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/vtxfmt.h"
114 #include "main/dispatch.h"
115 #include "main/state.h"
116 #include "main/varray.h"
117 #include "util/bitscan.h"
118 #include "util/u_memory.h"
119 #include "util/hash_table.h"
121 #include "gallium/include/pipe/p_state.h"
123 #include "vbo_noop.h"
124 #include "vbo_private.h"
131 /* An interesting VBO number/name to help with debugging */
132 #define VBO_BUF_ID 12345
134 static void GLAPIENTRY
135 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params);
137 static void GLAPIENTRY
138 _save_EvalCoord1f(GLfloat u);
140 static void GLAPIENTRY
141 _save_EvalCoord2f(GLfloat u, GLfloat v);
144 handle_out_of_memory(struct gl_context *ctx)
146 struct vbo_save_context *save = &vbo_context(ctx)->save;
147 _mesa_noop_vtxfmt_init(ctx, &save->vtxfmt);
148 save->out_of_memory = true;
152 * NOTE: Old 'parity' issue is gone, but copying can still be
153 * wrong-footed on replay.
156 copy_vertices(struct gl_context *ctx,
157 const struct vbo_save_vertex_list *node,
158 const fi_type * src_buffer)
160 struct vbo_save_context *save = &vbo_context(ctx)->save;
161 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
162 GLuint sz = save->vertex_size;
164 if (prim->end || !prim->count || !sz)
167 const fi_type *src = src_buffer + prim->start * sz;
168 assert(save->copied.buffer == NULL);
169 save->copied.buffer = malloc(sizeof(fi_type) * sz * prim->count);
171 return vbo_copy_vertices(ctx, prim->mode, prim->start, &prim->count,
172 prim->begin, sz, true, save->copied.buffer, src);
176 static struct vbo_save_primitive_store *
177 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
180 store = CALLOC_STRUCT(vbo_save_primitive_store);
182 uint32_t old_size = store->size;
183 store->size = prim_count;
184 assert (old_size < store->size);
185 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
186 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
193 reset_counters(struct gl_context *ctx)
195 struct vbo_save_context *save = &vbo_context(ctx)->save;
197 save->vertex_store->used = 0;
198 save->prim_store->used = 0;
199 save->dangling_attr_ref = GL_FALSE;
203 * For a list of prims, try merging prims that can just be extensions of the
207 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
211 struct _mesa_prim *prev_prim = prim_list;
213 for (i = 1; i < *prim_count; i++) {
214 struct _mesa_prim *this_prim = prim_list + i;
216 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
218 if (vbo_merge_draws(ctx, true,
219 prev_prim->mode, this_prim->mode,
220 prev_prim->start, this_prim->start,
221 &prev_prim->count, this_prim->count,
222 prev_prim->basevertex, this_prim->basevertex,
224 this_prim->begin, this_prim->end)) {
225 /* We've found a prim that just extend the previous one. Tack it
226 * onto the previous one, and let this primitive struct get dropped.
231 /* If any previous primitives have been dropped, then we need to copy
232 * this later one into the next available slot.
235 if (prev_prim != this_prim)
236 *prev_prim = *this_prim;
239 *prim_count = prev_prim - prim_list + 1;
244 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
245 * don't have to worry about handling the _mesa_prim::begin/end flags.
246 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
249 convert_line_loop_to_strip(struct vbo_save_context *save,
250 struct vbo_save_vertex_list *node)
252 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
254 assert(prim->mode == GL_LINE_LOOP);
257 /* Copy the 0th vertex to end of the buffer and extend the
258 * vertex count by one to finish the line loop.
260 const GLuint sz = save->vertex_size;
262 const fi_type *src = save->vertex_store->buffer_in_ram + prim->start * sz;
264 fi_type *dst = save->vertex_store->buffer_in_ram + (prim->start + prim->count) * sz;
266 memcpy(dst, src, sz * sizeof(float));
269 node->cold->vertex_count++;
270 save->vertex_store->used += sz;
274 /* Drawing the second or later section of a long line loop.
275 * Skip the 0th vertex.
281 prim->mode = GL_LINE_STRIP;
285 /* Compare the present vao if it has the same setup. */
287 compare_vao(gl_vertex_processing_mode mode,
288 const struct gl_vertex_array_object *vao,
289 const struct gl_buffer_object *bo, GLintptr buffer_offset,
290 GLuint stride, GLbitfield64 vao_enabled,
291 const GLubyte size[VBO_ATTRIB_MAX],
292 const GLenum16 type[VBO_ATTRIB_MAX],
293 const GLuint offset[VBO_ATTRIB_MAX])
298 /* If the enabled arrays are not the same we are not equal. */
299 if (vao_enabled != vao->Enabled)
302 /* Check the buffer binding at 0 */
303 if (vao->BufferBinding[0].BufferObj != bo)
305 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
306 if (vao->BufferBinding[0].Stride != stride)
308 assert(vao->BufferBinding[0].InstanceDivisor == 0);
310 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
311 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
313 /* Now check the enabled arrays */
314 GLbitfield mask = vao_enabled;
316 const int attr = u_bit_scan(&mask);
317 const unsigned char vbo_attr = vao_to_vbo_map[attr];
318 const GLenum16 tp = type[vbo_attr];
319 const GLintptr off = offset[vbo_attr] + buffer_offset;
320 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
321 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
323 if (attrib->Format.Type != tp)
325 if (attrib->Format.Size != size[vbo_attr])
327 assert(attrib->Format.Format == GL_RGBA);
328 assert(attrib->Format.Normalized == GL_FALSE);
329 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
330 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
331 assert(attrib->BufferBindingIndex == 0);
338 /* Create or reuse the vao for the vertex processing mode. */
340 update_vao(struct gl_context *ctx,
341 gl_vertex_processing_mode mode,
342 struct gl_vertex_array_object **vao,
343 struct gl_buffer_object *bo, GLintptr buffer_offset,
344 GLuint stride, GLbitfield64 vbo_enabled,
345 const GLubyte size[VBO_ATTRIB_MAX],
346 const GLenum16 type[VBO_ATTRIB_MAX],
347 const GLuint offset[VBO_ATTRIB_MAX])
349 /* Compute the bitmasks of vao_enabled arrays */
350 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
353 * Check if we can possibly reuse the exisiting one.
