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 realloc_vertex_store(struct vbo_save_vertex_store *store, uint32_t vertex_size, int vertex_count)
137 store = CALLOC_STRUCT(vbo_save_vertex_store);
139 int new_size = MAX2(vertex_count * vertex_size, VBO_SAVE_BUFFER_SIZE) * sizeof(GLfloat);
140 if (new_size > store->buffer_in_ram_size) {
141 store->buffer_in_ram_size = new_size;
142 store->buffer_in_ram = realloc(store->buffer_in_ram, store->buffer_in_ram_size);
149 static struct vbo_save_primitive_store *
150 realloc_prim_store(struct vbo_save_primitive_store *store, int prim_count)
153 store = CALLOC_STRUCT(vbo_save_primitive_store);
154 uint32_t old_size = store->size;
155 store->size = MAX3(store->size, prim_count, VBO_SAVE_PRIM_SIZE);
156 store->prims = realloc(store->prims, store->size * sizeof(struct _mesa_prim));
157 memset(&store->prims[old_size], 0, (store->size - old_size) * sizeof(struct _mesa_prim));
164 reset_counters(struct gl_context *ctx)
166 struct vbo_save_context *save = &vbo_context(ctx)->save;
168 save->vertex_store->used = 0;
169 save->prim_store->used = 0;
170 save->vert_count = 0;
171 save->dangling_attr_ref = GL_FALSE;
175 * For a list of prims, try merging prims that can just be extensions of the
179 merge_prims(struct gl_context *ctx, struct _mesa_prim *prim_list,
183 struct _mesa_prim *prev_prim = prim_list;
185 for (i = 1; i < *prim_count; i++) {
186 struct _mesa_prim *this_prim = prim_list + i;
188 vbo_try_prim_conversion(&this_prim->mode, &this_prim->count);
190 if (vbo_merge_draws(ctx, true,
191 prev_prim->mode, this_prim->mode,
192 prev_prim->start, this_prim->start,
193 &prev_prim->count, this_prim->count,
194 prev_prim->basevertex, this_prim->basevertex,
196 this_prim->begin, this_prim->end)) {
197 /* We've found a prim that just extend the previous one. Tack it
198 * onto the previous one, and let this primitive struct get dropped.
203 /* If any previous primitives have been dropped, then we need to copy
204 * this later one into the next available slot.
207 if (prev_prim != this_prim)
208 *prev_prim = *this_prim;
211 *prim_count = prev_prim - prim_list + 1;
216 * Convert GL_LINE_LOOP primitive into GL_LINE_STRIP so that drivers
217 * don't have to worry about handling the _mesa_prim::begin/end flags.
218 * See https://bugs.freedesktop.org/show_bug.cgi?id=81174
221 convert_line_loop_to_strip(struct vbo_save_context *save,
222 struct vbo_save_vertex_list *node)
224 struct _mesa_prim *prim = &node->cold->prims[node->cold->prim_count - 1];
226 assert(prim->mode == GL_LINE_LOOP);
229 /* Copy the 0th vertex to end of the buffer and extend the
230 * vertex count by one to finish the line loop.
232 const GLuint sz = save->vertex_size;
234 const fi_type *src = save->vertex_store->buffer_in_ram + prim->start * sz;
236 fi_type *dst = save->vertex_store->buffer_in_ram + (prim->start + prim->count) * sz;
238 memcpy(dst, src, sz * sizeof(float));
241 node->cold->vertex_count++;
243 save->vertex_store->used += sz;
247 /* Drawing the second or later section of a long line loop.
248 * Skip the 0th vertex.
254 prim->mode = GL_LINE_STRIP;
258 /* Compare the present vao if it has the same setup. */
260 compare_vao(gl_vertex_processing_mode mode,
261 const struct gl_vertex_array_object *vao,
262 const struct gl_buffer_object *bo, GLintptr buffer_offset,
263 GLuint stride, GLbitfield64 vao_enabled,
264 const GLubyte size[VBO_ATTRIB_MAX],
265 const GLenum16 type[VBO_ATTRIB_MAX],
266 const GLuint offset[VBO_ATTRIB_MAX])
271 /* If the enabled arrays are not the same we are not equal. */
272 if (vao_enabled != vao->Enabled)
275 /* Check the buffer binding at 0 */
276 if (vao->BufferBinding[0].BufferObj != bo)
278 /* BufferBinding[0].Offset != buffer_offset is checked per attribute */
279 if (vao->BufferBinding[0].Stride != stride)
281 assert(vao->BufferBinding[0].InstanceDivisor == 0);
283 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space */
284 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
286 /* Now check the enabled arrays */
287 GLbitfield mask = vao_enabled;
289 const int attr = u_bit_scan(&mask);
290 const unsigned char vbo_attr = vao_to_vbo_map[attr];
291 const GLenum16 tp = type[vbo_attr];
292 const GLintptr off = offset[vbo_attr] + buffer_offset;
293 const struct gl_array_attributes *attrib = &vao->VertexAttrib[attr];
294 if (attrib->RelativeOffset + vao->BufferBinding[0].Offset != off)
296 if (attrib->Format.Type != tp)
298 if (attrib->Format.Size != size[vbo_attr])
300 assert(attrib->Format.Format == GL_RGBA);
301 assert(attrib->Format.Normalized == GL_FALSE);
302 assert(attrib->Format.Integer == vbo_attrtype_to_integer_flag(tp));
303 assert(attrib->Format.Doubles == vbo_attrtype_to_double_flag(tp));
304 assert(attrib->BufferBindingIndex == 0);
311 /* Create or reuse the vao for the vertex processing mode. */
313 update_vao(struct gl_context *ctx,
314 gl_vertex_processing_mode mode,
315 struct gl_vertex_array_object **vao,
316 struct gl_buffer_object *bo, GLintptr buffer_offset,
317 GLuint stride, GLbitfield64 vbo_enabled,
318 const GLubyte size[VBO_ATTRIB_MAX],
319 const GLenum16 type[VBO_ATTRIB_MAX],
320 const GLuint offset[VBO_ATTRIB_MAX])
322 /* Compute the bitmasks of vao_enabled arrays */
323 GLbitfield vao_enabled = _vbo_get_vao_enabled_from_vbo(mode, vbo_enabled);
326 * Check if we can possibly reuse the exisiting one.
327 * In the long term we should reset them when something changes.
329 if (compare_vao(mode, *vao, bo, buffer_offset, stride,
330 vao_enabled, size, type, offset))
333 /* The initial refcount is 1 */
334 _mesa_reference_vao(ctx, vao, NULL);
335 *vao = _mesa_new_vao(ctx, ~((GLuint)0));
338 * assert(stride <= ctx->Const.MaxVertexAttribStride);
339 * MaxVertexAttribStride is not set for drivers that does not
340 * expose GL 44 or GLES 31.
343 /* Bind the buffer object at binding point 0 */
344 _mesa_bind_vertex_buffer(ctx, *vao, 0, bo, buffer_offset, stride, false,
347 /* Retrieve the mapping from VBO_ATTRIB to VERT_ATTRIB space
348 * Note that the position/generic0 aliasing is done in the VAO.
