1 /**************************************************************************
3 Copyright 2002-2008 VMware, Inc.
7 Permission is hereby granted, free of charge, to any person obtaining a
8 copy of this software and associated documentation files (the "Software"),
9 to deal in the Software without restriction, including without limitation
10 on the rights to use, copy, modify, merge, publish, distribute, sub
11 license, and/or sell copies of the Software, and to permit persons to whom
12 the Software is furnished to do so, subject to the following conditions:
14 The above copyright notice and this permission notice (including the next
15 paragraph) shall be included in all copies or substantial portions of the
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
35 /* Display list compiler attempts to store lists of vertices with the
36 * same vertex layout. Additionally it attempts to minimize the need
37 * for execute-time fixup of these vertex lists, allowing them to be
40 * There are still some circumstances where this can be thwarted, for
41 * example by building a list that consists of one very long primitive
42 * (eg Begin(Triangles), 1000 vertices, End), and calling that list
43 * from inside a different begin/end object (Begin(Lines), CallList,
46 * In that case the code will have to replay the list as individual
47 * commands through the Exec dispatch table, or fix up the copied
48 * vertices at execute-time.
50 * The other case where fixup is required is when a vertex attribute
51 * is introduced in the middle of a primitive. Eg:
53 * TexCoord1f() Vertex2f()
54 * TexCoord1f() Color3f() Vertex2f()
57 * If the current value of Color isn't known at compile-time, this
58 * primitive will require fixup.
61 * The list compiler currently doesn't attempt to compile lists
62 * containing EvalCoord or EvalPoint commands. On encountering one of
63 * these, compilation falls back to opcodes.
65 * This could be improved to fallback only when a mix of EvalCoord and
66 * Vertex commands are issued within a single primitive.
70 #include "main/glheader.h"
71 #include "main/arrayobj.h"
72 #include "main/bufferobj.h"
73 #include "main/context.h"
74 #include "main/dlist.h"
75 #include "main/enums.h"
76 #include "main/eval.h"
77 #include "main/macros.h"
78 #include "main/draw_validate.h"
79 #include "main/api_arrayelt.h"
80 #include "main/vtxfmt.h"
81 #include "main/dispatch.h"
82 #include "main/state.h"
83 #include "main/varray.h"
84 #include "util/bitscan.h"
85 #include "util/u_memory.h"
86 #include "util/hash_table.h"
88 #include "gallium/include/pipe/p_state.h"
91 #include "vbo_private.h"
98 /* An interesting VBO number/name to help with debugging */
99 #define VBO_BUF_ID 12345
101 static void GLAPIENTRY
102 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params);
104 static void GLAPIENTRY
105 _save_EvalCoord1f(GLfloat u);
107 static void GLAPIENTRY
108 _save_EvalCoord2f(GLfloat u, GLfloat v);
111 * NOTE: Old 'parity' issue is gone, but copying can still be
112 * wrong-footed on replay.
115 copy_vertices(struct gl_context *ctx,
116 const struct vbo_save_vertex_list *node,
117 const fi_type * src_buffer)
119 struct vbo_save_context *save = &vbo_context(ctx)->save;
120 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
121 GLuint sz = save->vertex_size;
122 const fi_type *src = src_buffer + prim->start * sz;
123 fi_type *dst = save->copied.buffer;
128 return vbo_copy_vertices(ctx, prim->mode, prim->start, &prim->count,
129 prim->begin, sz, true, dst, src);
133 static struct vbo_save_vertex_store *
134 alloc_vertex_store(struct gl_context *ctx, int vertex_count)
136 struct vbo_save_context *save = &vbo_context(ctx)->save;
137 struct vbo_save_vertex_store *vertex_store =
138 CALLOC_STRUCT(vbo_save_vertex_store);
140 int size = MAX2(vertex_count * save->vertex_size, VBO_SAVE_BUFFER_SIZE);
142 /* obj->Name needs to be non-zero, but won't ever be examined more
143 * closely than that. In particular these buffers won't be entered
144 * into the hash and can never be confused with ones visible to the
145 * user. Perhaps there could be a special number for internal
148 vertex_store->buffer_in_ram_size = size * sizeof(GLfloat);
149 vertex_store->buffer_in_ram = malloc(vertex_store->buffer_in_ram_size);
150 save->out_of_memory = vertex_store->buffer_in_ram == NULL;
152 if (save->out_of_memory) {
153 _mesa_error(ctx, GL_OUT_OF_MEMORY, "internal VBO allocation");
154 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
157 vertex_store->used = 0;
164 free_vertex_store(struct gl_context *ctx,
165 struct vbo_save_vertex_store *vertex_store)
167 free(vertex_store->buffer_in_ram);
172 static struct vbo_save_primitive_store *
173 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
176 store = CALLOC_STRUCT(vbo_save_primitive_store);
177 uint32_t old_size = store->size;
178 store->size = MAX3(store->size, prim_count, VBO_SAVE_PRIM_SIZE);
179 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
180 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
187 reset_counters(struct gl_context *ctx)
189 struct vbo_save_context *save = &vbo_context(ctx)->save;
191 save->buffer_map = save->vertex_store->buffer_in_ram + save->vertex_store->used;
193 if (save->vertex_size)
194 save->max_vert = (save->vertex_store->buffer_in_ram_size / sizeof(float) - save->vertex_store->used) /
199 save->prim_store->used = 0;
200 save->vert_count = 0;
201 save->dangling_attr_ref = GL_FALSE;
205 * For a list of prims, try merging prims that can just be extensions of the
209 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
213 struct _mesa_prim *prev_prim = prim_list;
215 for (i = 1; i < *prim_count; i++) {
216 struct _mesa_prim *this_prim = prim_list + i;
218 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
220 if (vbo_merge_draws(ctx, true,
221 prev_prim->mode, this_prim->mode,
222 prev_prim->start, this_prim->start,
223 &prev_prim->count, this_prim->count,
224 prev_prim->basevertex, this_prim->basevertex,
226 this_prim->begin, this_prim->end)) {
227 /* We've found a prim that just extend the previous one. Tack it
228 * onto the previous one, and let this primitive struct get dropped.
