git.oblomov.eu Git - wine/blob - dlls/wined3d/glsl_shader.c

   1 /*
   2  * GLSL pixel and vertex shader implementation
   3  *
   4  * Copyright 2006 Jason Green
   5  * Copyright 2006-2007 Henri Verbeet
   6  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
   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., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  21  */
  22
  23 /*
  24  * D3D shader asm has swizzles on source parameters, and write masks for
  25  * destination parameters. GLSL uses swizzles for both. The result of this is
  26  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  27  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  28  * mask for the destination parameter into account.
  29  */
  30
  31 #include "config.h"
  32 #include <stdio.h>
  33 #include "wined3d_private.h"
  34
  35 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  36 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  37 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  38
  39 #define GLINFO_LOCATION      (*gl_info)
  40
  41 typedef struct {
  42     char reg_name[150];
  43     char mask_str[6];
  44 } glsl_dst_param_t;
  45
  46 typedef struct {
  47     char reg_name[150];
  48     char param_str[100];
  49 } glsl_src_param_t;
  50
  51 typedef struct {
  52     const char *name;
  53     DWORD coord_mask;
  54 } glsl_sample_function_t;
  55
  56 /* GLSL shader private data */
  57 struct shader_glsl_priv {
  58     struct hash_table_t *glsl_program_lookup;
  59     struct glsl_shader_prog_link *glsl_program;
  60     GLhandleARB depth_blt_program[tex_type_count];
  61 };
  62
  63 /* Struct to maintain data about a linked GLSL program */
  64 struct glsl_shader_prog_link {
  65     struct list             vshader_entry;
  66     struct list             pshader_entry;
  67     GLhandleARB             programId;
  68     GLhandleARB             *vuniformF_locations;
  69     GLhandleARB             *puniformF_locations;
  70     GLhandleARB             vuniformI_locations[MAX_CONST_I];
  71     GLhandleARB             puniformI_locations[MAX_CONST_I];
  72     GLhandleARB             posFixup_location;
  73     GLhandleARB             bumpenvmat_location[MAX_TEXTURES];
  74     GLhandleARB             luminancescale_location[MAX_TEXTURES];
  75     GLhandleARB             luminanceoffset_location[MAX_TEXTURES];
  76     GLhandleARB             srgb_comparison_location;
  77     GLhandleARB             srgb_mul_low_location;
  78     GLhandleARB             ycorrection_location;
  79     GLenum                  vertex_color_clamp;
  80     GLhandleARB             vshader;
  81     GLhandleARB             pshader;
  82 };
  83
  84 typedef struct {
  85     GLhandleARB vshader;
  86     GLhandleARB pshader;
  87 } glsl_program_key_t;
  88
  89
  90 /** Prints the GLSL info log which will contain error messages if they exist */
  91 static void print_glsl_info_log(WineD3D_GL_Info *gl_info, GLhandleARB obj) {
  92     int infologLength = 0;
  93     char *infoLog;
  94     int i;
  95     BOOL is_spam;
  96
  97     const char *spam[] = {
  98         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
  99         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx          */
 100         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 101         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 102         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 103         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 104         "Fragment shader was successfully compiled to run on hardware.\nWARNING: 0:1: extension 'GL_ARB_draw_buffers' is not supported",
 105         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 106         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 107         "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported\n"  /* MacOS ati      */
 108     };
 109
 110     GL_EXTCALL(glGetObjectParameterivARB(obj,
 111                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 112                &infologLength));
 113
 114     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 115      * that if there are errors. */
 116     if (infologLength > 1)
 117     {
 118         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 119          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 120          */
 121         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 122         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 123         is_spam = FALSE;
 124
 125         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 126             if(strcmp(infoLog, spam[i]) == 0) {
 127                 is_spam = TRUE;
 128                 break;
 129             }
 130         }
 131         if(is_spam) {
 132             TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
 133         } else {
 134             FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
 135         }
 136         HeapFree(GetProcessHeap(), 0, infoLog);
 137     }
 138 }
 139
 140 /**
 141  * Loads (pixel shader) samplers
 142  */
 143 static void shader_glsl_load_psamplers(
 144     WineD3D_GL_Info *gl_info,
 145     IWineD3DStateBlock* iface,
 146     GLhandleARB programId) {
 147
 148     IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
 149     GLhandleARB name_loc;
 150     int i;
 151     char sampler_name[20];
 152
 153     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 154         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 155         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 156         if (name_loc != -1) {
 157             int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[i];
 158             if (mapped_unit != -1 && mapped_unit < GL_LIMITS(fragment_samplers)) {
 159                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 160                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 161                 checkGLcall("glUniform1iARB");
 162             } else {
 163                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 164             }
 165         }
 166     }
 167 }
 168
 169 static void shader_glsl_load_vsamplers(WineD3D_GL_Info *gl_info, IWineD3DStateBlock* iface, GLhandleARB programId) {
 170     IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
 171     GLhandleARB name_loc;
 172     char sampler_name[20];
 173     int i;
 174
 175     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 176         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 177         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 178         if (name_loc != -1) {
 179             int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[MAX_FRAGMENT_SAMPLERS + i];
 180             if (mapped_unit != -1 && mapped_unit < GL_LIMITS(combined_samplers)) {
 181                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 182                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 183                 checkGLcall("glUniform1iARB");
 184             } else {
 185                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 186             }
 187         }
 188     }
 189 }
 190
 191 /**
 192  * Loads floating point constants (aka uniforms) into the currently set GLSL program.
 193  * When constant_list == NULL, it will load all the constants.
 194  */
 195 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl* This, WineD3D_GL_Info *gl_info,
 196         unsigned int max_constants, float* constants, GLhandleARB *constant_locations,
 197         struct list *constant_list) {
 198     constants_entry *constant;
 199     local_constant* lconst;
 200     GLhandleARB tmp_loc;
 201     DWORD i, j, k;
 202     DWORD *idx;
 203
 204     if (TRACE_ON(d3d_shader)) {
 205         LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
 206             idx = constant->idx;
 207             j = constant->count;
 208             while (j--) {
 209                 i = *idx++;
 210                 tmp_loc = constant_locations[i];
 211                 if (tmp_loc != -1) {
 212                     TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 213                             constants[i * 4 + 0], constants[i * 4 + 1],
 214                             constants[i * 4 + 2], constants[i * 4 + 3]);
 215                 }
 216             }
 217         }
 218     }
 219
 220     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 221     if(WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1 &&
 222        shader_is_pshader_version(This->baseShader.hex_version)) {
 223         float lcl_const[4];
 224
 225         LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
 226             idx = constant->idx;
 227             j = constant->count;
 228             while (j--) {
 229                 i = *idx++;
 230                 tmp_loc = constant_locations[i];
 231                 if (tmp_loc != -1) {
 232                     /* We found this uniform name in the program - go ahead and send the data */
 233                     k = i * 4;
 234                     if(constants[k + 0] < -1.0) lcl_const[0] = -1.0;
 235                     else if(constants[k + 0] > 1.0) lcl_const[0] = 1.0;
 236                     else lcl_const[0] = constants[k + 0];
 237                     if(constants[k + 1] < -1.0) lcl_const[1] = -1.0;
 238                     else if(constants[k + 1] > 1.0) lcl_const[1] = 1.0;
 239                     else lcl_const[1] = constants[k + 1];
 240                     if(constants[k + 2] < -1.0) lcl_const[2] = -1.0;
 241                     else if(constants[k + 2] > 1.0) lcl_const[2] = 1.0;
 242                     else lcl_const[2] = constants[k + 2];
 243                     if(constants[k + 3] < -1.0) lcl_const[3] = -1.0;
 244                     else if(constants[k + 3] > 1.0) lcl_const[3] = 1.0;
 245                     else lcl_const[3] = constants[k + 3];
 246
 247                     GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, lcl_const));
 248                 }
 249             }
 250         }
 251     } else {
 252         LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
 253             idx = constant->idx;
 254             j = constant->count;
 255             while (j--) {
 256                 i = *idx++;
 257                 tmp_loc = constant_locations[i];
 258                 if (tmp_loc != -1) {
 259                     /* We found this uniform name in the program - go ahead and send the data */
 260                     GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, constants + (i * 4)));
 261                 }
 262             }
 263         }
 264     }
 265     checkGLcall("glUniform4fvARB()");
 266
 267     if(!This->baseShader.load_local_constsF) {
 268         TRACE("No need to load local float constants for this shader\n");
 269         return;
 270     }
 271
 272     /* Load immediate constants */
 273     if (TRACE_ON(d3d_shader)) {
 274         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 275             tmp_loc = constant_locations[lconst->idx];
 276             if (tmp_loc != -1) {
 277                 GLfloat* values = (GLfloat*)lconst->value;
 278                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 279                         values[0], values[1], values[2], values[3]);
 280             }
 281         }
 282     }
 283     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 284     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 285         tmp_loc = constant_locations[lconst->idx];
 286         if (tmp_loc != -1) {
 287             /* We found this uniform name in the program - go ahead and send the data */
 288             GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, (GLfloat*)lconst->value));
 289         }
 290     }
 291     checkGLcall("glUniform4fvARB()");
 292 }
 293
 294 /**
 295  * Loads integer constants (aka uniforms) into the currently set GLSL program.
 296  * When @constants_set == NULL, it will load all the constants.
 297  */
 298 static void shader_glsl_load_constantsI(
 299     IWineD3DBaseShaderImpl* This,
 300     WineD3D_GL_Info *gl_info,
 301     GLhandleARB programId,
 302     GLhandleARB locations[MAX_CONST_I],
 303     unsigned max_constants,
 304     int* constants,
 305     BOOL* constants_set) {
 306
 307     int i;
 308     struct list* ptr;
 309
 310     for (i=0; i<max_constants; ++i) {
 311         if (NULL == constants_set || constants_set[i]) {
 312
 313             TRACE_(d3d_constants)("Loading constants %i: %i, %i, %i, %i\n",
 314                   i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 315
 316             /* We found this uniform name in the program - go ahead and send the data */
 317             GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 318             checkGLcall("glUniform4ivARB");
 319         }
 320     }
 321
 322     /* Load immediate constants */
 323     ptr = list_head(&This->baseShader.constantsI);
 324     while (ptr) {
 325         local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
 326         unsigned int idx = lconst->idx;
 327         GLint* values = (GLint*) lconst->value;
 328
 329         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 330             values[0], values[1], values[2], values[3]);
 331
 332         /* We found this uniform name in the program - go ahead and send the data */
 333         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 334         checkGLcall("glUniform4ivARB");
 335         ptr = list_next(&This->baseShader.constantsI, ptr);
 336     }
 337 }
 338
 339 /**
 340  * Loads boolean constants (aka uniforms) into the currently set GLSL program.
 341  * When @constants_set == NULL, it will load all the constants.
 342  */
 343 static void shader_glsl_load_constantsB(
 344     IWineD3DBaseShaderImpl* This,
 345     WineD3D_GL_Info *gl_info,
 346     GLhandleARB programId,
 347     unsigned max_constants,
 348     BOOL* constants,
 349     BOOL* constants_set) {
 350
 351     GLhandleARB tmp_loc;
 352     int i;
 353     char tmp_name[8];
 354     char is_pshader = shader_is_pshader_version(This->baseShader.hex_version);
 355     const char* prefix = is_pshader? "PB":"VB";
 356     struct list* ptr;
 357
 358     for (i=0; i<max_constants; ++i) {
 359         if (NULL == constants_set || constants_set[i]) {
 360
 361             TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 362
 363             /* TODO: Benchmark and see if it would be beneficial to store the
 364              * locations of the constants to avoid looking up each time */
 365             snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 366             tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 367             if (tmp_loc != -1) {
 368                 /* We found this uniform name in the program - go ahead and send the data */
 369                 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 370                 checkGLcall("glUniform1ivARB");
 371             }
 372         }
 373     }
 374
 375     /* Load immediate constants */
 376     ptr = list_head(&This->baseShader.constantsB);
 377     while (ptr) {
 378         local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
 379         unsigned int idx = lconst->idx;
 380         GLint* values = (GLint*) lconst->value;
 381
 382         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 383
 384         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 385         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 386         if (tmp_loc != -1) {
 387             /* We found this uniform name in the program - go ahead and send the data */
 388             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 389             checkGLcall("glUniform1ivARB");
 390         }
 391         ptr = list_next(&This->baseShader.constantsB, ptr);
 392     }
 393 }
 394
 395
 396
 397 /**
 398  * Loads the app-supplied constants into the currently set GLSL program.
 399  */
 400 static void shader_glsl_load_constants(
 401     IWineD3DDevice* device,
 402     char usePixelShader,
 403     char useVertexShader) {
 404
 405     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
 406     struct shader_glsl_priv *priv = (struct shader_glsl_priv *)deviceImpl->shader_priv;
 407     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 408     WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 409
 410     GLhandleARB *constant_locations;
 411     struct list *constant_list;
 412     GLhandleARB programId;
 413     struct glsl_shader_prog_link *prog = priv->glsl_program;
 414     unsigned int i;
 415
 416     if (!prog) {
 417         /* No GLSL program set - nothing to do. */
 418         return;
 419     }
 420     programId = prog->programId;
 421
 422     if (useVertexShader) {
 423         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 424
 425         constant_locations = prog->vuniformF_locations;
 426         constant_list = &stateBlock->set_vconstantsF;
 427
 428         /* Load DirectX 9 float constants/uniforms for vertex shader */
 429         shader_glsl_load_constantsF(vshader, gl_info, GL_LIMITS(vshader_constantsF),
 430                 stateBlock->vertexShaderConstantF, constant_locations, constant_list);
 431
 432         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 433         shader_glsl_load_constantsI(vshader, gl_info, programId,
 434                                     prog->vuniformI_locations, MAX_CONST_I,
 435                                     stateBlock->vertexShaderConstantI,
 436                                     stateBlock->changed.vertexShaderConstantsI);
 437
 438         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 439         shader_glsl_load_constantsB(vshader, gl_info, programId, MAX_CONST_B,
 440                                     stateBlock->vertexShaderConstantB,
 441                                     stateBlock->changed.vertexShaderConstantsB);
 442
 443         /* Upload the position fixup params */
 444         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
 445         checkGLcall("glUniform4fvARB");
 446     }
 447
 448     if (usePixelShader) {
 449
 450         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 451
 452         constant_locations = prog->puniformF_locations;
 453         constant_list = &stateBlock->set_pconstantsF;
 454
 455         /* Load DirectX 9 float constants/uniforms for pixel shader */
 456         shader_glsl_load_constantsF(pshader, gl_info, GL_LIMITS(pshader_constantsF),
 457                 stateBlock->pixelShaderConstantF, constant_locations, constant_list);
 458
 459         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 460         shader_glsl_load_constantsI(pshader, gl_info, programId,
 461                                     prog->puniformI_locations, MAX_CONST_I,
 462                                     stateBlock->pixelShaderConstantI,
 463                                     stateBlock->changed.pixelShaderConstantsI);
 464
 465         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 466         shader_glsl_load_constantsB(pshader, gl_info, programId, MAX_CONST_B,
 467                                     stateBlock->pixelShaderConstantB,
 468                                     stateBlock->changed.pixelShaderConstantsB);
 469
 470         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 471          * It can't be 0 for a valid texbem instruction.
