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 unsigned r = vbo_copy_vertices(ctx, prim->mode, prim->start, &prim->count,
172 prim->begin, sz, true, save->copied.buffer, src);
174 free(save->copied.buffer);
175 save->copied.buffer = NULL;
181 static struct vbo_save_primitive_store *
182 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
185 store = CALLOC_STRUCT(vbo_save_primitive_store);
187 uint32_t old_size = store->size;
188 store->size = prim_count;
189 assert (old_size < store->size);
190 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
191 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
198 reset_counters(struct gl_context *ctx)
200 struct vbo_save_context *save = &vbo_context(ctx)->save;
202 save->vertex_store->used = 0;
203 save->prim_store->used = 0;
204 save->dangling_attr_ref = GL_FALSE;
208 * For a list of prims, try merging prims that can just be extensions of the
212 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
216 struct _mesa_prim *prev_prim = prim_list;
218 for (i = 1; i < *prim_count; i++) {
219 struct _mesa_prim *this_prim = prim_list + i;
221 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
223 if (vbo_merge_draws(ctx, true,
224 prev_prim->mode, this_prim->mode,
225 prev_prim->start, this_prim->start,
226 &prev_prim->count, this_prim->count,
227 prev_prim->basevertex, this_prim->basevertex,
229 this_prim->begin, this_prim->end)) {
230 /* We've found a prim that just extend the previous one. Tack it
231 * onto the previous one, and let this primitive struct get dropped.
236 /* If any previous primitives have been dropped, then we need to copy
237 * this later one into the next available slot.
240 if (prev_prim != this_prim)
241 *prev_prim = *this_prim;
244 *prim_count = prev_prim - prim_list + 1;
249 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
250 * don't have to worry about handling the _mesa_prim::begin/end flags.
251 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
254 convert_line_loop_to_strip(struct vbo_save_context *save,
255 struct vbo_save_vertex_list *node)
257 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
259 assert(prim->mode == GL_LINE_LOOP);
262 /* Copy the 0th vertex to end of the buffer and extend the
263 * vertex count by one to finish the line loop.
265 const GLuint sz = save->vertex_size;
267 const fi_type *src = save->vertex_store->buffer_in_ram + prim->start * sz;
269 fi_type *dst = save->vertex_store->buffer_in_ram + (prim->start + prim->count) * sz;
271 memcpy(dst, src, sz * sizeof(float));
274 node->cold->vertex_count++;
275 save->vertex_store->used += sz;
279 /* Drawing the second or later section of a long line loop.
280 * Skip the 0th vertex.
286 prim->mode = GL_LINE_STRIP;
290 /* Compare the present vao if it has the same setup. */
292 compare_vao(gl_vertex_processing_mode mode,
293 const struct gl_vertex_array_object *vao,
294 const struct gl_buffer_object *bo, GLintptr buffer_offset,
295 GLuint stride, GLbitfield64 vao_enabled,
296 const GLubyte size[VBO_ATTRIB_MAX],
297 const GLenum16 type[VBO_ATTRIB_MAX],
298 const GLuint offset[VBO_ATTRIB_MAX])
303 /* If the enabled arrays are not the same we are not equal. */
304 if (vao_enabled != vao->Enabled)
307 /* Check the buffer binding at 0 */
308 if (vao->BufferBinding[0].BufferObj != bo)
310 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
311 if (vao->BufferBinding[0].Stride != stride)
313 assert(vao->BufferBinding[0].InstanceDivisor == 0);
315 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
316 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
318 /* Now check the enabled arrays */
319 GLbitfield mask = vao_enabled;
321 const int attr = u_bit_scan(&mask);
322 const unsigned char vbo_attr = vao_to_vbo_map[attr];
323 const GLenum16 tp = type[vbo_attr];
324 const GLintptr off = offset[vbo_attr] + buffer_offset;
325 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
326 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
328 if (attrib->Format.Type != tp)
330 if (attrib->Format.Size != size[vbo_attr])
332 assert(attrib->Format.Format == GL_RGBA);
333 assert(attrib->Format.Normalized == GL_FALSE);
334 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
335 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
336 assert(attrib->BufferBindingIndex == 0);
343 /* Create or reuse the vao for the vertex processing mode. */
345 update_vao(struct gl_context *ctx,
346 gl_vertex_processing_mode mode,
347 struct gl_vertex_array_object **vao,
348 struct gl_buffer_object *bo, GLintptr buffer_offset,
349 GLuint stride, GLbitfield64 vbo_enabled,
350 const GLubyte size[VBO_ATTRIB_MAX],
351 const GLenum16 type[VBO_ATTRIB_MAX],
352 const GLuint offset[VBO_ATTRIB_MAX])
354 /* Compute the bitmasks of vao_enabled arrays */
355 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
358 * Check if we can possibly reuse the exisiting one.
359 * In the long term we should reset them when something changes.
361 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
362 vao_enabled, size, type, offset))
365 /* The initial refcount is 1 */
366 _mesa_reference_vao(ctx, vao, NULL);
367 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
370 * assert(stride <= ctx->Const.MaxVertexAttribStride);
371 * MaxVertexAttribStride is not set for drivers that does not
372 * expose GL 44 or GLES 31.
375 /* Bind the buffer object at binding point 0 */
376 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
379 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
380 * Note that the position/generic0 aliasing is done in the VAO.