233 /* If any previous primitives have been dropped, then we need to copy
234 * this later one into the next available slot.
237 if (prev_prim != this_prim)
238 *prev_prim = *this_prim;
241 *prim_count = prev_prim - prim_list + 1;
246 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
247 * don't have to worry about handling the _mesa_prim::begin/end flags.
248 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
251 convert_line_loop_to_strip(struct vbo_save_context *save,
252 struct vbo_save_vertex_list *node)
254 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
256 assert(prim->mode == GL_LINE_LOOP);
259 /* Copy the 0th vertex to end of the buffer and extend the
260 * vertex count by one to finish the line loop.
262 const GLuint sz = save->vertex_size;
264 const fi_type *src = save->buffer_map + prim->start * sz;
266 fi_type *dst = save->buffer_map + (prim->start + prim->count) * sz;
268 memcpy(dst, src, sz * sizeof(float));
271 node->cold->vertex_count++;
273 save->vertex_store->used += sz;
277 /* Drawing the second or later section of a long line loop.
278 * Skip the 0th vertex.
284 prim->mode = GL_LINE_STRIP;
288 /* Compare the present vao if it has the same setup. */
290 compare_vao(gl_vertex_processing_mode mode,
291 const struct gl_vertex_array_object *vao,
292 const struct gl_buffer_object *bo, GLintptr buffer_offset,
293 GLuint stride, GLbitfield64 vao_enabled,
294 const GLubyte size[VBO_ATTRIB_MAX],
295 const GLenum16 type[VBO_ATTRIB_MAX],
296 const GLuint offset[VBO_ATTRIB_MAX])
301 /* If the enabled arrays are not the same we are not equal. */
302 if (vao_enabled != vao->Enabled)
305 /* Check the buffer binding at 0 */
306 if (vao->BufferBinding[0].BufferObj != bo)
308 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
309 if (vao->BufferBinding[0].Stride != stride)
311 assert(vao->BufferBinding[0].InstanceDivisor == 0);
313 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
314 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
316 /* Now check the enabled arrays */
317 GLbitfield mask = vao_enabled;
319 const int attr = u_bit_scan(&mask);
320 const unsigned char vbo_attr = vao_to_vbo_map[attr];
321 const GLenum16 tp = type[vbo_attr];
322 const GLintptr off = offset[vbo_attr] + buffer_offset;
323 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
324 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
326 if (attrib->Format.Type != tp)
328 if (attrib->Format.Size != size[vbo_attr])
330 assert(attrib->Format.Format == GL_RGBA);
331 assert(attrib->Format.Normalized == GL_FALSE);
332 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
333 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
334 assert(attrib->BufferBindingIndex == 0);
341 /* Create or reuse the vao for the vertex processing mode. */
343 update_vao(struct gl_context *ctx,
344 gl_vertex_processing_mode mode,
345 struct gl_vertex_array_object **vao,
346 struct gl_buffer_object *bo, GLintptr buffer_offset,
347 GLuint stride, GLbitfield64 vbo_enabled,
348 const GLubyte size[VBO_ATTRIB_MAX],
349 const GLenum16 type[VBO_ATTRIB_MAX],
350 const GLuint offset[VBO_ATTRIB_MAX])
352 /* Compute the bitmasks of vao_enabled arrays */
353 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
356 * Check if we can possibly reuse the exisiting one.
357 * In the long term we should reset them when something changes.
359 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
360 vao_enabled, size, type, offset))
363 /* The initial refcount is 1 */
364 _mesa_reference_vao(ctx, vao, NULL);
365 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
368 * assert(stride <= ctx->Const.MaxVertexAttribStride);
369 * MaxVertexAttribStride is not set for drivers that does not
370 * expose GL 44 or GLES 31.
373 /* Bind the buffer object at binding point 0 */
374 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
377 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
378 * Note that the position/generic0 aliasing is done in the VAO.
