1 /**************************************************************************
3 Copyright 2002-2008 VMware, Inc.
7 Permission is hereby granted, free of charge, to any person obtaining a
8 copy of this software and associated documentation files (the "Software"),
9 to deal in the Software without restriction, including without limitation
10 on the rights to use, copy, modify, merge, publish, distribute, sub
11 license, and/or sell copies of the Software, and to permit persons to whom
12 the Software is furnished to do so, subject to the following conditions:
14 The above copyright notice and this permission notice (including the next
15 paragraph) shall be included in all copies or substantial portions of the
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Keith Whitwell <keithw@vmware.com>
35 /* Display list compiler attempts to store lists of vertices with the
36 * same vertex layout. Additionally it attempts to minimize the need
37 * for execute-time fixup of these vertex lists, allowing them to be
40 * There are still some circumstances where this can be thwarted, for
41 * example by building a list that consists of one very long primitive
42 * (eg Begin(Triangles), 1000 vertices, End), and calling that list
43 * from inside a different begin/end object (Begin(Lines), CallList,
46 * In that case the code will have to replay the list as individual
47 * commands through the Exec dispatch table, or fix up the copied
48 * vertices at execute-time.
50 * The other case where fixup is required is when a vertex attribute
51 * is introduced in the middle of a primitive. Eg:
53 * TexCoord1f() Vertex2f()
54 * TexCoord1f() Color3f() Vertex2f()
57 * If the current value of Color isn't known at compile-time, this
58 * primitive will require fixup.
61 * The list compiler currently doesn't attempt to compile lists
62 * containing EvalCoord or EvalPoint commands. On encountering one of
63 * these, compilation falls back to opcodes.
65 * This could be improved to fallback only when a mix of EvalCoord and
66 * Vertex commands are issued within a single primitive.
68 * The compilation process works as follows. All vertex attributes
69 * except position are copied to vbo_save_context::attrptr (see ATTR_UNION).
70 * 'attrptr' are pointers to vbo_save_context::vertex ordered according to the enabled
71 * attributes (se upgrade_vertex).
72 * When the position attribute is received, all the attributes are then
73 * copied to the vertex_store (see the end of ATTR_UNION).
74 * The vertex_store is simply an extensible float array.
75 * When the vertex list needs to be compiled (see compile_vertex_list),
76 * several transformations are performed:
77 * - some primitives are merged together (eg: two consecutive GL_TRIANGLES
78 * with 3 vertices can be merged in a single GL_TRIANGLES with 6 vertices).
79 * - an index buffer is built.
80 * - identical vertices are detected and only one is kept.
81 * At the end of this transformation, the index buffer and the vertex buffer
82 * are uploaded in vRAM in the same buffer object.
83 * This buffer object is shared between multiple display list to allow
84 * draw calls merging later.
86 * The layout of this buffer for two display lists is:
87 * V0A0|V0A1|V1A0|V1A1|P0I0|P0I1|V0A0V0A1V0A2|V1A1V1A1V1A2|...
89 * - VxAy: vertex x, attributes y
90 * - PxIy: draw x, index y
92 * To allow draw call merging, display list must use the same VAO, including
93 * the same Offset in the buffer object. To achieve this, the start values of
94 * the primitive are shifted and the indices adjusted (see offset_diff and
95 * start_offset in compile_vertex_list).
97 * Display list using the loopback code (see vbo_save_playback_vertex_list_loopback),
98 * can't be drawn with an index buffer so this transformation is disabled
103 #include "main/glheader.h"
104 #include "main/arrayobj.h"
105 #include "main/bufferobj.h"
106 #include "main/context.h"
107 #include "main/dlist.h"
108 #include "main/enums.h"
109 #include "main/eval.h"
110 #include "main/macros.h"
111 #include "main/draw_validate.h"
112 #include "main/api_arrayelt.h"
113 #include "main/dispatch.h"
114 #include "main/state.h"
115 #include "main/varray.h"
116 #include "util/bitscan.h"
117 #include "util/u_memory.h"
118 #include "util/hash_table.h"
119 #include "util/indices/u_indices.h"
121 #include "gallium/include/pipe/p_state.h"
123 #include "vbo_private.h"
124 #include "api_exec_decl.h"
125 #include "api_save.h"
131 /* An interesting VBO number/name to help with debugging */
132 #define VBO_BUF_ID 12345
134 static void GLAPIENTRY
135 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params);
137 static void GLAPIENTRY
138 _save_EvalCoord1f(GLfloat u);
140 static void GLAPIENTRY
141 _save_EvalCoord2f(GLfloat u, GLfloat v);
144 * NOTE: Old 'parity' issue is gone, but copying can still be
145 * wrong-footed on replay.
148 copy_vertices(struct gl_context *ctx,
149 const struct vbo_save_vertex_list *node,
150 const fi_type * src_buffer)
152 struct vbo_save_context *save = &vbo_context(ctx)->save;
153 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
154 GLuint sz = save->vertex_size;
156 if (prim->end || !prim->count || !sz)
159 const fi_type *src = src_buffer + prim->start * sz;
160 assert(save->copied.buffer == NULL);
161 save->copied.buffer = malloc(sizeof(fi_type) * sz * prim->count);
163 unsigned r = vbo_copy_vertices(ctx, prim->mode, prim->start, &prim->count,
164 prim->begin, sz, true, save->copied.buffer, src);
166 free(save->copied.buffer);
167 save->copied.buffer = NULL;
173 static struct vbo_save_primitive_store *
174 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
177 store = CALLOC_STRUCT(vbo_save_primitive_store);
179 uint32_t old_size = store->size;
180 store->size = prim_count;
181 assert (old_size < store->size);
182 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
183 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
190 reset_counters(struct gl_context *ctx)
192 struct vbo_save_context *save = &vbo_context(ctx)->save;
194 save->vertex_store->used = 0;
195 save->prim_store->used = 0;
196 save->dangling_attr_ref = GL_FALSE;
200 * For a list of prims, try merging prims that can just be extensions of the
204 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
208 struct _mesa_prim *prev_prim = prim_list;
210 for (i = 1; i < *prim_count; i++) {
211 struct _mesa_prim *this_prim = prim_list + i;
213 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
215 if (vbo_merge_draws(ctx, true,
216 prev_prim->mode, this_prim->mode,
217 prev_prim->start, this_prim->start,
218 &prev_prim->count, this_prim->count,
219 prev_prim->basevertex, this_prim->basevertex,
221 this_prim->begin, this_prim->end)) {
222 /* We've found a prim that just extend the previous one. Tack it
223 * onto the previous one, and let this primitive struct get dropped.