382 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
383 /* Now set the enable arrays */
384 GLbitfield mask = vao_enabled;
386 const int vao_attr = u_bit_scan(&mask);
387 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
388 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
390 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
391 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
392 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
394 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
395 assert(vao_enabled == (*vao)->Enabled);
396 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
398 /* Finalize and freeze the VAO */
399 _mesa_set_vao_immutable(ctx, *vao);
402 static void wrap_filled_vertex(struct gl_context *ctx);
404 /* Grow the vertex storage to accomodate for vertex_count new vertices */
406 grow_vertex_storage(struct gl_context *ctx, int vertex_count)
408 struct vbo_save_context *save = &vbo_context(ctx)->save;
409 assert (save->vertex_store);
411 int new_size = (save->vertex_store->used +
412 vertex_count * save->vertex_size) * sizeof(GLfloat);
414 /* Limit how much memory we allocate. */
415 if (save->prim_store->used > 0 &&
417 new_size > VBO_SAVE_BUFFER_SIZE) {
418 wrap_filled_vertex(ctx);
419 new_size = VBO_SAVE_BUFFER_SIZE;
422 if (new_size > save->vertex_store->buffer_in_ram_size) {
423 save->vertex_store->buffer_in_ram_size = new_size;
424 save->vertex_store->buffer_in_ram = realloc(save->vertex_store->buffer_in_ram,
425 save->vertex_store->buffer_in_ram_size);
426 if (save->vertex_store->buffer_in_ram == NULL)
427 handle_out_of_memory(ctx);
433 unsigned vertex_size;
434 fi_type *vertex_attributes;
437 static uint32_t _hash_vertex_key(const void *key)
439 struct vertex_key *k = (struct vertex_key*)key;
440 unsigned sz = k->vertex_size;
442 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
445 static bool _compare_vertex_key(const void *key1, const void *key2)
447 struct vertex_key *k1 = (struct vertex_key*)key1;
448 struct vertex_key *k2 = (struct vertex_key*)key2;
449 /* All the compared vertices are going to be drawn with the same VAO,
450 * so we can compare the attributes. */
451 assert (k1->vertex_size == k2->vertex_size);
452 return memcmp(k1->vertex_attributes,
453 k2->vertex_attributes,
454 k1->vertex_size * sizeof(float)) == 0;
457 static void _free_entry(struct hash_entry *entry)
459 free((void*)entry->key);
462 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
463 * of an existing vertex, return the original index instead.
466 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
467 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
469 /* If vertex deduplication is disabled return the original index. */
473 fi_type *vert = save->vertex_store->buffer_in_ram + save->vertex_size * index;
475 struct vertex_key *key = malloc(sizeof(struct vertex_key));
476 key->vertex_size = save->vertex_size;
477 key->vertex_attributes = vert;
479 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
482 /* We found an existing vertex with the same hash, return its index. */
483 return (uintptr_t) entry->data;
485 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
486 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
487 * starting at index 0.
489 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
490 *max_index = MAX2(n, *max_index);
492 memcpy(&new_buffer[save->vertex_size * n],
494 save->vertex_size * sizeof(fi_type));
496 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
498 /* The index buffer is shared between list compilations, so add the base index to get
507 get_vertex_count(struct vbo_save_context *save)
509 if (!save->vertex_size)
511 return save->vertex_store->used / save->vertex_size;
516 * Insert the active immediate struct onto the display list currently
520 compile_vertex_list(struct gl_context *ctx)
522 struct vbo_save_context *save = &vbo_context(ctx)->save;
523 struct vbo_save_vertex_list *node;
525 /* Allocate space for this structure in the display list currently
528 node = (struct vbo_save_vertex_list *)
529 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
534 node->cold = calloc(1, sizeof(*node->cold));
536 /* Make sure the pointer is aligned to the size of a pointer */
537 assert((GLintptr) node % sizeof(void *) == 0);
539 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
541 node->cold->vertex_count = get_vertex_count(save);
542 node->cold->wrap_count = save->copied.nr;
543 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
544 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
545 node->cold->ib.obj = NULL;
546 node->cold->prim_count = save->prim_store->used;
548 if (save->no_current_update) {
549 node->cold->current_data = NULL;
552 GLuint current_size = save->vertex_size - save->attrsz[0];
553 node->cold->current_data = NULL;
556 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
557 if (node->cold->current_data) {
558 const char *buffer = (const char *)save->vertex_store->buffer_in_ram;
559 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
560 unsigned vertex_offset = 0;
562 if (node->cold->vertex_count)
563 vertex_offset = (node->cold->vertex_count - 1) * stride;
565 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
566 current_size * sizeof(GLfloat));
568 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
569 handle_out_of_memory(ctx);
574 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
576 if (save->dangling_attr_ref)
577 ctx->ListState.Current.UseLoopback = true;
579 /* Copy duplicated vertices
581 save->copied.nr = copy_vertices(ctx, node, save->vertex_store->buffer_in_ram);
583 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
584 convert_line_loop_to_strip(save, node);
587 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
589 GLintptr buffer_offset = 0;
590 GLuint start_offset = 0;
592 /* Create an index buffer. */
593 node->cold->min_index = node->cold->max_index = 0;
594 if (node->cold->vertex_count == 0 || node->cold->prim_count == 0)
597 /* We won't modify node->prims, so use a const alias to avoid unintended
599 const struct _mesa_prim *original_prims = node->cold->prims;
601 int end = original_prims[node->cold->prim_count - 1].start +
602 original_prims[node->cold->prim_count - 1].count;
603 int total_vert_count = end - original_prims[0].start;
605 node->cold->min_index = node->cold->prims[0].start;
606 node->cold->max_index = end - 1;
608 int max_index_count = total_vert_count * 2;
610 int size = max_index_count * sizeof(uint32_t);
611 uint32_t* indices = (uint32_t*) malloc(size);
612 struct _mesa_prim *merged_prims = NULL;
615 struct hash_table *vertex_to_index = NULL;
616 fi_type *temp_vertices_buffer = NULL;
618 /* The loopback replay code doesn't use the index buffer, so we can't
619 * dedup vertices in this case.
