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.
70 #include "main/glheader.h"
71 #include "main/arrayobj.h"
72 #include "main/bufferobj.h"
73 #include "main/context.h"
74 #include "main/dlist.h"
75 #include "main/enums.h"
76 #include "main/eval.h"
77 #include "main/macros.h"
78 #include "main/draw_validate.h"
79 #include "main/api_arrayelt.h"
80 #include "main/vtxfmt.h"
81 #include "main/dispatch.h"
82 #include "main/state.h"
83 #include "main/varray.h"
84 #include "util/bitscan.h"
85 #include "util/u_memory.h"
86 #include "util/hash_table.h"
88 #include "gallium/include/pipe/p_state.h"
91 #include "vbo_private.h"
98 /* An interesting VBO number/name to help with debugging */
99 #define VBO_BUF_ID 12345
101 static void GLAPIENTRY
102 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params);
104 static void GLAPIENTRY
105 _save_EvalCoord1f(GLfloat u);
107 static void GLAPIENTRY
108 _save_EvalCoord2f(GLfloat u, GLfloat v);
111 * NOTE: Old 'parity' issue is gone, but copying can still be
112 * wrong-footed on replay.
115 copy_vertices(struct gl_context *ctx,
116 const struct vbo_save_vertex_list *node,
117 const fi_type * src_buffer)
119 struct vbo_save_context *save = &vbo_context(ctx)->save;
120 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
121 GLuint sz = save->vertex_size;
122 const fi_type *src = src_buffer + prim->start * sz;
123 fi_type *dst = save->copied.buffer;
128 return vbo_copy_vertices(ctx, prim->mode, prim->start, &prim->count,
129 prim->begin, sz, true, dst, src);
133 static struct vbo_save_vertex_store *
134 alloc_vertex_store(struct gl_context *ctx, int vertex_count)
136 struct vbo_save_context *save = &vbo_context(ctx)->save;
137 struct vbo_save_vertex_store *vertex_store =
138 CALLOC_STRUCT(vbo_save_vertex_store);
140 int size = MAX2(vertex_count * save->vertex_size, VBO_SAVE_BUFFER_SIZE);
142 /* obj->Name needs to be non-zero, but won't ever be examined more
143 * closely than that. In particular these buffers won't be entered
144 * into the hash and can never be confused with ones visible to the
145 * user. Perhaps there could be a special number for internal
148 vertex_store->buffer_in_ram_size = size * sizeof(GLfloat);
149 vertex_store->buffer_in_ram = malloc(vertex_store->buffer_in_ram_size);
150 save->out_of_memory = vertex_store->buffer_in_ram == NULL;
152 if (save->out_of_memory) {
153 _mesa_error(ctx, GL_OUT_OF_MEMORY, "internal VBO allocation");
154 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
157 vertex_store->used = 0;
164 free_vertex_store(struct gl_context *ctx,
165 struct vbo_save_vertex_store *vertex_store)
167 free(vertex_store->buffer_in_ram);
172 static struct vbo_save_primitive_store *
173 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
176 store = CALLOC_STRUCT(vbo_save_primitive_store);
177 uint32_t old_size = store->size;
178 store->size = MAX3(store->size, prim_count, VBO_SAVE_PRIM_SIZE);
179 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
180 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
187 reset_counters(struct gl_context *ctx)
189 struct vbo_save_context *save = &vbo_context(ctx)->save;
191 save->buffer_map = save->vertex_store->buffer_in_ram + save->vertex_store->used;
193 assert(save->buffer_map == save->buffer_ptr);
195 if (save->vertex_size)
196 save->max_vert = (save->vertex_store->buffer_in_ram_size / sizeof(float) - save->vertex_store->used) /
201 save->prim_store->used = 0;
202 save->vert_count = 0;
203 save->dangling_attr_ref = GL_FALSE;
207 * For a list of prims, try merging prims that can just be extensions of the
211 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
215 struct _mesa_prim *prev_prim = prim_list;
217 for (i = 1; i < *prim_count; i++) {
218 struct _mesa_prim *this_prim = prim_list + i;
220 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
222 if (vbo_merge_draws(ctx, true,
223 prev_prim->mode, this_prim->mode,
224 prev_prim->start, this_prim->start,
225 &prev_prim->count, this_prim->count,
226 prev_prim->basevertex, this_prim->basevertex,
228 this_prim->begin, this_prim->end)) {
229 /* We've found a prim that just extend the previous one. Tack it
230 * onto the previous one, and let this primitive struct get dropped.
235 /* If any previous primitives have been dropped, then we need to copy
236 * this later one into the next available slot.
239 if (prev_prim != this_prim)
240 *prev_prim = *this_prim;
243 *prim_count = prev_prim - prim_list + 1;
248 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
249 * don't have to worry about handling the _mesa_prim::begin/end flags.
250 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
253 convert_line_loop_to_strip(struct vbo_save_context *save,
254 struct vbo_save_vertex_list *node)
256 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
258 assert(prim->mode == GL_LINE_LOOP);
261 /* Copy the 0th vertex to end of the buffer and extend the
262 * vertex count by one to finish the line loop.
264 const GLuint sz = save->vertex_size;
266 const fi_type *src = save->buffer_map + prim->start * sz;
268 fi_type *dst = save->buffer_map + (prim->start + prim->count) * sz;
270 memcpy(dst, src, sz * sizeof(float));
273 node->cold->vertex_count++;
275 save->buffer_ptr += sz;
276 save->vertex_store->used += sz;
280 /* Drawing the second or later section of a long line loop.
281 * Skip the 0th vertex.