354 * In the long term we should reset them when something changes.
356 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
357 vao_enabled, size, type, offset))
360 /* The initial refcount is 1 */
361 _mesa_reference_vao(ctx, vao, NULL);
362 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
365 * assert(stride <= ctx->Const.MaxVertexAttribStride);
366 * MaxVertexAttribStride is not set for drivers that does not
367 * expose GL 44 or GLES 31.
370 /* Bind the buffer object at binding point 0 */
371 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
374 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
375 * Note that the position/generic0 aliasing is done in the VAO.
377 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
378 /* Now set the enable arrays */
379 GLbitfield mask = vao_enabled;
381 const int vao_attr = u_bit_scan(&mask);
382 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
383 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
385 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
386 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
387 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
389 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
390 assert(vao_enabled == (*vao)->Enabled);
391 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
393 /* Finalize and freeze the VAO */
394 _mesa_set_vao_immutable(ctx, *vao);
397 static void wrap_filled_vertex(struct gl_context *ctx);
399 /* Grow the vertex storage to accomodate for vertex_count new vertices */
401 grow_vertex_storage(struct gl_context *ctx, int vertex_count)
403 struct vbo_save_context *save = &vbo_context(ctx)->save;
404 assert (save->vertex_store);
406 int new_size = (save->vertex_store->used +
407 vertex_count * save->vertex_size) * sizeof(GLfloat);
409 /* Limit how much memory we allocate. */
410 if (save->prim_store->used > 0 &&
412 new_size > VBO_SAVE_BUFFER_SIZE) {
413 wrap_filled_vertex(ctx);
414 new_size = VBO_SAVE_BUFFER_SIZE;
417 if (new_size > save->vertex_store->buffer_in_ram_size) {
418 save->vertex_store->buffer_in_ram_size = new_size;
419 save->vertex_store->buffer_in_ram = realloc(save->vertex_store->buffer_in_ram,
420 save->vertex_store->buffer_in_ram_size);
421 if (save->vertex_store->buffer_in_ram == NULL)
422 handle_out_of_memory(ctx);
428 unsigned vertex_size;
429 fi_type *vertex_attributes;
432 static uint32_t _hash_vertex_key(const void *key)
434 struct vertex_key *k = (struct vertex_key*)key;
435 unsigned sz = k->vertex_size;
437 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
440 static bool _compare_vertex_key(const void *key1, const void *key2)
442 struct vertex_key *k1 = (struct vertex_key*)key1;
443 struct vertex_key *k2 = (struct vertex_key*)key2;
444 /* All the compared vertices are going to be drawn with the same VAO,
445 * so we can compare the attributes. */
446 assert (k1->vertex_size == k2->vertex_size);
447 return memcmp(k1->vertex_attributes,
448 k2->vertex_attributes,
449 k1->vertex_size * sizeof(float)) == 0;
452 static void _free_entry(struct hash_entry *entry)
454 free((void*)entry->key);
457 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
458 * of an existing vertex, return the original index instead.
461 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
462 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
464 /* If vertex deduplication is disabled return the original index. */
468 fi_type *vert = save->vertex_store->buffer_in_ram + save->vertex_size * index;
470 struct vertex_key *key = malloc(sizeof(struct vertex_key));
471 key->vertex_size = save->vertex_size;
472 key->vertex_attributes = vert;
474 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
477 /* We found an existing vertex with the same hash, return its index. */
478 return (uintptr_t) entry->data;
480 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
481 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
482 * starting at index 0.
484 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
485 *max_index = MAX2(n, *max_index);
487 memcpy(&new_buffer[save->vertex_size * n],
489 save->vertex_size * sizeof(fi_type));
491 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
493 /* The index buffer is shared between list compilations, so add the base index to get
502 get_vertex_count(struct vbo_save_context *save)
504 if (!save->vertex_size)
506 return save->vertex_store->used / save->vertex_size;
511 * Insert the active immediate struct onto the display list currently
515 compile_vertex_list(struct gl_context *ctx)
517 struct vbo_save_context *save = &vbo_context(ctx)->save;
518 struct vbo_save_vertex_list *node;
520 /* Allocate space for this structure in the display list currently
523 node = (struct vbo_save_vertex_list *)
524 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
529 memset(node, 0, sizeof(struct vbo_save_vertex_list));
530 node->cold = calloc(1, sizeof(*node->cold));
532 /* Make sure the pointer is aligned to the size of a pointer */
533 assert((GLintptr) node % sizeof(void *) == 0);
535 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
537 node->cold->vertex_count = get_vertex_count(save);
538 node->cold->wrap_count = save->copied.nr;
539 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
540 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
541 node->cold->ib.obj = NULL;
542 node->cold->prim_count = save->prim_store->used;
544 if (save->no_current_update) {
545 node->cold->current_data = NULL;
548 GLuint current_size = save->vertex_size - save->attrsz[0];
549 node->cold->current_data = NULL;
552 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
553 if (node->cold->current_data) {
554 const char *buffer = (const char *)save->vertex_store->buffer_in_ram;
555 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
556 unsigned vertex_offset = 0;
558 if (node->cold->vertex_count)
559 vertex_offset = (node->cold->vertex_count - 1) * stride;
561 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
562 current_size * sizeof(GLfloat));
564 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
565 handle_out_of_memory(ctx);
570 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
572 if (save->dangling_attr_ref)
573 ctx->ListState.Current.UseLoopback = true;
575 /* Copy duplicated vertices
577 save->copied.nr = copy_vertices(ctx, node, save->vertex_store->buffer_in_ram);
579 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
580 convert_line_loop_to_strip(save, node);
583 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
585 GLintptr buffer_offset = 0;
586 GLuint start_offset = 0;
588 /* Create an index buffer. */
589 node->cold->min_index = node->cold->max_index = 0;
590 if (node->cold->vertex_count == 0 || node->cold->prim_count == 0)
593 /* We won't modify node->prims, so use a const alias to avoid unintended
595 const struct _mesa_prim *original_prims = node->cold->prims;
597 int end = original_prims[node->cold->prim_count - 1].start +
598 original_prims[node->cold->prim_count - 1].count;
599 int total_vert_count = end - original_prims[0].start;
601 node->cold->min_index = node->cold->prims[0].start;
602 node->cold->max_index = end - 1;
604 int max_index_count = total_vert_count * 2;
606 int size = max_index_count * sizeof(uint32_t);
607 uint32_t* indices = (uint32_t*) malloc(size);
608 struct _mesa_prim *merged_prims = NULL;
611 struct hash_table *vertex_to_index = NULL;
612 fi_type *temp_vertices_buffer = NULL;
614 /* The loopback replay code doesn't use the index buffer, so we can't
615 * dedup vertices in this case.