350 const GLubyte *const vao_to_vbo_map = _vbo_attribute_alias_map[mode];
351 /* Now set the enable arrays */
352 GLbitfield mask = vao_enabled;
354 const int vao_attr = u_bit_scan(&mask);
355 const GLubyte vbo_attr = vao_to_vbo_map[vao_attr];
356 assert(offset[vbo_attr] <= ctx->Const.MaxVertexAttribRelativeOffset);
358 _vbo_set_attrib_format(ctx, *vao, vao_attr, buffer_offset,
359 size[vbo_attr], type[vbo_attr], offset[vbo_attr]);
360 _mesa_vertex_attrib_binding(ctx, *vao, vao_attr, 0);
362 _mesa_enable_vertex_array_attribs(ctx, *vao, vao_enabled);
363 assert(vao_enabled == (*vao)->Enabled);
364 assert((vao_enabled & ~(*vao)->VertexAttribBufferMask) == 0);
366 /* Finalize and freeze the VAO */
367 _mesa_set_vao_immutable(ctx, *vao);
371 static void wrap_filled_vertex(struct gl_context *ctx);
372 static void compile_vertex_list(struct gl_context *ctx);
375 realloc_storage(struct gl_context *ctx, int prim_count, int vertex_count)
377 struct vbo_save_context *save = &vbo_context(ctx)->save;
378 const int ten_MB = 10 * 1024 * 1024;
380 /* Limit how much memory we allocate. */
381 if (save->prim_store->used > 0 &&
383 vertex_count * save->vertex_size > ten_MB) {
384 wrap_filled_vertex(ctx);
385 vertex_count = ten_MB / save->vertex_size;
388 if (prim_count > 0 &&
389 prim_count * sizeof(struct _mesa_prim) > ten_MB) {
390 if (save->prim_store->used > 0)
391 compile_vertex_list(ctx);
392 prim_count = ten_MB / sizeof(struct _mesa_prim);
395 if (vertex_count >= 0)
396 save->vertex_store = realloc_vertex_store(save->vertex_store, save->vertex_size, vertex_count);
399 save->prim_store = realloc_prim_store(save->prim_store, prim_count);
403 unsigned vertex_size;
404 fi_type *vertex_attributes;
407 static uint32_t _hash_vertex_key(const void *key)
409 struct vertex_key *k = (struct vertex_key*)key;
410 unsigned sz = k->vertex_size;
412 return _mesa_hash_data(k->vertex_attributes, sz * sizeof(float));
415 static bool _compare_vertex_key(const void *key1, const void *key2)
417 struct vertex_key *k1 = (struct vertex_key*)key1;
418 struct vertex_key *k2 = (struct vertex_key*)key2;
419 /* All the compared vertices are going to be drawn with the same VAO,
420 * so we can compare the attributes. */
421 assert (k1->vertex_size == k2->vertex_size);
422 return memcmp(k1->vertex_attributes,
423 k2->vertex_attributes,
424 k1->vertex_size * sizeof(float)) == 0;
427 static void _free_entry(struct hash_entry *entry)
429 free((void*)entry->key);
432 /* Add vertex to the vertex buffer and return its index. If this vertex is a duplicate
433 * of an existing vertex, return the original index instead.
436 add_vertex(struct vbo_save_context *save, struct hash_table *hash_to_index,
437 uint32_t index, fi_type *new_buffer, uint32_t *max_index)
439 /* If vertex deduplication is disabled return the original index. */
443 fi_type *vert = save->vertex_store->buffer_in_ram + save->vertex_size * index;
445 struct vertex_key *key = malloc(sizeof(struct vertex_key));
446 key->vertex_size = save->vertex_size;
447 key->vertex_attributes = vert;
449 struct hash_entry *entry = _mesa_hash_table_search(hash_to_index, key);
452 /* We found an existing vertex with the same hash, return its index. */
453 return (uintptr_t) entry->data;
455 /* This is a new vertex. Determine a new index and copy its attributes to the vertex
456 * buffer. Note that 'new_buffer' is created at each list compilation so we write vertices
457 * starting at index 0.
459 uint32_t n = _mesa_hash_table_num_entries(hash_to_index);
460 *max_index = MAX2(n, *max_index);
462 memcpy(&new_buffer[save->vertex_size * n],
464 save->vertex_size * sizeof(fi_type));
466 _mesa_hash_table_insert(hash_to_index, key, (void*)(uintptr_t)(n));
468 /* The index buffer is shared between list compilations, so add the base index to get
477 * Insert the active immediate struct onto the display list currently
481 compile_vertex_list(struct gl_context *ctx)
483 struct vbo_save_context *save = &vbo_context(ctx)->save;
484 struct vbo_save_vertex_list *node;
486 /* Allocate space for this structure in the display list currently
489 node = (struct vbo_save_vertex_list *)
490 _mesa_dlist_alloc_vertex_list(ctx, !save->dangling_attr_ref && !save->no_current_update);
495 memset(node, 0, sizeof(struct vbo_save_vertex_list));
496 node->cold = calloc(1, sizeof(*node->cold));
498 /* Make sure the pointer is aligned to the size of a pointer */
499 assert((GLintptr) node % sizeof(void *) == 0);
501 const GLsizei stride = save->vertex_size*sizeof(GLfloat);
503 node->cold->vertex_count = save->vert_count;
504 node->cold->wrap_count = save->copied.nr;
505 node->cold->prims = malloc(sizeof(struct _mesa_prim) * save->prim_store->used);
506 memcpy(node->cold->prims, save->prim_store->prims, sizeof(struct _mesa_prim) * save->prim_store->used);
507 node->cold->ib.obj = NULL;
508 node->cold->prim_count = save->prim_store->used;
510 if (save->no_current_update) {
511 node->cold->current_data = NULL;
514 GLuint current_size = save->vertex_size - save->attrsz[0];
515 node->cold->current_data = NULL;
518 node->cold->current_data = malloc(current_size * sizeof(GLfloat));
519 if (node->cold->current_data) {
520 const char *buffer = (const char *)save->vertex_store->buffer_in_ram;
521 unsigned attr_offset = save->attrsz[0] * sizeof(GLfloat);
522 unsigned vertex_offset = 0;
524 if (node->cold->vertex_count)
525 vertex_offset = (node->cold->vertex_count - 1) * stride;
527 memcpy(node->cold->current_data, buffer + vertex_offset + attr_offset,
528 current_size * sizeof(GLfloat));
530 _mesa_error(ctx, GL_OUT_OF_MEMORY, "Current value allocation");
535 assert(save->attrsz[VBO_ATTRIB_POS] != 0 || node->cold->vertex_count == 0);
537 if (save->dangling_attr_ref)
538 ctx->ListState.Current.UseLoopback = true;
540 /* Copy duplicated vertices
542 save->copied.nr = copy_vertices(ctx, node, save->vertex_store->buffer_in_ram);
544 if (node->cold->prims[node->cold->prim_count - 1].mode == GL_LINE_LOOP) {
545 convert_line_loop_to_strip(save, node);
548 merge_prims(ctx, node->cold->prims, &node->cold->prim_count);
550 GLintptr buffer_offset = 0;
551 GLuint start_offset = 0;
553 /* Create an index buffer. */
554 node->cold->min_index = node->cold->max_index = 0;
555 if (save->vert_count == 0 || node->cold->prim_count == 0)
558 /* We won't modify node->prims, so use a const alias to avoid unintended
560 const struct _mesa_prim *original_prims = node->cold->prims;
562 int end = original_prims[node->cold->prim_count - 1].start +
563 original_prims[node->cold->prim_count - 1].count;
564 int total_vert_count = end - original_prims[0].start;
566 node->cold->min_index = node->cold->prims[0].start;
567 node->cold->max_index = end - 1;
569 /* Estimate for the worst case: all prims are line strips (the +1 is because
570 * wrap_buffers may call use but the last primitive may not be complete) */
571 int max_indices_count = MAX2(total_vert_count * 2 - (node->cold->prim_count * 2) + 1,
574 int size = max_indices_count * sizeof(uint32_t);
575 uint32_t* indices = (uint32_t*) malloc(size);
576 struct _mesa_prim *merged_prims = NULL;
579 struct hash_table *vertex_to_index = NULL;
580 fi_type *temp_vertices_buffer = NULL;
582 /* The loopback replay code doesn't use the index buffer, so we can't
583 * dedup vertices in this case.