287 prim->mode = GL_LINE_STRIP;
291 /* Compare the present vao if it has the same setup. */
293 compare_vao(gl_vertex_processing_mode mode,
294 const struct gl_vertex_array_object *vao,
295 const struct gl_buffer_object *bo, GLintptr buffer_offset,
296 GLuint stride, GLbitfield64 vao_enabled,
297 const GLubyte size[VBO_ATTRIB_MAX],
298 const GLenum16 type[VBO_ATTRIB_MAX],
299 const GLuint offset[VBO_ATTRIB_MAX])
304 /* If the enabled arrays are not the same we are not equal. */
305 if (vao_enabled != vao->Enabled)
308 /* Check the buffer binding at 0 */
309 if (vao->BufferBinding[0].BufferObj != bo)
311 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
312 if (vao->BufferBinding[0].Stride != stride)
314 assert(vao->BufferBinding[0].InstanceDivisor == 0);
316 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
317 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
319 /* Now check the enabled arrays */
320 GLbitfield mask = vao_enabled;
322 const int attr = u_bit_scan(&mask);
323 const unsigned char vbo_attr = vao_to_vbo_map[attr];
324 const GLenum16 tp = type[vbo_attr];
325 const GLintptr off = offset[vbo_attr] + buffer_offset;
326 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
327 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
329 if (attrib->Format.Type != tp)
331 if (attrib->Format.Size != size[vbo_attr])
333 assert(attrib->Format.Format == GL_RGBA);
334 assert(attrib->Format.Normalized == GL_FALSE);
335 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
336 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
337 assert(attrib->BufferBindingIndex == 0);
344 /* Create or reuse the vao for the vertex processing mode. */
346 update_vao(struct gl_context *ctx,
347 gl_vertex_processing_mode mode,
348 struct gl_vertex_array_object **vao,
349 struct gl_buffer_object *bo, GLintptr buffer_offset,
350 GLuint stride, GLbitfield64 vbo_enabled,
351 const GLubyte size[VBO_ATTRIB_MAX],
352 const GLenum16 type[VBO_ATTRIB_MAX],
353 const GLuint offset[VBO_ATTRIB_MAX])
355 /* Compute the bitmasks of vao_enabled arrays */
356 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
359 * Check if we can possibly reuse the exisiting one.
360 * In the long term we should reset them when something changes.
362 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
363 vao_enabled, size, type, offset))
366 /* The initial refcount is 1 */
367 _mesa_reference_vao(ctx, vao, NULL);
368 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
371 * assert(stride <= ctx->Const.MaxVertexAttribStride);
372 * MaxVertexAttribStride is not set for drivers that does not
373 * expose GL 44 or GLES 31.
376 /* Bind the buffer object at binding point 0 */
377 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
380 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
381 * Note that the position/generic0 aliasing is done in the VAO.
383 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
384 /* Now set the enable arrays */
385 GLbitfield mask = vao_enabled;
387 const int vao_attr = u_bit_scan(&mask);
388 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
389 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
391 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
392 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
393 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
395 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
396 assert(vao_enabled == (*vao)->Enabled);
397 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
399 /* Finalize and freeze the VAO */
400 _mesa_set_vao_immutable(ctx, *vao);
405 realloc_storage(struct gl_context *ctx, int prim_count, int vertex_count)
407 struct vbo_save_context *save = &vbo_context(ctx)->save;
408 if (vertex_count >= 0) {
409 /* Release old reference:
411 free_vertex_store(ctx, save->vertex_store);
412 save->vertex_store = NULL;
413 /* When we have a new vbo, we will for sure need a new vao */
414 for (gl_vertex_processing_mode vpm = 0; vpm < VP_MODE_MAX; ++vpm)
415 _mesa_reference_vao(ctx, &save->VAO[vpm], NULL);
417 /* Allocate and map new store:
419 save->vertex_store = alloc_vertex_store(ctx, vertex_count);
420 save->buffer_ptr = save->vertex_store->buffer_in_ram + save->vertex_store->used;
421 save->out_of_memory = save->buffer_ptr == NULL;
425 save->prim_store = realloc_prim_store(save->prim_store, prim_count);
429 unsigned vertex_size;
430 fi_type *vertex_attributes;
433 static uint32_t _hash_vertex_key(const void *key)
435 struct vertex_key *k = (struct vertex_key*)key;
436 unsigned sz = k->vertex_size;
438 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
441 static bool _compare_vertex_key(const void *key1, const void *key2)
443 struct vertex_key *k1 = (struct vertex_key*)key1;
444 struct vertex_key *k2 = (struct vertex_key*)key2;
445 /* All the compared vertices are going to be drawn with the same VAO,
446 * so we can compare the attributes. */
447 assert (k1->vertex_size == k2->vertex_size);
448 return memcmp(k1->vertex_attributes,
449 k2->vertex_attributes,
450 k1->vertex_size * sizeof(float)) == 0;
453 static void _free_entry(struct hash_entry *entry)
455 free((void*)entry->key);
458 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
459 * of an existing vertex, return the original index instead.
462 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
463 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
465 /* If vertex deduplication is disabled return the original index. */
469 fi_type *vert = save->buffer_map + save->vertex_size * index;
471 struct vertex_key *key = malloc(sizeof(struct vertex_key));
472 key->vertex_size = save->vertex_size;
473 key->vertex_attributes = vert;
475 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
478 /* We found an existing vertex with the same hash, return its index. */
479 return (uintptr_t) entry->data;
481 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
482 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
483 * starting at index 0.