 472          */
 473         for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
 474             IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
 475             int stage = ps->luminanceconst[i].texunit;
 476
 477             float *data = (float *) &stateBlock->textureState[(int) ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
 478             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 479             checkGLcall("glUniformMatrix2fvARB");
 480
 481             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 482              * is set too, so we can check that in the needsbumpmat check
 483              */
 484             if(ps->baseShader.reg_maps.luminanceparams[stage]) {
 485                 GLfloat *scale = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
 486                 GLfloat *offset = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
 487
 488                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 489                 checkGLcall("glUniform1fvARB");
 490                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 491                 checkGLcall("glUniform1fvARB");
 492             }
 493         }
 494
 495         if(((IWineD3DPixelShaderImpl *) pshader)->srgb_enabled &&
 496                   !((IWineD3DPixelShaderImpl *) pshader)->srgb_mode_hardcoded) {
 497             float comparison[4];
 498             float mul_low[4];
 499
 500             if(stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
 501                 comparison[0] = srgb_cmp; comparison[1] = srgb_cmp;
 502                 comparison[2] = srgb_cmp; comparison[3] = srgb_cmp;
 503
 504                 mul_low[0] = srgb_mul_low; mul_low[1] = srgb_mul_low;
 505                 mul_low[2] = srgb_mul_low; mul_low[3] = srgb_mul_low;
 506             } else {
 507                 comparison[0] = 1.0 / 0.0; comparison[1] = 1.0 / 0.0;
 508                 comparison[2] = 1.0 / 0.0; comparison[3] = 1.0 / 0.0;
 509
 510                 mul_low[0] = 1.0; mul_low[1] = 1.0;
 511                 mul_low[2] = 1.0; mul_low[3] = 1.0;
 512             }
 513
 514             GL_EXTCALL(glUniform4fvARB(prog->srgb_comparison_location, 1, comparison));
 515             GL_EXTCALL(glUniform4fvARB(prog->srgb_mul_low_location, 1, mul_low));
 516         }
 517         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 518             float correction_params[4];
 519             if(deviceImpl->render_offscreen) {
 520                 correction_params[0] = 0.0;
 521                 correction_params[1] = 1.0;
 522             } else {
 523                 /* position is window relative, not viewport relative */
 524                 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 525                 correction_params[1] = -1.0;
 526             }
 527             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 528         }
 529     }
 530 }
 531
 532 /** Generate the variable & register declarations for the GLSL output target */
 533 static void shader_generate_glsl_declarations(
 534     IWineD3DBaseShader *iface,
 535     shader_reg_maps* reg_maps,
 536     SHADER_BUFFER* buffer,
 537     WineD3D_GL_Info* gl_info) {
 538
 539     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 540     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 541     int i;
 542     unsigned int extra_constants_needed = 0;
 543     local_constant* lconst;
 544
 545     /* There are some minor differences between pixel and vertex shaders */
 546     char pshader = shader_is_pshader_version(This->baseShader.hex_version);
 547     char prefix = pshader ? 'P' : 'V';
 548
 549     /* Prototype the subroutines */
 550     for (i = 0; i < This->baseShader.limits.label; i++) {
 551         if (reg_maps->labels[i])
 552             shader_addline(buffer, "void subroutine%u();\n", i);
 553     }
 554
 555     /* Declare the constants (aka uniforms) */
 556     if (This->baseShader.limits.constant_float > 0) {
 557         unsigned max_constantsF = min(This->baseShader.limits.constant_float,
 558                 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
 559         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 560     }
 561
 562     if (This->baseShader.limits.constant_int > 0)
 563         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 564
 565     if (This->baseShader.limits.constant_bool > 0)
 566         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 567
 568     if(!pshader) {
 569         shader_addline(buffer, "uniform vec4 posFixup;\n");
 570         /* Predeclaration; This function is added at link time based on the pixel shader.
 571          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 572          * that. We know the input to the reorder function at vertex shader compile time, so
 573          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 574          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 575          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 576          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 577          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 578          * inout.
 579          */
 580         if(This->baseShader.hex_version >= WINED3DVS_VERSION(3, 0)) {
 581             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 582         } else {
 583             shader_addline(buffer, "void order_ps_input();\n");
 584         }
 585     } else {
 586         IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
 587
 588         ps_impl->numbumpenvmatconsts = 0;
 589         for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
 590             if(!reg_maps->bumpmat[i]) {
 591                 continue;
 592             }
 593
 594             ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
 595             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 596
 597             if(reg_maps->luminanceparams) {
 598                 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
 599                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
 600                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
 601                 extra_constants_needed++;
 602             } else {
 603                 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
 604             }
 605
 606             extra_constants_needed++;
 607             ps_impl->numbumpenvmatconsts++;
 608         }
 609
 610         if(device->stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
 611             ps_impl->srgb_enabled = 1;
 612             if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
 613                 shader_addline(buffer, "uniform vec4 srgb_mul_low;\n");
 614                 shader_addline(buffer, "uniform vec4 srgb_comparison;\n");
 615                 ps_impl->srgb_mode_hardcoded = 0;
 616                 extra_constants_needed++;
 617             } else {
 618                 ps_impl->srgb_mode_hardcoded = 1;
 619                 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
 620                                srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
 621                 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
 622                                srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
 623             }
 624         } else {
 625             IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
 626
 627             /* Do not write any srgb fixup into the shader to save shader size and processing time.
 628              * As a consequence, we can't toggle srgb write on without recompilation
 629              */
 630             ps_impl->srgb_enabled = 0;
 631             ps_impl->srgb_mode_hardcoded = 1;
 632         }
 633         if(reg_maps->vpos || reg_maps->usesdsy) {
 634             if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
 635                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
 636                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
 637                 extra_constants_needed++;
 638             } else {
 639                 /* This happens because we do not have proper tracking of the constant registers that are
 640                  * actually used, only the max limit of the shader version
 641                  */
 642                 FIXME("Cannot find a free uniform for vpos correction params\n");
 643                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
 644                                device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
 645                                device->render_offscreen ? 1.0 : -1.0);
 646             }
 647             shader_addline(buffer, "vec4 vpos;\n");
 648         }
 649     }
 650
 651     /* Declare texture samplers */
 652     for (i = 0; i < This->baseShader.limits.sampler; i++) {
 653         if (reg_maps->samplers[i]) {
 654
 655             DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
 656             switch (stype) {
 657
 658                 case WINED3DSTT_1D:
 659                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
 660                     break;
 661                 case WINED3DSTT_2D:
 662                     if(device->stateBlock->textures[i] &&
 663                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
 664                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
 665                     } else {
 666                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
 667                     }
 668                     break;
 669                 case WINED3DSTT_CUBE:
 670                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
 671                     break;
 672                 case WINED3DSTT_VOLUME:
 673                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
 674                     break;
 675                 default:
 676                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
 677                     FIXME("Unrecognized sampler type: %#x\n", stype);
 678                     break;
 679             }
 680         }
 681     }
 682
 683     /* Declare address variables */
 684     for (i = 0; i < This->baseShader.limits.address; i++) {
 685         if (reg_maps->address[i])
 686             shader_addline(buffer, "ivec4 A%d;\n", i);
 687     }
 688
 689     /* Declare texture coordinate temporaries and initialize them */
 690     for (i = 0; i < This->baseShader.limits.texcoord; i++) {
 691         if (reg_maps->texcoord[i])
 692             shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
 693     }
 694
 695     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
 696      * helper function shader that is linked in at link time
 697      */
 698     if(pshader && This->baseShader.hex_version >= WINED3DPS_VERSION(3, 0)) {
 699         if(use_vs(device)) {
 700             shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 701         } else {
 702             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
 703              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
 704              * pixel shader that reads the fixed function color into the packed input registers.
 705              */
 706             shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 707         }
 708     }
 709
 710     /* Declare output register temporaries */
 711     if(This->baseShader.limits.packed_output) {
 712         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
 713     }
 714
 715     /* Declare temporary variables */
 716     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 717         if (reg_maps->temporary[i])
 718             shader_addline(buffer, "vec4 R%u;\n", i);
 719     }
 720
 721     /* Declare attributes */
 722     for (i = 0; i < This->baseShader.limits.attributes; i++) {
 723         if (reg_maps->attributes[i])
 724             shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
 725     }
 726
 727     /* Declare loop registers aLx */
 728     for (i = 0; i < reg_maps->loop_depth; i++) {
 729         shader_addline(buffer, "int aL%u;\n", i);
 730         shader_addline(buffer, "int tmpInt%u;\n", i);
 731     }
 732
 733     /* Temporary variables for matrix operations */
 734     shader_addline(buffer, "vec4 tmp0;\n");
 735     shader_addline(buffer, "vec4 tmp1;\n");
 736
 737     /* Local constants use a different name so they can be loaded once at shader link time
 738      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
 739      * float -> string conversion can cause precision loss.
 740      */
 741     if(!This->baseShader.load_local_constsF) {
 742         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 743             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
 744         }
 745     }
 746
 747     /* Start the main program */
 748     shader_addline(buffer, "void main() {\n");
 749     if(pshader && reg_maps->vpos) {
 750         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
 751          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
 752          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
 753          * precision troubles when we just substract 0.5.
 754          *
 755          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
 756          *
 757          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
 758          *
 759          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
 760          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
 761          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
 762          * correctly on drivers that returns integer values.
 763          */
 764         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
 765     }
 766 }
 767
 768 /*****************************************************************************
 769  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
 770  *
 771  * For more information, see http://wiki.winehq.org/DirectX-Shaders
 772  ****************************************************************************/
 773
 774 /* Prototypes */
 775 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
 776         const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
 777
 778 /** Used for opcode modifiers - They multiply the result by the specified amount */
 779 static const char * const shift_glsl_tab[] = {
 780     "",           /*  0 (none) */
 781     "2.0 * ",     /*  1 (x2)   */
 782     "4.0 * ",     /*  2 (x4)   */
 783     "8.0 * ",     /*  3 (x8)   */
 784     "16.0 * ",    /*  4 (x16)  */
 785     "32.0 * ",    /*  5 (x32)  */
 786     "",           /*  6 (x64)  */
 787     "",           /*  7 (x128) */
 788     "",           /*  8 (d256) */
 789     "",           /*  9 (d128) */
 790     "",           /* 10 (d64)  */
 791     "",           /* 11 (d32)  */
 792     "0.0625 * ",  /* 12 (d16)  */
 793     "0.125 * ",   /* 13 (d8)   */
 794     "0.25 * ",    /* 14 (d4)   */
 795     "0.5 * "      /* 15 (d2)   */
 796 };
 797
 798 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
 799 static void shader_glsl_gen_modifier (
 800     const DWORD instr,
 801     const char *in_reg,
 802     const char *in_regswizzle,
 803     char *out_str) {
 804
 805     out_str[0] = 0;
 806
 807     if (instr == WINED3DSIO_TEXKILL)
 808         return;
 809
 810     switch (instr & WINED3DSP_SRCMOD_MASK) {
 811     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
 812     case WINED3DSPSM_DW:
 813     case WINED3DSPSM_NONE:
 814         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
 815         break;
 816     case WINED3DSPSM_NEG:
 817         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
 818         break;
 819     case WINED3DSPSM_NOT:
 820         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
 821         break;
 822     case WINED3DSPSM_BIAS:
 823         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
 824         break;
 825     case WINED3DSPSM_BIASNEG:
 826         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
 827         break;
 828     case WINED3DSPSM_SIGN:
 829         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
 830         break;
 831     case WINED3DSPSM_SIGNNEG:
 832         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
 833         break;
 834     case WINED3DSPSM_COMP:
 835         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
 836         break;
 837     case WINED3DSPSM_X2:
 838         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
 839         break;
 840     case WINED3DSPSM_X2NEG:
 841         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
 842         break;
 843     case WINED3DSPSM_ABS:
 844         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
 845         break;
 846     case WINED3DSPSM_ABSNEG:
 847         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
 848         break;
 849     default:
 850         FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
 851         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
 852     }
 853 }
 854
 855 /** Writes the GLSL variable name that corresponds to the register that the
 856  * DX opcode parameter is trying to access */
 857 static void shader_glsl_get_register_name(
 858     const DWORD param,
 859     const DWORD addr_token,
 860     char* regstr,
 861     BOOL* is_color,
 862     SHADER_OPCODE_ARG* arg) {
 863
 864     /* oPos, oFog and oPts in D3D */
 865     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
 866
 867     DWORD reg = param & WINED3DSP_REGNUM_MASK;
 868     DWORD regtype = shader_get_regtype(param);
 869     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
 870     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
 871     WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
 872
 873     char pshader = shader_is_pshader_version(This->baseShader.hex_version);
 874     char tmpStr[150];
 875
 876     *is_color = FALSE;
 877
 878     switch (regtype) {
 879     case WINED3DSPR_TEMP:
 880         sprintf(tmpStr, "R%u", reg);
 881     break;
 882     case WINED3DSPR_INPUT:
 883         if (pshader) {
 884             /* Pixel shaders >= 3.0 */
 885             if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3) {
 886                 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
 887
 888                 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
 889                     glsl_src_param_t rel_param;
 890                     shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
 891
 892                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
 893                      * operation there
 894                      */
 895                     if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
 896                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
 897                             sprintf(tmpStr, "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
 898                                     rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count - 1,
 899                                     rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count,
 900                                     rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
 901                         } else {
 902                             sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
 903                         }
 904                     } else {
 905                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
 906                             sprintf(tmpStr, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
 907                                     rel_param.param_str, in_count - 1,
 908                                     rel_param.param_str, in_count,
 909                                     rel_param.param_str);
 910                         } else {
 911                             sprintf(tmpStr, "IN[%s]", rel_param.param_str);
 912                         }
 913                     }
 914                 } else {
 915                     DWORD idx = ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg];
 916                     if (idx == in_count) {
 917                         sprintf(tmpStr, "gl_Color");
