] = add_vertex(save, vertex_to_index,
713 converted_prim ? CAST_INDEX(tmp_indices, index_size, j) : original_prims[i].start + j,
714 temp_vertices_buffer, &max_index);
718 /* We didn't convert to LINES, so restore the original mode */
720 mode = original_prims[i].mode;
722 for (unsigned j = 0; j < vertex_count; j++) {
723 indices[idx++] = add_vertex(save, vertex_to_index,
724 converted_prim ? CAST_INDEX(tmp_indices, index_size, j) : original_prims[i].start + j,
725 temp_vertices_buffer, &max_index);
729 /* Duplicate the last vertex for incomplete primitives */
730 if (vertex_count > 0) {
731 unsigned min_vert = u_prim_vertex_count(mode)->min;
732 for (unsigned j = vertex_count; j < min_vert; j++) {
733 indices[idx++] = add_vertex(save, vertex_to_index,
734 converted_prim ? CAST_INDEX(tmp_indices, index_size, vertex_count - 1) :
735 original_prims[i].start + vertex_count - 1,
736 temp_vertices_buffer, &max_index);
742 /* Update vertex count. */
743 merged_prims[last_valid_prim].count += idx - start;
745 /* Keep this primitive */
746 last_valid_prim += 1;
747 assert(last_valid_prim <= i);
748 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
749 merged_prims[last_valid_prim] = original_prims[i];
750 merged_prims[last_valid_prim].start = start;
751 merged_prims[last_valid_prim].count = idx - start;
753 merged_prims[last_valid_prim].mode = mode;
755 /* converted prims will filter incomplete primitives and may have no indices */
756 assert((idx > 0 || converted_prim) && idx <= max_index_count);
759 unsigned merged_prim_count = last_valid_prim + 1;
760 node->cold->ib.ptr = NULL;
761 node->cold->ib.count = idx;
762 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
764 /* How many bytes do we need to store the indices and the vertices */
765 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
766 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
767 total_vert_count * save->vertex_size * sizeof(fi_type);
769 const GLintptr old_offset = save->VAO[0] ?
770 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
771 if (old_offset != save->current_bo_bytes_used && stride > 0) {
772 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
773 while (offset_diff > 0 &&
774 save->current_bo_bytes_used < save->current_bo->Size &&
775 offset_diff % stride != 0) {
776 save->current_bo_bytes_used++;
777 offset_diff = save->current_bo_bytes_used - old_offset;
780 buffer_offset = save->current_bo_bytes_used;
782 /* Can we reuse the previous bo or should we allocate a new one? */
783 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
784 if (total_bytes_needed > available_bytes) {
785 if (save->current_bo)
786 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
787 save->current_bo = _mesa_bufferobj_alloc(ctx, VBO_BUF_ID + 1);
788 bool success = _mesa_bufferobj_data(ctx,
789 GL_ELEMENT_ARRAY_BUFFER_ARB,
790 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE),
792 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
793 MESA_GALLIUM_VERTEX_STATE_STORAGE,
796 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
797 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
798 save->out_of_memory = true;
800 save->current_bo_bytes_used = 0;
801 available_bytes = save->current_bo->Size;
805 assert(old_offset <= buffer_offset);
806 const GLintptr offset_diff = buffer_offset - old_offset;
807 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
808 /* The vertex size is an exact multiple of the buffer offset.
809 * This means that we can use zero-based vertex attribute pointers
810 * and specify the start of the primitive with the _mesa_prim::start
811 * field. This results in issuing several draw calls with identical
812 * vertex attribute information. This can result in fewer state
813 * changes in drivers. In particular, the Gallium CSO module will
814 * filter out redundant vertex buffer changes.
816 /* We cannot immediately update the primitives as some methods below
817 * still need the uncorrected start vertices
819 start_offset = offset_diff/stride;
820 assert(old_offset == buffer_offset - offset_diff);
821 buffer_offset = old_offset;
824 /* Correct the primitive starts, we can only do this here as copy_vertices
825 * and convert_line_loop_to_strip above consume the uncorrected starts.
826 * On the other hand the _vbo_loopback_vertex_list call below needs the
827 * primitives to be corrected already.
829 for (unsigned i = 0; i < node->cold->prim_count; i++) {
830 node->cold->prims[i].start += start_offset;
832 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
833 * to apply this transformation to all indices and max_index.
