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 alloc_prim_store(int prim_count)
175 struct vbo_save_primitive_store *store =
176 CALLOC_STRUCT(vbo_save_primitive_store);
177 store->size = MAX2(prim_count, VBO_SAVE_PRIM_SIZE);
178 store->prims = calloc(store->size, sizeof(struct _mesa_prim));
185 reset_counters(struct gl_context *ctx)
187 struct vbo_save_context *save = &vbo_context(ctx)->save;
189 save->prims = save->prim_store->prims + save->prim_store->used;
190 save->buffer_map = save->vertex_store->buffer_in_ram + save->vertex_store->used;
192 assert(save->buffer_map == save->buffer_ptr);
194 if (save->vertex_size)
195 save->max_vert = (save->vertex_store->buffer_in_ram_size / sizeof(float) - save->vertex_store->used) /
200 save->vert_count = 0;
201 save->prim_count = 0;
202 save->prim_max = save->prim_store->size - save->prim_store->used;
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;
424 if (prim_count >= 0) {
425 free(save->prim_store->prims);
426 free(save->prim_store);
427 save->prim_store = alloc_prim_store(prim_count);
432 unsigned vertex_size;
433 fi_type *vertex_attributes;
436 static uint32_t _hash_vertex_key(const void *key)
438 struct vertex_key *k = (struct vertex_key*)key;
439 unsigned sz = k->vertex_size;
441 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
444 static bool _compare_vertex_key(const void *key1, const void *key2)
446 struct vertex_key *k1 = (struct vertex_key*)key1;
447 struct vertex_key *k2 = (struct vertex_key*)key2;
448 /* All the compared vertices are going to be drawn with the same VAO,
449 * so we can compare the attributes. */
450 assert (k1->vertex_size == k2->vertex_size);
451 return memcmp(k1->vertex_attributes,
452 k2->vertex_attributes,
453 k1->vertex_size * sizeof(float)) == 0;
456 static void _free_entry(struct hash_entry *entry)
458 free((void*)entry->key);
461 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
462 * of an existing vertex, return the original index instead.
465 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
466 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
468 /* If vertex deduplication is disabled return the original index. */
472 fi_type *vert = save->buffer_map + save->vertex_size * index;
474 struct vertex_key *key = malloc(sizeof(struct vertex_key));
475 key->vertex_size = save->vertex_size;
476 key->vertex_attributes = vert;
478 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
481 /* We found an existing vertex with the same hash, return its index. */
482 return (uintptr_t) entry->data;
484 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
485 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
486 * starting at index 0.
488 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
489 *max_index = MAX2(n, *max_index);
491 memcpy(&new_buffer[save->vertex_size * n],
493 save->vertex_size * sizeof(fi_type));
495 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
497 /* The index buffer is shared between list compilations, so add the base index to get
506 * Insert the active immediate struct onto the display list currently
510 compile_vertex_list(struct gl_context *ctx)
512 struct vbo_save_context *save = &vbo_context(ctx)->save;
513 struct vbo_save_vertex_list *node;
515 /* Allocate space for this structure in the display list currently
518 node = (struct vbo_save_vertex_list *)
519 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
524 memset(node, 0, sizeof(struct vbo_save_vertex_list));
525 node->cold = calloc(1, sizeof(*node->cold));
527 /* Make sure the pointer is aligned to the size of a pointer */
528 assert((GLintptr) node % sizeof(void *) == 0);
530 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
532 node->cold->vertex_count = save->vert_count;
533 node->cold->wrap_count = save->copied.nr;
534 node->cold->prims = save->prims;
535 node->cold->ib.obj = NULL;
536 node->cold->prim_count = save->prim_count;
538 if (save->no_current_update) {
539 node->cold->current_data = NULL;
542 GLuint current_size = save->vertex_size - save->attrsz[0];
543 node->cold->current_data = NULL;
546 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
547 if (node->cold->current_data) {
548 const char *buffer = (const char *)save->buffer_map;
549 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
550 unsigned vertex_offset = 0;
552 if (node->cold->vertex_count)
553 vertex_offset = (node->cold->vertex_count - 1) * stride;
555 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
556 current_size * sizeof(GLfloat));
558 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
563 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
565 if (save->dangling_attr_ref)
566 ctx->ListState.Current.UseLoopback = true;
568 save->vertex_store->used += save->vertex_size * node->cold->vertex_count;
569 save->prim_store->used += node->cold->prim_count;
571 /* Copy duplicated vertices
573 save->copied.nr = copy_vertices(ctx, node, save->buffer_map);
575 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
576 convert_line_loop_to_strip(save, node);
579 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
581 GLintptr buffer_offset = 0;
582 GLuint start_offset = 0;
584 /* Create an index buffer. */
585 node->cold->min_index = node->cold->max_index = 0;
586 if (save->vert_count == 0 || save->prim_count == 0)
589 /* We won't modify node->prims, so use a const alias to avoid unintended
591 const struct _mesa_prim *original_prims = node->cold->prims;
593 int end = original_prims[node->cold->prim_count - 1].start +
594 original_prims[node->cold->prim_count - 1].count;
595 int total_vert_count = end - original_prims[0].start;
597 node->cold->min_index = node->cold->prims[0].start;
598 node->cold->max_index = end - 1;
600 /* Estimate for the worst case: all prims are line strips (the +1 is because
601 * wrap_buffers may call use but the last primitive may not be complete) */
602 int max_indices_count = MAX2(total_vert_count * 2 - (node->cold->prim_count * 2) + 1,
605 int size = max_indices_count * sizeof(uint32_t);
606 uint32_t* indices = (uint32_t*) malloc(size);
607 struct _mesa_prim *merged_prims = NULL;
610 struct hash_table *vertex_to_index = NULL;
611 fi_type *temp_vertices_buffer = NULL;
613 /* The loopback replay code doesn't use the index buffer, so we can't
614 * dedup vertices in this case.