228 /* If any previous primitives have been dropped, then we need to copy
229 * this later one into the next available slot.
232 if (prev_prim != this_prim)
233 *prev_prim = *this_prim;
236 *prim_count = prev_prim - prim_list + 1;
241 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
242 * don't have to worry about handling the _mesa_prim::begin/end flags.
243 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
246 convert_line_loop_to_strip(struct vbo_save_context *save,
247 struct vbo_save_vertex_list *node)
249 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
251 assert(prim->mode == GL_LINE_LOOP);
254 /* Copy the 0th vertex to end of the buffer and extend the
255 * vertex count by one to finish the line loop.
257 const GLuint sz = save->vertex_size;
259 const fi_type *src = save->vertex_store->buffer_in_ram + prim->start * sz;
261 fi_type *dst = save->vertex_store->buffer_in_ram + (prim->start + prim->count) * sz;
263 memcpy(dst, src, sz * sizeof(float));
266 node->cold->vertex_count++;
267 save->vertex_store->used += sz;
271 /* Drawing the second or later section of a long line loop.
272 * Skip the 0th vertex.
278 prim->mode = GL_LINE_STRIP;
282 /* Compare the present vao if it has the same setup. */
284 compare_vao(gl_vertex_processing_mode mode,
285 const struct gl_vertex_array_object *vao,
286 const struct gl_buffer_object *bo, GLintptr buffer_offset,
287 GLuint stride, GLbitfield64 vao_enabled,
288 const GLubyte size[VBO_ATTRIB_MAX],
289 const GLenum16 type[VBO_ATTRIB_MAX],
290 const GLuint offset[VBO_ATTRIB_MAX])
295 /* If the enabled arrays are not the same we are not equal. */
296 if (vao_enabled != vao->Enabled)
299 /* Check the buffer binding at 0 */
300 if (vao->BufferBinding[0].BufferObj != bo)
302 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
303 if (vao->BufferBinding[0].Stride != stride)
305 assert(vao->BufferBinding[0].InstanceDivisor == 0);
307 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
308 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
310 /* Now check the enabled arrays */
311 GLbitfield mask = vao_enabled;
313 const int attr = u_bit_scan(&mask);
314 const unsigned char vbo_attr = vao_to_vbo_map[attr];
315 const GLenum16 tp = type[vbo_attr];
316 const GLintptr off = offset[vbo_attr] + buffer_offset;
317 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
318 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
320 if (attrib->Format.Type != tp)
322 if (attrib->Format.Size != size[vbo_attr])
324 assert(attrib->Format.Format == GL_RGBA);
325 assert(attrib->Format.Normalized == GL_FALSE);
326 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
327 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
328 assert(attrib->BufferBindingIndex == 0);
335 /* Create or reuse the vao for the vertex processing mode. */
337 update_vao(struct gl_context *ctx,
338 gl_vertex_processing_mode mode,
339 struct gl_vertex_array_object **vao,
340 struct gl_buffer_object *bo, GLintptr buffer_offset,
341 GLuint stride, GLbitfield64 vbo_enabled,
342 const GLubyte size[VBO_ATTRIB_MAX],
343 const GLenum16 type[VBO_ATTRIB_MAX],
344 const GLuint offset[VBO_ATTRIB_MAX])
346 /* Compute the bitmasks of vao_enabled arrays */
347 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
350 * Check if we can possibly reuse the exisiting one.
351 * In the long term we should reset them when something changes.
353 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
354 vao_enabled, size, type, offset))
357 /* The initial refcount is 1 */
358 _mesa_reference_vao(ctx, vao, NULL);
359 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
362 * assert(stride <= ctx->Const.MaxVertexAttribStride);
363 * MaxVertexAttribStride is not set for drivers that does not
364 * expose GL 44 or GLES 31.
367 /* Bind the buffer object at binding point 0 */
368 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
371 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
372 * Note that the position/generic0 aliasing is done in the VAO.
374 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
375 /* Now set the enable arrays */
376 GLbitfield mask = vao_enabled;
378 const int vao_attr = u_bit_scan(&mask);
379 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
380 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
382 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
383 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
384 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
386 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
387 assert(vao_enabled == (*vao)->Enabled);
388 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
390 /* Finalize and freeze the VAO */
391 _mesa_set_vao_immutable(ctx, *vao);
394 static void wrap_filled_vertex(struct gl_context *ctx);
396 /* Grow the vertex storage to accomodate for vertex_count new vertices */
398 grow_vertex_storage(struct gl_context *ctx, int vertex_count)
400 struct vbo_save_context *save = &vbo_context(ctx)->save;
401 assert (save->vertex_store);
403 int new_size = (save->vertex_store->used +
404 vertex_count * save->vertex_size) * sizeof(GLfloat);
406 /* Limit how much memory we allocate. */
407 if (save->prim_store->used > 0 &&
409 new_size > VBO_SAVE_BUFFER_SIZE) {
410 wrap_filled_vertex(ctx);
411 new_size = VBO_SAVE_BUFFER_SIZE;
414 if (new_size > save->vertex_store->buffer_in_ram_size) {
415 save->vertex_store->buffer_in_ram_size = new_size;
416 save->vertex_store->buffer_in_ram = realloc(save->vertex_store->buffer_in_ram,
417 save->vertex_store->buffer_in_ram_size);
418 if (save->vertex_store->buffer_in_ram == NULL)
419 save->out_of_memory = true;
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->vertex_store->buffer_in_ram + 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 get_vertex_count(struct vbo_save_context *save)
500 if (!save->vertex_size)
502 return save->vertex_store->used / save->vertex_size;
507 * Insert the active immediate struct onto the display list currently
511 compile_vertex_list(struct gl_context *ctx)
513 struct vbo_save_context *save = &vbo_context(ctx)->save;
514 struct vbo_save_vertex_list *node;
516 /* Allocate space for this structure in the display list currently
519 node = (struct vbo_save_vertex_list *)
520 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
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 = get_vertex_count(save);
533 node->cold->wrap_count = save->copied.nr;
534 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
535 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
536 node->cold->ib.obj = NULL;
537 node->cold->prim_count = save->prim_store->used;
539 if (save->no_current_update) {
540 node->cold->current_data = NULL;
543 GLuint current_size = save->vertex_size - save->attrsz[0];
544 node->cold->current_data = NULL;
547 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