835 for (unsigned i = 0; i < idx; i++)
836 indices[i] += start_offset;
837 max_index += start_offset;
840 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
842 /* Upload the vertices first (see buffer_offset) */
843 _mesa_bufferobj_subdata(ctx,
844 save->current_bo_bytes_used,
845 total_vert_count * save->vertex_size * sizeof(fi_type),
846 vertex_to_index ? temp_vertices_buffer : save->vertex_store->buffer_in_ram,
848 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
850 if (vertex_to_index) {
851 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
852 free(temp_vertices_buffer);
855 /* Since we append the indices to an existing buffer, we need to adjust the start value of each
856 * primitive (not the indices themselves). */
857 if (!ctx->ListState.Current.UseLoopback) {
858 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
859 int indices_offset = save->current_bo_bytes_used / 4;
860 for (int i = 0; i < merged_prim_count; i++) {
861 merged_prims[i].start += indices_offset;
865 /* Then upload the indices. */
866 if (node->cold->ib.obj) {
867 _mesa_bufferobj_subdata(ctx,
868 save->current_bo_bytes_used,
869 idx * sizeof(uint32_t),
872 save->current_bo_bytes_used += idx * sizeof(uint32_t);
874 node->cold->vertex_count = 0;
875 node->cold->prim_count = 0;
878 /* Prepare for DrawGallium */
879 memset(&node->cold->info, 0, sizeof(struct pipe_draw_info));
880 /* The other info fields will be updated in vbo_save_playback_vertex_list */
881 node->cold->info.index_size = 4;
882 node->cold->info.instance_count = 1;
883 node->cold->info.index.gl_bo = node->cold->ib.obj;
884 if (merged_prim_count == 1) {
885 node->cold->info.mode = merged_prims[0].mode;
886 node->start_count.start = merged_prims[0].start;
887 node->start_count.count = merged_prims[0].count;
888 node->start_count.index_bias = 0;
891 node->modes = malloc(merged_prim_count * sizeof(unsigned char));
892 node->start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
893 for (unsigned i = 0; i < merged_prim_count; i++) {
894 node->start_counts[i].start = merged_prims[i].start;
895 node->start_counts[i].count = merged_prims[i].count;
896 node->start_counts[i].index_bias = 0;
897 node->modes[i] = merged_prims[i].mode;
900 node->num_draws = merged_prim_count;
901 if (node->num_draws > 1) {
902 bool same_mode = true;
903 for (unsigned i = 1; i < node->num_draws && same_mode; i++) {
904 same_mode = node->modes[i] == node->modes[0];
907 /* All primitives use the same mode, so we can simplify a bit */
908 node->cold->info.mode = node->modes[0];
919 node->draw_begins = node->cold->prims[0].begin;
921 if (!save->current_bo) {
922 save->current_bo = _mesa_bufferobj_alloc(ctx, VBO_BUF_ID + 1);
923 bool success = _mesa_bufferobj_data(ctx,
924 GL_ELEMENT_ARRAY_BUFFER_ARB,
925 VBO_SAVE_BUFFER_SIZE,
927 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
928 MESA_GALLIUM_VERTEX_STATE_STORAGE,
931 save->out_of_memory = true;
934 GLuint offsets[VBO_ATTRIB_MAX];
935 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
937 offset += save->attrsz[i] * sizeof(GLfloat);
939 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
940 * Note that this may reuse the previous one of possible.
942 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
943 /* create or reuse the vao */
944 update_vao(ctx, vpm, &save->VAO[vpm],
945 save->current_bo, buffer_offset, stride,
946 save->enabled, save->attrsz, save->attrtype, offsets);
947 /* Reference the vao in the dlist */
948 node->cold->VAO[vpm] = NULL;
949 _mesa_reference_vao(ctx, &node->cold->VAO[vpm], save->VAO[vpm]);
952 /* Prepare for DrawGalliumVertexState */
953 if (node->num_draws && ctx->Driver.DrawGalliumVertexState) {
954 for (unsigned i = 0; i < VP_MODE_MAX; i++) {
955 uint32_t enabled_attribs = _vbo_get_vao_filter(i) &
956 node->cold->VAO[i]->_EnabledWithMapMode;
959 ctx->Driver.CreateGalliumVertexState(ctx, node->cold->VAO[i],
962 node->private_refcount[i] = 0;
963 node->enabled_attribs[i] = enabled_attribs;
967 node->mode = node->cold->info.mode;
968 assert(node->cold->info.index_size == 4);
971 /* Deal with GL_COMPILE_AND_EXECUTE:
973 if (ctx->ExecuteFlag) {
974 struct _glapi_table *dispatch = GET_DISPATCH();
976 _glapi_set_dispatch(ctx->Exec);
978 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
979 * use buffer_in_ram (which contains all vertices) instead of current_bo
980 * (which contains deduplicated vertices *when* UseLoopback is false).
