/* $Id: t_imm_eval.c,v 1.20 2002/01/05 20:51:13 brianp Exp $ */ /* * Mesa 3-D graphics library * Version: 3.5 * * Copyright (C) 1999-2001 Brian Paul All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included * in all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * * Authors: * Keith Whitwell * */ #include "glheader.h" #include "colormac.h" #include "context.h" #include "macros.h" #include "mem.h" #include "mmath.h" #include "mtypes.h" #include "math/m_eval.h" #include "t_context.h" #include "t_imm_debug.h" #include "t_imm_eval.h" #include "t_imm_exec.h" #include "t_imm_fixup.h" #include "t_imm_alloc.h" static void eval_points1( GLfloat outcoord[][4], GLfloat coord[][4], const GLuint *flags, GLfloat du, GLfloat u1 ) { GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & VERT_EVAL_ANY) { outcoord[i][0] = coord[i][0]; outcoord[i][1] = coord[i][1]; if (flags[i] & VERT_EVAL_P1) outcoord[i][0] = coord[i][0] * du + u1; } } static void eval_points2( GLfloat outcoord[][4], GLfloat coord[][4], const GLuint *flags, GLfloat du, GLfloat u1, GLfloat dv, GLfloat v1 ) { GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) { if (flags[i] & VERT_EVAL_ANY) { outcoord[i][0] = coord[i][0]; outcoord[i][1] = coord[i][1]; if (flags[i] & VERT_EVAL_P2) { outcoord[i][0] = coord[i][0] * du + u1; outcoord[i][1] = coord[i][1] * dv + v1; } } } } static const GLubyte dirty_flags[5] = { 0, /* not possible */ VEC_DIRTY_0, VEC_DIRTY_1, VEC_DIRTY_2, VEC_DIRTY_3 }; static void eval1_4f( GLvector4f *dest, GLfloat coord[][4], const GLuint *flags, GLuint dimension, struct gl_1d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; GLfloat (*to)[4] = dest->data; GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C1|VERT_EVAL_P1)) { GLfloat u = (coord[i][0] - u1) * du; ASSIGN_4V(to[i], 0,0,0,1); _math_horner_bezier_curve(map->Points, to[i], u, dimension, map->Order); } dest->size = MAX2(dest->size, dimension); dest->flags |= dirty_flags[dimension]; } static void eval1_4f_ca( struct gl_client_array *dest, GLfloat coord[][4], const GLuint *flags, GLuint dimension, struct gl_1d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; GLfloat (*to)[4] = (GLfloat (*)[4])dest->Ptr; GLuint i; ASSERT(dest->Type == GL_FLOAT); ASSERT(dest->StrideB == 4 * sizeof(GLfloat)); for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C1|VERT_EVAL_P1)) { GLfloat u = (coord[i][0] - u1) * du; ASSIGN_4V(to[i], 0,0,0,1); _math_horner_bezier_curve(map->Points, to[i], u, dimension, map->Order); } dest->Size = MAX2(dest->Size, (GLint) dimension); } static void eval1_1ui( GLvector1ui *dest, GLfloat coord[][4], const GLuint *flags, struct gl_1d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; GLuint *to = dest->data; GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C1|VERT_EVAL_P1)) { GLfloat u = (coord[i][0] - u1) * du; GLfloat tmp; _math_horner_bezier_curve(map->Points, &tmp, u, 1, map->Order); to[i] = (GLuint) (GLint) tmp; } } static void eval1_norm( GLvector4f *dest, GLfloat coord[][4], const GLuint *flags, struct gl_1d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; GLfloat (*to)[4] = dest->data; GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C1|VERT_EVAL_P1)) { GLfloat u = (coord[i][0] - u1) * du; _math_horner_bezier_curve(map->Points, to[i], u, 3, map->Order); } } static void eval2_obj_norm( GLvector4f *obj_ptr, GLvector4f *norm_ptr, GLfloat coord[][4], GLuint *flags, GLuint