380 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
381 /* Now set the enable arrays */
382 GLbitfield mask = vao_enabled;
384 const int vao_attr = u_bit_scan(&mask);
385 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
386 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
388 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
389 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
390 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
392 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
393 assert(vao_enabled == (*vao)->Enabled);
394 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
396 /* Finalize and freeze the VAO */
397 _mesa_set_vao_immutable(ctx, *vao);
402 realloc_storage(struct gl_context *ctx, int prim_count, int vertex_count)
404 struct vbo_save_context *save = &vbo_context(ctx)->save;
405 if (vertex_count >= 0) {
406 /* Release old reference:
408 free_vertex_store(ctx, save->vertex_store);
409 save->vertex_store = NULL;
410 /* When we have a new vbo, we will for sure need a new vao */
411 for (gl_vertex_processing_mode vpm = 0; vpm < VP_MODE_MAX; ++vpm)
412 _mesa_reference_vao(ctx, &save->VAO[vpm], NULL);
414 /* Allocate and map new store:
416 save->vertex_store = alloc_vertex_store(ctx, vertex_count);
420 save->prim_store = realloc_prim_store(save->prim_store, prim_count);
424 unsigned vertex_size;
425 fi_type *vertex_attributes;
428 static uint32_t _hash_vertex_key(const void *key)
430 struct vertex_key *k = (struct vertex_key*)key;
431 unsigned sz = k->vertex_size;
433 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
436 static bool _compare_vertex_key(const void *key1, const void *key2)
438 struct vertex_key *k1 = (struct vertex_key*)key1;
439 struct vertex_key *k2 = (struct vertex_key*)key2;
440 /* All the compared vertices are going to be drawn with the same VAO,
441 * so we can compare the attributes. */
442 assert (k1->vertex_size == k2->vertex_size);
443 return memcmp(k1->vertex_attributes,
444 k2->vertex_attributes,
445 k1->vertex_size * sizeof(float)) == 0;
448 static void _free_entry(struct hash_entry *entry)
450 free((void*)entry->key);
453 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
454 * of an existing vertex, return the original index instead.
457 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
458 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
460 /* If vertex deduplication is disabled return the original index. */
464 fi_type *vert = save->buffer_map + save->vertex_size * index;
466 struct vertex_key *key = malloc(sizeof(struct vertex_key));
467 key->vertex_size = save->vertex_size;
468 key->vertex_attributes = vert;
470 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
473 /* We found an existing vertex with the same hash, return its index. */
474 return (uintptr_t) entry->data;
476 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
477 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
478 * starting at index 0.
480 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
481 *max_index = MAX2(n, *max_index);
483 memcpy(&new_buffer[save->vertex_size * n],
485 save->vertex_size * sizeof(fi_type));
487 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
489 /* The index buffer is shared between list compilations, so add the base index to get
498 * Insert the active immediate struct onto the display list currently
502 compile_vertex_list(struct gl_context *ctx)
504 struct vbo_save_context *save = &vbo_context(ctx)->save;
505 struct vbo_save_vertex_list *node;
507 /* Allocate space for this structure in the display list currently
510 node = (struct vbo_save_vertex_list *)
511 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
516 memset(node, 0, sizeof(struct vbo_save_vertex_list));
517 node->cold = calloc(1, sizeof(*node->cold));
519 /* Make sure the pointer is aligned to the size of a pointer */
520 assert((GLintptr) node % sizeof(void *) == 0);
522 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
524 node->cold->vertex_count = save->vert_count;
525 node->cold->wrap_count = save->copied.nr;
526 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
527 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
528 node->cold->ib.obj = NULL;
529 node->cold->prim_count = save->prim_store->used;
531 if (save->no_current_update) {
532 node->cold->current_data = NULL;
535 GLuint current_size = save->vertex_size - save->attrsz[0];
536 node->cold->current_data = NULL;
539 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
540 if (node->cold->current_data) {
541 const char *buffer = (const char *)save->buffer_map;
542 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
543 unsigned vertex_offset = 0;
545 if (node->cold->vertex_count)
546 vertex_offset = (node->cold->vertex_count - 1) * stride;
548 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
549 current_size * sizeof(GLfloat));
551 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
556 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
558 if (save->dangling_attr_ref)
559 ctx->ListState.Current.UseLoopback = true;
561 /* Copy duplicated vertices
563 save->copied.nr = copy_vertices(ctx, node, save->buffer_map);
565 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
566 convert_line_loop_to_strip(save, node);
569 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
571 GLintptr buffer_offset = 0;
572 GLuint start_offset = 0;
574 /* Create an index buffer. */
575 node->cold->min_index = node->cold->max_index = 0;
576 if (save->vert_count == 0 || node->cold->prim_count == 0)
579 /* We won't modify node->prims, so use a const alias to avoid unintended
581 const struct _mesa_prim *original_prims = node->cold->prims;
583 int end = original_prims[node->cold->prim_count - 1].start +
584 original_prims[node->cold->prim_count - 1].count;
585 int total_vert_count = end - original_prims[0].start;
587 node->cold->min_index = node->cold->prims[0].start;
588 node->cold->max_index = end - 1;
590 /* Estimate for the worst case: all prims are line strips (the +1 is because
591 * wrap_buffers may call use but the last primitive may not be complete) */
592 int max_indices_count = MAX2(total_vert_count * 2 - (node->cold->prim_count * 2) + 1,
595 int size = max_indices_count * sizeof(uint32_t);
596 uint32_t* indices = (uint32_t*) malloc(size);
597 struct _mesa_prim *merged_prims = NULL;
600 struct hash_table *vertex_to_index = NULL;
601 fi_type *temp_vertices_buffer = NULL;
603 /* The loopback replay code doesn't use the index buffer, so we can't
604 * dedup vertices in this case.