585 if (!ctx->ListState.Current.UseLoopback) {
586 vertex_to_index = _mesa_hash_table_create(NULL, _hash_vertex_key, _compare_vertex_key);
587 temp_vertices_buffer = malloc(save->vertex_store->buffer_in_ram_size);
590 uint32_t max_index = 0;
592 int last_valid_prim = -1;
593 /* Construct indices array. */
594 for (unsigned i = 0; i < node->cold->prim_count; i++) {
595 assert(original_prims[i].basevertex == 0);
596 GLubyte mode = original_prims[i].mode;
598 int vertex_count = original_prims[i].count;
603 /* Line strips may get converted to lines */
604 if (mode == GL_LINE_STRIP)
607 /* If 2 consecutive prims use the same mode => merge them. */
608 bool merge_prims = last_valid_prim >= 0 &&
609 mode == merged_prims[last_valid_prim].mode &&
610 mode != GL_LINE_LOOP && mode != GL_TRIANGLE_FAN &&
611 mode != GL_QUAD_STRIP && mode != GL_POLYGON &&
614 /* To be able to merge consecutive triangle strips we need to insert
615 * a degenerate triangle.
618 mode == GL_TRIANGLE_STRIP) {
619 /* Insert a degenerate triangle */
620 assert(merged_prims[last_valid_prim].mode == GL_TRIANGLE_STRIP);
621 unsigned tri_count = merged_prims[last_valid_prim].count - 2;
623 indices[idx] = indices[idx - 1];
624 indices[idx + 1] = add_vertex(save, vertex_to_index, original_prims[i].start,
625 temp_vertices_buffer, &max_index);
627 merged_prims[last_valid_prim].count += 2;
630 /* Add another index to preserve winding order */
631 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start,
632 temp_vertices_buffer, &max_index);
633 merged_prims[last_valid_prim].count++;
639 /* Convert line strips to lines if it'll allow if the previous
640 * prim mode is GL_LINES (so merge_prims is true) or if the next
641 * primitive mode is GL_LINES or GL_LINE_LOOP.
643 if (original_prims[i].mode == GL_LINE_STRIP &&
645 (i < node->cold->prim_count - 1 &&
646 (original_prims[i + 1].mode == GL_LINE_STRIP ||
647 original_prims[i + 1].mode == GL_LINES)))) {
648 for (unsigned j = 0; j < vertex_count; j++) {
649 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
650 temp_vertices_buffer, &max_index);
651 /* Repeat all but the first/last indices. */
652 if (j && j != vertex_count - 1) {
653 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
654 temp_vertices_buffer, &max_index);
658 /* We didn't convert to LINES, so restore the original mode */
659 mode = original_prims[i].mode;
661 for (unsigned j = 0; j < vertex_count; j++) {
662 indices[idx++] = add_vertex(save, vertex_to_index, original_prims[i].start + j,
663 temp_vertices_buffer, &max_index);
668 /* Update vertex count. */
669 merged_prims[last_valid_prim].count += idx - start;
671 /* Keep this primitive */
672 last_valid_prim += 1;
673 assert(last_valid_prim <= i);
674 merged_prims = realloc(merged_prims, (1 + last_valid_prim) * sizeof(struct _mesa_prim));
675 merged_prims[last_valid_prim] = original_prims[i];
676 merged_prims[last_valid_prim].start = start;
677 merged_prims[last_valid_prim].count = idx - start;
679 merged_prims[last_valid_prim].mode = mode;
682 assert(idx > 0 && idx <= max_indices_count);
684 unsigned merged_prim_count = last_valid_prim + 1;
685 node->cold->ib.ptr = NULL;
686 node->cold->ib.count = idx;
687 node->cold->ib.index_size_shift = (GL_UNSIGNED_INT - GL_UNSIGNED_BYTE) >> 1;
689 /* How many bytes do we need to store the indices and the vertices */
690 total_vert_count = vertex_to_index ? (max_index + 1) : idx;
691 unsigned total_bytes_needed = idx * sizeof(uint32_t) +
692 total_vert_count * save->vertex_size * sizeof(fi_type);
694 const GLintptr old_offset = save->VAO[0] ?
695 save->VAO[0]->BufferBinding[0].Offset + save->VAO[0]->VertexAttrib[VERT_ATTRIB_POS].RelativeOffset : 0;
696 if (old_offset != save->current_bo_bytes_used && stride > 0) {
697 GLintptr offset_diff = save->current_bo_bytes_used - old_offset;
698 while (offset_diff > 0 &&
699 save->current_bo_bytes_used < save->current_bo->Size &&
700 offset_diff % stride != 0) {
701 save->current_bo_bytes_used++;
702 offset_diff = save->current_bo_bytes_used - old_offset;
705 buffer_offset = save->current_bo_bytes_used;
707 /* Can we reuse the previous bo or should we allocate a new one? */
708 int available_bytes = save->current_bo ? save->current_bo->Size - save->current_bo_bytes_used : 0;
709 if (total_bytes_needed > available_bytes) {
710 if (save->current_bo)
711 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
712 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
713 bool success = ctx->Driver.BufferData(ctx,
714 GL_ELEMENT_ARRAY_BUFFER_ARB,
715 MAX2(total_bytes_needed, VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t)),
717 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
720 _mesa_reference_buffer_object(ctx, &save->current_bo, NULL);
721 _mesa_error(ctx, GL_OUT_OF_MEMORY, "IB allocation");
723 save->current_bo_bytes_used = 0;
724 available_bytes = save->current_bo->Size;
728 assert(old_offset <= buffer_offset);
729 const GLintptr offset_diff = buffer_offset - old_offset;
730 if (offset_diff > 0 && stride > 0 && offset_diff % stride == 0) {
731 /* The vertex size is an exact multiple of the buffer offset.
732 * This means that we can use zero-based vertex attribute pointers
733 * and specify the start of the primitive with the _mesa_prim::start
734 * field. This results in issuing several draw calls with identical
735 * vertex attribute information. This can result in fewer state
736 * changes in drivers. In particular, the Gallium CSO module will
737 * filter out redundant vertex buffer changes.
739 /* We cannot immediately update the primitives as some methods below
740 * still need the uncorrected start vertices
742 start_offset = offset_diff/stride;
743 assert(old_offset == buffer_offset - offset_diff);
744 buffer_offset = old_offset;
747 /* Correct the primitive starts, we can only do this here as copy_vertices
748 * and convert_line_loop_to_strip above consume the uncorrected starts.
749 * On the other hand the _vbo_loopback_vertex_list call below needs the
750 * primitives to be corrected already.
752 for (unsigned i = 0; i < node->cold->prim_count; i++) {
753 node->cold->prims[i].start += start_offset;
755 /* start_offset shifts vertices (so v[0] becomes v[start_offset]), so we have
756 * to apply this transformation to all indices and max_index.