617 if (!ctx->ListState.Current.UseLoopback) {
618 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
619 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
622 uint32_t max_index = 0;
624 int last_valid_prim = -1;
625 /* Construct indices array. */
626 for (unsigned i = 0; i < node->cold->prim_count; i++) {
627 assert(original_prims[i].basevertex == 0);
628 GLubyte mode = original_prims[i].mode;
630 int vertex_count = original_prims[i].count;
635 /* Line strips may get converted to lines */
636 if (mode == GL_LINE_STRIP)
639 /* If 2 consecutive prims use the same mode => merge them. */
640 bool merge_prims = last_valid_prim >= 0 &&
641 mode == merged_prims[last_valid_prim].mode &&
642 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
643 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
646 /* To be able to merge consecutive triangle strips we need to insert
647 * a degenerate triangle.
650 mode == GL_TRIANGLE_STRIP) {
651 /* Insert a degenerate triangle */
652 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
653 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
655 indices[idx] = indices[idx - 1];
656 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
657 temp_vertices_buffer, &max_index);
659 merged_prims[last_valid_prim].count += 2;
662 /* Add another index to preserve winding order */
663 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
664 temp_vertices_buffer, &max_index);
665 merged_prims[last_valid_prim].count++;
671 /* Convert line strips to lines if it'll allow if the previous
672 * prim mode is GL_LINES (so merge_prims is true) or if the next
673 * primitive mode is GL_LINES or GL_LINE_LOOP.
675 if (original_prims[i].mode == GL_LINE_STRIP &&
677 (i < node->cold->prim_count - 1 &&
678 (original_prims[i + 1].mode == GL_LINE_STRIP ||
679 original_prims[i + 1].mode == GL_LINES)))) {
680 for (unsigned j = 0; j < vertex_count; j++) {
681 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
682 temp_vertices_buffer, &max_index);
683 /* Repeat all but the first/last indices. */
684 if (j && j != vertex_count - 1) {
685 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
686 temp_vertices_buffer, &max_index);
690 /* We didn't convert to LINES, so restore the original mode */
691 mode = original_prims[i].mode;
693 for (unsigned j = 0; j < vertex_count; j++) {
694 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
695 temp_vertices_buffer, &max_index);
700 /* Update vertex count. */
701 merged_prims[last_valid_prim].count += idx - start;
703 /* Keep this primitive */
704 last_valid_prim += 1;
705 assert(last_valid_prim <= i);
706 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
707 merged_prims[last_valid_prim] = original_prims[i];
708 merged_prims[last_valid_prim].start = start;
709 merged_prims[last_valid_prim].count = idx - start;
711 merged_prims[last_valid_prim].mode = mode;
714 assert(idx > 0 && idx <= max_index_count);
716 unsigned merged_prim_count = last_valid_prim + 1;
717 node->cold->ib.ptr = NULL;
718 node->cold->ib.count = idx;
719 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
721 /* How many bytes do we need to store the indices and the vertices */
722 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
723 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
724 total_vert_count * save->vertex_size * sizeof(fi_type);
726 const GLintptr old_offset = save->VAO[0] ?
727 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
728 if (old_offset != save->current_bo_bytes_used && stride > 0) {
729 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
730 while (offset_diff > 0 &&
731 save->current_bo_bytes_used < save->current_bo->Size &&
732 offset_diff % stride != 0) {
733 save->current_bo_bytes_used++;
734 offset_diff = save->current_bo_bytes_used - old_offset;
737 buffer_offset = save->current_bo_bytes_used;
739 /* Can we reuse the previous bo or should we allocate a new one? */
740 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
741 if (total_bytes_needed > available_bytes) {
742 if (save->current_bo)
743 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
744 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
745 bool success = ctx->Driver.BufferData(ctx,
746 GL_ELEMENT_ARRAY_BUFFER_ARB,
747 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE),
749 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
752 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
753 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
754 handle_out_of_memory(ctx);
756 save->current_bo_bytes_used = 0;
757 available_bytes = save->current_bo->Size;
761 assert(old_offset <= buffer_offset);
762 const GLintptr offset_diff = buffer_offset - old_offset;
763 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
764 /* The vertex size is an exact multiple of the buffer offset.
765 * This means that we can use zero-based vertex attribute pointers
766 * and specify the start of the primitive with the _mesa_prim::start
767 * field. This results in issuing several draw calls with identical
768 * vertex attribute information. This can result in fewer state
769 * changes in drivers. In particular, the Gallium CSO module will
770 * filter out redundant vertex buffer changes.
772 /* We cannot immediately update the primitives as some methods below
773 * still need the uncorrected start vertices
775 start_offset = offset_diff/stride;
776 assert(old_offset == buffer_offset - offset_diff);
777 buffer_offset = old_offset;
780 /* Correct the primitive starts, we can only do this here as copy_vertices
781 * and convert_line_loop_to_strip above consume the uncorrected starts.
