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 "gallium/auxiliary/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);
849 node->cold->bo_bytes_used = save->current_bo_bytes_used;
851 if (vertex_to_index) {
852 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
853 free(temp_vertices_buffer);
856 /* Since we append the indices to an existing buffer, we need to adjust the start value of each
857 * primitive (not the indices themselves). */
858 if (!ctx->ListState.Current.UseLoopback) {
859 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
860 int indices_offset = save->current_bo_bytes_used / 4;
861 for (int i = 0; i < merged_prim_count; i++) {
862 merged_prims[i].start += indices_offset;
866 /* Then upload the indices. */
867 if (node->cold->ib.obj) {
868 _mesa_bufferobj_subdata(ctx,
869 save->current_bo_bytes_used,
870 idx * sizeof(uint32_t),
873 save->current_bo_bytes_used += idx * sizeof(uint32_t);
875 node->cold->vertex_count = 0;
876 node->cold->prim_count = 0;
879 /* Prepare for DrawGallium */
880 memset(&node->cold->info, 0, sizeof(struct pipe_draw_info));
881 /* The other info fields will be updated in vbo_save_playback_vertex_list */
882 node->cold->info.index_size = 4;
883 node->cold->info.instance_count = 1;
884 node->cold->info.index.gl_bo = node->cold->ib.obj;
885 if (merged_prim_count == 1) {
886 node->cold->info.mode = merged_prims[0].mode;
887 node->start_count.start = merged_prims[0].start;
888 node->start_count.count = merged_prims[0].count;
889 node->start_count.index_bias = 0;
892 node->modes = malloc(merged_prim_count * sizeof(unsigned char));
893 node->start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
894 for (unsigned i = 0; i < merged_prim_count; i++) {
895 node->start_counts[i].start = merged_prims[i].start;
896 node->start_counts[i].count = merged_prims[i].count;
897 node->start_counts[i].index_bias = 0;
898 node->modes[i] = merged_prims[i].mode;
901 node->num_draws = merged_prim_count;
902 if (node->num_draws > 1) {
903 bool same_mode = true;
904 for (unsigned i = 1; i < node->num_draws && same_mode; i++) {
905 same_mode = node->modes[i] == node->modes[0];
908 /* All primitives use the same mode, so we can simplify a bit */
909 node->cold->info.mode = node->modes[0];
920 node->draw_begins = node->cold->prims[0].begin;
922 if (!save->current_bo) {
923 save->current_bo = _mesa_bufferobj_alloc(ctx, VBO_BUF_ID + 1);
924 bool success = _mesa_bufferobj_data(ctx,
925 GL_ELEMENT_ARRAY_BUFFER_ARB,
926 VBO_SAVE_BUFFER_SIZE,
928 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
929 MESA_GALLIUM_VERTEX_STATE_STORAGE,
932 save->out_of_memory = true;
935 GLuint offsets[VBO_ATTRIB_MAX];
936 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
938 offset += save->attrsz[i] * sizeof(GLfloat);
940 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
941 * Note that this may reuse the previous one of possible.
943 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
944 /* create or reuse the vao */
945 update_vao(ctx, vpm, &save->VAO[vpm],
946 save->current_bo, buffer_offset, stride,
947 save->enabled, save->attrsz, save->attrtype, offsets);
948 /* Reference the vao in the dlist */
949 node->cold->VAO[vpm] = NULL;
950 _mesa_reference_vao(ctx, &node->cold->VAO[vpm], save->VAO[vpm]);
953 /* Prepare for DrawGalliumVertexState */
954 if (node->num_draws && ctx->Driver.DrawGalliumVertexState) {
955 for (unsigned i = 0; i < VP_MODE_MAX; i++) {
956 uint32_t enabled_attribs = _vbo_get_vao_filter(i) &
957 node->cold->VAO[i]->_EnabledWithMapMode;
960 ctx->Driver.CreateGalliumVertexState(ctx, node->cold->VAO[i],
963 node->private_refcount[i] = 0;
964 node->enabled_attribs[i] = enabled_attribs;
968 node->mode = node->cold->info.mode;
969 assert(node->cold->info.index_size == 4);
972 /* Deal with GL_COMPILE_AND_EXECUTE:
974 if (ctx->ExecuteFlag) {
975 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
976 * use buffer_in_ram (which contains all vertices) instead of current_bo
977 * (which contains deduplicated vertices *when* UseLoopback is false).
