dlls/wined3d/baseshader.c

   1 /*
   2  * shaders implementation
   3  *
   4  * Copyright 2002-2003 Jason Edmeades
   5  * Copyright 2002-2003 Raphael Junqueira
   6  * Copyright 2005 Oliver Stieber
   7  *
   8  * This library is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * This library is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with this library; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  */
  22
  23 #include "config.h"
  24 #include <string.h>
  25 #include <stdio.h>
  26 #include "wined3d_private.h"
  27
  28 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  29
  30 #define GLNAME_REQUIRE_GLSL  ((const char *)1)
  31
  32 inline static BOOL shader_is_version_token(DWORD token) {
  33     return shader_is_pshader_version(token) ||
  34            shader_is_vshader_version(token);
  35 }
  36
  37 int shader_addline(
  38     SHADER_BUFFER* buffer,
  39     const char *format, ...) {
  40
  41     char* base = buffer->buffer + buffer->bsize;
  42     int rc;
  43
  44     va_list args;
  45     va_start(args, format);
  46     rc = vsnprintf(base, SHADER_PGMSIZE - 1 - buffer->bsize, format, args);
  47     va_end(args);
  48
  49     if (rc < 0 ||                                   /* C89 */
  50         rc > SHADER_PGMSIZE - 1 - buffer->bsize) {  /* C99 */
  51
  52         ERR("The buffer allocated for the shader program string "
  53             "is too small at %d bytes.\n", SHADER_PGMSIZE);
  54         buffer->bsize = SHADER_PGMSIZE - 1;
  55         return -1;
  56     }
  57
  58     buffer->bsize += rc;
  59     buffer->lineNo++;
  60     TRACE("GL HW (%u, %u) : %s", buffer->lineNo, buffer->bsize, base);
  61     return 0;
  62 }
  63
  64 const SHADER_OPCODE* shader_get_opcode(
  65     IWineD3DBaseShader *iface, const DWORD code) {
  66
  67     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl*) iface;
  68
  69     DWORD i = 0;
  70     DWORD version = This->baseShader.version;
  71     DWORD hex_version = This->baseShader.hex_version;
  72     const SHADER_OPCODE *shader_ins = This->baseShader.shader_ins;
  73
  74     /** TODO: use dichotomic search */
  75     while (NULL != shader_ins[i].name) {
  76         if (((code & D3DSI_OPCODE_MASK) == shader_ins[i].opcode) &&
  77             (((hex_version >= shader_ins[i].min_version) && (hex_version <= shader_ins[i].max_version)) ||
  78             ((shader_ins[i].min_version == 0) && (shader_ins[i].max_version == 0)))) {
  79             return &shader_ins[i];
  80         }
  81         ++i;
  82     }
  83     FIXME("Unsupported opcode %lx(%ld) masked %lx version %ld\n",
  84        code, code, code & D3DSI_OPCODE_MASK, version);
  85     return NULL;
  86 }
  87
  88 /* Note: For vertex shaders,
  89  * texUsed = addrUsed, and
  90  * D3DSPR_TEXTURE = D3DSPR_ADDR.
