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->vertex_store->used = 0;
193 if (save->vertex_size)
194 save->max_vert = save->vertex_store->buffer_in_ram_size / (sizeof(float) * save->vertex_size);
198 save->prim_store->used = 0;
199 save->vert_count = 0;
200 save->dangling_attr_ref = GL_FALSE;
204 * For a list of prims, try merging prims that can just be extensions of the
208 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
212 struct _mesa_prim *prev_prim = prim_list;
214 for (i = 1; i < *prim_count; i++) {
215 struct _mesa_prim *this_prim = prim_list + i;
217 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
219 if (vbo_merge_draws(ctx, true,
220 prev_prim->mode, this_prim->mode,
221 prev_prim->start, this_prim->start,
222 &prev_prim->count, this_prim->count,
223 prev_prim->basevertex, this_prim->basevertex,
225 this_prim->begin, this_prim->end)) {
226 /* We've found a prim that just extend the previous one. Tack it
227 * onto the previous one, and let this primitive struct get dropped.
232 /* If any previous primitives have been dropped, then we need to copy
233 * this later one into the next available slot.
236 if (prev_prim != this_prim)
237 *prev_prim = *this_prim;
240 *prim_count = prev_prim - prim_list + 1;
245 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
246 * don't have to worry about handling the _mesa_prim::begin/end flags.
247 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
250 convert_line_loop_to_strip(struct vbo_save_context *save,
251 struct vbo_save_vertex_list *node)
253 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
255 assert(prim->mode == GL_LINE_LOOP);
258 /* Copy the 0th vertex to end of the buffer and extend the
259 * vertex count by one to finish the line loop.
261 const GLuint sz = save->vertex_size;
263 const fi_type *src = save->vertex_store->buffer_in_ram + prim->start * sz;
265 fi_type *dst = save->vertex_store->buffer_in_ram + (prim->start + prim->count) * sz;
267 memcpy(dst, src, sz * sizeof(float));
270 node->cold->vertex_count++;
272 save->vertex_store->used += sz;
276 /* Drawing the second or later section of a long line loop.
277 * Skip the 0th vertex.
283 prim->mode = GL_LINE_STRIP;
287 /* Compare the present vao if it has the same setup. */
289 compare_vao(gl_vertex_processing_mode mode,
290 const struct gl_vertex_array_object *vao,
291 const struct gl_buffer_object *bo, GLintptr buffer_offset,
292 GLuint stride, GLbitfield64 vao_enabled,
293 const GLubyte size[VBO_ATTRIB_MAX],
294 const GLenum16 type[VBO_ATTRIB_MAX],
295 const GLuint offset[VBO_ATTRIB_MAX])
300 /* If the enabled arrays are not the same we are not equal. */
301 if (vao_enabled != vao->Enabled)
304 /* Check the buffer binding at 0 */
305 if (vao->BufferBinding[0].BufferObj != bo)
307 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
308 if (vao->BufferBinding[0].Stride != stride)
310 assert(vao->BufferBinding[0].InstanceDivisor == 0);
312 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
313 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
315 /* Now check the enabled arrays */
316 GLbitfield mask = vao_enabled;
318 const int attr = u_bit_scan(&mask);
319 const unsigned char vbo_attr = vao_to_vbo_map[attr];
320 const GLenum16 tp = type[vbo_attr];
321 const GLintptr off = offset[vbo_attr] + buffer_offset;
322 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
323 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
325 if (attrib->Format.Type != tp)
327 if (attrib->Format.Size != size[vbo_attr])
329 assert(attrib->Format.Format == GL_RGBA);
330 assert(attrib->Format.Normalized == GL_FALSE);
331 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
332 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
333 assert(attrib->BufferBindingIndex == 0);
340 /* Create or reuse the vao for the vertex processing mode. */
342 update_vao(struct gl_context *ctx,
343 gl_vertex_processing_mode mode,
344 struct gl_vertex_array_object **vao,
345 struct gl_buffer_object *bo, GLintptr buffer_offset,
346 GLuint stride, GLbitfield64 vbo_enabled,
347 const GLubyte size[VBO_ATTRIB_MAX],
348 const GLenum16 type[VBO_ATTRIB_MAX],
349 const GLuint offset[VBO_ATTRIB_MAX])
351 /* Compute the bitmasks of vao_enabled arrays */
352 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
355 * Check if we can possibly reuse the exisiting one.
356 * In the long term we should reset them when something changes.
358 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
359 vao_enabled, size, type, offset))
362 /* The initial refcount is 1 */
363 _mesa_reference_vao(ctx, vao, NULL);
364 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
367 * assert(stride <= ctx->Const.MaxVertexAttribStride);
368 * MaxVertexAttribStride is not set for drivers that does not
369 * expose GL 44 or GLES 31.
372 /* Bind the buffer object at binding point 0 */
373 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
376 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
377 * Note that the position/generic0 aliasing is done in the VAO.