485 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
486 *max_index = MAX2(n, *max_index);
488 memcpy(&new_buffer[save->vertex_size * n],
490 save->vertex_size * sizeof(fi_type));
492 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
494 /* The index buffer is shared between list compilations, so add the base index to get
503 * Insert the active immediate struct onto the display list currently
507 compile_vertex_list(struct gl_context *ctx)
509 struct vbo_save_context *save = &vbo_context(ctx)->save;
510 struct vbo_save_vertex_list *node;
512 /* Allocate space for this structure in the display list currently
515 node = (struct vbo_save_vertex_list *)
516 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
521 memset(node, 0, sizeof(struct vbo_save_vertex_list));
522 node->cold = calloc(1, sizeof(*node->cold));
524 /* Make sure the pointer is aligned to the size of a pointer */
525 assert((GLintptr) node % sizeof(void *) == 0);
527 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
529 node->cold->vertex_count = save->vert_count;
530 node->cold->wrap_count = save->copied.nr;
531 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
532 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
533 node->cold->ib.obj = NULL;
534 node->cold->prim_count = save->prim_store->used;
536 if (save->no_current_update) {
537 node->cold->current_data = NULL;
540 GLuint current_size = save->vertex_size - save->attrsz[0];
541 node->cold->current_data = NULL;
544 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
545 if (node->cold->current_data) {
546 const char *buffer = (const char *)save->buffer_map;
547 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
548 unsigned vertex_offset = 0;
550 if (node->cold->vertex_count)
551 vertex_offset = (node->cold->vertex_count - 1) * stride;
553 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
554 current_size * sizeof(GLfloat));
556 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
561 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
563 if (save->dangling_attr_ref)
564 ctx->ListState.Current.UseLoopback = true;
566 save->vertex_store->used += save->vertex_size * node->cold->vertex_count;
568 /* Copy duplicated vertices
570 save->copied.nr = copy_vertices(ctx, node, save->buffer_map);
572 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
573 convert_line_loop_to_strip(save, node);
576 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
578 GLintptr buffer_offset = 0;
579 GLuint start_offset = 0;
581 /* Create an index buffer. */
582 node->cold->min_index = node->cold->max_index = 0;
583 if (save->vert_count == 0 || node->cold->prim_count == 0)
586 /* We won't modify node->prims, so use a const alias to avoid unintended
588 const struct _mesa_prim *original_prims = node->cold->prims;
590 int end = original_prims[node->cold->prim_count - 1].start +
591 original_prims[node->cold->prim_count - 1].count;
592 int total_vert_count = end - original_prims[0].start;
594 node->cold->min_index = node->cold->prims[0].start;
595 node->cold->max_index = end - 1;
597 /* Estimate for the worst case: all prims are line strips (the +1 is because
598 * wrap_buffers may call use but the last primitive may not be complete) */
599 int max_indices_count = MAX2(total_vert_count * 2 - (node->cold->prim_count * 2) + 1,
602 int size = max_indices_count * sizeof(uint32_t);
603 uint32_t* indices = (uint32_t*) malloc(size);
604 struct _mesa_prim *merged_prims = NULL;
607 struct hash_table *vertex_to_index = NULL;
608 fi_type *temp_vertices_buffer = NULL;
610 /* The loopback replay code doesn't use the index buffer, so we can't
611 * dedup vertices in this case.
613 if (!ctx->ListState.Current.UseLoopback) {
614 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
615 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
618 uint32_t max_index = 0;
620 int last_valid_prim = -1;
621 /* Construct indices array. */
622 for (unsigned i = 0; i < node->cold->prim_count; i++) {
623 assert(original_prims[i].basevertex == 0);
624 GLubyte mode = original_prims[i].mode;
626 int vertex_count = original_prims[i].count;
631 /* Line strips may get converted to lines */
632 if (mode == GL_LINE_STRIP)
635 /* If 2 consecutive prims use the same mode => merge them. */
636 bool merge_prims = last_valid_prim >= 0 &&
637 mode == merged_prims[last_valid_prim].mode &&
638 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
639 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
642 /* To be able to merge consecutive triangle strips we need to insert
643 * a degenerate triangle.
646 mode == GL_TRIANGLE_STRIP) {
647 /* Insert a degenerate triangle */
648 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
649 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
651 indices[idx] = indices[idx - 1];
652 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
653 temp_vertices_buffer, &max_index);
655 merged_prims[last_valid_prim].count += 2;
658 /* Add another index to preserve winding order */
659 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
660 temp_vertices_buffer, &max_index);
661 merged_prims[last_valid_prim].count++;
667 /* Convert line strips to lines if it'll allow if the previous
668 * prim mode is GL_LINES (so merge_prims is true) or if the next
669 * primitive mode is GL_LINES or GL_LINE_LOOP.
671 if (original_prims[i].mode == GL_LINE_STRIP &&
673 (i < node->cold->prim_count - 1 &&
674 (original_prims[i + 1].mode == GL_LINE_STRIP ||
675 original_prims[i + 1].mode == GL_LINES)))) {
676 for (unsigned j = 0; j < vertex_count; j++) {
677 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
678 temp_vertices_buffer, &max_index);
679 /* Repeat all but the first/last indices. */
680 if (j && j != vertex_count - 1) {
681 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
682 temp_vertices_buffer, &max_index);
686 /* We didn't convert to LINES, so restore the original mode */
687 mode = original_prims[i].mode;
689 for (unsigned j = 0; j < vertex_count; j++) {
690 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
691 temp_vertices_buffer, &max_index);
696 /* Update vertex count. */
697 merged_prims[last_valid_prim].count += idx - start;
699 /* Keep this primitive */
700 last_valid_prim += 1;
701 assert(last_valid_prim <= i);
702 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
703 merged_prims[last_valid_prim] = original_prims[i];
704 merged_prims[last_valid_prim].start = start;
705 merged_prims[last_valid_prim].count = idx - start;
707 merged_prims[last_valid_prim].mode = mode;
710 assert(idx > 0 && idx <= max_indices_count);
712 unsigned merged_prim_count = last_valid_prim + 1;
713 node->cold->ib.ptr = NULL;
714 node->cold->ib.count = idx;
715 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
717 /* How many bytes do we need to store the indices and the vertices */
718 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
719 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
720 total_vert_count * save->vertex_size * sizeof(fi_type);
722 const GLintptr old_offset = save->VAO[0] ?
723 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
724 if (old_offset != save->current_bo_bytes_used && stride > 0) {
725 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
726 while (offset_diff > 0 &&
727 save->current_bo_bytes_used < save->current_bo->Size &&
728 offset_diff % stride != 0) {
729 save->current_bo_bytes_used++;
730 offset_diff = save->current_bo_bytes_used - old_offset;
733 buffer_offset = save->current_bo_bytes_used;
735 /* Can we reuse the previous bo or should we allocate a new one? */
736 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
737 if (total_bytes_needed > available_bytes) {
738 if (save->current_bo)
739 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
740 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
741 bool success = ctx->Driver.BufferData(ctx,
742 GL_ELEMENT_ARRAY_BUFFER_ARB,
743 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t)),
745 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
748 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
749 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
751 save->current_bo_bytes_used = 0;
752 available_bytes = save->current_bo->Size;
756 assert(old_offset <= buffer_offset);
757 const GLintptr offset_diff = buffer_offset - old_offset;
758 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
759 /* The vertex size is an exact multiple of the buffer offset.