548 if (node->cold->current_data) {
549 const char *buffer = (const char *)save->vertex_store->buffer_in_ram;
550 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
551 unsigned vertex_offset = 0;
553 if (node->cold->vertex_count)
554 vertex_offset = (node->cold->vertex_count - 1) * stride;
556 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
557 current_size * sizeof(GLfloat));
559 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
560 save->out_of_memory = true;
565 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
567 if (save->dangling_attr_ref)
568 ctx->ListState.Current.UseLoopback = true;
570 /* Copy duplicated vertices
572 save->copied.nr = copy_vertices(ctx, node, save->vertex_store->buffer_in_ram);
574 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
575 convert_line_loop_to_strip(save, node);
578 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
580 GLintptr buffer_offset = 0;
581 GLuint start_offset = 0;
583 /* Create an index buffer. */
584 node->cold->min_index = node->cold->max_index = 0;
585 if (node->cold->vertex_count == 0 || node->cold->prim_count == 0)
588 /* We won't modify node->prims, so use a const alias to avoid unintended
590 const struct _mesa_prim *original_prims = node->cold->prims;
592 int end = original_prims[node->cold->prim_count - 1].start +
593 original_prims[node->cold->prim_count - 1].count;
594 int total_vert_count = end - original_prims[0].start;
596 node->cold->min_index = node->cold->prims[0].start;
597 node->cold->max_index = end - 1;
599 /* converting primitive types may result in many more indices */
600 bool all_prims_supported = (ctx->Const.DriverSupportedPrimMask & BITFIELD_MASK(PIPE_PRIM_MAX)) == BITFIELD_MASK(PIPE_PRIM_MAX);
601 int max_index_count = total_vert_count * (all_prims_supported ? 2 : 3);
603 int size = max_index_count * sizeof(uint32_t);
604 uint32_t* indices = (uint32_t*) malloc(size);
605 void *tmp_indices = all_prims_supported ? NULL : malloc(size);
606 struct _mesa_prim *merged_prims = NULL;
609 struct hash_table *vertex_to_index = NULL;
610 fi_type *temp_vertices_buffer = NULL;
612 /* The loopback replay code doesn't use the index buffer, so we can't
613 * dedup vertices in this case.
615 if (!ctx->ListState.Current.UseLoopback) {
616 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
617 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
620 uint32_t max_index = 0;
622 int last_valid_prim = -1;
623 /* Construct indices array. */
624 for (unsigned i = 0; i < node->cold->prim_count; i++) {
625 assert(original_prims[i].basevertex == 0);
626 GLubyte mode = original_prims[i].mode;
627 bool converted_prim = false;
630 int vertex_count = original_prims[i].count;
635 /* Line strips may get converted to lines */
636 if (mode == GL_LINE_STRIP)
639 if (!(ctx->Const.DriverSupportedPrimMask & BITFIELD_BIT(mode))) {
641 u_generate_func trans_func;
642 enum pipe_prim_type pmode = (enum pipe_prim_type)mode;
643 u_index_generator(ctx->Const.DriverSupportedPrimMask,
644 pmode, original_prims[i].start, vertex_count,
646 &pmode, &index_size, &new_count,
649 trans_func(original_prims[i].start, new_count, tmp_indices);
650 vertex_count = new_count;
651 mode = (GLubyte)pmode;
652 converted_prim = true;
655 /* If 2 consecutive prims use the same mode => merge them. */
656 bool merge_prims = last_valid_prim >= 0 &&
657 mode == merged_prims[last_valid_prim].mode &&
658 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
659 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
662 /* index generation uses uint16_t if the index count is small enough */
663 #define CAST_INDEX(BASE, SIZE, IDX) ((SIZE == 2 ? (uint32_t)(((uint16_t*)BASE)[IDX]) : ((uint32_t*)BASE)[IDX]))
664 /* To be able to merge consecutive triangle strips we need to insert
665 * a degenerate triangle.
668 mode == GL_TRIANGLE_STRIP) {
669 /* Insert a degenerate triangle */
670 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
671 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
673 indices[idx] = indices[idx - 1];
674 indices[idx + 1] = add_vertex(save, vertex_to_index,
675 converted_prim ? CAST_INDEX(tmp_indices, index_size, 0) : original_prims[i].start,
676 temp_vertices_buffer, &max_index);
678 merged_prims[last_valid_prim].count += 2;
681 /* Add another index to preserve winding order */
682 indices[idx++] = add_vertex(save, vertex_to_index,
683 converted_prim ? CAST_INDEX(tmp_indices, index_size, 0) : original_prims[i].start,
684 temp_vertices_buffer, &max_index);
685 merged_prims[last_valid_prim].count++;
691 /* Convert line strips to lines if it'll allow if the previous
692 * prim mode is GL_LINES (so merge_prims is true) or if the next
693 * primitive mode is GL_LINES or GL_LINE_LOOP.
695 if (original_prims[i].mode == GL_LINE_STRIP &&
697 (i < node->cold->prim_count - 1 &&
698 (original_prims[i + 1].mode == GL_LINE_STRIP ||
699 original_prims[i + 1].mode == GL_LINES)))) {
700 for (unsigned j = 0; j < vertex_count; j++) {
701 indices[idx++] = add_vertex(save, vertex_to_index,
702 converted_prim ? CAST_INDEX(tmp_indices, index_size, j) : original_prims[i].start + j,
703 temp_vertices_buffer, &max_index);
704 /* Repeat all but the first/last indices. */
705 if (j && j != vertex_count - 1) {
706 indices[idx++] = add_vertex(save, vertex_to_index,
707 converted_prim ? CAST_INDEX(tmp_indices, index_size, j) : original_prims[i].start + j,
708 temp_vertices_buffer, &max_index);
712 /* We didn't convert to LINES, so restore the original mode */
714 mode = original_prims[i].mode;
716 for (unsigned j = 0; j < vertex_count; j++) {
717 indices[idx++] = add_vertex(save, vertex_to_index,
718 converted_prim ? CAST_INDEX(tmp_indices, index_size, j) : original_prims[i].start + j,
719 temp_vertices_buffer, &max_index);
724 /* Update vertex count. */
725 merged_prims[last_valid_prim].count += idx - start;
727 /* Keep this primitive */
728 last_valid_prim += 1;
729 assert(last_valid_prim <= i);
730 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
731 merged_prims[last_valid_prim] = original_prims[i];
732 merged_prims[last_valid_prim].start = start;
733 merged_prims[last_valid_prim].count = idx - start;
735 merged_prims[last_valid_prim].mode = mode;
737 /* converted prims will filter incomplete primitives and may have no indices */
738 assert((idx > 0 || converted_prim) && idx <= max_index_count);
741 unsigned merged_prim_count = last_valid_prim + 1;
742 node->cold->ib.ptr = NULL;
743 node->cold->ib.count = idx;
744 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
746 /* How many bytes do we need to store the indices and the vertices */
747 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
748 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
749 total_vert_count * save->vertex_size * sizeof(fi_type);
751 const GLintptr old_offset = save->VAO[0] ?