379 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
380 /* Now set the enable arrays */
381 GLbitfield mask = vao_enabled;
383 const int vao_attr = u_bit_scan(&mask);
384 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
385 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
387 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
388 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
389 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
391 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
392 assert(vao_enabled == (*vao)->Enabled);
393 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
395 /* Finalize and freeze the VAO */
396 _mesa_set_vao_immutable(ctx, *vao);
401 realloc_storage(struct gl_context *ctx, int prim_count, int vertex_count)
403 struct vbo_save_context *save = &vbo_context(ctx)->save;
404 if (vertex_count >= 0) {
405 /* Release old reference:
407 free_vertex_store(ctx, save->vertex_store);
408 save->vertex_store = NULL;
409 /* When we have a new vbo, we will for sure need a new vao */
410 for (gl_vertex_processing_mode vpm = 0; vpm < VP_MODE_MAX; ++vpm)
411 _mesa_reference_vao(ctx, &save->VAO[vpm], NULL);
413 /* Allocate and map new store:
415 save->vertex_store = alloc_vertex_store(ctx, vertex_count);
419 save->prim_store = realloc_prim_store(save->prim_store, prim_count);
423 unsigned vertex_size;
424 fi_type *vertex_attributes;
427 static uint32_t _hash_vertex_key(const void *key)
429 struct vertex_key *k = (struct vertex_key*)key;
430 unsigned sz = k->vertex_size;
432 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
435 static bool _compare_vertex_key(const void *key1, const void *key2)
437 struct vertex_key *k1 = (struct vertex_key*)key1;
438 struct vertex_key *k2 = (struct vertex_key*)key2;
439 /* All the compared vertices are going to be drawn with the same VAO,
440 * so we can compare the attributes. */
441 assert (k1->vertex_size == k2->vertex_size);
442 return memcmp(k1->vertex_attributes,
443 k2->vertex_attributes,
444 k1->vertex_size * sizeof(float)) == 0;
447 static void _free_entry(struct hash_entry *entry)
449 free((void*)entry->key);
452 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
453 * of an existing vertex, return the original index instead.
456 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
457 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
459 /* If vertex deduplication is disabled return the original index. */
463 fi_type *vert = save->vertex_store->buffer_in_ram + save->vertex_size * index;
465 struct vertex_key *key = malloc(sizeof(struct vertex_key));
466 key->vertex_size = save->vertex_size;
467 key->vertex_attributes = vert;
469 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
472 /* We found an existing vertex with the same hash, return its index. */
473 return (uintptr_t) entry->data;
475 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
476 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
477 * starting at index 0.
479 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
480 *max_index = MAX2(n, *max_index);
482 memcpy(&new_buffer[save->vertex_size * n],
484 save->vertex_size * sizeof(fi_type));
486 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
488 /* The index buffer is shared between list compilations, so add the base index to get
497 * Insert the active immediate struct onto the display list currently
501 compile_vertex_list(struct gl_context *ctx)
503 struct vbo_save_context *save = &vbo_context(ctx)->save;
504 struct vbo_save_vertex_list *node;
506 /* Allocate space for this structure in the display list currently
509 node = (struct vbo_save_vertex_list *)
510 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
515 memset(node, 0, sizeof(struct vbo_save_vertex_list));
516 node->cold = calloc(1, sizeof(*node->cold));
518 /* Make sure the pointer is aligned to the size of a pointer */
519 assert((GLintptr) node % sizeof(void *) == 0);
521 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
523 node->cold->vertex_count = save->vert_count;
524 node->cold->wrap_count = save->copied.nr;
525 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
526 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
527 node->cold->ib.obj = NULL;
528 node->cold->prim_count = save->prim_store->used;
530 if (save->no_current_update) {
531 node->cold->current_data = NULL;
534 GLuint current_size = save->vertex_size - save->attrsz[0];
535 node->cold->current_data = NULL;
538 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
539 if (node->cold->current_data) {
540 const char *buffer = (const char *)save->vertex_store->buffer_in_ram;
541 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
542 unsigned vertex_offset = 0;
544 if (node->cold->vertex_count)
545 vertex_offset = (node->cold->vertex_count - 1) * stride;
547 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
548 current_size * sizeof(GLfloat));
550 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
555 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
557 if (save->dangling_attr_ref)
558 ctx->ListState.Current.UseLoopback = true;
560 /* Copy duplicated vertices
562 save->copied.nr = copy_vertices(ctx, node, save->vertex_store->buffer_in_ram);
564 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
565 convert_line_loop_to_strip(save, node);
568 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
570 GLintptr buffer_offset = 0;
571 GLuint start_offset = 0;
573 /* Create an index buffer. */
574 node->cold->min_index = node->cold->max_index = 0;
575 if (save->vert_count == 0 || node->cold->prim_count == 0)
578 /* We won't modify node->prims, so use a const alias to avoid unintended
580 const struct _mesa_prim *original_prims = node->cold->prims;
582 int end = original_prims[node->cold->prim_count - 1].start +
583 original_prims[node->cold->prim_count - 1].count;
584 int total_vert_count = end - original_prims[0].start;
586 node->cold->min_index = node->cold->prims[0].start;
587 node->cold->max_index = end - 1;
589 /* Estimate for the worst case: all prims are line strips (the +1 is because
590 * wrap_buffers may call use but the last primitive may not be complete) */
591 int max_indices_count = MAX2(total_vert_count * 2 - (node->cold->prim_count * 2) + 1,
594 int size = max_indices_count * sizeof(uint32_t);
595 uint32_t* indices = (uint32_t*) malloc(size);
596 struct _mesa_prim *merged_prims = NULL;
599 struct hash_table *vertex_to_index = NULL;
600 fi_type *temp_vertices_buffer = NULL;
602 /* The loopback replay code doesn't use the index buffer, so we can't
603 * dedup vertices in this case.
605 if (!ctx->ListState.Current.UseLoopback) {
606 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
607 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
610 uint32_t max_index = 0;
612 int last_valid_prim = -1;
613 /* Construct indices array. */
614 for (unsigned i = 0; i < node->cold->prim_count; i++) {
615 assert(original_prims[i].basevertex == 0);
616 GLubyte mode = original_prims[i].mode;
618 int vertex_count = original_prims[i].count;
623 /* Line strips may get converted to lines */
624 if (mode == GL_LINE_STRIP)
627 /* If 2 consecutive prims use the same mode => merge them. */
628 bool merge_prims = last_valid_prim >= 0 &&
629 mode == merged_prims[last_valid_prim].mode &&
630 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
631 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
634 /* To be able to merge consecutive triangle strips we need to insert
635 * a degenerate triangle.