760 * This means that we can use zero-based vertex attribute pointers
761 * and specify the start of the primitive with the _mesa_prim::start
762 * field. This results in issuing several draw calls with identical
763 * vertex attribute information. This can result in fewer state
764 * changes in drivers. In particular, the Gallium CSO module will
765 * filter out redundant vertex buffer changes.
767 /* We cannot immediately update the primitives as some methods below
768 * still need the uncorrected start vertices
770 start_offset = offset_diff/stride;
771 assert(old_offset == buffer_offset - offset_diff);
772 buffer_offset = old_offset;
775 /* Correct the primitive starts, we can only do this here as copy_vertices
776 * and convert_line_loop_to_strip above consume the uncorrected starts.
777 * On the other hand the _vbo_loopback_vertex_list call below needs the
778 * primitives to be corrected already.
780 for (unsigned i = 0; i < node->cold->prim_count; i++) {
781 node->cold->prims[i].start += start_offset;
783 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
784 * to apply this transformation to all indices and max_index.
786 for (unsigned i = 0; i < idx; i++)
787 indices[i] += start_offset;
788 max_index += start_offset;
791 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
793 /* Upload the vertices first (see buffer_offset) */
794 ctx->Driver.BufferSubData(ctx,
795 save->current_bo_bytes_used,
796 total_vert_count * save->vertex_size * sizeof(fi_type),
797 vertex_to_index ? temp_vertices_buffer : save->buffer_map,
799 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
801 if (vertex_to_index) {
802 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
803 free(temp_vertices_buffer);
806 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
807 * primitive (not the indices themselves). */
808 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
809 int indices_offset = save->current_bo_bytes_used / 4;
810 for (int i = 0; i < merged_prim_count; i++) {
811 merged_prims[i].start += indices_offset;
814 /* Then upload the indices. */
815 if (node->cold->ib.obj) {
816 ctx->Driver.BufferSubData(ctx,
817 save->current_bo_bytes_used,
818 idx * sizeof(uint32_t),
821 save->current_bo_bytes_used += idx * sizeof(uint32_t);
823 node->cold->vertex_count = 0;
824 node->cold->prim_count = 0;
827 /* Prepare for DrawGallium */
828 memset(&node->merged.info, 0, sizeof(struct pipe_draw_info));
829 /* The other info fields will be updated in vbo_save_playback_vertex_list */
830 node->merged.info.index_size = 4;
831 node->merged.info.instance_count = 1;
832 node->merged.info.index.gl_bo = node->cold->ib.obj;
833 if (merged_prim_count == 1) {
834 node->merged.info.mode = merged_prims[0].mode;
835 node->merged.start_count.start = merged_prims[0].start;
836 node->merged.start_count.count = merged_prims[0].count;
837 node->merged.start_count.index_bias = 0;
838 node->merged.mode = NULL;
840 node->merged.mode = malloc(merged_prim_count * sizeof(unsigned char));
841 node->merged.start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
842 for (unsigned i = 0; i < merged_prim_count; i++) {
843 node->merged.start_counts[i].start = merged_prims[i].start;
844 node->merged.start_counts[i].count = merged_prims[i].count;
845 node->merged.start_counts[i].index_bias = 0;
846 node->merged.mode[i] = merged_prims[i].mode;
849 node->merged.num_draws = merged_prim_count;
850 if (node->merged.num_draws > 1) {
851 bool same_mode = true;
852 for (unsigned i = 1; i < node->merged.num_draws && same_mode; i++) {
853 same_mode = node->merged.mode[i] == node->merged.mode[0];
856 /* All primitives use the same mode, so we can simplify a bit */
857 node->merged.info.mode = node->merged.mode[0];
858 free(node->merged.mode);
859 node->merged.mode = NULL;
868 if (!save->current_bo) {
869 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
870 bool success = ctx->Driver.BufferData(ctx,
871 GL_ELEMENT_ARRAY_BUFFER_ARB,
872 VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t),
874 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
878 GLuint offsets[VBO_ATTRIB_MAX];
879 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
881 offset += save->attrsz[i] * sizeof(GLfloat);
883 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
884 * Note that this may reuse the previous one of possible.
886 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
887 /* create or reuse the vao */
888 update_vao(ctx, vpm, &save->VAO[vpm],
889 save->current_bo, buffer_offset, stride,
890 save->enabled, save->attrsz, save->attrtype, offsets);
891 /* Reference the vao in the dlist */
892 node->VAO[vpm] = NULL;
893 _mesa_reference_vao(ctx, &node->VAO[vpm], save->VAO[vpm]);
897 /* Deal with GL_COMPILE_AND_EXECUTE:
899 if (ctx->ExecuteFlag) {
900 struct _glapi_table *dispatch = GET_DISPATCH();
902 _glapi_set_dispatch(ctx->Exec);
904 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
905 * use buffer_in_ram instead of current_bo which contains all vertices instead
906 * of the deduplicated vertices only in the !UseLoopback case.
908 * The problem is that the VAO offset is based on current_bo's layout,
909 * so we have to use a temp value.
911 struct gl_vertex_array_object *vao = node->VAO[VP_MODE_SHADER];
912 GLintptr original = vao->BufferBinding[0].Offset;
913 if (!ctx->ListState.Current.UseLoopback) {
914 GLintptr new_offset = (save->buffer_map - save->vertex_store->buffer_in_ram) *
916 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
917 new_offset -= start_offset * stride;
918 vao->BufferBinding[0].Offset = new_offset;
920 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
921 vao->BufferBinding[0].Offset = original;
923 _glapi_set_dispatch(dispatch);
926 /* Decide whether the storage structs are full, or can be used for
927 * the next vertex lists as well.
929 if (save->vertex_store->used >
930 save->vertex_store->buffer_in_ram_size / sizeof(float) - 16 * (save->vertex_size + 4)) {
931 realloc_storage(ctx, -1, 0);
934 /* update buffer_ptr for next vertex */
935 save->buffer_ptr = save->vertex_store->buffer_in_ram
936 + save->vertex_store->used;
939 /* Reset our structures for the next run of vertices:
946 * This is called when we fill a vertex buffer before we hit a glEnd().