982 * The problem is that the VAO offset is based on current_bo's layout,
983 * so we have to use a temp value.
985 struct gl_vertex_array_object *vao = node->cold->VAO[VP_MODE_SHADER];
986 GLintptr original = vao->BufferBinding[0].Offset;
987 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
988 vao->BufferBinding[0].Offset = -(GLintptr)(start_offset * stride);
989 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
990 vao->BufferBinding[0].Offset = original;
992 _glapi_set_dispatch(dispatch);
995 /* Reset our structures for the next run of vertices:
1002 * This is called when we fill a vertex buffer before we hit a glEnd().
1004 * TODO -- If no new vertices have been stored, don't bother saving it.
1007 wrap_buffers(struct gl_context *ctx)
1009 struct vbo_save_context *save = &vbo_context(ctx)->save;
1010 GLint i = save->prim_store->used - 1;
1013 assert(i < (GLint) save->prim_store->size);
1016 /* Close off in-progress primitive.
1018 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1019 mode = save->prim_store->prims[i].mode;
1021 /* store the copied vertices, and allocate a new list.
1023 compile_vertex_list(ctx);
1025 /* Restart interrupted primitive
1027 save->prim_store->prims[0].mode = mode;
1028 save->prim_store->prims[0].begin = 0;
1029 save->prim_store->prims[0].end = 0;
1030 save->prim_store->prims[0].start = 0;
1031 save->prim_store->prims[0].count = 0;
1032 save->prim_store->used = 1;
1037 * Called only when buffers are wrapped as the result of filling the
1038 * vertex_store struct.
1041 wrap_filled_vertex(struct gl_context *ctx)
1043 struct vbo_save_context *save = &vbo_context(ctx)->save;
1044 unsigned numComponents;
1046 /* Emit a glEnd to close off the last vertex list.
1050 assert(save->vertex_store->used == 0 && save->vertex_store->used == 0);
1052 /* Copy stored stored vertices to start of new list.
1054 numComponents = save->copied.nr * save->vertex_size;
1056 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
1057 if (numComponents) {
1058 assert(save->copied.buffer);
1060 save->copied.buffer,
1061 numComponents * sizeof(fi_type));
1062 free(save->copied.buffer);
1063 save->copied.buffer = NULL;
1065 save->vertex_store->used = numComponents;
1070 copy_to_current(struct gl_context *ctx)
1072 struct vbo_save_context *save = &vbo_context(ctx)->save;
1073 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1076 const int i = u_bit_scan64(&enabled);
1077 assert(save->attrsz[i]);
1079 if (save->attrtype[i] == GL_DOUBLE ||
1080 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1081 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1083 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1084 save->attrptr[i], save->attrtype[i]);
1090 copy_from_current(struct gl_context *ctx)
1092 struct vbo_save_context *save = &vbo_context(ctx)->save;
1093 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1096 const int i = u_bit_scan64(&enabled);
1098 switch (save->attrsz[i]) {
1100 save->attrptr[i][3] = save->current[i][3];
1103 save->attrptr[i][2] = save->current[i][2];
1106 save->attrptr[i][1] = save->current[i][1];
1109 save->attrptr[i][0] = save->current[i][0];
1112 unreachable("Unexpected vertex attribute size");
1119 * Called when we increase the size of a vertex attribute. For example,
1120 * if we've seen one or more glTexCoord2f() calls and now we get a
1121 * glTexCoord3f() call.
1122 * Flush existing data, set new attrib size, replay copied vertices.
1125 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1127 struct vbo_save_context *save = &vbo_context(ctx)->save;
1132 /* Store the current run of vertices, and emit a GL_END. Emit a
1133 * BEGIN in the new buffer.