638 mode == GL_TRIANGLE_STRIP) {
639 /* Insert a degenerate triangle */
640 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
641 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
643 indices[idx] = indices[idx - 1];
644 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
645 temp_vertices_buffer, &max_index);
647 merged_prims[last_valid_prim].count += 2;
650 /* Add another index to preserve winding order */
651 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
652 temp_vertices_buffer, &max_index);
653 merged_prims[last_valid_prim].count++;
659 /* Convert line strips to lines if it'll allow if the previous
660 * prim mode is GL_LINES (so merge_prims is true) or if the next
661 * primitive mode is GL_LINES or GL_LINE_LOOP.
663 if (original_prims[i].mode == GL_LINE_STRIP &&
665 (i < node->cold->prim_count - 1 &&
666 (original_prims[i + 1].mode == GL_LINE_STRIP ||
667 original_prims[i + 1].mode == GL_LINES)))) {
668 for (unsigned j = 0; j < vertex_count; j++) {
669 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
670 temp_vertices_buffer, &max_index);
671 /* Repeat all but the first/last indices. */
672 if (j && j != vertex_count - 1) {
673 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
674 temp_vertices_buffer, &max_index);
678 /* We didn't convert to LINES, so restore the original mode */
679 mode = original_prims[i].mode;
681 for (unsigned j = 0; j < vertex_count; j++) {
682 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
683 temp_vertices_buffer, &max_index);
688 /* Update vertex count. */
689 merged_prims[last_valid_prim].count += idx - start;
691 /* Keep this primitive */
692 last_valid_prim += 1;
693 assert(last_valid_prim <= i);
694 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
695 merged_prims[last_valid_prim] = original_prims[i];
696 merged_prims[last_valid_prim].start = start;
697 merged_prims[last_valid_prim].count = idx - start;
699 merged_prims[last_valid_prim].mode = mode;
702 assert(idx > 0 && idx <= max_indices_count);
704 unsigned merged_prim_count = last_valid_prim + 1;
705 node->cold->ib.ptr = NULL;
706 node->cold->ib.count = idx;
707 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
709 /* How many bytes do we need to store the indices and the vertices */
710 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
711 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
712 total_vert_count * save->vertex_size * sizeof(fi_type);
714 const GLintptr old_offset = save->VAO[0] ?
715 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
716 if (old_offset != save->current_bo_bytes_used && stride > 0) {
717 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
718 while (offset_diff > 0 &&
719 save->current_bo_bytes_used < save->current_bo->Size &&
720 offset_diff % stride != 0) {
721 save->current_bo_bytes_used++;
722 offset_diff = save->current_bo_bytes_used - old_offset;
725 buffer_offset = save->current_bo_bytes_used;
727 /* Can we reuse the previous bo or should we allocate a new one? */
728 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
729 if (total_bytes_needed > available_bytes) {
730 if (save->current_bo)
731 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
732 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
733 bool success = ctx->Driver.BufferData(ctx,
734 GL_ELEMENT_ARRAY_BUFFER_ARB,
735 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t)),
737 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
740 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
741 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
743 save->current_bo_bytes_used = 0;
744 available_bytes = save->current_bo->Size;
748 assert(old_offset <= buffer_offset);
749 const GLintptr offset_diff = buffer_offset - old_offset;
750 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
751 /* The vertex size is an exact multiple of the buffer offset.
752 * This means that we can use zero-based vertex attribute pointers
753 * and specify the start of the primitive with the _mesa_prim::start
754 * field. This results in issuing several draw calls with identical
755 * vertex attribute information. This can result in fewer state
756 * changes in drivers. In particular, the Gallium CSO module will
757 * filter out redundant vertex buffer changes.
759 /* We cannot immediately update the primitives as some methods below
760 * still need the uncorrected start vertices
762 start_offset = offset_diff/stride;
763 assert(old_offset == buffer_offset - offset_diff);
764 buffer_offset = old_offset;
767 /* Correct the primitive starts, we can only do this here as copy_vertices
768 * and convert_line_loop_to_strip above consume the uncorrected starts.
769 * On the other hand the _vbo_loopback_vertex_list call below needs the
770 * primitives to be corrected already.
772 for (unsigned i = 0; i < node->cold->prim_count; i++) {
773 node->cold->prims[i].start += start_offset;
775 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
776 * to apply this transformation to all indices and max_index.
