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
  48 void csr_rt_save_tga( csr_target *rt, const char *path, u32 offset, u32 nc );
  49 void csr_rt_save_c32f( csr_target *rt, const char *path, u32 offset );
  50
  51 // Obsolete
  52 __attribute__ ((deprecated))
  53 void csr_rt_save_buffers( csr_target *rt, const char *basename, const char *subname );
  54
  55 // Implementation
  56 //=======================================================================================================================
  57
  58 struct csr_shader
  59 {
  60         u32 stride;
  61         csr_frag_program frag;
  62         csr_frag_clear clear;
  63 };
  64
  65 struct csr_target
  66 {
  67         void  *colour;
  68         float *depth;
  69
  70         u32 x, y;
  71         boxf bounds;
  72         float scale;
  73
  74         v2f subsamples[ 8 ];
  75         int num_samples;
  76         v2f *sample_src;
  77
  78         csr_shader *shader;
  79 };
  80
  81 struct csr_filter
  82 {
  83         const char *visgroup;           // Limit to this visgroup only
  84         const char *classname;          // Limit to this exact classname. will not draw world
  85
  86         int compute_bounds_only;
  87 };
  88
  89 enum EMSAA
  90 {
  91         k_EMSAA_none,
  92         k_EMSAA_2x2,
  93         k_EMSAA_RGSS,
  94         k_EMSAA_8R
  95 };
  96
  97 #ifdef CSR_EXECUTABLE
  98
  99 // MSAA patterns
 100 v2f csr_msaa_1[] =
 101 {
 102         {0.f, 0.f}
 103 };
 104
 105 // XX
 106 // XX
 107 v2f csr_msaa_2x2[] =
 108 {
 109         {  0x0.4p0f,  0x0.4p0f },
 110         {  0x0.4p0f, -0x0.4p0f },
 111         { -0x0.4p0f, -0x0.4p0f },
 112         { -0x0.4p0f,  0x0.4p0f }
 113 };
 114
 115 //   X
 116 // X
 117 //    X
 118 // X
 119 v2f csr_msaa_2x2rgss[] =
 120 {
 121         {  0x0.2p0f,  0x0.6p0f },
 122         { -0x0.6p0f,  0x0.2p0f },
 123         { -0x0.2p0f, -0x0.6p0f },
 124         {  0x0.6p0f, -0x0.2p0f }
 125 };
 126
 127 //     X
 128 //   X
 129 // X
 130 //       X
 131 //  X
 132 //        X
 133 //      X
 134 //    X
 135 v2f csr_msaa_8rook[] =
 136 {
 137         {  0x0.1p0f,  0x0.7p0f },
 138         {  0x0.5p0f,  0x0.1p0f },
 139         {  0x0.7p0f, -0x0.3p0f },
 140         {  0x0.3p0f, -0x0.5p0f },
 141         { -0x0.1p0f, -0x0.7p0f },
 142         { -0x0.5p0f, -0x0.1p0f },
 143         { -0x0.7p0f,  0x0.3p0f },
 144         { -0x0.3p0f,  0x0.5p0f }
 145 };
 146
 147
 148 void csr_create_target( csr_target *rt, u32 x, u32 y, EMSAA aa, csr_shader *shader )
 149 {
 150         rt->x = x;
 151         rt->y = y;
 152
 153         switch( aa )
 154         {
 155                 default:
 156                 case k_EMSAA_none:
 157                         rt->num_samples = 1;
 158                         rt->sample_src = csr_msaa_1;
 159                         break;
 160
 161                 case k_EMSAA_2x2:
 162                         rt->num_samples = 4;
 163                         rt->sample_src = csr_msaa_2x2;
 164                         break;
 165
 166                 case k_EMSAA_RGSS:
 167                         rt->num_samples = 4;
 168                         rt->sample_src = csr_msaa_2x2rgss;
 169                         break;
 170
 171                 case k_EMSAA_8R:
 172                         rt->num_samples = 8;
 173                         rt->sample_src = csr_msaa_8rook;
 174                         break;
 175         }
 176
 177         rt->shader = shader;
 178         rt->depth = (float *)csr_malloc( x*y*rt->num_samples * sizeof(float) );
 179         rt->colour = csr_malloc( x * y * rt->shader->stride * rt->num_samples );
 180
 181         v3_fill( rt->bounds[0], INFINITY );
 182         v3_fill( rt->bounds[1], -INFINITY );
 183 }
 184
 185 void csr_update_subsamples( csr_target *rt )
 186 {
 187         float range_x = (rt->bounds[1][0]-rt->bounds[0][0]);
 188         float range_y = (rt->bounds[1][1]-rt->bounds[0][1]);
 189
