csrDraw.h

   1 // Copyright (C) 2021 Harry Godden (hgn)
   2
   3 // Extremely simple software renderer. Only supports orthographic
   4 //=======================================================================================================================
   5
   6 typedef struct csr_target csr_target;
   7 typedef struct csr_filter csr_filter;
   8 typedef struct csr_shader csr_shader;
   9 typedef enum EMSAA EMSAA;
  10
  11 typedef void (* csr_frag_program)( void *, vmf_vert[3], float, float, float );
  12 typedef void (* csr_frag_clear)( void * );
  13
  14 // API
  15 //=======================================================================================================================
  16
  17 // Create a render target. Resolution, msaa, and shader must be known at this point!
  18 void csr_create_target( csr_target *rt, u32 x, u32 y, EMSAA aa, csr_shader *shader );
  19 void csr_rt_clear( csr_target *rt );
  20 void csr_rt_free( csr_target *rt );
  21
  22 // Refit bounds so that it is square, and crops to center with padding
  23 void csr_auto_fit( csr_target *rt, float padding );
  24
  25 // Run this after bounds have been adjusted on the RT to update the size of the msaa
  26 // Auto fit will call this.
  27 void csr_update_subsamples( csr_target *rt );
  28
  29 // Write CS:GO radar txt
  30 void csr_write_txt( char const *path, const char *name, csr_target *rt  );
  31
  32 // Render calls
  33 // ------------
  34
  35 // Render a finalzied triangle into target. Coordinates are world space
  36 void simple_raster( csr_target *rt, vmf_vert tri[3] );
  37
  38 // Draw a batch of triangles with an affine world transformation
  39 void csr_draw( csr_target *rt, vmf_vert *triangles, u32 triangle_count, m4x3f transform );
  40
  41 // Draw VMF with filtering options. Will automatically branch into instances
  42 // You should call this with the last two recursive arguments (prev,inst), set to NULL
  43 //
  44 // Filter is optional, it can be st to NULL to just render everything.
  45 void csr_vmf_render( csr_target *rt, vmf_map *map, vdf_node *root, csr_filter *filter, m4x3f prev, m4x3f inst );
  46
  47 // Obsolete
  48 void csr_rt_save_buffers( csr_target *rt, const char *basename, const char *subname );
  49
  50 // Implementation
  51 //=======================================================================================================================
  52
  53 struct csr_shader
  54 {
  55         u32 stride;
  56         csr_frag_program frag;
  57         csr_frag_clear clear;
  58 };
  59
  60 struct csr_target
  61 {
  62         void  *colour;
  63         float *depth;
  64
  65         u32 x, y;
  66         boxf bounds;
  67         float scale;
  68
  69         v2f subsamples[ 8 ];
  70         int num_samples;
  71         v2f *sample_src;
  72
  73         csr_shader *shader;
  74 };
  75
  76 struct csr_filter
  77 {
  78         const char *visgroup;           // Limit to this visgroup only
  79         const char *classname;          // Limit to this exact classname. will not draw world
  80
  81         int compute_bounds_only;
  82 };
  83
  84 enum EMSAA
  85 {
  86         k_EMSAA_none,
  87         k_EMSAA_2x2,
  88         k_EMSAA_RGSS,
  89         k_EMSAA_8R
  90 };
  91
  92 #ifdef CSR_EXECUTABLE
  93
  94 // MSAA patterns
  95 v2f csr_msaa_1[] =
  96 {
  97         {0.f, 0.f}
  98 };
  99
 100 // XX
 101 // XX
 102 v2f csr_msaa_2x2[] =
 103 {
 104         {  0x0.4p0f,  0x0.4p0f },
 105         {  0x0.4p0f, -0x0.4p0f },
 106         { -0x0.4p0f, -0x0.4p0f },
 107         { -0x0.4p0f,  0x0.4p0f }
 108 };
 109
 110 //   X
 111 // X
 112 //    X
 113 // X
 114 v2f csr_msaa_2x2rgss[] =
 115 {
 116         {  0x0.2p0f,  0x0.6p0f },
 117         { -0x0.6p0f,  0x0.2p0f },
 118         { -0x0.2p0f, -0x0.6p0f },
 119         {  0x0.6p0f, -0x0.2p0f }
 120 };
 121
 122 //     X