  91  *
  92  * Also note that this does not count the loop register
  93  * as an address register. */
  94
  95 void shader_get_registers_used(
  96     IWineD3DBaseShader *iface,
  97     CONST DWORD* pToken) {
  98
  99     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 100     DWORD* tempsUsed = &This->baseShader.temps_used;
 101     DWORD* texUsed = &This->baseShader.textures_used;
 102
 103     if (pToken == NULL)
 104         return;
 105
 106     *tempsUsed = 0;
 107     *texUsed = 0;
 108
 109     while (D3DVS_END() != *pToken) {
 110         CONST SHADER_OPCODE* curOpcode;
 111
 112         /* Skip version */
 113         if (shader_is_version_token(*pToken)) {
 114              ++pToken;
 115              continue;
 116
 117         /* Skip comments */
 118         } else if (shader_is_comment(*pToken)) {
 119              DWORD comment_len = (*pToken & D3DSI_COMMENTSIZE_MASK) >> D3DSI_COMMENTSIZE_SHIFT;
 120              ++pToken;
 121              pToken += comment_len;
 122              continue;
 123         }
 124
 125         /* Fetch opcode */
 126         curOpcode = shader_get_opcode(iface, *pToken);
 127         ++pToken;
 128
 129         /* Unhandled opcode, and its parameters */
 130         if (NULL == curOpcode) {
 131            while (*pToken & 0x80000000)
 132                ++pToken;
 133            continue;
 134
 135         /* Skip declarations (for now) */
 136         } else if (D3DSIO_DCL == curOpcode->opcode) {
 137             pToken += curOpcode->num_params;
 138             continue;
 139
 140         /* Skip definitions (for now) */
 141         } else if (D3DSIO_DEF == curOpcode->opcode) {
 142             pToken += curOpcode->num_params;
 143             continue;
 144
 145         /* Set texture registers, and temporary registers */
 146         } else {
 147             int i;
 148
 149             for (i = 0; i < curOpcode->num_params; ++i) {
 150                 DWORD regtype = (((*pToken) & D3DSP_REGTYPE_MASK) >> D3DSP_REGTYPE_SHIFT);
 151                 DWORD reg = (*pToken) & D3DSP_REGNUM_MASK;
 152                 if (D3DSPR_TEXTURE == regtype)
 153                     *texUsed |= (1 << reg);
 154                 if (D3DSPR_TEMP == regtype)
 155                     *tempsUsed |= (1 << reg);
 156                 ++pToken;
 157              }
 158         }
 159     }
 160 }
 161
 162 void shader_program_dump_decl_usage(
 163     DWORD decl,
 164     DWORD param) {
 165
 166     DWORD regtype = shader_get_regtype(param);
 167     TRACE("dcl_");
 168
 169     if (regtype == D3DSPR_SAMPLER) {
 170         DWORD ttype = decl & D3DSP_TEXTURETYPE_MASK;
 171
 172         switch (ttype) {
 173             case D3DSTT_2D: TRACE("2d"); break;
 174             case D3DSTT_CUBE: TRACE("cube"); break;
 175             case D3DSTT_VOLUME: TRACE("volume"); break;
 176             default: TRACE("unknown_ttype(%08lx)", ttype);
 177        }
 178
 179     } else {
 180
 181         DWORD usage = decl & D3DSP_DCL_USAGE_MASK;
 182         DWORD idx = (decl & D3DSP_DCL_USAGEINDEX_MASK) >> D3DSP_DCL_USAGEINDEX_SHIFT;
 183
 184         switch(usage) {
 185         case D3DDECLUSAGE_POSITION:
 186             TRACE("%s%ld", "position", idx);
 187             break;
 188         case D3DDECLUSAGE_BLENDINDICES:
 189             TRACE("%s", "blend");
 190             break;
 191         case D3DDECLUSAGE_BLENDWEIGHT:
 192             TRACE("%s", "weight");
 193             break;
 194         case D3DDECLUSAGE_NORMAL:
 195             TRACE("%s%ld", "normal", idx);
 196             break;
 197         case D3DDECLUSAGE_PSIZE:
 198             TRACE("%s", "psize");