1135 if (save->vertex_store->used)
1138 assert(save->copied.nr == 0);
1140 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1141 * when the attribute already exists in the vertex and is having
1142 * its size increased.
1144 copy_to_current(ctx);
1148 oldsz = save->attrsz[attr];
1149 save->attrsz[attr] = newsz;
1150 save->enabled |= BITFIELD64_BIT(attr);
1152 save->vertex_size += newsz - oldsz;
1154 /* Recalculate all the attrptr[] values:
1157 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1158 if (save->attrsz[i]) {
1159 save->attrptr[i] = tmp;
1160 tmp += save->attrsz[i];
1163 save->attrptr[i] = NULL; /* will not be dereferenced. */
1167 /* Copy from current to repopulate the vertex with correct values.
1169 copy_from_current(ctx);
1171 /* Replay stored vertices to translate them to new format here.
1173 * If there are copied vertices and the new (upgraded) attribute
1174 * has not been defined before, this list is somewhat degenerate,
1175 * and will need fixup at runtime.
1177 if (save->copied.nr) {
1178 assert(save->copied.buffer);
1179 const fi_type *data = save->copied.buffer;
1180 grow_vertex_storage(ctx, save->copied.nr);
1181 fi_type *dest = save->vertex_store->buffer_in_ram;
1183 /* Need to note this and fix up at runtime (or loopback):
1185 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1187 save->dangling_attr_ref = GL_TRUE;
1190 for (i = 0; i < save->copied.nr; i++) {
1191 GLbitfield64 enabled = save->enabled;
1193 const int j = u_bit_scan64(&enabled);
1194 assert(save->attrsz[j]);
1197 const fi_type *src = oldsz ? data : save->current[attr];
1198 int copy = oldsz ? oldsz : newsz;
1199 for (k = 0; k < copy; k++)
1201 for (; k < newsz; k++) {
1202 switch (save->attrtype[j]) {
1204 dest[k] = FLOAT_AS_UNION(k == 3);
1207 dest[k] = INT_AS_UNION(k == 3);
1209 case GL_UNSIGNED_INT:
1210 dest[k] = UINT_AS_UNION(k == 3);
1213 dest[k] = FLOAT_AS_UNION(k == 3);
1214 assert(!"Unexpected type in upgrade_vertex");
1221 GLint sz = save->attrsz[j];
1222 for (int k = 0; k < sz; k++)
1230 save->vertex_store->used += save->vertex_size * save->copied.nr;
1231 free(save->copied.buffer);
1232 save->copied.buffer = NULL;
1238 * This is called when the size of a vertex attribute changes.
1239 * For example, after seeing one or more glTexCoord2f() calls we
1240 * get a glTexCoord4f() or glTexCoord1f() call.
1243 fixup_vertex(struct gl_context *ctx, GLuint attr,
1244 GLuint sz, GLenum newType)
1246 struct vbo_save_context *save = &vbo_context(ctx)->save;
1248 if (sz > save->attrsz[attr] ||
1249 newType != save->attrtype[attr]) {
1250 /* New size is larger. Need to flush existing vertices and get
1251 * an enlarged vertex format.
1253 upgrade_vertex(ctx, attr, sz);
1255 else if (sz < save->active_sz[attr]) {
1257 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1259 /* New size is equal or smaller - just need to fill in some
1262 for (i = sz; i <= save->attrsz[attr]; i++)
1263 save->attrptr[attr][i - 1] = id[i - 1];
1266 save->active_sz[attr] = sz;
1268 grow_vertex_storage(ctx, 1);
1273 * Reset the current size of all vertex attributes to the default
1274 * value of 0. This signals that we haven't yet seen any per-vertex
1275 * commands such as glNormal3f() or glTexCoord2f().
1278 reset_vertex(struct gl_context *ctx)
1280 struct vbo_save_context *save = &vbo_context(ctx)->save;
1282 while (save->enabled) {
1283 const int i = u_bit_scan64(&save->enabled);
1284 assert(save->attrsz[i]);
1285 save->attrsz[i] = 0;
1286 save->active_sz[i] = 0;
1289 save->vertex_size = 0;
1294 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1295 * It depends on a few things, including whether we're inside or outside
1299 is_vertex_position(const struct gl_context *ctx, GLuint index)
1301 return (index == 0 &&
1302 _mesa_attr_zero_aliases_vertex(ctx) &&
1303 _mesa_inside_dlist_begin_end(ctx));
1308 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1311 /* Only one size for each attribute may be active at once. Eg. if
1312 * Color3f is installed/active, then Color4f may not be, even if the
1313 * vertex actually contains 4 color coordinates. This is because the
1314 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1315 * of the chooser function when switching between Color4f and Color3f.