979 * The problem is that the VAO offset is based on current_bo's layout,
980 * so we have to use a temp value.
982 struct gl_vertex_array_object *vao = node->cold->VAO[VP_MODE_SHADER];
983 GLintptr original = vao->BufferBinding[0].Offset;
984 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
985 vao->BufferBinding[0].Offset = -(GLintptr)(start_offset * stride);
986 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
987 vao->BufferBinding[0].Offset = original;
990 /* Reset our structures for the next run of vertices:
997 * This is called when we fill a vertex buffer before we hit a glEnd().
999 * TODO -- If no new vertices have been stored, don't bother saving it.
1002 wrap_buffers(struct gl_context *ctx)
1004 struct vbo_save_context *save = &vbo_context(ctx)->save;
1005 GLint i = save->prim_store->used - 1;
1008 assert(i < (GLint) save->prim_store->size);
1011 /* Close off in-progress primitive.
1013 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1014 mode = save->prim_store->prims[i].mode;
1016 /* store the copied vertices, and allocate a new list.
1018 compile_vertex_list(ctx);
1020 /* Restart interrupted primitive
1022 save->prim_store->prims[0].mode = mode;
1023 save->prim_store->prims[0].begin = 0;
1024 save->prim_store->prims[0].end = 0;
1025 save->prim_store->prims[0].start = 0;
1026 save->prim_store->prims[0].count = 0;
1027 save->prim_store->used = 1;
1032 * Called only when buffers are wrapped as the result of filling the
1033 * vertex_store struct.
1036 wrap_filled_vertex(struct gl_context *ctx)
1038 struct vbo_save_context *save = &vbo_context(ctx)->save;
1039 unsigned numComponents;
1041 /* Emit a glEnd to close off the last vertex list.
1045 assert(save->vertex_store->used == 0 && save->vertex_store->used == 0);
1047 /* Copy stored stored vertices to start of new list.
1049 numComponents = save->copied.nr * save->vertex_size;
1051 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
1052 if (numComponents) {
1053 assert(save->copied.buffer);
1055 save->copied.buffer,
1056 numComponents * sizeof(fi_type));
1057 free(save->copied.buffer);
1058 save->copied.buffer = NULL;
1060 save->vertex_store->used = numComponents;
1065 copy_to_current(struct gl_context *ctx)
1067 struct vbo_save_context *save = &vbo_context(ctx)->save;
1068 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1071 const int i = u_bit_scan64(&enabled);
1072 assert(save->attrsz[i]);
1074 if (save->attrtype[i] == GL_DOUBLE ||
1075 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1076 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1078 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1079 save->attrptr[i], save->attrtype[i]);
1085 copy_from_current(struct gl_context *ctx)
1087 struct vbo_save_context *save = &vbo_context(ctx)->save;
1088 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1091 const int i = u_bit_scan64(&enabled);
1093 switch (save->attrsz[i]) {
1095 save->attrptr[i][3] = save->current[i][3];
1098 save->attrptr[i][2] = save->current[i][2];
1101 save->attrptr[i][1] = save->current[i][1];
1104 save->attrptr[i][0] = save->current[i][0];
1107 unreachable("Unexpected vertex attribute size");
1114 * Called when we increase the size of a vertex attribute. For example,
1115 * if we've seen one or more glTexCoord2f() calls and now we get a
1116 * glTexCoord3f() call.
1117 * Flush existing data, set new attrib size, replay copied vertices.
1120 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1122 struct vbo_save_context *save = &vbo_context(ctx)->save;
1127 /* Store the current run of vertices, and emit a GL_END. Emit a
1128 * BEGIN in the new buffer.
1130 if (save->vertex_store->used)
1133 assert(save->copied.nr == 0);
1135 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1136 * when the attribute already exists in the vertex and is having
1137 * its size increased.
