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  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  20  */
  21
  22 /*
  23  * D3D shader asm has swizzles on source parameters, and write masks for
  24  * destination parameters. GLSL uses swizzles for both. The result of this is
  25  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  26  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  27  * mask for the destination parameter into account.
  28  */
  29
  30 #include "config.h"
  31 #include <stdio.h>
  32 #include "wined3d_private.h"
  33
  34 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  35 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  36
  37 #define GLINFO_LOCATION      (*gl_info)
  38
  39 typedef struct {
  40     char reg_name[50];
  41     char mask_str[6];
  42 } glsl_dst_param_t;
  43
  44 typedef struct {
  45     char reg_name[50];
  46     char param_str[100];
  47 } glsl_src_param_t;
  48
  49 typedef struct {
  50     const char *name;
  51     DWORD coord_mask;
  52 } glsl_sample_function_t;
  53
  54 /** Prints the GLSL info log which will contain error messages if they exist */
  55 void print_glsl_info_log(WineD3D_GL_Info *gl_info, GLhandleARB obj) {
  56
  57     int infologLength = 0;
  58     char *infoLog;
  59     int i;
  60     BOOL is_spam;
  61
  62     const char *spam[] = {
  63         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx        */
  64         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx        */
  65         "Fragment shader(s) linked, vertex shader(s) linked."               /* fglrx, no \n */
  66     };
  67
  68     GL_EXTCALL(glGetObjectParameterivARB(obj,
  69                GL_OBJECT_INFO_LOG_LENGTH_ARB,
  70                &infologLength));
  71
  72     /* A size of 1 is just a null-terminated string, so the log should be bigger than
  73      * that if there are errors. */
  74     if (infologLength > 1)
  75     {
  76         infoLog = HeapAlloc(GetProcessHeap(), 0, infologLength);
  77         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
  78         is_spam = FALSE;
  79
  80         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
  81             if(strcmp(infoLog, spam[i]) == 0) {
  82                 is_spam = TRUE;
  83                 break;
  84             }
  85         }
  86         if(is_spam) {
  87             TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
  88         } else {
  89             FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
  90         }
  91         HeapFree(GetProcessHeap(), 0, infoLog);
  92     }
  93 }
  94
  95 /**
  96  * Loads (pixel shader) samplers
  97  */
  98 static void shader_glsl_load_psamplers(
  99     WineD3D_GL_Info *gl_info,
 100     IWineD3DStateBlock* iface,
 101     GLhandleARB programId) {
 102
 103     IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
 104     GLhandleARB name_loc;
 105     int i;
 106     char sampler_name[20];
 107
 108     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 109         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 110         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 111         if (name_loc != -1) {
 112             int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[i];
 113             if (mapped_unit != -1 && mapped_unit < GL_LIMITS(fragment_samplers)) {
 114                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 115                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 116                 checkGLcall("glUniform1iARB");
 117             } else {
 118                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 119             }
 120         }
 121     }
 122 }
 123
 124 static void shader_glsl_load_vsamplers(WineD3D_GL_Info *gl_info, IWineD3DStateBlock* iface, GLhandleARB programId) {
 125     IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
 126     GLhandleARB name_loc;
 127     char sampler_name[20];
 128     int i;
 129
 130     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 131         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 132         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 133         if (name_loc != -1) {
 134             int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[MAX_FRAGMENT_SAMPLERS + i];
 135             if (mapped_unit != -1 && mapped_unit < GL_LIMITS(combined_samplers)) {
 136                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 137                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 138                 checkGLcall("glUniform1iARB");
 139             } else {
 140                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 141             }
 142         }
 143     }
 144 }
 145
 146 /**
 147  * Loads floating point constants (aka uniforms) into the currently set GLSL program.
 148  * When constant_list == NULL, it will load all the constants.
 149  */
 150 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl* This, WineD3D_GL_Info *gl_info,
 151         unsigned int max_constants, float* constants, GLhandleARB *constant_locations,
 152         struct list *constant_list) {
 153     constants_entry *constant;
 154     local_constant* lconst;
 155     GLhandleARB tmp_loc;
 156     DWORD i, j, k;
 157     DWORD *idx;
 158
 159     if (TRACE_ON(d3d_shader)) {
 160         LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
 161             idx = constant->idx;
 162             j = constant->count;
 163             while (j--) {
 164                 i = *idx++;
 165                 tmp_loc = constant_locations[i];
 166                 if (tmp_loc != -1) {
 167                     TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 168                             constants[i * 4 + 0], constants[i * 4 + 1],
 169                             constants[i * 4 + 2], constants[i * 4 + 3]);
 170                 }
 171             }
 172         }
 173     }
 174
 175     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 176     if(WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1 &&
 177        shader_is_pshader_version(This->baseShader.hex_version)) {
 178         float lcl_const[4];
 179
 180         LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
 181             idx = constant->idx;
 182             j = constant->count;
 183             while (j--) {
 184                 i = *idx++;
 185                 tmp_loc = constant_locations[i];
 186                 if (tmp_loc != -1) {
 187                     /* We found this uniform name in the program - go ahead and send the data */
 188                     k = i * 4;
 189                     if(constants[k + 0] < -1.0) lcl_const[0] = -1.0;
 190                     else if(constants[k + 0] > 1.0) lcl_const[0] = 1.0;
 191                     else lcl_const[0] = constants[k + 0];
 192                     if(constants[k + 1] < -1.0) lcl_const[1] = -1.0;
 193                     else if(constants[k + 1] > 1.0) lcl_const[1] = 1.0;
 194                     else lcl_const[1] = constants[k + 1];
 195                     if(constants[k + 2] < -1.0) lcl_const[2] = -1.0;
 196                     else if(constants[k + 2] > 1.0) lcl_const[2] = 1.0;
 197                     else lcl_const[2] = constants[k + 2];
 198                     if(constants[k + 3] < -1.0) lcl_const[3] = -1.0;
 199                     else if(constants[k + 3] > 1.0) lcl_const[3] = 1.0;
 200                     else lcl_const[3] = constants[k + 3];
 201
 202                     GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, lcl_const));
 203                 }
 204             }
 205         }
 206     } else {
 207         LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
 208             idx = constant->idx;
 209             j = constant->count;
 210             while (j--) {
 211                 i = *idx++;
 212                 tmp_loc = constant_locations[i];
 213                 if (tmp_loc != -1) {
 214                     /* We found this uniform name in the program - go ahead and send the data */
 215                     GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, constants + (i * 4)));
 216                 }
 217             }
 218         }
 219     }
 220     checkGLcall("glUniform4fvARB()");
 221
 222     if(!This->baseShader.load_local_constsF) {
 223         TRACE("No need to load local float constants for this shader\n");
 224         return;
 225     }
 226
 227     /* Load immediate constants */
 228     if (TRACE_ON(d3d_shader)) {
 229         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 230             tmp_loc = constant_locations[lconst->idx];
 231             if (tmp_loc != -1) {
 232                 GLfloat* values = (GLfloat*)lconst->value;
 233                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 234                         values[0], values[1], values[2], values[3]);
 235             }
 236         }
 237     }
 238     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 239     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 240         tmp_loc = constant_locations[lconst->idx];
 241         if (tmp_loc != -1) {
 242             /* We found this uniform name in the program - go ahead and send the data */
 243             GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, (GLfloat*)lconst->value));
 244         }
 245     }
 246     checkGLcall("glUniform4fvARB()");
 247 }
 248
 249 /**
 250  * Loads integer constants (aka uniforms) into the currently set GLSL program.
 251  * When @constants_set == NULL, it will load all the constants.
 252  */
 253 static void shader_glsl_load_constantsI(
 254     IWineD3DBaseShaderImpl* This,
 255     WineD3D_GL_Info *gl_info,
 256     GLhandleARB programId,
 257     GLhandleARB locations[MAX_CONST_I],
 258     unsigned max_constants,
 259     int* constants,
 260     BOOL* constants_set) {
 261
 262     int i;
 263     struct list* ptr;
 264
 265     for (i=0; i<max_constants; ++i) {
 266         if (NULL == constants_set || constants_set[i]) {
 267
 268             TRACE_(d3d_constants)("Loading constants %i: %i, %i, %i, %i\n",
 269                   i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 270
 271             /* We found this uniform name in the program - go ahead and send the data */
 272             GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 273             checkGLcall("glUniform4ivARB");
 274         }
 275     }
 276
 277     /* Load immediate constants */
 278     ptr = list_head(&This->baseShader.constantsI);
 279     while (ptr) {
 280         local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
 281         unsigned int idx = lconst->idx;
 282         GLint* values = (GLint*) lconst->value;
 283
 284         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 285             values[0], values[1], values[2], values[3]);
 286
 287         /* We found this uniform name in the program - go ahead and send the data */
 288         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 289         checkGLcall("glUniform4ivARB");
 290         ptr = list_next(&This->baseShader.constantsI, ptr);
 291     }
 292 }
 293
 294 /**
 295  * Loads boolean 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_constantsB(
 299     IWineD3DBaseShaderImpl* This,
 300     WineD3D_GL_Info *gl_info,
 301     GLhandleARB programId,
 302     unsigned max_constants,
 303     BOOL* constants,
 304     BOOL* constants_set) {
 305
 306     GLhandleARB tmp_loc;
 307     int i;
 308     char tmp_name[8];
 309     char is_pshader = shader_is_pshader_version(This->baseShader.hex_version);
 310     const char* prefix = is_pshader? "PB":"VB";
 311     struct list* ptr;
 312
 313     for (i=0; i<max_constants; ++i) {
 314         if (NULL == constants_set || constants_set[i]) {
 315
 316             TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 317
 318             /* TODO: Benchmark and see if it would be beneficial to store the
 319              * locations of the constants to avoid looking up each time */
 320             snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 321             tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 322             if (tmp_loc != -1) {
 323                 /* We found this uniform name in the program - go ahead and send the data */
 324                 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 325                 checkGLcall("glUniform1ivARB");
 326             }
 327         }
 328     }
 329
 330     /* Load immediate constants */
 331     ptr = list_head(&This->baseShader.constantsB);
 332     while (ptr) {
 333         local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
 334         unsigned int idx = lconst->idx;
 335         GLint* values = (GLint*) lconst->value;
 336
 337         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 338
 339         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 340         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 341         if (tmp_loc != -1) {
 342             /* We found this uniform name in the program - go ahead and send the data */
 343             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 344             checkGLcall("glUniform1ivARB");
 345         }
 346         ptr = list_next(&This->baseShader.constantsB, ptr);
 347     }
 348 }
 349
 350
 351
 352 /**
 353  * Loads the app-supplied constants into the currently set GLSL program.