1317 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1319 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1320 int sz = (sizeof(C) / sizeof(GLfloat)); \
1322 if (save->active_sz[A] != N) \
1323 fixup_vertex(ctx, A, N * sz, T); \
1326 C *dest = (C *)save->attrptr[A]; \
1327 if (N>0) dest[0] = V0; \
1328 if (N>1) dest[1] = V1; \
1329 if (N>2) dest[2] = V2; \
1330 if (N>3) dest[3] = V3; \
1331 save->attrtype[A] = T; \
1334 if ((A) == VBO_ATTRIB_POS) { \
1335 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + \
1336 save->vertex_store->used; \
1338 for (int i = 0; i < save->vertex_size; i++) \
1339 buffer_ptr[i] = save->vertex[i]; \
1341 save->vertex_store->used += save->vertex_size; \
1342 unsigned used_next = (save->vertex_store->used + \
1343 save->vertex_size) * sizeof(float); \
1344 if (used_next > save->vertex_store->buffer_in_ram_size) { \
1345 grow_vertex_storage(ctx, get_vertex_count(save)); \
1346 assert(used_next <= \
1347 save->vertex_store->buffer_in_ram_size); \
1352 #define TAG(x) _save_##x
1354 #include "vbo_attrib_tmp.h"
1357 #define MAT( ATTR, N, face, params ) \
1359 if (face != GL_BACK) \
1360 MAT_ATTR( ATTR, N, params ); /* front */ \
1361 if (face != GL_FRONT) \
1362 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1367 * Save a glMaterial call found between glBegin/End.
1368 * glMaterial calls outside Begin/End are handled in dlist.c.
1370 static void GLAPIENTRY
1371 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1373 GET_CURRENT_CONTEXT(ctx);
1375 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1376 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1382 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1385 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1388 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1391 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1394 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1395 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1398 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1401 case GL_COLOR_INDEXES:
1402 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1404 case GL_AMBIENT_AND_DIFFUSE:
1405 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1406 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1409 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1416 vbo_install_save_vtxfmt(struct gl_context *ctx);
1419 /* Cope with EvalCoord/CallList called within a begin/end object:
1420 * -- Flush current buffer
1421 * -- Fallback to opcodes for the rest of the begin/end object.
1424 dlist_fallback(struct gl_context *ctx)
1426 struct vbo_save_context *save = &vbo_context(ctx)->save;
1428 if (save->vertex_store->used || save->prim_store->used) {
1429 if (save->prim_store->used > 0 && save->vertex_store->used > 0) {
1430 assert(save->vertex_size);
1431 /* Close off in-progress primitive. */
1432 GLint i = save->prim_store->used - 1;
1433 save->prim_store->prims[i].count =
1434 get_vertex_count(save) -
1435 save->prim_store->prims[i].start;
1438 /* Need to replay this display list with loopback,
1439 * unfortunately, otherwise this primitive won't be handled
1442 save->dangling_attr_ref = GL_TRUE;
1444 compile_vertex_list(ctx);
1447 copy_to_current(ctx);
1449 if (save->out_of_memory) {
1450 vbo_install_save_vtxfmt_noop(ctx);
1453 _mesa_install_save_vtxfmt(ctx);
1455 ctx->Driver.SaveNeedFlush = GL_FALSE;
1459 static void GLAPIENTRY
1460 _save_EvalCoord1f(GLfloat u)
1462 GET_CURRENT_CONTEXT(ctx);
1463 dlist_fallback(ctx);
1464 CALL_EvalCoord1f(ctx->Save, (u));
1467 static void GLAPIENTRY
1468 _save_EvalCoord1fv(const GLfloat * v)