616 if (!ctx->ListState.Current.UseLoopback) {
617 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
618 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
621 uint32_t max_index = 0;
623 int last_valid_prim = -1;
624 /* Construct indices array. */
625 for (unsigned i = 0; i < node->cold->prim_count; i++) {
626 assert(original_prims[i].basevertex == 0);
627 GLubyte mode = original_prims[i].mode;
629 int vertex_count = original_prims[i].count;
634 /* Line strips may get converted to lines */
635 if (mode == GL_LINE_STRIP)
638 /* If 2 consecutive prims use the same mode => merge them. */
639 bool merge_prims = last_valid_prim >= 0 &&
640 mode == merged_prims[last_valid_prim].mode &&
641 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
642 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
645 /* To be able to merge consecutive triangle strips we need to insert
646 * a degenerate triangle.
649 mode == GL_TRIANGLE_STRIP) {
650 /* Insert a degenerate triangle */
651 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
652 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
654 indices[idx] = indices[idx - 1];
655 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
656 temp_vertices_buffer, &max_index);
658 merged_prims[last_valid_prim].count += 2;
661 /* Add another index to preserve winding order */
662 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
663 temp_vertices_buffer, &max_index);
664 merged_prims[last_valid_prim].count++;
670 /* Convert line strips to lines if it'll allow if the previous
671 * prim mode is GL_LINES (so merge_prims is true) or if the next
672 * primitive mode is GL_LINES or GL_LINE_LOOP.
674 if (original_prims[i].mode == GL_LINE_STRIP &&
676 (i < node->cold->prim_count - 1 &&
677 (original_prims[i + 1].mode == GL_LINE_STRIP ||
678 original_prims[i + 1].mode == GL_LINES)))) {
679 for (unsigned j = 0; j < vertex_count; j++) {
680 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
681 temp_vertices_buffer, &max_index);
682 /* Repeat all but the first/last indices. */
683 if (j && j != vertex_count - 1) {
684 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
685 temp_vertices_buffer, &max_index);
689 /* We didn't convert to LINES, so restore the original mode */
690 mode = original_prims[i].mode;
692 for (unsigned j = 0; j < vertex_count; j++) {
693 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
694 temp_vertices_buffer, &max_index);
699 /* Update vertex count. */
700 merged_prims[last_valid_prim].count += idx - start;
702 /* Keep this primitive */
703 last_valid_prim += 1;
704 assert(last_valid_prim <= i);
705 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
706 merged_prims[last_valid_prim] = original_prims[i];
707 merged_prims[last_valid_prim].start = start;
708 merged_prims[last_valid_prim].count = idx - start;
710 merged_prims[last_valid_prim].mode = mode;
713 assert(idx > 0 && idx <= max_indices_count);
715 unsigned merged_prim_count = last_valid_prim + 1;
716 node->cold->ib.ptr = NULL;
717 node->cold->ib.count = idx;
718 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
720 /* How many bytes do we need to store the indices and the vertices */
721 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
722 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
723 total_vert_count * save->vertex_size * sizeof(fi_type);
725 const GLintptr old_offset = save->VAO[0] ?
726 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
727 if (old_offset != save->current_bo_bytes_used && stride > 0) {
728 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
729 while (offset_diff > 0 &&
730 save->current_bo_bytes_used < save->current_bo->Size &&
731 offset_diff % stride != 0) {
732 save->current_bo_bytes_used++;
733 offset_diff = save->current_bo_bytes_used - old_offset;
736 buffer_offset = save->current_bo_bytes_used;
738 /* Can we reuse the previous bo or should we allocate a new one? */
739 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
740 if (total_bytes_needed > available_bytes) {
741 if (save->current_bo)
742 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
743 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
744 bool success = ctx->Driver.BufferData(ctx,
745 GL_ELEMENT_ARRAY_BUFFER_ARB,
746 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t)),
748 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
751 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
752 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
754 save->current_bo_bytes_used = 0;
755 available_bytes = save->current_bo->Size;
759 assert(old_offset <= buffer_offset);
760 const GLintptr offset_diff = buffer_offset - old_offset;
761 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
762 /* The vertex size is an exact multiple of the buffer offset.