621 if (!ctx->ListState.Current.UseLoopback) {
622 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
623 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
626 uint32_t max_index = 0;
628 int last_valid_prim = -1;
629 /* Construct indices array. */
630 for (unsigned i = 0; i < node->cold->prim_count; i++) {
631 assert(original_prims[i].basevertex == 0);
632 GLubyte mode = original_prims[i].mode;
634 int vertex_count = original_prims[i].count;
639 /* Line strips may get converted to lines */
640 if (mode == GL_LINE_STRIP)
643 /* If 2 consecutive prims use the same mode => merge them. */
644 bool merge_prims = last_valid_prim >= 0 &&
645 mode == merged_prims[last_valid_prim].mode &&
646 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
647 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
650 /* To be able to merge consecutive triangle strips we need to insert
651 * a degenerate triangle.
654 mode == GL_TRIANGLE_STRIP) {
655 /* Insert a degenerate triangle */
656 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
657 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
659 indices[idx] = indices[idx - 1];
660 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
661 temp_vertices_buffer, &max_index);
663 merged_prims[last_valid_prim].count += 2;
666 /* Add another index to preserve winding order */
667 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
668 temp_vertices_buffer, &max_index);
669 merged_prims[last_valid_prim].count++;
675 /* Convert line strips to lines if it'll allow if the previous
676 * prim mode is GL_LINES (so merge_prims is true) or if the next
677 * primitive mode is GL_LINES or GL_LINE_LOOP.
679 if (original_prims[i].mode == GL_LINE_STRIP &&
681 (i < node->cold->prim_count - 1 &&
682 (original_prims[i + 1].mode == GL_LINE_STRIP ||
683 original_prims[i + 1].mode == GL_LINES)))) {
684 for (unsigned j = 0; j < vertex_count; j++) {
685 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
686 temp_vertices_buffer, &max_index);
687 /* Repeat all but the first/last indices. */
688 if (j && j != vertex_count - 1) {
689 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
690 temp_vertices_buffer, &max_index);
694 /* We didn't convert to LINES, so restore the original mode */
695 mode = original_prims[i].mode;
697 for (unsigned j = 0; j < vertex_count; j++) {
698 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
699 temp_vertices_buffer, &max_index);
704 /* Update vertex count. */
705 merged_prims[last_valid_prim].count += idx - start;
707 /* Keep this primitive */
708 last_valid_prim += 1;
709 assert(last_valid_prim <= i);
710 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
711 merged_prims[last_valid_prim] = original_prims[i];
712 merged_prims[last_valid_prim].start = start;
713 merged_prims[last_valid_prim].count = idx - start;
715 merged_prims[last_valid_prim].mode = mode;
718 assert(idx > 0 && idx <= max_index_count);
720 unsigned merged_prim_count = last_valid_prim + 1;
721 node->cold->ib.ptr = NULL;
722 node->cold->ib.count = idx;
723 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
725 /* How many bytes do we need to store the indices and the vertices */
726 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
727 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
728 total_vert_count * save->vertex_size * sizeof(fi_type);
730 const GLintptr old_offset = save->VAO[0] ?
731 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
732 if (old_offset != save->current_bo_bytes_used && stride > 0) {
733 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
734 while (offset_diff > 0 &&
735 save->current_bo_bytes_used < save->current_bo->Size &&
736 offset_diff % stride != 0) {
737 save->current_bo_bytes_used++;
738 offset_diff = save->current_bo_bytes_used - old_offset;
741 buffer_offset = save->current_bo_bytes_used;
743 /* Can we reuse the previous bo or should we allocate a new one? */
744 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
745 if (total_bytes_needed > available_bytes) {
746 if (save->current_bo)
747 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
748 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
749 bool success = ctx->Driver.BufferData(ctx,
750 GL_ELEMENT_ARRAY_BUFFER_ARB,
751 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE),
753 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
754 MESA_GALLIUM_VERTEX_STATE_STORAGE,
757 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
758 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
759 handle_out_of_memory(ctx);
761 save->current_bo_bytes_used = 0;
762 available_bytes = save->current_bo->Size;
766 assert(old_offset <= buffer_offset);
767 const GLintptr offset_diff = buffer_offset - old_offset;
768 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
769 /* The vertex size is an exact multiple of the buffer offset.
770 * This means that we can use zero-based vertex attribute pointers
771 * and specify the start of the primitive with the _mesa_prim::start
772 * field. This results in issuing several draw calls with identical
773 * vertex attribute information. This can result in fewer state
774 * changes in drivers. In particular, the Gallium CSO module will
775 * filter out redundant vertex buffer changes.
777 /* We cannot immediately update the primitives as some methods below
778 * still need the uncorrected start vertices
780 start_offset = offset_diff/stride;
781 assert(old_offset == buffer_offset - offset_diff);
782 buffer_offset = old_offset;
785 /* Correct the primitive starts, we can only do this here as copy_vertices
786 * and convert_line_loop_to_strip above consume the uncorrected starts.
787 * On the other hand the _vbo_loopback_vertex_list call below needs the
788 * primitives to be corrected already.
790 for (unsigned i = 0; i < node->cold->prim_count; i++) {
791 node->cold->prims[i].start += start_offset;
793 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
794 * to apply this transformation to all indices and max_index.