 199             break;
 200         case D3DDECLUSAGE_COLOR:
 201             if(idx == 0)  {
 202                 TRACE("%s", "color");
 203             } else {
 204                 TRACE("%s%ld", "specular", (idx - 1));
 205             }
 206             break;
 207         case D3DDECLUSAGE_TEXCOORD:
 208             TRACE("%s%ld", "texture", idx);
 209             break;
 210         case D3DDECLUSAGE_TANGENT:
 211             TRACE("%s", "tangent");
 212             break;
 213         case D3DDECLUSAGE_BINORMAL:
 214             TRACE("%s", "binormal");
 215             break;
 216         case D3DDECLUSAGE_TESSFACTOR:
 217             TRACE("%s", "tessfactor");
 218             break;
 219         case D3DDECLUSAGE_POSITIONT:
 220             TRACE("%s%ld", "positionT", idx);
 221             break;
 222         case D3DDECLUSAGE_FOG:
 223             TRACE("%s", "fog");
 224             break;
 225         case D3DDECLUSAGE_DEPTH:
 226             TRACE("%s", "depth");
 227             break;
 228         case D3DDECLUSAGE_SAMPLE:
 229             TRACE("%s", "sample");
 230             break;
 231         default:
 232             FIXME("unknown_semantics(%08lx)", usage);
 233         }
 234     }
 235 }
 236
 237 void shader_dump_param(
 238     IWineD3DBaseShader *iface,
 239     const DWORD param,
 240     int input) {
 241
 242     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 243     static const char* rastout_reg_names[] = { "oPos", "oFog", "oPts" };
 244     char swizzle_reg_chars[4];
 245
 246     DWORD reg = param & D3DSP_REGNUM_MASK;
 247     DWORD regtype = shader_get_regtype(param);
 248
 249     /* There are some minor differences between pixel and vertex shaders */
 250     BOOL pshader = shader_is_pshader_version(This->baseShader.hex_version);
 251
 252     /* For one, we'd prefer color components to be shown for pshaders.
 253      * FIXME: use the swizzle function for this */
 254
 255     swizzle_reg_chars[0] = pshader? 'r': 'x';
 256     swizzle_reg_chars[1] = pshader? 'g': 'y';
 257     swizzle_reg_chars[2] = pshader? 'b': 'z';
 258     swizzle_reg_chars[3] = pshader? 'a': 'w';
 259
 260     if (input) {
 261         if ( ((param & D3DSP_SRCMOD_MASK) == D3DSPSM_NEG) ||
 262              ((param & D3DSP_SRCMOD_MASK) == D3DSPSM_BIASNEG) ||
 263              ((param & D3DSP_SRCMOD_MASK) == D3DSPSM_SIGNNEG) ||
 264              ((param & D3DSP_SRCMOD_MASK) == D3DSPSM_X2NEG) )
 265             TRACE("-");
 266         else if ((param & D3DSP_SRCMOD_MASK) == D3DSPSM_COMP)
 267             TRACE("1-");
 268     }
 269
 270     switch (regtype) {
 271         case D3DSPR_TEMP:
 272             TRACE("r%lu", reg);
 273             break;
 274         case D3DSPR_INPUT:
 275             TRACE("v%lu", reg);
 276             break;
 277         case D3DSPR_CONST:
 278             TRACE("c%s%lu", (param & D3DVS_ADDRMODE_RELATIVE) ? "a0.x + " : "", reg);
 279             break;
 280         case D3DSPR_TEXTURE: /* vs: case D3DSPR_ADDR */
 281             TRACE("%c%lu", (pshader? 't':'a'), reg);
 282             break;
 283         case D3DSPR_RASTOUT:
 284             TRACE("%s", rastout_reg_names[reg]);
 285             break;
 286         case D3DSPR_COLOROUT:
 287             TRACE("oC%lu", reg);
 288             break;
 289         case D3DSPR_DEPTHOUT:
 290             TRACE("oDepth");
 291             break;
 292         case D3DSPR_ATTROUT:
 293             TRACE("oD%lu", reg);
 294             break;
 295         case D3DSPR_TEXCRDOUT:
 296             TRACE("oT%lu", reg);
 297             break;
 298         case D3DSPR_CONSTINT:
 299             TRACE("i%s%lu", (param & D3DVS_ADDRMODE_RELATIVE) ? "a0.x + " : "", reg);
 300             break;
 301         case D3DSPR_CONSTBOOL:
 302             TRACE("b%s%lu", (param & D3DVS_ADDRMODE_RELATIVE) ? "a0.x + " : "", reg);
 303             break;
 304         case D3DSPR_LABEL:
 305             TRACE("l%lu", reg);
 306             break;
 307         case D3DSPR_LOOP:
 308             TRACE("aL%s%lu", (param & D3DVS_ADDRMODE_RELATIVE) ? "a0.x + " : "", reg);
 309             break;
 310         case D3DSPR_SAMPLER:
 311             TRACE("s%lu", reg);
 312             break;
 313         default:
 314             TRACE("unhandled_rtype(%lx)", regtype);
 315             break;
 316    }
 317
 318    if (!input) {
 319        /* operand output (for modifiers and shift, see dump_ins_modifiers) */
 320
 321        if ((param & D3DSP_WRITEMASK_ALL) != D3DSP_WRITEMASK_ALL) {
 322            TRACE(".");
 323            if (param & D3DSP_WRITEMASK_0) TRACE("%c", swizzle_reg_chars[0]);
 324            if (param & D3DSP_WRITEMASK_1) TRACE("%c", swizzle_reg_chars[1]);
 325            if (param & D3DSP_WRITEMASK_2) TRACE("%c", swizzle_reg_chars[2]);
 326            if (param & D3DSP_WRITEMASK_3) TRACE("%c", swizzle_reg_chars[3]);
 327        }
 328
 329    } else {
 330         /** operand input */
 331         DWORD swizzle = (param & D3DSP_SWIZZLE_MASK) >> D3DSP_SWIZZLE_SHIFT;
 332         DWORD swizzle_r = swizzle & 0x03;
 333         DWORD swizzle_g = (swizzle >> 2) & 0x03;
 334         DWORD swizzle_b = (swizzle >> 4) & 0x03;
 335         DWORD swizzle_a = (swizzle >> 6) & 0x03;
 336
 337         if (0 != (param & D3DSP_SRCMOD_MASK)) {
 338             DWORD mask = param & D3DSP_SRCMOD_MASK;
 339             /*TRACE("_modifier(0x%08lx) ", mask);*/
 340             switch (mask) {
 341                 case D3DSPSM_NONE:    break;
 342                 case D3DSPSM_NEG:     break;
 343                 case D3DSPSM_BIAS:    TRACE("_bias"); break;
 344                 case D3DSPSM_BIASNEG: TRACE("_bias"); break;
 345                 case D3DSPSM_SIGN:    TRACE("_bx2"); break;
 346                 case D3DSPSM_SIGNNEG: TRACE("_bx2"); break;
 347                 case D3DSPSM_COMP:    break;
 348                 case D3DSPSM_X2:      TRACE("_x2"); break;
 349                 case D3DSPSM_X2NEG:   TRACE("_x2"); break;
 350                 case D3DSPSM_DZ:      TRACE("_dz"); break;
 351                 case D3DSPSM_DW:      TRACE("_dw"); break;
 352                 default:
 353                     TRACE("_unknown(0x%08lx)", mask);
 354             }
 355         }
 356
 357         /**
 358         * swizzle bits fields:
 359         *  RRGGBBAA
 360         */
 361         if ((D3DVS_NOSWIZZLE >> D3DVS_SWIZZLE_SHIFT) != swizzle) { /* ! D3DVS_NOSWIZZLE == 0xE4 << D3DVS_SWIZZLE_SHIFT */
 362             if (swizzle_r == swizzle_g &&
 363                 swizzle_r == swizzle_b &&
 364                 swizzle_r == swizzle_a) {
 365                     TRACE(".%c", swizzle_reg_chars[swizzle_r]);
 366             } else {
 367                 TRACE(".%c%c%c%c",
 368                 swizzle_reg_chars[swizzle_r],
 369                 swizzle_reg_chars[swizzle_g],
 370                 swizzle_reg_chars[swizzle_b],
 371                 swizzle_reg_chars[swizzle_a]);
 372             }
 373         }
 374     }
 375 }
 376
 377 /** Generate the variable & register declarations for the ARB_vertex_program
 378     output target */
 379 void generate_arb_declarations(IWineD3DBaseShader *iface, SHADER_BUFFER* buffer) {
 380
 381     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 382     DWORD i;