782 * On the other hand the _vbo_loopback_vertex_list call below needs the
783 * primitives to be corrected already.
785 for (unsigned i = 0; i < node->cold->prim_count; i++) {
786 node->cold->prims[i].start += start_offset;
788 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
789 * to apply this transformation to all indices and max_index.
791 for (unsigned i = 0; i < idx; i++)
792 indices[i] += start_offset;
793 max_index += start_offset;
796 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
798 /* Upload the vertices first (see buffer_offset) */
799 ctx->Driver.BufferSubData(ctx,
800 save->current_bo_bytes_used,
801 total_vert_count * save->vertex_size * sizeof(fi_type),
802 vertex_to_index ? temp_vertices_buffer : save->vertex_store->buffer_in_ram,
804 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
806 if (vertex_to_index) {
807 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
808 free(temp_vertices_buffer);
811 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
812 * primitive (not the indices themselves). */
813 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
814 int indices_offset = save->current_bo_bytes_used / 4;
815 for (int i = 0; i < merged_prim_count; i++) {
816 merged_prims[i].start += indices_offset;
819 /* Then upload the indices. */
820 if (node->cold->ib.obj) {
821 ctx->Driver.BufferSubData(ctx,
822 save->current_bo_bytes_used,
823 idx * sizeof(uint32_t),
826 save->current_bo_bytes_used += idx * sizeof(uint32_t);
828 node->cold->vertex_count = 0;
829 node->cold->prim_count = 0;
832 /* Prepare for DrawGallium */
833 memset(&node->merged.info, 0, sizeof(struct pipe_draw_info));
834 /* The other info fields will be updated in vbo_save_playback_vertex_list */
835 node->merged.info.index_size = 4;
836 node->merged.info.instance_count = 1;
837 node->merged.info.index.gl_bo = node->cold->ib.obj;
838 if (merged_prim_count == 1) {
839 node->merged.info.mode = merged_prims[0].mode;
840 node->merged.start_count.start = merged_prims[0].start;
841 node->merged.start_count.count = merged_prims[0].count;
842 node->merged.start_count.index_bias = 0;
843 node->merged.mode = NULL;
845 node->merged.mode = malloc(merged_prim_count * sizeof(unsigned char));
846 node->merged.start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
847 for (unsigned i = 0; i < merged_prim_count; i++) {
848 node->merged.start_counts[i].start = merged_prims[i].start;
849 node->merged.start_counts[i].count = merged_prims[i].count;
850 node->merged.start_counts[i].index_bias = 0;
851 node->merged.mode[i] = merged_prims[i].mode;
854 node->merged.num_draws = merged_prim_count;
855 if (node->merged.num_draws > 1) {
856 bool same_mode = true;
857 for (unsigned i = 1; i < node->merged.num_draws && same_mode; i++) {
858 same_mode = node->merged.mode[i] == node->merged.mode[0];
861 /* All primitives use the same mode, so we can simplify a bit */
862 node->merged.info.mode = node->merged.mode[0];
863 free(node->merged.mode);
864 node->merged.mode = NULL;
873 if (!save->current_bo) {
874 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
875 bool success = ctx->Driver.BufferData(ctx,
876 GL_ELEMENT_ARRAY_BUFFER_ARB,
877 VBO_SAVE_BUFFER_SIZE,
879 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
882 handle_out_of_memory(ctx);
885 GLuint offsets[VBO_ATTRIB_MAX];
886 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
888 offset += save->attrsz[i] * sizeof(GLfloat);
890 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
891 * Note that this may reuse the previous one of possible.
893 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
894 /* create or reuse the vao */
895 update_vao(ctx, vpm, &save->VAO[vpm],
896 save->current_bo, buffer_offset, stride,
897 save->enabled, save->attrsz, save->attrtype, offsets);
898 /* Reference the vao in the dlist */
899 node->VAO[vpm] = NULL;
900 _mesa_reference_vao(ctx, &node->VAO[vpm], save->VAO[vpm]);
904 /* Deal with GL_COMPILE_AND_EXECUTE:
906 if (ctx->ExecuteFlag) {
907 struct _glapi_table *dispatch = GET_DISPATCH();
909 _glapi_set_dispatch(ctx->Exec);
911 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
912 * use buffer_in_ram instead of current_bo which contains all vertices instead
913 * of the deduplicated vertices only in the !UseLoopback case.
915 * The problem is that the VAO offset is based on current_bo's layout,
916 * so we have to use a temp value.
918 struct gl_vertex_array_object *vao = node->VAO[VP_MODE_SHADER];
919 GLintptr original = vao->BufferBinding[0].Offset;
920 if (!ctx->ListState.Current.UseLoopback) {
921 GLintptr new_offset = 0;
922 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
923 new_offset -= start_offset * stride;
924 vao->BufferBinding[0].Offset = new_offset;
926 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
927 vao->BufferBinding[0].Offset = original;
929 _glapi_set_dispatch(dispatch);
932 /* Reset our structures for the next run of vertices:
939 * This is called when we fill a vertex buffer before we hit a glEnd().
941 * TODO -- If no new vertices have been stored, don't bother saving it.
944 wrap_buffers(struct gl_context *ctx)
946 struct vbo_save_context *save = &vbo_context(ctx)->save;
947 GLint i = save->prim_store->used - 1;
950 assert(i < (GLint) save->prim_store->size);
953 /* Close off in-progress primitive.
955 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
956 mode = save->prim_store->prims[i].mode;
958 /* store the copied vertices, and allocate a new list.
960 compile_vertex_list(ctx);
962 /* Restart interrupted primitive
964 save->prim_store->prims[0].mode = mode;
965 save->prim_store->prims[0].begin = 0;
966 save->prim_store->prims[0].end = 0;
967 save->prim_store->prims[0].start = 0;
968 save->prim_store->prims[0].count = 0;
969 save->prim_store->used = 1;
974 * Called only when buffers are wrapped as the result of filling the
975 * vertex_store struct.