763 * This means that we can use zero-based vertex attribute pointers
764 * and specify the start of the primitive with the _mesa_prim::start
765 * field. This results in issuing several draw calls with identical
766 * vertex attribute information. This can result in fewer state
767 * changes in drivers. In particular, the Gallium CSO module will
768 * filter out redundant vertex buffer changes.
770 /* We cannot immediately update the primitives as some methods below
771 * still need the uncorrected start vertices
773 start_offset = offset_diff/stride;
774 assert(old_offset == buffer_offset - offset_diff);
775 buffer_offset = old_offset;
778 /* Correct the primitive starts, we can only do this here as copy_vertices
779 * and convert_line_loop_to_strip above consume the uncorrected starts.
780 * On the other hand the _vbo_loopback_vertex_list call below needs the
781 * primitives to be corrected already.
783 for (unsigned i = 0; i < node->cold->prim_count; i++) {
784 node->cold->prims[i].start += start_offset;
786 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
787 * to apply this transformation to all indices and max_index.
789 for (unsigned i = 0; i < idx; i++)
790 indices[i] += start_offset;
791 max_index += start_offset;
794 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
796 /* Upload the vertices first (see buffer_offset) */
797 ctx->Driver.BufferSubData(ctx,
798 save->current_bo_bytes_used,
799 total_vert_count * save->vertex_size * sizeof(fi_type),
800 vertex_to_index ? temp_vertices_buffer : save->buffer_map,
802 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
804 if (vertex_to_index) {
805 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
806 free(temp_vertices_buffer);
809 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
810 * primitive (not the indices themselves). */
811 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
812 int indices_offset = save->current_bo_bytes_used / 4;
813 for (int i = 0; i < merged_prim_count; i++) {
814 merged_prims[i].start += indices_offset;
817 /* Then upload the indices. */
818 if (node->cold->ib.obj) {
819 ctx->Driver.BufferSubData(ctx,
820 save->current_bo_bytes_used,
821 idx * sizeof(uint32_t),
824 save->current_bo_bytes_used += idx * sizeof(uint32_t);
826 node->cold->vertex_count = 0;
827 node->cold->prim_count = 0;
830 /* Prepare for DrawGallium */
831 memset(&node->merged.info, 0, sizeof(struct pipe_draw_info));
832 /* The other info fields will be updated in vbo_save_playback_vertex_list */
833 node->merged.info.index_size = 4;
834 node->merged.info.instance_count = 1;
835 node->merged.info.index.gl_bo = node->cold->ib.obj;
836 if (merged_prim_count == 1) {
837 node->merged.info.mode = merged_prims[0].mode;
838 node->merged.start_count.start = merged_prims[0].start;
839 node->merged.start_count.count = merged_prims[0].count;
840 node->merged.start_count.index_bias = 0;
841 node->merged.mode = NULL;
843 node->merged.mode = malloc(merged_prim_count * sizeof(unsigned char));
844 node->merged.start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
845 for (unsigned i = 0; i < merged_prim_count; i++) {
846 node->merged.start_counts[i].start = merged_prims[i].start;
847 node->merged.start_counts[i].count = merged_prims[i].count;
848 node->merged.start_counts[i].index_bias = 0;
849 node->merged.mode[i] = merged_prims[i].mode;
852 node->merged.num_draws = merged_prim_count;
853 if (node->merged.num_draws > 1) {
854 bool same_mode = true;
855 for (unsigned i = 1; i < node->merged.num_draws && same_mode; i++) {
856 same_mode = node->merged.mode[i] == node->merged.mode[0];
859 /* All primitives use the same mode, so we can simplify a bit */
860 node->merged.info.mode = node->merged.mode[0];
861 free(node->merged.mode);
862 node->merged.mode = NULL;
871 if (!save->current_bo) {
872 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
873 bool success = ctx->Driver.BufferData(ctx,
874 GL_ELEMENT_ARRAY_BUFFER_ARB,
875 VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t),
877 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
881 GLuint offsets[VBO_ATTRIB_MAX];
882 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
884 offset += save->attrsz[i] * sizeof(GLfloat);
886 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
887 * Note that this may reuse the previous one of possible.
889 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
890 /* create or reuse the vao */
891 update_vao(ctx, vpm, &save->VAO[vpm],
892 save->current_bo, buffer_offset, stride,
893 save->enabled, save->attrsz, save->attrtype, offsets);
894 /* Reference the vao in the dlist */
895 node->VAO[vpm] = NULL;
896 _mesa_reference_vao(ctx, &node->VAO[vpm], save->VAO[vpm]);
900 /* Deal with GL_COMPILE_AND_EXECUTE:
902 if (ctx->ExecuteFlag) {
903 struct _glapi_table *dispatch = GET_DISPATCH();
905 _glapi_set_dispatch(ctx->Exec);
907 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
908 * use buffer_in_ram instead of current_bo which contains all vertices instead
909 * of the deduplicated vertices only in the !UseLoopback case.
911 * The problem is that the VAO offset is based on current_bo's layout,
912 * so we have to use a temp value.