 918                     } else if (idx == in_count + 1) {
 919                         sprintf(tmpStr, "gl_SecondaryColor");
 920                     } else {
 921                         sprintf(tmpStr, "IN[%u]", idx);
 922                     }
 923                 }
 924             } else {
 925                 if (reg==0)
 926                     strcpy(tmpStr, "gl_Color");
 927                 else
 928                     strcpy(tmpStr, "gl_SecondaryColor");
 929             }
 930         } else {
 931             if (vshader_input_is_color((IWineD3DVertexShader*) This, reg))
 932                *is_color = TRUE;
 933             sprintf(tmpStr, "attrib%u", reg);
 934         }
 935         break;
 936     case WINED3DSPR_CONST:
 937     {
 938         const char prefix = pshader? 'P':'V';
 939
 940         /* Relative addressing */
 941         if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
 942
 943            /* Relative addressing on shaders 2.0+ have a relative address token,
 944             * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
 945            if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 2)  {
 946                glsl_src_param_t rel_param;
 947                shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
 948                if(reg) {
 949                    sprintf(tmpStr, "%cC[%s + %u]", prefix, rel_param.param_str, reg);
 950                } else {
 951                    sprintf(tmpStr, "%cC[%s]", prefix, rel_param.param_str);
 952                }
 953            } else {
 954                if(reg) {
 955                    sprintf(tmpStr, "%cC[A0.x + %u]", prefix, reg);
 956                } else {
 957                    sprintf(tmpStr, "%cC[A0.x]", prefix);
 958                }
 959            }
 960
 961         } else {
 962             if(shader_constant_is_local(This, reg)) {
 963                 sprintf(tmpStr, "%cLC%u", prefix, reg);
 964             } else {
 965                 sprintf(tmpStr, "%cC[%u]", prefix, reg);
 966             }
 967         }
 968
 969         break;
 970     }
 971     case WINED3DSPR_CONSTINT:
 972         if (pshader)
 973             sprintf(tmpStr, "PI[%u]", reg);
 974         else
 975             sprintf(tmpStr, "VI[%u]", reg);
 976         break;
 977     case WINED3DSPR_CONSTBOOL:
 978         if (pshader)
 979             sprintf(tmpStr, "PB[%u]", reg);
 980         else
 981             sprintf(tmpStr, "VB[%u]", reg);
 982         break;
 983     case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 984         if (pshader) {
 985             sprintf(tmpStr, "T%u", reg);
 986         } else {
 987             sprintf(tmpStr, "A%u", reg);
 988         }
 989     break;
 990     case WINED3DSPR_LOOP:
 991         sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
 992     break;
 993     case WINED3DSPR_SAMPLER:
 994         if (pshader)
 995             sprintf(tmpStr, "Psampler%u", reg);
 996         else
 997             sprintf(tmpStr, "Vsampler%u", reg);
 998     break;
 999     case WINED3DSPR_COLOROUT:
1000         if (reg >= GL_LIMITS(buffers)) {
1001             WARN("Write to render target %u, only %d supported\n", reg, 4);
1002         }
1003         if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
1004             sprintf(tmpStr, "gl_FragData[%u]", reg);
1005         } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
1006             sprintf(tmpStr, "gl_FragColor");
1007         }
1008     break;
1009     case WINED3DSPR_RASTOUT:
1010         sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
1011     break;
1012     case WINED3DSPR_DEPTHOUT:
1013         sprintf(tmpStr, "gl_FragDepth");
1014     break;
1015     case WINED3DSPR_ATTROUT:
1016         if (reg == 0) {
1017             sprintf(tmpStr, "gl_FrontColor");
1018         } else {
1019             sprintf(tmpStr, "gl_FrontSecondaryColor");
1020         }
1021     break;
1022     case WINED3DSPR_TEXCRDOUT:
1023         /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1024         if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3)
1025             sprintf(tmpStr, "OUT[%u]", reg);
1026         else
1027             sprintf(tmpStr, "gl_TexCoord[%u]", reg);
1028     break;
1029     case WINED3DSPR_MISCTYPE:
1030         if (reg == 0) {
1031             /* vPos */
1032             sprintf(tmpStr, "vpos");
1033         } else if (reg == 1){
1034             /* Note that gl_FrontFacing is a bool, while vFace is
1035              * a float for which the sign determines front/back
1036              */
1037             sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
1038         } else {
1039             FIXME("Unhandled misctype register %d\n", reg);
1040             sprintf(tmpStr, "unrecognized_register");
1041         }
1042         break;
1043     default:
1044         FIXME("Unhandled register name Type(%d)\n", regtype);
1045         sprintf(tmpStr, "unrecognized_register");
1046     break;
1047     }
1048
1049     strcat(regstr, tmpStr);
1050 }
1051
1052 /* Get the GLSL write mask for the destination register */
1053 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
1054     char *ptr = write_mask;
1055     DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
1056
1057     if (shader_is_scalar(param)) {
1058         mask = WINED3DSP_WRITEMASK_0;
1059     } else {
1060         *ptr++ = '.';
1061         if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1062         if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1063         if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1064         if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1065     }
1066
1067     *ptr = '\0';
1068
1069     return mask;
1070 }
1071
1072 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1073     unsigned int size = 0;
1074
1075     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1076     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1077     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1078     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1079
1080     return size;
1081 }
1082
1083 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1084     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1085      * but addressed as "rgba". To fix this we need to swap the register's x
1086      * and z components. */
1087     DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1088     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1089     char *ptr = swizzle_str;
1090
1091     if (!shader_is_scalar(param)) {
1092         *ptr++ = '.';
1093         /* swizzle bits fields: wwzzyyxx */
1094         if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1095         if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1096         if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1097         if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1098     }
1099
1100     *ptr = '\0';
1101 }
1102
1103 /* From a given parameter token, generate the corresponding GLSL string.
1104  * Also, return the actual register name and swizzle in case the
1105  * caller needs this information as well. */
1106 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1107         const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param) {
1108     BOOL is_color = FALSE;
1109     char swizzle_str[6];
1110
1111     src_param->reg_name[0] = '\0';
1112     src_param->param_str[0] = '\0';
1113     swizzle_str[0] = '\0';
1114
1115     shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1116
1117     shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1118     shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1119 }
1120
1121 /* From a given parameter token, generate the corresponding GLSL string.
1122  * Also, return the actual register name and swizzle in case the
1123  * caller needs this information as well. */
1124 static DWORD shader_glsl_add_dst_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1125         const DWORD addr_token, glsl_dst_param_t *dst_param) {
1126     BOOL is_color = FALSE;
1127
1128     dst_param->mask_str[0] = '\0';
1129     dst_param->reg_name[0] = '\0';
1130
1131     shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1132     return shader_glsl_get_write_mask(param, dst_param->mask_str);
1133 }
1134
1135 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1136 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg, const DWORD param) {
1137     glsl_dst_param_t dst_param;
1138     DWORD mask;
1139     int shift;
1140
1141     mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1142
1143     if(mask) {
1144         shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1145         shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1146     }
1147
1148     return mask;
1149 }
1150
1151 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1152 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg) {
1153     return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1154 }
1155
1156 /** Process GLSL instruction modifiers */
1157 void shader_glsl_add_instruction_modifiers(SHADER_OPCODE_ARG* arg) {
1158
1159     DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1160
1161     if (arg->opcode->dst_token && mask != 0) {
1162         glsl_dst_param_t dst_param;
1163
1164         shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1165
1166         if (mask & WINED3DSPDM_SATURATE) {
1167             /* _SAT means to clamp the value of the register to between 0 and 1 */
1168             shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1169                     dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1170         }
1171         if (mask & WINED3DSPDM_MSAMPCENTROID) {
1172             FIXME("_centroid modifier not handled\n");
1173         }
1174         if (mask & WINED3DSPDM_PARTIALPRECISION) {
1175             /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1176         }
1177     }
1178 }
1179
1180 static inline const char* shader_get_comp_op(
1181     const DWORD opcode) {
1182
1183     DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1184     switch (op) {
1185         case COMPARISON_GT: return ">";
1186         case COMPARISON_EQ: return "==";
1187         case COMPARISON_GE: return ">=";
1188         case COMPARISON_LT: return "<";
1189         case COMPARISON_NE: return "!=";
1190         case COMPARISON_LE: return "<=";
1191         default:
1192             FIXME("Unrecognized comparison value: %u\n", op);
1193             return "(\?\?)";
1194     }
1195 }
1196
1197 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1198     /* Note that there's no such thing as a projected cube texture. */
1199     switch(sampler_type) {
1200         case WINED3DSTT_1D:
1201             sample_function->name = projected ? "texture1DProj" : "texture1D";
1202             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1203             break;
1204         case WINED3DSTT_2D:
1205             if(texrect) {
1206                 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1207             } else {
1208                 sample_function->name = projected ? "texture2DProj" : "texture2D";
1209             }
1210             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1211             break;
1212         case WINED3DSTT_CUBE:
1213             sample_function->name = "textureCube";
1214             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1215             break;
1216         case WINED3DSTT_VOLUME:
1217             sample_function->name = projected ? "texture3DProj" : "texture3D";
1218             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1219             break;
1220         default:
1221             sample_function->name = "";
1222             sample_function->coord_mask = 0;
1223             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1224             break;
1225     }
1226 }
1227
1228 static void shader_glsl_color_correction(SHADER_OPCODE_ARG* arg) {
1229     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1230     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) shader->baseShader.device;
1231     WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
1232     glsl_dst_param_t dst_param;
1233     glsl_dst_param_t dst_param2;
1234     WINED3DFORMAT fmt;
1235     WINED3DFORMAT conversion_group;
1236     IWineD3DBaseTextureImpl *texture;
1237     DWORD mask, mask_size;
1238     UINT i;
1239     BOOL recorded = FALSE;
1240     DWORD sampler_idx;
1241     DWORD hex_version = shader->baseShader.hex_version;
1242
1243     switch(arg->opcode->opcode) {
1244         case WINED3DSIO_TEX:
1245             if (hex_version < WINED3DPS_VERSION(2,0)) {
1246                 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1247             } else {
1248                 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
1249             }
1250             break;
1251
1252         case WINED3DSIO_TEXLDL:
1253             FIXME("Add color fixup for vertex texture WINED3DSIO_TEXLDL\n");
1254             return;
1255
1256         case WINED3DSIO_TEXDP3TEX:
1257         case WINED3DSIO_TEXM3x3TEX:
1258         case WINED3DSIO_TEXM3x3SPEC:
1259         case WINED3DSIO_TEXM3x3VSPEC:
1260         case WINED3DSIO_TEXBEM:
1261         case WINED3DSIO_TEXREG2AR:
1262         case WINED3DSIO_TEXREG2GB:
1263         case WINED3DSIO_TEXREG2RGB:
1264             sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1265             break;
1266
1267         default:
1268             /* Not a texture sampling instruction, nothing to do */
1269             return;
1270     };
1271
1272     texture = (IWineD3DBaseTextureImpl *) deviceImpl->stateBlock->textures[sampler_idx];
1273     if(texture) {
1274         fmt = texture->resource.format;
1275         conversion_group = texture->baseTexture.shader_conversion_group;
1276     } else {
1277         fmt = WINED3DFMT_UNKNOWN;
1278         conversion_group = WINED3DFMT_UNKNOWN;
1279     }
1280
1281     /* before doing anything, record the sampler with the format in the format conversion list,
1282      * but check if it's not there already
1283      */
1284     for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
1285         if(shader->baseShader.sampled_samplers[i] == sampler_idx) {
1286             recorded = TRUE;
1287             break;
1288         }
1289     }
1290     if(!recorded) {
1291         shader->baseShader.sampled_samplers[shader->baseShader.num_sampled_samplers] = sampler_idx;
1292         shader->baseShader.num_sampled_samplers++;
1293         shader->baseShader.sampled_format[sampler_idx] = conversion_group;
1294     }
1295
1296     switch(fmt) {
1297         case WINED3DFMT_V8U8:
1298         case WINED3DFMT_V16U16:
1299             if(GL_SUPPORT(NV_TEXTURE_SHADER) && fmt == WINED3DFMT_V8U8) {
1300                 /* The 3rd channel returns 1.0 in d3d, but 0.0 in gl. Fix this while we're at it :-) */
1301                 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_2, &dst_param);
1302                 mask_size = shader_glsl_get_write_mask_size(mask);
1303                 if(mask_size >= 3) {
1304                     shader_addline(arg->buffer, "%s.%c = 1.0;\n", dst_param.reg_name, dst_param.mask_str[3]);
1305                 }
1306             } else {
1307                 /* Correct the sign, but leave the blue as it is - it was loaded correctly already */
1308                 mask = shader_glsl_add_dst_param(arg, arg->dst,
1309                                           WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1,
1310                                           &dst_param);
1311                 mask_size = shader_glsl_get_write_mask_size(mask);
1312                 if(mask_size >= 2) {
1313                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1314                     dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1315                     dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1316                 } else if(mask_size == 1) {
1317                     shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str[1],
1318                     dst_param.reg_name, dst_param.mask_str[1]);
1319                 }
1320             }
1321             break;
1322
1323         case WINED3DFMT_X8L8V8U8:
1324             if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1325                 /* Red and blue are the signed channels, fix them up; Blue(=L) is correct already,
1326                  * and a(X) is always 1.0
1327                  */
1328                 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1329                 mask_size = shader_glsl_get_write_mask_size(mask);
1330                 if(mask_size >= 2) {
1331                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1332                                    dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1333                                    dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1334                 } else if(mask_size == 1) {
1335                     shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1336                                    dst_param.reg_name, dst_param.mask_str[1],
1337                                    dst_param.reg_name, dst_param.mask_str[1]);
1338                 }
1339             }
1340             break;
1341
1342         case WINED3DFMT_L6V5U5:
1343             if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1344                 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1345                 mask_size = shader_glsl_get_write_mask_size(mask);
1346                 shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_2, &dst_param2);
1347                 if(mask_size >= 3) {
1348                     /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1349                     shader_addline(arg->buffer, "tmp0.g = %s.%c;\n",
1350                                    dst_param.reg_name, dst_param.mask_str[2]);
1351                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1352                                    dst_param.reg_name, dst_param.mask_str[2], dst_param.mask_str[1],
1353                                    dst_param2.reg_name, dst_param.mask_str[1], dst_param.mask_str[3]);
1354                     shader_addline(arg->buffer, "%s.%c = tmp0.g;\n", dst_param.reg_name,
1355                                    dst_param.mask_str[3]);
1356                 } else if(mask_size == 2) {
1357                     /* This is bad: We have VL, but we need VU */
1358                     FIXME("2 components sampled from a converted L6V5U5 texture\n");
1359                 } else {
1360                     shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1361                                    dst_param.reg_name, dst_param.mask_str[1],
1362                                    dst_param2.reg_name, dst_param.mask_str[1]);
1363                 }
1364             }
1365             break;
1366
1367         case WINED3DFMT_Q8W8V8U8:
1368             if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1369                 /* Correct the sign in all channels. The writemask just applies as-is, no
1370                  * need for checking the mask size
1371                  */
1372                 shader_glsl_add_dst_param(arg, arg->dst,
1373                                           WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 |
1374                                           WINED3DSP_WRITEMASK_2 | WINED3DSP_WRITEMASK_3,
1375                                           &dst_param);
1376                 shader_addline(arg->buffer, "%s%s = %s%s * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str,
1377                                dst_param.reg_name, dst_param.mask_str);
1378             }
1379             break;
1380
1381         case WINED3DFMT_ATI2N:
1382             /* GL_ATI_texture_compression_3dc returns the two channels as luminance-alpha,
1383              * which means the first one is replicated across .rgb, and the 2nd one is in
1384              * .a. We need the 2nd in .g
1385              *
1386              * GL_EXT_texture_compression_rgtc returns the values in .rg, however, they
1387              * are swapped compared to d3d. So swap red and green.