978 wrap_filled_vertex(struct gl_context *ctx)
980 struct vbo_save_context *save = &vbo_context(ctx)->save;
981 unsigned numComponents;
983 /* Emit a glEnd to close off the last vertex list.
987 assert(save->vertex_store->used == 0 && save->vertex_store->used == 0);
989 /* Copy stored stored vertices to start of new list.
991 numComponents = save->copied.nr * save->vertex_size;
993 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
995 assert(save->copied.buffer);
998 numComponents * sizeof(fi_type));
999 free(save->copied.buffer);
1000 save->copied.buffer = NULL;
1002 save->vertex_store->used = numComponents;
1007 copy_to_current(struct gl_context *ctx)
1009 struct vbo_save_context *save = &vbo_context(ctx)->save;
1010 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1013 const int i = u_bit_scan64(&enabled);
1014 assert(save->attrsz[i]);
1016 if (save->attrtype[i] == GL_DOUBLE ||
1017 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1018 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1020 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1021 save->attrptr[i], save->attrtype[i]);
1027 copy_from_current(struct gl_context *ctx)
1029 struct vbo_save_context *save = &vbo_context(ctx)->save;
1030 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1033 const int i = u_bit_scan64(&enabled);
1035 switch (save->attrsz[i]) {
1037 save->attrptr[i][3] = save->current[i][3];
1040 save->attrptr[i][2] = save->current[i][2];
1043 save->attrptr[i][1] = save->current[i][1];
1046 save->attrptr[i][0] = save->current[i][0];
1049 unreachable("Unexpected vertex attribute size");
1056 * Called when we increase the size of a vertex attribute. For example,
1057 * if we've seen one or more glTexCoord2f() calls and now we get a
1058 * glTexCoord3f() call.
1059 * Flush existing data, set new attrib size, replay copied vertices.
1062 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1064 struct vbo_save_context *save = &vbo_context(ctx)->save;
1069 /* Store the current run of vertices, and emit a GL_END. Emit a
1070 * BEGIN in the new buffer.
1072 if (save->vertex_store->used)
1075 assert(save->copied.nr == 0);
1077 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1078 * when the attribute already exists in the vertex and is having
1079 * its size increased.
1081 copy_to_current(ctx);
1085 oldsz = save->attrsz[attr];
1086 save->attrsz[attr] = newsz;
1087 save->enabled |= BITFIELD64_BIT(attr);
1089 save->vertex_size += newsz - oldsz;
1091 /* Recalculate all the attrptr[] values:
1094 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1095 if (save->attrsz[i]) {
1096 save->attrptr[i] = tmp;
1097 tmp += save->attrsz[i];
1100 save->attrptr[i] = NULL; /* will not be dereferenced. */
1104 /* Copy from current to repopulate the vertex with correct values.
1106 copy_from_current(ctx);
1108 /* Replay stored vertices to translate them to new format here.
1110 * If there are copied vertices and the new (upgraded) attribute
1111 * has not been defined before, this list is somewhat degenerate,
1112 * and will need fixup at runtime.
1114 if (save->copied.nr) {
1115 assert(save->copied.buffer);
1116 const fi_type *data = save->copied.buffer;
1117 grow_vertex_storage(ctx, save->copied.nr);
1118 fi_type *dest = save->vertex_store->buffer_in_ram;
1120 /* Need to note this and fix up at runtime (or loopback):
1122 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1124 save->dangling_attr_ref = GL_TRUE;
1127 for (i = 0; i < save->copied.nr; i++) {
1128 GLbitfield64 enabled = save->enabled;
1130 const int j = u_bit_scan64(&enabled);
1131 assert(save->attrsz[j]);
1134 const fi_type *src = oldsz ? data : save->current[attr];
1135 int copy = oldsz ? oldsz : newsz;
1136 for (k = 0; k < copy; k++)
1138 for (; k < newsz; k++) {
1139 switch (save->attrtype[j]) {
1141 dest[k] = FLOAT_AS_UNION(k == 3);
1144 dest[k] = INT_AS_UNION(k == 3);
1146 case GL_UNSIGNED_INT:
1147 dest[k] = UINT_AS_UNION(k == 3);
1150 dest[k] = FLOAT_AS_UNION(k == 3);
1151 assert(!"Unexpected type in upgrade_vertex");
1158 GLint sz = save->attrsz[j];
1159 for (int k = 0; k < sz; k++)
1167 save->vertex_store->used += save->vertex_size * save->copied.nr;
1168 free(save->copied.buffer);
1169 save->copied.buffer = NULL;
1175 * This is called when the size of a vertex attribute changes.
1176 * For example, after seeing one or more glTexCoord2f() calls we
1177 * get a glTexCoord4f() or glTexCoord1f() call.
1180 fixup_vertex(struct gl_context *ctx, GLuint attr,
1181 GLuint sz, GLenum newType)
1183 struct vbo_save_context *save = &vbo_context(ctx)->save;
1185 if (sz > save->attrsz[attr] ||
1186 newType != save->attrtype[attr]) {
1187 /* New size is larger. Need to flush existing vertices and get
1188 * an enlarged vertex format.
1190 upgrade_vertex(ctx, attr, sz);
1192 else if (sz < save->active_sz[attr]) {
1194 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1196 /* New size is equal or smaller - just need to fill in some
1199 for (i = sz; i <= save->attrsz[attr]; i++)
1200 save->attrptr[attr][i - 1] = id[i - 1];
1203 save->active_sz[attr] = sz;
1205 grow_vertex_storage(ctx, 1);
1210 * Reset the current size of all vertex attributes to the default
1211 * value of 0. This signals that we haven't yet seen any per-vertex
1212 * commands such as glNormal3f() or glTexCoord2f().