914 struct gl_vertex_array_object *vao = node->VAO[VP_MODE_SHADER];
915 GLintptr original = vao->BufferBinding[0].Offset;
916 if (!ctx->ListState.Current.UseLoopback) {
917 GLintptr new_offset = (save->buffer_map - save->vertex_store->buffer_in_ram) *
919 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
920 new_offset -= start_offset * stride;
921 vao->BufferBinding[0].Offset = new_offset;
923 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
924 vao->BufferBinding[0].Offset = original;
926 _glapi_set_dispatch(dispatch);
929 /* Decide whether the storage structs are full, or can be used for
930 * the next vertex lists as well.
932 if (save->vertex_store->used >
933 save->vertex_store->buffer_in_ram_size / sizeof(float) - 16 * (save->vertex_size + 4)) {
934 realloc_storage(ctx, -1, 0);
937 /* update buffer_ptr for next vertex */
938 save->buffer_ptr = save->vertex_store->buffer_in_ram
939 + save->vertex_store->used;
942 if (save->prim_store->used > save->prim_store->size - 6) {
943 realloc_storage(ctx, 0, -1);
946 /* Reset our structures for the next run of vertices:
953 * This is called when we fill a vertex buffer before we hit a glEnd().
955 * TODO -- If no new vertices have been stored, don't bother saving it.
958 wrap_buffers(struct gl_context *ctx)
960 struct vbo_save_context *save = &vbo_context(ctx)->save;
961 GLint i = save->prim_count - 1;
964 assert(i < (GLint) save->prim_max);
967 /* Close off in-progress primitive.
969 save->prims[i].count = (save->vert_count - save->prims[i].start);
970 mode = save->prims[i].mode;
972 /* store the copied vertices, and allocate a new list.
974 compile_vertex_list(ctx);
976 /* Restart interrupted primitive
978 save->prims[0].mode = mode;
979 save->prims[0].begin = 0;
980 save->prims[0].end = 0;
981 save->prims[0].start = 0;
982 save->prims[0].count = 0;
983 save->prim_count = 1;
988 * Called only when buffers are wrapped as the result of filling the
989 * vertex_store struct.
992 wrap_filled_vertex(struct gl_context *ctx)
994 struct vbo_save_context *save = &vbo_context(ctx)->save;
995 unsigned numComponents;
997 /* Emit a glEnd to close off the last vertex list.
1001 /* Copy stored stored vertices to start of new list.
1003 assert(save->max_vert - save->vert_count > save->copied.nr);
1005 numComponents = save->copied.nr * save->vertex_size;
1006 memcpy(save->buffer_ptr,
1007 save->copied.buffer,
1008 numComponents * sizeof(fi_type));
1009 save->buffer_ptr += numComponents;
1010 save->vert_count += save->copied.nr;
1015 copy_to_current(struct gl_context *ctx)
1017 struct vbo_save_context *save = &vbo_context(ctx)->save;
1018 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1021 const int i = u_bit_scan64(&enabled);
1022 assert(save->attrsz[i]);
1024 if (save->attrtype[i] == GL_DOUBLE ||
1025 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1026 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1028 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1029 save->attrptr[i], save->attrtype[i]);
1035 copy_from_current(struct gl_context *ctx)
1037 struct vbo_save_context *save = &vbo_context(ctx)->save;
1038 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1041 const int i = u_bit_scan64(&enabled);
1043 switch (save->attrsz[i]) {
1045 save->attrptr[i][3] = save->current[i][3];
1048 save->attrptr[i][2] = save->current[i][2];
1051 save->attrptr[i][1] = save->current[i][1];
1054 save->attrptr[i][0] = save->current[i][0];
1057 unreachable("Unexpected vertex attribute size");
1064 * Called when we increase the size of a vertex attribute. For example,
1065 * if we've seen one or more glTexCoord2f() calls and now we get a
1066 * glTexCoord3f() call.
1067 * Flush existing data, set new attrib size, replay copied vertices.
1070 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1072 struct vbo_save_context *save = &vbo_context(ctx)->save;
1077 /* Store the current run of vertices, and emit a GL_END. Emit a
1078 * BEGIN in the new buffer.
1080 if (save->vert_count)
1083 assert(save->copied.nr == 0);
1085 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1086 * when the attribute already exists in the vertex and is having
1087 * its size increased.
1089 copy_to_current(ctx);
1093 oldsz = save->attrsz[attr];
1094 save->attrsz[attr] = newsz;
1095 save->enabled |= BITFIELD64_BIT(attr);
1097 save->vertex_size += newsz - oldsz;
1098 save->max_vert = ((save->vertex_store->buffer_in_ram_size / sizeof(float) -
1099 save->vertex_store->used) /
1101 save->vert_count = 0;
1103 /* Recalculate all the attrptr[] values:
1106 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1107 if (save->attrsz[i]) {
1108 save->attrptr[i] = tmp;
1109 tmp += save->attrsz[i];
1112 save->attrptr[i] = NULL; /* will not be dereferenced. */
1116 /* Copy from current to repopulate the vertex with correct values.
1118 copy_from_current(ctx);
1120 /* Replay stored vertices to translate them to new format here.
1122 * If there are copied vertices and the new (upgraded) attribute
1123 * has not been defined before, this list is somewhat degenerate,
1124 * and will need fixup at runtime.