778 for (unsigned i = 0; i < idx; i++)
779 indices[i] += start_offset;
780 max_index += start_offset;
783 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
785 /* Upload the vertices first (see buffer_offset) */
786 ctx->Driver.BufferSubData(ctx,
787 save->current_bo_bytes_used,
788 total_vert_count * save->vertex_size * sizeof(fi_type),
789 vertex_to_index ? temp_vertices_buffer : save->vertex_store->buffer_in_ram,
791 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
793 if (vertex_to_index) {
794 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
795 free(temp_vertices_buffer);
798 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
799 * primitive (not the indices themselves). */
800 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
801 int indices_offset = save->current_bo_bytes_used / 4;
802 for (int i = 0; i < merged_prim_count; i++) {
803 merged_prims[i].start += indices_offset;
806 /* Then upload the indices. */
807 if (node->cold->ib.obj) {
808 ctx->Driver.BufferSubData(ctx,
809 save->current_bo_bytes_used,
810 idx * sizeof(uint32_t),
813 save->current_bo_bytes_used += idx * sizeof(uint32_t);
815 node->cold->vertex_count = 0;
816 node->cold->prim_count = 0;
819 /* Prepare for DrawGallium */
820 memset(&node->merged.info, 0, sizeof(struct pipe_draw_info));
821 /* The other info fields will be updated in vbo_save_playback_vertex_list */
822 node->merged.info.index_size = 4;
823 node->merged.info.instance_count = 1;
824 node->merged.info.index.gl_bo = node->cold->ib.obj;
825 if (merged_prim_count == 1) {
826 node->merged.info.mode = merged_prims[0].mode;
827 node->merged.start_count.start = merged_prims[0].start;
828 node->merged.start_count.count = merged_prims[0].count;
829 node->merged.start_count.index_bias = 0;
830 node->merged.mode = NULL;
832 node->merged.mode = malloc(merged_prim_count * sizeof(unsigned char));
833 node->merged.start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
834 for (unsigned i = 0; i < merged_prim_count; i++) {
835 node->merged.start_counts[i].start = merged_prims[i].start;
836 node->merged.start_counts[i].count = merged_prims[i].count;
837 node->merged.start_counts[i].index_bias = 0;
838 node->merged.mode[i] = merged_prims[i].mode;
841 node->merged.num_draws = merged_prim_count;
842 if (node->merged.num_draws > 1) {
843 bool same_mode = true;
844 for (unsigned i = 1; i < node->merged.num_draws && same_mode; i++) {
845 same_mode = node->merged.mode[i] == node->merged.mode[0];
848 /* All primitives use the same mode, so we can simplify a bit */
849 node->merged.info.mode = node->merged.mode[0];
850 free(node->merged.mode);
851 node->merged.mode = NULL;
860 if (!save->current_bo) {
861 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
862 bool success = ctx->Driver.BufferData(ctx,
863 GL_ELEMENT_ARRAY_BUFFER_ARB,
864 VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t),
866 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
870 GLuint offsets[VBO_ATTRIB_MAX];
871 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
873 offset += save->attrsz[i] * sizeof(GLfloat);
875 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
876 * Note that this may reuse the previous one of possible.
878 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
879 /* create or reuse the vao */
880 update_vao(ctx, vpm, &save->VAO[vpm],
881 save->current_bo, buffer_offset, stride,
882 save->enabled, save->attrsz, save->attrtype, offsets);
883 /* Reference the vao in the dlist */
884 node->VAO[vpm] = NULL;
885 _mesa_reference_vao(ctx, &node->VAO[vpm], save->VAO[vpm]);
889 /* Deal with GL_COMPILE_AND_EXECUTE:
891 if (ctx->ExecuteFlag) {
892 struct _glapi_table *dispatch = GET_DISPATCH();
894 _glapi_set_dispatch(ctx->Exec);
896 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
897 * use buffer_in_ram instead of current_bo which contains all vertices instead
898 * of the deduplicated vertices only in the !UseLoopback case.
900 * The problem is that the VAO offset is based on current_bo's layout,
901 * so we have to use a temp value.
903 struct gl_vertex_array_object *vao = node->VAO[VP_MODE_SHADER];
904 GLintptr original = vao->BufferBinding[0].Offset;
905 if (!ctx->ListState.Current.UseLoopback) {
906 GLintptr new_offset = 0;
907 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
908 new_offset -= start_offset * stride;
909 vao->BufferBinding[0].Offset = new_offset;
911 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
912 vao->BufferBinding[0].Offset = original;
914 _glapi_set_dispatch(dispatch);
917 /* Decide whether the storage structs are full, or can be used for
918 * the next vertex lists as well.
920 if (save->vertex_store->used >
921 save->vertex_store->buffer_in_ram_size / sizeof(float) - 16 * (save->vertex_size + 4)) {
922 realloc_storage(ctx, -1, 0);
925 /* Reset our structures for the next run of vertices:
932 * This is called when we fill a vertex buffer before we hit a glEnd().
934 * TODO -- If no new vertices have been stored, don't bother saving it.
937 wrap_buffers(struct gl_context *ctx)
939 struct vbo_save_context *save = &vbo_context(ctx)->save;
940 GLint i = save->prim_store->used - 1;
943 assert(i < (GLint) save->prim_store->size);
946 /* Close off in-progress primitive.
948 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
949 mode = save->prim_store->prims[i].mode;
951 /* store the copied vertices, and allocate a new list.
953 compile_vertex_list(ctx);
955 /* Restart interrupted primitive
957 save->prim_store->prims[0].mode = mode;
958 save->prim_store->prims[0].begin = 0;
959 save->prim_store->prims[0].end = 0;
960 save->prim_store->prims[0].start = 0;
961 save->prim_store->prims[0].count = 0;
962 save->prim_store->used = 1;
967 * Called only when buffers are wrapped as the result of filling the
968 * vertex_store struct.
971 wrap_filled_vertex(struct gl_context *ctx)
973 struct vbo_save_context *save = &vbo_context(ctx)->save;
974 unsigned numComponents;
976 /* Emit a glEnd to close off the last vertex list.
980 /* Copy stored stored vertices to start of new list.