948 * TODO -- If no new vertices have been stored, don't bother saving it.
951 wrap_buffers(struct gl_context *ctx)
953 struct vbo_save_context *save = &vbo_context(ctx)->save;
954 GLint i = save->prim_store->used - 1;
957 assert(i < (GLint) save->prim_store->size);
960 /* Close off in-progress primitive.
962 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
963 mode = save->prim_store->prims[i].mode;
965 /* store the copied vertices, and allocate a new list.
967 compile_vertex_list(ctx);
969 /* Restart interrupted primitive
971 save->prim_store->prims[0].mode = mode;
972 save->prim_store->prims[0].begin = 0;
973 save->prim_store->prims[0].end = 0;
974 save->prim_store->prims[0].start = 0;
975 save->prim_store->prims[0].count = 0;
976 save->prim_store->used = 1;
981 * Called only when buffers are wrapped as the result of filling the
982 * vertex_store struct.
985 wrap_filled_vertex(struct gl_context *ctx)
987 struct vbo_save_context *save = &vbo_context(ctx)->save;
988 unsigned numComponents;
990 /* Emit a glEnd to close off the last vertex list.
994 /* Copy stored stored vertices to start of new list.
996 assert(save->max_vert - save->vert_count > save->copied.nr);
998 numComponents = save->copied.nr * save->vertex_size;
999 memcpy(save->buffer_ptr,
1000 save->copied.buffer,
1001 numComponents * sizeof(fi_type));
1002 save->buffer_ptr += numComponents;
1003 save->vert_count += save->copied.nr;
1008 copy_to_current(struct gl_context *ctx)
1010 struct vbo_save_context *save = &vbo_context(ctx)->save;
1011 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1014 const int i = u_bit_scan64(&enabled);
1015 assert(save->attrsz[i]);
1017 if (save->attrtype[i] == GL_DOUBLE ||
1018 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1019 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1021 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1022 save->attrptr[i], save->attrtype[i]);
1028 copy_from_current(struct gl_context *ctx)
1030 struct vbo_save_context *save = &vbo_context(ctx)->save;
1031 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1034 const int i = u_bit_scan64(&enabled);
1036 switch (save->attrsz[i]) {
1038 save->attrptr[i][3] = save->current[i][3];
1041 save->attrptr[i][2] = save->current[i][2];
1044 save->attrptr[i][1] = save->current[i][1];
1047 save->attrptr[i][0] = save->current[i][0];
1050 unreachable("Unexpected vertex attribute size");
1057 * Called when we increase the size of a vertex attribute. For example,
1058 * if we've seen one or more glTexCoord2f() calls and now we get a
1059 * glTexCoord3f() call.
1060 * Flush existing data, set new attrib size, replay copied vertices.
1063 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1065 struct vbo_save_context *save = &vbo_context(ctx)->save;
1070 /* Store the current run of vertices, and emit a GL_END. Emit a
1071 * BEGIN in the new buffer.
1073 if (save->vert_count)
1076 assert(save->copied.nr == 0);
1078 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1079 * when the attribute already exists in the vertex and is having
1080 * its size increased.
1082 copy_to_current(ctx);
1086 oldsz = save->attrsz[attr];
1087 save->attrsz[attr] = newsz;
1088 save->enabled |= BITFIELD64_BIT(attr);
1090 save->vertex_size += newsz - oldsz;
1091 save->max_vert = ((save->vertex_store->buffer_in_ram_size / sizeof(float) -
1092 save->vertex_store->used) /
1094 save->vert_count = 0;
1096 /* Recalculate all the attrptr[] values:
1099 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1100 if (save->attrsz[i]) {
1101 save->attrptr[i] = tmp;
1102 tmp += save->attrsz[i];
1105 save->attrptr[i] = NULL; /* will not be dereferenced. */
1109 /* Copy from current to repopulate the vertex with correct values.
1111 copy_from_current(ctx);
1113 /* Replay stored vertices to translate them to new format here.
1115 * If there are copied vertices and the new (upgraded) attribute
1116 * has not been defined before, this list is somewhat degenerate,
1117 * and will need fixup at runtime.
1119 if (save->copied.nr) {
1120 const fi_type *data = save->copied.buffer;
1121 fi_type *dest = save->buffer_map;
1123 /* Need to note this and fix up at runtime (or loopback):
1125 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1127 save->dangling_attr_ref = GL_TRUE;
1130 for (i = 0; i < save->copied.nr; i++) {
1131 GLbitfield64 enabled = save->enabled;
1133 const int j = u_bit_scan64(&enabled);
1134 assert(save->attrsz[j]);
1137 COPY_CLEAN_4V_TYPE_AS_UNION(dest, oldsz, data,
1143 COPY_SZ_4V(dest, newsz, save->current[attr]);
1148 GLint sz = save->attrsz[j];
1149 COPY_SZ_4V(dest, sz, data);
1156 save->buffer_ptr = dest;
1157 save->vert_count += save->copied.nr;
1163 * This is called when the size of a vertex attribute changes.
1164 * For example, after seeing one or more glTexCoord2f() calls we
1165 * get a glTexCoord4f() or glTexCoord1f() call.
1168 fixup_vertex(struct gl_context *ctx, GLuint attr,
1169 GLuint sz, GLenum newType)
1171 struct vbo_save_context *save = &vbo_context(ctx)->save;
1173 if (sz > save->attrsz[attr] ||
1174 newType != save->attrtype[attr]) {
1175 /* New size is larger. Need to flush existing vertices and get
1176 * an enlarged vertex format.
1178 upgrade_vertex(ctx, attr, sz);
1180 else if (sz < save->active_sz[attr]) {
1182 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1184 /* New size is equal or smaller - just need to fill in some
1187 for (i = sz; i <= save->attrsz[attr]; i++)
1188 save->attrptr[attr][i - 1] = id[i - 1];
1191 save->active_sz[attr] = sz;
1196 * Reset the current size of all vertex attributes to the default
1197 * value of 0. This signals that we haven't yet seen any per-vertex
1198 * commands such as glNormal3f() or glTexCoord2f().