 354  */
 355 void shader_glsl_load_constants(
 356     IWineD3DDevice* device,
 357     char usePixelShader,
 358     char useVertexShader) {
 359
 360     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
 361     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 362     WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 363
 364     GLhandleARB *constant_locations;
 365     struct list *constant_list;
 366     GLhandleARB programId;
 367     struct glsl_shader_prog_link *prog = stateBlock->glsl_program;
 368
 369     if (!prog) {
 370         /* No GLSL program set - nothing to do. */
 371         return;
 372     }
 373     programId = prog->programId;
 374
 375     if (useVertexShader) {
 376         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 377
 378         constant_locations = prog->vuniformF_locations;
 379         constant_list = &stateBlock->set_vconstantsF;
 380
 381         /* Load DirectX 9 float constants/uniforms for vertex shader */
 382         shader_glsl_load_constantsF(vshader, gl_info, GL_LIMITS(vshader_constantsF),
 383                 stateBlock->vertexShaderConstantF, constant_locations, constant_list);
 384
 385         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 386         shader_glsl_load_constantsI(vshader, gl_info, programId,
 387                                     prog->vuniformI_locations, MAX_CONST_I,
 388                                     stateBlock->vertexShaderConstantI,
 389                                     stateBlock->changed.vertexShaderConstantsI);
 390
 391         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 392         shader_glsl_load_constantsB(vshader, gl_info, programId, MAX_CONST_B,
 393                                     stateBlock->vertexShaderConstantB,
 394                                     stateBlock->changed.vertexShaderConstantsB);
 395
 396         /* Upload the position fixup params */
 397         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
 398         checkGLcall("glUniform4fvARB");
 399     }
 400
 401     if (usePixelShader) {
 402
 403         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 404
 405         constant_locations = prog->puniformF_locations;
 406         constant_list = &stateBlock->set_pconstantsF;
 407
 408         /* Load DirectX 9 float constants/uniforms for pixel shader */
 409         shader_glsl_load_constantsF(pshader, gl_info, GL_LIMITS(pshader_constantsF),
 410                 stateBlock->pixelShaderConstantF, constant_locations, constant_list);
 411
 412         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 413         shader_glsl_load_constantsI(pshader, gl_info, programId,
 414                                     prog->puniformI_locations, MAX_CONST_I,
 415                                     stateBlock->pixelShaderConstantI,
 416                                     stateBlock->changed.pixelShaderConstantsI);
 417
 418         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 419         shader_glsl_load_constantsB(pshader, gl_info, programId, MAX_CONST_B,
 420                                     stateBlock->pixelShaderConstantB,
 421                                     stateBlock->changed.pixelShaderConstantsB);
 422
 423         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 424          * It can't be 0 for a valid texbem instruction.
 425          */
 426         if(((IWineD3DPixelShaderImpl *) pshader)->needsbumpmat != -1) {
 427             float *data = (float *) &stateBlock->textureState[(int) ((IWineD3DPixelShaderImpl *) pshader)->needsbumpmat][WINED3DTSS_BUMPENVMAT00];
 428             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location, 1, 0, data));
 429             checkGLcall("glUniformMatrix2fvARB");
 430
 431             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 432              * is set too, so we can check that in the needsbumpmat check
 433              */
 434             if(((IWineD3DPixelShaderImpl *) pshader)->baseShader.reg_maps.luminanceparams != -1) {
 435                 int stage = ((IWineD3DPixelShaderImpl *) pshader)->baseShader.reg_maps.luminanceparams;
 436                 GLfloat *scale = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
 437                 GLfloat *offset = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
 438
 439                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location, 1, scale));
 440                 checkGLcall("glUniform1fvARB");
 441                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location, 1, offset));
 442                 checkGLcall("glUniform1fvARB");
 443             }
 444         } else if(((IWineD3DPixelShaderImpl *) pshader)->srgb_enabled &&
 445                   !((IWineD3DPixelShaderImpl *) pshader)->srgb_mode_hardcoded) {
 446             float comparison[4];
 447             float mul_low[4];
 448
 449             if(stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
 450                 comparison[0] = srgb_cmp; comparison[1] = srgb_cmp;
 451                 comparison[2] = srgb_cmp; comparison[3] = srgb_cmp;
 452
 453                 mul_low[0] = srgb_mul_low; mul_low[1] = srgb_mul_low;
 454                 mul_low[2] = srgb_mul_low; mul_low[3] = srgb_mul_low;
 455             } else {
 456                 comparison[0] = 1.0 / 0.0; comparison[1] = 1.0 / 0.0;
 457                 comparison[2] = 1.0 / 0.0; comparison[3] = 1.0 / 0.0;
 458
 459                 mul_low[0] = 1.0; mul_low[1] = 1.0;
 460                 mul_low[2] = 1.0; mul_low[3] = 1.0;
 461             }
 462
 463             GL_EXTCALL(glUniform4fvARB(prog->srgb_comparison_location, 1, comparison));
 464             GL_EXTCALL(glUniform4fvARB(prog->srgb_mul_low_location, 1, mul_low));
 465         }
 466         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 467             float correction_params[4];
 468             if(deviceImpl->render_offscreen) {
 469                 correction_params[0] = 0.0;
 470                 correction_params[1] = 1.0;
 471             } else {
 472                 /* position is window relative, not viewport relative */
 473                 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 474                 correction_params[1] = -1.0;
 475             }
 476             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 477         }
 478     }
 479 }
 480
 481 /** Generate the variable & register declarations for the GLSL output target */
 482 void shader_generate_glsl_declarations(
 483     IWineD3DBaseShader *iface,
 484     shader_reg_maps* reg_maps,
 485     SHADER_BUFFER* buffer,
 486     WineD3D_GL_Info* gl_info) {
 487
 488     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 489     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 490     int i;
 491     unsigned int extra_constants_needed = 0;
 492     local_constant* lconst;
 493
 494     /* There are some minor differences between pixel and vertex shaders */
 495     char pshader = shader_is_pshader_version(This->baseShader.hex_version);
 496     char prefix = pshader ? 'P' : 'V';
 497
 498     /* Prototype the subroutines */
 499     for (i = 0; i < This->baseShader.limits.label; i++) {
 500         if (reg_maps->labels[i])
 501             shader_addline(buffer, "void subroutine%u();\n", i);
 502     }
 503
 504     /* Declare the constants (aka uniforms) */
 505     if (This->baseShader.limits.constant_float > 0) {
 506         unsigned max_constantsF = min(This->baseShader.limits.constant_float,
 507                 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
 508         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 509     }
 510
 511     if (This->baseShader.limits.constant_int > 0)
 512         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 513
 514     if (This->baseShader.limits.constant_bool > 0)
 515         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 516
 517     if(!pshader) {
 518         shader_addline(buffer, "uniform vec4 posFixup;\n");
 519         /* Predeclaration; This function is added at link time based on the pixel shader.
 520          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 521          * that. We know the input to the reorder function at vertex shader compile time, so
 522          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 523          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 524          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 525          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 526          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 527          * inout.
 528          */
 529         if(This->baseShader.hex_version >= WINED3DVS_VERSION(3, 0)) {
 530             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 531         } else {
 532             shader_addline(buffer, "void order_ps_input();\n");
 533         }
 534     } else {
 535         IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
 536
 537         if(reg_maps->bumpmat != -1) {
 538             shader_addline(buffer, "uniform mat2 bumpenvmat;\n");
 539             if(reg_maps->luminanceparams) {
 540                 shader_addline(buffer, "uniform float luminancescale;\n");
 541                 shader_addline(buffer, "uniform float luminanceoffset;\n");
 542                 extra_constants_needed++;
 543             }
 544             extra_constants_needed++;
 545         }
 546
 547         if(device->stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
 548             ps_impl->srgb_enabled = 1;
 549             if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
 550                 shader_addline(buffer, "uniform vec4 srgb_mul_low;\n");
 551                 shader_addline(buffer, "uniform vec4 srgb_comparison;\n");
 552                 ps_impl->srgb_mode_hardcoded = 0;
 553                 extra_constants_needed++;
 554             } else {
 555                 ps_impl->srgb_mode_hardcoded = 1;
 556                 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
 557                                srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
 558                 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
 559                                srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
 560             }
 561         } else {
 562             IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
 563
 564             /* Do not write any srgb fixup into the shader to save shader size and processing time.
 565              * As a consequence, we can't toggle srgb write on without recompilation
 566              */
 567             ps_impl->srgb_enabled = 0;
 568             ps_impl->srgb_mode_hardcoded = 1;
 569         }
 570         if(reg_maps->vpos || reg_maps->usesdsy) {
 571             if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
 572                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
 573                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
 574                 extra_constants_needed++;
 575             } else {
 576                 /* This happens because we do not have proper tracking of the constant registers that are
 577                  * actually used, only the max limit of the shader version
 578                  */
 579                 FIXME("Cannot find a free uniform for vpos correction params\n");
 580                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
 581                                device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
 582                                device->render_offscreen ? 1.0 : -1.0);
 583             }
 584             shader_addline(buffer, "vec4 vpos;\n");
 585         }
 586     }
 587
 588     /* Declare texture samplers */
 589     for (i = 0; i < This->baseShader.limits.sampler; i++) {
 590         if (reg_maps->samplers[i]) {
 591
 592             DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
 593             switch (stype) {
 594
 595                 case WINED3DSTT_1D:
 596                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
 597                     break;
 598                 case WINED3DSTT_2D:
 599                     if(device->stateBlock->textures[i] &&
 600                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
 601                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
 602                     } else {
 603                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
 604                     }
 605                     break;
 606                 case WINED3DSTT_CUBE:
 607                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
 608                     break;
 609                 case WINED3DSTT_VOLUME:
 610                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
 611                     break;
 612                 default:
 613                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
 614                     FIXME("Unrecognized sampler type: %#x\n", stype);
 615                     break;
 616             }
 617         }
 618     }
 619
 620     /* Declare address variables */
 621     for (i = 0; i < This->baseShader.limits.address; i++) {
 622         if (reg_maps->address[i])
 623             shader_addline(buffer, "ivec4 A%d;\n", i);
 624     }
 625
 626     /* Declare texture coordinate temporaries and initialize them */
 627     for (i = 0; i < This->baseShader.limits.texcoord; i++) {
 628         if (reg_maps->texcoord[i])
 629             shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
 630     }
 631
 632     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
 633      * helper function shader that is linked in at link time
 634      */
 635     if(pshader && This->baseShader.hex_version >= WINED3DPS_VERSION(3, 0)) {
 636         if(use_vs(device)) {
 637             shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 638         } else {
 639             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
 640              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
 641              * pixel shader that reads the fixed function color into the packed input registers.