1215 reset_vertex(struct gl_context *ctx)
1217 struct vbo_save_context *save = &vbo_context(ctx)->save;
1219 while (save->enabled) {
1220 const int i = u_bit_scan64(&save->enabled);
1221 assert(save->attrsz[i]);
1222 save->attrsz[i] = 0;
1223 save->active_sz[i] = 0;
1226 save->vertex_size = 0;
1231 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1232 * It depends on a few things, including whether we're inside or outside
1236 is_vertex_position(const struct gl_context *ctx, GLuint index)
1238 return (index == 0 &&
1239 _mesa_attr_zero_aliases_vertex(ctx) &&
1240 _mesa_inside_dlist_begin_end(ctx));
1245 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1248 /* Only one size for each attribute may be active at once. Eg. if
1249 * Color3f is installed/active, then Color4f may not be, even if the
1250 * vertex actually contains 4 color coordinates. This is because the
1251 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1252 * of the chooser function when switching between Color4f and Color3f.
1254 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1256 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1257 int sz = (sizeof(C) / sizeof(GLfloat)); \
1259 if (save->active_sz[A] != N) \
1260 fixup_vertex(ctx, A, N * sz, T); \
1263 C *dest = (C *)save->attrptr[A]; \
1264 if (N>0) dest[0] = V0; \
1265 if (N>1) dest[1] = V1; \
1266 if (N>2) dest[2] = V2; \
1267 if (N>3) dest[3] = V3; \
1268 save->attrtype[A] = T; \
1271 if ((A) == VBO_ATTRIB_POS) { \
1272 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + \
1273 save->vertex_store->used; \
1275 for (int i = 0; i < save->vertex_size; i++) \
1276 buffer_ptr[i] = save->vertex[i]; \
1278 save->vertex_store->used += save->vertex_size; \
1279 unsigned used_next = (save->vertex_store->used + \
1280 save->vertex_size) * sizeof(float); \
1281 if (used_next > save->vertex_store->buffer_in_ram_size) { \
1282 grow_vertex_storage(ctx, get_vertex_count(save)); \
1283 assert(used_next <= \
1284 save->vertex_store->buffer_in_ram_size); \
1289 #define TAG(x) _save_##x
1291 #include "vbo_attrib_tmp.h"
1294 #define MAT( ATTR, N, face, params ) \
1296 if (face != GL_BACK) \
1297 MAT_ATTR( ATTR, N, params ); /* front */ \
1298 if (face != GL_FRONT) \
1299 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1304 * Save a glMaterial call found between glBegin/End.
1305 * glMaterial calls outside Begin/End are handled in dlist.c.
1307 static void GLAPIENTRY
1308 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1310 GET_CURRENT_CONTEXT(ctx);
1312 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1313 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1319 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1322 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1325 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1328 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1331 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1332 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1335 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1338 case GL_COLOR_INDEXES:
1339 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1341 case GL_AMBIENT_AND_DIFFUSE:
1342 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1343 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1346 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1352 /* Cope with EvalCoord/CallList called within a begin/end object:
1353 * -- Flush current buffer
1354 * -- Fallback to opcodes for the rest of the begin/end object.
1357 dlist_fallback(struct gl_context *ctx)
1359 struct vbo_save_context *save = &vbo_context(ctx)->save;
1361 if (save->vertex_store->used || save->prim_store->used) {
1362 if (save->prim_store->used > 0 && save->vertex_store->used > 0) {
1363 assert(save->vertex_size);
1364 /* Close off in-progress primitive. */
1365 GLint i = save->prim_store->used - 1;
1366 save->prim_store->prims[i].count =
1367 get_vertex_count(save) -
1368 save->prim_store->prims[i].start;
1371 /* Need to replay this display list with loopback,
1372 * unfortunately, otherwise this primitive won't be handled
1375 save->dangling_attr_ref = GL_TRUE;
1377 compile_vertex_list(ctx);
1380 copy_to_current(ctx);
1382 if (save->out_of_memory) {
1383 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1386 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1388 ctx->Driver.SaveNeedFlush = GL_FALSE;
1392 static void GLAPIENTRY
1393 _save_EvalCoord1f(GLfloat u)
1395 GET_CURRENT_CONTEXT(ctx);
1396 dlist_fallback(ctx);
1397 CALL_EvalCoord1f(ctx->Save, (u));
1400 static void GLAPIENTRY
1401 _save_EvalCoord1fv(const GLfloat * v)
1403 GET_CURRENT_CONTEXT(ctx);
1404 dlist_fallback(ctx);
1405 CALL_EvalCoord1fv(ctx->Save, (v));
1408 static void GLAPIENTRY
1409 _save_EvalCoord2f(GLfloat u, GLfloat v)
1411 GET_CURRENT_CONTEXT(ctx);
1412 dlist_fallback(ctx);
1413 CALL_EvalCoord2f(ctx->Save, (u, v));
1416 static void GLAPIENTRY
1417 _save_EvalCoord2fv(const GLfloat * v)
1419 GET_CURRENT_CONTEXT(ctx);
1420 dlist_fallback(ctx);
1421 CALL_EvalCoord2fv(ctx->Save, (v));
1424 static void GLAPIENTRY
1425 _save_EvalPoint1(GLint i)
1427 GET_CURRENT_CONTEXT(ctx);
1428 dlist_fallback(ctx);
1429 CALL_EvalPoint1(ctx->Save, (i));
1432 static void GLAPIENTRY
1433 _save_EvalPoint2(GLint i, GLint j)
1435 GET_CURRENT_CONTEXT(ctx);
1436 dlist_fallback(ctx);
1437 CALL_EvalPoint2(ctx->Save, (i, j));
1440 static void GLAPIENTRY
1441 _save_CallList(GLuint l)
1443 GET_CURRENT_CONTEXT(ctx);
1444 dlist_fallback(ctx);
1445 CALL_CallList(ctx->Save, (l));
1448 static void GLAPIENTRY
1449 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1451 GET_CURRENT_CONTEXT(ctx);
1452 dlist_fallback(ctx);
1453 CALL_CallLists(ctx->Save, (n, type, v));
1459 * Called when a glBegin is getting compiled into a display list.