1201 reset_vertex(struct gl_context *ctx)
1203 struct vbo_save_context *save = &vbo_context(ctx)->save;
1205 while (save->enabled) {
1206 const int i = u_bit_scan64(&save->enabled);
1207 assert(save->attrsz[i]);
1208 save->attrsz[i] = 0;
1209 save->active_sz[i] = 0;
1212 save->vertex_size = 0;
1217 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1218 * It depends on a few things, including whether we're inside or outside
1222 is_vertex_position(const struct gl_context *ctx, GLuint index)
1224 return (index == 0 &&
1225 _mesa_attr_zero_aliases_vertex(ctx) &&
1226 _mesa_inside_dlist_begin_end(ctx));
1231 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1234 /* Only one size for each attribute may be active at once. Eg. if
1235 * Color3f is installed/active, then Color4f may not be, even if the
1236 * vertex actually contains 4 color coordinates. This is because the
1237 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1238 * of the chooser function when switching between Color4f and Color3f.
1240 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1242 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1243 int sz = (sizeof(C) / sizeof(GLfloat)); \
1245 if (save->active_sz[A] != N) \
1246 fixup_vertex(ctx, A, N * sz, T); \
1249 C *dest = (C *)save->attrptr[A]; \
1250 if (N>0) dest[0] = V0; \
1251 if (N>1) dest[1] = V1; \
1252 if (N>2) dest[2] = V2; \
1253 if (N>3) dest[3] = V3; \
1254 save->attrtype[A] = T; \
1260 for (i = 0; i < save->vertex_size; i++) \
1261 save->buffer_ptr[i] = save->vertex[i]; \
1263 save->buffer_ptr += save->vertex_size; \
1265 if (++save->vert_count >= save->max_vert) \
1266 wrap_filled_vertex(ctx); \
1270 #define TAG(x) _save_##x
1272 #include "vbo_attrib_tmp.h"
1276 #define MAT( ATTR, N, face, params ) \
1278 if (face != GL_BACK) \
1279 MAT_ATTR( ATTR, N, params ); /* front */ \
1280 if (face != GL_FRONT) \
1281 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1286 * Save a glMaterial call found between glBegin/End.
1287 * glMaterial calls outside Begin/End are handled in dlist.c.
1289 static void GLAPIENTRY
1290 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1292 GET_CURRENT_CONTEXT(ctx);
1294 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1295 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1301 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1304 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1307 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1310 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1313 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1314 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1317 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1320 case GL_COLOR_INDEXES:
1321 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1323 case GL_AMBIENT_AND_DIFFUSE:
1324 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1325 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1328 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1334 /* Cope with EvalCoord/CallList called within a begin/end object:
1335 * -- Flush current buffer
1336 * -- Fallback to opcodes for the rest of the begin/end object.
1339 dlist_fallback(struct gl_context *ctx)
1341 struct vbo_save_context *save = &vbo_context(ctx)->save;
1343 if (save->vert_count || save->prim_store->used) {
1344 if (save->prim_store->used > 0) {
1345 /* Close off in-progress primitive. */
1346 GLint i = save->prim_store->used - 1;
1347 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1350 /* Need to replay this display list with loopback,
1351 * unfortunately, otherwise this primitive won't be handled
1354 save->dangling_attr_ref = GL_TRUE;
1356 compile_vertex_list(ctx);
1359 copy_to_current(ctx);
1361 reset_counters(ctx);
1362 if (save->out_of_memory) {
1363 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1366 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1368 ctx->Driver.SaveNeedFlush = GL_FALSE;
1372 static void GLAPIENTRY
1373 _save_EvalCoord1f(GLfloat u)
1375 GET_CURRENT_CONTEXT(ctx);
1376 dlist_fallback(ctx);
1377 CALL_EvalCoord1f(ctx->Save, (u));
1380 static void GLAPIENTRY
1381 _save_EvalCoord1fv(const GLfloat * v)
1383 GET_CURRENT_CONTEXT(ctx);
1384 dlist_fallback(ctx);
1385 CALL_EvalCoord1fv(ctx->Save, (v));
1388 static void GLAPIENTRY
1389 _save_EvalCoord2f(GLfloat u, GLfloat v)
1391 GET_CURRENT_CONTEXT(ctx);
1392 dlist_fallback(ctx);
1393 CALL_EvalCoord2f(ctx->Save, (u, v));
1396 static void GLAPIENTRY
1397 _save_EvalCoord2fv(const GLfloat * v)
1399 GET_CURRENT_CONTEXT(ctx);
1400 dlist_fallback(ctx);
1401 CALL_EvalCoord2fv(ctx->Save, (v));
1404 static void GLAPIENTRY
1405 _save_EvalPoint1(GLint i)
1407 GET_CURRENT_CONTEXT(ctx);
1408 dlist_fallback(ctx);
1409 CALL_EvalPoint1(ctx->Save, (i));
1412 static void GLAPIENTRY
1413 _save_EvalPoint2(GLint i, GLint j)
1415 GET_CURRENT_CONTEXT(ctx);
1416 dlist_fallback(ctx);
1417 CALL_EvalPoint2(ctx->Save, (i, j));
1420 static void GLAPIENTRY
1421 _save_CallList(GLuint l)
1423 GET_CURRENT_CONTEXT(ctx);
1424 dlist_fallback(ctx);
1425 CALL_CallList(ctx->Save, (l));
1428 static void GLAPIENTRY
1429 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1431 GET_CURRENT_CONTEXT(ctx);
1432 dlist_fallback(ctx);
1433 CALL_CallLists(ctx->Save, (n, type, v));
1439 * Called when a glBegin is getting compiled into a display list.