982 assert(save->max_vert - save->vert_count > save->copied.nr);
984 numComponents = save->copied.nr * save->vertex_size;
986 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
989 numComponents * sizeof(fi_type));
990 assert(save->vertex_store->used == 0 && save->vert_count == 0);
991 save->vert_count = save->copied.nr;
992 save->vertex_store->used = numComponents;
997 copy_to_current(struct gl_context *ctx)
999 struct vbo_save_context *save = &vbo_context(ctx)->save;
1000 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1003 const int i = u_bit_scan64(&enabled);
1004 assert(save->attrsz[i]);
1006 if (save->attrtype[i] == GL_DOUBLE ||
1007 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1008 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1010 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1011 save->attrptr[i], save->attrtype[i]);
1017 copy_from_current(struct gl_context *ctx)
1019 struct vbo_save_context *save = &vbo_context(ctx)->save;
1020 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1023 const int i = u_bit_scan64(&enabled);
1025 switch (save->attrsz[i]) {
1027 save->attrptr[i][3] = save->current[i][3];
1030 save->attrptr[i][2] = save->current[i][2];
1033 save->attrptr[i][1] = save->current[i][1];
1036 save->attrptr[i][0] = save->current[i][0];
1039 unreachable("Unexpected vertex attribute size");
1046 * Called when we increase the size of a vertex attribute. For example,
1047 * if we've seen one or more glTexCoord2f() calls and now we get a
1048 * glTexCoord3f() call.
1049 * Flush existing data, set new attrib size, replay copied vertices.
1052 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1054 struct vbo_save_context *save = &vbo_context(ctx)->save;
1059 /* Store the current run of vertices, and emit a GL_END. Emit a
1060 * BEGIN in the new buffer.
1062 if (save->vert_count)
1065 assert(save->copied.nr == 0);
1067 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1068 * when the attribute already exists in the vertex and is having
1069 * its size increased.
1071 copy_to_current(ctx);
1075 oldsz = save->attrsz[attr];
1076 save->attrsz[attr] = newsz;
1077 save->enabled |= BITFIELD64_BIT(attr);
1079 save->vertex_size += newsz - oldsz;
1080 save->max_vert = ((save->vertex_store->buffer_in_ram_size / sizeof(float) -
1081 save->vertex_store->used) /
1083 save->vert_count = 0;
1085 /* Recalculate all the attrptr[] values:
1088 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1089 if (save->attrsz[i]) {
1090 save->attrptr[i] = tmp;
1091 tmp += save->attrsz[i];
1094 save->attrptr[i] = NULL; /* will not be dereferenced. */
1098 /* Copy from current to repopulate the vertex with correct values.
1100 copy_from_current(ctx);
1102 /* Replay stored vertices to translate them to new format here.
1104 * If there are copied vertices and the new (upgraded) attribute
1105 * has not been defined before, this list is somewhat degenerate,
1106 * and will need fixup at runtime.
1108 if (save->copied.nr) {
1109 const fi_type *data = save->copied.buffer;
1110 fi_type *dest = save->vertex_store->buffer_in_ram;
1112 /* Need to note this and fix up at runtime (or loopback):
1114 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1116 save->dangling_attr_ref = GL_TRUE;
1119 for (i = 0; i < save->copied.nr; i++) {
1120 GLbitfield64 enabled = save->enabled;
1122 const int j = u_bit_scan64(&enabled);
1123 assert(save->attrsz[j]);
1126 COPY_CLEAN_4V_TYPE_AS_UNION(dest, oldsz, data,
1132 COPY_SZ_4V(dest, newsz, save->current[attr]);
1137 GLint sz = save->attrsz[j];
1138 COPY_SZ_4V(dest, sz, data);
1145 save->vert_count += save->copied.nr;
1146 save->vertex_store->used += save->vertex_size * save->copied.nr;
1152 * This is called when the size of a vertex attribute changes.
1153 * For example, after seeing one or more glTexCoord2f() calls we
1154 * get a glTexCoord4f() or glTexCoord1f() call.
1157 fixup_vertex(struct gl_context *ctx, GLuint attr,
1158 GLuint sz, GLenum newType)
1160 struct vbo_save_context *save = &vbo_context(ctx)->save;
1162 if (sz > save->attrsz[attr] ||
1163 newType != save->attrtype[attr]) {
1164 /* New size is larger. Need to flush existing vertices and get
1165 * an enlarged vertex format.
1167 upgrade_vertex(ctx, attr, sz);
1169 else if (sz < save->active_sz[attr]) {
1171 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1173 /* New size is equal or smaller - just need to fill in some
1176 for (i = sz; i <= save->attrsz[attr]; i++)
1177 save->attrptr[attr][i - 1] = id[i - 1];
1180 save->active_sz[attr] = sz;
1185 * Reset the current size of all vertex attributes to the default
1186 * value of 0. This signals that we haven't yet seen any per-vertex
1187 * commands such as glNormal3f() or glTexCoord2f().
1190 reset_vertex(struct gl_context *ctx)
1192 struct vbo_save_context *save = &vbo_context(ctx)->save;
1194 while (save->enabled) {
1195 const int i = u_bit_scan64(&save->enabled);
1196 assert(save->attrsz[i]);
1197 save->attrsz[i] = 0;
1198 save->active_sz[i] = 0;
1201 save->vertex_size = 0;
1206 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1207 * It depends on a few things, including whether we're inside or outside
1211 is_vertex_position(const struct gl_context *ctx, GLuint index)
1213 return (index == 0 &&
1214 _mesa_attr_zero_aliases_vertex(ctx) &&
1215 _mesa_inside_dlist_begin_end(ctx));
1220 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1223 /* Only one size for each attribute may be active at once. Eg. if
1224 * Color3f is installed/active, then Color4f may not be, even if the
1225 * vertex actually contains 4 color coordinates. This is because the
1226 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1227 * of the chooser function when switching between Color4f and Color3f.