 123 //   X
 124 // X
 125 //       X
 126 //  X
 127 //        X
 128 //      X
 129 //    X
 130 v2f csr_msaa_8rook[] =
 131 {
 132         {  0x0.1p0f,  0x0.7p0f },
 133         {  0x0.5p0f,  0x0.1p0f },
 134         {  0x0.7p0f, -0x0.3p0f },
 135         {  0x0.3p0f, -0x0.5p0f },
 136         { -0x0.1p0f, -0x0.7p0f },
 137         { -0x0.5p0f, -0x0.1p0f },
 138         { -0x0.7p0f,  0x0.3p0f },
 139         { -0x0.3p0f,  0x0.5p0f }
 140 };
 141
 142
 143 void csr_create_target( csr_target *rt, u32 x, u32 y, EMSAA aa, csr_shader *shader )
 144 {
 145         rt->x = x;
 146         rt->y = y;
 147
 148         switch( aa )
 149         {
 150                 default:
 151                 case k_EMSAA_none:
 152                         rt->num_samples = 1;
 153                         rt->sample_src = csr_msaa_1;
 154                         break;
 155
 156                 case k_EMSAA_2x2:
 157                         rt->num_samples = 4;
 158                         rt->sample_src = csr_msaa_2x2;
 159                         break;
 160
 161                 case k_EMSAA_RGSS:
 162                         rt->num_samples = 4;
 163                         rt->sample_src = csr_msaa_2x2rgss;
 164                         break;
 165
 166                 case k_EMSAA_8R:
 167                         rt->num_samples = 8;
 168                         rt->sample_src = csr_msaa_8rook;
 169                         break;
 170         }
 171
 172         rt->shader = shader;
 173         rt->depth = (float *)csr_malloc( x*y*rt->num_samples * sizeof(float) );
 174         rt->colour = csr_malloc( x * y * rt->shader->stride * rt->num_samples );
 175
 176         v3_fill( rt->bounds[0], INFINITY );
 177         v3_fill( rt->bounds[1], -INFINITY );
 178 }
 179
 180 void csr_update_subsamples( csr_target *rt )
 181 {
 182         float range_x = (rt->bounds[1][0]-rt->bounds[0][0]);
 183         float range_y = (rt->bounds[1][1]-rt->bounds[0][1]);
 184
 185         v2f pixel_size = { range_x/(float)rt->x, range_y/(float)rt->y };
 186
 187         for( int i = 0; i < rt->num_samples; i ++ )
 188         {
 189                 v2_mul( rt->sample_src[i], pixel_size, rt->subsamples[i] );
 190         }
 191 }
 192
 193 void csr_rt_free( csr_target *rt )
 194 {
 195         free( rt->depth );
 196         free( rt->colour );
 197 }
 198
 199 void csr_rt_clear( csr_target *rt )
 200 {
 201         for( u32 i = 0; i < rt->x*rt->y*rt->num_samples; i ++ )
 202         {
 203                 rt->shader->clear( rt->colour + i * rt->shader->stride );
 204                 rt->depth[i] = 0.f;
 205         }
 206 }
 207
 208 void csr_auto_fit( csr_target *rt, float padding )
 209 {
 210         // Correct aspect ratio to be square
 211         float dx, dy, l, cx, cy;
 212
 213         dx = rt->bounds[1][0] - rt->bounds[0][0];
 214         dy = rt->bounds[1][1] - rt->bounds[0][1];
 215
 216         l = fmaxf( dx, dy ) * .5f;
 217
 218         cx = (rt->bounds[1][0] + rt->bounds[0][0]) * .5f;
 219         cy = (rt->bounds[1][1] + rt->bounds[0][1]) * .5f;
 220
 221         rt->bounds[0][0] = cx - l - padding;
 222         rt->bounds[1][0] = cx + l + padding;
 223         rt->bounds[0][1] = cy - l - padding;
 224         rt->bounds[1][1] = cy + l + padding;
 225
 226         rt->scale = l + padding;
 227
 228         csr_update_subsamples( rt );
 229 }
 230
 231 void csr_write_txt( char const *path, const char *name, csr_target *rt  )
 232 {
 233         FILE *write_ptr;
 234
 235         write_ptr = fopen( path, "w" );
 236
 237         fprintf( write_ptr, "\"%s\"\n\{\n", name );
 238         fprintf( write_ptr, "\t\"material\" \"overviews/%s\"\n", name );
 239         fprintf( write_ptr, "\t\"pos_x\" \"%.8f\"\n", rt->bounds[0][0] );
 240         fprintf( write_ptr, "\t\"pos_y\" \"%.8f\"\n", rt->bounds[0][1] );