1470 GET_CURRENT_CONTEXT(ctx);
1471 dlist_fallback(ctx);
1472 CALL_EvalCoord1fv(ctx->Save, (v));
1475 static void GLAPIENTRY
1476 _save_EvalCoord2f(GLfloat u, GLfloat v)
1478 GET_CURRENT_CONTEXT(ctx);
1479 dlist_fallback(ctx);
1480 CALL_EvalCoord2f(ctx->Save, (u, v));
1483 static void GLAPIENTRY
1484 _save_EvalCoord2fv(const GLfloat * v)
1486 GET_CURRENT_CONTEXT(ctx);
1487 dlist_fallback(ctx);
1488 CALL_EvalCoord2fv(ctx->Save, (v));
1491 static void GLAPIENTRY
1492 _save_EvalPoint1(GLint i)
1494 GET_CURRENT_CONTEXT(ctx);
1495 dlist_fallback(ctx);
1496 CALL_EvalPoint1(ctx->Save, (i));
1499 static void GLAPIENTRY
1500 _save_EvalPoint2(GLint i, GLint j)
1502 GET_CURRENT_CONTEXT(ctx);
1503 dlist_fallback(ctx);
1504 CALL_EvalPoint2(ctx->Save, (i, j));
1507 static void GLAPIENTRY
1508 _save_CallList(GLuint l)
1510 GET_CURRENT_CONTEXT(ctx);
1511 dlist_fallback(ctx);
1512 CALL_CallList(ctx->Save, (l));
1515 static void GLAPIENTRY
1516 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1518 GET_CURRENT_CONTEXT(ctx);
1519 dlist_fallback(ctx);
1520 CALL_CallLists(ctx->Save, (n, type, v));
1526 * Called when a glBegin is getting compiled into a display list.
1527 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1530 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1531 bool no_current_update)
1533 struct vbo_save_context *save = &vbo_context(ctx)->save;
1534 const GLuint i = save->prim_store->used++;
1536 ctx->Driver.CurrentSavePrimitive = mode;
1538 if (!save->prim_store || i >= save->prim_store->size) {
1539 save->prim_store = realloc_prim_store(save->prim_store, i * 2);
1541 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1542 save->prim_store->prims[i].begin = 1;
1543 save->prim_store->prims[i].end = 0;
1544 save->prim_store->prims[i].start = get_vertex_count(save);
1545 save->prim_store->prims[i].count = 0;
1547 save->no_current_update = no_current_update;
1549 vbo_install_save_vtxfmt(ctx);
1551 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1552 ctx->Driver.SaveNeedFlush = GL_TRUE;
1556 static void GLAPIENTRY
1559 GET_CURRENT_CONTEXT(ctx);
1560 struct vbo_save_context *save = &vbo_context(ctx)->save;
1561 const GLint i = save->prim_store->used - 1;
1563 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1564 save->prim_store->prims[i].end = 1;
1565 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1567 /* Swap out this vertex format while outside begin/end. Any color,
1568 * etc. received between here and the next begin will be compiled
1571 if (save->out_of_memory) {
1572 vbo_install_save_vtxfmt_noop(ctx);
1575 _mesa_install_save_vtxfmt(ctx);
1580 static void GLAPIENTRY
1581 _save_Begin(GLenum mode)
1583 GET_CURRENT_CONTEXT(ctx);
1585 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1589 static void GLAPIENTRY
1590 _save_PrimitiveRestartNV(void)
1592 GET_CURRENT_CONTEXT(ctx);
1593 struct vbo_save_context *save = &vbo_context(ctx)->save;
1595 if (save->prim_store->used == 0) {
1596 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1599 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1600 "glPrimitiveRestartNV called outside glBegin/End");
1602 /* get current primitive mode */
1603 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1604 bool no_current_update = save->no_current_update;
1606 /* restart primitive */
1607 CALL_End(ctx->CurrentServerDispatch, ());
1608 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1613 /* Unlike the functions above, these are to be hooked into the vtxfmt
1614 * maintained in ctx->ListState, active when the list is known or
1615 * suspected to be outside any begin/end primitive.