1229 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1231 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1232 int sz = (sizeof(C) / sizeof(GLfloat)); \
1234 if (save->active_sz[A] != N) \
1235 fixup_vertex(ctx, A, N * sz, T); \
1238 C *dest = (C *)save->attrptr[A]; \
1239 if (N>0) dest[0] = V0; \
1240 if (N>1) dest[1] = V1; \
1241 if (N>2) dest[2] = V2; \
1242 if (N>3) dest[3] = V3; \
1243 save->attrtype[A] = T; \
1248 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + save->vertex_store->used; \
1250 for (i = 0; i < save->vertex_size; i++) \
1251 buffer_ptr[i] = save->vertex[i]; \
1253 save->vertex_store->used += save->vertex_size; \
1254 if (++save->vert_count >= save->max_vert) \
1255 wrap_filled_vertex(ctx); \
1259 #define TAG(x) _save_##x
1261 #include "vbo_attrib_tmp.h"
1265 #define MAT( ATTR, N, face, params ) \
1267 if (face != GL_BACK) \
1268 MAT_ATTR( ATTR, N, params ); /* front */ \
1269 if (face != GL_FRONT) \
1270 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1275 * Save a glMaterial call found between glBegin/End.
1276 * glMaterial calls outside Begin/End are handled in dlist.c.
1278 static void GLAPIENTRY
1279 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1281 GET_CURRENT_CONTEXT(ctx);
1283 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1284 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1290 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1293 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1296 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1299 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1302 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1303 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1306 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1309 case GL_COLOR_INDEXES:
1310 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1312 case GL_AMBIENT_AND_DIFFUSE:
1313 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1314 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1317 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1323 /* Cope with EvalCoord/CallList called within a begin/end object:
1324 * -- Flush current buffer
1325 * -- Fallback to opcodes for the rest of the begin/end object.
1328 dlist_fallback(struct gl_context *ctx)
1330 struct vbo_save_context *save = &vbo_context(ctx)->save;
1332 if (save->vert_count || save->prim_store->used) {
1333 if (save->prim_store->used > 0) {
1334 /* Close off in-progress primitive. */
1335 GLint i = save->prim_store->used - 1;
1336 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1339 /* Need to replay this display list with loopback,
1340 * unfortunately, otherwise this primitive won't be handled
1343 save->dangling_attr_ref = GL_TRUE;
1345 compile_vertex_list(ctx);
1348 copy_to_current(ctx);
1350 if (save->out_of_memory) {
1351 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1354 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1356 ctx->Driver.SaveNeedFlush = GL_FALSE;
1360 static void GLAPIENTRY
1361 _save_EvalCoord1f(GLfloat u)
1363 GET_CURRENT_CONTEXT(ctx);
1364 dlist_fallback(ctx);
1365 CALL_EvalCoord1f(ctx->Save, (u));
1368 static void GLAPIENTRY
1369 _save_EvalCoord1fv(const GLfloat * v)
1371 GET_CURRENT_CONTEXT(ctx);
1372 dlist_fallback(ctx);
1373 CALL_EvalCoord1fv(ctx->Save, (v));
1376 static void GLAPIENTRY
1377 _save_EvalCoord2f(GLfloat u, GLfloat v)
1379 GET_CURRENT_CONTEXT(ctx);
1380 dlist_fallback(ctx);
1381 CALL_EvalCoord2f(ctx->Save, (u, v));
1384 static void GLAPIENTRY
1385 _save_EvalCoord2fv(const GLfloat * v)
1387 GET_CURRENT_CONTEXT(ctx);
1388 dlist_fallback(ctx);
1389 CALL_EvalCoord2fv(ctx->Save, (v));
1392 static void GLAPIENTRY
1393 _save_EvalPoint1(GLint i)
1395 GET_CURRENT_CONTEXT(ctx);
1396 dlist_fallback(ctx);
1397 CALL_EvalPoint1(ctx->Save, (i));
1400 static void GLAPIENTRY
1401 _save_EvalPoint2(GLint i, GLint j)
1403 GET_CURRENT_CONTEXT(ctx);
1404 dlist_fallback(ctx);
1405 CALL_EvalPoint2(ctx->Save, (i, j));
1408 static void GLAPIENTRY
1409 _save_CallList(GLuint l)
1411 GET_CURRENT_CONTEXT(ctx);
1412 dlist_fallback(ctx);
1413 CALL_CallList(ctx->Save, (l));
1416 static void GLAPIENTRY
1417 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1419 GET_CURRENT_CONTEXT(ctx);
1420 dlist_fallback(ctx);
1421 CALL_CallLists(ctx->Save, (n, type, v));
1427 * Called when a glBegin is getting compiled into a display list.