752 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
753 if (old_offset != save->current_bo_bytes_used && stride > 0) {
754 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
755 while (offset_diff > 0 &&
756 save->current_bo_bytes_used < save->current_bo->Size &&
757 offset_diff % stride != 0) {
758 save->current_bo_bytes_used++;
759 offset_diff = save->current_bo_bytes_used - old_offset;
762 buffer_offset = save->current_bo_bytes_used;
764 /* Can we reuse the previous bo or should we allocate a new one? */
765 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
766 if (total_bytes_needed > available_bytes) {
767 if (save->current_bo)
768 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
769 save->current_bo = _mesa_bufferobj_alloc(ctx, VBO_BUF_ID + 1);
770 bool success = _mesa_bufferobj_data(ctx,
771 GL_ELEMENT_ARRAY_BUFFER_ARB,
772 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE),
774 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
775 MESA_GALLIUM_VERTEX_STATE_STORAGE,
778 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
779 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
780 save->out_of_memory = true;
782 save->current_bo_bytes_used = 0;
783 available_bytes = save->current_bo->Size;
787 assert(old_offset <= buffer_offset);
788 const GLintptr offset_diff = buffer_offset - old_offset;
789 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
790 /* The vertex size is an exact multiple of the buffer offset.
791 * This means that we can use zero-based vertex attribute pointers
792 * and specify the start of the primitive with the _mesa_prim::start
793 * field. This results in issuing several draw calls with identical
794 * vertex attribute information. This can result in fewer state
795 * changes in drivers. In particular, the Gallium CSO module will
796 * filter out redundant vertex buffer changes.
798 /* We cannot immediately update the primitives as some methods below
799 * still need the uncorrected start vertices
801 start_offset = offset_diff/stride;
802 assert(old_offset == buffer_offset - offset_diff);
803 buffer_offset = old_offset;
806 /* Correct the primitive starts, we can only do this here as copy_vertices
807 * and convert_line_loop_to_strip above consume the uncorrected starts.
808 * On the other hand the _vbo_loopback_vertex_list call below needs the
809 * primitives to be corrected already.
811 for (unsigned i = 0; i < node->cold->prim_count; i++) {
812 node->cold->prims[i].start += start_offset;
814 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
815 * to apply this transformation to all indices and max_index.
817 for (unsigned i = 0; i < idx; i++)
818 indices[i] += start_offset;
819 max_index += start_offset;
822 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
824 /* Upload the vertices first (see buffer_offset) */
825 _mesa_bufferobj_subdata(ctx,
826 save->current_bo_bytes_used,
827 total_vert_count * save->vertex_size * sizeof(fi_type),
828 vertex_to_index ? temp_vertices_buffer : save->vertex_store->buffer_in_ram,
830 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
832 if (vertex_to_index) {
833 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
834 free(temp_vertices_buffer);
837 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
838 * primitive (not the indices themselves). */
839 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
840 int indices_offset = save->current_bo_bytes_used / 4;
841 for (int i = 0; i < merged_prim_count; i++) {
842 merged_prims[i].start += indices_offset;
845 /* Then upload the indices. */
846 if (node->cold->ib.obj) {
847 _mesa_bufferobj_subdata(ctx,
848 save->current_bo_bytes_used,
849 idx * sizeof(uint32_t),
852 save->current_bo_bytes_used += idx * sizeof(uint32_t);
854 node->cold->vertex_count = 0;
855 node->cold->prim_count = 0;
858 /* Prepare for DrawGallium */
859 memset(&node->cold->info, 0, sizeof(struct pipe_draw_info));
860 /* The other info fields will be updated in vbo_save_playback_vertex_list */
861 node->cold->info.index_size = 4;
862 node->cold->info.instance_count = 1;
863 node->cold->info.index.gl_bo = node->cold->ib.obj;
864 if (merged_prim_count == 1) {
865 node->cold->info.mode = merged_prims[0].mode;
866 node->start_count.start = merged_prims[0].start;
867 node->start_count.count = merged_prims[0].count;
868 node->start_count.index_bias = 0;
871 node->modes = malloc(merged_prim_count * sizeof(unsigned char));
872 node->start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
873 for (unsigned i = 0; i < merged_prim_count; i++) {
874 node->start_counts[i].start = merged_prims[i].start;
875 node->start_counts[i].count = merged_prims[i].count;
876 node->start_counts[i].index_bias = 0;
877 node->modes[i] = merged_prims[i].mode;
880 node->num_draws = merged_prim_count;
881 if (node->num_draws > 1) {
882 bool same_mode = true;
883 for (unsigned i = 1; i < node->num_draws && same_mode; i++) {
884 same_mode = node->modes[i] == node->modes[0];
887 /* All primitives use the same mode, so we can simplify a bit */
888 node->cold->info.mode = node->modes[0];
899 node->draw_begins = node->cold->prims[0].begin;
901 if (!save->current_bo) {
902 save->current_bo = _mesa_bufferobj_alloc(ctx, VBO_BUF_ID + 1);
903 bool success = _mesa_bufferobj_data(ctx,
904 GL_ELEMENT_ARRAY_BUFFER_ARB,
905 VBO_SAVE_BUFFER_SIZE,
907 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT |
908 MESA_GALLIUM_VERTEX_STATE_STORAGE,
911 save->out_of_memory = true;
914 GLuint offsets[VBO_ATTRIB_MAX];
915 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
917 offset += save->attrsz[i] * sizeof(GLfloat);
919 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
920 * Note that this may reuse the previous one of possible.