 642              */
 643             shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 644         }
 645     }
 646
 647     /* Declare output register temporaries */
 648     if(This->baseShader.limits.packed_output) {
 649         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
 650     }
 651
 652     /* Declare temporary variables */
 653     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 654         if (reg_maps->temporary[i])
 655             shader_addline(buffer, "vec4 R%u;\n", i);
 656     }
 657
 658     /* Declare attributes */
 659     for (i = 0; i < This->baseShader.limits.attributes; i++) {
 660         if (reg_maps->attributes[i])
 661             shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
 662     }
 663
 664     /* Declare loop registers aLx */
 665     for (i = 0; i < reg_maps->loop_depth; i++) {
 666         shader_addline(buffer, "int aL%u;\n", i);
 667         shader_addline(buffer, "int tmpInt%u;\n", i);
 668     }
 669
 670     /* Temporary variables for matrix operations */
 671     shader_addline(buffer, "vec4 tmp0;\n");
 672     shader_addline(buffer, "vec4 tmp1;\n");
 673
 674     /* Hardcodable local constants */
 675     if(!This->baseShader.load_local_constsF) {
 676         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 677             float *value = (float *) lconst->value;
 678             shader_addline(buffer, "const vec4 LC%u = vec4(%f, %f, %f, %f);\n", lconst->idx,
 679                            value[0], value[1], value[2], value[3]);
 680         }
 681     }
 682
 683     /* Start the main program */
 684     shader_addline(buffer, "void main() {\n");
 685     if(pshader && reg_maps->vpos) {
 686         shader_addline(buffer, "vpos = vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1) - 0.5;\n");
 687     }
 688 }
 689
 690 /*****************************************************************************
 691  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
 692  *
 693  * For more information, see http://wiki.winehq.org/DirectX-Shaders
 694  ****************************************************************************/
 695
 696 /* Prototypes */
 697 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
 698         const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
 699
 700 /** Used for opcode modifiers - They multiply the result by the specified amount */
 701 static const char * const shift_glsl_tab[] = {
 702     "",           /*  0 (none) */
 703     "2.0 * ",     /*  1 (x2)   */
 704     "4.0 * ",     /*  2 (x4)   */
 705     "8.0 * ",     /*  3 (x8)   */
 706     "16.0 * ",    /*  4 (x16)  */
 707     "32.0 * ",    /*  5 (x32)  */
 708     "",           /*  6 (x64)  */
 709     "",           /*  7 (x128) */
 710     "",           /*  8 (d256) */
 711     "",           /*  9 (d128) */
 712     "",           /* 10 (d64)  */
 713     "",           /* 11 (d32)  */
 714     "0.0625 * ",  /* 12 (d16)  */
 715     "0.125 * ",   /* 13 (d8)   */
 716     "0.25 * ",    /* 14 (d4)   */
 717     "0.5 * "      /* 15 (d2)   */
 718 };
 719
 720 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
 721 static void shader_glsl_gen_modifier (
 722     const DWORD instr,
 723     const char *in_reg,
 724     const char *in_regswizzle,
 725     char *out_str) {
 726
 727     out_str[0] = 0;
 728
 729     if (instr == WINED3DSIO_TEXKILL)
 730         return;
 731
 732     switch (instr & WINED3DSP_SRCMOD_MASK) {
 733     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
 734     case WINED3DSPSM_DW:
 735     case WINED3DSPSM_NONE:
 736         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
 737         break;
 738     case WINED3DSPSM_NEG:
 739         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
 740         break;
 741     case WINED3DSPSM_NOT:
 742         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
 743         break;
 744     case WINED3DSPSM_BIAS:
 745         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
 746         break;
 747     case WINED3DSPSM_BIASNEG:
 748         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
 749         break;
 750     case WINED3DSPSM_SIGN:
 751         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
 752         break;
 753     case WINED3DSPSM_SIGNNEG:
 754         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
 755         break;
 756     case WINED3DSPSM_COMP:
 757         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
 758         break;
 759     case WINED3DSPSM_X2:
 760         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
 761         break;
 762     case WINED3DSPSM_X2NEG:
 763         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
 764         break;
 765     case WINED3DSPSM_ABS:
 766         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
 767         break;
 768     case WINED3DSPSM_ABSNEG:
 769         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
 770         break;
 771     default:
 772         FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
 773         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
 774     }
 775 }
 776
 777 static BOOL constant_is_local(IWineD3DBaseShaderImpl* This, DWORD reg) {
 778     local_constant* lconst;
 779
 780     if(This->baseShader.load_local_constsF) return FALSE;
 781     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 782         if(lconst->idx == reg) return TRUE;
 783     }
 784     return FALSE;
 785
 786 }
 787
 788 /** Writes the GLSL variable name that corresponds to the register that the
 789  * DX opcode parameter is trying to access */
 790 static void shader_glsl_get_register_name(
 791     const DWORD param,
 792     const DWORD addr_token,
 793     char* regstr,
 794     BOOL* is_color,
 795     SHADER_OPCODE_ARG* arg) {
 796
 797     /* oPos, oFog and oPts in D3D */
 798     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
 799
 800     DWORD reg = param & WINED3DSP_REGNUM_MASK;
 801     DWORD regtype = shader_get_regtype(param);
 802     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
 803     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
 804     WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
 805
 806     char pshader = shader_is_pshader_version(This->baseShader.hex_version);
 807     char tmpStr[50];
 808
 809     *is_color = FALSE;
 810
 811     switch (regtype) {
 812     case WINED3DSPR_TEMP:
 813         sprintf(tmpStr, "R%u", reg);
 814     break;
 815     case WINED3DSPR_INPUT:
 816         if (pshader) {
 817             /* Pixel shaders >= 3.0 */
 818             if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3) {
 819                 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
 820                     glsl_src_param_t rel_param;
 821                     shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
 822
 823                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
 824                      * operation there
 825                      */
 826                     if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
 827                         sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str,
 828                                 ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]);
 829                     } else {
 830                         sprintf(tmpStr, "IN[%s]", rel_param.param_str);
 831                     }
 832                 } else {
 833                     sprintf(tmpStr, "IN[%u]",
 834                             ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]);
 835                 }
 836             } else {
 837                 if (reg==0)
 838                     strcpy(tmpStr, "gl_Color");
 839                 else
 840                     strcpy(tmpStr, "gl_SecondaryColor");
 841             }
 842         } else {
 843             if (vshader_input_is_color((IWineD3DVertexShader*) This, reg))
 844                *is_color = TRUE;
 845             sprintf(tmpStr, "attrib%u", reg);
 846         }
 847         break;
 848     case WINED3DSPR_CONST:
 849     {
 850         const char* prefix = pshader? "PC":"VC";
 851
 852         /* Relative addressing */
 853         if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
 854
 855            /* Relative addressing on shaders 2.0+ have a relative address token,
 856             * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
 857            if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 2)  {
 858                glsl_src_param_t rel_param;
 859                shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
 860                if(reg) {
 861                    sprintf(tmpStr, "%s[%s + %u]", prefix, rel_param.param_str, reg);
 862                } else {
 863                    sprintf(tmpStr, "%s[%s]", prefix, rel_param.param_str);
 864                }
 865            } else {
 866                if(reg) {
 867                    sprintf(tmpStr, "%s[A0.x + %u]", prefix, reg);
 868                } else {
 869                    sprintf(tmpStr, "%s[A0.x]", prefix);
 870                }
 871            }
 872
 873         } else {
 874             if(constant_is_local(This, reg)) {
 875                 sprintf(tmpStr, "LC%u", reg);
 876             } else {
 877                 sprintf(tmpStr, "%s[%u]", prefix, reg);
 878             }
 879         }
 880
 881         break;
 882     }
 883     case WINED3DSPR_CONSTINT:
 884         if (pshader)
 885             sprintf(tmpStr, "PI[%u]", reg);
 886         else
 887             sprintf(tmpStr, "VI[%u]", reg);
 888         break;
 889     case WINED3DSPR_CONSTBOOL:
 890         if (pshader)
 891             sprintf(tmpStr, "PB[%u]", reg);
 892         else
 893             sprintf(tmpStr, "VB[%u]", reg);
 894         break;
 895     case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
 896         if (pshader) {
 897             sprintf(tmpStr, "T%u", reg);
 898         } else {
 899             sprintf(tmpStr, "A%u", reg);
 900         }
 901     break;
 902     case WINED3DSPR_LOOP:
 903         sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
 904     break;
 905     case WINED3DSPR_SAMPLER:
 906         if (pshader)
 907             sprintf(tmpStr, "Psampler%u", reg);
 908         else
 909             sprintf(tmpStr, "Vsampler%u", reg);
 910     break;
 911     case WINED3DSPR_COLOROUT:
 912         if (reg >= GL_LIMITS(buffers)) {
 913             WARN("Write to render target %u, only %d supported\n", reg, 4);
 914         }
 915         if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
 916             sprintf(tmpStr, "gl_FragData[%u]", reg);
 917         } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
 918             sprintf(tmpStr, "gl_FragColor");
 919         }
 920     break;
 921     case WINED3DSPR_RASTOUT:
 922         sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
 923     break;
 924     case WINED3DSPR_DEPTHOUT:
 925         sprintf(tmpStr, "gl_FragDepth");
 926     break;
 927     case WINED3DSPR_ATTROUT:
 928         if (reg == 0) {
 929             sprintf(tmpStr, "gl_FrontColor");
 930         } else {
 931             sprintf(tmpStr, "gl_FrontSecondaryColor");
 932         }
 933     break;
 934     case WINED3DSPR_TEXCRDOUT:
 935         /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
 936         if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3)
 937             sprintf(tmpStr, "OUT[%u]", reg);
 938         else
 939             sprintf(tmpStr, "gl_TexCoord[%u]", reg);
 940     break;
 941     case WINED3DSPR_MISCTYPE:
 942         if (reg == 0) {
 943             /* vPos */
 944             sprintf(tmpStr, "vpos");
 945         } else if (reg == 1){
 946             /* Note that gl_FrontFacing is a bool, while vFace is
 947              * a float for which the sign determines front/back
 948              */
 949             sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
 950         } else {
 951             FIXME("Unhandled misctype register %d\n", reg);
 952             sprintf(tmpStr, "unrecognized_register");
 953         }
 954         break;
 955     default:
 956         FIXME("Unhandled register name Type(%d)\n", regtype);
 957         sprintf(tmpStr, "unrecognized_register");
 958     break;
 959     }
 960
 961     strcat(regstr, tmpStr);
 962 }
 963
 964 /* Get the GLSL write mask for the destination register */
 965 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
 966     char *ptr = write_mask;
 967     DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
 968
 969     if (shader_is_scalar(param)) {
 970         mask = WINED3DSP_WRITEMASK_0;
 971     } else {
 972         *ptr++ = '.';
 973         if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 974         if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 975         if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 976         if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 977     }
 978
 979     *ptr = '\0';
 980
 981     return mask;
 982 }
 983
 984 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
 985     unsigned int size = 0;
 986
 987     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
 988     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
 989     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
 990     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
 991
 992     return size;
 993 }
 994
 995 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
 996     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 997      * but addressed as "rgba". To fix this we need to swap the register's x
 998      * and z components. */
 999     DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1000     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1001     char *ptr = swizzle_str;
1002
1003     if (!shader_is_scalar(param)) {
1004         *ptr++ = '.';
1005         /* swizzle bits fields: wwzzyyxx */
1006         if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1007         if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1008         if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1009         if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1010     }
1011
1012     *ptr = '\0';
1013 }
1014
1015 /* From a given parameter token, generate the corresponding GLSL string.