1126 if (save->copied.nr) {
1127 const fi_type *data = save->copied.buffer;
1128 fi_type *dest = save->buffer_map;
1130 /* Need to note this and fix up at runtime (or loopback):
1132 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1134 save->dangling_attr_ref = GL_TRUE;
1137 for (i = 0; i < save->copied.nr; i++) {
1138 GLbitfield64 enabled = save->enabled;
1140 const int j = u_bit_scan64(&enabled);
1141 assert(save->attrsz[j]);
1144 COPY_CLEAN_4V_TYPE_AS_UNION(dest, oldsz, data,
1150 COPY_SZ_4V(dest, newsz, save->current[attr]);
1155 GLint sz = save->attrsz[j];
1156 COPY_SZ_4V(dest, sz, data);
1163 save->buffer_ptr = dest;
1164 save->vert_count += save->copied.nr;
1170 * This is called when the size of a vertex attribute changes.
1171 * For example, after seeing one or more glTexCoord2f() calls we
1172 * get a glTexCoord4f() or glTexCoord1f() call.
1175 fixup_vertex(struct gl_context *ctx, GLuint attr,
1176 GLuint sz, GLenum newType)
1178 struct vbo_save_context *save = &vbo_context(ctx)->save;
1180 if (sz > save->attrsz[attr] ||
1181 newType != save->attrtype[attr]) {
1182 /* New size is larger. Need to flush existing vertices and get
1183 * an enlarged vertex format.
1185 upgrade_vertex(ctx, attr, sz);
1187 else if (sz < save->active_sz[attr]) {
1189 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1191 /* New size is equal or smaller - just need to fill in some
1194 for (i = sz; i <= save->attrsz[attr]; i++)
1195 save->attrptr[attr][i - 1] = id[i - 1];
1198 save->active_sz[attr] = sz;
1203 * Reset the current size of all vertex attributes to the default
1204 * value of 0. This signals that we haven't yet seen any per-vertex
1205 * commands such as glNormal3f() or glTexCoord2f().
1208 reset_vertex(struct gl_context *ctx)
1210 struct vbo_save_context *save = &vbo_context(ctx)->save;
1212 while (save->enabled) {
1213 const int i = u_bit_scan64(&save->enabled);
1214 assert(save->attrsz[i]);
1215 save->attrsz[i] = 0;
1216 save->active_sz[i] = 0;
1219 save->vertex_size = 0;
1224 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1225 * It depends on a few things, including whether we're inside or outside
1229 is_vertex_position(const struct gl_context *ctx, GLuint index)
1231 return (index == 0 &&
1232 _mesa_attr_zero_aliases_vertex(ctx) &&
1233 _mesa_inside_dlist_begin_end(ctx));
1238 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1241 /* Only one size for each attribute may be active at once. Eg. if
1242 * Color3f is installed/active, then Color4f may not be, even if the
1243 * vertex actually contains 4 color coordinates. This is because the
1244 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1245 * of the chooser function when switching between Color4f and Color3f.
1247 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1249 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1250 int sz = (sizeof(C) / sizeof(GLfloat)); \
1252 if (save->active_sz[A] != N) \
1253 fixup_vertex(ctx, A, N * sz, T); \
1256 C *dest = (C *)save->attrptr[A]; \
1257 if (N>0) dest[0] = V0; \
1258 if (N>1) dest[1] = V1; \
1259 if (N>2) dest[2] = V2; \
1260 if (N>3) dest[3] = V3; \
1261 save->attrtype[A] = T; \
1267 for (i = 0; i < save->vertex_size; i++) \
1268 save->buffer_ptr[i] = save->vertex[i]; \
1270 save->buffer_ptr += save->vertex_size; \
1272 if (++save->vert_count >= save->max_vert) \
1273 wrap_filled_vertex(ctx); \
1277 #define TAG(x) _save_##x
1279 #include "vbo_attrib_tmp.h"
1283 #define MAT( ATTR, N, face, params ) \
1285 if (face != GL_BACK) \
1286 MAT_ATTR( ATTR, N, params ); /* front */ \
1287 if (face != GL_FRONT) \
1288 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1293 * Save a glMaterial call found between glBegin/End.
1294 * glMaterial calls outside Begin/End are handled in dlist.c.
1296 static void GLAPIENTRY
1297 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1299 GET_CURRENT_CONTEXT(ctx);
1301 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1302 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1308 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1311 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1314 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1317 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1320 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1321 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1324 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1327 case GL_COLOR_INDEXES:
1328 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1330 case GL_AMBIENT_AND_DIFFUSE:
1331 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1332 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1335 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1341 /* Cope with EvalCoord/CallList called within a begin/end object:
1342 * -- Flush current buffer
1343 * -- Fallback to opcodes for the rest of the begin/end object.