922 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
923 /* create or reuse the vao */
924 update_vao(ctx, vpm, &save->VAO[vpm],
925 save->current_bo, buffer_offset, stride,
926 save->enabled, save->attrsz, save->attrtype, offsets);
927 /* Reference the vao in the dlist */
928 node->cold->VAO[vpm] = NULL;
929 _mesa_reference_vao(ctx, &node->cold->VAO[vpm], save->VAO[vpm]);
932 /* Prepare for DrawGalliumVertexState */
933 if (node->num_draws && ctx->Driver.DrawGalliumVertexState) {
934 for (unsigned i = 0; i < VP_MODE_MAX; i++) {
935 uint32_t enabled_attribs = _vbo_get_vao_filter(i) &
936 node->cold->VAO[i]->_EnabledWithMapMode;
939 ctx->Driver.CreateGalliumVertexState(ctx, node->cold->VAO[i],
942 node->private_refcount[i] = 0;
943 node->enabled_attribs[i] = enabled_attribs;
947 node->mode = node->cold->info.mode;
948 assert(node->cold->info.index_size == 4);
951 /* Deal with GL_COMPILE_AND_EXECUTE:
953 if (ctx->ExecuteFlag) {
954 struct _glapi_table *dispatch = GET_DISPATCH();
956 _glapi_set_dispatch(ctx->Exec);
958 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
959 * use buffer_in_ram instead of current_bo which contains all vertices instead
960 * of the deduplicated vertices only in the !UseLoopback case.
962 * The problem is that the VAO offset is based on current_bo's layout,
963 * so we have to use a temp value.
965 struct gl_vertex_array_object *vao = node->cold->VAO[VP_MODE_SHADER];
966 GLintptr original = vao->BufferBinding[0].Offset;
967 if (!ctx->ListState.Current.UseLoopback) {
968 GLintptr new_offset = 0;
969 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
970 new_offset -= start_offset * stride;
971 vao->BufferBinding[0].Offset = new_offset;
973 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
974 vao->BufferBinding[0].Offset = original;
976 _glapi_set_dispatch(dispatch);
979 /* Reset our structures for the next run of vertices:
986 * This is called when we fill a vertex buffer before we hit a glEnd().
988 * TODO -- If no new vertices have been stored, don't bother saving it.
991 wrap_buffers(struct gl_context *ctx)
993 struct vbo_save_context *save = &vbo_context(ctx)->save;
994 GLint i = save->prim_store->used - 1;
997 assert(i < (GLint) save->prim_store->size);
1000 /* Close off in-progress primitive.
1002 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1003 mode = save->prim_store->prims[i].mode;
1005 /* store the copied vertices, and allocate a new list.
1007 compile_vertex_list(ctx);
1009 /* Restart interrupted primitive
1011 save->prim_store->prims[0].mode = mode;
1012 save->prim_store->prims[0].begin = 0;
1013 save->prim_store->prims[0].end = 0;
1014 save->prim_store->prims[0].start = 0;
1015 save->prim_store->prims[0].count = 0;
1016 save->prim_store->used = 1;
1021 * Called only when buffers are wrapped as the result of filling the
1022 * vertex_store struct.
1025 wrap_filled_vertex(struct gl_context *ctx)
1027 struct vbo_save_context *save = &vbo_context(ctx)->save;
1028 unsigned numComponents;
1030 /* Emit a glEnd to close off the last vertex list.
1034 assert(save->vertex_store->used == 0 && save->vertex_store->used == 0);
1036 /* Copy stored stored vertices to start of new list.
1038 numComponents = save->copied.nr * save->vertex_size;
1040 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
1041 if (numComponents) {
1042 assert(save->copied.buffer);
1044 save->copied.buffer,
1045 numComponents * sizeof(fi_type));
1046 free(save->copied.buffer);
1047 save->copied.buffer = NULL;
1049 save->vertex_store->used = numComponents;
1054 copy_to_current(struct gl_context *ctx)
1056 struct vbo_save_context *save = &vbo_context(ctx)->save;
1057 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1060 const int i = u_bit_scan64(&enabled);
1061 assert(save->attrsz[i]);
1063 if (save->attrtype[i] == GL_DOUBLE ||
1064 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
1065 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
1067 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
1068 save->attrptr[i], save->attrtype[i]);
1074 copy_from_current(struct gl_context *ctx)
1076 struct vbo_save_context *save = &vbo_context(ctx)->save;
1077 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1080 const int i = u_bit_scan64(&enabled);
1082 switch (save->attrsz[i]) {
1084 save->attrptr[i][3] = save->current[i][3];
1087 save->attrptr[i][2] = save->current[i][2];
1090 save->attrptr[i][1] = save->current[i][1];
1093 save->attrptr[i][0] = save->current[i][0];
1096 unreachable("Unexpected vertex attribute size");
1103 * Called when we increase the size of a vertex attribute. For example,
1104 * if we've seen one or more glTexCoord2f() calls and now we get a
1105 * glTexCoord3f() call.
1106 * Flush existing data, set new attrib size, replay copied vertices.
1109 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1111 struct vbo_save_context *save = &vbo_context(ctx)->save;
1116 /* Store the current run of vertices, and emit a GL_END. Emit a
1117 * BEGIN in the new buffer.
1119 if (save->vertex_store->used)
1122 assert(save->copied.nr == 0);
1124 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1125 * when the attribute already exists in the vertex and is having
1126 * its size increased.
1128 copy_to_current(ctx);
1132 oldsz = save->attrsz[attr];
1133 save->attrsz[attr] = newsz;
1134 save->enabled |= BITFIELD64_BIT(attr);
1136 save->vertex_size += newsz - oldsz;
1138 /* Recalculate all the attrptr[] values:
1141 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1142 if (save->attrsz[i]) {
1143 save->attrptr[i] = tmp;
1144 tmp += save->attrsz[i];
1147 save->attrptr[i] = NULL; /* will not be dereferenced. */
1151 /* Copy from current to repopulate the vertex with correct values.
1153 copy_from_current(ctx);
1155 /* Replay stored vertices to translate them to new format here.
1157 * If there are copied vertices and the new (upgraded) attribute
1158 * has not been defined before, this list is somewhat degenerate,
1159 * and will need fixup at runtime.