1388              */
1389             mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1390             mask_size = shader_glsl_get_write_mask_size(mask);
1391             if(GL_SUPPORT(EXT_TEXTURE_COMPRESSION_RGTC)) {
1392                 if(mask_size >= 2) {
1393                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c;\n",
1394                                 dst_param.reg_name, dst_param.mask_str[1],
1395                                                     dst_param.mask_str[2],
1396                                 dst_param.reg_name, dst_param.mask_str[2],
1397                                                     dst_param.mask_str[1]);
1398                 } else {
1399                     FIXME("%u components sampled from a converted ATI2N texture\n", mask_size);
1400                 }
1401             } else {
1402                 if(mask_size == 4) {
1403                     /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1404                     shader_addline(arg->buffer, "%s.%c = %s.%c;\n",
1405                                 dst_param.reg_name, dst_param.mask_str[2],
1406                                 dst_param.reg_name, dst_param.mask_str[4]);
1407                 } else if(mask_size == 1) {
1408                     /* Nothing to do */
1409                 } else {
1410                     FIXME("%u components sampled from a converted ATI2N texture\n", mask_size);
1411                     /* This is bad: We have .r[gb], but we need .ra */
1412                 }
1413             }
1414             break;
1415
1416             /* stupid compiler */
1417         default:
1418             break;
1419     }
1420 }
1421
1422 /*****************************************************************************
1423  *
1424  * Begin processing individual instruction opcodes
1425  *
1426  ****************************************************************************/
1427
1428 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1429 static void shader_glsl_arith(SHADER_OPCODE_ARG* arg) {
1430     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1431     SHADER_BUFFER* buffer = arg->buffer;
1432     glsl_src_param_t src0_param;
1433     glsl_src_param_t src1_param;
1434     DWORD write_mask;
1435     char op;
1436
1437     /* Determine the GLSL operator to use based on the opcode */
1438     switch (curOpcode->opcode) {
1439         case WINED3DSIO_MUL: op = '*'; break;
1440         case WINED3DSIO_ADD: op = '+'; break;
1441         case WINED3DSIO_SUB: op = '-'; break;
1442         default:
1443             op = ' ';
1444             FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1445             break;
1446     }
1447
1448     write_mask = shader_glsl_append_dst(buffer, arg);
1449     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1450     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1451     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1452 }
1453
1454 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1455 static void shader_glsl_mov(SHADER_OPCODE_ARG* arg) {
1456     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1457     SHADER_BUFFER* buffer = arg->buffer;
1458     glsl_src_param_t src0_param;
1459     DWORD write_mask;
1460
1461     write_mask = shader_glsl_append_dst(buffer, arg);
1462     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1463
1464     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1465      * shader versions WINED3DSIO_MOVA is used for this. */
1466     if ((WINED3DSHADER_VERSION_MAJOR(shader->baseShader.hex_version) == 1 &&
1467             !shader_is_pshader_version(shader->baseShader.hex_version) &&
1468             shader_get_regtype(arg->dst) == WINED3DSPR_ADDR)) {
1469         /* This is a simple floor() */
1470         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1471         if (mask_size > 1) {
1472             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1473         } else {
1474             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1475         }
1476     } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1477         /* We need to *round* to the nearest int here. */
1478         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1479         if (mask_size > 1) {
1480             shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n", mask_size, src0_param.param_str, mask_size, src0_param.param_str);
1481         } else {
1482             shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1483         }
1484     } else {
1485         shader_addline(buffer, "%s);\n", src0_param.param_str);
1486     }
1487 }
1488
1489 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1490 static void shader_glsl_dot(SHADER_OPCODE_ARG* arg) {
1491     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1492     SHADER_BUFFER* buffer = arg->buffer;
1493     glsl_src_param_t src0_param;
1494     glsl_src_param_t src1_param;
1495     DWORD dst_write_mask, src_write_mask;
1496     unsigned int dst_size = 0;
1497
1498     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1499     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1500
1501     /* dp3 works on vec3, dp4 on vec4 */
1502     if (curOpcode->opcode == WINED3DSIO_DP4) {
1503         src_write_mask = WINED3DSP_WRITEMASK_ALL;
1504     } else {
1505         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1506     }
1507
1508     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1509     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1510
1511     if (dst_size > 1) {
1512         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1513     } else {
1514         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1515     }
1516 }
1517
1518 /* Note that this instruction has some restrictions. The destination write mask
1519  * can't contain the w component, and the source swizzles have to be .xyzw */
1520 static void shader_glsl_cross(SHADER_OPCODE_ARG *arg) {
1521     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1522     glsl_src_param_t src0_param;
1523     glsl_src_param_t src1_param;
1524     char dst_mask[6];
1525
1526     shader_glsl_get_write_mask(arg->dst, dst_mask);
1527     shader_glsl_append_dst(arg->buffer, arg);
1528     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1529     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1530     shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1531 }
1532
1533 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1534  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1535  * GLSL uses the value as-is. */
1536 static void shader_glsl_pow(SHADER_OPCODE_ARG *arg) {
1537     SHADER_BUFFER *buffer = arg->buffer;
1538     glsl_src_param_t src0_param;
1539     glsl_src_param_t src1_param;
1540     DWORD dst_write_mask;
1541     unsigned int dst_size;
1542
1543     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1544     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1545
1546     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1547     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1548
1549     if (dst_size > 1) {
1550         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1551     } else {
1552         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1553     }
1554 }
1555
1556 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1557  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1558  * GLSL uses the value as-is. */
1559 static void shader_glsl_log(SHADER_OPCODE_ARG *arg) {
1560     SHADER_BUFFER *buffer = arg->buffer;
1561     glsl_src_param_t src0_param;
1562     DWORD dst_write_mask;
1563     unsigned int dst_size;
1564
1565     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1566     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1567
1568     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1569
1570     if (dst_size > 1) {
1571         shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1572     } else {
1573         shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1574     }
1575 }
1576
1577 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1578 static void shader_glsl_map2gl(SHADER_OPCODE_ARG* arg) {
1579     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1580     SHADER_BUFFER* buffer = arg->buffer;
1581     glsl_src_param_t src_param;
1582     const char *instruction;
1583     char arguments[256];
1584     DWORD write_mask;
1585     unsigned i;
1586
1587     /* Determine the GLSL function to use based on the opcode */
1588     /* TODO: Possibly make this a table for faster lookups */
1589     switch (curOpcode->opcode) {
1590         case WINED3DSIO_MIN: instruction = "min"; break;
1591         case WINED3DSIO_MAX: instruction = "max"; break;
1592         case WINED3DSIO_ABS: instruction = "abs"; break;
1593         case WINED3DSIO_FRC: instruction = "fract"; break;
1594         case WINED3DSIO_NRM: instruction = "normalize"; break;
1595         case WINED3DSIO_EXP: instruction = "exp2"; break;
1596         case WINED3DSIO_SGN: instruction = "sign"; break;
1597         case WINED3DSIO_DSX: instruction = "dFdx"; break;
1598         case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1599         default: instruction = "";
1600             FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1601             break;
1602     }
1603
1604     write_mask = shader_glsl_append_dst(buffer, arg);
1605
1606     arguments[0] = '\0';
1607     if (curOpcode->num_params > 0) {
1608         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1609         strcat(arguments, src_param.param_str);
1610         for (i = 2; i < curOpcode->num_params; ++i) {
1611             strcat(arguments, ", ");
1612             shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1613             strcat(arguments, src_param.param_str);
1614         }
1615     }
1616
1617     shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1618 }
1619
1620 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1621  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1622  *   dst.x = 2^(floor(src))
1623  *   dst.y = src - floor(src)
1624  *   dst.z = 2^src   (partial precision is allowed, but optional)
1625  *   dst.w = 1.0;
1626  * For 2.0 shaders, just do this (honoring writemask and swizzle):
1627  *   dst = 2^src;    (partial precision is allowed, but optional)
1628  */
1629 static void shader_glsl_expp(SHADER_OPCODE_ARG* arg) {
1630     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1631     glsl_src_param_t src_param;
1632
1633     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1634
1635     if (shader->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
1636         char dst_mask[6];
1637
1638         shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1639         shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1640         shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1641         shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1642
1643         shader_glsl_append_dst(arg->buffer, arg);
1644         shader_glsl_get_write_mask(arg->dst, dst_mask);
1645         shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1646     } else {
1647         DWORD write_mask;
1648         unsigned int mask_size;
1649
1650         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1651         mask_size = shader_glsl_get_write_mask_size(write_mask);
1652
1653         if (mask_size > 1) {
1654             shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1655         } else {
1656             shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1657         }
1658     }
1659 }
1660
1661 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1662 static void shader_glsl_rcp(SHADER_OPCODE_ARG* arg) {
1663     glsl_src_param_t src_param;
1664     DWORD write_mask;
1665     unsigned int mask_size;
1666
1667     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1668     mask_size = shader_glsl_get_write_mask_size(write_mask);
1669     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1670
1671     if (mask_size > 1) {
1672         shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1673     } else {
1674         shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1675     }
1676 }
1677
1678 static void shader_glsl_rsq(SHADER_OPCODE_ARG* arg) {
1679     SHADER_BUFFER* buffer = arg->buffer;
1680     glsl_src_param_t src_param;
1681     DWORD write_mask;
1682     unsigned int mask_size;
1683
1684     write_mask = shader_glsl_append_dst(buffer, arg);
1685     mask_size = shader_glsl_get_write_mask_size(write_mask);
1686
1687     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1688
1689     if (mask_size > 1) {
1690         shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1691     } else {
1692         shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1693     }
1694 }
1695
1696 /** Process signed comparison opcodes in GLSL. */
1697 static void shader_glsl_compare(SHADER_OPCODE_ARG* arg) {
1698     glsl_src_param_t src0_param;
1699     glsl_src_param_t src1_param;
1700     DWORD write_mask;
1701     unsigned int mask_size;
1702
1703     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1704     mask_size = shader_glsl_get_write_mask_size(write_mask);
1705     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1706     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1707
1708     if (mask_size > 1) {
1709         const char *compare;
1710
1711         switch(arg->opcode->opcode) {
1712             case WINED3DSIO_SLT: compare = "lessThan"; break;
1713             case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1714             default: compare = "";
1715                 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1716         }
1717
1718         shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1719                 src0_param.param_str, src1_param.param_str);
1720     } else {
1721         switch(arg->opcode->opcode) {
1722             case WINED3DSIO_SLT:
1723                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1724                  * to return 0.0 but step returns 1.0 because step is not < x
1725                  * An alternative is a bvec compare padded with an unused second component.
1726                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1727                  * issue. Playing with not() is not possible either because not() does not accept
1728                  * a scalar.
1729                  */
1730                 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1731                 break;
1732             case WINED3DSIO_SGE:
1733                 /* Here we can use the step() function and safe a conditional */
1734                 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1735                 break;
1736             default:
1737                 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1738         }
1739
1740     }
1741 }
1742
1743 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1744 static void shader_glsl_cmp(SHADER_OPCODE_ARG* arg) {
1745     glsl_src_param_t src0_param;
1746     glsl_src_param_t src1_param;
1747     glsl_src_param_t src2_param;
1748     DWORD write_mask, cmp_channel = 0;
1749     unsigned int i, j;
1750     char mask_char[6];
1751     BOOL temp_destination = FALSE;
1752
1753     if(shader_is_scalar(arg->src[0])) {
1754         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1755
1756         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1757         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1758         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1759
1760         shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1761                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
1762     } else {
1763         DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1764         DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1765         DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1766         DWORD src0regtype = shader_get_regtype(arg->src[0]);
1767         DWORD src1regtype = shader_get_regtype(arg->src[1]);
1768         DWORD src2regtype = shader_get_regtype(arg->src[2]);
1769         DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1770         DWORD dstregtype = shader_get_regtype(arg->dst);
1771
1772         /* Cycle through all source0 channels */
1773         for (i=0; i<4; i++) {
1774             write_mask = 0;
1775             /* Find the destination channels which use the current source0 channel */
1776             for (j=0; j<4; j++) {
1777                 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1778                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
1779                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1780                 }
1781             }
1782
1783             /* Splitting the cmp instruction up in multiple lines imposes a problem:
1784             * The first lines may overwrite source parameters of the following lines.
1785             * Deal with that by using a temporary destination register if needed
1786             */
1787             if((src0reg == dstreg && src0regtype == dstregtype) ||
1788             (src1reg == dstreg && src1regtype == dstregtype) ||
1789             (src2reg == dstreg && src2regtype == dstregtype)) {
1790
1791                 write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1792                 if (!write_mask) continue;
1793                 shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1794                 temp_destination = TRUE;
1795             } else {
1796                 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1797                 if (!write_mask) continue;
1798             }
1799
1800             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1801             shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1802             shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1803
1804             shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1805                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
1806         }
1807
1808         if(temp_destination) {
1809             shader_glsl_get_write_mask(arg->dst, mask_char);
1810             shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1811             shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1812         }
1813     }
1814
1815 }
1816
1817 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1818 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1819  * the compare is done per component of src0. */
1820 static void shader_glsl_cnd(SHADER_OPCODE_ARG* arg) {
1821     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1822     glsl_src_param_t src0_param;
1823     glsl_src_param_t src1_param;
1824     glsl_src_param_t src2_param;
1825     DWORD write_mask, cmp_channel = 0;
1826     unsigned int i, j;
1827
1828     if (shader->baseShader.hex_version < WINED3DPS_VERSION(1, 4)) {
1829         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1830         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1831         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1832         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1833
1834         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1835         if(arg->opcode_token & WINED3DSI_COISSUE) {
1836             shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1837         } else {
1838             shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1839                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
1840         }
1841         return;
1842     }
1843     /* Cycle through all source0 channels */
1844     for (i=0; i<4; i++) {
1845         write_mask = 0;
1846         /* Find the destination channels which use the current source0 channel */
1847         for (j=0; j<4; j++) {
1848             if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1849                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1850                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1851             }
1852         }
1853         write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1854         if (!write_mask) continue;
1855
1856         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1857         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1858         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1859
1860         shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1861                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1862     }
1863 }
1864
1865 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1866 static void shader_glsl_mad(SHADER_OPCODE_ARG* arg) {
1867     glsl_src_param_t src0_param;
1868     glsl_src_param_t src1_param;
1869     glsl_src_param_t src2_param;
1870     DWORD write_mask;
1871
1872     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1873     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1874     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1875     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1876     shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1877             src0_param.param_str, src1_param.param_str, src2_param.param_str);
1878 }
1879
1880 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1881     Vertex shaders to GLSL codes */
1882 static void shader_glsl_mnxn(SHADER_OPCODE_ARG* arg) {
1883     int i;
1884     int nComponents = 0;
1885     SHADER_OPCODE_ARG tmpArg;
1886
1887     memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1888
1889     /* Set constants for the temporary argument */
1890     tmpArg.shader      = arg->shader;
1891     tmpArg.buffer      = arg->buffer;
1892     tmpArg.src[0]      = arg->src[0];
1893     tmpArg.src_addr[0] = arg->src_addr[0];
1894     tmpArg.src_addr[1] = arg->src_addr[1];
1895     tmpArg.reg_maps = arg->reg_maps;
1896
1897     switch(arg->opcode->opcode) {
1898         case WINED3DSIO_M4x4:
1899             nComponents = 4;
1900             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1901             break;
1902         case WINED3DSIO_M4x3:
1903             nComponents = 3;
1904             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1905             break;
1906         case WINED3DSIO_M3x4:
1907             nComponents = 4;
1908             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1909             break;
1910         case WINED3DSIO_M3x3:
1911             nComponents = 3;
1912             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1913             break;
1914         case WINED3DSIO_M3x2:
1915             nComponents = 2;
1916             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1917             break;
1918         default:
1919             break;
1920     }
1921
1922     for (i = 0; i < nComponents; i++) {
1923         tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1924         tmpArg.src[1]      = arg->src[1]+i;
1925         shader_glsl_dot(&tmpArg);
1926     }
1927 }
1928
1929 /**
1930     The LRP instruction performs a component-wise linear interpolation
1931     between the second and third operands using the first operand as the
1932     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
1933     This is equivalent to mix(src2, src1, src0);
1934 */
1935 static void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
1936     glsl_src_param_t src0_param;
1937     glsl_src_param_t src1_param;
1938     glsl_src_param_t src2_param;
1939     DWORD write_mask;
1940
1941     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1942
1943     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1944     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1945     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1946
1947     shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1948             src2_param.param_str, src1_param.param_str, src0_param.param_str);
1949 }
1950
1951 /** Process the WINED3DSIO_LIT instruction in GLSL:
1952  * dst.x = dst.w = 1.0
1953  * dst.y = (src0.x > 0) ? src0.x
1954  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1955  *                                        where src.w is clamped at +- 128
1956  */
1957 static void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
1958     glsl_src_param_t src0_param;
1959     glsl_src_param_t src1_param;
1960     glsl_src_param_t src3_param;
1961     char dst_mask[6];
1962
1963     shader_glsl_append_dst(arg->buffer, arg);
1964     shader_glsl_get_write_mask(arg->dst, dst_mask);
1965
1966     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1967     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
1968     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
1969
1970     /* The sdk specifies the instruction like this
1971      * dst.x = 1.0;
1972      * if(src.x > 0.0) dst.y = src.x
1973      * else dst.y = 0.0.