 241         fprintf( write_ptr, "\t\"scale\" \"%.8f\"\n", rt->scale / (float)rt->x );
 242         fprintf( write_ptr, "}\n" );
 243
 244         fclose( write_ptr );
 245 }
 246
 247 void simple_raster( csr_target *rt, vmf_vert tri[3] )
 248 {
 249         // Very very simplified rasterizing algorithm
 250         v2f bmin = { 0.f, 0.f };
 251         v2f bmax = { rt->x, rt->y };
 252
 253         v2_minv( tri[0].co, tri[1].co, bmin );
 254         v2_minv( tri[2].co, bmin, bmin );
 255
 256         v2_maxv( tri[0].co, tri[1].co, bmax );
 257         v2_maxv( tri[2].co, bmax, bmax );
 258
 259         float range_x = (rt->bounds[1][0]-rt->bounds[0][0])/(float)rt->x;
 260         float range_y = (rt->bounds[1][1]-rt->bounds[0][1])/(float)rt->y;
 261
 262         int start_x = csr_min( rt->x-1, csr_max( 0,       floorf( (bmin[0]-rt->bounds[0][0])/range_x)));
 263         int end_x =   csr_max( 0,       csr_min( rt->x-1,  ceilf( (bmax[0]-rt->bounds[0][0])/range_x)));
 264         int start_y = csr_min( rt->y-1, csr_max( 0,       floorf( (bmin[1]-rt->bounds[0][1])/range_y)));
 265         int end_y =   csr_max( 0,       csr_min( rt->y-1,  ceilf( (bmax[1]-rt->bounds[0][1])/range_y)));
 266
 267         v2f v0, v1, v2, vp;
 268         float d, bca = 0.f, bcb = 0.f, bcc = 0.f;
 269
 270         v2_sub( tri[1].co, tri[0].co, v0 );
 271         v2_sub( tri[2].co, tri[0].co, v1 );
 272         v2_sub( tri[1].co, tri[2].co, v2 );
 273         d = 1.f / (v0[0]*v1[1] - v1[0]*v0[1]);
 274
 275         // Backface culling
 276         if( v2_cross( v0, v1 ) > 0.f )
 277                 return;
 278
 279         v2f trace_origin;
 280
 281         for( u32 py = start_y; py <= end_y; py ++ )
 282         {
 283                 trace_origin[1] = csr_lerpf( rt->bounds[0][1], rt->bounds[1][1], (float)py/(float)rt->y );
 284
 285                 for( u32 px = start_x; px <= end_x; px ++ )
 286                 {
 287                         u32 sample_index = (py * rt->y + px) * rt->num_samples;
 288
 289                         void *frag = rt->colour + sample_index*rt->shader->stride;
 290                         float *depth = &rt->depth[ sample_index ];
 291
 292                         trace_origin[0] = csr_lerpf( rt->bounds[0][0], rt->bounds[1][0], (float)px/(float)rt->x );
 293
 294                         // Determine coverage
 295                         for( int i = 0; i < rt->num_samples; i ++ )
 296                         {
 297                                 v3f sample_origin;
 298
 299                                 v2_add( rt->subsamples[ i ], trace_origin, sample_origin );
 300                                 v2_sub( sample_origin, tri[0].co, vp );
 301
 302                                 if( v2_cross( v0, vp ) > 0.f )
 303                                         continue;
 304                                 if( v2_cross( vp, v1 ) > 0.f )
 305                                         continue;
 306
 307                                 v2f vp2;
 308                                 v2_sub( sample_origin, tri[2].co, vp2 );
 309
 310                                 if( v2_cross( vp2, v2 ) > 0.f )
 311                                         continue;
 312
 313                                 bcb = (vp[0]*v1[1] - v1[0]*vp[1]) * d;
 314                                 bcc = (v0[0]*vp[1] - vp[0]*v0[1]) * d;
 315                                 bca = 1.f - bcb - bcc;
 316
 317                                 float hit = (tri[0].co[2] * bca + tri[1].co[2] * bcb + tri[2].co[2] * bcc) +16385.f;
 318
 319                                 if( hit > depth[i] )
 320                                 {
 321                                         depth[i] = hit;
 322                                         rt->shader->frag( frag+i*rt->shader->stride, tri, bca, bcb, bcc );
 323                                 }
 324                         }
 325                 }
 326         }