758 for (unsigned i = 0; i < idx; i++)
759 indices[i] += start_offset;
760 max_index += start_offset;
763 _mesa_reference_buffer_object(ctx, &node->cold->ib.obj, save->current_bo);
765 /* Upload the vertices first (see buffer_offset) */
766 ctx->Driver.BufferSubData(ctx,
767 save->current_bo_bytes_used,
768 total_vert_count * save->vertex_size * sizeof(fi_type),
769 vertex_to_index ? temp_vertices_buffer : save->vertex_store->buffer_in_ram,
771 save->current_bo_bytes_used += total_vert_count * save->vertex_size * sizeof(fi_type);
773 if (vertex_to_index) {
774 _mesa_hash_table_destroy(vertex_to_index, _free_entry);
775 free(temp_vertices_buffer);
778 /* Since we're append the indices to an existing buffer, we need to adjust the start value of each
779 * primitive (not the indices themselves). */
780 save->current_bo_bytes_used += align(save->current_bo_bytes_used, 4) - save->current_bo_bytes_used;
781 int indices_offset = save->current_bo_bytes_used / 4;
782 for (int i = 0; i < merged_prim_count; i++) {
783 merged_prims[i].start += indices_offset;
786 /* Then upload the indices. */
787 if (node->cold->ib.obj) {
788 ctx->Driver.BufferSubData(ctx,
789 save->current_bo_bytes_used,
790 idx * sizeof(uint32_t),
793 save->current_bo_bytes_used += idx * sizeof(uint32_t);
795 node->cold->vertex_count = 0;
796 node->cold->prim_count = 0;
799 /* Prepare for DrawGallium */
800 memset(&node->merged.info, 0, sizeof(struct pipe_draw_info));
801 /* The other info fields will be updated in vbo_save_playback_vertex_list */
802 node->merged.info.index_size = 4;
803 node->merged.info.instance_count = 1;
804 node->merged.info.index.gl_bo = node->cold->ib.obj;
805 if (merged_prim_count == 1) {
806 node->merged.info.mode = merged_prims[0].mode;
807 node->merged.start_count.start = merged_prims[0].start;
808 node->merged.start_count.count = merged_prims[0].count;
809 node->merged.start_count.index_bias = 0;
810 node->merged.mode = NULL;
812 node->merged.mode = malloc(merged_prim_count * sizeof(unsigned char));
813 node->merged.start_counts = malloc(merged_prim_count * sizeof(struct pipe_draw_start_count_bias));
814 for (unsigned i = 0; i < merged_prim_count; i++) {
815 node->merged.start_counts[i].start = merged_prims[i].start;
816 node->merged.start_counts[i].count = merged_prims[i].count;
817 node->merged.start_counts[i].index_bias = 0;
818 node->merged.mode[i] = merged_prims[i].mode;
821 node->merged.num_draws = merged_prim_count;
822 if (node->merged.num_draws > 1) {
823 bool same_mode = true;
824 for (unsigned i = 1; i < node->merged.num_draws && same_mode; i++) {
825 same_mode = node->merged.mode[i] == node->merged.mode[0];
828 /* All primitives use the same mode, so we can simplify a bit */
829 node->merged.info.mode = node->merged.mode[0];
830 free(node->merged.mode);
831 node->merged.mode = NULL;
840 if (!save->current_bo) {
841 save->current_bo = ctx->Driver.NewBufferObject(ctx, VBO_BUF_ID + 1);
842 bool success = ctx->Driver.BufferData(ctx,
843 GL_ELEMENT_ARRAY_BUFFER_ARB,
844 VBO_SAVE_BUFFER_SIZE * sizeof(uint32_t),
846 GL_STATIC_DRAW_ARB, GL_MAP_WRITE_BIT,
850 GLuint offsets[VBO_ATTRIB_MAX];
851 for (unsigned i = 0, offset = 0; i < VBO_ATTRIB_MAX; ++i) {
853 offset += save->attrsz[i] * sizeof(GLfloat);
855 /* Create a pair of VAOs for the possible VERTEX_PROCESSING_MODEs
856 * Note that this may reuse the previous one of possible.
858 for (gl_vertex_processing_mode vpm = VP_MODE_FF; vpm < VP_MODE_MAX; ++vpm) {
859 /* create or reuse the vao */
860 update_vao(ctx, vpm, &save->VAO[vpm],
861 save->current_bo, buffer_offset, stride,
862 save->enabled, save->attrsz, save->attrtype, offsets);
863 /* Reference the vao in the dlist */
864 node->VAO[vpm] = NULL;
865 _mesa_reference_vao(ctx, &node->VAO[vpm], save->VAO[vpm]);
869 /* Deal with GL_COMPILE_AND_EXECUTE:
871 if (ctx->ExecuteFlag) {
872 struct _glapi_table *dispatch = GET_DISPATCH();
874 _glapi_set_dispatch(ctx->Exec);
876 /* _vbo_loopback_vertex_list doesn't use the index buffer, so we have to
877 * use buffer_in_ram instead of current_bo which contains all vertices instead
878 * of the deduplicated vertices only in the !UseLoopback case.
880 * The problem is that the VAO offset is based on current_bo's layout,
881 * so we have to use a temp value.
883 struct gl_vertex_array_object *vao = node->VAO[VP_MODE_SHADER];
884 GLintptr original = vao->BufferBinding[0].Offset;
885 if (!ctx->ListState.Current.UseLoopback) {
886 GLintptr new_offset = 0;
887 /* 'start_offset' has been added to all primitives 'start', so undo it here. */
888 new_offset -= start_offset * stride;
889 vao->BufferBinding[0].Offset = new_offset;
891 _vbo_loopback_vertex_list(ctx, node, save->vertex_store->buffer_in_ram);
892 vao->BufferBinding[0].Offset = original;
894 _glapi_set_dispatch(dispatch);
897 /* Decide whether the storage structs are full, or can be used for
898 * the next vertex lists as well.
900 if (save->vertex_store->used >
901 save->vertex_store->buffer_in_ram_size / sizeof(float) - 16 * (save->vertex_size + 4)) {
902 realloc_storage(ctx, -1, 0);
905 /* Reset our structures for the next run of vertices:
912 * This is called when we fill a vertex buffer before we hit a glEnd().
914 * TODO -- If no new vertices have been stored, don't bother saving it.
917 wrap_buffers(struct gl_context *ctx)
919 struct vbo_save_context *save = &vbo_context(ctx)->save;
920 GLint i = save->prim_store->used - 1;
923 assert(i < (GLint) save->prim_store->size);
926 /* Close off in-progress primitive.
928 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
929 mode = save->prim_store->prims[i].mode;
931 /* store the copied vertices, and allocate a new list.
933 compile_vertex_list(ctx);
935 /* Restart interrupted primitive
937 save->prim_store->prims[0].mode = mode;
938 save->prim_store->prims[0].begin = 0;
939 save->prim_store->prims[0].end = 0;
940 save->prim_store->prims[0].start = 0;
941 save->prim_store->prims[0].count = 0;
942 save->prim_store->used = 1;
947 * Called only when buffers are wrapped as the result of filling the
948 * vertex_store struct.
951 wrap_filled_vertex(struct gl_context *ctx)
953 struct vbo_save_context *save = &vbo_context(ctx)->save;
954 unsigned numComponents;
955 ASSERTED uint32_t max_vert = save->vertex_size ?
956 save->vertex_store->buffer_in_ram_size / (sizeof(float) * save->vertex_size) : 0;
958 /* Emit a glEnd to close off the last vertex list.