 190         v2f pixel_size = { range_x/(float)rt->x, range_y/(float)rt->y };
 191
 192         for( int i = 0; i < rt->num_samples; i ++ )
 193         {
 194                 v2_mul( rt->sample_src[i], pixel_size, rt->subsamples[i] );
 195         }
 196 }
 197
 198 void csr_rt_free( csr_target *rt )
 199 {
 200         free( rt->depth );
 201         free( rt->colour );
 202 }
 203
 204 void csr_rt_clear( csr_target *rt )
 205 {
 206         for( u32 i = 0; i < rt->x*rt->y*rt->num_samples; i ++ )
 207         {
 208                 rt->shader->clear( rt->colour + i * rt->shader->stride );
 209                 rt->depth[i] = 0.f;
 210         }
 211 }
 212
 213 void csr_auto_fit( csr_target *rt, float padding )
 214 {
 215         // Correct aspect ratio to be square
 216         float dx, dy, l, cx, cy;
 217
 218         dx = rt->bounds[1][0] - rt->bounds[0][0];
 219         dy = rt->bounds[1][1] - rt->bounds[0][1];
 220
 221         l = fmaxf( dx, dy ) * .5f;
 222
 223         cx = (rt->bounds[1][0] + rt->bounds[0][0]) * .5f;
 224         cy = (rt->bounds[1][1] + rt->bounds[0][1]) * .5f;
 225
 226         rt->bounds[0][0] = cx - l - padding;
 227         rt->bounds[1][0] = cx + l + padding;
 228         rt->bounds[0][1] = cy - l - padding;
 229         rt->bounds[1][1] = cy + l + padding;
 230
 231         rt->scale = l + padding;
 232
 233         csr_update_subsamples( rt );
 234 }
 235
 236 void csr_write_txt( char const *path, const char *name, csr_target *rt  )
 237 {
 238         FILE *write_ptr;
 239
 240         write_ptr = fopen( path, "w" );
 241
 242         fprintf( write_ptr, "\"%s\"\n\{\n", name );
 243         fprintf( write_ptr, "\t\"material\" \"overviews/%s\"\n", name );
 244         fprintf( write_ptr, "\t\"pos_x\" \"%.8f\"\n", rt->bounds[0][0] );
 245         fprintf( write_ptr, "\t\"pos_y\" \"%.8f\"\n", rt->bounds[0][1] );
 246         fprintf( write_ptr, "\t\"scale\" \"%.8f\"\n", rt->scale / (float)rt->x );
 247         fprintf( write_ptr, "}\n" );
 248
 249         fclose( write_ptr );
 250 }
 251
 252 void simple_raster( csr_target *rt, vmf_vert tri[3] )
 253 {
 254         // Very very simplified rasterizing algorithm
 255         v2f bmin = { 0.f, 0.f };
 256         v2f bmax = { rt->x, rt->y };
 257
 258         v2_minv( tri[0].co, tri[1].co, bmin );
 259         v2_minv( tri[2].co, bmin, bmin );
 260
 261         v2_maxv( tri[0].co, tri[1].co, bmax );
 262         v2_maxv( tri[2].co, bmax, bmax );
 263
 264         float range_x = (rt->bounds[1][0]-rt->bounds[0][0])/(float)rt->x;
 265         float range_y = (rt->bounds[1][1]-rt->bounds[0][1])/(float)rt->y;
 266
 267         int start_x = csr_min( rt->x-1, csr_max( 0,       floorf( (bmin[0]-rt->bounds[0][0])/range_x)));
 268         int end_x =   csr_max( 0,       csr_min( rt->x-1,  ceilf( (bmax[0]-rt->bounds[0][0])/range_x)));
 269         int start_y = csr_min( rt->y-1, csr_max( 0,       floorf( (bmin[1]-rt->bounds[0][1])/range_y)));
 270         int end_y =   csr_max( 0,       csr_min( rt->y-1,  ceilf( (bmax[1]-rt->bounds[0][1])/range_y)));
 271
 272         v2f v0, v1, v2, vp;
 273         float d, bca = 0.f, bcb = 0.f, bcc = 0.f;
 274
 275         v2_sub( tri[1].co, tri[0].co, v0 );
 276         v2_sub( tri[2].co, tri[0].co, v1 );
 277         v2_sub( tri[1].co, tri[2].co, v2 );
 278         d = 1.f / (v0[0]*v1[1] - v1[0]*v0[1]);
 279
 280         // Backface culling
 281         if( v2_cross( v0, v1 ) > 0.f )
 282                 return;
 283
 284         v2f trace_origin;
 285
 286         for( u32 py = start_y; py <= end_y; py ++ )
 287         {
 288                 trace_origin[1] = csr_lerpf( rt->bounds[0][1], rt->bounds[1][1], (float)py/(float)rt->y );
 289
 290                 for( u32 px = start_x; px <= end_x; px ++ )
 291                 {
 292                         u32 sample_index = ((rt->y-py-1)*rt->x+px) * rt->num_samples;
 293
 294                         void *frag = rt->colour + sample_index*rt->shader->stride;