606 if (!ctx->ListState.Current.UseLoopback) {
607 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
608 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
611 uint32_t max_index = 0;
613 int last_valid_prim = -1;
614 /* Construct indices array. */
615 for (unsigned i = 0; i < node->cold->prim_count; i++) {
616 assert(original_prims[i].basevertex == 0);
617 GLubyte mode = original_prims[i].mode;
619 int vertex_count = original_prims[i].count;
624 /* Line strips may get converted to lines */
625 if (mode == GL_LINE_STRIP)
628 /* If 2 consecutive prims use the same mode => merge them. */
629 bool merge_prims = last_valid_prim >= 0 &&
630 mode == merged_prims[last_valid_prim].mode &&
631 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
632 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
635 /* To be able to merge consecutive triangle strips we need to insert
636 * a degenerate triangle.
639 mode == GL_TRIANGLE_STRIP) {
640 /* Insert a degenerate triangle */
641 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
642 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
644 indices[idx] = indices[idx - 1];
645 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
646 temp_vertices_buffer, &max_index);
648 merged_prims[last_valid_prim].count += 2;
651 /* Add another index to preserve winding order */
652 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
653 temp_vertices_buffer, &max_index);
654 merged_prims[last_valid_prim].count++;
660 /* Convert line strips to lines if it'll allow if the previous
661 * prim mode is GL_LINES (so merge_prims is true) or if the next
662 * primitive mode is GL_LINES or GL_LINE_LOOP.
664 if (original_prims[i].mode == GL_LINE_STRIP &&
666 (i < node->cold->prim_count - 1 &&
667 (original_prims[i + 1].mode == GL_LINE_STRIP ||
668 original_prims[i + 1].mode == GL_LINES)))) {
669 for (unsigned j = 0; j < vertex_count; j++) {
670 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
671 temp_vertices_buffer, &max_index);
672 /* Repeat all but the first/last indices. */
673 if (j && j != vertex_count - 1) {
674 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
675 temp_vertices_buffer, &max_index);
679 /* We didn't convert to LINES, so restore the original mode */
680 mode = original_prims[i].mode;
682 for (unsigned j = 0; j < vertex_count; j++) {
683 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
684 temp_vertices_buffer, &max_index);
689 /* Update vertex count. */
690 merged_prims[last_valid_prim].count += idx - start;
692 /* Keep this primitive */
693 last_valid_prim += 1;
694 assert(last_valid_prim <= i);
695 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
696 merged_prims[last_valid_prim] = original_prims[i];
697 merged_prims[last_valid_prim].start = start;
698 merged_prims[last_valid_prim].count = idx - start;
700 merged_prims[last_valid_prim].mode = mode;
703 assert(idx > 0 && idx <= max_indices_count);
705 unsigned merged_prim_count = last_valid_prim + 1;
706 node->cold->ib.ptr = NULL;
707 node->cold->ib.count = idx;
708 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
710 /* How many bytes do we need to store the indices and the vertices */
711 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
712 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
713 total_vert_count * save->vertex_size * sizeof(fi_type);
715 const GLintptr old_offset = save->VAO[0] ?
716 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
717 if (old_offset != save->current_bo_bytes_used && stride > 0) {
718 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
719 while (offset_diff > 0 &&
720 save->current_bo_bytes_used < save->current_bo->Size &&
721 offset_diff % stride != 0) {
722 save->current_bo_bytes_used++;
723 offset_diff = save->current_bo_bytes_used - old_offset;
726 buffer_offset = save->current_bo_bytes_used;
728 /* Can we reuse the previous bo or should we allocate a new one? */
729 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
730 if (total_bytes_needed > available_bytes) {
731 if (save->current_bo)
732 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
733 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
734 bool success = ctx->Driver.BufferData(ctx,
735 GL_ELEMENT_ARRAY_BUFFER_ARB,
736 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t)),
738 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
741 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
742 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
744 save->current_bo_bytes_used = 0;
745 available_bytes = save->current_bo->Size;
749 assert(old_offset <= buffer_offset);
750 const GLintptr offset_diff = buffer_offset - old_offset;
751 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
752 /* The vertex size is an exact multiple of the buffer offset.
753 * This means that we can use zero-based vertex attribute pointers
754 * and specify the start of the primitive with the _mesa_prim::start
755 * field. This results in issuing several draw calls with identical
756 * vertex attribute information. This can result in fewer state
757 * changes in drivers. In particular, the Gallium CSO module will
758 * filter out redundant vertex buffer changes.
760 /* We cannot immediately update the primitives as some methods below
761 * still need the uncorrected start vertices
763 start_offset = offset_diff/stride;
764 assert(old_offset == buffer_offset - offset_diff);
765 buffer_offset = old_offset;
768 /* Correct the primitive starts, we can only do this here as copy_vertices
769 * and convert_line_loop_to_strip above consume the uncorrected starts.
770 * On the other hand the _vbo_loopback_vertex_list call below needs the
771 * primitives to be corrected already.
773 for (unsigned i = 0; i < node->cold->prim_count; i++) {
774 node->cold->prims[i].start += start_offset;
776 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
777 * to apply this transformation to all indices and max_index.