962 /* Copy stored stored vertices to start of new list.
964 assert(max_vert - save->vert_count > save->copied.nr);
966 numComponents = save->copied.nr * save->vertex_size;
968 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram;
971 numComponents * sizeof(fi_type));
972 assert(save->vertex_store->used == 0 && save->vert_count == 0);
973 save->vert_count = save->copied.nr;
974 save->vertex_store->used = numComponents;
979 copy_to_current(struct gl_context *ctx)
981 struct vbo_save_context *save = &vbo_context(ctx)->save;
982 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
985 const int i = u_bit_scan64(&enabled);
986 assert(save->attrsz[i]);
988 if (save->attrtype[i] == GL_DOUBLE ||
989 save->attrtype[i] == GL_UNSIGNED_INT64_ARB)
990 memcpy(save->current[i], save->attrptr[i], save->attrsz[i] * sizeof(GLfloat));
992 COPY_CLEAN_4V_TYPE_AS_UNION(save->current[i], save->attrsz[i],
993 save->attrptr[i], save->attrtype[i]);
999 copy_from_current(struct gl_context *ctx)
1001 struct vbo_save_context *save = &vbo_context(ctx)->save;
1002 GLbitfield64 enabled = save->enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
1005 const int i = u_bit_scan64(&enabled);
1007 switch (save->attrsz[i]) {
1009 save->attrptr[i][3] = save->current[i][3];
1012 save->attrptr[i][2] = save->current[i][2];
1015 save->attrptr[i][1] = save->current[i][1];
1018 save->attrptr[i][0] = save->current[i][0];
1021 unreachable("Unexpected vertex attribute size");
1028 * Called when we increase the size of a vertex attribute. For example,
1029 * if we've seen one or more glTexCoord2f() calls and now we get a
1030 * glTexCoord3f() call.
1031 * Flush existing data, set new attrib size, replay copied vertices.
1034 upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
1036 struct vbo_save_context *save = &vbo_context(ctx)->save;
1041 /* Store the current run of vertices, and emit a GL_END. Emit a
1042 * BEGIN in the new buffer.
1044 if (save->vert_count)
1047 assert(save->copied.nr == 0);
1049 /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
1050 * when the attribute already exists in the vertex and is having
1051 * its size increased.
1053 copy_to_current(ctx);
1057 oldsz = save->attrsz[attr];
1058 save->attrsz[attr] = newsz;
1059 save->enabled |= BITFIELD64_BIT(attr);
1061 save->vertex_size += newsz - oldsz;
1062 save->vert_count = 0;
1064 /* Recalculate all the attrptr[] values:
1067 for (i = 0; i < VBO_ATTRIB_MAX; i++) {
1068 if (save->attrsz[i]) {
1069 save->attrptr[i] = tmp;
1070 tmp += save->attrsz[i];
1073 save->attrptr[i] = NULL; /* will not be dereferenced. */
1077 /* Copy from current to repopulate the vertex with correct values.
1079 copy_from_current(ctx);
1081 /* Replay stored vertices to translate them to new format here.
1083 * If there are copied vertices and the new (upgraded) attribute
1084 * has not been defined before, this list is somewhat degenerate,
1085 * and will need fixup at runtime.
1087 if (save->copied.nr) {
1088 const fi_type *data = save->copied.buffer;
1089 fi_type *dest = save->vertex_store->buffer_in_ram;
1091 /* Need to note this and fix up at runtime (or loopback):
1093 if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
1095 save->dangling_attr_ref = GL_TRUE;
1098 for (i = 0; i < save->copied.nr; i++) {
1099 GLbitfield64 enabled = save->enabled;
1101 const int j = u_bit_scan64(&enabled);
1102 assert(save->attrsz[j]);
1105 COPY_CLEAN_4V_TYPE_AS_UNION(dest, oldsz, data,
1111 COPY_SZ_4V(dest, newsz, save->current[attr]);
1116 GLint sz = save->attrsz[j];
1117 COPY_SZ_4V(dest, sz, data);
1124 save->vert_count += save->copied.nr;
1125 save->vertex_store->used += save->vertex_size * save->copied.nr;
1131 * This is called when the size of a vertex attribute changes.
1132 * For example, after seeing one or more glTexCoord2f() calls we
1133 * get a glTexCoord4f() or glTexCoord1f() call.
1136 fixup_vertex(struct gl_context *ctx, GLuint attr,
1137 GLuint sz, GLenum newType)
1139 struct vbo_save_context *save = &vbo_context(ctx)->save;
1141 if (sz > save->attrsz[attr] ||
1142 newType != save->attrtype[attr]) {
1143 /* New size is larger. Need to flush existing vertices and get
1144 * an enlarged vertex format.
1146 upgrade_vertex(ctx, attr, sz);
1148 else if (sz < save->active_sz[attr]) {
1150 const fi_type *id = vbo_get_default_vals_as_union(save->attrtype[attr]);
1152 /* New size is equal or smaller - just need to fill in some
1155 for (i = sz; i <= save->attrsz[attr]; i++)
1156 save->attrptr[attr][i - 1] = id[i - 1];
1159 save->active_sz[attr] = sz;
1164 * Reset the current size of all vertex attributes to the default
1165 * value of 0. This signals that we haven't yet seen any per-vertex
1166 * commands such as glNormal3f() or glTexCoord2f().
1169 reset_vertex(struct gl_context *ctx)
1171 struct vbo_save_context *save = &vbo_context(ctx)->save;
1173 while (save->enabled) {
1174 const int i = u_bit_scan64(&save->enabled);
1175 assert(save->attrsz[i]);
1176 save->attrsz[i] = 0;
1177 save->active_sz[i] = 0;
1180 save->vertex_size = 0;
1185 * If index=0, does glVertexAttrib*() alias glVertex() to emit a vertex?
1186 * It depends on a few things, including whether we're inside or outside
1190 is_vertex_position(const struct gl_context *ctx, GLuint index)
1192 return (index == 0 &&
1193 _mesa_attr_zero_aliases_vertex(ctx) &&
1194 _mesa_inside_dlist_begin_end(ctx));
1199 #define ERROR(err) _mesa_compile_error(ctx, err, __func__);
1202 /* Only one size for each attribute may be active at once. Eg. if
1203 * Color3f is installed/active, then Color4f may not be, even if the
1204 * vertex actually contains 4 color coordinates. This is because the
1205 * 3f version won't otherwise set color[3] to 1.0 -- this is the job
1206 * of the chooser function when switching between Color4f and Color3f.