1616 * Note: OBE = Outside Begin/End
1619 save_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1621 GET_CURRENT_CONTEXT(ctx);
1622 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1624 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1625 CALL_Vertex2f(dispatch, (x1, y1));
1626 CALL_Vertex2f(dispatch, (x2, y1));
1627 CALL_Vertex2f(dispatch, (x2, y2));
1628 CALL_Vertex2f(dispatch, (x1, y2));
1629 CALL_End(dispatch, ());
1634 save_Rectdv(const GLdouble *v1, const GLdouble *v2)
1636 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1640 save_Rectfv(const GLfloat *v1, const GLfloat *v2)
1642 save_Rectf(v1[0], v1[1], v2[0], v2[1]);
1646 save_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1648 save_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1652 save_Rectiv(const GLint *v1, const GLint *v2)
1654 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1658 save_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1660 save_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1664 save_Rectsv(const GLshort *v1, const GLshort *v2)
1666 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1670 save_DrawArrays(GLenum mode, GLint start, GLsizei count)
1672 GET_CURRENT_CONTEXT(ctx);
1673 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1674 struct vbo_save_context *save = &vbo_context(ctx)->save;
1677 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1678 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1682 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1686 if (save->out_of_memory)
1689 grow_vertex_storage(ctx, count);
1691 /* Make sure to process any VBO binding changes */
1692 _mesa_update_state(ctx);
1694 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1696 vbo_save_NotifyBegin(ctx, mode, true);
1698 for (i = 0; i < count; i++)
1699 _mesa_array_element(ctx, start + i);
1700 CALL_End(ctx->CurrentServerDispatch, ());
1702 _mesa_vao_unmap_arrays(ctx, vao);
1707 save_MultiDrawArrays(GLenum mode, const GLint *first,
1708 const GLsizei *count, GLsizei primcount)
1710 GET_CURRENT_CONTEXT(ctx);
1713 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1714 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1718 if (primcount < 0) {
1719 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1720 "glMultiDrawArrays(primcount<0)");
1724 unsigned vertcount = 0;
1725 for (i = 0; i < primcount; i++) {
1727 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1728 "glMultiDrawArrays(count[i]<0)");
1731 vertcount += count[i];
1734 grow_vertex_storage(ctx, vertcount);
1736 for (i = 0; i < primcount; i++) {
1738 save_DrawArrays(mode, first[i], count[i]);
1745 array_element(struct gl_context *ctx,
1746 GLint basevertex, GLuint elt, unsigned index_size_shift)
1748 /* Section 10.3.5 Primitive Restart:
1750 * When one of the *BaseVertex drawing commands specified in section 10.5
1751 * is used, the primitive restart comparison occurs before the basevertex
1752 * offset is added to the array index.
1754 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1755 * then we call PrimitiveRestartNV and return.
1757 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1758 elt == ctx->Array._RestartIndex[index_size_shift]) {
1759 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1763 _mesa_array_element(ctx, basevertex + elt);
1767 /* Could do better by copying the arrays and element list intact and
1768 * then emitting an indexed prim at runtime.
1771 save_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1772 const GLvoid * indices, GLint basevertex)
1774 GET_CURRENT_CONTEXT(ctx);
1775 struct vbo_save_context *save = &vbo_context(ctx)->save;
1776 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1777 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1780 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1781 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1785 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1788 if (type != GL_UNSIGNED_BYTE &&
1789 type != GL_UNSIGNED_SHORT &&
1790 type != GL_UNSIGNED_INT) {
1791 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1795 if (save->out_of_memory)
1798 grow_vertex_storage(ctx, count);
1800 /* Make sure to process any VBO binding changes */
1801 _mesa_update_state(ctx);
1803 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1807 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1809 vbo_save_NotifyBegin(ctx, mode, true);
1812 case GL_UNSIGNED_BYTE:
1813 for (i = 0; i < count; i++)
1814 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1816 case GL_UNSIGNED_SHORT:
1817 for (i = 0; i < count; i++)
1818 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1820 case GL_UNSIGNED_INT:
1821 for (i = 0; i < count; i++)
1822 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1825 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1829 CALL_End(ctx->CurrentServerDispatch, ());
1831 _mesa_vao_unmap(ctx, vao);
1835 save_DrawElements(GLenum mode, GLsizei count, GLenum type,