1428 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1431 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1432 bool no_current_update)
1434 struct vbo_save_context *save = &vbo_context(ctx)->save;
1435 const GLuint i = save->prim_store->used++;
1437 ctx->Driver.CurrentSavePrimitive = mode;
1439 assert(i < save->prim_store->size);
1440 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1441 save->prim_store->prims[i].begin = 1;
1442 save->prim_store->prims[i].end = 0;
1443 save->prim_store->prims[i].start = save->vert_count;
1444 save->prim_store->prims[i].count = 0;
1446 save->no_current_update = no_current_update;
1448 if (save->out_of_memory) {
1449 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1452 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1455 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1456 ctx->Driver.SaveNeedFlush = GL_TRUE;
1460 static void GLAPIENTRY
1463 GET_CURRENT_CONTEXT(ctx);
1464 struct vbo_save_context *save = &vbo_context(ctx)->save;
1465 const GLint i = save->prim_store->used - 1;
1467 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1468 save->prim_store->prims[i].end = 1;
1469 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
1471 if (i == (GLint) save->prim_store->size - 1) {
1472 compile_vertex_list(ctx);
1473 assert(save->copied.nr == 0);
1476 /* Swap out this vertex format while outside begin/end. Any color,
1477 * etc. received between here and the next begin will be compiled
1480 if (save->out_of_memory) {
1481 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1484 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1489 static void GLAPIENTRY
1490 _save_Begin(GLenum mode)
1492 GET_CURRENT_CONTEXT(ctx);
1494 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1498 static void GLAPIENTRY
1499 _save_PrimitiveRestartNV(void)
1501 GET_CURRENT_CONTEXT(ctx);
1502 struct vbo_save_context *save = &vbo_context(ctx)->save;
1504 if (save->prim_store->used == 0) {
1505 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1508 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1509 "glPrimitiveRestartNV called outside glBegin/End");
1511 /* get current primitive mode */
1512 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1513 bool no_current_update = save->no_current_update;
1515 /* restart primitive */
1516 CALL_End(ctx->CurrentServerDispatch, ());
1517 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1522 /* Unlike the functions above, these are to be hooked into the vtxfmt
1523 * maintained in ctx->ListState, active when the list is known or
1524 * suspected to be outside any begin/end primitive.
1525 * Note: OBE = Outside Begin/End
1527 static void GLAPIENTRY
1528 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1530 GET_CURRENT_CONTEXT(ctx);
1531 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1533 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1534 CALL_Vertex2f(dispatch, (x1, y1));
1535 CALL_Vertex2f(dispatch, (x2, y1));
1536 CALL_Vertex2f(dispatch, (x2, y2));
1537 CALL_Vertex2f(dispatch, (x1, y2));
1538 CALL_End(dispatch, ());
1542 static void GLAPIENTRY
1543 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1545 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1548 static void GLAPIENTRY
1549 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1551 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1554 static void GLAPIENTRY
1555 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1557 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1560 static void GLAPIENTRY
1561 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1563 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1566 static void GLAPIENTRY
1567 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1569 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1572 static void GLAPIENTRY
1573 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1575 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1578 static void GLAPIENTRY
1579 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1581 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1585 _ensure_draws_fits_in_storage(struct gl_context *ctx, int primcount, int vertcount)
1587 struct vbo_save_context *save = &vbo_context(ctx)->save;
1589 bool realloc_prim = save->prim_store->used + primcount > save->prim_store->size;
1590 bool realloc_vert = save->vertex_size && (save->vert_count + vertcount >= save->max_vert);
1592 if (realloc_prim || realloc_vert) {
1593 if (realloc_vert && (save->vert_count || save->prim_store->used)) {
1594 /* TODO: this really isn't needed. We should realloc only the CPU-side memory. */
1595 compile_vertex_list(ctx);
1597 realloc_storage(ctx, realloc_prim ? primcount : -1, realloc_vert ? vertcount : -1);
1602 static void GLAPIENTRY
1603 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1605 GET_CURRENT_CONTEXT(ctx);
1606 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1607 struct vbo_save_context *save = &vbo_context(ctx)->save;
1610 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1611 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1615 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1619 if (save->out_of_memory)
1622 _ensure_draws_fits_in_storage(ctx, 1, count);
1624 /* Make sure to process any VBO binding changes */
1625 _mesa_update_state(ctx);
1627 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1629 vbo_save_NotifyBegin(ctx, mode, true);
1631 for (i = 0; i < count; i++)
1632 _mesa_array_element(ctx, start + i);
1633 CALL_End(ctx->CurrentServerDispatch, ());
1635 _mesa_vao_unmap_arrays(ctx, vao);
1639 static void GLAPIENTRY
1640 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1641 const GLsizei *count, GLsizei primcount)
1643 GET_CURRENT_CONTEXT(ctx);
1646 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1647 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1651 if (primcount < 0) {
1652 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1653 "glMultiDrawArrays(primcount<0)");
1657 unsigned vertcount = 0;
1658 for (i = 0; i < primcount; i++) {
1660 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1661 "glMultiDrawArrays(count[i]<0)");
1664 vertcount += count[i];
1667 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1669 for (i = 0; i < primcount; i++) {
1671 _save_OBE_DrawArrays(mode, first[i], count[i]);
1678 array_element(struct gl_context *ctx,
1679 GLint basevertex, GLuint elt, unsigned index_size_shift)
1681 /* Section 10.3.5 Primitive Restart:
1683 * When one of the *BaseVertex drawing commands specified in section 10.5
1684 * is used, the primitive restart comparison occurs before the basevertex
1685 * offset is added to the array index.
1687 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1688 * then we call PrimitiveRestartNV and return.
1690 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1691 elt == ctx->Array._RestartIndex[index_size_shift]) {
1692 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1696 _mesa_array_element(ctx, basevertex + elt);
1700 /* Could do better by copying the arrays and element list intact and
1701 * then emitting an indexed prim at runtime.