 295                         float *depth = &rt->depth[ sample_index ];
 296
 297                         trace_origin[0] = csr_lerpf( rt->bounds[0][0], rt->bounds[1][0], (float)px/(float)rt->x );
 298
 299                         // Determine coverage
 300                         for( int i = 0; i < rt->num_samples; i ++ )
 301                         {
 302                                 v3f sample_origin;
 303
 304                                 v2_add( rt->subsamples[ i ], trace_origin, sample_origin );
 305                                 v2_sub( sample_origin, tri[0].co, vp );
 306
 307                                 if( v2_cross( v0, vp ) > 0.f )
 308                                         continue;
 309                                 if( v2_cross( vp, v1 ) > 0.f )
 310                                         continue;
 311
 312                                 v2f vp2;
 313                                 v2_sub( sample_origin, tri[2].co, vp2 );
 314
 315                                 if( v2_cross( vp2, v2 ) > 0.f )
 316                                         continue;
 317
 318                                 bcb = (vp[0]*v1[1] - v1[0]*vp[1]) * d;
 319                                 bcc = (v0[0]*vp[1] - vp[0]*v0[1]) * d;
 320                                 bca = 1.f - bcb - bcc;
 321
 322                                 float hit = tri[0].co[2] * bca + tri[1].co[2] * bcb + tri[2].co[2] * bcc;
 323                                 float hit_depth = hit + 16385.f;
 324
 325                                 if( hit_depth > depth[i] && hit >= rt->bounds[0][2] && hit <= rt->bounds[1][2] )
 326                                 {
 327                                         depth[i] = hit_depth;
 328                                         rt->shader->frag( frag+i*rt->shader->stride, tri, bca, bcb, bcc );
 329                                 }
 330                         }
 331                 }
 332         }
 333 }
 334
 335 void csr_draw( csr_target *rt, vmf_vert *triangles, u32 triangle_count, m4x3f transform )
 336 {
 337         m3x3f normal;
 338         vmf_vert new_tri[3];
 339
 340         // Derive normal matrix
 341         m4x3_to_3x3( transform, normal );
 342         m3x3_inv_transpose( normal, normal );
 343
 344         for( u32 i = 0; i < triangle_count; i ++ )
 345         {
 346                 vmf_vert *triangle = triangles + i*3;
 347
 348                 m4x3_mulv( transform, triangle[0].co, new_tri[0].co );
 349                 m4x3_mulv( transform, triangle[1].co, new_tri[1].co );
 350                 m4x3_mulv( transform, triangle[2].co, new_tri[2].co );
 351
 352                 m3x3_mulv( normal, triangle[0].nrm, new_tri[0].nrm );
 353                 m3x3_mulv( normal, triangle[1].nrm, new_tri[1].nrm );
 354                 m3x3_mulv( normal, triangle[2].nrm, new_tri[2].nrm );
 355
 356                 v3_normalize( new_tri[0].nrm );
 357                 v3_normalize( new_tri[1].nrm );
 358                 v3_normalize( new_tri[2].nrm );
 359
 360                 m4x3_mulv( transform, triangles[0].origin, new_tri[0].origin );
 361
 362                 simple_raster( rt, new_tri );
 363         }
 364 }
 365
 366 void csr_vmf_render( csr_target *rt, vmf_map *map, vdf_node *root, csr_filter *filter, m4x3f prev, m4x3f inst )
 367 {
 368         m4x3f transform = M4X3_IDENTITY;
 369         vmf_solid solid;
 370         vmf_vert tri[3];
 371         boxf trf_bounds;
 372
 373         u32 group_id = 0;
 374         int filter_visgroups = 0, filter_classname = 0, compute_bounds_only = 0;
 375
 376         if( filter )
 377         {
 378                 if( filter->visgroup )
 379                 {
 380                         filter_visgroups = 1;
 381                         group_id = vmf_visgroup_id( root, filter->visgroup );
 382                 }
 383
 384                 if( filter->classname )
 385                 {
 386                         filter_classname = 1;
 387                 }
 388