1161 if (save->copied.nr) {
1162 assert(save->copied.buffer);
1163 const fi_type *data = save->copied.buffer;
1164 grow_vertex_storage(ctx, save->copied.nr);
1165 fi_type *dest = save->vertex_store->buffer_in_ram;
1167 /* Need to note this and fix up at runtime (or loopback):
1169 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1171 save->dangling_attr_ref = GL_TRUE;
1174 for (i = 0; i < save->copied.nr; i++) {
1175 GLbitfield64 enabled = save->enabled;
1177 const int j = u_bit_scan64(&enabled);
1178 assert(save->attrsz[j]);
1181 const fi_type *src = oldsz ? data : save->current[attr];
1182 int copy = oldsz ? oldsz : newsz;
1183 for (k = 0; k < copy; k++)
1185 for (; k < newsz; k++) {
1186 switch (save->attrtype[j]) {
1188 dest[k] = FLOAT_AS_UNION(k == 3);
1191 dest[k] = INT_AS_UNION(k == 3);
1193 case GL_UNSIGNED_INT:
1194 dest[k] = UINT_AS_UNION(k == 3);
1197 dest[k] = FLOAT_AS_UNION(k == 3);
1198 assert(!"Unexpected type in upgrade_vertex");
1205 GLint sz = save->attrsz[j];
1206 for (int k = 0; k < sz; k++)
1214 save->vertex_store->used += save->vertex_size * save->copied.nr;
1215 free(save->copied.buffer);
1216 save->copied.buffer = NULL;
1222 * This is called when the size of a vertex attribute changes.
1223 * For example, after seeing one or more glTexCoord2f() calls we
1224 * get a glTexCoord4f() or glTexCoord1f() call.
1227 fixup_vertex(struct gl_context *ctx, GLuint attr,
1228 GLuint sz, GLenum newType)
1230 struct vbo_save_context *save = &vbo_context(ctx)->save;
1232 if (sz > save->attrsz[attr] ||
1233 newType != save->attrtype[attr]) {
1234 /* New size is larger. Need to flush existing vertices and get
1235 * an enlarged vertex format.
1237 upgrade_vertex(ctx, attr, sz);
1239 else if (sz < save->active_sz[attr]) {
1241 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1243 /* New size is equal or smaller - just need to fill in some
1246 for (i = sz; i <= save->attrsz[attr]; i++)
1247 save->attrptr[attr][i - 1] = id[i - 1];
1250 save->active_sz[attr] = sz;
1252 grow_vertex_storage(ctx, 1);
1257 * Reset the current size of all vertex attributes to the default
1258 * value of 0. This signals that we haven't yet seen any per-vertex
1259 * commands such as glNormal3f() or glTexCoord2f().
1262 reset_vertex(struct gl_context *ctx)
1264 struct vbo_save_context *save = &vbo_context(ctx)->save;
1266 while (save->enabled) {
1267 const int i = u_bit_scan64(&save->enabled);
1268 assert(save->attrsz[i]);
1269 save->attrsz[i] = 0;
1270 save->active_sz[i] = 0;
1273 save->vertex_size = 0;
1278 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1279 * It depends on a few things, including whether we're inside or outside
1283 is_vertex_position(const struct gl_context *ctx, GLuint index)
1285 return (index == 0 &&
1286 _mesa_attr_zero_aliases_vertex(ctx) &&
1287 _mesa_inside_dlist_begin_end(ctx));
1292 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1295 /* Only one size for each attribute may be active at once. Eg. if
1296 * Color3f is installed/active, then Color4f may not be, even if the
1297 * vertex actually contains 4 color coordinates. This is because the
1298 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1299 * of the chooser function when switching between Color4f and Color3f.
1301 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1303 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1304 int sz = (sizeof(C) / sizeof(GLfloat)); \
1306 if (save->active_sz[A] != N) \
1307 fixup_vertex(ctx, A, N * sz, T); \
1310 C *dest = (C *)save->attrptr[A]; \
1311 if (N>0) dest[0] = V0; \
1312 if (N>1) dest[1] = V1; \
1313 if (N>2) dest[2] = V2; \
1314 if (N>3) dest[3] = V3; \
1315 save->attrtype[A] = T; \
1318 if ((A) == VBO_ATTRIB_POS) { \
1319 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + \
1320 save->vertex_store->used; \
1322 for (int i = 0; i < save->vertex_size; i++) \
1323 buffer_ptr[i] = save->vertex[i]; \
1325 save->vertex_store->used += save->vertex_size; \
1326 unsigned used_next = (save->vertex_store->used + \
1327 save->vertex_size) * sizeof(float); \
1328 if (used_next > save->vertex_store->buffer_in_ram_size) { \
1329 grow_vertex_storage(ctx, get_vertex_count(save)); \
1330 assert(used_next <= \
1331 save->vertex_store->buffer_in_ram_size); \
1336 #define TAG(x) _save_##x
1338 #include "vbo_attrib_tmp.h"
1341 #define MAT( ATTR, N, face, params ) \
1343 if (face != GL_BACK) \
1344 MAT_ATTR( ATTR, N, params ); /* front */ \
1345 if (face != GL_FRONT) \
1346 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1351 * Save a glMaterial call found between glBegin/End.
1352 * glMaterial calls outside Begin/End are handled in dlist.c.
1354 static void GLAPIENTRY
1355 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1357 GET_CURRENT_CONTEXT(ctx);
1359 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1360 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1366 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1369 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1372 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1375 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1378 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1379 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1382 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1385 case GL_COLOR_INDEXES:
1386 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1388 case GL_AMBIENT_AND_DIFFUSE:
1389 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1390 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1393 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1400 vbo_install_save_vtxfmt(struct gl_context *ctx);
1403 /* Cope with EvalCoord/CallList called within a begin/end object:
1404 * -- Flush current buffer
1405 * -- Fallback to opcodes for the rest of the begin/end object.