 327 }
 328
 329 void csr_draw( csr_target *rt, vmf_vert *triangles, u32 triangle_count, m4x3f transform )
 330 {
 331         m3x3f normal;
 332         vmf_vert new_tri[3];
 333
 334         // Derive normal matrix
 335         m4x3_to_3x3( transform, normal );
 336
 337         // NOTE: This isn't strictly necessary since CS:GO only uses uniform scaling.
 338         m3x3_inv_transpose( normal, normal );
 339
 340         for( u32 i = 0; i < triangle_count; i ++ )
 341         {
 342                 vmf_vert *triangle = triangles + i*3;
 343
 344                 m4x3_mulv( transform, triangle[0].co, new_tri[0].co );
 345                 m4x3_mulv( transform, triangle[1].co, new_tri[1].co );
 346                 m4x3_mulv( transform, triangle[2].co, new_tri[2].co );
 347                 m3x3_mulv( normal, triangle[0].nrm, new_tri[0].nrm );
 348                 m3x3_mulv( normal, triangle[1].nrm, new_tri[1].nrm );
 349                 m3x3_mulv( normal, triangle[2].nrm, new_tri[2].nrm );
 350
 351                 simple_raster( rt, new_tri );
 352         }
 353 }
 354
 355 void csr_vmf_render( csr_target *rt, vmf_map *map, vdf_node *root, csr_filter *filter, m4x3f prev, m4x3f inst )
 356 {
 357         m4x3f transform = M4X3_IDENTITY;
 358         vmf_solid solid;
 359         vmf_vert tri[3];
 360         boxf trf_bounds;
 361
 362         u32 group_id = 0;
 363         int filter_visgroups = 0, filter_classname = 0, compute_bounds_only = 0;
 364
 365         if( filter )
 366         {
 367                 if( filter->visgroup )
 368                 {
 369                         filter_visgroups = 1;
 370                         group_id = vmf_visgroup_id( root, filter->visgroup );
 371                 }
 372
 373                 if( filter->classname )
 374                 {
 375                         filter_classname = 1;
 376                 }
 377
 378                 compute_bounds_only = filter->compute_bounds_only;
 379         }
 380
 381         // Multiply previous transform with instance transform to create basis
 382         if( prev )
 383         {
 384                 m4x3_mul( prev, inst, transform );
 385         }
 386
 387         // Gather world brushes
 388         solidgen_ctx_init( &solid );
 389
 390         if( !filter_classname )
 391         {
 392                 vdf_node *world = vdf_next( root, "world", NULL );
 393
 394                 vdf_foreach( world, "solid", brush )
 395                 {
 396                         if( filter_visgroups && !vmf_visgroup_match( brush, group_id ) )
 397                                 continue;
 398
 399                         // TODO: heap-use-after-free
 400                         solidgen_push( &solid, brush );
 401                 }
 402         }
 403
 404         // Actual entity loop
 405         m4x3f model;
 406
 407         vdf_foreach( root, "entity", ent )
 408         {
 409                 if( filter_visgroups && !vmf_visgroup_match( ent, group_id ) )
 410                         continue;
 411
 412                 if( filter_classname )
 413                         if( strcmp( kv_get( ent, "classname", "" ), filter->classname ) )
 414                                 continue;
 415
 416                 if( ent->user & VMF_FLAG_IS_PROP )
 417                 {
 418                         // Create model transform
 419                         m4x3_identity( model );
 420
 421                         vmf_entity_transform( ent, model );
 422                         m4x3_mul( transform, model, model );
 423
 424                         // Draw model
 425                         mdl_mesh_t *mdl = &map->models[ ent->user1 ].mdl;
 426
 427                         if( compute_bounds_only )
 428                         {
 429                                 map->models[ ent->user1 ].need_load = 1;