1208 #define ATTR_UNION(A, N, T, C, V0, V1, V2, V3) \
1210 struct vbo_save_context *save = &vbo_context(ctx)->save; \
1211 int sz = (sizeof(C) / sizeof(GLfloat)); \
1213 if (save->active_sz[A] != N) \
1214 fixup_vertex(ctx, A, N * sz, T); \
1217 C *dest = (C *)save->attrptr[A]; \
1218 if (N>0) dest[0] = V0; \
1219 if (N>1) dest[1] = V1; \
1220 if (N>2) dest[2] = V2; \
1221 if (N>3) dest[3] = V3; \
1222 save->attrtype[A] = T; \
1227 uint32_t max_vert = save->vertex_size ? \
1228 save->vertex_store->buffer_in_ram_size / (sizeof(float) * save->vertex_size) : 0; \
1229 fi_type *buffer_ptr = save->vertex_store->buffer_in_ram + save->vertex_store->used; \
1231 for (i = 0; i < save->vertex_size; i++) \
1232 buffer_ptr[i] = save->vertex[i]; \
1234 save->vertex_store->used += save->vertex_size; \
1235 if (++save->vert_count >= max_vert) \
1236 wrap_filled_vertex(ctx); \
1240 #define TAG(x) _save_##x
1242 #include "vbo_attrib_tmp.h"
1246 #define MAT( ATTR, N, face, params ) \
1248 if (face != GL_BACK) \
1249 MAT_ATTR( ATTR, N, params ); /* front */ \
1250 if (face != GL_FRONT) \
1251 MAT_ATTR( ATTR + 1, N, params ); /* back */ \
1256 * Save a glMaterial call found between glBegin/End.
1257 * glMaterial calls outside Begin/End are handled in dlist.c.
1259 static void GLAPIENTRY
1260 _save_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
1262 GET_CURRENT_CONTEXT(ctx);
1264 if (face != GL_FRONT && face != GL_BACK && face != GL_FRONT_AND_BACK) {
1265 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(face)");
1271 MAT(VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params);
1274 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1277 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1280 MAT(VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params);
1283 if (*params < 0 || *params > ctx->Const.MaxShininess) {
1284 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glMaterial(shininess)");
1287 MAT(VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params);
1290 case GL_COLOR_INDEXES:
1291 MAT(VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params);
1293 case GL_AMBIENT_AND_DIFFUSE:
1294 MAT(VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params);
1295 MAT(VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params);
1298 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMaterial(pname)");
1304 /* Cope with EvalCoord/CallList called within a begin/end object:
1305 * -- Flush current buffer
1306 * -- Fallback to opcodes for the rest of the begin/end object.
1309 dlist_fallback(struct gl_context *ctx)
1311 struct vbo_save_context *save = &vbo_context(ctx)->save;
1313 if (save->vert_count || save->prim_store->used) {
1314 if (save->prim_store->used > 0) {
1315 /* Close off in-progress primitive. */
1316 GLint i = save->prim_store->used - 1;
1317 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1320 /* Need to replay this display list with loopback,
1321 * unfortunately, otherwise this primitive won't be handled
1324 save->dangling_attr_ref = GL_TRUE;
1326 compile_vertex_list(ctx);
1329 copy_to_current(ctx);
1331 if (save->out_of_memory) {
1332 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1335 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1337 ctx->Driver.SaveNeedFlush = GL_FALSE;
1341 static void GLAPIENTRY
1342 _save_EvalCoord1f(GLfloat u)
1344 GET_CURRENT_CONTEXT(ctx);
1345 dlist_fallback(ctx);
1346 CALL_EvalCoord1f(ctx->Save, (u));
1349 static void GLAPIENTRY
1350 _save_EvalCoord1fv(const GLfloat * v)
1352 GET_CURRENT_CONTEXT(ctx);
1353 dlist_fallback(ctx);
1354 CALL_EvalCoord1fv(ctx->Save, (v));
1357 static void GLAPIENTRY
1358 _save_EvalCoord2f(GLfloat u, GLfloat v)
1360 GET_CURRENT_CONTEXT(ctx);
1361 dlist_fallback(ctx);
1362 CALL_EvalCoord2f(ctx->Save, (u, v));
1365 static void GLAPIENTRY
1366 _save_EvalCoord2fv(const GLfloat * v)
1368 GET_CURRENT_CONTEXT(ctx);
1369 dlist_fallback(ctx);
1370 CALL_EvalCoord2fv(ctx->Save, (v));
1373 static void GLAPIENTRY
1374 _save_EvalPoint1(GLint i)
1376 GET_CURRENT_CONTEXT(ctx);
1377 dlist_fallback(ctx);
1378 CALL_EvalPoint1(ctx->Save, (i));
1381 static void GLAPIENTRY
1382 _save_EvalPoint2(GLint i, GLint j)
1384 GET_CURRENT_CONTEXT(ctx);
1385 dlist_fallback(ctx);
1386 CALL_EvalPoint2(ctx->Save, (i, j));
1389 static void GLAPIENTRY
1390 _save_CallList(GLuint l)
1392 GET_CURRENT_CONTEXT(ctx);
1393 dlist_fallback(ctx);
1394 CALL_CallList(ctx->Save, (l));
1397 static void GLAPIENTRY
1398 _save_CallLists(GLsizei n, GLenum type, const GLvoid * v)
1400 GET_CURRENT_CONTEXT(ctx);
1401 dlist_fallback(ctx);
1402 CALL_CallLists(ctx->Save, (n, type, v));
1408 * Called when a glBegin is getting compiled into a display list.
1409 * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
1412 vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
1413 bool no_current_update)
1415 struct vbo_save_context *save = &vbo_context(ctx)->save;
1416 const GLuint i = save->prim_store->used++;
1418 ctx->Driver.CurrentSavePrimitive = mode;
1420 assert(i < save->prim_store->size);
1421 save->prim_store->prims[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK;
1422 save->prim_store->prims[i].begin = 1;
1423 save->prim_store->prims[i].end = 0;
1424 save->prim_store->prims[i].start = save->vert_count;
1425 save->prim_store->prims[i].count = 0;
1427 save->no_current_update = no_current_update;
1429 if (save->out_of_memory) {
1430 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1433 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt);
1436 /* We need to call vbo_save_SaveFlushVertices() if there's state change */
1437 ctx->Driver.SaveNeedFlush = GL_TRUE;
1441 static void GLAPIENTRY
1444 GET_CURRENT_CONTEXT(ctx);
1445 struct vbo_save_context *save = &vbo_context(ctx)->save;
1446 const GLint i = save->prim_store->used - 1;
1448 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1449 save->prim_store->prims[i].end = 1;
1450 save->prim_store->prims[i].count = (save->vert_count - save->prim_store->prims[i].start);
1452 if (i == (GLint) save->prim_store->size - 1) {
1453 compile_vertex_list(ctx);
1454 assert(save->copied.nr == 0);
1457 /* Swap out this vertex format while outside begin/end. Any color,
1458 * etc. received between here and the next begin will be compiled
1461 if (save->out_of_memory) {
1462 _mesa_install_save_vtxfmt(ctx, &save->vtxfmt_noop);
1465 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1470 static void GLAPIENTRY
1471 _save_Begin(GLenum mode)
1473 GET_CURRENT_CONTEXT(ctx);
1475 _mesa_compile_error(ctx, GL_INVALID_OPERATION, "Recursive glBegin");
1479 static void GLAPIENTRY
1480 _save_PrimitiveRestartNV(void)
1482 GET_CURRENT_CONTEXT(ctx);
1483 struct vbo_save_context *save = &vbo_context(ctx)->save;
1485 if (save->prim_store->used == 0) {
1486 /* We're not inside a glBegin/End pair, so calling glPrimitiverRestartNV
1489 _mesa_compile_error(ctx, GL_INVALID_OPERATION,
1490 "glPrimitiveRestartNV called outside glBegin/End");
1492 /* get current primitive mode */
1493 GLenum curPrim = save->prim_store->prims[save->prim_store->used - 1].mode;
1494 bool no_current_update = save->no_current_update;
1496 /* restart primitive */
1497 CALL_End(ctx->CurrentServerDispatch, ());
1498 vbo_save_NotifyBegin(ctx, curPrim, no_current_update);
1503 /* Unlike the functions above, these are to be hooked into the vtxfmt
1504 * maintained in ctx->ListState, active when the list is known or
1505 * suspected to be outside any begin/end primitive.