1139 copy_to_current(ctx);
1143 oldsz = save->attrsz[attr];
1144 save->attrsz[attr] = newsz;
1145 save->enabled |= BITFIELD64_BIT(attr);
1147 save->vertex_size += newsz - oldsz;
1149 /* Recalculate all the attrptr[] values:
1152 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1153 if (save->attrsz[i]) {
1154 save->attrptr[i] = tmp;
1155 tmp += save->attrsz[i];
1158 save->attrptr[i] = NULL; /* will not be dereferenced. */
1162 /* Copy from current to repopulate the vertex with correct values.
1164 copy_from_current(ctx);
1166 /* Replay stored vertices to translate them to new format here.
1168 * If there are copied vertices and the new (upgraded) attribute
1169 * has not been defined before, this list is somewhat degenerate,
1170 * and will need fixup at runtime.
1172 if (save->copied.nr) {
1173 assert(save->copied.buffer);
1174 const fi_type *data = save->copied.buffer;
1175 grow_vertex_storage(ctx, save->copied.nr);
1176 fi_type *dest = save->vertex_store->buffer_in_ram;
1178 /* Need to note this and fix up at runtime (or loopback):
1180 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1182 save->dangling_attr_ref = GL_TRUE;
1185 for (i = 0; i < save->copied.nr; i++) {
1186 GLbitfield64 enabled = save->enabled;
1188 const int j = u_bit_scan64(&enabled);
1189 assert(save->attrsz[j]);
1192 const fi_type *src = oldsz ? data : save->current[attr];
1193 int copy = oldsz ? oldsz : newsz;
1194 for (k = 0; k < copy; k++)
1196 for (; k < newsz; k++) {
1197 switch (save->attrtype[j]) {
1199 dest[k] = FLOAT_AS_UNION(k == 3);
1202 dest[k] = INT_AS_UNION(k == 3);
1204 case GL_UNSIGNED_INT:
1205 dest[k] = UINT_AS_UNION(k == 3);
1208 dest[k] = FLOAT_AS_UNION(k == 3);
1209 assert(!"Unexpected type in upgrade_vertex");
1216 GLint sz = save->attrsz[j];
1217 for (int k = 0; k < sz; k++)
1225 save->vertex_store->used += save->vertex_size * save->copied.nr;
1226 free(save->copied.buffer);
1227 save->copied.buffer = NULL;
1233 * This is called when the size of a vertex attribute changes.
1234 * For example, after seeing one or more glTexCoord2f() calls we
1235 * get a glTexCoord4f() or glTexCoord1f() call.
1238 fixup_vertex(struct gl_context *ctx, GLuint attr,
1239 GLuint sz, GLenum newType)
1241 struct vbo_save_context *save = &vbo_context(ctx)->save;
1243 if (sz > save->attrsz[attr] ||
1244 newType != save->attrtype[attr]) {
1245 /* New size is larger. Need to flush existing vertices and get
1246 * an enlarged vertex format.
1248 upgrade_vertex(ctx, attr, sz);
1250 else if (sz < save->active_sz[attr]) {
1252 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1254 /* New size is equal or smaller - just need to fill in some
1257 for (i = sz; i <= save->attrsz[attr]; i++)
1258 save->attrptr[attr][i - 1] = id[i - 1];
1261 save->active_sz[attr] = sz;
1263 grow_vertex_storage(ctx, 1);
1268 * Reset the current size of all vertex attributes to the default
1269 * value of 0. This signals that we haven't yet seen any per-vertex
1270 * commands such as glNormal3f() or glTexCoord2f().
1273 reset_vertex(struct gl_context *ctx)
1275 struct vbo_save_context *save = &vbo_context(ctx)->save;
1277 while (save->enabled) {
1278 const int i = u_bit_scan64(&save->enabled);
1279 assert(save->attrsz[i]);
1280 save->attrsz[i] = 0;
1281 save->active_sz[i] = 0;
1284 save->vertex_size = 0;
1289 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1290 * It depends on a few things, including whether we're inside or outside
1294 is_vertex_position(const struct gl_context *ctx, GLuint index)
1296 return (index == 0 &&
1297 _mesa_attr_zero_aliases_vertex(ctx) &&
1298 _mesa_inside_dlist_begin_end(ctx));
1303 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1306 /* Only one size for each attribute may be active at once. Eg. if
1307 * Color3f is installed/active, then Color4f may not be, even if the
1308 * vertex actually contains 4 color coordinates. This is because the
1309 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1310 * of the chooser function when switching between Color4f and Color3f.