 430                                 m4x3_expand_aabb_point( model, rt->bounds, (v3f){0.f,0.f,0.f} );
 431                         }
 432                         else
 433                         {
 434                                 for( int i = 0; i < mdl->num_indices/3; i ++ )
 435                                 {
 436                                         for( int j = 0; j < 3; j ++ )
 437                                         {
 438                                                 v3_copy( &mdl->vertices[ mdl->indices[ i*3+j ] *8 ],   tri[j].co );
 439                                                 v3_copy( &mdl->vertices[ mdl->indices[ i*3+j ] *8+3 ], tri[j].nrm );
 440                                                 tri[j].xy[0] = 0.f;
 441                                                 tri[j].xy[1] = 0.f;
 442                                         }
 443
 444                                         csr_draw( rt, tri, 1, model );
 445                                 }
 446                         }
 447                 }
 448                 else if( ent->user & VMF_FLAG_IS_INSTANCE )
 449                 {
 450                         m4x3_identity( model );
 451                         vmf_entity_transform( ent, model );
 452
 453                         csr_vmf_render( rt, map, map->cache[ ent->user1 ].root, filter, transform, model );
 454                 }
 455                 else
 456                 {
 457                         // Brush entity
 458                         vdf_foreach( ent, "solid", ent_solid )
 459                         {
 460                                 solidgen_push( &solid, ent_solid );
 461                         }
 462                 }
 463         }
 464
 465         if( compute_bounds_only )
 466         {
 467                 solidgen_bounds( &solid, trf_bounds );
 468                 m4x3_transform_aabb( transform, trf_bounds );
 469                 box_concat( rt->bounds, trf_bounds );
 470         }
 471         else
 472         {
 473                 // Draw brushes
 474                 for( int i = 0; i < csr_sb_count( solid.indices )/3; i ++ )
 475                 {
 476                         u32 * base = solid.indices + i*3;
 477
 478                         tri[0] = solid.verts[ base[0] ];
 479                         tri[1] = solid.verts[ base[1] ];
 480                         tri[2] = solid.verts[ base[2] ];
 481
 482                         csr_draw( rt, tri, 1, transform );
 483                 }
 484         }
 485
 486         solidgen_ctx_reset( &solid );
 487         solidgen_ctx_free( &solid );
 488 }
 489
 490 // Obsolete
 491 void csr_rt_save_buffers( csr_target *rt, const char *basename, const char *subname )
 492 {
 493         char output[ 512 ];
 494
 495         float *image = (float *)csr_malloc( 1024*1024*sizeof(float)*3 );
 496
 497         float contrib = 1.f/(float)rt->num_samples;
 498
 499         for( int l = 0; l < rt->x; l ++ )
 500         {
 501                 for( int x = 0; x < rt->y; x ++ )
 502                 {
 503                         float *dst = &image[ (l*1024+x)*3 ];
 504                         void *src = rt->colour + ((1023-l)*1024+x) * rt->num_samples * rt->shader->stride;
 505
 506                         v3_muls( (float *)src, contrib, dst );
 507
 508                         for( int j = 1; j < rt->num_samples; j ++ )
 509                         {
 510                                 v3_muladds( dst, (float *)(src + j*rt->shader->stride), contrib, dst );
 511                         }
 512                 }
 513         }
 514
 515         // Save position buffer
 516         strcpy( output, basename );
 517         strcat( output, "." );
 518         strcat( output, subname );
 519         strcat( output, "_position.pfm" );
 520         csr_32f_write( output, rt->x, rt->y, image );
 521
 522         free( image );
 523 }
 524
 525 #endif