1016  * Also, return the actual register name and swizzle in case the
1017  * caller needs this information as well. */
1018 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1019         const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param) {
1020     BOOL is_color = FALSE;
1021     char swizzle_str[6];
1022
1023     src_param->reg_name[0] = '\0';
1024     src_param->param_str[0] = '\0';
1025     swizzle_str[0] = '\0';
1026
1027     shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1028
1029     shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1030     shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1031 }
1032
1033 /* From a given parameter token, generate the corresponding GLSL string.
1034  * Also, return the actual register name and swizzle in case the
1035  * caller needs this information as well. */
1036 static DWORD shader_glsl_add_dst_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1037         const DWORD addr_token, glsl_dst_param_t *dst_param) {
1038     BOOL is_color = FALSE;
1039
1040     dst_param->mask_str[0] = '\0';
1041     dst_param->reg_name[0] = '\0';
1042
1043     shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1044     return shader_glsl_get_write_mask(param, dst_param->mask_str);
1045 }
1046
1047 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1048 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg, const DWORD param) {
1049     glsl_dst_param_t dst_param;
1050     DWORD mask;
1051     int shift;
1052
1053     mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1054
1055     if(mask) {
1056         shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1057         shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1058     }
1059
1060     return mask;
1061 }
1062
1063 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1064 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg) {
1065     return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1066 }
1067
1068 /** Process GLSL instruction modifiers */
1069 void shader_glsl_add_instruction_modifiers(SHADER_OPCODE_ARG* arg) {
1070
1071     DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1072
1073     if (arg->opcode->dst_token && mask != 0) {
1074         glsl_dst_param_t dst_param;
1075
1076         shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1077
1078         if (mask & WINED3DSPDM_SATURATE) {
1079             /* _SAT means to clamp the value of the register to between 0 and 1 */
1080             shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1081                     dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1082         }
1083         if (mask & WINED3DSPDM_MSAMPCENTROID) {
1084             FIXME("_centroid modifier not handled\n");
1085         }
1086         if (mask & WINED3DSPDM_PARTIALPRECISION) {
1087             /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1088         }
1089     }
1090 }
1091
1092 static inline const char* shader_get_comp_op(
1093     const DWORD opcode) {
1094
1095     DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1096     switch (op) {
1097         case COMPARISON_GT: return ">";
1098         case COMPARISON_EQ: return "==";
1099         case COMPARISON_GE: return ">=";
1100         case COMPARISON_LT: return "<";
1101         case COMPARISON_NE: return "!=";
1102         case COMPARISON_LE: return "<=";
1103         default:
1104             FIXME("Unrecognized comparison value: %u\n", op);
1105             return "(\?\?)";
1106     }
1107 }
1108
1109 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1110     /* Note that there's no such thing as a projected cube texture. */
1111     switch(sampler_type) {
1112         case WINED3DSTT_1D:
1113             sample_function->name = projected ? "texture1DProj" : "texture1D";
1114             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1115             break;
1116         case WINED3DSTT_2D:
1117             if(texrect) {
1118                 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1119             } else {
1120                 sample_function->name = projected ? "texture2DProj" : "texture2D";
1121             }
1122             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1123             break;
1124         case WINED3DSTT_CUBE:
1125             sample_function->name = "textureCube";
1126             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1127             break;
1128         case WINED3DSTT_VOLUME:
1129             sample_function->name = projected ? "texture3DProj" : "texture3D";
1130             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1131             break;
1132         default:
1133             sample_function->name = "";
1134             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1135             break;
1136     }
1137 }
1138
1139 static void shader_glsl_color_correction(SHADER_OPCODE_ARG* arg) {
1140     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1141     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) shader->baseShader.device;
1142     WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
1143     glsl_dst_param_t dst_param;
1144     glsl_dst_param_t dst_param2;
1145     WINED3DFORMAT fmt;
1146     WINED3DFORMAT conversion_group;
1147     IWineD3DBaseTextureImpl *texture;
1148     DWORD mask, mask_size;
1149     UINT i;
1150     BOOL recorded = FALSE;
1151     DWORD sampler_idx;
1152     DWORD hex_version = shader->baseShader.hex_version;
1153
1154     switch(arg->opcode->opcode) {
1155         case WINED3DSIO_TEX:
1156             if (hex_version < WINED3DPS_VERSION(2,0)) {
1157                 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1158             } else {
1159                 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
1160             }
1161             break;
1162
1163         case WINED3DSIO_TEXLDL:
1164             FIXME("Add color fixup for vertex texture WINED3DSIO_TEXLDL\n");
1165             return;
1166
1167         case WINED3DSIO_TEXDP3TEX:
1168         case WINED3DSIO_TEXM3x3TEX:
1169         case WINED3DSIO_TEXM3x3SPEC:
1170         case WINED3DSIO_TEXM3x3VSPEC:
1171         case WINED3DSIO_TEXBEM:
1172         case WINED3DSIO_TEXREG2AR:
1173         case WINED3DSIO_TEXREG2GB:
1174         case WINED3DSIO_TEXREG2RGB:
1175             sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1176             break;
1177
1178         default:
1179             /* Not a texture sampling instruction, nothing to do */
1180             return;
1181     };
1182
1183     texture = (IWineD3DBaseTextureImpl *) deviceImpl->stateBlock->textures[sampler_idx];
1184     if(texture) {
1185         fmt = texture->resource.format;
1186         conversion_group = texture->baseTexture.shader_conversion_group;
1187     } else {
1188         fmt = WINED3DFMT_UNKNOWN;
1189         conversion_group = WINED3DFMT_UNKNOWN;
1190     }
1191
1192     /* before doing anything, record the sampler with the format in the format conversion list,
1193      * but check if it's not there already
1194      */
1195     for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
1196         if(shader->baseShader.sampled_samplers[i] == sampler_idx) {
1197             recorded = TRUE;
1198             break;
1199         }
1200     }
1201     if(!recorded) {
1202         shader->baseShader.sampled_samplers[shader->baseShader.num_sampled_samplers] = sampler_idx;
1203         shader->baseShader.num_sampled_samplers++;
1204         shader->baseShader.sampled_format[sampler_idx] = conversion_group;
1205     }
1206
1207     switch(fmt) {
1208         case WINED3DFMT_V8U8:
1209         case WINED3DFMT_V16U16:
1210             if(GL_SUPPORT(NV_TEXTURE_SHADER) ||
1211               (GL_SUPPORT(ATI_ENVMAP_BUMPMAP) && fmt == WINED3DFMT_V8U8)) {
1212                 /* The 3rd channel returns 1.0 in d3d, but 0.0 in gl. Fix this while we're at it :-) */
1213                 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_2, &dst_param);
1214                 mask_size = shader_glsl_get_write_mask_size(mask);
1215                 if(mask_size >= 3) {
1216                     shader_addline(arg->buffer, "%s.%c = 1.0;\n", dst_param.reg_name, dst_param.mask_str[3]);
1217                 }
1218             } else {
1219                 /* Correct the sign, but leave the blue as it is - it was loaded correctly already */
1220                 mask = shader_glsl_add_dst_param(arg, arg->dst,
1221                                           WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1,
1222                                           &dst_param);
1223                 mask_size = shader_glsl_get_write_mask_size(mask);
1224                 if(mask_size >= 2) {
1225                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1226                     dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1227                     dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1228                 } else if(mask_size == 1) {
1229                     shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str[1],
1230                     dst_param.reg_name, dst_param.mask_str[1]);
1231                 }
1232             }
1233             break;
1234
1235         case WINED3DFMT_X8L8V8U8:
1236             if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1237                 /* Red and blue are the signed channels, fix them up; Blue(=L) is correct already,
1238                  * and a(X) is always 1.0
1239                  */
1240                 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1241                 mask_size = shader_glsl_get_write_mask_size(mask);
1242                 if(mask_size >= 2) {
1243                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1244                                    dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1245                                    dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1246                 } else if(mask_size == 1) {
1247                     shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1248                                    dst_param.reg_name, dst_param.mask_str[1],
1249                                    dst_param.reg_name, dst_param.mask_str[1]);
1250                 }
1251             }
1252             break;
1253
1254         case WINED3DFMT_L6V5U5:
1255             if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1256                 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1257                 mask_size = shader_glsl_get_write_mask_size(mask);
1258                 shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_2, &dst_param2);
1259                 if(mask_size >= 3) {
1260                     /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1261                     shader_addline(arg->buffer, "tmp0.g = %s.%c;\n",
1262                                    dst_param.reg_name, dst_param.mask_str[2]);
1263                     shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1264                                    dst_param.reg_name, dst_param.mask_str[2], dst_param.mask_str[1],
1265                                    dst_param2.reg_name, dst_param.mask_str[1], dst_param.mask_str[3]);
1266                     shader_addline(arg->buffer, "%s.%c = tmp0.g;\n", dst_param.reg_name,
1267                                    dst_param.mask_str[3]);
1268                 } else if(mask_size == 2) {
1269                     /* This is bad: We have VL, but we need VU */
1270                     FIXME("2 components sampled from a converted L6V5U5 texture\n");
1271                 } else {
1272                     shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1273                                    dst_param.reg_name, dst_param.mask_str[1],
1274                                    dst_param2.reg_name, dst_param.mask_str[1]);
1275                 }
1276             }
1277             break;
1278
1279         case WINED3DFMT_Q8W8V8U8:
1280             if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1281                 /* Correct the sign in all channels. The writemask just applies as-is, no
1282                  * need for checking the mask size
1283                  */
1284                 shader_glsl_add_dst_param(arg, arg->dst,
1285                                           WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 |
1286                                           WINED3DSP_WRITEMASK_2 | WINED3DSP_WRITEMASK_3,
1287                                           &dst_param);
1288                 shader_addline(arg->buffer, "%s%s = %s%s * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str,
1289                                dst_param.reg_name, dst_param.mask_str);
1290             }
1291             break;
1292
1293             /* stupid compiler */
1294         default:
1295             break;
1296     }
1297 }
1298
1299 /*****************************************************************************
1300  *
1301  * Begin processing individual instruction opcodes
1302  *
1303  ****************************************************************************/
1304
1305 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1306 void shader_glsl_arith(SHADER_OPCODE_ARG* arg) {
1307     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1308     SHADER_BUFFER* buffer = arg->buffer;
1309     glsl_src_param_t src0_param;
1310     glsl_src_param_t src1_param;
1311     DWORD write_mask;
1312     char op;
1313
1314     /* Determine the GLSL operator to use based on the opcode */
1315     switch (curOpcode->opcode) {
1316         case WINED3DSIO_MUL: op = '*'; break;
1317         case WINED3DSIO_ADD: op = '+'; break;
1318         case WINED3DSIO_SUB: op = '-'; break;
1319         default:
1320             op = ' ';