1460 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1463 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1464 bool no_current_update)
1466 struct vbo_save_context *save = &vbo_context(ctx)->save;
1467 const GLuint i = save->prim_store->used++;
1469 ctx->Driver.CurrentSavePrimitive = mode;
1471 if (!save->prim_store || i >= save->prim_store->size) {
1472 save->prim_store = realloc_prim_store(save->prim_store, i * 2);
1474 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1475 save->prim_store->prims[i].begin = 1;
1476 save->prim_store->prims[i].end = 0;
1477 save->prim_store->prims[i].start = get_vertex_count(save);
1478 save->prim_store->prims[i].count = 0;
1480 save->no_current_update = no_current_update;
1482 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1484 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1485 ctx->Driver.SaveNeedFlush = GL_TRUE;
1489 static void GLAPIENTRY
1492 GET_CURRENT_CONTEXT(ctx);
1493 struct vbo_save_context *save = &vbo_context(ctx)->save;
1494 const GLint i = save->prim_store->used - 1;
1496 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1497 save->prim_store->prims[i].end = 1;
1498 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1500 /* Swap out this vertex format while outside begin/end. Any color,
1501 * etc. received between here and the next begin will be compiled
1504 if (save->out_of_memory) {
1505 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1508 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1513 static void GLAPIENTRY
1514 _save_Begin(GLenum mode)
1516 GET_CURRENT_CONTEXT(ctx);
1518 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1522 static void GLAPIENTRY
1523 _save_PrimitiveRestartNV(void)
1525 GET_CURRENT_CONTEXT(ctx);
1526 struct vbo_save_context *save = &vbo_context(ctx)->save;
1528 if (save->prim_store->used == 0) {
1529 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1532 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1533 "glPrimitiveRestartNV called outside glBegin/End");
1535 /* get current primitive mode */
1536 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1537 bool no_current_update = save->no_current_update;
1539 /* restart primitive */
1540 CALL_End(ctx->CurrentServerDispatch, ());
1541 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1546 /* Unlike the functions above, these are to be hooked into the vtxfmt
1547 * maintained in ctx->ListState, active when the list is known or
1548 * suspected to be outside any begin/end primitive.
1549 * Note: OBE = Outside Begin/End
1551 static void GLAPIENTRY
1552 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1554 GET_CURRENT_CONTEXT(ctx);
1555 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1557 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1558 CALL_Vertex2f(dispatch, (x1, y1));
1559 CALL_Vertex2f(dispatch, (x2, y1));
1560 CALL_Vertex2f(dispatch, (x2, y2));
1561 CALL_Vertex2f(dispatch, (x1, y2));
1562 CALL_End(dispatch, ());
1566 static void GLAPIENTRY
1567 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1569 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1572 static void GLAPIENTRY
1573 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1575 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1578 static void GLAPIENTRY
1579 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1581 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1584 static void GLAPIENTRY
1585 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1587 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1590 static void GLAPIENTRY
1591 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1593 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1596 static void GLAPIENTRY
1597 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1599 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1602 static void GLAPIENTRY
1603 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1605 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1608 static void GLAPIENTRY
1609 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1611 GET_CURRENT_CONTEXT(ctx);
1612 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1613 struct vbo_save_context *save = &vbo_context(ctx)->save;
1616 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1617 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1621 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1625 if (save->out_of_memory)
1628 grow_vertex_storage(ctx, count);
1630 /* Make sure to process any VBO binding changes */
1631 _mesa_update_state(ctx);
1633 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1635 vbo_save_NotifyBegin(ctx, mode, true);
1637 for (i = 0; i < count; i++)
1638 _mesa_array_element(ctx, start + i);
1639 CALL_End(ctx->CurrentServerDispatch, ());
1641 _mesa_vao_unmap_arrays(ctx, vao);
1645 static void GLAPIENTRY
1646 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1647 const GLsizei *count, GLsizei primcount)
1649 GET_CURRENT_CONTEXT(ctx);
1652 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1653 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1657 if (primcount < 0) {
1658 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1659 "glMultiDrawArrays(primcount<0)");
1663 unsigned vertcount = 0;
1664 for (i = 0; i < primcount; i++) {
1666 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1667 "glMultiDrawArrays(count[i]<0)");
1670 vertcount += count[i];
1673 grow_vertex_storage(ctx, vertcount);
1675 for (i = 0; i < primcount; i++) {
1677 _save_OBE_DrawArrays(mode, first[i], count[i]);
1684 array_element(struct gl_context *ctx,
1685 GLint basevertex, GLuint elt, unsigned index_size_shift)
1687 /* Section 10.3.5 Primitive Restart:
1689 * When one of the *BaseVertex drawing commands specified in section 10.5
1690 * is used, the primitive restart comparison occurs before the basevertex
1691 * offset is added to the array index.
1693 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1694 * then we call PrimitiveRestartNV and return.
1696 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1697 elt == ctx->Array._RestartIndex[index_size_shift]) {
1698 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1702 _mesa_array_element(ctx, basevertex + elt);
1706 /* Could do better by copying the arrays and element list intact and
1707 * then emitting an indexed prim at runtime.