 383
 384     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 385         if (This->baseShader.temps_used & (1 << i))
 386             shader_addline(buffer, "TEMP R%lu;\n", i);
 387     }
 388
 389     for (i = 0; i < This->baseShader.limits.address; i++) {
 390         if (This->baseShader.textures_used & (1 << i))
 391             shader_addline(buffer, "ADDRESS A%ld;\n", i);
 392     }
 393
 394     for(i = 0; i < This->baseShader.limits.texture; i++) {
 395         if (This->baseShader.textures_used & (1 << i))
 396             shader_addline(buffer,"TEMP T%lu;\n", i);
 397     }
 398
 399     /* Texture coordinate registers must be pre-loaded */
 400     for (i = 0; i < This->baseShader.limits.texture; i++) {
 401     if (This->baseShader.textures_used & (1 << i))
 402         shader_addline(buffer, "MOV T%lu, fragment.texcoord[%lu];\n", i, i);
 403     }
 404 }
 405
 406 /** Generate the variable & register declarations for the GLSL
 407     output target */
 408 void generate_glsl_declarations(IWineD3DBaseShader *iface, SHADER_BUFFER* buffer) {
 409
 410     FIXME("GLSL not fully implemented yet.\n");
 411
 412 }
 413
 414 /** Shared code in order to generate the bulk of the shader string.
 415     Use the shader_header_fct & shader_footer_fct to add strings
 416     that are specific to pixel or vertex functions
 417     NOTE: A description of how to parse tokens can be found at:
 418           http://msdn.microsoft.com/library/default.asp?url=/library/en-us/graphics/hh/graphics/usermodedisplaydriver_shader_cc8e4e05-f5c3-4ec0-8853-8ce07c1551b2.xml.asp */
 419 void generate_base_shader(
 420     IWineD3DBaseShader *iface,
 421     SHADER_BUFFER* buffer,
 422     CONST DWORD* pFunction) {
 423
 424     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 425     const DWORD *pToken = pFunction;
 426     const SHADER_OPCODE *curOpcode = NULL;
 427     DWORD i;
 428
 429     /* Initialize current parsing state */
 430     This->baseShader.parse_state.current_row = 0;
 431
 432     /* First pass: figure out which temporary and texture registers are used */
 433     shader_get_registers_used(iface, pToken);
 434     TRACE("Texture/Address registers used: %#lx, Temp registers used %#lx\n",
 435           This->baseShader.textures_used, This->baseShader.temps_used);
 436
 437     /* TODO: check register usage against GL/Directx limits, and fail if they're exceeded
 438         nUseAddressRegister < = GL_MAX_PROGRAM_ADDRESS_REGISTERS_AR
 439         nUseTempRegister    <=  GL_MAX_PROGRAM_LOCAL_PARAMETERS_ARB
 440     */
 441
 442     /* Pre-declare registers */
 443     if (USING_GLSL)
 444         generate_glsl_declarations(iface, buffer);
 445     else
 446         generate_arb_declarations(iface, buffer);
 447
 448     /* Second pass, process opcodes */
 449     if (NULL != pToken) {
 450         while (D3DPS_END() != *pToken) {
 451
 452             /* Skip version token */
 453             if (shader_is_version_token(*pToken)) {
 454                 ++pToken;
 455                 continue;
 456             }
 457
 458             /* Skip comment tokens */
 459             if (shader_is_comment(*pToken)) {
 460                 DWORD comment_len = (*pToken & D3DSI_COMMENTSIZE_MASK) >> D3DSI_COMMENTSIZE_SHIFT;
 461                 ++pToken;
 462                 TRACE("#%s\n", (char*)pToken);
 463                 pToken += comment_len;
 464                 continue;
 465             }
 466
 467             /* Read opcode */
 468             curOpcode = shader_get_opcode(iface, *pToken);
 469             ++pToken;
 470
 471             /* Unknown opcode and its parameters */