796 for (unsigned i = 0; i < idx; i++)
797 indices[i] += start_offset;
798 max_index += start_offset;
801 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
803 /* Upload the vertices first (see buffer_offset) */
804 ctx->Driver.BufferSubData(ctx,
805 save->current_bo_bytes_used,
806 total_vert_count * save->vertex_size * sizeof(fi_type),
807 vertex_to_index ? temp_vertices_buffer : save->vertex_store->buffer_in_ram,
809 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
811 if (vertex_to_index) {
812 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
813 free(temp_vertices_buffer);
816 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
817 * primitive (not the indices themselves). */
818 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
819 int indices_offset = save->current_bo_bytes_used / 4;
820 for (int i = 0; i < merged_prim_count; i++) {
821 merged_prims[i].start += indices_offset;
824 /* Then upload the indices. */
825 if (node->cold->ib.obj) {
826 ctx->Driver.BufferSubData(ctx,
827 save->current_bo_bytes_used,
828 idx * sizeof(uint32_t),
831 save->current_bo_bytes_used += idx * sizeof(uint32_t);
833 node->cold->vertex_count = 0;
834 node->cold->prim_count = 0;
837 /* Prepare for DrawGallium */
838 memset(&node->cold->info, 0, sizeof(struct pipe_draw_info));
839 /* The other info fields will be updated in vbo_save_playback_vertex_list */
840 node->cold->info.index_size = 4;
841 node->cold->info.instance_count = 1;
842 node->cold->info.index.gl_bo = node->cold->ib.obj;
843 if (merged_prim_count == 1) {
844 node->cold->info.mode = merged_prims[0].mode;
845 node->start_count.start = merged_prims[0].start;
846 node->start_count.count = merged_prims[0].count;
847 node->start_count.index_bias = 0;
850 node->modes = malloc(merged_prim_count * sizeof(unsigned char));
851 node->start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
852 for (unsigned i = 0; i < merged_prim_count; i++) {
853 node->start_counts[i].start = merged_prims[i].start;
854 node->start_counts[i].count = merged_prims[i].count;
855 node->start_counts[i].index_bias = 0;
856 node->modes[i] = merged_prims[i].mode;
859 node->num_draws = merged_prim_count;
860 if (node->num_draws > 1) {
861 bool same_mode = true;
862 for (unsigned i = 1; i < node->num_draws && same_mode; i++) {
863 same_mode = node->modes[i] == node->modes[0];
866 /* All primitives use the same mode, so we can simplify a bit */
867 node->cold->info.mode = node->modes[0];
878 if (!save->current_bo) {
879 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
880 bool success = ctx->Driver.BufferData(ctx,
881 GL_ELEMENT_ARRAY_BUFFER_ARB,
882 VBO_SAVE_BUFFER_SIZE,
884 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
885 MESA_GALLIUM_VERTEX_STATE_STORAGE,
888 handle_out_of_memory(ctx);
891 GLuint offsets[VBO_ATTRIB_MAX];
892 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
894 offset += save->attrsz[i] * sizeof(GLfloat);
896 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
897 * Note that this may reuse the previous one of possible.
899 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
900 /* create or reuse the vao */
901 update_vao(ctx, vpm, &save->VAO[vpm],
902 save->current_bo, buffer_offset, stride,
903 save->enabled, save->attrsz, save->attrtype, offsets);
904 /* Reference the vao in the dlist */
905 node->cold->VAO[vpm] = NULL;
906 _mesa_reference_vao(ctx, &node->cold->VAO[vpm], save->VAO[vpm]);
909 /* Prepare for DrawGalliumVertexState */
910 if (node->num_draws && ctx->Driver.DrawGalliumVertexState) {
911 for (unsigned i = 0; i < VP_MODE_MAX; i++) {
912 uint32_t enabled_attribs = _vbo_get_vao_filter(i) &
913 node->cold->VAO[i]->_EnabledWithMapMode;
916 ctx->Driver.CreateGalliumVertexState(ctx, node->cold->VAO[i],
919 node->private_refcount[i] = 0;
920 node->enabled_attribs[i] = enabled_attribs;
924 node->mode = node->cold->info.mode;
925 assert(node->cold->info.index_size == 4);
928 /* Deal with GL_COMPILE_AND_EXECUTE:
930 if (ctx->ExecuteFlag) {
931 struct _glapi_table *dispatch = GET_DISPATCH();
933 _glapi_set_dispatch(ctx->Exec);
935 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
936 * use buffer_in_ram instead of current_bo which contains all vertices instead
937 * of the deduplicated vertices only in the !UseLoopback case.
939 * The problem is that the VAO offset is based on current_bo's layout,
940 * so we have to use a temp value.
942 struct gl_vertex_array_object *vao = node->cold->VAO[VP_MODE_SHADER];
943 GLintptr original = vao->BufferBinding[0].Offset;
944 if (!ctx->ListState.Current.UseLoopback) {
945 GLintptr new_offset = 0;
946 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
947 new_offset -= start_offset * stride;
948 vao->BufferBinding[0].Offset = new_offset;
950 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
951 vao->BufferBinding[0].Offset = original;
953 _glapi_set_dispatch(dispatch);
956 /* Reset our structures for the next run of vertices:
963 * This is called when we fill a vertex buffer before we hit a glEnd().
965 * TODO -- If no new vertices have been stored, don't bother saving it.
968 wrap_buffers(struct gl_context *ctx)
970 struct vbo_save_context *save = &vbo_context(ctx)->save;
971 GLint i = save->prim_store->used - 1;
974 assert(i < (GLint) save->prim_store->size);
977 /* Close off in-progress primitive.
979 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
980 mode = save->prim_store->prims[i].mode;
982 /* store the copied vertices, and allocate a new list.
984 compile_vertex_list(ctx);
986 /* Restart interrupted primitive
988 save->prim_store->prims[0].mode = mode;
989 save->prim_store->prims[0].begin = 0;
990 save->prim_store->prims[0].end = 0;
991 save->prim_store->prims[0].start = 0;
992 save->prim_store->prims[0].count = 0;
993 save->prim_store->used = 1;
998 * Called only when buffers are wrapped as the result of filling the
999 * vertex_store struct.
1002 wrap_filled_vertex(struct gl_context *ctx)
1004 struct vbo_save_context *save = &vbo_context(ctx)->save;
1005 unsigned numComponents;
1007 /* Emit a glEnd to close off the last vertex list.
1011 assert(save->vertex_store->used == 0 && save->vertex_store->used == 0);
1013 /* Copy stored stored vertices to start of new list.