1408 dlist_fallback(struct gl_context *ctx)
1410 struct vbo_save_context *save = &vbo_context(ctx)->save;
1412 if (save->vertex_store->used || save->prim_store->used) {
1413 if (save->prim_store->used > 0 && save->vertex_store->used > 0) {
1414 assert(save->vertex_size);
1415 /* Close off in-progress primitive. */
1416 GLint i = save->prim_store->used - 1;
1417 save->prim_store->prims[i].count =
1418 get_vertex_count(save) -
1419 save->prim_store->prims[i].start;
1422 /* Need to replay this display list with loopback,
1423 * unfortunately, otherwise this primitive won't be handled
1426 save->dangling_attr_ref = GL_TRUE;
1428 compile_vertex_list(ctx);
1431 copy_to_current(ctx);
1433 if (save->out_of_memory) {
1434 vbo_install_save_vtxfmt_noop(ctx);
1437 _mesa_install_save_vtxfmt(ctx);
1439 ctx->Driver.SaveNeedFlush = GL_FALSE;
1443 static void GLAPIENTRY
1444 _save_EvalCoord1f(GLfloat u)
1446 GET_CURRENT_CONTEXT(ctx);
1447 dlist_fallback(ctx);
1448 CALL_EvalCoord1f(ctx->Save, (u));
1451 static void GLAPIENTRY
1452 _save_EvalCoord1fv(const GLfloat * v)
1454 GET_CURRENT_CONTEXT(ctx);
1455 dlist_fallback(ctx);
1456 CALL_EvalCoord1fv(ctx->Save, (v));
1459 static void GLAPIENTRY
1460 _save_EvalCoord2f(GLfloat u, GLfloat v)
1462 GET_CURRENT_CONTEXT(ctx);
1463 dlist_fallback(ctx);
1464 CALL_EvalCoord2f(ctx->Save, (u, v));
1467 static void GLAPIENTRY
1468 _save_EvalCoord2fv(const GLfloat * v)
1470 GET_CURRENT_CONTEXT(ctx);
1471 dlist_fallback(ctx);
1472 CALL_EvalCoord2fv(ctx->Save, (v));
1475 static void GLAPIENTRY
1476 _save_EvalPoint1(GLint i)
1478 GET_CURRENT_CONTEXT(ctx);
1479 dlist_fallback(ctx);
1480 CALL_EvalPoint1(ctx->Save, (i));
1483 static void GLAPIENTRY
1484 _save_EvalPoint2(GLint i, GLint j)
1486 GET_CURRENT_CONTEXT(ctx);
1487 dlist_fallback(ctx);
1488 CALL_EvalPoint2(ctx->Save, (i, j));
1491 static void GLAPIENTRY
1492 _save_CallList(GLuint l)
1494 GET_CURRENT_CONTEXT(ctx);
1495 dlist_fallback(ctx);
1496 CALL_CallList(ctx->Save, (l));
1499 static void GLAPIENTRY
1500 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1502 GET_CURRENT_CONTEXT(ctx);
1503 dlist_fallback(ctx);
1504 CALL_CallLists(ctx->Save, (n, type, v));
1510 * Called when a glBegin is getting compiled into a display list.
1511 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1514 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1515 bool no_current_update)
1517 struct vbo_save_context *save = &vbo_context(ctx)->save;
1518 const GLuint i = save->prim_store->used++;
1520 ctx->Driver.CurrentSavePrimitive = mode;
1522 if (!save->prim_store || i >= save->prim_store->size) {
1523 save->prim_store = realloc_prim_store(save->prim_store, i * 2);
1525 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1526 save->prim_store->prims[i].begin = 1;
1527 save->prim_store->prims[i].end = 0;
1528 save->prim_store->prims[i].start = get_vertex_count(save);
1529 save->prim_store->prims[i].count = 0;
1531 save->no_current_update = no_current_update;
1533 vbo_install_save_vtxfmt(ctx);
1535 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1536 ctx->Driver.SaveNeedFlush = GL_TRUE;
1540 static void GLAPIENTRY
1543 GET_CURRENT_CONTEXT(ctx);
1544 struct vbo_save_context *save = &vbo_context(ctx)->save;
1545 const GLint i = save->prim_store->used - 1;
1547 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1548 save->prim_store->prims[i].end = 1;
1549 save->prim_store->prims[i].count = (get_vertex_count(save) - save->prim_store->prims[i].start);
1551 /* Swap out this vertex format while outside begin/end. Any color,
1552 * etc. received between here and the next begin will be compiled
1555 if (save->out_of_memory) {
1556 vbo_install_save_vtxfmt_noop(ctx);
1559 _mesa_install_save_vtxfmt(ctx);
1564 static void GLAPIENTRY
1565 _save_Begin(GLenum mode)
1567 GET_CURRENT_CONTEXT(ctx);
1569 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1573 static void GLAPIENTRY
1574 _save_PrimitiveRestartNV(void)
1576 GET_CURRENT_CONTEXT(ctx);
1577 struct vbo_save_context *save = &vbo_context(ctx)->save;
1579 if (save->prim_store->used == 0) {
1580 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1583 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1584 "glPrimitiveRestartNV called outside glBegin/End");
1586 /* get current primitive mode */
1587 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1588 bool no_current_update = save->no_current_update;
1590 /* restart primitive */
1591 CALL_End(ctx->CurrentServerDispatch, ());
1592 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1597 /* Unlike the functions above, these are to be hooked into the vtxfmt
1598 * maintained in ctx->ListState, active when the list is known or
1599 * suspected to be outside any begin/end primitive.
1600 * Note: OBE = Outside Begin/End
1603 save_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1605 GET_CURRENT_CONTEXT(ctx);
1606 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1608 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1609 CALL_Vertex2f(dispatch, (x1, y1));
1610 CALL_Vertex2f(dispatch, (x2, y1));
1611 CALL_Vertex2f(dispatch, (x2, y2));
1612 CALL_Vertex2f(dispatch, (x1, y2));
1613 CALL_End(dispatch, ());
1618 save_Rectdv(const GLdouble *v1, const GLdouble *v2)
1620 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1624 save_Rectfv(const GLfloat *v1, const GLfloat *v2)
1626 save_Rectf(v1[0], v1[1], v2[0], v2[1]);
1630 save_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1632 save_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1636 save_Rectiv(const GLint *v1, const GLint *v2)
1638 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1642 save_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1644 save_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1648 save_Rectsv(const GLshort *v1, const GLshort *v2)
1650 save_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1654 save_DrawArrays(GLenum mode, GLint start, GLsizei count)
1656 GET_CURRENT_CONTEXT(ctx);
1657 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1658 struct vbo_save_context *save = &vbo_context(ctx)->save;
1661 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1662 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1666 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1670 if (save->out_of_memory)
1673 grow_vertex_storage(ctx, count);
1675 /* Make sure to process any VBO binding changes */
1676 _mesa_update_state(ctx);
1678 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1680 vbo_save_NotifyBegin(ctx, mode, true);
1682 for (i = 0; i < count; i++)
1683 _mesa_array_element(ctx, start + i);
1684 CALL_End(ctx->CurrentServerDispatch, ());
1686 _mesa_vao_unmap_arrays(ctx, vao);
1691 save_MultiDrawArrays(GLenum mode, const GLint *first,
1692 const GLsizei *count, GLsizei primcount)
1694 GET_CURRENT_CONTEXT(ctx);
1697 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1698 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1702 if (primcount < 0) {
1703 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1704 "glMultiDrawArrays(primcount<0)");
1708 unsigned vertcount = 0;
1709 for (i = 0; i < primcount; i++) {
1711 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1712 "glMultiDrawArrays(count[i]<0)");
1715 vertcount += count[i];
1718 grow_vertex_storage(ctx, vertcount);
1720 for (i = 0; i < primcount; i++) {
1722 save_DrawArrays(mode, first[i], count[i]);
1729 array_element(struct gl_context *ctx,
1730 GLint basevertex, GLuint elt, unsigned index_size_shift)
1732 /* Section 10.3.5 Primitive Restart:
1734 * When one of the *BaseVertex drawing commands specified in section 10.5
1735 * is used, the primitive restart comparison occurs before the basevertex
1736 * offset is added to the array index.