1321             FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1322             break;
1323     }
1324
1325     write_mask = shader_glsl_append_dst(buffer, arg);
1326     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1327     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1328     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1329 }
1330
1331 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1332 void shader_glsl_mov(SHADER_OPCODE_ARG* arg) {
1333     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1334     SHADER_BUFFER* buffer = arg->buffer;
1335     glsl_src_param_t src0_param;
1336     DWORD write_mask;
1337
1338     write_mask = shader_glsl_append_dst(buffer, arg);
1339     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1340
1341     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1342      * shader versions WINED3DSIO_MOVA is used for this. */
1343     if ((WINED3DSHADER_VERSION_MAJOR(shader->baseShader.hex_version) == 1 &&
1344             !shader_is_pshader_version(shader->baseShader.hex_version) &&
1345             shader_get_regtype(arg->dst) == WINED3DSPR_ADDR)) {
1346         /* This is a simple floor() */
1347         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1348         if (mask_size > 1) {
1349             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1350         } else {
1351             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1352         }
1353     } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1354         /* We need to *round* to the nearest int here. */
1355         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1356         if (mask_size > 1) {
1357             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);
1358         } else {
1359             shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1360         }
1361     } else {
1362         shader_addline(buffer, "%s);\n", src0_param.param_str);
1363     }
1364 }
1365
1366 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1367 void shader_glsl_dot(SHADER_OPCODE_ARG* arg) {
1368     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1369     SHADER_BUFFER* buffer = arg->buffer;
1370     glsl_src_param_t src0_param;
1371     glsl_src_param_t src1_param;
1372     DWORD dst_write_mask, src_write_mask;
1373     unsigned int dst_size = 0;
1374
1375     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1376     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1377
1378     /* dp3 works on vec3, dp4 on vec4 */
1379     if (curOpcode->opcode == WINED3DSIO_DP4) {
1380         src_write_mask = WINED3DSP_WRITEMASK_ALL;
1381     } else {
1382         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1383     }
1384
1385     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1386     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1387
1388     if (dst_size > 1) {
1389         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1390     } else {
1391         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1392     }
1393 }
1394
1395 /* Note that this instruction has some restrictions. The destination write mask
1396  * can't contain the w component, and the source swizzles have to be .xyzw */
1397 void shader_glsl_cross(SHADER_OPCODE_ARG *arg) {
1398     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1399     glsl_src_param_t src0_param;
1400     glsl_src_param_t src1_param;
1401     char dst_mask[6];
1402
1403     shader_glsl_get_write_mask(arg->dst, dst_mask);
1404     shader_glsl_append_dst(arg->buffer, arg);
1405     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1406     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1407     shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1408 }
1409
1410 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1411  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1412  * GLSL uses the value as-is. */
1413 void shader_glsl_pow(SHADER_OPCODE_ARG *arg) {
1414     SHADER_BUFFER *buffer = arg->buffer;
1415     glsl_src_param_t src0_param;
1416     glsl_src_param_t src1_param;
1417     DWORD dst_write_mask;
1418     unsigned int dst_size;
1419
1420     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1421     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1422
1423     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1424     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1425
1426     if (dst_size > 1) {
1427         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1428     } else {
1429         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1430     }
1431 }
1432
1433 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1434  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1435  * GLSL uses the value as-is. */
1436 void shader_glsl_log(SHADER_OPCODE_ARG *arg) {
1437     SHADER_BUFFER *buffer = arg->buffer;
1438     glsl_src_param_t src0_param;
1439     DWORD dst_write_mask;
1440     unsigned int dst_size;
1441
1442     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1443     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1444
1445     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1446
1447     if (dst_size > 1) {
1448         shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1449     } else {
1450         shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1451     }
1452 }
1453
1454 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1455 void shader_glsl_map2gl(SHADER_OPCODE_ARG* arg) {
1456     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1457     SHADER_BUFFER* buffer = arg->buffer;
1458     glsl_src_param_t src_param;
1459     const char *instruction;
1460     char arguments[256];
1461     DWORD write_mask;
1462     unsigned i;
1463
1464     /* Determine the GLSL function to use based on the opcode */
1465     /* TODO: Possibly make this a table for faster lookups */
1466     switch (curOpcode->opcode) {
1467         case WINED3DSIO_MIN: instruction = "min"; break;
1468         case WINED3DSIO_MAX: instruction = "max"; break;
1469         case WINED3DSIO_ABS: instruction = "abs"; break;
1470         case WINED3DSIO_FRC: instruction = "fract"; break;
1471         case WINED3DSIO_NRM: instruction = "normalize"; break;
1472         case WINED3DSIO_LOGP:
1473         case WINED3DSIO_LOG: instruction = "log2"; break;
1474         case WINED3DSIO_EXP: instruction = "exp2"; break;
1475         case WINED3DSIO_SGN: instruction = "sign"; break;
1476         case WINED3DSIO_DSX: instruction = "dFdx"; break;
1477         case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1478         default: instruction = "";
1479             FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1480             break;
1481     }
1482
1483     write_mask = shader_glsl_append_dst(buffer, arg);
1484
1485     arguments[0] = '\0';
1486     if (curOpcode->num_params > 0) {
1487         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1488         strcat(arguments, src_param.param_str);
1489         for (i = 2; i < curOpcode->num_params; ++i) {
1490             strcat(arguments, ", ");
1491             shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1492             strcat(arguments, src_param.param_str);
1493         }
1494     }
1495
1496     shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1497 }
1498
1499 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1500  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1501  *   dst.x = 2^(floor(src))
1502  *   dst.y = src - floor(src)
1503  *   dst.z = 2^src   (partial precision is allowed, but optional)
1504  *   dst.w = 1.0;
1505  * For 2.0 shaders, just do this (honoring writemask and swizzle):
1506  *   dst = 2^src;    (partial precision is allowed, but optional)
1507  */
1508 void shader_glsl_expp(SHADER_OPCODE_ARG* arg) {
1509     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1510     glsl_src_param_t src_param;
1511
1512     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1513
1514     if (shader->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
1515         char dst_mask[6];
1516
1517         shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1518         shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1519         shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1520         shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1521
1522         shader_glsl_append_dst(arg->buffer, arg);
1523         shader_glsl_get_write_mask(arg->dst, dst_mask);
1524         shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1525     } else {
1526         DWORD write_mask;
1527         unsigned int mask_size;
1528
1529         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1530         mask_size = shader_glsl_get_write_mask_size(write_mask);
1531
1532         if (mask_size > 1) {
1533             shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1534         } else {
1535             shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1536         }
1537     }
1538 }
1539
1540 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1541 void shader_glsl_rcp(SHADER_OPCODE_ARG* arg) {
1542     glsl_src_param_t src_param;
1543     DWORD write_mask;
1544     unsigned int mask_size;
1545
1546     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1547     mask_size = shader_glsl_get_write_mask_size(write_mask);
1548     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1549
1550     if (mask_size > 1) {
1551         shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1552     } else {
1553         shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1554     }
1555 }
1556
1557 void shader_glsl_rsq(SHADER_OPCODE_ARG* arg) {
1558     SHADER_BUFFER* buffer = arg->buffer;
1559     glsl_src_param_t src_param;
1560     DWORD write_mask;
1561     unsigned int mask_size;
1562
1563     write_mask = shader_glsl_append_dst(buffer, arg);
1564     mask_size = shader_glsl_get_write_mask_size(write_mask);
1565
1566     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1567
1568     if (mask_size > 1) {
1569         shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1570     } else {
1571         shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1572     }
1573 }
1574
1575 /** Process signed comparison opcodes in GLSL. */
1576 void shader_glsl_compare(SHADER_OPCODE_ARG* arg) {
1577     glsl_src_param_t src0_param;
1578     glsl_src_param_t src1_param;
1579     DWORD write_mask;
1580     unsigned int mask_size;
1581
1582     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1583     mask_size = shader_glsl_get_write_mask_size(write_mask);
1584     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1585     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1586
1587     if (mask_size > 1) {
1588         const char *compare;
1589
1590         switch(arg->opcode->opcode) {
1591             case WINED3DSIO_SLT: compare = "lessThan"; break;
1592             case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1593             default: compare = "";
1594                 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1595         }
1596
1597         shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1598                 src0_param.param_str, src1_param.param_str);
1599     } else {
1600         switch(arg->opcode->opcode) {
1601             case WINED3DSIO_SLT:
1602                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1603                  * to return 0.0 but step returns 1.0 because step is not < x
1604                  * An alternative is a bvec compare padded with an unused second component.
1605                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1606                  * issue. Playing with not() is not possible either because not() does not accept
1607                  * a scalar.
1608                  */
1609                 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1610                 break;
1611             case WINED3DSIO_SGE:
1612                 /* Here we can use the step() function and safe a conditional */
1613                 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1614                 break;
1615             default:
1616                 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1617         }
1618
1619     }
1620 }
1621
1622 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1623 void shader_glsl_cmp(SHADER_OPCODE_ARG* arg) {
1624     glsl_src_param_t src0_param;
1625     glsl_src_param_t src1_param;
1626     glsl_src_param_t src2_param;
1627     DWORD write_mask, cmp_channel = 0;
1628     unsigned int i, j;
1629     char mask_char[6];
1630     BOOL temp_destination = FALSE;
1631
1632     DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1633     DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1634     DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1635     DWORD src0regtype = shader_get_regtype(arg->src[0]);
1636     DWORD src1regtype = shader_get_regtype(arg->src[1]);
1637     DWORD src2regtype = shader_get_regtype(arg->src[2]);
1638     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1639     DWORD dstregtype = shader_get_regtype(arg->dst);
1640
1641     /* Cycle through all source0 channels */
1642     for (i=0; i<4; i++) {
1643         write_mask = 0;
1644         /* Find the destination channels which use the current source0 channel */
1645         for (j=0; j<4; j++) {
1646             if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1647                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1648                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1649             }
1650         }
1651
1652         /* Splitting the cmp instruction up in multiple lines imposes a problem:
1653          * The first lines may overwrite source parameters of the following lines.