1440 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1443 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1444 bool no_current_update)
1446 struct vbo_save_context *save = &vbo_context(ctx)->save;
1447 const GLuint i = save->prim_store->used++;
1449 ctx->Driver.CurrentSavePrimitive = mode;
1451 assert(i < save->prim_store->size);
1452 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1453 save->prim_store->prims[i].begin = 1;
1454 save->prim_store->prims[i].end = 0;
1455 save->prim_store->prims[i].start = save->vert_count;
1456 save->prim_store->prims[i].count = 0;
1458 save->no_current_update = no_current_update;
1460 if (save->out_of_memory) {
1461 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1464 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1467 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1468 ctx->Driver.SaveNeedFlush = GL_TRUE;
1472 static void GLAPIENTRY
1475 GET_CURRENT_CONTEXT(ctx);
1476 struct vbo_save_context *save = &vbo_context(ctx)->save;
1477 const GLint i = save->prim_store->used - 1;
1479 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1480 save->prim_store->prims[i].end = 1;
1481 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
1483 if (i == (GLint) save->prim_store->size - 1) {
1484 compile_vertex_list(ctx);
1485 assert(save->copied.nr == 0);
1488 /* Swap out this vertex format while outside begin/end. Any color,
1489 * etc. received between here and the next begin will be compiled
1492 if (save->out_of_memory) {
1493 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1496 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1501 static void GLAPIENTRY
1502 _save_Begin(GLenum mode)
1504 GET_CURRENT_CONTEXT(ctx);
1506 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1510 static void GLAPIENTRY
1511 _save_PrimitiveRestartNV(void)
1513 GET_CURRENT_CONTEXT(ctx);
1514 struct vbo_save_context *save = &vbo_context(ctx)->save;
1516 if (save->prim_store->used == 0) {
1517 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1520 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1521 "glPrimitiveRestartNV called outside glBegin/End");
1523 /* get current primitive mode */
1524 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1525 bool no_current_update = save->no_current_update;
1527 /* restart primitive */
1528 CALL_End(ctx->CurrentServerDispatch, ());
1529 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1534 /* Unlike the functions above, these are to be hooked into the vtxfmt
1535 * maintained in ctx->ListState, active when the list is known or
1536 * suspected to be outside any begin/end primitive.
1537 * Note: OBE = Outside Begin/End
1539 static void GLAPIENTRY
1540 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1542 GET_CURRENT_CONTEXT(ctx);
1543 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1545 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1546 CALL_Vertex2f(dispatch, (x1, y1));
1547 CALL_Vertex2f(dispatch, (x2, y1));
1548 CALL_Vertex2f(dispatch, (x2, y2));
1549 CALL_Vertex2f(dispatch, (x1, y2));
1550 CALL_End(dispatch, ());
1554 static void GLAPIENTRY
1555 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1557 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1560 static void GLAPIENTRY
1561 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1563 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1566 static void GLAPIENTRY
1567 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1569 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1572 static void GLAPIENTRY
1573 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1575 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1578 static void GLAPIENTRY
1579 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1581 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1584 static void GLAPIENTRY
1585 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1587 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1590 static void GLAPIENTRY
1591 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1593 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1597 _ensure_draws_fits_in_storage(struct gl_context *ctx, int primcount, int vertcount)
1599 struct vbo_save_context *save = &vbo_context(ctx)->save;
1601 bool realloc_prim = save->prim_store->used + primcount > save->prim_store->size;
1602 bool realloc_vert = save->vertex_size && (save->vert_count + vertcount >= save->max_vert);
1604 if (realloc_prim || realloc_vert) {
1605 if (realloc_vert && (save->vert_count || save->prim_store->used)) {
1606 /* TODO: this really isn't needed. We should realloc only the CPU-side memory. */
1607 compile_vertex_list(ctx);
1609 realloc_storage(ctx, realloc_prim ? primcount : -1, realloc_vert ? vertcount : -1);
1610 reset_counters(ctx);
1615 static void GLAPIENTRY
1616 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1618 GET_CURRENT_CONTEXT(ctx);
1619 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1620 struct vbo_save_context *save = &vbo_context(ctx)->save;
1623 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1624 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1628 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1632 if (save->out_of_memory)
1635 _ensure_draws_fits_in_storage(ctx, 1, count);
1637 /* Make sure to process any VBO binding changes */
1638 _mesa_update_state(ctx);
1640 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1642 vbo_save_NotifyBegin(ctx, mode, true);
1644 for (i = 0; i < count; i++)
1645 _mesa_array_element(ctx, start + i);
1646 CALL_End(ctx->CurrentServerDispatch, ());
1648 _mesa_vao_unmap_arrays(ctx, vao);
1652 static void GLAPIENTRY
1653 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1654 const GLsizei *count, GLsizei primcount)
1656 GET_CURRENT_CONTEXT(ctx);
1659 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1660 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1664 if (primcount < 0) {
1665 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1666 "glMultiDrawArrays(primcount<0)");
1670 unsigned vertcount = 0;
1671 for (i = 0; i < primcount; i++) {
1673 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1674 "glMultiDrawArrays(count[i]<0)");
1677 vertcount += count[i];
1680 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1682 for (i = 0; i < primcount; i++) {
1684 _save_OBE_DrawArrays(mode, first[i], count[i]);
1691 array_element(struct gl_context *ctx,
1692 GLint basevertex, GLuint elt, unsigned index_size_shift)
1694 /* Section 10.3.5 Primitive Restart:
1696 * When one of the *BaseVertex drawing commands specified in section 10.5
1697 * is used, the primitive restart comparison occurs before the basevertex
1698 * offset is added to the array index.
1700 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1701 * then we call PrimitiveRestartNV and return.
1703 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1704 elt == ctx->Array._RestartIndex[index_size_shift]) {
1705 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1709 _mesa_array_element(ctx, basevertex + elt);
1713 /* Could do better by copying the arrays and element list intact and
1714 * then emitting an indexed prim at runtime.