1974      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
1975      * else dst.z = 0.0;
1976      * dst.w = 1.0;
1977      *
1978      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
1979      * dst.x = 1.0                                  ... No further explanation needed
1980      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
1981      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
1982      * dst.w = 1.0.                                 ... Nothing fancy.
1983      *
1984      * So we still have one conditional in there. So do this:
1985      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
1986      *
1987      * step(0.0, x) will return 1 if src.x > 0.0, and 0 otherwise. So if y is 0 we get pow(0.0 * 1.0, power),
1988      * which sets dst.z to 0. If y > 0, but x = 0.0, we get pow(y * 0.0, power), which results in 0 too.
1989      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
1990      */
1991     shader_addline(arg->buffer, "vec4(1.0, max(%s, 0.0), pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
1992                    src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
1993 }
1994
1995 /** Process the WINED3DSIO_DST instruction in GLSL:
1996  * dst.x = 1.0
1997  * dst.y = src0.x * src0.y
1998  * dst.z = src0.z
1999  * dst.w = src1.w
2000  */
2001 static void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
2002     glsl_src_param_t src0y_param;
2003     glsl_src_param_t src0z_param;
2004     glsl_src_param_t src1y_param;
2005     glsl_src_param_t src1w_param;
2006     char dst_mask[6];
2007
2008     shader_glsl_append_dst(arg->buffer, arg);
2009     shader_glsl_get_write_mask(arg->dst, dst_mask);
2010
2011     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2012     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2013     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2014     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2015
2016     shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2017             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2018 }
2019
2020 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2021  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2022  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2023  *
2024  * dst.x = cos(src0.?)
2025  * dst.y = sin(src0.?)
2026  * dst.z = dst.z
2027  * dst.w = dst.w
2028  */
2029 static void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
2030     glsl_src_param_t src0_param;
2031     DWORD write_mask;
2032
2033     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2034     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2035
2036     switch (write_mask) {
2037         case WINED3DSP_WRITEMASK_0:
2038             shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
2039             break;
2040
2041         case WINED3DSP_WRITEMASK_1:
2042             shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
2043             break;
2044
2045         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2046             shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2047             break;
2048
2049         default:
2050             ERR("Write mask should be .x, .y or .xy\n");
2051             break;
2052     }
2053 }
2054
2055 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2056  * Start a for() loop where src1.y is the initial value of aL,
2057  *  increment aL by src1.z for a total of src1.x iterations.
2058  *  Need to use a temporary variable for this operation.
2059  */
2060 /* FIXME: I don't think nested loops will work correctly this way. */
2061 static void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
2062     glsl_src_param_t src1_param;
2063     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2064     DWORD regtype = shader_get_regtype(arg->src[1]);
2065     DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
2066     const DWORD *control_values = NULL;
2067     local_constant *constant;
2068
2069     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2070
2071     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2072      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2073      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2074      * addressing.
2075      */
2076     if(regtype == WINED3DSPR_CONSTINT) {
2077         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2078             if(constant->idx == reg) {
2079                 control_values = constant->value;
2080                 break;
2081             }
2082         }
2083     }
2084
2085     if(control_values) {
2086         if(control_values[2] > 0) {
2087             shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
2088                            shader->baseShader.cur_loop_depth, control_values[1],
2089                            shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2090                            shader->baseShader.cur_loop_depth, control_values[2]);
2091         } else if(control_values[2] == 0) {
2092             shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2093                            shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2094                            shader->baseShader.cur_loop_depth, control_values[0],
2095                            shader->baseShader.cur_loop_depth);
2096         } else {
2097             shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2098                            shader->baseShader.cur_loop_depth, control_values[1],
2099                            shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2100                            shader->baseShader.cur_loop_depth, control_values[2]);
2101         }
2102     } else {
2103         shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2104                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2105                        src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2106                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2107     }
2108
2109     shader->baseShader.cur_loop_depth++;
2110     shader->baseShader.cur_loop_regno++;
2111 }
2112
2113 static void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
2114     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2115
2116     shader_addline(arg->buffer, "}\n");
2117
2118     if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
2119         shader->baseShader.cur_loop_depth--;
2120         shader->baseShader.cur_loop_regno--;
2121     }
2122     if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
2123         shader->baseShader.cur_loop_depth--;
2124     }
2125 }
2126
2127 static void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
2128     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2129     glsl_src_param_t src0_param;
2130
2131     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2132     shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2133                    shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2134                    src0_param.param_str, shader->baseShader.cur_loop_depth);
2135     shader->baseShader.cur_loop_depth++;
2136 }
2137
2138 static void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
2139     glsl_src_param_t src0_param;
2140
2141     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2142     shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2143 }
2144
2145 static void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
2146     glsl_src_param_t src0_param;
2147     glsl_src_param_t src1_param;
2148
2149     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2150     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2151
2152     shader_addline(arg->buffer, "if (%s %s %s) {\n",
2153             src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2154 }
2155
2156 static void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
2157     shader_addline(arg->buffer, "} else {\n");
2158 }
2159
2160 static void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
2161     shader_addline(arg->buffer, "break;\n");
2162 }
2163
2164 /* FIXME: According to MSDN the compare is done per component. */
2165 static void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
2166     glsl_src_param_t src0_param;
2167     glsl_src_param_t src1_param;
2168
2169     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2170     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2171
2172     shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2173             src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2174 }
2175
2176 static void shader_glsl_label(SHADER_OPCODE_ARG* arg) {
2177
2178     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2179     shader_addline(arg->buffer, "}\n");
2180     shader_addline(arg->buffer, "void subroutine%u () {\n",  snum);
2181 }
2182
2183 static void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
2184     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2185     shader_addline(arg->buffer, "subroutine%u();\n", snum);
2186 }
2187
2188 static void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
2189     glsl_src_param_t src1_param;
2190
2191     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2192     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2193     shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2194 }
2195
2196 /*********************************************
2197  * Pixel Shader Specific Code begins here
2198  ********************************************/
2199 static void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
2200     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2201     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2202     DWORD hex_version = This->baseShader.hex_version;
2203     char dst_swizzle[6];
2204     glsl_sample_function_t sample_function;
2205     DWORD sampler_type;
2206     DWORD sampler_idx;
2207     BOOL projected, texrect = FALSE;
2208     DWORD mask = 0;
2209
2210     /* All versions have a destination register */
2211     shader_glsl_append_dst(arg->buffer, arg);
2212
2213     /* 1.0-1.4: Use destination register as sampler source.
2214      * 2.0+: Use provided sampler source. */
2215     if (hex_version < WINED3DPS_VERSION(2,0)) {
2216         sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2217     } else {
2218         sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2219     }
2220     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2221
2222     if (hex_version < WINED3DPS_VERSION(1,4)) {
2223         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2224
2225         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2226         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2227             projected = TRUE;
2228             switch (flags & ~WINED3DTTFF_PROJECTED) {
2229                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2230                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2231                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2232                 case WINED3DTTFF_COUNT4:
2233                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2234             }
2235         } else {
2236             projected = FALSE;
2237         }
2238     } else if (hex_version < WINED3DPS_VERSION(2,0)) {
2239         DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2240
2241         if (src_mod == WINED3DSPSM_DZ) {
2242             projected = TRUE;
2243             mask = WINED3DSP_WRITEMASK_2;
2244         } else if (src_mod == WINED3DSPSM_DW) {
2245             projected = TRUE;
2246             mask = WINED3DSP_WRITEMASK_3;
2247         } else {
2248             projected = FALSE;
2249         }
2250     } else {
2251         if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2252                 /* ps 2.0 texldp instruction always divides by the fourth component. */
2253                 projected = TRUE;
2254                 mask = WINED3DSP_WRITEMASK_3;
2255         } else {
2256             projected = FALSE;
2257         }
2258     }
2259
2260     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2261        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2262         texrect = TRUE;
2263     }
2264
2265     shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2266     mask |= sample_function.coord_mask;
2267
2268     if (hex_version < WINED3DPS_VERSION(2,0)) {
2269         shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2270     } else {
2271         shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2272     }
2273
2274     /* 1.0-1.3: Use destination register as coordinate source.
2275        1.4+: Use provided coordinate source register. */
2276     if (hex_version < WINED3DPS_VERSION(1,4)) {
2277         char coord_mask[6];
2278         shader_glsl_get_write_mask(mask, coord_mask);
2279         shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2280                 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2281     } else {
2282         glsl_src_param_t coord_param;
2283         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2284         if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2285             glsl_src_param_t bias;
2286             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2287
2288             shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2289                     sample_function.name, sampler_idx, coord_param.param_str,
2290                     bias.param_str, dst_swizzle);
2291         } else {
2292             shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2293                     sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2294         }
2295     }
2296 }
2297
2298 static void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
2299     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2300     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2301     glsl_sample_function_t sample_function;
2302     glsl_src_param_t coord_param, lod_param;
2303     char dst_swizzle[6];
2304     DWORD sampler_type;
2305     DWORD sampler_idx;
2306     BOOL texrect = FALSE;
2307
2308     shader_glsl_append_dst(arg->buffer, arg);
2309     shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2310
2311     sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2312     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2313     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2314        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2315         texrect = TRUE;
2316     }
2317     shader_glsl_get_sample_function(sampler_type, FALSE, texrect, &sample_function);    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &coord_param);
2318
2319     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2320
2321     if (shader_is_pshader_version(This->baseShader.hex_version)) {
2322         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2323          * However, they seem to work just fine in fragment shaders as well. */
2324         WARN("Using %sLod in fragment shader.\n", sample_function.name);
2325         shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2326                 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2327     } else {
2328         shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2329                 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2330     }
2331 }
2332
2333 static void pshader_glsl_texcoord(SHADER_OPCODE_ARG* arg) {
2334
2335     /* FIXME: Make this work for more than just 2D textures */
2336
2337     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2338     SHADER_BUFFER* buffer = arg->buffer;
2339     DWORD hex_version = This->baseShader.hex_version;
2340     DWORD write_mask;
2341     char dst_mask[6];
2342
2343     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2344     shader_glsl_get_write_mask(write_mask, dst_mask);
2345
2346     if (hex_version != WINED3DPS_VERSION(1,4)) {
2347         DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2348         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2349     } else {
2350         DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2351         DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2352         char dst_swizzle[6];
2353
2354         shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2355
2356         if (src_mod == WINED3DSPSM_DZ) {
2357             glsl_src_param_t div_param;
2358             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2359             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2360
2361             if (mask_size > 1) {
2362                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2363             } else {
2364                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2365             }
2366         } else if (src_mod == WINED3DSPSM_DW) {
2367             glsl_src_param_t div_param;
2368             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2369             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2370
2371             if (mask_size > 1) {
2372                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2373             } else {
2374                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2375             }
2376         } else {
2377             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2378         }
2379     }
2380 }
2381
2382 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2383  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2384  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2385 static void pshader_glsl_texdp3tex(SHADER_OPCODE_ARG* arg) {
2386     glsl_src_param_t src0_param;
2387     char dst_mask[6];
2388     glsl_sample_function_t sample_function;
2389     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2390     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2391     DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2392
2393     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2394
2395     shader_glsl_append_dst(arg->buffer, arg);
2396     shader_glsl_get_write_mask(arg->dst, dst_mask);
2397
2398     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2399      * scalar, and projected sampling would require 4.