1312 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1314 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1315 int sz = (sizeof(C) / sizeof(GLfloat)); \
1317 if (save->active_sz[A] != N) \
1318 fixup_vertex(ctx, A, N * sz, T); \
1321 C *dest = (C *)save->attrptr[A]; \
1322 if (N>0) dest[0] = V0; \
1323 if (N>1) dest[1] = V1; \
1324 if (N>2) dest[2] = V2; \
1325 if (N>3) dest[3] = V3; \
1326 save->attrtype[A] = T; \
1329 if ((A) == VBO_ATTRIB_POS) { \
1330 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + \
1331 save->vertex_store->used; \
1333 for (int i = 0; i < save->vertex_size; i++) \
1334 buffer_ptr[i] = save->vertex[i]; \
1336 save->vertex_store->used += save->vertex_size; \
1337 unsigned used_next = (save->vertex_store->used + \
1338 save->vertex_size) * sizeof(float); \
1339 if (used_next > save->vertex_store->buffer_in_ram_size) { \
1340 grow_vertex_storage(ctx, get_vertex_count(save)); \
1341 assert(used_next <= \
1342 save->vertex_store->buffer_in_ram_size); \
1347 #define TAG(x) _save_##x
1349 #include "vbo_attrib_tmp.h"
1352 #define MAT( ATTR, N, face, params ) \
1354 if (face != GL_BACK) \
1355 MAT_ATTR( ATTR, N, params ); /* front */ \
1356 if (face != GL_FRONT) \
1357 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1362 * Save a glMaterial call found between glBegin/End.
1363 * glMaterial calls outside Begin/End are handled in dlist.c.
1365 static void GLAPIENTRY
1366 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1368 GET_CURRENT_CONTEXT(ctx);
1370 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1371 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1377 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1380 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1383 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1386 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1389 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1390 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1393 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1396 case GL_COLOR_INDEXES:
1397 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1399 case GL_AMBIENT_AND_DIFFUSE:
1400 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1401 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1404 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1411 vbo_init_dispatch_save_begin_end(struct gl_context *ctx);
1414 /* Cope with EvalCoord/CallList called within a begin/end object:
1415 * -- Flush current buffer
1416 * -- Fallback to opcodes for the rest of the begin/end object.
1419 dlist_fallback(struct gl_context *ctx)
1421 struct vbo_save_context *save = &vbo_context(ctx)->save;
1423 if (save->vertex_store->used || save->prim_store->used) {
1424 if (save->prim_store->used > 0 && save->vertex_store->used > 0) {
1425 assert(save->vertex_size);
1426 /* Close off in-progress primitive. */
1427 GLint i = save->prim_store->used - 1;
1428 save->prim_store->prims[i].count =
1429 get_vertex_count(save) -
1430 save->prim_store->prims[i].start;
1433 /* Need to replay this display list with loopback,
1434 * unfortunately, otherwise this primitive won't be handled
1437 save->dangling_attr_ref = GL_TRUE;
1439 compile_vertex_list(ctx);
1442 copy_to_current(ctx);
1444 if (save->out_of_memory) {
1445 vbo_install_save_vtxfmt_noop(ctx);
1448 _mesa_init_dispatch_save_begin_end(ctx);
1450 ctx->Driver.SaveNeedFlush = GL_FALSE;
1454 static void GLAPIENTRY
1455 _save_EvalCoord1f(GLfloat u)
1457 GET_CURRENT_CONTEXT(ctx);
1458 dlist_fallback(ctx);
1459 CALL_EvalCoord1f(ctx->Save, (u));
1462 static void GLAPIENTRY
1463 _save_EvalCoord1fv(const GLfloat * v)
1465 GET_CURRENT_CONTEXT(ctx);
1466 dlist_fallback(ctx);
1467 CALL_EvalCoord1fv(ctx->Save, (v));
1470 static void GLAPIENTRY
1471 _save_EvalCoord2f(GLfloat u, GLfloat v)
1473 GET_CURRENT_CONTEXT(ctx);
1474 dlist_fallback(ctx);
1475 CALL_EvalCoord2f(ctx->Save, (u, v));
1478 static void GLAPIENTRY
1479 _save_EvalCoord2fv(const GLfloat * v)
1481 GET_CURRENT_CONTEXT(ctx);
1482 dlist_fallback(ctx);
1483 CALL_EvalCoord2fv(ctx->Save, (v));
1486 static void GLAPIENTRY
1487 _save_EvalPoint1(GLint i)
1489 GET_CURRENT_CONTEXT(ctx);
1490 dlist_fallback(ctx);
1491 CALL_EvalPoint1(ctx->Save, (i));