1015 numComponents = save->copied.nr * save->vertex_size;
1017 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
1018 if (numComponents) {
1019 assert(save->copied.buffer);
1021 save->copied.buffer,
1022 numComponents * sizeof(fi_type));
1023 free(save->copied.buffer);
1024 save->copied.buffer = NULL;
1026 save->vertex_store->used = numComponents;
1031 copy_to_current(struct gl_context *ctx)
1033 struct vbo_save_context *save = &vbo_context(ctx)->save;
1034 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1037 const int i = u_bit_scan64(&enabled);
1038 assert(save->attrsz[i]);
1040 if (save->attrtype[i] == GL_DOUBLE ||
1041 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1042 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1044 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1045 save->attrptr[i], save->attrtype[i]);
1051 copy_from_current(struct gl_context *ctx)
1053 struct vbo_save_context *save = &vbo_context(ctx)->save;
1054 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1057 const int i = u_bit_scan64(&enabled);
1059 switch (save->attrsz[i]) {
1061 save->attrptr[i][3] = save->current[i][3];
1064 save->attrptr[i][2] = save->current[i][2];
1067 save->attrptr[i][1] = save->current[i][1];
1070 save->attrptr[i][0] = save->current[i][0];
1073 unreachable("Unexpected vertex attribute size");
1080 * Called when we increase the size of a vertex attribute. For example,
1081 * if we've seen one or more glTexCoord2f() calls and now we get a
1082 * glTexCoord3f() call.
1083 * Flush existing data, set new attrib size, replay copied vertices.
1086 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1088 struct vbo_save_context *save = &vbo_context(ctx)->save;
1093 /* Store the current run of vertices, and emit a GL_END. Emit a
1094 * BEGIN in the new buffer.
1096 if (save->vertex_store->used)
1099 assert(save->copied.nr == 0);
1101 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1102 * when the attribute already exists in the vertex and is having
1103 * its size increased.
1105 copy_to_current(ctx);
1109 oldsz = save->attrsz[attr];
1110 save->attrsz[attr] = newsz;
1111 save->enabled |= BITFIELD64_BIT(attr);
1113 save->vertex_size += newsz - oldsz;
1115 /* Recalculate all the attrptr[] values:
1118 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1119 if (save->attrsz[i]) {
1120 save->attrptr[i] = tmp;
1121 tmp += save->attrsz[i];
1124 save->attrptr[i] = NULL; /* will not be dereferenced. */
1128 /* Copy from current to repopulate the vertex with correct values.
1130 copy_from_current(ctx);
1132 /* Replay stored vertices to translate them to new format here.
1134 * If there are copied vertices and the new (upgraded) attribute
1135 * has not been defined before, this list is somewhat degenerate,
1136 * and will need fixup at runtime.
1138 if (save->copied.nr) {
1139 assert(save->copied.buffer);
1140 const fi_type *data = save->copied.buffer;
1141 grow_vertex_storage(ctx, save->copied.nr);
1142 fi_type *dest = save->vertex_store->buffer_in_ram;
1144 /* Need to note this and fix up at runtime (or loopback):
1146 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1148 save->dangling_attr_ref = GL_TRUE;
1151 for (i = 0; i < save->copied.nr; i++) {
1152 GLbitfield64 enabled = save->enabled;
1154 const int j = u_bit_scan64(&enabled);
1155 assert(save->attrsz[j]);
1158 const fi_type *src = oldsz ? data : save->current[attr];
1159 int copy = oldsz ? oldsz : newsz;
1160 for (k = 0; k < copy; k++)
1162 for (; k < newsz; k++) {
1163 switch (save->attrtype[j]) {
1165 dest[k] = FLOAT_AS_UNION(k == 3);
1168 dest[k] = INT_AS_UNION(k == 3);
1170 case GL_UNSIGNED_INT:
1171 dest[k] = UINT_AS_UNION(k == 3);
1174 dest[k] = FLOAT_AS_UNION(k == 3);
1175 assert(!"Unexpected type in upgrade_vertex");
1182 GLint sz = save->attrsz[j];
1183 for (int k = 0; k < sz; k++)
1191 save->vertex_store->used += save->vertex_size * save->copied.nr;
1192 free(save->copied.buffer);
1193 save->copied.buffer = NULL;
1199 * This is called when the size of a vertex attribute changes.
1200 * For example, after seeing one or more glTexCoord2f() calls we
1201 * get a glTexCoord4f() or glTexCoord1f() call.
1204 fixup_vertex(struct gl_context *ctx, GLuint attr,
1205 GLuint sz, GLenum newType)
1207 struct vbo_save_context *save = &vbo_context(ctx)->save;
1209 if (sz > save->attrsz[attr] ||
1210 newType != save->attrtype[attr]) {
1211 /* New size is larger. Need to flush existing vertices and get
1212 * an enlarged vertex format.
1214 upgrade_vertex(ctx, attr, sz);
1216 else if (sz < save->active_sz[attr]) {
1218 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1220 /* New size is equal or smaller - just need to fill in some
1223 for (i = sz; i <= save->attrsz[attr]; i++)
1224 save->attrptr[attr][i - 1] = id[i - 1];
1227 save->active_sz[attr] = sz;
1229 grow_vertex_storage(ctx, 1);
1234 * Reset the current size of all vertex attributes to the default
1235 * value of 0. This signals that we haven't yet seen any per-vertex
1236 * commands such as glNormal3f() or glTexCoord2f().