2400      *
2401      * It is a dependent read - not valid with conditional NP2 textures
2402      */
2403     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2404
2405     switch(count_bits(sample_function.coord_mask)) {
2406         case 1:
2407             shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2408                            sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2409             break;
2410
2411         case 2:
2412             shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2413                           sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2414             break;
2415
2416         case 3:
2417             shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2418                            sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2419             break;
2420         default:
2421             FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2422     }
2423 }
2424
2425 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2426  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2427 static void pshader_glsl_texdp3(SHADER_OPCODE_ARG* arg) {
2428     glsl_src_param_t src0_param;
2429     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2430     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2431     DWORD dst_mask;
2432     unsigned int mask_size;
2433
2434     dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2435     mask_size = shader_glsl_get_write_mask_size(dst_mask);
2436     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2437
2438     if (mask_size > 1) {
2439         shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2440     } else {
2441         shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2442     }
2443 }
2444
2445 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2446  * Calculate the depth as dst.x / dst.y   */
2447 static void pshader_glsl_texdepth(SHADER_OPCODE_ARG* arg) {
2448     glsl_dst_param_t dst_param;
2449
2450     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2451
2452     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2453      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2454      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2455      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2456      * >= 1.0 or < 0.0
2457      */
2458     shader_addline(arg->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n", dst_param.reg_name, dst_param.reg_name);
2459 }
2460
2461 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2462  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2463  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2464  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2465  */
2466 static void pshader_glsl_texm3x2depth(SHADER_OPCODE_ARG* arg) {
2467     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2468     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2469     glsl_src_param_t src0_param;
2470
2471     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2472
2473     shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2474     shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2475 }
2476
2477 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2478  * Calculate the 1st of a 2-row matrix multiplication. */
2479 static void pshader_glsl_texm3x2pad(SHADER_OPCODE_ARG* arg) {
2480     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2481     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2482     SHADER_BUFFER* buffer = arg->buffer;
2483     glsl_src_param_t src0_param;
2484
2485     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2486     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2487 }
2488
2489 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2490  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2491 static void pshader_glsl_texm3x3pad(SHADER_OPCODE_ARG* arg) {
2492
2493     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2494     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2495     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2496     SHADER_BUFFER* buffer = arg->buffer;
2497     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2498     glsl_src_param_t src0_param;
2499
2500     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2501     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2502     current_state->texcoord_w[current_state->current_row++] = reg;
2503 }
2504
2505 static void pshader_glsl_texm3x2tex(SHADER_OPCODE_ARG* arg) {
2506     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2507     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2508     SHADER_BUFFER* buffer = arg->buffer;
2509     glsl_src_param_t src0_param;
2510     char dst_mask[6];
2511
2512     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2513     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2514
2515     shader_glsl_append_dst(buffer, arg);
2516     shader_glsl_get_write_mask(arg->dst, dst_mask);
2517
2518     /* Sample the texture using the calculated coordinates */
2519     shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2520 }
2521
2522 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2523  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2524 static void pshader_glsl_texm3x3tex(SHADER_OPCODE_ARG* arg) {
2525     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2526     glsl_src_param_t src0_param;
2527     char dst_mask[6];
2528     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2529     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2530     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2531     DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2532     glsl_sample_function_t sample_function;
2533
2534     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2535     shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2536
2537     shader_glsl_append_dst(arg->buffer, arg);
2538     shader_glsl_get_write_mask(arg->dst, dst_mask);
2539     /* Dependent read, not valid with conditional NP2 */
2540     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2541
2542     /* Sample the texture using the calculated coordinates */
2543     shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2544
2545     current_state->current_row = 0;
2546 }
2547
2548 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2549  * Perform the 3rd row of a 3x3 matrix multiply */
2550 static void pshader_glsl_texm3x3(SHADER_OPCODE_ARG* arg) {
2551     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2552     glsl_src_param_t src0_param;
2553     char dst_mask[6];
2554     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2555     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2556     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2557
2558     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2559
2560     shader_glsl_append_dst(arg->buffer, arg);
2561     shader_glsl_get_write_mask(arg->dst, dst_mask);
2562     shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2563
2564     current_state->current_row = 0;
2565 }
2566
2567 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2568  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2569 static void pshader_glsl_texm3x3spec(SHADER_OPCODE_ARG* arg) {
2570
2571     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2572     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2573     glsl_src_param_t src0_param;
2574     glsl_src_param_t src1_param;
2575     char dst_mask[6];
2576     SHADER_BUFFER* buffer = arg->buffer;
2577     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2578     DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2579     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2580     glsl_sample_function_t sample_function;
2581
2582     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2583     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2584
2585     /* Perform the last matrix multiply operation */
2586     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2587     /* Reflection calculation */
2588     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2589
2590     shader_glsl_append_dst(buffer, arg);
2591     shader_glsl_get_write_mask(arg->dst, dst_mask);
2592     /* Dependent read, not valid with conditional NP2 */
2593     shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2594
2595     /* Sample the texture */
2596     shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2597
2598     current_state->current_row = 0;
2599 }
2600
2601 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2602  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2603 static void pshader_glsl_texm3x3vspec(SHADER_OPCODE_ARG* arg) {
2604
2605     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2606     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2607     SHADER_BUFFER* buffer = arg->buffer;
2608     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2609     glsl_src_param_t src0_param;
2610     char dst_mask[6];
2611     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2612     DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2613     glsl_sample_function_t sample_function;
2614
2615     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2616
2617     /* Perform the last matrix multiply operation */
2618     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2619
2620     /* Construct the eye-ray vector from w coordinates */
2621     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2622             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2623     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2624
2625     shader_glsl_append_dst(buffer, arg);
2626     shader_glsl_get_write_mask(arg->dst, dst_mask);
2627     /* Dependent read, not valid with conditional NP2 */
2628     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2629
2630     /* Sample the texture using the calculated coordinates */
2631     shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2632
2633     current_state->current_row = 0;
2634 }
2635
2636 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2637  * Apply a fake bump map transform.
2638  * texbem is pshader <= 1.3 only, this saves a few version checks
2639  */
2640 static void pshader_glsl_texbem(SHADER_OPCODE_ARG* arg) {
2641     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2642     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2643     char dst_swizzle[6];
2644     glsl_sample_function_t sample_function;
2645     glsl_src_param_t coord_param;
2646     DWORD sampler_type;
2647     DWORD sampler_idx;
2648     DWORD mask;
2649     DWORD flags;
2650     char coord_mask[6];
2651
2652     sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2653     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2654
2655     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2656     /* Dependent read, not valid with conditional NP2 */
2657     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2658     mask = sample_function.coord_mask;
2659
2660     shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2661
2662     shader_glsl_get_write_mask(mask, coord_mask);
2663
2664     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2665          * so we can't let the GL handle this.
2666          */
2667     if (flags & WINED3DTTFF_PROJECTED) {
2668         DWORD div_mask=0;
2669         char coord_div_mask[3];
2670         switch (flags & ~WINED3DTTFF_PROJECTED) {
2671             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2672             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2673             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2674             case WINED3DTTFF_COUNT4:
2675             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2676         }
2677         shader_glsl_get_write_mask(div_mask, coord_div_mask);
2678         shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2679     }
2680
2681     shader_glsl_append_dst(arg->buffer, arg);
2682     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2683     if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2684         glsl_src_param_t luminance_param;
2685         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2686         shader_addline(arg->buffer, "(%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )*(%s * luminancescale%d + luminanceoffset%d))%s);\n",
2687                        sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask,
2688                        luminance_param.param_str, sampler_idx, sampler_idx, dst_swizzle);
2689     } else {
2690         shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )%s);\n",
2691                        sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask, dst_swizzle);
2692     }
2693 }
2694
2695 static void pshader_glsl_bem(SHADER_OPCODE_ARG* arg) {
2696     glsl_src_param_t src0_param, src1_param;
2697     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2698
2699     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2700     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2701
2702     shader_glsl_append_dst(arg->buffer, arg);
2703     shader_addline(arg->buffer, "%s + bumpenvmat%d * %s);\n",
2704                    src0_param.param_str, sampler_idx, src1_param.param_str);
2705 }
2706
2707 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2708  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2709 static void pshader_glsl_texreg2ar(SHADER_OPCODE_ARG* arg) {
2710
2711     glsl_src_param_t src0_param;
2712     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2713     char dst_mask[6];
2714
2715     shader_glsl_append_dst(arg->buffer, arg);
2716     shader_glsl_get_write_mask(arg->dst, dst_mask);
2717     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2718
2719     shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2720 }
2721
2722 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2723  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2724 static void pshader_glsl_texreg2gb(SHADER_OPCODE_ARG* arg) {
2725     glsl_src_param_t src0_param;
2726     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2727     char dst_mask[6];
2728
2729     shader_glsl_append_dst(arg->buffer, arg);
2730     shader_glsl_get_write_mask(arg->dst, dst_mask);
2731     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2732
2733     shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2734 }
2735
2736 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2737  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2738 static void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
2739     glsl_src_param_t src0_param;
2740     char dst_mask[6];
2741     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2742     DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2743     glsl_sample_function_t sample_function;
2744
2745     shader_glsl_append_dst(arg->buffer, arg);
2746     shader_glsl_get_write_mask(arg->dst, dst_mask);
2747     /* Dependent read, not valid with conditional NP2 */
2748     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2749     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2750
2751     shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2752 }
2753
2754 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2755  * If any of the first 3 components are < 0, discard this pixel */
2756 static void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
2757     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2758     DWORD hex_version = This->baseShader.hex_version;
2759     glsl_dst_param_t dst_param;
2760
2761     /* The argument is a destination parameter, and no writemasks are allowed */
2762     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2763     if((hex_version >= WINED3DPS_VERSION(2,0))) {
2764         /* 2.0 shaders compare all 4 components in texkill */
2765         shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2766     } else {
2767         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2768          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2769          * 4 components are defined, only the first 3 are used
2770          */
2771         shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2772     }
2773 }
2774
2775 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2776  * dst = dot2(src0, src1) + src2 */
2777 static void pshader_glsl_dp2add(SHADER_OPCODE_ARG* arg) {
2778     glsl_src_param_t src0_param;
2779     glsl_src_param_t src1_param;
2780     glsl_src_param_t src2_param;
2781     DWORD write_mask;
2782     unsigned int mask_size;
2783
2784     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2785     mask_size = shader_glsl_get_write_mask_size(write_mask);
2786
2787     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2788     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2789     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2790
2791     if (mask_size > 1) {
2792         shader_addline(arg->buffer, "vec%d(dot(%s, %s) + %s));\n", mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
2793     } else {
2794         shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2795     }
2796 }
2797
2798 static void pshader_glsl_input_pack(
2799    SHADER_BUFFER* buffer,
2800    semantic* semantics_in,
2801    IWineD3DPixelShader *iface) {
2802
2803    unsigned int i;
2804    IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2805
2806    for (i = 0; i < MAX_REG_INPUT; i++) {
2807
2808        DWORD usage_token = semantics_in[i].usage;
2809        DWORD register_token = semantics_in[i].reg;
2810        DWORD usage, usage_idx;
2811        char reg_mask[6];
2812
2813        /* Uninitialized */
2814        if (!usage_token) continue;
2815        usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2816        usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2817        shader_glsl_get_write_mask(register_token, reg_mask);
2818
2819        switch(usage) {
2820
2821            case WINED3DDECLUSAGE_TEXCOORD:
2822                if(usage_idx < 8 && This->vertexprocessing == pretransformed) {
2823                    shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2824                                   This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2825                } else {
2826                    shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2827                                   This->input_reg_map[i], reg_mask, reg_mask);
2828                }
2829                break;
2830
2831            case WINED3DDECLUSAGE_COLOR:
2832                if (usage_idx == 0)
2833                    shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2834                        This->input_reg_map[i], reg_mask, reg_mask);
2835                else if (usage_idx == 1)
2836                    shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2837                        This->input_reg_map[i], reg_mask, reg_mask);
2838                else
2839                    shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2840                        This->input_reg_map[i], reg_mask, reg_mask);
2841                break;
2842
2843            default:
2844                shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2845                    This->input_reg_map[i], reg_mask, reg_mask);
2846         }
2847     }
2848 }
2849
2850 /*********************************************
2851  * Vertex Shader Specific Code begins here
2852  ********************************************/
2853
2854 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
2855     glsl_program_key_t *key;
2856
2857     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2858     key->vshader = entry->vshader;
2859     key->pshader = entry->pshader;
2860
2861     hash_table_put(priv->glsl_program_lookup, key, entry);
2862 }
2863
2864 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
2865         GLhandleARB vshader, GLhandleARB pshader) {
2866     glsl_program_key_t key;
2867
2868     key.vshader = vshader;
2869     key.pshader = pshader;
2870
2871     return (struct glsl_shader_prog_link *)hash_table_get(priv->glsl_program_lookup, &key);
2872 }
2873
2874 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, WineD3D_GL_Info *gl_info, struct glsl_shader_prog_link *entry) {
2875     glsl_program_key_t *key;
2876
2877     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2878     key->vshader = entry->vshader;
2879     key->pshader = entry->pshader;
2880     hash_table_remove(priv->glsl_program_lookup, key);
2881
2882     GL_EXTCALL(glDeleteObjectARB(entry->programId));
2883     if (entry->vshader) list_remove(&entry->vshader_entry);
2884     if (entry->pshader) list_remove(&entry->pshader_entry);
2885     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2886     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2887     HeapFree(GetProcessHeap(), 0, entry);
2888 }
2889
2890 static void handle_ps3_input(SHADER_BUFFER *buffer, semantic *semantics_in, semantic *semantics_out, WineD3D_GL_Info *gl_info, DWORD *map) {
2891     unsigned int i, j;
2892     DWORD usage_token, usage_token_out;
2893     DWORD register_token, register_token_out;
2894     DWORD usage, usage_idx, usage_out, usage_idx_out;
2895     DWORD *set;
2896     DWORD in_idx;
2897     DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
2898     char reg_mask[6], reg_mask_out[6];
2899     char destination[50];
2900
2901     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
2902
2903     if (!semantics_out) {
2904         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
2905         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
2906         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
2907     }
2908
2909     for(i = 0; i < MAX_REG_INPUT; i++) {
2910         usage_token = semantics_in[i].usage;
2911         if (!usage_token) continue;
2912
2913         in_idx = map[i];
2914         if (in_idx >= (in_count + 2)) {
2915             FIXME("More input varyings declared than supported, expect issues\n");
2916             continue;
2917         } else if(map[i] == -1) {
2918             /* Declared, but not read register */
2919             continue;
2920         }
2921
2922         if (in_idx == in_count) {
2923             sprintf(destination, "gl_FrontColor");
2924         } else if (in_idx == in_count + 1) {
2925             sprintf(destination, "gl_FrontSecondaryColor");
2926         } else {
2927             sprintf(destination, "IN[%u]", in_idx);
2928         }
2929
2930         register_token = semantics_in[i].reg;
2931
2932         usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2933         usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2934         set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
2935
2936         if(!semantics_out) {
2937             switch(usage) {
2938                 case WINED3DDECLUSAGE_COLOR:
2939                     if (usage_idx == 0)
2940                         shader_addline(buffer, "%s%s = front_color%s;\n",
2941                                        destination, reg_mask, reg_mask);
2942                     else if (usage_idx == 1)
2943                         shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
2944                                        destination, reg_mask, reg_mask);
2945                     else
2946                         shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2947                                        destination, reg_mask, reg_mask);
2948                     break;
2949
2950                 case WINED3DDECLUSAGE_TEXCOORD:
2951                     if (usage_idx < 8) {
2952                         shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
2953                                        destination, reg_mask, usage_idx, reg_mask);
2954                     } else {
2955                         shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2956                                        destination, reg_mask, reg_mask);
2957                     }
2958                     break;
2959
2960                 case WINED3DDECLUSAGE_FOG:
2961                     shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
2962                                    destination, reg_mask, reg_mask);
2963                     break;
2964
2965                 default:
2966                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2967                                    destination, reg_mask, reg_mask);
2968             }
2969         } else {
2970             BOOL found = FALSE;
2971             for(j = 0; j < MAX_REG_OUTPUT; j++) {
2972                 usage_token_out = semantics_out[j].usage;
2973                 if (!usage_token_out) continue;
2974                 register_token_out = semantics_out[j].reg;
2975
2976                 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2977                 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2978                 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
2979
2980                 if(usage == usage_out &&
2981                    usage_idx == usage_idx_out) {
2982                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
2983                                    destination, reg_mask, j, reg_mask);
2984                     found = TRUE;
2985                 }
2986             }
2987             if(!found) {
2988                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2989                                destination, reg_mask, reg_mask);
2990             }
2991         }
2992     }
2993
2994     /* This is solely to make the compiler / linker happy and avoid warning about undefined
2995      * varyings. It shouldn't result in any real code executed on the GPU, since all read
2996      * input varyings are assigned above, if the optimizer works properly.