1703 static void GLAPIENTRY
1704 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1705 const GLvoid * indices, GLint basevertex)
1707 GET_CURRENT_CONTEXT(ctx);
1708 struct vbo_save_context *save = &vbo_context(ctx)->save;
1709 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1710 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1713 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1714 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1718 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1721 if (type != GL_UNSIGNED_BYTE &&
1722 type != GL_UNSIGNED_SHORT &&
1723 type != GL_UNSIGNED_INT) {
1724 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1728 if (save->out_of_memory)
1731 _ensure_draws_fits_in_storage(ctx, 1, count);
1733 /* Make sure to process any VBO binding changes */
1734 _mesa_update_state(ctx);
1736 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1740 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1742 vbo_save_NotifyBegin(ctx, mode, true);
1745 case GL_UNSIGNED_BYTE:
1746 for (i = 0; i < count; i++)
1747 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1749 case GL_UNSIGNED_SHORT:
1750 for (i = 0; i < count; i++)
1751 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1753 case GL_UNSIGNED_INT:
1754 for (i = 0; i < count; i++)
1755 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1758 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1762 CALL_End(ctx->CurrentServerDispatch, ());
1764 _mesa_vao_unmap(ctx, vao);
1767 static void GLAPIENTRY
1768 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1769 const GLvoid * indices)
1771 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1775 static void GLAPIENTRY
1776 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1777 GLsizei count, GLenum type,
1778 const GLvoid * indices)
1780 GET_CURRENT_CONTEXT(ctx);
1781 struct vbo_save_context *save = &vbo_context(ctx)->save;
1783 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1784 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1788 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1789 "glDrawRangeElements(count<0)");
1792 if (type != GL_UNSIGNED_BYTE &&
1793 type != GL_UNSIGNED_SHORT &&
1794 type != GL_UNSIGNED_INT) {
1795 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1799 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1800 "glDrawRangeElements(end < start)");
1804 if (save->out_of_memory)
1807 _save_OBE_DrawElements(mode, count, type, indices);
1811 static void GLAPIENTRY
1812 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1813 const GLvoid * const *indices, GLsizei primcount)
1815 GET_CURRENT_CONTEXT(ctx);
1816 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1820 for (i = 0; i < primcount; i++) {
1821 vertcount += count[i];
1823 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1825 for (i = 0; i < primcount; i++) {
1827 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1833 static void GLAPIENTRY
1834 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1836 const GLvoid * const *indices,
1838 const GLint *basevertex)
1840 GET_CURRENT_CONTEXT(ctx);
1841 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1845 for (i = 0; i < primcount; i++) {
1846 vertcount += count[i];
1848 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1850 for (i = 0; i < primcount; i++) {
1852 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1861 vtxfmt_init(struct gl_context *ctx)
1863 struct vbo_save_context *save = &vbo_context(ctx)->save;
1864 GLvertexformat *vfmt = &save->vtxfmt;
1866 #define NAME_AE(x) _ae_##x
1867 #define NAME_CALLLIST(x) _save_##x
1868 #define NAME(x) _save_##x
1869 #define NAME_ES(x) _save_##x##ARB
1871 #include "vbo_init_tmp.h"
1876 * Initialize the dispatch table with the VBO functions for display
1880 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1881 struct _glapi_table *exec)
1883 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1884 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1885 SET_DrawElements(exec, _save_OBE_DrawElements);
1886 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1887 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1888 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1889 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1890 SET_Rectf(exec, _save_OBE_Rectf);
1891 SET_Rectd(exec, _save_OBE_Rectd);
1892 SET_Rectdv(exec, _save_OBE_Rectdv);
1893 SET_Rectfv(exec, _save_OBE_Rectfv);
1894 SET_Recti(exec, _save_OBE_Recti);
1895 SET_Rectiv(exec, _save_OBE_Rectiv);
1896 SET_Rects(exec, _save_OBE_Rects);
1897 SET_Rectsv(exec, _save_OBE_Rectsv);
1899 /* Note: other glDraw functins aren't compiled into display lists */
1905 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1907 struct vbo_save_context *save = &vbo_context(ctx)->save;
1909 /* Noop when we are actually active:
1911 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1914 if (save->vert_count || save->prim_store->used)
1915 compile_vertex_list(ctx);
1917 copy_to_current(ctx);
1919 ctx->Driver.SaveNeedFlush = GL_FALSE;
1924 * Called from glNewList when we're starting to compile a display list.
1927 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1929 struct vbo_save_context *save = &vbo_context(ctx)->save;
1934 if (!save->prim_store)
1935 save->prim_store = realloc_prim_store(NULL, 8);
1937 if (!save->vertex_store)
1938 save->vertex_store = alloc_vertex_store(ctx, 0);
1941 ctx->Driver.SaveNeedFlush = GL_FALSE;
1946 * Called from glEndList when we're finished compiling a display list.
1949 vbo_save_EndList(struct gl_context *ctx)
1951 struct vbo_save_context *save = &vbo_context(ctx)->save;
1953 /* EndList called inside a (saved) Begin/End pair?
1955 if (_mesa_inside_dlist_begin_end(ctx)) {
1956 if (save->prim_store->used > 0) {
1957 GLint i = save->prim_store->used - 1;
1958 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1959 save->prim_store->prims[i].end = 0;
1960 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1963 /* Make sure this vertex list gets replayed by the "loopback"
1966 save->dangling_attr_ref = GL_TRUE;
1967 vbo_save_SaveFlushVertices(ctx);
1969 /* Swap out this vertex format while outside begin/end. Any color,
1970 * etc. received between here and the next begin will be compiled
1973 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1976 assert(save->vertex_size == 0);
1980 * Called during context creation/init.
1983 current_init(struct gl_context *ctx)
1985 struct vbo_save_context *save = &vbo_context(ctx)->save;
1988 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
1989 const GLuint j = i - VBO_ATTRIB_POS;
1990 assert(j < VERT_ATTRIB_MAX);
1991 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
1992 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
1995 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
1996 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
1997 assert(j < MAT_ATTRIB_MAX);
1998 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
1999 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2005 * Initialize the display list compiler. Called during context creation.
2008 vbo_save_api_init(struct vbo_save_context *save)
2010 struct gl_context *ctx = gl_context_from_vbo_save(save);
2014 _mesa_noop_vtxfmt_init(ctx, &save->vtxfmt_noop);