1716 static void GLAPIENTRY
1717 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1718 const GLvoid * indices, GLint basevertex)
1720 GET_CURRENT_CONTEXT(ctx);
1721 struct vbo_save_context *save = &vbo_context(ctx)->save;
1722 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1723 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1726 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1727 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1731 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1734 if (type != GL_UNSIGNED_BYTE &&
1735 type != GL_UNSIGNED_SHORT &&
1736 type != GL_UNSIGNED_INT) {
1737 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1741 if (save->out_of_memory)
1744 _ensure_draws_fits_in_storage(ctx, 1, count);
1746 /* Make sure to process any VBO binding changes */
1747 _mesa_update_state(ctx);
1749 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1753 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1755 vbo_save_NotifyBegin(ctx, mode, true);
1758 case GL_UNSIGNED_BYTE:
1759 for (i = 0; i < count; i++)
1760 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1762 case GL_UNSIGNED_SHORT:
1763 for (i = 0; i < count; i++)
1764 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1766 case GL_UNSIGNED_INT:
1767 for (i = 0; i < count; i++)
1768 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1771 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1775 CALL_End(ctx->CurrentServerDispatch, ());
1777 _mesa_vao_unmap(ctx, vao);
1780 static void GLAPIENTRY
1781 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1782 const GLvoid * indices)
1784 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1788 static void GLAPIENTRY
1789 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1790 GLsizei count, GLenum type,
1791 const GLvoid * indices)
1793 GET_CURRENT_CONTEXT(ctx);
1794 struct vbo_save_context *save = &vbo_context(ctx)->save;
1796 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1797 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1801 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1802 "glDrawRangeElements(count<0)");
1805 if (type != GL_UNSIGNED_BYTE &&
1806 type != GL_UNSIGNED_SHORT &&
1807 type != GL_UNSIGNED_INT) {
1808 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1812 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1813 "glDrawRangeElements(end < start)");
1817 if (save->out_of_memory)
1820 _save_OBE_DrawElements(mode, count, type, indices);
1824 static void GLAPIENTRY
1825 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1826 const GLvoid * const *indices, GLsizei primcount)
1828 GET_CURRENT_CONTEXT(ctx);
1829 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1833 for (i = 0; i < primcount; i++) {
1834 vertcount += count[i];
1836 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1838 for (i = 0; i < primcount; i++) {
1840 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1846 static void GLAPIENTRY
1847 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1849 const GLvoid * const *indices,
1851 const GLint *basevertex)
1853 GET_CURRENT_CONTEXT(ctx);
1854 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1858 for (i = 0; i < primcount; i++) {
1859 vertcount += count[i];
1861 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1863 for (i = 0; i < primcount; i++) {
1865 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1874 vtxfmt_init(struct gl_context *ctx)
1876 struct vbo_save_context *save = &vbo_context(ctx)->save;
1877 GLvertexformat *vfmt = &save->vtxfmt;
1879 #define NAME_AE(x) _ae_##x
1880 #define NAME_CALLLIST(x) _save_##x
1881 #define NAME(x) _save_##x
1882 #define NAME_ES(x) _save_##x##ARB
1884 #include "vbo_init_tmp.h"
1889 * Initialize the dispatch table with the VBO functions for display
1893 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1894 struct _glapi_table *exec)
1896 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1897 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1898 SET_DrawElements(exec, _save_OBE_DrawElements);
1899 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1900 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1901 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1902 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1903 SET_Rectf(exec, _save_OBE_Rectf);
1904 SET_Rectd(exec, _save_OBE_Rectd);
1905 SET_Rectdv(exec, _save_OBE_Rectdv);
1906 SET_Rectfv(exec, _save_OBE_Rectfv);
1907 SET_Recti(exec, _save_OBE_Recti);
1908 SET_Rectiv(exec, _save_OBE_Rectiv);
1909 SET_Rects(exec, _save_OBE_Rects);
1910 SET_Rectsv(exec, _save_OBE_Rectsv);
1912 /* Note: other glDraw functins aren't compiled into display lists */
1918 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1920 struct vbo_save_context *save = &vbo_context(ctx)->save;
1922 /* Noop when we are actually active:
1924 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1927 if (save->vert_count || save->prim_store->used)
1928 compile_vertex_list(ctx);
1930 copy_to_current(ctx);
1932 reset_counters(ctx);
1933 ctx->Driver.SaveNeedFlush = GL_FALSE;
1938 * Called from glNewList when we're starting to compile a display list.
1941 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1943 struct vbo_save_context *save = &vbo_context(ctx)->save;
1948 if (!save->prim_store)
1949 save->prim_store = realloc_prim_store(NULL, 8);
1951 if (!save->vertex_store)
1952 save->vertex_store = alloc_vertex_store(ctx, 0);
1954 save->buffer_ptr = save->vertex_store->buffer_in_ram + save->vertex_store->used;
1957 reset_counters(ctx);
1958 ctx->Driver.SaveNeedFlush = GL_FALSE;
1963 * Called from glEndList when we're finished compiling a display list.
1966 vbo_save_EndList(struct gl_context *ctx)
1968 struct vbo_save_context *save = &vbo_context(ctx)->save;
1970 /* EndList called inside a (saved) Begin/End pair?
1972 if (_mesa_inside_dlist_begin_end(ctx)) {
1973 if (save->prim_store->used > 0) {
1974 GLint i = save->prim_store->used - 1;
1975 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1976 save->prim_store->prims[i].end = 0;
1977 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1980 /* Make sure this vertex list gets replayed by the "loopback"
1983 save->dangling_attr_ref = GL_TRUE;
1984 vbo_save_SaveFlushVertices(ctx);
1986 /* Swap out this vertex format while outside begin/end. Any color,
1987 * etc. received between here and the next begin will be compiled
1990 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1993 assert(save->vertex_size == 0);
1997 * Called during context creation/init.
2000 current_init(struct gl_context *ctx)
2002 struct vbo_save_context *save = &vbo_context(ctx)->save;
2005 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
2006 const GLuint j = i - VBO_ATTRIB_POS;
2007 assert(j < VERT_ATTRIB_MAX);
2008 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
2009 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2012 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2013 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2014 assert(j < MAT_ATTRIB_MAX);
2015 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2016 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2022 * Initialize the display list compiler. Called during context creation.
2025 vbo_save_api_init(struct vbo_save_context *save)
2027 struct gl_context *ctx = gl_context_from_vbo_save(save);
2031 _mesa_noop_vtxfmt_init(ctx, &save->vtxfmt_noop);