1738 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1739 * then we call PrimitiveRestartNV and return.
1741 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1742 elt == ctx->Array._RestartIndex[index_size_shift]) {
1743 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1747 _mesa_array_element(ctx, basevertex + elt);
1751 /* Could do better by copying the arrays and element list intact and
1752 * then emitting an indexed prim at runtime.
1755 save_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1756 const GLvoid * indices, GLint basevertex)
1758 GET_CURRENT_CONTEXT(ctx);
1759 struct vbo_save_context *save = &vbo_context(ctx)->save;
1760 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1761 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1764 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1765 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1769 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1772 if (type != GL_UNSIGNED_BYTE &&
1773 type != GL_UNSIGNED_SHORT &&
1774 type != GL_UNSIGNED_INT) {
1775 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1779 if (save->out_of_memory)
1782 grow_vertex_storage(ctx, count);
1784 /* Make sure to process any VBO binding changes */
1785 _mesa_update_state(ctx);
1787 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1791 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1793 vbo_save_NotifyBegin(ctx, mode, true);
1796 case GL_UNSIGNED_BYTE:
1797 for (i = 0; i < count; i++)
1798 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1800 case GL_UNSIGNED_SHORT:
1801 for (i = 0; i < count; i++)
1802 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1804 case GL_UNSIGNED_INT:
1805 for (i = 0; i < count; i++)
1806 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1809 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1813 CALL_End(ctx->CurrentServerDispatch, ());
1815 _mesa_vao_unmap(ctx, vao);
1819 save_DrawElements(GLenum mode, GLsizei count, GLenum type,
1820 const GLvoid * indices)
1822 save_DrawElementsBaseVertex(mode, count, type, indices, 0);
1827 save_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1828 GLsizei count, GLenum type,
1829 const GLvoid * indices)
1831 GET_CURRENT_CONTEXT(ctx);
1832 struct vbo_save_context *save = &vbo_context(ctx)->save;
1834 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1835 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1839 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1840 "glDrawRangeElements(count<0)");
1843 if (type != GL_UNSIGNED_BYTE &&
1844 type != GL_UNSIGNED_SHORT &&
1845 type != GL_UNSIGNED_INT) {
1846 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1850 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1851 "glDrawRangeElements(end < start)");
1855 if (save->out_of_memory)
1858 save_DrawElements(mode, count, type, indices);
1863 save_MultiDrawElementsEXT(GLenum mode, const GLsizei *count, GLenum type,
1864 const GLvoid * const *indices, GLsizei primcount)
1866 GET_CURRENT_CONTEXT(ctx);
1867 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1871 for (i = 0; i < primcount; i++) {
1872 vertcount += count[i];
1874 grow_vertex_storage(ctx, vertcount);
1876 for (i = 0; i < primcount; i++) {
1878 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1885 save_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1887 const GLvoid * const *indices,
1889 const GLint *basevertex)
1891 GET_CURRENT_CONTEXT(ctx);
1892 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1896 for (i = 0; i < primcount; i++) {
1897 vertcount += count[i];
1899 grow_vertex_storage(ctx, vertcount);
1901 for (i = 0; i < primcount; i++) {
1903 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1912 vbo_install_save_vtxfmt(struct gl_context *ctx)
1914 #define NAME_AE(x) _mesa_##x
1915 #define NAME_CALLLIST(x) _save_##x
1916 #define NAME(x) _save_##x
1917 #define NAME_ES(x) _save_##x
1919 struct _glapi_table *tab = ctx->Save;
1920 #include "api_vtxfmt_init.h"
1925 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1927 struct vbo_save_context *save = &vbo_context(ctx)->save;
1929 /* Noop when we are actually active:
1931 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1934 if (save->vertex_store->used || save->prim_store->used)
1935 compile_vertex_list(ctx);
1937 copy_to_current(ctx);
1939 ctx->Driver.SaveNeedFlush = GL_FALSE;
1944 * Called from glNewList when we're starting to compile a display list.
1947 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1949 struct vbo_save_context *save = &vbo_context(ctx)->save;
1954 if (!save->prim_store)
1955 save->prim_store = realloc_prim_store(NULL, 8);
1957 if (!save->vertex_store)
1958 save->vertex_store = CALLOC_STRUCT(vbo_save_vertex_store);
1961 ctx->Driver.SaveNeedFlush = GL_FALSE;
1966 * Called from glEndList when we're finished compiling a display list.
1969 vbo_save_EndList(struct gl_context *ctx)
1971 struct vbo_save_context *save = &vbo_context(ctx)->save;
1973 /* EndList called inside a (saved) Begin/End pair?
1975 if (_mesa_inside_dlist_begin_end(ctx)) {
1976 if (save->prim_store->used > 0) {
1977 GLint i = save->prim_store->used - 1;
1978 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1979 save->prim_store->prims[i].end = 0;
1980 save->prim_store->prims[i].count = get_vertex_count(save) - save->prim_store->prims[i].start;
1983 /* Make sure this vertex list gets replayed by the "loopback"
1986 save->dangling_attr_ref = GL_TRUE;
1987 vbo_save_SaveFlushVertices(ctx);
1989 /* Swap out this vertex format while outside begin/end. Any color,
1990 * etc. received between here and the next begin will be compiled
1993 _mesa_install_save_vtxfmt(ctx);
1996 assert(save->vertex_size == 0);
2000 * Called during context creation/init.
2003 current_init(struct gl_context *ctx)
2005 struct vbo_save_context *save = &vbo_context(ctx)->save;
2008 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
2009 const GLuint j = i - VBO_ATTRIB_POS;
2010 assert(j < VERT_ATTRIB_MAX);
2011 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
2012 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
2015 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
2016 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
2017 assert(j < MAT_ATTRIB_MAX);
2018 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
2019 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
2025 * Initialize the display list compiler. Called during context creation.
2028 vbo_save_api_init(struct vbo_save_context *save)
2030 struct gl_context *ctx = gl_context_from_vbo_save(save);