2997      */
2998     for(i = 0; i < in_count + 2; i++) {
2999         if(set[i] != WINED3DSP_WRITEMASK_ALL) {
3000             unsigned int size = 0;
3001             memset(reg_mask, 0, sizeof(reg_mask));
3002             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3003                 reg_mask[size] = 'x';
3004                 size++;
3005             }
3006             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3007                 reg_mask[size] = 'y';
3008                 size++;
3009             }
3010             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3011                 reg_mask[size] = 'z';
3012                 size++;
3013             }
3014             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3015                 reg_mask[size] = 'w';
3016                 size++;
3017             }
3018
3019             if (i == in_count) {
3020                 sprintf(destination, "gl_FrontColor");
3021             } else if (i == in_count + 1) {
3022                 sprintf(destination, "gl_FrontSecondaryColor");
3023             } else {
3024                 sprintf(destination, "IN[%u]", i);
3025             }
3026
3027             if (size == 1) {
3028                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3029             } else {
3030                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3031             }
3032         }
3033     }
3034
3035     HeapFree(GetProcessHeap(), 0, set);
3036 }
3037
3038 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
3039         IWineD3DPixelShader *pixelshader,
3040         WineD3D_GL_Info *gl_info) {
3041     GLhandleARB ret = 0;
3042     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3043     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3044     IWineD3DDeviceImpl *device;
3045     DWORD vs_major = WINED3DSHADER_VERSION_MAJOR(vs->baseShader.hex_version);
3046     DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.hex_version) : 0;
3047     unsigned int i;
3048     SHADER_BUFFER buffer;
3049     DWORD usage_token;
3050     DWORD register_token;
3051     DWORD usage, usage_idx, writemask;
3052     char reg_mask[6];
3053     semantic *semantics_out, *semantics_in;
3054
3055     buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
3056     buffer.bsize = 0;
3057     buffer.lineNo = 0;
3058     buffer.newline = TRUE;
3059
3060     shader_addline(&buffer, "#version 120\n");
3061
3062     if(vs_major < 3 && ps_major < 3) {
3063         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3064          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3065          */
3066         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3067         if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0 &&
3068             !device->frag_pipe->ffp_proj_control) {
3069             shader_addline(&buffer, "void order_ps_input() {\n");
3070             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3071                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3072                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3073                     shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3074                 }
3075             }
3076             shader_addline(&buffer, "}\n");
3077         } else {
3078             shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
3079         }
3080     } else if(ps_major < 3 && vs_major >= 3) {
3081         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3082         semantics_out = vs->semantics_out;
3083
3084         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3085         for(i = 0; i < MAX_REG_OUTPUT; i++) {
3086             usage_token = semantics_out[i].usage;
3087             if (!usage_token) continue;
3088             register_token = semantics_out[i].reg;
3089
3090             usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3091             usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3092             writemask = shader_glsl_get_write_mask(register_token, reg_mask);
3093
3094             switch(usage) {
3095                 case WINED3DDECLUSAGE_COLOR:
3096                     if (usage_idx == 0)
3097                         shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3098                     else if (usage_idx == 1)
3099                         shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3100                     break;
3101
3102                 case WINED3DDECLUSAGE_POSITION:
3103                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3104                     break;
3105
3106                 case WINED3DDECLUSAGE_TEXCOORD:
3107                     if (usage_idx < 8) {
3108                         if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
3109
3110                         shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3111                                         usage_idx, reg_mask, i, reg_mask);
3112                         if(!(writemask & WINED3DSP_WRITEMASK_3)) {
3113                             shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
3114                         }
3115                     }
3116                     break;
3117
3118                 case WINED3DDECLUSAGE_PSIZE:
3119                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3120                     break;
3121
3122                 case WINED3DDECLUSAGE_FOG:
3123                     shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3124                     break;
3125
3126                 default:
3127                     break;
3128             }
3129         }
3130         shader_addline(&buffer, "}\n");
3131
3132     } else if(ps_major >= 3 && vs_major >= 3) {
3133         semantics_out = vs->semantics_out;
3134         semantics_in = ps->semantics_in;
3135
3136         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3137         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3138         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3139
3140         /* First, sort out position and point size. Those are not passed to the pixel shader */
3141         for(i = 0; i < MAX_REG_OUTPUT; i++) {
3142             usage_token = semantics_out[i].usage;
3143             if (!usage_token) continue;
3144             register_token = semantics_out[i].reg;
3145
3146             usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3147             usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3148             shader_glsl_get_write_mask(register_token, reg_mask);
3149
3150             switch(usage) {
3151                 case WINED3DDECLUSAGE_POSITION:
3152                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3153                     break;
3154
3155                 case WINED3DDECLUSAGE_PSIZE:
3156                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3157                     break;
3158
3159                 default:
3160                     break;
3161             }
3162         }
3163
3164         /* Then, fix the pixel shader input */
3165         handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3166
3167         shader_addline(&buffer, "}\n");
3168     } else if(ps_major >= 3 && vs_major < 3) {
3169         semantics_in = ps->semantics_in;
3170
3171         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3172         shader_addline(&buffer, "void order_ps_input() {\n");
3173         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3174          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3175          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3176          */
3177         handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3178         shader_addline(&buffer, "}\n");
3179     } else {
3180         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3181     }
3182
3183     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3184     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3185     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3186     checkGLcall("glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL)");
3187     GL_EXTCALL(glCompileShaderARB(ret));
3188     checkGLcall("glCompileShaderARB(ret)");
3189
3190     HeapFree(GetProcessHeap(), 0, buffer.buffer);
3191     return ret;
3192 }
3193
3194 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, WineD3D_GL_Info *gl_info, GLhandleARB programId, char prefix) {
3195     local_constant* lconst;
3196     GLuint tmp_loc;
3197     float *value;
3198     char glsl_name[8];
3199
3200     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3201         value = (float *) lconst->value;
3202         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3203         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3204         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3205     }
3206     checkGLcall("Hardcoding local constants\n");
3207 }
3208
3209 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3210  * It sets the programId on the current StateBlock (because it should be called
3211  * inside of the DrawPrimitive() part of the render loop).
3212  *
3213  * If a program for the given combination does not exist, create one, and store
3214  * the program in the hash table.  If it creates a program, it will link the
3215  * given objects, too.
3216  */
3217 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3218     IWineD3DDeviceImpl *This               = (IWineD3DDeviceImpl *)iface;
3219     struct shader_glsl_priv *priv          = (struct shader_glsl_priv *)This->shader_priv;
3220     WineD3D_GL_Info *gl_info               = &This->adapter->gl_info;
3221     IWineD3DPixelShader  *pshader          = This->stateBlock->pixelShader;
3222     IWineD3DVertexShader *vshader          = This->stateBlock->vertexShader;
3223     struct glsl_shader_prog_link *entry    = NULL;
3224     GLhandleARB programId                  = 0;
3225     GLhandleARB reorder_shader_id          = 0;
3226     int i;
3227     char glsl_name[8];
3228
3229     GLhandleARB vshader_id = use_vs ? ((IWineD3DBaseShaderImpl*)vshader)->baseShader.prgId : 0;
3230     GLhandleARB pshader_id = use_ps ? ((IWineD3DBaseShaderImpl*)pshader)->baseShader.prgId : 0;
3231     entry = get_glsl_program_entry(priv, vshader_id, pshader_id);
3232     if (entry) {
3233         priv->glsl_program = entry;
3234         return;
3235     }
3236
3237     /* If we get to this point, then no matching program exists, so we create one */
3238     programId = GL_EXTCALL(glCreateProgramObjectARB());
3239     TRACE("Created new GLSL shader program %u\n", programId);
3240
3241     /* Create the entry */
3242     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3243     entry->programId = programId;
3244     entry->vshader = vshader_id;
3245     entry->pshader = pshader_id;
3246     /* Add the hash table entry */
3247     add_glsl_program_entry(priv, entry);
3248
3249     /* Set the current program */
3250     priv->glsl_program = entry;
3251
3252     /* Attach GLSL vshader */
3253     if (vshader_id) {
3254         int max_attribs = 16;   /* TODO: Will this always be the case? It is at the moment... */
3255         char tmp_name[10];
3256
3257         reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3258         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3259         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3260         checkGLcall("glAttachObjectARB");
3261         /* Flag the reorder function for deletion, then it will be freed automatically when the program
3262          * is destroyed
3263          */
3264         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3265
3266         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3267         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3268         checkGLcall("glAttachObjectARB");
3269
3270         /* Bind vertex attributes to a corresponding index number to match
3271          * the same index numbers as ARB_vertex_programs (makes loading
3272          * vertex attributes simpler).  With this method, we can use the
3273          * exact same code to load the attributes later for both ARB and
3274          * GLSL shaders.
3275          *
3276          * We have to do this here because we need to know the Program ID
3277          * in order to make the bindings work, and it has to be done prior
3278          * to linking the GLSL program. */
3279         for (i = 0; i < max_attribs; ++i) {
3280             if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3281                 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3282                 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3283             }
3284         }
3285         checkGLcall("glBindAttribLocationARB");
3286
3287         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3288     }
3289
3290     /* Attach GLSL pshader */
3291     if (pshader_id) {
3292         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3293         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3294         checkGLcall("glAttachObjectARB");
3295
3296         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3297     }
3298
3299     /* Link the program */
3300     TRACE("Linking GLSL shader program %u\n", programId);
3301     GL_EXTCALL(glLinkProgramARB(programId));
3302     print_glsl_info_log(&GLINFO_LOCATION, programId);
3303
3304     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3305     for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3306         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3307         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3308     }
3309     for (i = 0; i < MAX_CONST_I; ++i) {
3310         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3311         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3312     }
3313     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3314     for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3315         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3316         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3317     }
3318     for (i = 0; i < MAX_CONST_I; ++i) {
3319         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3320         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3321     }
3322
3323     if(pshader) {
3324         for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3325             char name[32];
3326             sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3327             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3328             sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3329             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3330             sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3331             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3332         }
3333     }
3334
3335
3336     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3337     entry->srgb_comparison_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_comparison"));
3338     entry->srgb_mul_low_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_mul_low"));
3339     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3340     checkGLcall("Find glsl program uniform locations");
3341
3342     if (pshader && WINED3DSHADER_VERSION_MAJOR(((IWineD3DPixelShaderImpl *)pshader)->baseShader.hex_version) >= 3
3343             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > GL_LIMITS(glsl_varyings) / 4) {
3344         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
3345         entry->vertex_color_clamp = GL_FALSE;
3346     } else {
3347         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
3348     }
3349
3350     /* Set the shader to allow uniform loading on it */
3351     GL_EXTCALL(glUseProgramObjectARB(programId));
3352     checkGLcall("glUseProgramObjectARB(programId)");
3353
3354     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3355      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3356      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3357      * vertex shader with fixed function pixel processing is used we make sure that the card
3358      * supports enough samplers to allow the max number of vertex samplers with all possible
3359      * fixed function fragment processing setups. So once the program is linked these samplers
3360      * won't change.