779 for (unsigned i = 0; i < idx; i++)
780 indices[i] += start_offset;
781 max_index += start_offset;
784 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
786 /* Upload the vertices first (see buffer_offset) */
787 ctx->Driver.BufferSubData(ctx,
788 save->current_bo_bytes_used,
789 total_vert_count * save->vertex_size * sizeof(fi_type),
790 vertex_to_index ? temp_vertices_buffer : save->buffer_map,
792 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
794 if (vertex_to_index) {
795 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
796 free(temp_vertices_buffer);
799 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
800 * primitive (not the indices themselves). */
801 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
802 int indices_offset = save->current_bo_bytes_used / 4;
803 for (int i = 0; i < merged_prim_count; i++) {
804 merged_prims[i].start += indices_offset;
807 /* Then upload the indices. */
808 if (node->cold->ib.obj) {
809 ctx->Driver.BufferSubData(ctx,
810 save->current_bo_bytes_used,
811 idx * sizeof(uint32_t),
814 save->current_bo_bytes_used += idx * sizeof(uint32_t);
816 node->cold->vertex_count = 0;
817 node->cold->prim_count = 0;
820 /* Prepare for DrawGallium */
821 memset(&node->merged.info, 0, sizeof(struct pipe_draw_info));
822 /* The other info fields will be updated in vbo_save_playback_vertex_list */
823 node->merged.info.index_size = 4;
824 node->merged.info.instance_count = 1;
825 node->merged.info.index.gl_bo = node->cold->ib.obj;
826 if (merged_prim_count == 1) {
827 node->merged.info.mode = merged_prims[0].mode;
828 node->merged.start_count.start = merged_prims[0].start;
829 node->merged.start_count.count = merged_prims[0].count;
830 node->merged.start_count.index_bias = 0;
831 node->merged.mode = NULL;
833 node->merged.mode = malloc(merged_prim_count * sizeof(unsigned char));
834 node->merged.start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
835 for (unsigned i = 0; i < merged_prim_count; i++) {
836 node->merged.start_counts[i].start = merged_prims[i].start;
837 node->merged.start_counts[i].count = merged_prims[i].count;
838 node->merged.start_counts[i].index_bias = 0;
839 node->merged.mode[i] = merged_prims[i].mode;
842 node->merged.num_draws = merged_prim_count;
843 if (node->merged.num_draws > 1) {
844 bool same_mode = true;
845 for (unsigned i = 1; i < node->merged.num_draws && same_mode; i++) {
846 same_mode = node->merged.mode[i] == node->merged.mode[0];
849 /* All primitives use the same mode, so we can simplify a bit */
850 node->merged.info.mode = node->merged.mode[0];
851 free(node->merged.mode);
852 node->merged.mode = NULL;
861 if (!save->current_bo) {
862 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
863 bool success = ctx->Driver.BufferData(ctx,
864 GL_ELEMENT_ARRAY_BUFFER_ARB,
865 VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t),
867 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
871 GLuint offsets[VBO_ATTRIB_MAX];
872 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
874 offset += save->attrsz[i] * sizeof(GLfloat);
876 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
877 * Note that this may reuse the previous one of possible.
879 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
880 /* create or reuse the vao */
881 update_vao(ctx, vpm, &save->VAO[vpm],
882 save->current_bo, buffer_offset, stride,
883 save->enabled, save->attrsz, save->attrtype, offsets);
884 /* Reference the vao in the dlist */
885 node->VAO[vpm] = NULL;
886 _mesa_reference_vao(ctx, &node->VAO[vpm], save->VAO[vpm]);
890 /* Deal with GL_COMPILE_AND_EXECUTE:
892 if (ctx->ExecuteFlag) {
893 struct _glapi_table *dispatch = GET_DISPATCH();
895 _glapi_set_dispatch(ctx->Exec);
897 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
898 * use buffer_in_ram instead of current_bo which contains all vertices instead
899 * of the deduplicated vertices only in the !UseLoopback case.
901 * The problem is that the VAO offset is based on current_bo's layout,
902 * so we have to use a temp value.
904 struct gl_vertex_array_object *vao = node->VAO[VP_MODE_SHADER];
905 GLintptr original = vao->BufferBinding[0].Offset;
906 if (!ctx->ListState.Current.UseLoopback) {
907 GLintptr new_offset = (save->buffer_map - save->vertex_store->buffer_in_ram) *
909 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
910 new_offset -= start_offset * stride;
911 vao->BufferBinding[0].Offset = new_offset;
913 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
914 vao->BufferBinding[0].Offset = original;
916 _glapi_set_dispatch(dispatch);
919 /* Decide whether the storage structs are full, or can be used for
920 * the next vertex lists as well.
922 if (save->vertex_store->used >
923 save->vertex_store->buffer_in_ram_size / sizeof(float) - 16 * (save->vertex_size + 4)) {
924 realloc_storage(ctx, -1, 0);
927 /* Reset our structures for the next run of vertices:
934 * This is called when we fill a vertex buffer before we hit a glEnd().
936 * TODO -- If no new vertices have been stored, don't bother saving it.
939 wrap_buffers(struct gl_context *ctx)
941 struct vbo_save_context *save = &vbo_context(ctx)->save;
942 GLint i = save->prim_store->used - 1;
945 assert(i < (GLint) save->prim_store->size);
948 /* Close off in-progress primitive.