1494 static void GLAPIENTRY
1495 _save_EvalPoint2(GLint i, GLint j)
1497 GET_CURRENT_CONTEXT(ctx);
1498 dlist_fallback(ctx);
1499 CALL_EvalPoint2(ctx->Save, (i, j));
1502 static void GLAPIENTRY
1503 _save_CallList(GLuint l)
1505 GET_CURRENT_CONTEXT(ctx);
1506 dlist_fallback(ctx);
1507 CALL_CallList(ctx->Save, (l));
1510 static void GLAPIENTRY
1511 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1513 GET_CURRENT_CONTEXT(ctx);
1514 dlist_fallback(ctx);
1515 CALL_CallLists(ctx->Save, (n, type, v));
1521 * Called when a glBegin is getting compiled into a display list.
1522 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1525 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1526 bool no_current_update)
1528 struct vbo_save_context *save = &vbo_context(ctx)->save;
1529 const GLuint i = save->prim_store->used++;
1531 ctx->Driver.CurrentSavePrimitive = mode;
1533 if (!save->prim_store || i >= save->prim_store->size) {
1534 save->prim_store = realloc_prim_store(save->prim_store, i * 2);
1536 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1537 save->prim_store->prims[i].begin = 1;
1538 save->prim_store->prims[i].end = 0;
1539 save->prim_store->prims[i].start = get_vertex_count(save);
1540 save->prim_store->prims[i].count = 0;
1542 save->no_current_update = no_current_update;
1544 vbo_init_dispatch_save_begin_end(ctx);
1546 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1547 ctx->Driver.SaveNeedFlush = GL_TRUE;
1551 static void GLAPIENTRY
1554 GET_CURRENT_CONTEXT(ctx);
1555 struct vbo_save_context *save = &vbo_context(ctx)->save;
1556 const GLint i = save->prim_store->used - 1;
1558 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1559 save->prim_store->prims[i].end = 1;
1560 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1562 /* Swap out this vertex format while outside begin/end. Any color,
1563 * etc. received between here and the next begin will be compiled
1566 if (save->out_of_memory) {
1567 vbo_install_save_vtxfmt_noop(ctx);
1570 _mesa_init_dispatch_save_begin_end(ctx);
1575 static void GLAPIENTRY
1576 _save_Begin(GLenum mode)
1578 GET_CURRENT_CONTEXT(ctx);
1580 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1584 static void GLAPIENTRY
1585 _save_PrimitiveRestartNV(void)
1587 GET_CURRENT_CONTEXT(ctx);
1588 struct vbo_save_context *save = &vbo_context(ctx)->save;
1590 if (save->prim_store->used == 0) {
1591 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1594 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1595 "glPrimitiveRestartNV called outside glBegin/End");
1597 /* get current primitive mode */
1598 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1599 bool no_current_update = save->no_current_update;
1601 /* restart primitive */
1602 CALL_End(ctx->CurrentServerDispatch, ());
1603 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1609 save_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1611 GET_CURRENT_CONTEXT(ctx);
1612 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1614 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1615 CALL_Vertex2f(dispatch, (x1, y1));
1616 CALL_Vertex2f(dispatch, (x2, y1));
1617 CALL_Vertex2f(dispatch, (x2, y2));
1618 CALL_Vertex2f(dispatch, (x1, y2));
1619 CALL_End(dispatch, ());
1624 save_Rectdv(const GLdouble *v1, const GLdouble *v2)
1626 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1630 save_Rectfv(const GLfloat *v1, const GLfloat *v2)
1632 save_Rectf(v1[0], v1[1], v2[0], v2[1]);
1636 save_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1638 save_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1642 save_Rectiv(const GLint *v1, const GLint *v2)
1644 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1648 save_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1650 save_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1654 save_Rectsv(const GLshort *v1, const GLshort *v2)
1656 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1660 save_DrawArrays(GLenum mode, GLint start, GLsizei count)
1662 GET_CURRENT_CONTEXT(ctx);
1663 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1664 struct vbo_save_context *save = &vbo_context(ctx)->save;