1506 * Note: OBE = Outside Begin/End
1508 static void GLAPIENTRY
1509 _save_OBE_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
1511 GET_CURRENT_CONTEXT(ctx);
1512 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1514 vbo_save_NotifyBegin(ctx, GL_QUADS, false);
1515 CALL_Vertex2f(dispatch, (x1, y1));
1516 CALL_Vertex2f(dispatch, (x2, y1));
1517 CALL_Vertex2f(dispatch, (x2, y2));
1518 CALL_Vertex2f(dispatch, (x1, y2));
1519 CALL_End(dispatch, ());
1523 static void GLAPIENTRY
1524 _save_OBE_Rectd(GLdouble x1, GLdouble y1, GLdouble x2, GLdouble y2)
1526 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1529 static void GLAPIENTRY
1530 _save_OBE_Rectdv(const GLdouble *v1, const GLdouble *v2)
1532 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1535 static void GLAPIENTRY
1536 _save_OBE_Rectfv(const GLfloat *v1, const GLfloat *v2)
1538 _save_OBE_Rectf(v1[0], v1[1], v2[0], v2[1]);
1541 static void GLAPIENTRY
1542 _save_OBE_Recti(GLint x1, GLint y1, GLint x2, GLint y2)
1544 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1547 static void GLAPIENTRY
1548 _save_OBE_Rectiv(const GLint *v1, const GLint *v2)
1550 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1553 static void GLAPIENTRY
1554 _save_OBE_Rects(GLshort x1, GLshort y1, GLshort x2, GLshort y2)
1556 _save_OBE_Rectf((GLfloat) x1, (GLfloat) y1, (GLfloat) x2, (GLfloat) y2);
1559 static void GLAPIENTRY
1560 _save_OBE_Rectsv(const GLshort *v1, const GLshort *v2)
1562 _save_OBE_Rectf((GLfloat) v1[0], (GLfloat) v1[1], (GLfloat) v2[0], (GLfloat) v2[1]);
1566 _ensure_draws_fits_in_storage(struct gl_context *ctx, int primcount, int vertcount)
1568 struct vbo_save_context *save = &vbo_context(ctx)->save;
1569 uint32_t max_vert = save->vertex_size ?
1570 save->vertex_store->buffer_in_ram_size / (sizeof(float) * save->vertex_size) : 0;
1572 bool realloc_prim = save->prim_store->used + primcount > save->prim_store->size;
1573 bool realloc_vert = save->vertex_size && (save->vert_count + vertcount >= max_vert);
1575 if (realloc_prim || realloc_vert) {
1576 if (realloc_vert && (save->vert_count || save->prim_store->used)) {
1577 /* TODO: this really isn't needed. We should realloc only the CPU-side memory. */
1578 compile_vertex_list(ctx);
1580 realloc_storage(ctx, realloc_prim ? primcount : -1, realloc_vert ? vertcount : -1);
1585 static void GLAPIENTRY
1586 _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
1588 GET_CURRENT_CONTEXT(ctx);
1589 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1590 struct vbo_save_context *save = &vbo_context(ctx)->save;
1593 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1594 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawArrays(mode)");
1598 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawArrays(count<0)");
1602 if (save->out_of_memory)
1605 _ensure_draws_fits_in_storage(ctx, 1, count);
1607 /* Make sure to process any VBO binding changes */
1608 _mesa_update_state(ctx);
1610 _mesa_vao_map_arrays(ctx, vao, GL_MAP_READ_BIT);
1612 vbo_save_NotifyBegin(ctx, mode, true);
1614 for (i = 0; i < count; i++)
1615 _mesa_array_element(ctx, start + i);
1616 CALL_End(ctx->CurrentServerDispatch, ());
1618 _mesa_vao_unmap_arrays(ctx, vao);
1622 static void GLAPIENTRY
1623 _save_OBE_MultiDrawArrays(GLenum mode, const GLint *first,
1624 const GLsizei *count, GLsizei primcount)
1626 GET_CURRENT_CONTEXT(ctx);
1629 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1630 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glMultiDrawArrays(mode)");
1634 if (primcount < 0) {
1635 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1636 "glMultiDrawArrays(primcount<0)");
1640 unsigned vertcount = 0;
1641 for (i = 0; i < primcount; i++) {
1643 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1644 "glMultiDrawArrays(count[i]<0)");
1647 vertcount += count[i];
1650 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1652 for (i = 0; i < primcount; i++) {
1654 _save_OBE_DrawArrays(mode, first[i], count[i]);
1661 array_element(struct gl_context *ctx,
1662 GLint basevertex, GLuint elt, unsigned index_size_shift)
1664 /* Section 10.3.5 Primitive Restart:
1666 * When one of the *BaseVertex drawing commands specified in section 10.5
1667 * is used, the primitive restart comparison occurs before the basevertex
1668 * offset is added to the array index.
1670 /* If PrimitiveRestart is enabled and the index is the RestartIndex
1671 * then we call PrimitiveRestartNV and return.
1673 if (ctx->Array._PrimitiveRestart[index_size_shift] &&
1674 elt == ctx->Array._RestartIndex[index_size_shift]) {
1675 CALL_PrimitiveRestartNV(ctx->CurrentServerDispatch, ());
1679 _mesa_array_element(ctx, basevertex + elt);
1683 /* Could do better by copying the arrays and element list intact and
1684 * then emitting an indexed prim at runtime.