 472             if (NULL == curOpcode) {
 473                 while (*pToken & 0x80000000) { /* TODO: Think of a sensible name for 0x80000000 */
 474                     FIXME("unrecognized opcode: %08lx\n", *pToken);
 475                     ++pToken;
 476                 }
 477
 478             /* Using GLSL & no generator function exists */
 479             } else if (USING_GLSL && curOpcode->hw_glsl_fct == NULL) {
 480
 481                 FIXME("Token %s is not yet implemented with GLSL\n", curOpcode->name);
 482                 pToken += curOpcode->num_params;
 483
 484             /* Unhandled opcode in ARB */
 485             } else if ( !USING_GLSL && GLNAME_REQUIRE_GLSL == curOpcode->glname) {
 486
 487                 FIXME("Token %s requires greater functionality than "
 488                     "Vertex or Fragment_Program_ARB supports\n", curOpcode->name);
 489                 pToken += curOpcode->num_params;
 490
 491             /* If a generator function is set for current shader target, use it */
 492             } else if ((!USING_GLSL && curOpcode->hw_fct != NULL) ||
 493                        (USING_GLSL && curOpcode->hw_glsl_fct != NULL)) {
 494
 495                 SHADER_OPCODE_ARG hw_arg;
 496
 497                 hw_arg.shader = iface;
 498                 hw_arg.opcode = curOpcode;
 499                 hw_arg.buffer = buffer;
 500                 if (curOpcode->num_params > 0) {
 501                     hw_arg.dst = *pToken;
 502
 503                     /* FIXME: this does not account for relative address tokens */
 504                     for (i = 1; i < curOpcode->num_params; i++)
 505                        hw_arg.src[i-1] = *(pToken + i);
 506                 }
 507
 508                 /* Call appropriate function for output target */
 509                 if (USING_GLSL)
 510                     curOpcode->hw_glsl_fct(&hw_arg);
 511                 else
 512                     curOpcode->hw_fct(&hw_arg);
 513
 514                 pToken += curOpcode->num_params;
 515
 516             } else {
 517
 518                 TRACE("Found opcode D3D:%s GL:%s, PARAMS:%d,\n",
 519                 curOpcode->name, curOpcode->glname, curOpcode->num_params);
 520
 521                 /* Unless we encounter a no-op command, this opcode is unrecognized */
 522                 if (curOpcode->opcode != D3DSIO_NOP) {
 523                     FIXME("Can't handle opcode %s in hwShader\n", curOpcode->name);
 524                     pToken += curOpcode->num_params;
 525                 }
 526             }
 527         }
 528         /* TODO: What about result.depth? */
 529
 530     }
 531 }
 532
 533
 534 void shader_dump_ins_modifiers(const DWORD output) {
 535
 536     DWORD shift = (output & D3DSP_DSTSHIFT_MASK) >> D3DSP_DSTSHIFT_SHIFT;
 537     DWORD mmask = output & D3DSP_DSTMOD_MASK;
 538
 539     switch (shift) {
 540         case 0: break;
 541         case 13: TRACE("_d8"); break;
 542         case 14: TRACE("_d4"); break;
 543         case 15: TRACE("_d2"); break;
 544         case 1: TRACE("_x2"); break;
 545         case 2: TRACE("_x4"); break;
 546         case 3: TRACE("_x8"); break;
 547         default: TRACE("_unhandled_shift(%ld)", shift); break;
 548     }
 549
 550     switch(mmask) {
 551         case D3DSPDM_NONE: break;
 552         case D3DSPDM_SATURATE: TRACE("_sat"); break;
 553         case D3DSPDM_PARTIALPRECISION: TRACE("_pp"); break;
 554         case D3DSPDM_MSAMPCENTROID: TRACE("_centroid"); break;
 555         default: TRACE("_unhandled_modifier(%#lx)", mmask); break;
 556     }
 557 }
 558
 559 /* TODO: Move other shared code here */