1654          * Deal with that by using a temporary destination register if needed
1655          */
1656         if((src0reg == dstreg && src0regtype == dstregtype) ||
1657            (src1reg == dstreg && src1regtype == dstregtype) ||
1658            (src2reg == dstreg && src2regtype == dstregtype)) {
1659
1660             write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1661             if (!write_mask) continue;
1662             shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1663             temp_destination = TRUE;
1664         } else {
1665             write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1666             if (!write_mask) continue;
1667         }
1668
1669         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1670         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1671         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1672
1673             shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1674                            src0_param.param_str, src1_param.param_str, src2_param.param_str);
1675     }
1676
1677     if(temp_destination) {
1678         shader_glsl_get_write_mask(arg->dst, mask_char);
1679         shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1680         shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1681     }
1682
1683 }
1684
1685 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1686 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1687  * the compare is done per component of src0. */
1688 void shader_glsl_cnd(SHADER_OPCODE_ARG* arg) {
1689     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1690     glsl_src_param_t src0_param;
1691     glsl_src_param_t src1_param;
1692     glsl_src_param_t src2_param;
1693     DWORD write_mask, cmp_channel = 0;
1694     unsigned int i, j;
1695
1696     if (shader->baseShader.hex_version < WINED3DPS_VERSION(1, 4)) {
1697         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1698         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1699         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1700         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1701
1702         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1703         if(arg->opcode_token & WINED3DSI_COISSUE) {
1704             shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1705         } else {
1706             shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1707                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
1708         }
1709         return;
1710     }
1711     /* Cycle through all source0 channels */
1712     for (i=0; i<4; i++) {
1713         write_mask = 0;
1714         /* Find the destination channels which use the current source0 channel */
1715         for (j=0; j<4; j++) {
1716             if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1717                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1718                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1719             }
1720         }
1721         write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1722         if (!write_mask) continue;
1723
1724         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1725         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1726         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1727
1728         shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1729                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1730     }
1731 }
1732
1733 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1734 void shader_glsl_mad(SHADER_OPCODE_ARG* arg) {
1735     glsl_src_param_t src0_param;
1736     glsl_src_param_t src1_param;
1737     glsl_src_param_t src2_param;
1738     DWORD write_mask;
1739
1740     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1741     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1742     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1743     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1744     shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1745             src0_param.param_str, src1_param.param_str, src2_param.param_str);
1746 }
1747
1748 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1749     Vertex shaders to GLSL codes */
1750 void shader_glsl_mnxn(SHADER_OPCODE_ARG* arg) {
1751     int i;
1752     int nComponents = 0;
1753     SHADER_OPCODE_ARG tmpArg;
1754
1755     memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1756
1757     /* Set constants for the temporary argument */
1758     tmpArg.shader      = arg->shader;
1759     tmpArg.buffer      = arg->buffer;
1760     tmpArg.src[0]      = arg->src[0];
1761     tmpArg.src_addr[0] = arg->src_addr[0];
1762     tmpArg.src_addr[1] = arg->src_addr[1];
1763     tmpArg.reg_maps = arg->reg_maps;
1764
1765     switch(arg->opcode->opcode) {
1766         case WINED3DSIO_M4x4:
1767             nComponents = 4;
1768             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1769             break;
1770         case WINED3DSIO_M4x3:
1771             nComponents = 3;
1772             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1773             break;
1774         case WINED3DSIO_M3x4:
1775             nComponents = 4;
1776             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1777             break;
1778         case WINED3DSIO_M3x3:
1779             nComponents = 3;
1780             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1781             break;
1782         case WINED3DSIO_M3x2:
1783             nComponents = 2;
1784             tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1785             break;
1786         default:
1787             break;
1788     }
1789
1790     for (i = 0; i < nComponents; i++) {
1791         tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1792         tmpArg.src[1]      = arg->src[1]+i;
1793         shader_glsl_dot(&tmpArg);
1794     }
1795 }
1796
1797 /**
1798     The LRP instruction performs a component-wise linear interpolation
1799     between the second and third operands using the first operand as the
1800     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
1801     This is equivalent to mix(src2, src1, src0);
1802 */
1803 void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
1804     glsl_src_param_t src0_param;
1805     glsl_src_param_t src1_param;
1806     glsl_src_param_t src2_param;
1807     DWORD write_mask;
1808
1809     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1810
1811     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1812     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1813     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1814
1815     shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1816             src2_param.param_str, src1_param.param_str, src0_param.param_str);
1817 }
1818
1819 /** Process the WINED3DSIO_LIT instruction in GLSL:
1820  * dst.x = dst.w = 1.0
1821  * dst.y = (src0.x > 0) ? src0.x
1822  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1823  *                                        where src.w is clamped at +- 128
1824  */
1825 void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
1826     glsl_src_param_t src0_param;
1827     glsl_src_param_t src1_param;
1828     glsl_src_param_t src3_param;
1829     char dst_mask[6];
1830
1831     shader_glsl_append_dst(arg->buffer, arg);
1832     shader_glsl_get_write_mask(arg->dst, dst_mask);
1833
1834     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1835     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
1836     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
1837
1838     /* The sdk specifies the instruction like this
1839      * dst.x = 1.0;
1840      * if(src.x > 0.0) dst.y = src.x
1841      * else dst.y = 0.0.
1842      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
1843      * else dst.z = 0.0;
1844      * dst.w = 1.0;
1845      *
1846      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
1847      * dst.x = 1.0                                  ... No further explanation needed
1848      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
1849      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
1850      * dst.w = 1.0.                                 ... Nothing fancy.
1851      *
1852      * So we still have one conditional in there. So do this:
1853      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
1854      *
1855      * 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),
1856      * 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.
1857      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
1858      */
1859     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",
1860                    src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
1861 }
1862
1863 /** Process the WINED3DSIO_DST instruction in GLSL:
1864  * dst.x = 1.0
1865  * dst.y = src0.x * src0.y
1866  * dst.z = src0.z
1867  * dst.w = src1.w
1868  */
1869 void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
1870     glsl_src_param_t src0y_param;
1871     glsl_src_param_t src0z_param;
1872     glsl_src_param_t src1y_param;
1873     glsl_src_param_t src1w_param;
1874     char dst_mask[6];
1875
1876     shader_glsl_append_dst(arg->buffer, arg);
1877     shader_glsl_get_write_mask(arg->dst, dst_mask);
1878
1879     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
1880     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
1881     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
1882     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
1883
1884     shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
1885             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
1886 }
1887
1888 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
1889  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
1890  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
1891  *
1892  * dst.x = cos(src0.?)
1893  * dst.y = sin(src0.?)
1894  * dst.z = dst.z
1895  * dst.w = dst.w
1896  */
1897 void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
1898     glsl_src_param_t src0_param;
1899     DWORD write_mask;
1900
1901     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1902     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1903
1904     switch (write_mask) {
1905         case WINED3DSP_WRITEMASK_0:
1906             shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
1907             break;
1908
1909         case WINED3DSP_WRITEMASK_1:
1910             shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
1911             break;
1912
1913         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
1914             shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
1915             break;
1916
1917         default:
1918             ERR("Write mask should be .x, .y or .xy\n");
1919             break;
1920     }
1921 }
1922
1923 /** Process the WINED3DSIO_LOOP instruction in GLSL:
1924  * Start a for() loop where src1.y is the initial value of aL,
1925  *  increment aL by src1.z for a total of src1.x iterations.
1926  *  Need to use a temporary variable for this operation.
1927  */
1928 /* FIXME: I don't think nested loops will work correctly this way. */
1929 void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
1930     glsl_src_param_t src1_param;
1931     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1932     DWORD regtype = shader_get_regtype(arg->src[1]);
1933     DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1934     const DWORD *control_values = NULL;
1935     local_constant *constant;
1936
1937     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
1938
1939     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
1940      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
1941      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
1942      * addressing.
1943      */
1944     if(regtype == WINED3DSPR_CONSTINT) {
1945         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
1946             if(constant->idx == reg) {
1947                 control_values = constant->value;
1948                 break;
1949             }
1950         }
1951     }
1952
1953     if(control_values) {
1954         if(control_values[2] > 0) {
1955             shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
1956                            shader->baseShader.cur_loop_depth, control_values[1],
1957                            shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
1958                            shader->baseShader.cur_loop_depth, control_values[2]);
1959         } else if(control_values[2] == 0) {
1960             shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
1961                            shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
1962                            shader->baseShader.cur_loop_depth, control_values[0],
1963                            shader->baseShader.cur_loop_depth);
1964         } else {
1965             shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
1966                            shader->baseShader.cur_loop_depth, control_values[1],
1967                            shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
1968                            shader->baseShader.cur_loop_depth, control_values[2]);
1969         }
1970     } else {
1971         shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
1972                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
1973                        src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
1974                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
1975     }
1976
1977     shader->baseShader.cur_loop_depth++;
1978     shader->baseShader.cur_loop_regno++;
1979 }
1980
1981 void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
1982     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1983
1984     shader_addline(arg->buffer, "}\n");
1985
1986     if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
1987         shader->baseShader.cur_loop_depth--;
1988         shader->baseShader.cur_loop_regno--;
1989     }
1990     if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
1991         shader->baseShader.cur_loop_depth--;
1992     }
1993 }
1994
1995 void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
1996     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1997     glsl_src_param_t src0_param;
1998
1999     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2000     shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2001                    shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2002                    src0_param.param_str, shader->baseShader.cur_loop_depth);
2003     shader->baseShader.cur_loop_depth++;
2004 }
2005
2006 void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
2007     glsl_src_param_t src0_param;
2008
2009     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2010     shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2011 }
2012
2013 void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
2014     glsl_src_param_t src0_param;
2015     glsl_src_param_t src1_param;
2016
2017     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2018     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2019
2020     shader_addline(arg->buffer, "if (%s %s %s) {\n",
2021             src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2022 }
2023
2024 void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
2025     shader_addline(arg->buffer, "} else {\n");
2026 }
2027
2028 void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
2029     shader_addline(arg->buffer, "break;\n");
2030 }
2031
2032 /* FIXME: According to MSDN the compare is done per component. */
2033 void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
2034     glsl_src_param_t src0_param;
2035     glsl_src_param_t src1_param;
2036
2037     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2038     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2039
2040     shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2041             src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2042 }
2043
2044 void shader_glsl_label(SHADER_OPCODE_ARG* arg) {
2045
2046     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2047     shader_addline(arg->buffer, "}\n");
2048     shader_addline(arg->buffer, "void subroutine%u () {\n",  snum);
2049 }
2050
2051 void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
2052     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2053     shader_addline(arg->buffer, "subroutine%u();\n", snum);
2054 }
2055
2056 void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
2057     glsl_src_param_t src1_param;
2058
2059     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2060     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2061     shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2062 }
2063
2064 /*********************************************
2065  * Pixel Shader Specific Code begins here
2066  ********************************************/
2067 void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
2068     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2069     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2070     DWORD hex_version = This->baseShader.hex_version;
2071     char dst_swizzle[6];
2072     glsl_sample_function_t sample_function;
2073     DWORD sampler_type;
2074     DWORD sampler_idx;
2075     BOOL projected, texrect = FALSE;
2076     DWORD mask = 0;
2077
2078     /* All versions have a destination register */
2079     shader_glsl_append_dst(arg->buffer, arg);
2080
2081     /* 1.0-1.4: Use destination register as sampler source.