dimension, struct gl_2d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; const GLfloat v1 = map->v1; const GLfloat dv = map->dv; GLfloat (*obj)[4] = obj_ptr->data; GLfloat (*normal)[4] = norm_ptr->data; GLuint i; /* fprintf(stderr, "%s\n", __FUNCTION__); */ for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C2|VERT_EVAL_P2)) { GLfloat u = (coord[i][0] - u1) * du; GLfloat v = (coord[i][1] - v1) * dv; GLfloat du[4], dv[4]; ASSIGN_4V(obj[i], 0,0,0,1); _math_de_casteljau_surf(map->Points, obj[i], du, dv, u, v, dimension, map->Uorder, map->Vorder); if (dimension == 4) { du[0] = du[0]*obj[i][3] - du[3]*obj[i][0]; du[1] = du[1]*obj[i][3] - du[3]*obj[i][1]; du[2] = du[2]*obj[i][3] - du[3]*obj[i][2]; dv[0] = dv[0]*obj[i][3] - dv[3]*obj[i][0]; dv[1] = dv[1]*obj[i][3] - dv[3]*obj[i][1]; dv[2] = dv[2]*obj[i][3] - dv[3]*obj[i][2]; } CROSS3(normal[i], du, dv); NORMALIZE_3FV(normal[i]); } obj_ptr->size = MAX2(obj_ptr->size, dimension); obj_ptr->flags |= dirty_flags[dimension]; } static void eval2_4f( GLvector4f *dest, GLfloat coord[][4], const GLuint *flags, GLuint dimension, struct gl_2d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; const GLfloat v1 = map->v1; const GLfloat dv = map->dv; GLfloat (*to)[4] = dest->data; GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C2|VERT_EVAL_P2)) { GLfloat u = (coord[i][0] - u1) * du; GLfloat v = (coord[i][1] - v1) * dv; /* fprintf(stderr, "coord %d: %f %f\n", i, coord[i][0], coord[i][1]); */ _math_horner_bezier_surf(map->Points, to[i], u, v, dimension, map->Uorder, map->Vorder); } dest->size = MAX2(dest->size, dimension); dest->flags |= dirty_flags[dimension]; } static void eval2_4f_ca( struct gl_client_array *dest, GLfloat coord[][4], const GLuint *flags, GLuint dimension, struct gl_2d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; const GLfloat v1 = map->v1; const GLfloat dv = map->dv; GLfloat (*to)[4] = (GLfloat (*)[4])dest->Ptr; GLuint i; ASSERT(dest->Type == GL_FLOAT); ASSERT(dest->StrideB == 4 * sizeof(GLfloat)); for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C2|VERT_EVAL_P2)) { GLfloat u = (coord[i][0] - u1) * du; GLfloat v = (coord[i][1] - v1) * dv; _math_horner_bezier_surf(map->Points, to[i], u, v, dimension, map->Uorder, map->Vorder); } dest->Size = MAX2(dest->Size, (GLint) dimension); } static void eval2_norm( GLvector4f *dest, GLfloat coord[][4], GLuint *flags, struct gl_2d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; const GLfloat v1 = map->v1; const GLfloat dv = map->dv; GLfloat (*to)[4] = dest->data; GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) { if (flags[i] & (VERT_EVAL_C2|VERT_EVAL_P2)) { GLfloat u = (coord[i][0] - u1) * du; GLfloat v = (coord[i][1] - v1) * dv; _math_horner_bezier_surf(map->Points, to[i], u, v, 3, map->Uorder, map->Vorder); } } } static void eval2_1ui( GLvector1ui *dest, GLfloat coord[][4], const GLuint *flags, struct gl_2d_map *map ) { const GLfloat u1 = map->u1; const GLfloat du = map->du; const GLfloat v1 = map->v1; const GLfloat dv = map->dv; GLuint *to = dest->data; GLuint i; for (i = 0 ; !