1709 static void GLAPIENTRY
1710 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1711 const GLvoid * indices, GLint basevertex)
1713 GET_CURRENT_CONTEXT(ctx);
1714 struct vbo_save_context *save = &vbo_context(ctx)->save;
1715 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1716 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1719 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1720 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1724 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1727 if (type != GL_UNSIGNED_BYTE &&
1728 type != GL_UNSIGNED_SHORT &&
1729 type != GL_UNSIGNED_INT) {
1730 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1734 if (save->out_of_memory)
1737 grow_vertex_storage(ctx, count);
1739 /* Make sure to process any VBO binding changes */
1740 _mesa_update_state(ctx);
1742 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1746 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1748 vbo_save_NotifyBegin(ctx, mode, true);
1751 case GL_UNSIGNED_BYTE:
1752 for (i = 0; i < count; i++)
1753 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1755 case GL_UNSIGNED_SHORT:
1756 for (i = 0; i < count; i++)
1757 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1759 case GL_UNSIGNED_INT:
1760 for (i = 0; i < count; i++)
1761 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1764 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1768 CALL_End(ctx->CurrentServerDispatch, ());
1770 _mesa_vao_unmap(ctx, vao);
1773 static void GLAPIENTRY
1774 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1775 const GLvoid * indices)
1777 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1781 static void GLAPIENTRY
1782 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1783 GLsizei count, GLenum type,
1784 const GLvoid * indices)
1786 GET_CURRENT_CONTEXT(ctx);
1787 struct vbo_save_context *save = &vbo_context(ctx)->save;
1789 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1790 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1794 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1795 "glDrawRangeElements(count<0)");
1798 if (type != GL_UNSIGNED_BYTE &&
1799 type != GL_UNSIGNED_SHORT &&
1800 type != GL_UNSIGNED_INT) {
1801 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1805 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1806 "glDrawRangeElements(end < start)");
1810 if (save->out_of_memory)
1813 _save_OBE_DrawElements(mode, count, type, indices);
1817 static void GLAPIENTRY
1818 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1819 const GLvoid * const *indices, GLsizei primcount)
1821 GET_CURRENT_CONTEXT(ctx);
1822 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1826 for (i = 0; i < primcount; i++) {
1827 vertcount += count[i];
1829 grow_vertex_storage(ctx, vertcount);
1831 for (i = 0; i < primcount; i++) {
1833 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1839 static void GLAPIENTRY
1840 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1842 const GLvoid * const *indices,
1844 const GLint *basevertex)
1846 GET_CURRENT_CONTEXT(ctx);
1847 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1851 for (i = 0; i < primcount; i++) {
1852 vertcount += count[i];
1854 grow_vertex_storage(ctx, vertcount);
1856 for (i = 0; i < primcount; i++) {
1858 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1867 vtxfmt_init(struct gl_context *ctx)
1869 struct vbo_save_context *save = &vbo_context(ctx)->save;
1870 GLvertexformat *vfmt = &save->vtxfmt;
1872 #define NAME_AE(x) _ae_##x
1873 #define NAME_CALLLIST(x) _save_##x
1874 #define NAME(x) _save_##x
1875 #define NAME_ES(x) _save_##x##ARB
1877 #include "vbo_init_tmp.h"
1882 * Initialize the dispatch table with the VBO functions for display
1886 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1887 struct _glapi_table *exec)
1889 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1890 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1891 SET_DrawElements(exec, _save_OBE_DrawElements);
1892 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1893 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1894 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1895 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1896 SET_Rectf(exec, _save_OBE_Rectf);
1897 SET_Rectd(exec, _save_OBE_Rectd);
1898 SET_Rectdv(exec, _save_OBE_Rectdv);
1899 SET_Rectfv(exec, _save_OBE_Rectfv);
1900 SET_Recti(exec, _save_OBE_Recti);
1901 SET_Rectiv(exec, _save_OBE_Rectiv);
1902 SET_Rects(exec, _save_OBE_Rects);
1903 SET_Rectsv(exec, _save_OBE_Rectsv);
1905 /* Note: other glDraw functins aren't compiled into display lists */
1911 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1913 struct vbo_save_context *save = &vbo_context(ctx)->save;
1915 /* Noop when we are actually active:
1917 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1920 if (save->vertex_store->used || save->prim_store->used)
1921 compile_vertex_list(ctx);
1923 copy_to_current(ctx);
1925 ctx->Driver.SaveNeedFlush = GL_FALSE;
1930 * Called from glNewList when we're starting to compile a display list.
1933 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1935 struct vbo_save_context *save = &vbo_context(ctx)->save;
1940 if (!save->prim_store)
1941 save->prim_store = realloc_prim_store(NULL, 8);
1943 if (!save->vertex_store)
1944 save->vertex_store = CALLOC_STRUCT(vbo_save_vertex_store);
1947 ctx->Driver.SaveNeedFlush = GL_FALSE;
1952 * Called from glEndList when we're finished compiling a display list.
1955 vbo_save_EndList(struct gl_context *ctx)
1957 struct vbo_save_context *save = &vbo_context(ctx)->save;
1959 /* EndList called inside a (saved) Begin/End pair?
1961 if (_mesa_inside_dlist_begin_end(ctx)) {
1962 if (save->prim_store->used > 0) {
1963 GLint i = save->prim_store->used - 1;
1964 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1965 save->prim_store->prims[i].end = 0;
1966 save->prim_store->prims[i].count = get_vertex_count(save) - save->prim_store->prims[i].start;
1969 /* Make sure this vertex list gets replayed by the "loopback"
1972 save->dangling_attr_ref = GL_TRUE;
1973 vbo_save_SaveFlushVertices(ctx);
1975 /* Swap out this vertex format while outside begin/end. Any color,
1976 * etc. received between here and the next begin will be compiled
1979 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1982 assert(save->vertex_size == 0);
1986 * Called during context creation/init.
1989 current_init(struct gl_context *ctx)
1991 struct vbo_save_context *save = &vbo_context(ctx)->save;
1994 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
1995 const GLuint j = i - VBO_ATTRIB_POS;
1996 assert(j < VERT_ATTRIB_MAX);
1997 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
1998 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2001 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2002 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2003 assert(j < MAT_ATTRIB_MAX);
2004 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2005 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2011 * Initialize the display list compiler. Called during context creation.
2014 vbo_save_api_init(struct vbo_save_context *save)
2016 struct gl_context *ctx = gl_context_from_vbo_save(save);