2082      * 2.0+: Use provided sampler source. */
2083     if (hex_version < WINED3DPS_VERSION(1,4)) {
2084         DWORD flags;
2085
2086         sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2087         flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2088
2089         if (flags & WINED3DTTFF_PROJECTED) {
2090             projected = TRUE;
2091             switch (flags & ~WINED3DTTFF_PROJECTED) {
2092                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2093                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2094                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2095                 case WINED3DTTFF_COUNT4:
2096                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2097             }
2098         } else {
2099             projected = FALSE;
2100         }
2101     } else if (hex_version < WINED3DPS_VERSION(2,0)) {
2102         DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2103         sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2104
2105         if (src_mod == WINED3DSPSM_DZ) {
2106             projected = TRUE;
2107             mask = WINED3DSP_WRITEMASK_2;
2108         } else if (src_mod == WINED3DSPSM_DW) {
2109             projected = TRUE;
2110             mask = WINED3DSP_WRITEMASK_3;
2111         } else {
2112             projected = FALSE;
2113         }
2114     } else {
2115         sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2116         if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2117                 /* ps 2.0 texldp instruction always divides by the fourth component. */
2118                 projected = TRUE;
2119                 mask = WINED3DSP_WRITEMASK_3;
2120         } else {
2121             projected = FALSE;
2122         }
2123     }
2124
2125     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2126        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2127         texrect = TRUE;
2128     }
2129
2130     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2131     shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2132     mask |= sample_function.coord_mask;
2133
2134     if (hex_version < WINED3DPS_VERSION(2,0)) {
2135         shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2136     } else {
2137         shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2138     }
2139
2140     /* 1.0-1.3: Use destination register as coordinate source.
2141        1.4+: Use provided coordinate source register. */
2142     if (hex_version < WINED3DPS_VERSION(1,4)) {
2143         char coord_mask[6];
2144         shader_glsl_get_write_mask(mask, coord_mask);
2145         shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2146                 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2147     } else {
2148         glsl_src_param_t coord_param;
2149         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2150         if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2151             glsl_src_param_t bias;
2152             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2153
2154             shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2155                     sample_function.name, sampler_idx, coord_param.param_str,
2156                     bias.param_str, dst_swizzle);
2157         } else {
2158             shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2159                     sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2160         }
2161     }
2162 }
2163
2164 void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
2165     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2166     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2167     glsl_sample_function_t sample_function;
2168     glsl_src_param_t coord_param, lod_param;
2169     char dst_swizzle[6];
2170     DWORD sampler_type;
2171     DWORD sampler_idx;
2172     BOOL texrect = FALSE;
2173
2174     shader_glsl_append_dst(arg->buffer, arg);
2175     shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2176
2177     sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2178     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2179     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2180        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2181         texrect = TRUE;
2182     }
2183     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);
2184
2185     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2186
2187     if (shader_is_pshader_version(This->baseShader.hex_version)) {
2188         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2189          * However, they seem to work just fine in fragment shaders as well. */
2190         WARN("Using %sLod in fragment shader.\n", sample_function.name);
2191         shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2192                 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2193     } else {
2194         shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2195                 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2196     }
2197 }
2198
2199 void pshader_glsl_texcoord(SHADER_OPCODE_ARG* arg) {
2200
2201     /* FIXME: Make this work for more than just 2D textures */
2202
2203     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2204     SHADER_BUFFER* buffer = arg->buffer;
2205     DWORD hex_version = This->baseShader.hex_version;
2206     DWORD write_mask;
2207     char dst_mask[6];
2208
2209     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2210     shader_glsl_get_write_mask(write_mask, dst_mask);
2211
2212     if (hex_version != WINED3DPS_VERSION(1,4)) {
2213         DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2214         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2215     } else {
2216         DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2217         DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2218         char dst_swizzle[6];
2219
2220         shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2221
2222         if (src_mod == WINED3DSPSM_DZ) {
2223             glsl_src_param_t div_param;
2224             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2225             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2226
2227             if (mask_size > 1) {
2228                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2229             } else {
2230                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2231             }
2232         } else if (src_mod == WINED3DSPSM_DW) {
2233             glsl_src_param_t div_param;
2234             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2235             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2236
2237             if (mask_size > 1) {
2238                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2239             } else {
2240                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2241             }
2242         } else {
2243             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2244         }
2245     }
2246 }
2247
2248 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2249  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2250  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2251 void pshader_glsl_texdp3tex(SHADER_OPCODE_ARG* arg) {
2252     glsl_src_param_t src0_param;
2253     char dst_mask[6];
2254     glsl_sample_function_t sample_function;
2255     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2256     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2257     DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2258
2259     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2260
2261     shader_glsl_append_dst(arg->buffer, arg);
2262     shader_glsl_get_write_mask(arg->dst, dst_mask);
2263
2264     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2265      * scalar, and projected sampling would require 4.
2266      *
2267      * It is a dependent read - not valid with conditional NP2 textures
2268      */
2269     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2270
2271     switch(count_bits(sample_function.coord_mask)) {
2272         case 1:
2273             shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2274                            sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2275             break;
2276
2277         case 2:
2278             shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2279                           sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2280             break;
2281
2282         case 3:
2283             shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2284                            sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2285             break;
2286         default:
2287             FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2288     }
2289 }
2290
2291 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2292  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2293 void pshader_glsl_texdp3(SHADER_OPCODE_ARG* arg) {
2294     glsl_src_param_t src0_param;
2295     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2296     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2297     DWORD dst_mask;
2298     unsigned int mask_size;
2299
2300     dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2301     mask_size = shader_glsl_get_write_mask_size(dst_mask);
2302     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2303
2304     if (mask_size > 1) {
2305         shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2306     } else {
2307         shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2308     }
2309 }
2310
2311 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2312  * Calculate the depth as dst.x / dst.y   */
2313 void pshader_glsl_texdepth(SHADER_OPCODE_ARG* arg) {
2314     glsl_dst_param_t dst_param;
2315
2316     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2317
2318     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2319      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2320      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2321      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2322      * >= 1.0 or < 0.0
2323      */
2324     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);
2325 }
2326
2327 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2328  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2329  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2330  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2331  */
2332 void pshader_glsl_texm3x2depth(SHADER_OPCODE_ARG* arg) {
2333     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2334     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2335     glsl_src_param_t src0_param;
2336
2337     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2338
2339     shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2340     shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2341 }
2342
2343 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2344  * Calculate the 1st of a 2-row matrix multiplication. */
2345 void pshader_glsl_texm3x2pad(SHADER_OPCODE_ARG* arg) {
2346     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2347     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2348     SHADER_BUFFER* buffer = arg->buffer;
2349     glsl_src_param_t src0_param;
2350
2351     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2352     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2353 }
2354
2355 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2356  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2357 void pshader_glsl_texm3x3pad(SHADER_OPCODE_ARG* arg) {
2358
2359     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2360     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2361     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2362     SHADER_BUFFER* buffer = arg->buffer;
2363     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2364     glsl_src_param_t src0_param;
2365
2366     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2367     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2368     current_state->texcoord_w[current_state->current_row++] = reg;
2369 }
2370
2371 void pshader_glsl_texm3x2tex(SHADER_OPCODE_ARG* arg) {
2372     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2373     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2374     SHADER_BUFFER* buffer = arg->buffer;
2375     glsl_src_param_t src0_param;
2376     char dst_mask[6];
2377
2378     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2379     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2380
2381     shader_glsl_append_dst(buffer, arg);
2382     shader_glsl_get_write_mask(arg->dst, dst_mask);
2383
2384     /* Sample the texture using the calculated coordinates */
2385     shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2386 }
2387
2388 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2389  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2390 void pshader_glsl_texm3x3tex(SHADER_OPCODE_ARG* arg) {
2391     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2392     glsl_src_param_t src0_param;
2393     char dst_mask[6];
2394     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2395     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2396     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2397     DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2398     glsl_sample_function_t sample_function;
2399
2400     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2401     shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2402
2403     shader_glsl_append_dst(arg->buffer, arg);
2404     shader_glsl_get_write_mask(arg->dst, dst_mask);
2405     /* Dependent read, not valid with conditional NP2 */
2406     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2407
2408     /* Sample the texture using the calculated coordinates */
2409     shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2410
2411     current_state->current_row = 0;
2412 }
2413
2414 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2415  * Perform the 3rd row of a 3x3 matrix multiply */
2416 void pshader_glsl_texm3x3(SHADER_OPCODE_ARG* arg) {
2417     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2418     glsl_src_param_t src0_param;
2419     char dst_mask[6];
2420     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2421     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2422     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2423
2424     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2425
2426     shader_glsl_append_dst(arg->buffer, arg);
2427     shader_glsl_get_write_mask(arg->dst, dst_mask);
2428     shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2429
2430     current_state->current_row = 0;
2431 }
2432
2433 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2434  * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2435 void pshader_glsl_texm3x3spec(SHADER_OPCODE_ARG* arg) {
2436
2437     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2438     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2439     glsl_src_param_t src0_param;
2440     glsl_src_param_t src1_param;
2441     char dst_mask[6];
2442     SHADER_BUFFER* buffer = arg->buffer;
2443     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2444     DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2445     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2446     glsl_sample_function_t sample_function;
2447
2448     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2449     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2450
2451     /* Perform the last matrix multiply operation */
2452     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2453     /* Reflection calculation */
2454     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2455
2456     shader_glsl_append_dst(buffer, arg);
2457     shader_glsl_get_write_mask(arg->dst, dst_mask);
2458     /* Dependent read, not valid with conditional NP2 */
2459     shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2460
2461     /* Sample the texture */
2462     shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2463
2464     current_state->current_row = 0;
2465 }
2466
2467 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2468  * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2469 void pshader_glsl_texm3x3vspec(SHADER_OPCODE_ARG* arg) {
2470
2471     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2472     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2473     SHADER_BUFFER* buffer = arg->buffer;
2474     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2475     glsl_src_param_t src0_param;
2476     char dst_mask[6];
2477     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2478     DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2479     glsl_sample_function_t sample_function;
2480
2481     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2482
2483     /* Perform the last matrix multiply operation */
2484     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2485
2486     /* Construct the eye-ray vector from w coordinates */
2487     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2488             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2489     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2490
2491     shader_glsl_append_dst(buffer, arg);
2492     shader_glsl_get_write_mask(arg->dst, dst_mask);
2493     /* Dependent read, not valid with conditional NP2 */
2494     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2495
2496     /* Sample the texture using the calculated coordinates */
2497     shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2498
2499     current_state->current_row = 0;
2500 }
2501
2502 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2503  * Apply a fake bump map transform.
2504  * texbem is pshader <= 1.3 only, this saves a few version checks
2505  */
2506 void pshader_glsl_texbem(SHADER_OPCODE_ARG* arg) {
2507     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2508     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2509     char dst_swizzle[6];
2510     glsl_sample_function_t sample_function;
2511     glsl_src_param_t coord_param;
2512     DWORD sampler_type;
2513     DWORD sampler_idx;
2514     DWORD mask;
2515     DWORD flags;
2516     char coord_mask[6];
2517
2518     sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2519     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2520
2521     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2522     /* Dependent read, not valid with conditional NP2 */
2523     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2524     mask = sample_function.coord_mask;
2525
2526     shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2527
2528     shader_glsl_get_write_mask(mask, coord_mask);
2529
2530     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2531          * so we can't let the GL handle this.