950 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
951 mode = save->prim_store->prims[i].mode;
953 /* store the copied vertices, and allocate a new list.
955 compile_vertex_list(ctx);
957 /* Restart interrupted primitive
959 save->prim_store->prims[0].mode = mode;
960 save->prim_store->prims[0].begin = 0;
961 save->prim_store->prims[0].end = 0;
962 save->prim_store->prims[0].start = 0;
963 save->prim_store->prims[0].count = 0;
964 save->prim_store->used = 1;
969 * Called only when buffers are wrapped as the result of filling the
970 * vertex_store struct.
973 wrap_filled_vertex(struct gl_context *ctx)
975 struct vbo_save_context *save = &vbo_context(ctx)->save;
976 unsigned numComponents;
978 /* Emit a glEnd to close off the last vertex list.
982 /* Copy stored stored vertices to start of new list.
984 assert(save->max_vert - save->vert_count > save->copied.nr);
986 numComponents = save->copied.nr * save->vertex_size;
988 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + save->vertex_store->used;
991 numComponents * sizeof(fi_type));
992 save->vert_count += save->copied.nr;
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->buffer_map;
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 reset_counters(ctx);
1351 if (save->out_of_memory) {
1352 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1355 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1357 ctx->Driver.SaveNeedFlush = GL_FALSE;
1361 static void GLAPIENTRY
1362 _save_EvalCoord1f(GLfloat u)
1364 GET_CURRENT_CONTEXT(ctx);
1365 dlist_fallback(ctx);
1366 CALL_EvalCoord1f(ctx->Save, (u));
1369 static void GLAPIENTRY
1370 _save_EvalCoord1fv(const GLfloat * v)
1372 GET_CURRENT_CONTEXT(ctx);
1373 dlist_fallback(ctx);
1374 CALL_EvalCoord1fv(ctx->Save, (v));
1377 static void GLAPIENTRY
1378 _save_EvalCoord2f(GLfloat u, GLfloat v)
1380 GET_CURRENT_CONTEXT(ctx);
1381 dlist_fallback(ctx);
1382 CALL_EvalCoord2f(ctx->Save, (u, v));
1385 static void GLAPIENTRY
1386 _save_EvalCoord2fv(const GLfloat * v)
1388 GET_CURRENT_CONTEXT(ctx);
1389 dlist_fallback(ctx);
1390 CALL_EvalCoord2fv(ctx->Save, (v));
1393 static void GLAPIENTRY
1394 _save_EvalPoint1(GLint i)
1396 GET_CURRENT_CONTEXT(ctx);
1397 dlist_fallback(ctx);
1398 CALL_EvalPoint1(ctx->Save, (i));
1401 static void GLAPIENTRY
1402 _save_EvalPoint2(GLint i, GLint j)
1404 GET_CURRENT_CONTEXT(ctx);
1405 dlist_fallback(ctx);
1406 CALL_EvalPoint2(ctx->Save, (i, j));
1409 static void GLAPIENTRY
1410 _save_CallList(GLuint l)
1412 GET_CURRENT_CONTEXT(ctx);
1413 dlist_fallback(ctx);
1414 CALL_CallList(ctx->Save, (l));
1417 static void GLAPIENTRY
1418 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1420 GET_CURRENT_CONTEXT(ctx);
1421 dlist_fallback(ctx);
1422 CALL_CallLists(ctx->Save, (n, type, v));
1428 * Called when a glBegin is getting compiled into a display list.
1429 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1432 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1433 bool no_current_update)
1435 struct vbo_save_context *save = &vbo_context(ctx)->save;
1436 const GLuint i = save->prim_store->used++;
1438 ctx->Driver.CurrentSavePrimitive = mode;
1440 assert(i < save->prim_store->size);
1441 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1442 save->prim_store->prims[i].begin = 1;
1443 save->prim_store->prims[i].end = 0;
1444 save->prim_store->prims[i].start = save->vert_count;
1445 save->prim_store->prims[i].count = 0;
1447 save->no_current_update = no_current_update;
1449 if (save->out_of_memory) {
1450 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1453 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1456 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1457 ctx->Driver.SaveNeedFlush = GL_TRUE;
1461 static void GLAPIENTRY
1464 GET_CURRENT_CONTEXT(ctx);
1465 struct vbo_save_context *save = &vbo_context(ctx)->save;
1466 const GLint i = save->prim_store->used - 1;
1468 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1469 save->prim_store->prims[i].end = 1;
1470 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
1472 if (i == (GLint) save->prim_store->size - 1) {
1473 compile_vertex_list(ctx);
1474 assert(save->copied.nr == 0);
1477 /* Swap out this vertex format while outside begin/end. Any color,
1478 * etc. received between here and the next begin will be compiled
1481 if (save->out_of_memory) {
1482 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1485 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1490 static void GLAPIENTRY
1491 _save_Begin(GLenum mode)
1493 GET_CURRENT_CONTEXT(ctx);
1495 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1499 static void GLAPIENTRY
1500 _save_PrimitiveRestartNV(void)
1502 GET_CURRENT_CONTEXT(ctx);
1503 struct vbo_save_context *save = &vbo_context(ctx)->save;
1505 if (save->prim_store->used == 0) {
1506 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1509 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1510 "glPrimitiveRestartNV called outside glBegin/End");
1512 /* get current primitive mode */
1513 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1514 bool no_current_update = save->no_current_update;
1516 /* restart primitive */
1517 CALL_End(ctx->CurrentServerDispatch, ());
1518 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1523 /* Unlike the functions above, these are to be hooked into the vtxfmt
1524 * maintained in ctx->ListState, active when the list is known or
1525 * suspected to be outside any begin/end primitive.