1836 const GLvoid * indices)
1838 save_DrawElementsBaseVertex(mode, count, type, indices, 0);
1843 save_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1844 GLsizei count, GLenum type,
1845 const GLvoid * indices)
1847 GET_CURRENT_CONTEXT(ctx);
1848 struct vbo_save_context *save = &vbo_context(ctx)->save;
1850 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1851 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1855 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1856 "glDrawRangeElements(count<0)");
1859 if (type != GL_UNSIGNED_BYTE &&
1860 type != GL_UNSIGNED_SHORT &&
1861 type != GL_UNSIGNED_INT) {
1862 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1866 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1867 "glDrawRangeElements(end < start)");
1871 if (save->out_of_memory)
1874 save_DrawElements(mode, count, type, indices);
1878 save_DrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end,
1879 GLsizei count, GLenum type,
1880 const GLvoid *indices, GLint basevertex)
1882 GET_CURRENT_CONTEXT(ctx);
1885 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1886 "glDrawRangeElementsBaseVertex(end < start)");
1890 save_DrawElementsBaseVertex(mode, count, type, indices, basevertex);
1894 save_MultiDrawElementsEXT(GLenum mode, const GLsizei *count, GLenum type,
1895 const GLvoid * const *indices, GLsizei primcount)
1897 GET_CURRENT_CONTEXT(ctx);
1898 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1902 for (i = 0; i < primcount; i++) {
1903 vertcount += count[i];
1905 grow_vertex_storage(ctx, vertcount);
1907 for (i = 0; i < primcount; i++) {
1909 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1916 save_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1918 const GLvoid * const *indices,
1920 const GLint *basevertex)
1922 GET_CURRENT_CONTEXT(ctx);
1923 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1927 for (i = 0; i < primcount; i++) {
1928 vertcount += count[i];
1930 grow_vertex_storage(ctx, vertcount);
1932 for (i = 0; i < primcount; i++) {
1934 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1943 vbo_install_save_vtxfmt(struct gl_context *ctx)
1945 #define NAME_AE(x) _mesa_##x
1946 #define NAME_CALLLIST(x) _save_##x
1947 #define NAME(x) _save_##x
1948 #define NAME_ES(x) _save_##x
1950 struct _glapi_table *tab = ctx->Save;
1951 #include "api_vtxfmt_init.h"
1956 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1958 struct vbo_save_context *save = &vbo_context(ctx)->save;
1960 /* Noop when we are actually active:
1962 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1965 if (save->vertex_store->used || save->prim_store->used)
1966 compile_vertex_list(ctx);
1968 copy_to_current(ctx);
1970 ctx->Driver.SaveNeedFlush = GL_FALSE;
1975 * Called from glNewList when we're starting to compile a display list.
1978 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1980 struct vbo_save_context *save = &vbo_context(ctx)->save;
1985 if (!save->prim_store)
1986 save->prim_store = realloc_prim_store(NULL, 8);
1988 if (!save->vertex_store)
1989 save->vertex_store = CALLOC_STRUCT(vbo_save_vertex_store);
1992 ctx->Driver.SaveNeedFlush = GL_FALSE;
1997 * Called from glEndList when we're finished compiling a display list.
2000 vbo_save_EndList(struct gl_context *ctx)
2002 struct vbo_save_context *save = &vbo_context(ctx)->save;
2004 /* EndList called inside a (saved) Begin/End pair?
2006 if (_mesa_inside_dlist_begin_end(ctx)) {
2007 if (save->prim_store->used > 0) {
2008 GLint i = save->prim_store->used - 1;
2009 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
2010 save->prim_store->prims[i].end = 0;
2011 save->prim_store->prims[i].count = get_vertex_count(save) - save->prim_store->prims[i].start;
2014 /* Make sure this vertex list gets replayed by the "loopback"
2017 save->dangling_attr_ref = GL_TRUE;
2018 vbo_save_SaveFlushVertices(ctx);
2020 /* Swap out this vertex format while outside begin/end. Any color,
2021 * etc. received between here and the next begin will be compiled
2024 _mesa_install_save_vtxfmt(ctx);
2027 assert(save->vertex_size == 0);
2031 * Called during context creation/init.
2034 current_init(struct gl_context *ctx)
2036 struct vbo_save_context *save = &vbo_context(ctx)->save;
2039 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
2040 const GLuint j = i - VBO_ATTRIB_POS;
2041 assert(j < VERT_ATTRIB_MAX);
2042 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
2043 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2046 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2047 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2048 assert(j < MAT_ATTRIB_MAX);
2049 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2050 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2056 * Initialize the display list compiler. Called during context creation.
2059 vbo_save_api_init(struct vbo_save_context *save)
2061 struct gl_context *ctx = gl_context_from_vbo_save(save);