1346 dlist_fallback(struct gl_context *ctx)
1348 struct vbo_save_context *save = &vbo_context(ctx)->save;
1350 if (save->vert_count || save->prim_count) {
1351 if (save->prim_count > 0) {
1352 /* Close off in-progress primitive. */
1353 GLint i = save->prim_count - 1;
1354 save->prims[i].count = save->vert_count - save->prims[i].start;
1357 /* Need to replay this display list with loopback,
1358 * unfortunately, otherwise this primitive won't be handled
1361 save->dangling_attr_ref = GL_TRUE;
1363 compile_vertex_list(ctx);
1366 copy_to_current(ctx);
1368 reset_counters(ctx);
1369 if (save->out_of_memory) {
1370 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1373 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1375 ctx->Driver.SaveNeedFlush = GL_FALSE;
1379 static void GLAPIENTRY
1380 _save_EvalCoord1f(GLfloat u)
1382 GET_CURRENT_CONTEXT(ctx);
1383 dlist_fallback(ctx);
1384 CALL_EvalCoord1f(ctx->Save, (u));
1387 static void GLAPIENTRY
1388 _save_EvalCoord1fv(const GLfloat * v)
1390 GET_CURRENT_CONTEXT(ctx);
1391 dlist_fallback(ctx);
1392 CALL_EvalCoord1fv(ctx->Save, (v));
1395 static void GLAPIENTRY
1396 _save_EvalCoord2f(GLfloat u, GLfloat v)
1398 GET_CURRENT_CONTEXT(ctx);
1399 dlist_fallback(ctx);
1400 CALL_EvalCoord2f(ctx->Save, (u, v));
1403 static void GLAPIENTRY
1404 _save_EvalCoord2fv(const GLfloat * v)
1406 GET_CURRENT_CONTEXT(ctx);
1407 dlist_fallback(ctx);
1408 CALL_EvalCoord2fv(ctx->Save, (v));
1411 static void GLAPIENTRY
1412 _save_EvalPoint1(GLint i)
1414 GET_CURRENT_CONTEXT(ctx);
1415 dlist_fallback(ctx);
1416 CALL_EvalPoint1(ctx->Save, (i));
1419 static void GLAPIENTRY
1420 _save_EvalPoint2(GLint i, GLint j)
1422 GET_CURRENT_CONTEXT(ctx);
1423 dlist_fallback(ctx);
1424 CALL_EvalPoint2(ctx->Save, (i, j));
1427 static void GLAPIENTRY
1428 _save_CallList(GLuint l)
1430 GET_CURRENT_CONTEXT(ctx);
1431 dlist_fallback(ctx);
1432 CALL_CallList(ctx->Save, (l));
1435 static void GLAPIENTRY
1436 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1438 GET_CURRENT_CONTEXT(ctx);
1439 dlist_fallback(ctx);
1440 CALL_CallLists(ctx->Save, (n, type, v));
1446 * Called when a glBegin is getting compiled into a display list.
1447 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1450 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1451 bool no_current_update)
1453 struct vbo_save_context *save = &vbo_context(ctx)->save;
1454 const GLuint i = save->prim_count++;
1456 ctx->Driver.CurrentSavePrimitive = mode;
1458 assert(i < save->prim_max);
1459 save->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1460 save->prims[i].begin = 1;
1461 save->prims[i].end = 0;
1462 save->prims[i].start = save->vert_count;
1463 save->prims[i].count = 0;
1465 save->no_current_update = no_current_update;
1467 if (save->out_of_memory) {
1468 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1471 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1474 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1475 ctx->Driver.SaveNeedFlush = GL_TRUE;
1479 static void GLAPIENTRY
1482 GET_CURRENT_CONTEXT(ctx);
1483 struct vbo_save_context *save = &vbo_context(ctx)->save;
1484 const GLint i = save->prim_count - 1;
1486 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1487 save->prims[i].end = 1;
1488 save->prims[i].count = (save->vert_count - save->prims[i].start);
1490 if (i == (GLint) save->prim_max - 1) {
1491 compile_vertex_list(ctx);
1492 assert(save->copied.nr == 0);
1495 /* Swap out this vertex format while outside begin/end. Any color,
1496 * etc. received between here and the next begin will be compiled
1499 if (save->out_of_memory) {
1500 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1503 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1508 static void GLAPIENTRY
1509 _save_Begin(GLenum mode)
1511 GET_CURRENT_CONTEXT(ctx);
1513 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1517 static void GLAPIENTRY
1518 _save_PrimitiveRestartNV(void)
1520 GET_CURRENT_CONTEXT(ctx);
1521 struct vbo_save_context *save = &vbo_context(ctx)->save;
1523 if (save->prim_count == 0) {
1524 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1527 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1528 "glPrimitiveRestartNV called outside glBegin/End");
1530 /* get current primitive mode */
1531 GLenum curPrim = save->prims[save->prim_count - 1].mode;
1532 bool no_current_update = save->no_current_update;
1534 /* restart primitive */
1535 CALL_End(ctx->CurrentServerDispatch, ());
1536 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1541 /* Unlike the functions above, these are to be hooked into the vtxfmt
1542 * maintained in ctx->ListState, active when the list is known or
1543 * suspected to be outside any begin/end primitive.