3361      */
3362     if(vshader_id) {
3363         /* Load vertex shader samplers */
3364         shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3365     }
3366     if(pshader_id) {
3367         /* Load pixel shader samplers */
3368         shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3369     }
3370
3371     /* If the local constants do not have to be loaded with the environment constants,
3372      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
3373      * later
3374      */
3375     if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
3376         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
3377     }
3378     if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
3379         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
3380     }
3381 }
3382
3383 static GLhandleARB create_glsl_blt_shader(WineD3D_GL_Info *gl_info, enum tex_types tex_type) {
3384     GLhandleARB program_id;
3385     GLhandleARB vshader_id, pshader_id;
3386     const char *blt_vshader[] = {
3387         "#version 120\n"
3388         "void main(void)\n"
3389         "{\n"
3390         "    gl_Position = gl_Vertex;\n"
3391         "    gl_FrontColor = vec4(1.0);\n"
3392         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
3393         "}\n"
3394     };
3395
3396     const char *blt_pshaders[tex_type_count] = {
3397         /* tex_1d */
3398         NULL,
3399         /* tex_2d */
3400         "#version 120\n"
3401         "uniform sampler2D sampler;\n"
3402         "void main(void)\n"
3403         "{\n"
3404         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3405         "}\n",
3406         /* tex_3d */
3407         NULL,
3408         /* tex_cube */
3409         "#version 120\n"
3410         "uniform samplerCube sampler;\n"
3411         "void main(void)\n"
3412         "{\n"
3413         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
3414         "}\n",
3415         /* tex_rect */
3416         "#version 120\n"
3417         "#extension GL_ARB_texture_rectangle : enable\n"
3418         "uniform sampler2DRect sampler;\n"
3419         "void main(void)\n"
3420         "{\n"
3421         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
3422         "}\n",
3423     };
3424
3425     if (!blt_pshaders[tex_type])
3426     {
3427         FIXME("tex_type %#x not supported\n", tex_type);
3428         tex_type = tex_2d;
3429     }
3430
3431     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3432     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3433     GL_EXTCALL(glCompileShaderARB(vshader_id));
3434
3435     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3436     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
3437     GL_EXTCALL(glCompileShaderARB(pshader_id));
3438
3439     program_id = GL_EXTCALL(glCreateProgramObjectARB());
3440     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3441     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3442     GL_EXTCALL(glLinkProgramARB(program_id));
3443
3444     print_glsl_info_log(&GLINFO_LOCATION, program_id);
3445
3446     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3447      * is destroyed
3448      */
3449     GL_EXTCALL(glDeleteObjectARB(vshader_id));
3450     GL_EXTCALL(glDeleteObjectARB(pshader_id));
3451     return program_id;
3452 }
3453
3454 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3455     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3456     struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3457     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3458     GLhandleARB program_id = 0;
3459     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
3460
3461     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3462
3463     if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3464     else priv->glsl_program = NULL;
3465
3466     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3467
3468     if (old_vertex_color_clamp != current_vertex_color_clamp) {
3469         if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
3470             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
3471             checkGLcall("glClampColorARB");
3472         } else {
3473             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
3474         }
3475     }
3476
3477     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3478     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3479     GL_EXTCALL(glUseProgramObjectARB(program_id));
3480     checkGLcall("glUseProgramObjectARB");
3481 }
3482
3483 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
3484     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3485     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3486     struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3487     GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
3488
3489     if (!*blt_program) {
3490         GLhandleARB loc;
3491         *blt_program = create_glsl_blt_shader(gl_info, tex_type);
3492         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
3493         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3494         GL_EXTCALL(glUniform1iARB(loc, 0));
3495     } else {
3496         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3497     }
3498 }
3499
3500 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
3501     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3502     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3503     struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3504     GLhandleARB program_id;
3505
3506     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3507     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3508
3509     GL_EXTCALL(glUseProgramObjectARB(program_id));
3510     checkGLcall("glUseProgramObjectARB");
3511 }
3512
3513 static void shader_glsl_cleanup(IWineD3DDevice *iface) {
3514     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3515     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3516     GL_EXTCALL(glUseProgramObjectARB(0));
3517 }
3518
3519 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3520     struct list *linked_programs;
3521     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3522     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3523     struct shader_glsl_priv *priv = (struct shader_glsl_priv *)device->shader_priv;
3524     WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
3525
3526     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3527      * can be called from IWineD3DBaseShader::Release
3528      */
3529     char pshader = shader_is_pshader_version(This->baseShader.hex_version);
3530
3531     if(This->baseShader.prgId == 0) return;
3532     linked_programs = &This->baseShader.linked_programs;
3533
3534     TRACE("Deleting linked programs\n");
3535     if (linked_programs->next) {
3536         struct glsl_shader_prog_link *entry, *entry2;
3537
3538         if(pshader) {
3539             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3540                 delete_glsl_program_entry(priv, gl_info, entry);
3541             }
3542         } else {
3543             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3544                 delete_glsl_program_entry(priv, gl_info, entry);
3545             }
3546         }
3547     }
3548
3549     TRACE("Deleting shader object %u\n", This->baseShader.prgId);
3550     GL_EXTCALL(glDeleteObjectARB(This->baseShader.prgId));
3551     checkGLcall("glDeleteObjectARB");
3552     This->baseShader.prgId = 0;
3553     This->baseShader.is_compiled = FALSE;
3554 }
3555
3556 static unsigned int glsl_program_key_hash(void *key) {
3557     glsl_program_key_t *k = (glsl_program_key_t *)key;
3558
3559     unsigned int hash = k->vshader | k->pshader << 16;
3560     hash += ~(hash << 15);
3561     hash ^=  (hash >> 10);
3562     hash +=  (hash << 3);
3563     hash ^=  (hash >> 6);
3564     hash += ~(hash << 11);
3565     hash ^=  (hash >> 16);
3566
3567     return hash;
3568 }
3569
3570 static BOOL glsl_program_key_compare(void *keya, void *keyb) {
3571     glsl_program_key_t *ka = (glsl_program_key_t *)keya;
3572     glsl_program_key_t *kb = (glsl_program_key_t *)keyb;
3573
3574     return ka->vshader == kb->vshader && ka->pshader == kb->pshader;
3575 }
3576
3577 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3578     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3579     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3580     priv->glsl_program_lookup = hash_table_create(glsl_program_key_hash, glsl_program_key_compare);
3581     This->shader_priv = priv;
3582     return WINED3D_OK;
3583 }
3584
3585 static void shader_glsl_free(IWineD3DDevice *iface) {
3586     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3587     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3588     struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3589     int i;
3590
3591     for (i = 0; i < tex_type_count; ++i)
3592     {
3593         if (priv->depth_blt_program[i])
3594         {
3595             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
3596         }
3597     }
3598
3599     hash_table_destroy(priv->glsl_program_lookup, NULL, NULL);
3600
3601     HeapFree(GetProcessHeap(), 0, This->shader_priv);
3602     This->shader_priv = NULL;
3603 }
3604
3605 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3606     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3607     return FALSE;
3608 }
3609
3610 static void shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer) {
3611     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3612     shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3613     CONST DWORD *function = This->baseShader.function;
3614     const char *fragcolor;
3615     WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3616
3617     /* Create the hw GLSL shader object and assign it as the baseShader.prgId */
3618     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3619
3620     shader_addline(buffer, "#version 120\n");
3621
3622     if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3623         shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3624     }
3625     if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3626         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3627          * drivers write a warning if we don't do so
3628          */
3629         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3630     }
3631
3632     /* Base Declarations */
3633     shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3634
3635     /* Pack 3.0 inputs */
3636     if (This->baseShader.hex_version >= WINED3DPS_VERSION(3,0)) {
3637
3638         if(((IWineD3DDeviceImpl *) This->baseShader.device)->strided_streams.u.s.position_transformed) {
3639             This->vertexprocessing = pretransformed;
3640             pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3641         } else if(!use_vs((IWineD3DDeviceImpl *) This->baseShader.device)) {
3642             This->vertexprocessing = fixedfunction;
3643             pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3644         } else {
3645             This->vertexprocessing = vertexshader;
3646         }
3647     }
3648
3649     /* Base Shader Body */
3650     shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3651
3652     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3653     if (This->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
3654         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3655         if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3656             shader_addline(buffer, "gl_FragData[0] = R0;\n");
3657         else
3658             shader_addline(buffer, "gl_FragColor = R0;\n");
3659     }
3660
3661     if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3662         fragcolor = "gl_FragData[0]";
3663     } else {
3664         fragcolor = "gl_FragColor";
3665     }
3666     if(This->srgb_enabled) {
3667         shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3668                         fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3669                         srgb_sub_high, srgb_sub_high, srgb_sub_high);
3670         shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3671         shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3672         shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3673         shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3674         shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3675     }
3676     /* Pixel shader < 3.0 do not replace the fog stage.
3677      * This implements linear fog computation and blending.
3678      * TODO: non linear fog
3679      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3680      * -1/(e-s) and e/(e-s) respectively.
3681      */
3682     if(This->baseShader.hex_version < WINED3DPS_VERSION(3,0)) {
3683         shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * gl_Fog.start + gl_Fog.end, 0.0, 1.0);\n");
3684         shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
3685     }
3686
3687     shader_addline(buffer, "}\n");
3688
3689     TRACE("Compiling shader object %u\n", shader_obj);
3690     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3691     GL_EXTCALL(glCompileShaderARB(shader_obj));
3692     print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3693
3694     /* Store the shader object */
3695     This->baseShader.prgId = shader_obj;
3696 }
3697
3698 static void shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer) {
3699     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
3700     shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3701     CONST DWORD *function = This->baseShader.function;
3702     WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3703
3704     /* Create the hw GLSL shader program and assign it as the baseShader.prgId */
3705     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3706
3707     shader_addline(buffer, "#version 120\n");
3708
3709     /* Base Declarations */
3710     shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3711
3712     /* Base Shader Body */
3713     shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3714
3715     /* Unpack 3.0 outputs */
3716     if (This->baseShader.hex_version >= WINED3DVS_VERSION(3,0)) {
3717         shader_addline(buffer, "order_ps_input(OUT);\n");
3718     } else {
3719         shader_addline(buffer, "order_ps_input();\n");
3720     }
3721
3722     /* If this shader doesn't use fog copy the z coord to the fog coord so that we can use table fog */
3723     if (!reg_maps->fog)
3724         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3725
3726     /* Write the final position.
3727      *
3728      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3729      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3730      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3731      * contains 1.0 to allow a mad.
3732      */
3733     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3734     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3735
3736     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3737      *
3738      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3739      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3740      * which is the same as z = z * 2 - w.
3741      */
3742     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3743
3744     shader_addline(buffer, "}\n");
3745
3746     TRACE("Compiling shader object %u\n", shader_obj);
3747     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3748     GL_EXTCALL(glCompileShaderARB(shader_obj));
3749     print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3750
3751     /* Store the shader object */
3752     This->baseShader.prgId = shader_obj;
3753 }
3754
3755 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, WineD3D_GL_Info *gl_info, struct shader_caps *pCaps) {
3756     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
3757      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
3758      * vs_nv_version which is based on NV_vertex_program.
3759      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
3760      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
3761      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
3762      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
3763      */
3764     if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3765         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
3766     else
3767         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
3768     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
3769     pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
3770
3771     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
3772      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
3773      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
3774      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
3775      * in max native instructions. Intel and others also offer the info in this extension but they
3776      * don't support GLSL (at least on Windows).
3777      *
3778      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
3779      * of instructions is 512 or less we have to do with ps2.0 hardware.
3780      * NOTE: ps3.0 hardware requires 512 or more instructions but ati and nvidia offer 'enough' (1024 vs 4096) on their most basic ps3.0 hardware.
3781      */
3782     if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3783         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
3784     else
3785         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
3786
3787     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
3788      * Direct3D minimum requirement.
3789      *
3790      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
3791      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
3792      *
3793      * The problem is that the refrast clamps temporary results in the shader to
3794      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
3795      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
3796      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
3797      * offer a way to query this.
3798      */
3799     pCaps->PixelShader1xMaxValue = 8.0;
3800     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
3801 }
3802
3803 static BOOL shader_glsl_conv_supported(WINED3DFORMAT fmt) {
3804     TRACE("Checking shader format support for format %s:", debug_d3dformat(fmt));
3805     switch(fmt) {
3806         case WINED3DFMT_V8U8:
3807         case WINED3DFMT_V16U16:
3808         case WINED3DFMT_X8L8V8U8:
3809         case WINED3DFMT_L6V5U5:
3810         case WINED3DFMT_Q8W8V8U8:
3811         case WINED3DFMT_ATI2N:
3812             TRACE("[OK]\n");
3813             return TRUE;
3814         default:
3815             TRACE("[FAILED\n");
3816             return FALSE;
3817     }
3818 }
3819
3820 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
3821 {
3822     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
3823     /* WINED3DSIH_ADD           */ shader_glsl_arith,
3824     /* WINED3DSIH_BEM           */ pshader_glsl_bem,
3825     /* WINED3DSIH_BREAK         */ shader_glsl_break,
3826     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
3827     /* WINED3DSIH_BREAKP        */ NULL,
3828     /* WINED3DSIH_CALL          */ shader_glsl_call,
3829     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
3830     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
3831     /* WINED3DSIH_CND           */ shader_glsl_cnd,
3832     /* WINED3DSIH_CRS           */ shader_glsl_cross,
3833     /* WINED3DSIH_DCL           */ NULL,
3834     /* WINED3DSIH_DEF           */ NULL,
3835     /* WINED3DSIH_DEFB          */ NULL,
3836     /* WINED3DSIH_DEFI          */ NULL,
3837     /* WINED3DSIH_DP2ADD        */ pshader_glsl_dp2add,
3838     /* WINED3DSIH_DP3           */ shader_glsl_dot,
3839     /* WINED3DSIH_DP4           */ shader_glsl_dot,
3840     /* WINED3DSIH_DST           */ shader_glsl_dst,
3841     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
3842     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
3843     /* WINED3DSIH_ELSE          */ shader_glsl_else,
3844     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
3845     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
3846     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
3847     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
3848     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
3849     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
3850     /* WINED3DSIH_IF            */ shader_glsl_if,
3851     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
3852     /* WINED3DSIH_LABEL         */ shader_glsl_label,
3853     /* WINED3DSIH_LIT           */ shader_glsl_lit,
3854     /* WINED3DSIH_LOG           */ shader_glsl_log,
3855     /* WINED3DSIH_LOGP          */ shader_glsl_log,
3856     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
3857     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
3858     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
3859     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
3860     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
3861     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
3862     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
3863     /* WINED3DSIH_MAD           */ shader_glsl_mad,
3864     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
3865     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
3866     /* WINED3DSIH_MOV           */ shader_glsl_mov,
3867     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
3868     /* WINED3DSIH_MUL           */ shader_glsl_arith,
3869     /* WINED3DSIH_NOP           */ NULL,
3870     /* WINED3DSIH_NRM           */ shader_glsl_map2gl,
3871     /* WINED3DSIH_PHASE         */ NULL,
3872     /* WINED3DSIH_POW           */ shader_glsl_pow,
3873     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
3874     /* WINED3DSIH_REP           */ shader_glsl_rep,
3875     /* WINED3DSIH_RET           */ NULL,
3876     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
3877     /* WINED3DSIH_SETP          */ NULL,
3878     /* WINED3DSIH_SGE           */ shader_glsl_compare,
3879     /* WINED3DSIH_SGN           */ shader_glsl_map2gl,
3880     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
3881     /* WINED3DSIH_SLT           */ shader_glsl_compare,
3882     /* WINED3DSIH_SUB           */ shader_glsl_arith,
3883     /* WINED3DSIH_TEX           */ pshader_glsl_tex,
3884     /* WINED3DSIH_TEXBEM        */ pshader_glsl_texbem,
3885     /* WINED3DSIH_TEXBEML       */ pshader_glsl_texbem,
3886     /* WINED3DSIH_TEXCOORD      */ pshader_glsl_texcoord,
3887     /* WINED3DSIH_TEXDEPTH      */ pshader_glsl_texdepth,
3888     /* WINED3DSIH_TEXDP3        */ pshader_glsl_texdp3,
3889     /* WINED3DSIH_TEXDP3TEX     */ pshader_glsl_texdp3tex,
3890     /* WINED3DSIH_TEXKILL       */ pshader_glsl_texkill,
3891     /* WINED3DSIH_TEXLDD        */ NULL,
3892     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
3893     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_glsl_texm3x2depth,
3894     /* WINED3DSIH_TEXM3x2PAD    */ pshader_glsl_texm3x2pad,
3895     /* WINED3DSIH_TEXM3x2TEX    */ pshader_glsl_texm3x2tex,
3896     /* WINED3DSIH_TEXM3x3       */ pshader_glsl_texm3x3,
3897     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
3898     /* WINED3DSIH_TEXM3x3PAD    */ pshader_glsl_texm3x3pad,
3899     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_glsl_texm3x3spec,
3900     /* WINED3DSIH_TEXM3x3TEX    */ pshader_glsl_texm3x3tex,
3901     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_glsl_texm3x3vspec,
3902     /* WINED3DSIH_TEXREG2AR     */ pshader_glsl_texreg2ar,
3903     /* WINED3DSIH_TEXREG2GB     */ pshader_glsl_texreg2gb,
3904     /* WINED3DSIH_TEXREG2RGB    */ pshader_glsl_texreg2rgb,
3905 };
3906
3907 const shader_backend_t glsl_shader_backend = {
3908     shader_glsl_instruction_handler_table,
3909     shader_glsl_select,
3910     shader_glsl_select_depth_blt,
3911     shader_glsl_deselect_depth_blt,
3912     shader_glsl_load_constants,
3913     shader_glsl_cleanup,
3914     shader_glsl_color_correction,
3915     shader_glsl_destroy,
3916     shader_glsl_alloc,
3917     shader_glsl_free,
3918     shader_glsl_dirty_const,
3919     shader_glsl_generate_pshader,
3920     shader_glsl_generate_vshader,
3921     shader_glsl_get_caps,
3922     shader_glsl_conv_supported,
3923 };