 389                 compute_bounds_only = filter->compute_bounds_only;
 390         }
 391
 392         // Multiply previous transform with instance transform to create basis
 393         if( prev )
 394         {
 395                 m4x3_mul( prev, inst, transform );
 396         }
 397
 398         // Gather world brushes
 399         solidgen_ctx_init( &solid );
 400
 401         if( !filter_classname )
 402         {
 403                 vdf_node *world = vdf_next( root, "world", NULL );
 404
 405                 vdf_foreach( world, "solid", brush )
 406                 {
 407                         if( filter_visgroups && !vmf_visgroup_match( brush, group_id ) )
 408                                 continue;
 409
 410                         // TODO: heap-use-after-free
 411                         solidgen_push( &solid, brush );
 412                 }
 413         }
 414
 415         // Actual entity loop
 416         m4x3f model;
 417
 418         vdf_foreach( root, "entity", ent )
 419         {
 420                 if( filter_visgroups && !vmf_visgroup_match( ent, group_id ) )
 421                         continue;
 422
 423                 if( filter_classname )
 424                         if( strcmp( kv_get( ent, "classname", "" ), filter->classname ) )
 425                                 continue;
 426
 427                 if( ent->user & VMF_FLAG_IS_PROP )
 428                 {
 429                         // Create model transform
 430                         m4x3_identity( model );
 431
 432                         vmf_entity_transform( ent, model );
 433                         m4x3_mul( transform, model, model );
 434
 435                         // Draw model
 436                         mdl_mesh_t *mdl = &map->models[ ent->user1 ].mdl;
 437
 438                         if( compute_bounds_only )
 439                         {
 440                                 map->models[ ent->user1 ].need_load = 1;
 441                                 m4x3_expand_aabb_point( model, rt->bounds, (v3f){0.f,0.f,0.f} );
 442                         }
 443                         else
 444                         {
 445                                 for( int i = 0; i < mdl->num_indices/3; i ++ )
 446                                 {
 447                                         for( int j = 0; j < 3; j ++ )
 448                                         {
 449                                                 v3_copy( &mdl->vertices[ mdl->indices[ i*3+j ] *8 ],   tri[j].co );
 450                                                 v3_copy( &mdl->vertices[ mdl->indices[ i*3+j ] *8+3 ], tri[j].nrm );
 451                                                 v3_zero( tri[j].origin );
 452                                         }
 453
 454                                         csr_draw( rt, tri, 1, model );
 455                                 }
 456                         }
 457                 }
 458                 else if( ent->user & VMF_FLAG_IS_INSTANCE )
 459                 {
 460                         m4x3_identity( model );
 461                         vmf_entity_transform( ent, model );
 462
 463                         csr_vmf_render( rt, map, map->cache[ ent->user1 ].root, filter, transform, model );
 464                 }
 465                 else
 466                 {
 467                         // Brush entity
 468                         vdf_foreach( ent, "solid", ent_solid )
 469                         {
 470                                 solidgen_push( &solid, ent_solid );
 471                         }
 472                 }
 473         }
 474
 475         if( compute_bounds_only )
 476         {
 477                 solidgen_bounds( &solid, trf_bounds );
 478                 m4x3_transform_aabb( transform, trf_bounds );
 479                 box_concat( rt->bounds, trf_bounds );
 480         }
 481         else
 482         {
 483                 // Draw brushes
 484                 for( int i = 0; i < csr_sb_count( solid.indices )/3; i ++ )