1544 * Note: OBE = Outside Begin/End
1546 static void GLAPIENTRY
1547 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1549 GET_CURRENT_CONTEXT(ctx);
1550 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1552 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1553 CALL_Vertex2f(dispatch, (x1, y1));
1554 CALL_Vertex2f(dispatch, (x2, y1));
1555 CALL_Vertex2f(dispatch, (x2, y2));
1556 CALL_Vertex2f(dispatch, (x1, y2));
1557 CALL_End(dispatch, ());
1561 static void GLAPIENTRY
1562 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1564 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1567 static void GLAPIENTRY
1568 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1570 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1573 static void GLAPIENTRY
1574 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1576 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1579 static void GLAPIENTRY
1580 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1582 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1585 static void GLAPIENTRY
1586 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1588 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1591 static void GLAPIENTRY
1592 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1594 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1597 static void GLAPIENTRY
1598 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1600 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1604 _ensure_draws_fits_in_storage(struct gl_context *ctx, int primcount, int vertcount)
1606 struct vbo_save_context *save = &vbo_context(ctx)->save;
1608 bool realloc_prim = save->prim_count + primcount > save->prim_max;
1609 bool realloc_vert = save->vertex_size && (save->vert_count + vertcount >= save->max_vert);
1611 if (realloc_prim || realloc_vert) {
1612 if (save->vert_count || save->prim_count) {
1613 /* TODO: this really isn't needed. We should realloc only the CPU-side memory. */
1614 compile_vertex_list(ctx);
1616 realloc_storage(ctx, realloc_prim ? primcount : -1, realloc_vert ? vertcount : -1);
1617 reset_counters(ctx);
1618 assert(save->prim_max);
1623 static void GLAPIENTRY
1624 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1626 GET_CURRENT_CONTEXT(ctx);
1627 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1628 struct vbo_save_context *save = &vbo_context(ctx)->save;
1631 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1632 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1636 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1640 if (save->out_of_memory)
1643 _ensure_draws_fits_in_storage(ctx, 1, count);
1645 /* Make sure to process any VBO binding changes */
1646 _mesa_update_state(ctx);
1648 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1650 vbo_save_NotifyBegin(ctx, mode, true);
1652 for (i = 0; i < count; i++)
1653 _mesa_array_element(ctx, start + i);
1654 CALL_End(ctx->CurrentServerDispatch, ());
1656 _mesa_vao_unmap_arrays(ctx, vao);
1660 static void GLAPIENTRY
1661 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1662 const GLsizei *count, GLsizei primcount)
1664 GET_CURRENT_CONTEXT(ctx);
1667 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1668 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1672 if (primcount < 0) {
1673 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1674 "glMultiDrawArrays(primcount<0)");
1678 unsigned vertcount = 0;
1679 for (i = 0; i < primcount; i++) {
1681 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1682 "glMultiDrawArrays(count[i]<0)");
1685 vertcount += count[i];
1688 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1690 for (i = 0; i < primcount; i++) {
1692 _save_OBE_DrawArrays(mode, first[i], count[i]);
1699 array_element(struct gl_context *ctx,
1700 GLint basevertex, GLuint elt, unsigned index_size_shift)
1702 /* Section 10.3.5 Primitive Restart:
1704 * When one of the *BaseVertex drawing commands specified in section 10.5
1705 * is used, the primitive restart comparison occurs before the basevertex
1706 * offset is added to the array index.
1708 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1709 * then we call PrimitiveRestartNV and return.
1711 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1712 elt == ctx->Array._RestartIndex[index_size_shift]) {
1713 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1717 _mesa_array_element(ctx, basevertex + elt);
1721 /* Could do better by copying the arrays and element list intact and
1722 * then emitting an indexed prim at runtime.