1667 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1668 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1672 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1676 if (save->out_of_memory)
1679 grow_vertex_storage(ctx, count);
1681 /* Make sure to process any VBO binding changes */
1682 _mesa_update_state(ctx);
1684 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1686 vbo_save_NotifyBegin(ctx, mode, true);
1688 for (i = 0; i < count; i++)
1689 _mesa_array_element(ctx, start + i);
1690 CALL_End(ctx->CurrentServerDispatch, ());
1692 _mesa_vao_unmap_arrays(ctx, vao);
1697 save_MultiDrawArrays(GLenum mode, const GLint *first,
1698 const GLsizei *count, GLsizei primcount)
1700 GET_CURRENT_CONTEXT(ctx);
1703 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1704 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1708 if (primcount < 0) {
1709 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1710 "glMultiDrawArrays(primcount<0)");
1714 unsigned vertcount = 0;
1715 for (i = 0; i < primcount; i++) {
1717 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1718 "glMultiDrawArrays(count[i]<0)");
1721 vertcount += count[i];
1724 grow_vertex_storage(ctx, vertcount);
1726 for (i = 0; i < primcount; i++) {
1728 save_DrawArrays(mode, first[i], count[i]);
1735 array_element(struct gl_context *ctx,
1736 GLint basevertex, GLuint elt, unsigned index_size_shift)
1738 /* Section 10.3.5 Primitive Restart:
1740 * When one of the *BaseVertex drawing commands specified in section 10.5
1741 * is used, the primitive restart comparison occurs before the basevertex
1742 * offset is added to the array index.
1744 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1745 * then we call PrimitiveRestartNV and return.
1747 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1748 elt == ctx->Array._RestartIndex[index_size_shift]) {
1749 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1753 _mesa_array_element(ctx, basevertex + elt);
1757 /* Could do better by copying the arrays and element list intact and
1758 * then emitting an indexed prim at runtime.
1761 save_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1762 const GLvoid * indices, GLint basevertex)
1764 GET_CURRENT_CONTEXT(ctx);
1765 struct vbo_save_context *save = &vbo_context(ctx)->save;
1766 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1767 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1770 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1771 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1775 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1778 if (type != GL_UNSIGNED_BYTE &&
1779 type != GL_UNSIGNED_SHORT &&
1780 type != GL_UNSIGNED_INT) {
1781 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1785 if (save->out_of_memory)
1788 grow_vertex_storage(ctx, count);
1790 /* Make sure to process any VBO binding changes */
1791 _mesa_update_state(ctx);
1793 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1797 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1799 vbo_save_NotifyBegin(ctx, mode, true);
1802 case GL_UNSIGNED_BYTE:
1803 for (i = 0; i < count; i++)
1804 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1806 case GL_UNSIGNED_SHORT:
1807 for (i = 0; i < count; i++)
1808 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1810 case GL_UNSIGNED_INT:
1811 for (i = 0; i < count; i++)
1812 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1815 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1819 CALL_End(ctx->CurrentServerDispatch, ());
1821 _mesa_vao_unmap(ctx, vao);
1825 save_DrawElements(GLenum mode, GLsizei count, GLenum type,
1826 const GLvoid * indices)
1828 save_DrawElementsBaseVertex(mode, count, type, indices, 0);
1833 save_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1834 GLsizei count, GLenum type,
1835 const GLvoid * indices)
1837 GET_CURRENT_CONTEXT(ctx);
1838 struct vbo_save_context *save = &vbo_context(ctx)->save;
1840 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1841 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1845 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1846 "glDrawRangeElements(count<0)");
1849 if (type != GL_UNSIGNED_BYTE &&
1850 type != GL_UNSIGNED_SHORT &&
1851 type != GL_UNSIGNED_INT) {
1852 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1856 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1857 "glDrawRangeElements(end < start)");