1526 * Note: OBE = Outside Begin/End
1528 static void GLAPIENTRY
1529 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1531 GET_CURRENT_CONTEXT(ctx);
1532 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1534 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1535 CALL_Vertex2f(dispatch, (x1, y1));
1536 CALL_Vertex2f(dispatch, (x2, y1));
1537 CALL_Vertex2f(dispatch, (x2, y2));
1538 CALL_Vertex2f(dispatch, (x1, y2));
1539 CALL_End(dispatch, ());
1543 static void GLAPIENTRY
1544 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1546 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1549 static void GLAPIENTRY
1550 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1552 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1555 static void GLAPIENTRY
1556 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1558 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1561 static void GLAPIENTRY
1562 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1564 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1567 static void GLAPIENTRY
1568 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1570 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1573 static void GLAPIENTRY
1574 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1576 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1579 static void GLAPIENTRY
1580 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1582 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1586 _ensure_draws_fits_in_storage(struct gl_context *ctx, int primcount, int vertcount)
1588 struct vbo_save_context *save = &vbo_context(ctx)->save;
1590 bool realloc_prim = save->prim_store->used + primcount > save->prim_store->size;
1591 bool realloc_vert = save->vertex_size && (save->vert_count + vertcount >= save->max_vert);
1593 if (realloc_prim || realloc_vert) {
1594 if (realloc_vert && (save->vert_count || save->prim_store->used)) {
1595 /* TODO: this really isn't needed. We should realloc only the CPU-side memory. */
1596 compile_vertex_list(ctx);
1598 realloc_storage(ctx, realloc_prim ? primcount : -1, realloc_vert ? vertcount : -1);
1599 reset_counters(ctx);
1604 static void GLAPIENTRY
1605 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1607 GET_CURRENT_CONTEXT(ctx);
1608 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1609 struct vbo_save_context *save = &vbo_context(ctx)->save;
1612 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1613 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1617 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1621 if (save->out_of_memory)
1624 _ensure_draws_fits_in_storage(ctx, 1, count);
1626 /* Make sure to process any VBO binding changes */
1627 _mesa_update_state(ctx);
1629 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1631 vbo_save_NotifyBegin(ctx, mode, true);
1633 for (i = 0; i < count; i++)
1634 _mesa_array_element(ctx, start + i);
1635 CALL_End(ctx->CurrentServerDispatch, ());
1637 _mesa_vao_unmap_arrays(ctx, vao);
1641 static void GLAPIENTRY
1642 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1643 const GLsizei *count, GLsizei primcount)
1645 GET_CURRENT_CONTEXT(ctx);
1648 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1649 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1653 if (primcount < 0) {
1654 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1655 "glMultiDrawArrays(primcount<0)");
1659 unsigned vertcount = 0;
1660 for (i = 0; i < primcount; i++) {
1662 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1663 "glMultiDrawArrays(count[i]<0)");
1666 vertcount += count[i];
1669 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1671 for (i = 0; i < primcount; i++) {
1673 _save_OBE_DrawArrays(mode, first[i], count[i]);
1680 array_element(struct gl_context *ctx,
1681 GLint basevertex, GLuint elt, unsigned index_size_shift)
1683 /* Section 10.3.5 Primitive Restart:
1685 * When one of the *BaseVertex drawing commands specified in section 10.5
1686 * is used, the primitive restart comparison occurs before the basevertex
1687 * offset is added to the array index.
1689 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1690 * then we call PrimitiveRestartNV and return.
1692 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1693 elt == ctx->Array._RestartIndex[index_size_shift]) {
1694 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1698 _mesa_array_element(ctx, basevertex + elt);
1702 /* Could do better by copying the arrays and element list intact and
1703 * then emitting an indexed prim at runtime.