1724 static void GLAPIENTRY
1725 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1726 const GLvoid * indices, GLint basevertex)
1728 GET_CURRENT_CONTEXT(ctx);
1729 struct vbo_save_context *save = &vbo_context(ctx)->save;
1730 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1731 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1734 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1735 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1739 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1742 if (type != GL_UNSIGNED_BYTE &&
1743 type != GL_UNSIGNED_SHORT &&
1744 type != GL_UNSIGNED_INT) {
1745 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1749 if (save->out_of_memory)
1752 _ensure_draws_fits_in_storage(ctx, 1, count);
1754 /* Make sure to process any VBO binding changes */
1755 _mesa_update_state(ctx);
1757 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1761 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1763 vbo_save_NotifyBegin(ctx, mode, true);
1766 case GL_UNSIGNED_BYTE:
1767 for (i = 0; i < count; i++)
1768 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1770 case GL_UNSIGNED_SHORT:
1771 for (i = 0; i < count; i++)
1772 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1774 case GL_UNSIGNED_INT:
1775 for (i = 0; i < count; i++)
1776 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1779 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1783 CALL_End(ctx->CurrentServerDispatch, ());
1785 _mesa_vao_unmap(ctx, vao);
1788 static void GLAPIENTRY
1789 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1790 const GLvoid * indices)
1792 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1796 static void GLAPIENTRY
1797 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1798 GLsizei count, GLenum type,
1799 const GLvoid * indices)
1801 GET_CURRENT_CONTEXT(ctx);
1802 struct vbo_save_context *save = &vbo_context(ctx)->save;
1804 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1805 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1809 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1810 "glDrawRangeElements(count<0)");
1813 if (type != GL_UNSIGNED_BYTE &&
1814 type != GL_UNSIGNED_SHORT &&
1815 type != GL_UNSIGNED_INT) {
1816 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1820 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1821 "glDrawRangeElements(end < start)");
1825 if (save->out_of_memory)
1828 _save_OBE_DrawElements(mode, count, type, indices);
1832 static void GLAPIENTRY
1833 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1834 const GLvoid * const *indices, GLsizei primcount)
1836 GET_CURRENT_CONTEXT(ctx);
1837 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1841 for (i = 0; i < primcount; i++) {
1842 vertcount += count[i];
1844 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1846 for (i = 0; i < primcount; i++) {
1848 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1854 static void GLAPIENTRY
1855 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1857 const GLvoid * const *indices,
1859 const GLint *basevertex)
1861 GET_CURRENT_CONTEXT(ctx);
1862 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1866 for (i = 0; i < primcount; i++) {
1867 vertcount += count[i];
1869 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1871 for (i = 0; i < primcount; i++) {
1873 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1882 vtxfmt_init(struct gl_context *ctx)
1884 struct vbo_save_context *save = &vbo_context(ctx)->save;
1885 GLvertexformat *vfmt = &save->vtxfmt;
1887 #define NAME_AE(x) _ae_##x
1888 #define NAME_CALLLIST(x) _save_##x
1889 #define NAME(x) _save_##x
1890 #define NAME_ES(x) _save_##x##ARB
1892 #include "vbo_init_tmp.h"
1897 * Initialize the dispatch table with the VBO functions for display
1901 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1902 struct _glapi_table *exec)
1904 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1905 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1906 SET_DrawElements(exec, _save_OBE_DrawElements);
1907 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1908 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1909 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1910 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1911 SET_Rectf(exec, _save_OBE_Rectf);
1912 SET_Rectd(exec, _save_OBE_Rectd);
1913 SET_Rectdv(exec, _save_OBE_Rectdv);
1914 SET_Rectfv(exec, _save_OBE_Rectfv);
1915 SET_Recti(exec, _save_OBE_Recti);
1916 SET_Rectiv(exec, _save_OBE_Rectiv);
1917 SET_Rects(exec, _save_OBE_Rects);
1918 SET_Rectsv(exec, _save_OBE_Rectsv);
1920 /* Note: other glDraw functins aren't compiled into display lists */
1926 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1928 struct vbo_save_context *save = &vbo_context(ctx)->save;
1930 /* Noop when we are actually active:
1932 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1935 if (save->vert_count || save->prim_count)
1936 compile_vertex_list(ctx);
1938 copy_to_current(ctx);
1940 reset_counters(ctx);
1941 ctx->Driver.SaveNeedFlush = GL_FALSE;
1946 * Called from glNewList when we're starting to compile a display list.
1949 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1951 struct vbo_save_context *save = &vbo_context(ctx)->save;
1956 if (!save->prim_store)
1957 save->prim_store = alloc_prim_store(0);
1959 if (!save->vertex_store)
1960 save->vertex_store = alloc_vertex_store(ctx, 0);
1962 save->buffer_ptr = save->vertex_store->buffer_in_ram + save->vertex_store->used;
1965 reset_counters(ctx);
1966 ctx->Driver.SaveNeedFlush = GL_FALSE;
1971 * Called from glEndList when we're finished compiling a display list.
1974 vbo_save_EndList(struct gl_context *ctx)
1976 struct vbo_save_context *save = &vbo_context(ctx)->save;
1978 /* EndList called inside a (saved) Begin/End pair?
1980 if (_mesa_inside_dlist_begin_end(ctx)) {
1981 if (save->prim_count > 0) {
1982 GLint i = save->prim_count - 1;
1983 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1984 save->prims[i].end = 0;
1985 save->prims[i].count = save->vert_count - save->prims[i].start;
1988 /* Make sure this vertex list gets replayed by the "loopback"
1991 save->dangling_attr_ref = GL_TRUE;
1992 vbo_save_SaveFlushVertices(ctx);
1994 /* Swap out this vertex format while outside begin/end. Any color,
1995 * etc. received between here and the next begin will be compiled
1998 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
2001 assert(save->vertex_size == 0);
2005 * Called during context creation/init.
2008 current_init(struct gl_context *ctx)
2010 struct vbo_save_context *save = &vbo_context(ctx)->save;
2013 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
2014 const GLuint j = i - VBO_ATTRIB_POS;
2015 assert(j < VERT_ATTRIB_MAX);
2016 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
2017 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2020 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2021 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2022 assert(j < MAT_ATTRIB_MAX);
2023 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2024 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2030 * Initialize the display list compiler. Called during context creation.
2033 vbo_save_api_init(struct vbo_save_context *save)
2035 struct gl_context *ctx = gl_context_from_vbo_save(save);
2039 _mesa_noop_vtxfmt_init(ctx, &save->vtxfmt_noop);