1686 static void GLAPIENTRY
1687 _save_OBE_DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type,
1688 const GLvoid * indices, GLint basevertex)
1690 GET_CURRENT_CONTEXT(ctx);
1691 struct vbo_save_context *save = &vbo_context(ctx)->save;
1692 struct gl_vertex_array_object *vao = ctx->Array.VAO;
1693 struct gl_buffer_object *indexbuf = vao->IndexBufferObj;
1696 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1697 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawElements(mode)");
1701 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1704 if (type != GL_UNSIGNED_BYTE &&
1705 type != GL_UNSIGNED_SHORT &&
1706 type != GL_UNSIGNED_INT) {
1707 _mesa_compile_error(ctx, GL_INVALID_VALUE, "glDrawElements(count<0)");
1711 if (save->out_of_memory)
1714 _ensure_draws_fits_in_storage(ctx, 1, count);
1716 /* Make sure to process any VBO binding changes */
1717 _mesa_update_state(ctx);
1719 _mesa_vao_map(ctx, vao, GL_MAP_READ_BIT);
1723 ADD_POINTERS(indexbuf->Mappings[MAP_INTERNAL].Pointer, indices);
1725 vbo_save_NotifyBegin(ctx, mode, true);
1728 case GL_UNSIGNED_BYTE:
1729 for (i = 0; i < count; i++)
1730 array_element(ctx, basevertex, ((GLubyte *) indices)[i], 0);
1732 case GL_UNSIGNED_SHORT:
1733 for (i = 0; i < count; i++)
1734 array_element(ctx, basevertex, ((GLushort *) indices)[i], 1);
1736 case GL_UNSIGNED_INT:
1737 for (i = 0; i < count; i++)
1738 array_element(ctx, basevertex, ((GLuint *) indices)[i], 2);
1741 _mesa_error(ctx, GL_INVALID_ENUM, "glDrawElements(type)");
1745 CALL_End(ctx->CurrentServerDispatch, ());
1747 _mesa_vao_unmap(ctx, vao);
1750 static void GLAPIENTRY
1751 _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
1752 const GLvoid * indices)
1754 _save_OBE_DrawElementsBaseVertex(mode, count, type, indices, 0);
1758 static void GLAPIENTRY
1759 _save_OBE_DrawRangeElements(GLenum mode, GLuint start, GLuint end,
1760 GLsizei count, GLenum type,
1761 const GLvoid * indices)
1763 GET_CURRENT_CONTEXT(ctx);
1764 struct vbo_save_context *save = &vbo_context(ctx)->save;
1766 if (!_mesa_is_valid_prim_mode(ctx, mode)) {
1767 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(mode)");
1771 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1772 "glDrawRangeElements(count<0)");
1775 if (type != GL_UNSIGNED_BYTE &&
1776 type != GL_UNSIGNED_SHORT &&
1777 type != GL_UNSIGNED_INT) {
1778 _mesa_compile_error(ctx, GL_INVALID_ENUM, "glDrawRangeElements(type)");
1782 _mesa_compile_error(ctx, GL_INVALID_VALUE,
1783 "glDrawRangeElements(end < start)");
1787 if (save->out_of_memory)
1790 _save_OBE_DrawElements(mode, count, type, indices);
1794 static void GLAPIENTRY
1795 _save_OBE_MultiDrawElements(GLenum mode, const GLsizei *count, GLenum type,
1796 const GLvoid * const *indices, GLsizei primcount)
1798 GET_CURRENT_CONTEXT(ctx);
1799 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1803 for (i = 0; i < primcount; i++) {
1804 vertcount += count[i];
1806 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1808 for (i = 0; i < primcount; i++) {
1810 CALL_DrawElements(dispatch, (mode, count[i], type, indices[i]));
1816 static void GLAPIENTRY
1817 _save_OBE_MultiDrawElementsBaseVertex(GLenum mode, const GLsizei *count,
1819 const GLvoid * const *indices,
1821 const GLint *basevertex)
1823 GET_CURRENT_CONTEXT(ctx);
1824 struct _glapi_table *dispatch = ctx->CurrentServerDispatch;
1828 for (i = 0; i < primcount; i++) {
1829 vertcount += count[i];
1831 _ensure_draws_fits_in_storage(ctx, primcount, vertcount);
1833 for (i = 0; i < primcount; i++) {
1835 CALL_DrawElementsBaseVertex(dispatch, (mode, count[i], type,
1844 vtxfmt_init(struct gl_context *ctx)
1846 struct vbo_save_context *save = &vbo_context(ctx)->save;
1847 GLvertexformat *vfmt = &save->vtxfmt;
1849 #define NAME_AE(x) _ae_##x
1850 #define NAME_CALLLIST(x) _save_##x
1851 #define NAME(x) _save_##x
1852 #define NAME_ES(x) _save_##x##ARB
1854 #include "vbo_init_tmp.h"
1859 * Initialize the dispatch table with the VBO functions for display
1863 vbo_initialize_save_dispatch(const struct gl_context *ctx,
1864 struct _glapi_table *exec)
1866 SET_DrawArrays(exec, _save_OBE_DrawArrays);
1867 SET_MultiDrawArrays(exec, _save_OBE_MultiDrawArrays);
1868 SET_DrawElements(exec, _save_OBE_DrawElements);
1869 SET_DrawElementsBaseVertex(exec, _save_OBE_DrawElementsBaseVertex);
1870 SET_DrawRangeElements(exec, _save_OBE_DrawRangeElements);
1871 SET_MultiDrawElementsEXT(exec, _save_OBE_MultiDrawElements);
1872 SET_MultiDrawElementsBaseVertex(exec, _save_OBE_MultiDrawElementsBaseVertex);
1873 SET_Rectf(exec, _save_OBE_Rectf);
1874 SET_Rectd(exec, _save_OBE_Rectd);
1875 SET_Rectdv(exec, _save_OBE_Rectdv);
1876 SET_Rectfv(exec, _save_OBE_Rectfv);
1877 SET_Recti(exec, _save_OBE_Recti);
1878 SET_Rectiv(exec, _save_OBE_Rectiv);
1879 SET_Rects(exec, _save_OBE_Rects);
1880 SET_Rectsv(exec, _save_OBE_Rectsv);
1882 /* Note: other glDraw functins aren't compiled into display lists */
1888 vbo_save_SaveFlushVertices(struct gl_context *ctx)
1890 struct vbo_save_context *save = &vbo_context(ctx)->save;
1892 /* Noop when we are actually active:
1894 if (ctx->Driver.CurrentSavePrimitive <= PRIM_MAX)
1897 if (save->vert_count || save->prim_store->used)
1898 compile_vertex_list(ctx);
1900 copy_to_current(ctx);
1902 ctx->Driver.SaveNeedFlush = GL_FALSE;
1907 * Called from glNewList when we're starting to compile a display list.
1910 vbo_save_NewList(struct gl_context *ctx, GLuint list, GLenum mode)
1912 struct vbo_save_context *save = &vbo_context(ctx)->save;
1917 if (!save->prim_store)
1918 save->prim_store = realloc_prim_store(NULL, 8);
1920 if (!save->vertex_store)
1921 save->vertex_store = realloc_vertex_store(NULL, save->vertex_size, 8);
1924 ctx->Driver.SaveNeedFlush = GL_FALSE;
1929 * Called from glEndList when we're finished compiling a display list.
1932 vbo_save_EndList(struct gl_context *ctx)
1934 struct vbo_save_context *save = &vbo_context(ctx)->save;
1936 /* EndList called inside a (saved) Begin/End pair?
1938 if (_mesa_inside_dlist_begin_end(ctx)) {
1939 if (save->prim_store->used > 0) {
1940 GLint i = save->prim_store->used - 1;
1941 ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
1942 save->prim_store->prims[i].end = 0;
1943 save->prim_store->prims[i].count = save->vert_count - save->prim_store->prims[i].start;
1946 /* Make sure this vertex list gets replayed by the "loopback"
1949 save->dangling_attr_ref = GL_TRUE;
1950 vbo_save_SaveFlushVertices(ctx);
1952 /* Swap out this vertex format while outside begin/end. Any color,
1953 * etc. received between here and the next begin will be compiled
1956 _mesa_install_save_vtxfmt(ctx, &ctx->ListState.ListVtxfmt);
1959 assert(save->vertex_size == 0);
1963 * Called during context creation/init.
1966 current_init(struct gl_context *ctx)
1968 struct vbo_save_context *save = &vbo_context(ctx)->save;
1971 for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_EDGEFLAG; i++) {
1972 const GLuint j = i - VBO_ATTRIB_POS;
1973 assert(j < VERT_ATTRIB_MAX);
1974 save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
1975 save->current[i] = (fi_type *) ctx->ListState.CurrentAttrib[j];
1978 for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
1979 const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
1980 assert(j < MAT_ATTRIB_MAX);
1981 save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
1982 save->current[i] = (fi_type *) ctx->ListState.CurrentMaterial[j];
1988 * Initialize the display list compiler. Called during context creation.
1991 vbo_save_api_init(struct vbo_save_context *save)
1993 struct gl_context *ctx = gl_context_from_vbo_save(save);
1997 _mesa_noop_vtxfmt_init(ctx, &save->vtxfmt_noop);