(flags[i] & VERT_END_VB) ; i++) if (flags[i] & (VERT_EVAL_C2|VERT_EVAL_P2)) { GLfloat u = (coord[i][0] - u1) * du; GLfloat v = (coord[i][1] - v1) * dv; GLfloat tmp; _math_horner_bezier_surf(map->Points, &tmp, u, v, 1, map->Uorder, map->Vorder); to[i] = (GLuint) (GLint) tmp; } } static void copy_4f( GLfloat to[][4], GLfloat from[][4], GLuint count ) { MEMCPY( to, from, count * sizeof(to[0])); } static void copy_4f_stride( GLfloat to[][4], GLfloat *from, GLuint stride, GLuint count ) { if (stride == 4 * sizeof(GLfloat)) MEMCPY( to, from, count * sizeof(to[0])); else { GLuint i; /* fprintf(stderr, "%s stride %d count %d\n", __FUNCTION__, */ /* stride, count); */ for (i = 0 ; i < count ; i++, STRIDE_F(from, stride)) COPY_4FV( to[i], from ); } } static void copy_3f( GLfloat to[][4], GLfloat from[][4], GLuint count ) { GLuint i; /* MEMCPY( to, from, (count) * sizeof(to[0])); */ for (i = 0 ; i < count ; i++) { /* fprintf(stderr, "copy norm %d from %p: %f %f %f\n", i, */ /* from[i], */ /* from[i][0], from[i][1], from[i][2]); */ COPY_3FV(to[i], from[i]); } } static void copy_1ui( GLuint to[], GLuint from[], GLuint count ) { MEMCPY( to, from, (count) * sizeof(to[0])); } /* Translate eval enabled flags to VERT_* flags. */ static void update_eval( GLcontext *ctx ) { TNLcontext *tnl = TNL_CONTEXT(ctx); GLuint eval1 = 0, eval2 = 0; if (ctx->Eval.Map1Index) eval1 |= VERT_INDEX_BIT; if (ctx->Eval.Map2Index) eval2 |= VERT_INDEX_BIT; if (ctx->Eval.Map1Color4) eval1 |= VERT_COLOR0_BIT; if (ctx->Eval.Map2Color4) eval2 |= VERT_COLOR0_BIT; if (ctx->Eval.Map1Normal) eval1 |= VERT_NORMAL_BIT; if (ctx->Eval.Map2Normal) eval2 |= VERT_NORMAL_BIT; if (ctx->Eval.Map1TextureCoord4 || ctx->Eval.Map1TextureCoord3 || ctx->Eval.Map1TextureCoord2 || ctx->Eval.Map1TextureCoord1) eval1 |= VERT_TEX0_BIT; if (ctx->Eval.Map2TextureCoord4 || ctx->Eval.Map2TextureCoord3 || ctx->Eval.Map2TextureCoord2 || ctx->Eval.Map2TextureCoord1) eval2 |= VERT_TEX0_BIT; if (ctx->Eval.Map1Vertex4) eval1 |= VERT_OBJ_234; if (ctx->Eval.Map1Vertex3) eval1 |= VERT_OBJ_23; if (ctx->Eval.Map2Vertex4) { if (ctx->Eval.AutoNormal) eval2 |= VERT_OBJ_234 | VERT_NORMAL_BIT; else eval2 |= VERT_OBJ_234; } else if (ctx->Eval.Map2Vertex3) { if (ctx->Eval.AutoNormal) eval2 |= VERT_OBJ_23 | VERT_NORMAL_BIT; else eval2 |= VERT_OBJ_23; } tnl->eval.EvalMap1Flags = eval1; tnl->eval.EvalMap2Flags = eval2; tnl->eval.EvalNewState = 0; } /* This looks a lot like a pipeline stage, but for various reasons is * better handled outside the pipeline, and considered the final stage * of fixing up an immediate struct for execution. * * Really want to cache the results of this function in display lists, * at least for EvalMesh commands. */ void _tnl_eval_immediate( GLcontext *ctx, struct immediate *IM ) { TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_arrays *tmp = &tnl->imm_inputs; struct immediate *store = tnl->eval.im; GLuint *flags = IM->Flag + IM->CopyStart; GLuint copycount; GLuint orflag = IM->OrFlag; GLuint any_eval1 = orflag & (VERT_EVAL_C1|VERT_EVAL_P1); GLuint any_eval2 = orflag & (VERT_EVAL_C2|VERT_EVAL_P2); GLuint req = 0; GLuint purge_flags = 0; GLfloat (*coord)[4] = IM->Attrib[VERT_ATTRIB_POS] + IM->CopyStart; if (IM->AndFlag & VERT_EVAL_ANY) copycount = IM->Start - IM->CopyStart; /* just copy copied vertices */ else copycount = IM->Count - IM->CopyStart; /* copy all vertices */ /* fprintf(stderr, "%s copystart %d start %d count %d copycount %d\n", */ /* __FUNCTION__, IM->CopyStart, IM->Start, IM->Count, copycount); */ if (!store) store = tnl->eval.im = _tnl_alloc_immediate( ctx ); if (tnl->eval.EvalNewState & _NEW_EVAL) update_eval( ctx ); if (any_eval1) { req |= tnl->pipeline.inputs & tnl->eval.EvalMap1Flags; if (!ctx->Eval.Map1Vertex4 && !ctx->Eval.Map1Vertex3) purge_flags = (VERT_EVAL_P1|VERT_EVAL_C1); if (orflag & VERT_EVAL_P1) { eval_points1( store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart, coord, flags, ctx->Eval.MapGrid1du, ctx->Eval.MapGrid1u1); coord = store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart; } } if (any_eval2) { req |= tnl->pipeline.inputs & tnl->eval.EvalMap2Flags; if (!ctx->Eval.Map2Vertex4 && !ctx->Eval.Map2Vertex3) purge_flags |= (VERT_EVAL_P2|VERT_EVAL_C2); if (orflag & VERT_EVAL_P2) { eval_points2( store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart, coord, flags, ctx->Eval.MapGrid2du, ctx->Eval.MapGrid2u1, ctx->Eval.MapGrid2dv, ctx->Eval.MapGrid2v1 ); coord = store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart; } } /* _tnl_print_vert_flags(__FUNCTION__, req); */ /* Perform the evaluations on active data elements. */ if (req & VERT_INDEX_BIT) { GLuint generated = 0; if (copycount) copy_1ui( store->Index + IM->CopyStart, tmp->Index.data, copycount ); tmp->Index.data = store->Index + IM->CopyStart; tmp->Index.start = store->Index + IM->CopyStart; if (ctx->Eval.Map1Index && any_eval1) { eval1_1ui( &tmp->Index, coord, flags, &ctx->EvalMap.Map1Index ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } if (ctx->Eval.Map2Index && any_eval2) { eval2_1ui( &tmp->Index, coord, flags, &ctx->EvalMap.Map2Index ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } } if (req & VERT_COLOR0_BIT) { GLuint generated = 0; if (copycount) copy_4f_stride( store->Attrib[VERT_ATTRIB_COLOR0] + IM->CopyStart, (GLfloat *)tmp->Color.Ptr, tmp->Color.StrideB, copycount ); tmp->Color.Ptr = store->Attrib[VERT_ATTRIB_COLOR0] + IM->CopyStart; tmp->Color.StrideB = 4 * sizeof(GLfloat); tmp->Color.Flags = 0; tnl->vb.importable_data &= ~VERT_COLOR0_BIT; if (ctx->Eval.Map1Color4 && any_eval1) { eval1_4f_ca( &tmp->Color, coord, flags, 4, &ctx->EvalMap.Map1Color4 ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } if (ctx->Eval.Map2Color4 && any_eval2) { eval2_4f_ca( &tmp->Color, coord, flags, 4, &ctx->EvalMap.Map2Color4 ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } } if (req & VERT_TEX(0)) { GLuint generated = 0; if (copycount) copy_4f( store->Attrib[VERT_ATTRIB_TEX0] + IM->CopyStart, tmp->TexCoord[0].data, copycount ); else tmp->TexCoord[0].size = 0; tmp->TexCoord[0].data = store->Attrib[VERT_ATTRIB_TEX0] + IM->CopyStart; tmp->TexCoord[0].start = (GLfloat *)tmp->TexCoord[0].data; if (any_eval1) { if (ctx->Eval.Map1TextureCoord4) { eval1_4f( &tmp->TexCoord[0], coord, flags, 4, &ctx->EvalMap.Map1Texture4 ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } else if (ctx->Eval.Map1TextureCoord3) { eval1_4f( &tmp->TexCoord[0], coord, flags, 3, &ctx->EvalMap.Map1Texture3 ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } else if (ctx->Eval.Map1TextureCoord2) { eval1_4f( &tmp->TexCoord[0], coord, flags, 2, &ctx->EvalMap.Map1Texture2 ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } else if (ctx->Eval.Map1TextureCoord1) { eval1_4f( &tmp->TexCoord[0], coord, flags, 1, &ctx->EvalMap.Map1Texture1 ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } } if (any_eval2) { if (ctx->Eval.Map2TextureCoord4) { eval2_4f( &tmp->TexCoord[0], coord, flags, 4, &ctx->EvalMap.Map2Texture4 ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } else if (ctx->Eval.Map2TextureCoord3) { eval2_4f( &tmp->TexCoord[0], coord, flags, 3, &ctx->EvalMap.Map2Texture3 ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } else if (ctx->Eval.Map2TextureCoord2) { eval2_4f( &tmp->TexCoord[0], coord, flags, 2, &ctx->EvalMap.Map2Texture2 ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } else if (ctx->Eval.Map2TextureCoord1) { eval2_4f( &tmp->TexCoord[0], coord, flags, 1, &ctx->EvalMap.Map2Texture1 ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } } } if (req & VERT_NORMAL_BIT) { GLuint generated = 0; if (copycount) { /* fprintf(stderr, "%s: Copy normals\n", __FUNCTION__); */ copy_3f( store->Attrib[VERT_ATTRIB_NORMAL] + IM->CopyStart, tmp->Normal.data, copycount ); } tmp->Normal.data = store->Attrib[VERT_ATTRIB_NORMAL] + IM->CopyStart; tmp->Normal.start = (GLfloat *)tmp->Normal.data; if (ctx->Eval.Map1Normal && any_eval1) { eval1_norm( &tmp->Normal, coord, flags, &ctx->EvalMap.Map1Normal ); generated |= VERT_EVAL_C1|VERT_EVAL_P1; } if (ctx->Eval.Map2Normal && any_eval2) { eval2_norm( &tmp->Normal, coord, flags, &ctx->EvalMap.Map2Normal ); generated |= VERT_EVAL_C2|VERT_EVAL_P2; } } /* In the AutoNormal case, the copy and assignment of tmp->NormalPtr * are done above. */ if (req & VERT_OBJ_BIT) { if (copycount) { /* This copy may already have occurred when eliminating * glEvalPoint calls: */ if (coord != store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart) { copy_4f( store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart, tmp->Obj.data, copycount ); } } else { tmp->Obj.size = 0; } tmp->Obj.data = store->Attrib[VERT_ATTRIB_POS] + IM->CopyStart; tmp->Obj.start = (GLfloat *)tmp->Obj.data; /* Note: Normal data is already prepared above. */ if (any_eval1) { if (ctx->Eval.Map1Vertex4) { eval1_4f( &tmp->Obj, coord, flags, 4, &ctx->EvalMap.Map1Vertex4 ); } else if (ctx->Eval.Map1Vertex3) { eval1_4f( &tmp->Obj, coord, flags, 3, &ctx->EvalMap.Map1Vertex3 ); } } if (any_eval2) { if (ctx->Eval.Map2Vertex4) { if (ctx->Eval.AutoNormal && (req & VERT_NORMAL_BIT)) eval2_obj_norm( &tmp->Obj, &tmp->Normal, coord, flags, 4, &ctx->EvalMap.Map2Vertex4 ); else eval2_4f( &tmp->Obj, coord, flags, 4, &ctx->EvalMap.Map2Vertex4 ); } else if (ctx->Eval.Map2Vertex3) { if (ctx->Eval.AutoNormal && (req & VERT_NORMAL_BIT)) eval2_obj_norm( &tmp->Obj, &tmp->Normal, coord, flags, 3, &ctx->EvalMap.Map2Vertex3 ); else eval2_4f( &tmp->Obj, coord, flags, 3, &ctx->EvalMap.Map2Vertex3 ); } } } /* Calculate new IM->Elts, IM->Primitive, IM->PrimitiveLength for * the case where vertex maps are not enabled for some received * eval coordinates. In this case those slots in the immediate * must be ignored. */ if (purge_flags) { GLuint vertex = VERT_OBJ_BIT|(VERT_EVAL_ANY & ~purge_flags); GLuint last_new_prim = 0; GLuint new_prim_length = 0; GLuint next_old_prim = 0; struct vertex_buffer *VB = &tnl->vb; GLuint i,j,count = VB->Count; /* fprintf(stderr, "PURGING\n"); */ for (i = 0, j = 0 ; i < count ; i++) { if (flags[i] & vertex) { store->Elt[j++] = i; new_prim_length++; } if (i == next_old_prim) { next_old_prim += VB->PrimitiveLength[i]; VB->PrimitiveLength[last_new_prim] = new_prim_length; VB->Primitive[j] = VB->Primitive[i]; last_new_prim = j; } } VB->Elts = store->Elt; _tnl_get_purged_copy_verts( ctx, store ); } /* Produce new flags array: */ { GLuint i; GLuint count = tnl->vb.Count + 1; copy_1ui( store->Flag, flags, count ); tnl->vb.Flag = store->Flag; for (i = 0 ; i < count ; i++) store->Flag[i] |= req; IM->Evaluated = req; /* hack for copying. */ } }