1239 reset_vertex(struct gl_context *ctx)
1241 struct vbo_save_context *save = &vbo_context(ctx)->save;
1243 while (save->enabled) {
1244 const int i = u_bit_scan64(&save->enabled);
1245 assert(save->attrsz[i]);
1246 save->attrsz[i] = 0;
1247 save->active_sz[i] = 0;
1250 save->vertex_size = 0;
1255 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1256 * It depends on a few things, including whether we're inside or outside
1260 is_vertex_position(const struct gl_context *ctx, GLuint index)
1262 return (index == 0 &&
1263 _mesa_attr_zero_aliases_vertex(ctx) &&
1264 _mesa_inside_dlist_begin_end(ctx));
1269 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1272 /* Only one size for each attribute may be active at once. Eg. if
1273 * Color3f is installed/active, then Color4f may not be, even if the
1274 * vertex actually contains 4 color coordinates. This is because the
1275 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1276 * of the chooser function when switching between Color4f and Color3f.
1278 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1280 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1281 int sz = (sizeof(C) / sizeof(GLfloat)); \
1283 if (save->active_sz[A] != N) \
1284 fixup_vertex(ctx, A, N * sz, T); \
1287 C *dest = (C *)save->attrptr[A]; \
1288 if (N>0) dest[0] = V0; \
1289 if (N>1) dest[1] = V1; \
1290 if (N>2) dest[2] = V2; \
1291 if (N>3) dest[3] = V3; \
1292 save->attrtype[A] = T; \
1295 if ((A) == VBO_ATTRIB_POS) { \
1296 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + \
1297 save->vertex_store->used; \
1299 for (int i = 0; i < save->vertex_size; i++) \
1300 buffer_ptr[i] = save->vertex[i]; \
1302 save->vertex_store->used += save->vertex_size; \
1303 unsigned used_next = (save->vertex_store->used + \
1304 save->vertex_size) * sizeof(float); \
1305 if (used_next > save->vertex_store->buffer_in_ram_size) { \
1306 grow_vertex_storage(ctx, get_vertex_count(save)); \
1307 assert(used_next <= \
1308 save->vertex_store->buffer_in_ram_size); \
1313 #define TAG(x) _save_##x
1315 #include "vbo_attrib_tmp.h"
1318 #define MAT( ATTR, N, face, params ) \
1320 if (face != GL_BACK) \
1321 MAT_ATTR( ATTR, N, params ); /* front */ \
1322 if (face != GL_FRONT) \
1323 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1328 * Save a glMaterial call found between glBegin/End.
1329 * glMaterial calls outside Begin/End are handled in dlist.c.
1331 static void GLAPIENTRY
1332 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1334 GET_CURRENT_CONTEXT(ctx);
1336 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1337 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1343 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1346 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1349 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1352 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1355 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1356 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1359 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1362 case GL_COLOR_INDEXES:
1363 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1365 case GL_AMBIENT_AND_DIFFUSE:
1366 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1367 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1370 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1376 /* Cope with EvalCoord/CallList called within a begin/end object:
1377 * -- Flush current buffer
1378 * -- Fallback to opcodes for the rest of the begin/end object.
1381 dlist_fallback(struct gl_context *ctx)
1383 struct vbo_save_context *save = &vbo_context(ctx)->save;
1385 if (save->vertex_store->used || save->prim_store->used) {
1386 if (save->prim_store->used > 0 && save->vertex_store->used > 0) {
1387 assert(save->vertex_size);
1388 /* Close off in-progress primitive. */
1389 GLint i = save->prim_store->used - 1;
1390 save->prim_store->prims[i].count =
1391 get_vertex_count(save) -
1392 save->prim_store->prims[i].start;
1395 /* Need to replay this display list with loopback,
1396 * unfortunately, otherwise this primitive won't be handled
1399 save->dangling_attr_ref = GL_TRUE;
1401 compile_vertex_list(ctx);
1404 copy_to_current(ctx);
1406 if (save->out_of_memory) {
1407 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1410 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1412 ctx->Driver.SaveNeedFlush = GL_FALSE;
1416 static void GLAPIENTRY
1417 _save_EvalCoord1f(GLfloat u)
1419 GET_CURRENT_CONTEXT(ctx);
1420 dlist_fallback(ctx);
1421 CALL_EvalCoord1f(ctx->Save, (u));
1424 static void GLAPIENTRY
1425 _save_EvalCoord1fv(const GLfloat * v)
1427 GET_CURRENT_CONTEXT(ctx);
1428 dlist_fallback(ctx);
1429 CALL_EvalCoord1fv(ctx->Save, (v));
1432 static void GLAPIENTRY
1433 _save_EvalCoord2f(GLfloat u, GLfloat v)
1435 GET_CURRENT_CONTEXT(ctx);
1436 dlist_fallback(ctx);
1437 CALL_EvalCoord2f(ctx->Save, (u, v));
1440 static void GLAPIENTRY
1441 _save_EvalCoord2fv(const GLfloat * v)
1443 GET_CURRENT_CONTEXT(ctx);
1444 dlist_fallback(ctx);
1445 CALL_EvalCoord2fv(ctx->Save, (v));
1448 static void GLAPIENTRY
1449 _save_EvalPoint1(GLint i)
1451 GET_CURRENT_CONTEXT(ctx);
1452 dlist_fallback(ctx);
1453 CALL_EvalPoint1(ctx->Save, (i));
1456 static void GLAPIENTRY
1457 _save_EvalPoint2(GLint i, GLint j)
1459 GET_CURRENT_CONTEXT(ctx);
1460 dlist_fallback(ctx);
1461 CALL_EvalPoint2(ctx->Save, (i, j));
1464 static void GLAPIENTRY
1465 _save_CallList(GLuint l)
1467 GET_CURRENT_CONTEXT(ctx);
1468 dlist_fallback(ctx);
1469 CALL_CallList(ctx->Save, (l));
1472 static void GLAPIENTRY
1473 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1475 GET_CURRENT_CONTEXT(ctx);
1476 dlist_fallback(ctx);
1477 CALL_CallLists(ctx->Save, (n, type, v));
1483 * Called when a glBegin is getting compiled into a display list.