1705 static void GLAPIENTRY
1706 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1707 const GLvoid * indices, GLint basevertex)
1709 GET_CURRENT_CONTEXT(ctx);
1710 struct vbo_save_context *save = &vbo_context(ctx)->save;
1711 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1712 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1715 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1716 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1720 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1723 if (type != GL_UNSIGNED_BYTE &&
1724 type != GL_UNSIGNED_SHORT &&
1725 type != GL_UNSIGNED_INT) {
1726 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1730 if (save->out_of_memory)
1733 _ensure_draws_fits_in_storage(ctx, 1, count);
1735 /* Make sure to process any VBO binding changes */
1736 _mesa_update_state(ctx);
1738 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1742 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1744 vbo_save_NotifyBegin(ctx, mode, true);
1747 case GL_UNSIGNED_BYTE:
1748 for (i = 0; i < count; i++)
1749 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1751 case GL_UNSIGNED_SHORT:
1752 for (i = 0; i < count; i++)
1753 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1755 case GL_UNSIGNED_INT:
1756 for (i = 0; i < count; i++)
1757 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1760 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1764 CALL_End(ctx->CurrentServerDispatch, ());
1766 _mesa_vao_unmap(ctx, vao);
1769 static void GLAPIENTRY
1770 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1771 const GLvoid * indices)
1773 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1777 static void GLAPIENTRY
1778 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1779 GLsizei count, GLenum type,
1780 const GLvoid * indices)
1782 GET_CURRENT_CONTEXT(ctx);
1783 struct vbo_save_context *save = &vbo_context(ctx)->save;
1785 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1786 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1790 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1791 "glDrawRangeElements(count<0)");
1794 if (type != GL_UNSIGNED_BYTE &&
1795 type != GL_UNSIGNED_SHORT &&
1796 type != GL_UNSIGNED_INT) {
1797 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1801 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1802 "glDrawRangeElements(end < start)");
1806 if (save->out_of_memory)
1809 _save_OBE_DrawElements(mode, count, type, indices);
1813 static void GLAPIENTRY
1814 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1815 const GLvoid * const *indices, GLsizei primcount)
1817 GET_CURRENT_CONTEXT(ctx);
1818 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1822 for (i = 0; i < primcount; i++) {
1823 vertcount += count[i];
1825 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1827 for (i = 0; i < primcount; i++) {
1829 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1835 static void GLAPIENTRY
1836 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1838 const GLvoid * const *indices,
1840 const GLint *basevertex)
1842 GET_CURRENT_CONTEXT(ctx);
1843 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1847 for (i = 0; i < primcount; i++) {
1848 vertcount += count[i];
1850 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1852 for (i = 0; i < primcount; i++) {
1854 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1863 vtxfmt_init(struct gl_context *ctx)
1865 struct vbo_save_context *save = &vbo_context(ctx)->save;
1866 GLvertexformat *vfmt = &save->vtxfmt;
1868 #define NAME_AE(x) _ae_##x
1869 #define NAME_CALLLIST(x) _save_##x
1870 #define NAME(x) _save_##x
1871 #define NAME_ES(x) _save_##x##ARB
1873 #include "vbo_init_tmp.h"
1878 * Initialize the dispatch table with the VBO functions for display
1882 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1883 struct _glapi_table *exec)
1885 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1886 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1887 SET_DrawElements(exec, _save_OBE_DrawElements);
1888 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1889 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1890 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1891 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1892 SET_Rectf(exec, _save_OBE_Rectf);
1893 SET_Rectd(exec, _save_OBE_Rectd);
1894 SET_Rectdv(exec, _save_OBE_Rectdv);
1895 SET_Rectfv(exec, _save_OBE_Rectfv);
1896 SET_Recti(exec, _save_OBE_Recti);
1897 SET_Rectiv(exec, _save_OBE_Rectiv);
1898 SET_Rects(exec, _save_OBE_Rects);
1899 SET_Rectsv(exec, _save_OBE_Rectsv);
1901 /* Note: other glDraw functins aren't compiled into display lists */
1907 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1909 struct vbo_save_context *save = &vbo_context(ctx)->save;
1911 /* Noop when we are actually active:
1913 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1916 if (save->vert_count || save->prim_store->used)
1917 compile_vertex_list(ctx);
1919 copy_to_current(ctx);
1921 reset_counters(ctx);
1922 ctx->Driver.SaveNeedFlush = GL_FALSE;
1927 * Called from glNewList when we're starting to compile a display list.
1930 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1932 struct vbo_save_context *save = &vbo_context(ctx)->save;
1937 if (!save->prim_store)
1938 save->prim_store = realloc_prim_store(NULL, 8);
1940 if (!save->vertex_store)
1941 save->vertex_store = alloc_vertex_store(ctx, 0);
1944 reset_counters(ctx);
1945 ctx->Driver.SaveNeedFlush = GL_FALSE;
1950 * Called from glEndList when we're finished compiling a display list.
1953 vbo_save_EndList(struct gl_context *ctx)
1955 struct vbo_save_context *save = &vbo_context(ctx)->save;
1957 /* EndList called inside a (saved) Begin/End pair?
1959 if (_mesa_inside_dlist_begin_end(ctx)) {
1960 if (save->prim_store->used > 0) {
1961 GLint i = save->prim_store->used - 1;
1962 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1963 save->prim_store->prims[i].end = 0;
1964 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1967 /* Make sure this vertex list gets replayed by the "loopback"
1970 save->dangling_attr_ref = GL_TRUE;
1971 vbo_save_SaveFlushVertices(ctx);
1973 /* Swap out this vertex format while outside begin/end. Any color,
1974 * etc. received between here and the next begin will be compiled
1977 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1980 assert(save->vertex_size == 0);
1984 * Called during context creation/init.
1987 current_init(struct gl_context *ctx)
1989 struct vbo_save_context *save = &vbo_context(ctx)->save;
1992 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
1993 const GLuint j = i - VBO_ATTRIB_POS;
1994 assert(j < VERT_ATTRIB_MAX);
1995 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
1996 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
1999 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2000 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2001 assert(j < MAT_ATTRIB_MAX);
2002 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2003 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2009 * Initialize the display list compiler. Called during context creation.
2012 vbo_save_api_init(struct vbo_save_context *save)
2014 struct gl_context *ctx = gl_context_from_vbo_save(save);
2018 _mesa_noop_vtxfmt_init(ctx, &save->vtxfmt_noop);