 485                 {
 486                         u32 * base = solid.indices + i*3;
 487
 488                         tri[0] = solid.verts[ base[0] ];
 489                         tri[1] = solid.verts[ base[1] ];
 490                         tri[2] = solid.verts[ base[2] ];
 491
 492                         csr_draw( rt, tri, 1, transform );
 493                 }
 494         }
 495
 496         solidgen_ctx_reset( &solid );
 497         solidgen_ctx_free( &solid );
 498 }
 499
 500 // Obsolete
 501 void csr_rt_save_buffers( csr_target *rt, const char *basename, const char *subname )
 502 {
 503         char output[ 512 ];
 504
 505         float *image = (float *)csr_malloc( 1024*1024*sizeof(float)*3 );
 506
 507         float contrib = 1.f/(float)rt->num_samples;
 508
 509         for( int l = 0; l < rt->x; l ++ )
 510         {
 511                 for( int x = 0; x < rt->y; x ++ )
 512                 {
 513                         float *dst = &image[ (l*1024+x)*3 ];
 514                         void *src = rt->colour + ((1023-l)*1024+x) * rt->num_samples * rt->shader->stride;
 515
 516                         v3_muls( (float *)src, contrib, dst );
 517
 518                         for( int j = 1; j < rt->num_samples; j ++ )
 519                         {
 520                                 v3_muladds( dst, (float *)(src + j*rt->shader->stride), contrib, dst );
 521                         }
 522                 }
 523         }
 524
 525         // Save position buffer
 526         strcpy( output, basename );
 527         strcat( output, "." );
 528         strcat( output, subname );
 529         strcat( output, "_position.pfm" );
 530         csr_32f_write( output, rt->x, rt->y, image );
 531
 532         free( image );
 533 }
 534
 535 // ALWAYS RGB32
 536 void csr_rt_save_c32f( csr_target *rt, const char *path, u32 offset )
 537 {
 538         float *image = (float *)csr_malloc( rt->x*rt->y*3*sizeof(float) );
 539
 540         float contrib = 255.f/(float)rt->num_samples;
 541
 542         for( int i = 0; i < rt->x*rt->y; i ++ )
 543         {
 544                 void *src = rt->colour + offset + i * rt->num_samples * rt->shader->stride;
 545                 float *dst = image + i*3;
 546
 547                 v3_zero( dst );
 548                 for( int k = 0; k < rt->num_samples; k ++ )
 549                 {
 550                         v3_muladds( dst, (float *)(src + k*rt->shader->stride), contrib, dst );
 551                 }
 552         }
 553
 554         csr_32f_write( path, rt->x, rt->y, image );
 555         free( image );
 556 }
 557
 558 // Save floating point buffer to tga. Must be in range (0-1)
 559 // Offset and stride are in bytes
 560 void csr_rt_save_tga( csr_target *rt, const char *path, u32 offset, u32 nc )
 561 {
 562         u8 *image = (u8 *)csr_malloc( rt->x*rt->y * 4 );
 563
 564         float contrib = 255.f/(float)rt->num_samples;
 565
 566         for( int i = 0; i < rt->x*rt->y; i ++ )
 567         {
 568                 void *src = rt->colour + offset + i * rt->num_samples * rt->shader->stride;
 569                 u8 *dst = image + i*4;
 570
 571                 v4f accum = { 0.f, 0.f, 0.f, 0.f };
 572
 573                 for( int k = 0; k < rt->num_samples; k ++ )
 574                 {
 575                         float *src_sample = (float *)(src + k*rt->shader->stride);
 576
 577                         for( int j = 0; j < nc; j ++ )
 578                         {
 579                                 accum[ j ] += src_sample[ j ] * contrib;
 580                         }
 581                 }
 582
 583                 // TODO: Clamp this value
 584                 dst[0] = accum[0];
 585                 dst[1] = accum[1];
 586                 dst[2] = accum[2];
 587                 dst[3] = accum[3];
 588         }
 589
 590         csr_tga_write( path, rt->x, rt->y, nc, image );
 591         free( image );
 592 }
 593
 594 #endif