1484 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1487 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1488 bool no_current_update)
1490 struct vbo_save_context *save = &vbo_context(ctx)->save;
1491 const GLuint i = save->prim_store->used++;
1493 ctx->Driver.CurrentSavePrimitive = mode;
1495 if (!save->prim_store || i >= save->prim_store->size) {
1496 save->prim_store = realloc_prim_store(save->prim_store, i * 2);
1498 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1499 save->prim_store->prims[i].begin = 1;
1500 save->prim_store->prims[i].end = 0;
1501 save->prim_store->prims[i].start = get_vertex_count(save);
1502 save->prim_store->prims[i].count = 0;
1504 save->no_current_update = no_current_update;
1506 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1508 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1509 ctx->Driver.SaveNeedFlush = GL_TRUE;
1513 static void GLAPIENTRY
1516 GET_CURRENT_CONTEXT(ctx);
1517 struct vbo_save_context *save = &vbo_context(ctx)->save;
1518 const GLint i = save->prim_store->used - 1;
1520 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1521 save->prim_store->prims[i].end = 1;
1522 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1524 /* Swap out this vertex format while outside begin/end. Any color,
1525 * etc. received between here and the next begin will be compiled
1528 if (save->out_of_memory) {
1529 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1532 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1537 static void GLAPIENTRY
1538 _save_Begin(GLenum mode)
1540 GET_CURRENT_CONTEXT(ctx);
1542 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1546 static void GLAPIENTRY
1547 _save_PrimitiveRestartNV(void)
1549 GET_CURRENT_CONTEXT(ctx);
1550 struct vbo_save_context *save = &vbo_context(ctx)->save;
1552 if (save->prim_store->used == 0) {
1553 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1556 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1557 "glPrimitiveRestartNV called outside glBegin/End");
1559 /* get current primitive mode */
1560 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1561 bool no_current_update = save->no_current_update;
1563 /* restart primitive */
1564 CALL_End(ctx->CurrentServerDispatch, ());
1565 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1570 /* Unlike the functions above, these are to be hooked into the vtxfmt
1571 * maintained in ctx->ListState, active when the list is known or
1572 * suspected to be outside any begin/end primitive.
1573 * Note: OBE = Outside Begin/End
1575 static void GLAPIENTRY
1576 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1578 GET_CURRENT_CONTEXT(ctx);
1579 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1581 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1582 CALL_Vertex2f(dispatch, (x1, y1));
1583 CALL_Vertex2f(dispatch, (x2, y1));
1584 CALL_Vertex2f(dispatch, (x2, y2));
1585 CALL_Vertex2f(dispatch, (x1, y2));
1586 CALL_End(dispatch, ());
1590 static void GLAPIENTRY
1591 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1593 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1596 static void GLAPIENTRY
1597 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1599 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1602 static void GLAPIENTRY
1603 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1605 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1608 static void GLAPIENTRY
1609 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1611 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1614 static void GLAPIENTRY
1615 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1617 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1620 static void GLAPIENTRY
1621 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1623 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1626 static void GLAPIENTRY
1627 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1629 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1632 static void GLAPIENTRY
1633 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1635 GET_CURRENT_CONTEXT(ctx);
1636 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1637 struct vbo_save_context *save = &vbo_context(ctx)->save;
1640 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1641 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1645 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1649 if (save->out_of_memory)
1652 grow_vertex_storage(ctx, count);
1654 /* Make sure to process any VBO binding changes */
1655 _mesa_update_state(ctx);
1657 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1659 vbo_save_NotifyBegin(ctx, mode, true);
1661 for (i = 0; i < count; i++)
1662 _mesa_array_element(ctx, start + i);
1663 CALL_End(ctx->CurrentServerDispatch, ());
1665 _mesa_vao_unmap_arrays(ctx, vao);
1669 static void GLAPIENTRY
1670 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1671 const GLsizei *count, GLsizei primcount)
1673 GET_CURRENT_CONTEXT(ctx);
1676 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1677 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1681 if (primcount < 0) {
1682 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1683 "glMultiDrawArrays(primcount<0)");
1687 unsigned vertcount = 0;
1688 for (i = 0; i < primcount; i++) {
1690 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1691 "glMultiDrawArrays(count[i]<0)");
1694 vertcount += count[i];
1697 grow_vertex_storage(ctx, vertcount);
1699 for (i = 0; i < primcount; i++) {
1701 _save_OBE_DrawArrays(mode, first[i], count[i]);
1708 array_element(struct gl_context *ctx,
1709 GLint basevertex, GLuint elt, unsigned index_size_shift)
1711 /* Section 10.3.5 Primitive Restart:
1713 * When one of the *BaseVertex drawing commands specified in section 10.5
1714 * is used, the primitive restart comparison occurs before the basevertex
1715 * offset is added to the array index.
1717 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1718 * then we call PrimitiveRestartNV and return.
1720 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1721 elt == ctx->Array._RestartIndex[index_size_shift]) {
1722 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1726 _mesa_array_element(ctx, basevertex + elt);
1730 /* Could do better by copying the arrays and element list intact and
1731 * then emitting an indexed prim at runtime.