1861 if (save->out_of_memory)
1864 save_DrawElements(mode, count, type, indices);
1868 save_DrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end,
1869 GLsizei count, GLenum type,
1870 const GLvoid *indices, GLint basevertex)
1872 GET_CURRENT_CONTEXT(ctx);
1875 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1876 "glDrawRangeElementsBaseVertex(end < start)");
1880 save_DrawElementsBaseVertex(mode, count, type, indices, basevertex);
1884 save_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1885 const GLvoid * const *indices, GLsizei primcount)
1887 GET_CURRENT_CONTEXT(ctx);
1888 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1892 for (i = 0; i < primcount; i++) {
1893 vertcount += count[i];
1895 grow_vertex_storage(ctx, vertcount);
1897 for (i = 0; i < primcount; i++) {
1899 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1906 save_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1908 const GLvoid * const *indices,
1910 const GLint *basevertex)
1912 GET_CURRENT_CONTEXT(ctx);
1913 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1917 for (i = 0; i < primcount; i++) {
1918 vertcount += count[i];
1920 grow_vertex_storage(ctx, vertcount);
1922 for (i = 0; i < primcount; i++) {
1924 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1933 vbo_init_dispatch_save_begin_end(struct gl_context *ctx)
1935 #define NAME_AE(x) _mesa_##x
1936 #define NAME_CALLLIST(x) _save_##x
1937 #define NAME(x) _save_##x
1938 #define NAME_ES(x) _save_##x
1940 struct _glapi_table *tab = ctx->Save;
1941 #include "api_beginend_init.h"
1946 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1948 struct vbo_save_context *save = &vbo_context(ctx)->save;
1950 /* Noop when we are actually active:
1952 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1955 if (save->vertex_store->used || save->prim_store->used)
1956 compile_vertex_list(ctx);
1958 copy_to_current(ctx);
1960 ctx->Driver.SaveNeedFlush = GL_FALSE;
1965 * Called from glNewList when we're starting to compile a display list.
1968 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1970 struct vbo_save_context *save = &vbo_context(ctx)->save;
1975 if (!save->prim_store)
1976 save->prim_store = realloc_prim_store(NULL, 8);
1978 if (!save->vertex_store)
1979 save->vertex_store = CALLOC_STRUCT(vbo_save_vertex_store);
1982 ctx->Driver.SaveNeedFlush = GL_FALSE;
1987 * Called from glEndList when we're finished compiling a display list.
1990 vbo_save_EndList(struct gl_context *ctx)
1992 struct vbo_save_context *save = &vbo_context(ctx)->save;
1994 /* EndList called inside a (saved) Begin/End pair?
1996 if (_mesa_inside_dlist_begin_end(ctx)) {
1997 if (save->prim_store->used > 0) {
1998 GLint i = save->prim_store->used - 1;
1999 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
2000 save->prim_store->prims[i].end = 0;
2001 save->prim_store->prims[i].count = get_vertex_count(save) - save->prim_store->prims[i].start;
2004 /* Make sure this vertex list gets replayed by the "loopback"
2007 save->dangling_attr_ref = GL_TRUE;
2008 vbo_save_SaveFlushVertices(ctx);
2010 /* Swap out this vertex format while outside begin/end. Any color,
2011 * etc. received between here and the next begin will be compiled
2014 _mesa_init_dispatch_save_begin_end(ctx);
2017 assert(save->vertex_size == 0);
2021 * Called during context creation/init.
2024 current_init(struct gl_context *ctx)
2026 struct vbo_save_context *save = &vbo_context(ctx)->save;
2029 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
2030 const GLuint j = i - VBO_ATTRIB_POS;
2031 assert(j < VERT_ATTRIB_MAX);
2032 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
2033 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2036 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2037 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2038 assert(j < MAT_ATTRIB_MAX);
2039 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2040 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2046 * Initialize the display list compiler. Called during context creation.
2049 vbo_save_api_init(struct vbo_save_context *save)
2051 struct gl_context *ctx = gl_context_from_vbo_save(save);