1733 static void GLAPIENTRY
1734 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1735 const GLvoid * indices, GLint basevertex)
1737 GET_CURRENT_CONTEXT(ctx);
1738 struct vbo_save_context *save = &vbo_context(ctx)->save;
1739 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1740 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1743 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1744 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1748 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1751 if (type != GL_UNSIGNED_BYTE &&
1752 type != GL_UNSIGNED_SHORT &&
1753 type != GL_UNSIGNED_INT) {
1754 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1758 if (save->out_of_memory)
1761 grow_vertex_storage(ctx, count);
1763 /* Make sure to process any VBO binding changes */
1764 _mesa_update_state(ctx);
1766 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1770 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1772 vbo_save_NotifyBegin(ctx, mode, true);
1775 case GL_UNSIGNED_BYTE:
1776 for (i = 0; i < count; i++)
1777 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1779 case GL_UNSIGNED_SHORT:
1780 for (i = 0; i < count; i++)
1781 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1783 case GL_UNSIGNED_INT:
1784 for (i = 0; i < count; i++)
1785 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1788 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1792 CALL_End(ctx->CurrentServerDispatch, ());
1794 _mesa_vao_unmap(ctx, vao);
1797 static void GLAPIENTRY
1798 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1799 const GLvoid * indices)
1801 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1805 static void GLAPIENTRY
1806 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1807 GLsizei count, GLenum type,
1808 const GLvoid * indices)
1810 GET_CURRENT_CONTEXT(ctx);
1811 struct vbo_save_context *save = &vbo_context(ctx)->save;
1813 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1814 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1818 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1819 "glDrawRangeElements(count<0)");
1822 if (type != GL_UNSIGNED_BYTE &&
1823 type != GL_UNSIGNED_SHORT &&
1824 type != GL_UNSIGNED_INT) {
1825 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1829 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1830 "glDrawRangeElements(end < start)");
1834 if (save->out_of_memory)
1837 _save_OBE_DrawElements(mode, count, type, indices);
1841 static void GLAPIENTRY
1842 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1843 const GLvoid * const *indices, GLsizei primcount)
1845 GET_CURRENT_CONTEXT(ctx);
1846 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1850 for (i = 0; i < primcount; i++) {
1851 vertcount += count[i];
1853 grow_vertex_storage(ctx, vertcount);
1855 for (i = 0; i < primcount; i++) {
1857 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1863 static void GLAPIENTRY
1864 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1866 const GLvoid * const *indices,
1868 const GLint *basevertex)
1870 GET_CURRENT_CONTEXT(ctx);
1871 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1875 for (i = 0; i < primcount; i++) {
1876 vertcount += count[i];
1878 grow_vertex_storage(ctx, vertcount);
1880 for (i = 0; i < primcount; i++) {
1882 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1891 vtxfmt_init(struct gl_context *ctx)
1893 struct vbo_save_context *save = &vbo_context(ctx)->save;
1894 GLvertexformat *vfmt = &save->vtxfmt;
1896 #define NAME_AE(x) _ae_##x
1897 #define NAME_CALLLIST(x) _save_##x
1898 #define NAME(x) _save_##x
1899 #define NAME_ES(x) _save_##x##ARB
1901 #include "vbo_init_tmp.h"
1906 * Initialize the dispatch table with the VBO functions for display
1910 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1911 struct _glapi_table *exec)
1913 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1914 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1915 SET_DrawElements(exec, _save_OBE_DrawElements);
1916 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1917 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1918 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1919 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1920 SET_Rectf(exec, _save_OBE_Rectf);
1921 SET_Rectd(exec, _save_OBE_Rectd);
1922 SET_Rectdv(exec, _save_OBE_Rectdv);
1923 SET_Rectfv(exec, _save_OBE_Rectfv);
1924 SET_Recti(exec, _save_OBE_Recti);
1925 SET_Rectiv(exec, _save_OBE_Rectiv);
1926 SET_Rects(exec, _save_OBE_Rects);
1927 SET_Rectsv(exec, _save_OBE_Rectsv);
1929 /* Note: other glDraw functins aren't compiled into display lists */
1935 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1937 struct vbo_save_context *save = &vbo_context(ctx)->save;
1939 /* Noop when we are actually active:
1941 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1944 if (save->vertex_store->used || save->prim_store->used)
1945 compile_vertex_list(ctx);
1947 copy_to_current(ctx);
1949 ctx->Driver.SaveNeedFlush = GL_FALSE;
1954 * Called from glNewList when we're starting to compile a display list.
1957 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1959 struct vbo_save_context *save = &vbo_context(ctx)->save;
1964 if (!save->prim_store)
1965 save->prim_store = realloc_prim_store(NULL, 8);
1967 if (!save->vertex_store)
1968 save->vertex_store = CALLOC_STRUCT(vbo_save_vertex_store);
1971 ctx->Driver.SaveNeedFlush = GL_FALSE;
1976 * Called from glEndList when we're finished compiling a display list.
1979 vbo_save_EndList(struct gl_context *ctx)
1981 struct vbo_save_context *save = &vbo_context(ctx)->save;
1983 /* EndList called inside a (saved) Begin/End pair?
1985 if (_mesa_inside_dlist_begin_end(ctx)) {
1986 if (save->prim_store->used > 0) {
1987 GLint i = save->prim_store->used - 1;
1988 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1989 save->prim_store->prims[i].end = 0;
1990 save->prim_store->prims[i].count = get_vertex_count(save) - save->prim_store->prims[i].start;
1993 /* Make sure this vertex list gets replayed by the "loopback"
1996 save->dangling_attr_ref = GL_TRUE;
1997 vbo_save_SaveFlushVertices(ctx);
1999 /* Swap out this vertex format while outside begin/end. Any color,
2000 * etc. received between here and the next begin will be compiled
2003 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
2006 assert(save->vertex_size == 0);
2010 * Called during context creation/init.
2013 current_init(struct gl_context *ctx)
2015 struct vbo_save_context *save = &vbo_context(ctx)->save;
2018 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
2019 const GLuint j = i - VBO_ATTRIB_POS;
2020 assert(j < VERT_ATTRIB_MAX);
2021 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
2022 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2025 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2026 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2027 assert(j < MAT_ATTRIB_MAX);
2028 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2029 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2035 * Initialize the display list compiler. Called during context creation.
2038 vbo_save_api_init(struct vbo_save_context *save)
2040 struct gl_context *ctx = gl_context_from_vbo_save(save);