2532          */
2533     if (flags & WINED3DTTFF_PROJECTED) {
2534         DWORD div_mask=0;
2535         char coord_div_mask[3];
2536         switch (flags & ~WINED3DTTFF_PROJECTED) {
2537             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2538             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2539             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2540             case WINED3DTTFF_COUNT4:
2541             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2542         }
2543         shader_glsl_get_write_mask(div_mask, coord_div_mask);
2544         shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2545     }
2546
2547     shader_glsl_append_dst(arg->buffer, arg);
2548     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2549     if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2550         glsl_src_param_t luminance_param;
2551         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2552         shader_addline(arg->buffer, "(%s(Psampler%u, T%u%s + vec4(bumpenvmat * %s, 0.0, 0.0)%s )*(%s * luminancescale + luminanceoffset))%s);\n",
2553                        sample_function.name, sampler_idx, sampler_idx, coord_mask, coord_param.param_str, coord_mask,
2554                        luminance_param.param_str, dst_swizzle);
2555     } else {
2556         shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat * %s, 0.0, 0.0)%s )%s);\n",
2557                        sample_function.name, sampler_idx, sampler_idx, coord_mask, coord_param.param_str, coord_mask, dst_swizzle);
2558     }
2559 }
2560
2561 void pshader_glsl_bem(SHADER_OPCODE_ARG* arg) {
2562     glsl_src_param_t src0_param, src1_param;
2563
2564     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2565     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2566
2567     shader_glsl_append_dst(arg->buffer, arg);
2568     shader_addline(arg->buffer, "%s + bumpenvmat * %s);\n",
2569                    src0_param.param_str, src1_param.param_str);
2570 }
2571
2572 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2573  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2574 void pshader_glsl_texreg2ar(SHADER_OPCODE_ARG* arg) {
2575
2576     glsl_src_param_t src0_param;
2577     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2578     char dst_mask[6];
2579
2580     shader_glsl_append_dst(arg->buffer, arg);
2581     shader_glsl_get_write_mask(arg->dst, dst_mask);
2582     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2583
2584     shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2585 }
2586
2587 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2588  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2589 void pshader_glsl_texreg2gb(SHADER_OPCODE_ARG* arg) {
2590     glsl_src_param_t src0_param;
2591     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2592     char dst_mask[6];
2593
2594     shader_glsl_append_dst(arg->buffer, arg);
2595     shader_glsl_get_write_mask(arg->dst, dst_mask);
2596     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2597
2598     shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2599 }
2600
2601 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2602  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2603 void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
2604     glsl_src_param_t src0_param;
2605     char dst_mask[6];
2606     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2607     DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2608     glsl_sample_function_t sample_function;
2609
2610     shader_glsl_append_dst(arg->buffer, arg);
2611     shader_glsl_get_write_mask(arg->dst, dst_mask);
2612     /* Dependent read, not valid with conditional NP2 */
2613     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2614     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2615
2616     shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2617 }
2618
2619 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2620  * If any of the first 3 components are < 0, discard this pixel */
2621 void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
2622     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2623     DWORD hex_version = This->baseShader.hex_version;
2624     glsl_dst_param_t dst_param;
2625
2626     /* The argument is a destination parameter, and no writemasks are allowed */
2627     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2628     if((hex_version >= WINED3DPS_VERSION(2,0))) {
2629         /* 2.0 shaders compare all 4 components in texkill */
2630         shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2631     } else {
2632         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2633          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2634          * 4 components are defined, only the first 3 are used
2635          */
2636         shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2637     }
2638 }
2639
2640 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2641  * dst = dot2(src0, src1) + src2 */
2642 void pshader_glsl_dp2add(SHADER_OPCODE_ARG* arg) {
2643     glsl_src_param_t src0_param;
2644     glsl_src_param_t src1_param;
2645     glsl_src_param_t src2_param;
2646     DWORD write_mask;
2647     unsigned int mask_size;
2648
2649     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2650     mask_size = shader_glsl_get_write_mask_size(write_mask);
2651
2652     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2653     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2654     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2655
2656     shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2657 }
2658
2659 void pshader_glsl_input_pack(
2660    SHADER_BUFFER* buffer,
2661    semantic* semantics_in,
2662    IWineD3DPixelShader *iface) {
2663
2664    unsigned int i;
2665    IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2666
2667    for (i = 0; i < MAX_REG_INPUT; i++) {
2668
2669        DWORD usage_token = semantics_in[i].usage;
2670        DWORD register_token = semantics_in[i].reg;
2671        DWORD usage, usage_idx;
2672        char reg_mask[6];
2673
2674        /* Uninitialized */
2675        if (!usage_token) continue;
2676        usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2677        usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2678        shader_glsl_get_write_mask(register_token, reg_mask);
2679
2680        switch(usage) {
2681
2682            case WINED3DDECLUSAGE_TEXCOORD:
2683                if(usage_idx < 8 && This->vertexprocessing == pretransformed) {
2684                    shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2685                                   This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2686                } else {
2687                    shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2688                                   This->input_reg_map[i], reg_mask, reg_mask);
2689                }
2690                break;
2691
2692            case WINED3DDECLUSAGE_COLOR:
2693                if (usage_idx == 0)
2694                    shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2695                        This->input_reg_map[i], reg_mask, reg_mask);
2696                else if (usage_idx == 1)
2697                    shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2698                        This->input_reg_map[i], reg_mask, reg_mask);
2699                else
2700                    shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2701                        This->input_reg_map[i], reg_mask, reg_mask);
2702                break;
2703
2704            default:
2705                shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2706                    This->input_reg_map[i], reg_mask, reg_mask);
2707         }
2708     }
2709 }
2710
2711 /*********************************************
2712  * Vertex Shader Specific Code begins here
2713  ********************************************/
2714
2715 static void add_glsl_program_entry(IWineD3DDeviceImpl *device, struct glsl_shader_prog_link *entry) {
2716     glsl_program_key_t *key;
2717
2718     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2719     key->vshader = entry->vshader;
2720     key->pshader = entry->pshader;
2721
2722     hash_table_put(device->glsl_program_lookup, key, entry);
2723 }
2724
2725 static struct glsl_shader_prog_link *get_glsl_program_entry(IWineD3DDeviceImpl *device,
2726         GLhandleARB vshader, GLhandleARB pshader) {
2727     glsl_program_key_t key;
2728
2729     key.vshader = vshader;
2730     key.pshader = pshader;
2731
2732     return (struct glsl_shader_prog_link *)hash_table_get(device->glsl_program_lookup, &key);
2733 }
2734
2735 void delete_glsl_program_entry(IWineD3DDevice *iface, struct glsl_shader_prog_link *entry) {
2736     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
2737     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
2738     glsl_program_key_t *key;
2739
2740     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2741     key->vshader = entry->vshader;
2742     key->pshader = entry->pshader;
2743     hash_table_remove(This->glsl_program_lookup, key);
2744
2745     GL_EXTCALL(glDeleteObjectARB(entry->programId));
2746     if (entry->vshader) list_remove(&entry->vshader_entry);
2747     if (entry->pshader) list_remove(&entry->pshader_entry);
2748     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2749     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2750     HeapFree(GetProcessHeap(), 0, entry);
2751 }
2752
2753 static void handle_ps3_input(SHADER_BUFFER *buffer, semantic *semantics_in, semantic *semantics_out, WineD3D_GL_Info *gl_info, DWORD *map) {
2754     unsigned int i, j;
2755     DWORD usage_token, usage_token_out;
2756     DWORD register_token, register_token_out;
2757     DWORD usage, usage_idx, usage_out, usage_idx_out;
2758     DWORD *set;
2759     char reg_mask[6], reg_mask_out[6];
2760
2761     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (GL_LIMITS(glsl_varyings) / 4));
2762
2763     for(i = 0; i < MAX_REG_INPUT; i++) {
2764         usage_token = semantics_in[i].usage;
2765         if (!usage_token) continue;
2766         if(map[i] >= (GL_LIMITS(glsl_varyings) / 4)) {
2767             FIXME("More input varyings declared than supported, expect issues\n");
2768             continue;
2769         } else if(map[i] == -1) {
2770             /* Declared, but not read register */
2771             continue;
2772         }
2773         register_token = semantics_in[i].reg;
2774
2775         usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2776         usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2777         set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
2778
2779         if(!semantics_out) {
2780             switch(usage) {
2781                 case WINED3DDECLUSAGE_COLOR:
2782                     if (usage_idx == 0)
2783                         shader_addline(buffer, "IN[%u]%s = gl_FrontColor%s;\n",
2784                                        map[i], reg_mask, reg_mask);
2785                     else if (usage_idx == 1)
2786                         shader_addline(buffer, "IN[%u]%s = gl_FrontSecondaryColor%s;\n",
2787                                        map[i], reg_mask, reg_mask);
2788                     else
2789                         shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2790                                        map[i], reg_mask, reg_mask);
2791                     break;
2792
2793                 case WINED3DDECLUSAGE_TEXCOORD:
2794                     if (usage_idx < 8) {
2795                         shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2796                                        map[i], reg_mask, usage_idx, reg_mask);
2797                     } else {
2798                         shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2799                                        map[i], reg_mask, reg_mask);
2800                     }
2801                     break;
2802
2803                 case WINED3DDECLUSAGE_FOG:
2804                     shader_addline(buffer, "IN[%u]%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
2805                                    map[i], reg_mask, reg_mask);
2806                     break;
2807
2808                 default:
2809                     shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2810                                    map[i], reg_mask, reg_mask);
2811             }
2812         } else {
2813             BOOL found = FALSE;
2814             for(j = 0; j < MAX_REG_OUTPUT; j++) {
2815                 usage_token_out = semantics_out[j].usage;
2816                 if (!usage_token_out) continue;
2817                 register_token_out = semantics_out[j].reg;
2818
2819                 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2820                 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2821                 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
2822
2823                 if(usage == usage_out &&
2824                    usage_idx == usage_idx_out) {
2825                     shader_addline(buffer, "IN[%u]%s = OUT[%u]%s;\n",
2826                                    map[i], reg_mask, j, reg_mask);
2827                     found = TRUE;
2828                 }
2829             }
2830             if(!found) {
2831                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2832                                map[i], reg_mask, reg_mask);
2833             }
2834         }
2835     }
2836
2837     /* This is solely to make the compiler / linker happy and avoid warning about undefined
2838      * varyings. It shouldn't result in any real code executed on the GPU, since all read
2839      * input varyings are assigned above, if the optimizer works properly.
2840      */
2841     for(i = 0; i < GL_LIMITS(glsl_varyings) / 4; i++) {
2842         if(set[i] != WINED3DSP_WRITEMASK_ALL) {
2843             unsigned int size = 0;
2844             memset(reg_mask, 0, sizeof(reg_mask));
2845             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
2846                 reg_mask[size] = 'x';
2847                 size++;
2848             }
2849             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
2850                 reg_mask[size] = 'y';
2851                 size++;
2852             }
2853             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
2854                 reg_mask[size] = 'z';
2855                 size++;
2856             }
2857             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
2858                 reg_mask[size] = 'w';
2859                 size++;
2860             }
2861             switch(size) {
2862                 case 1:
2863                     shader_addline(buffer, "IN[%u].%s = 0.0;\n", i, reg_mask);
2864                     break;
2865                 case 2:
2866                     shader_addline(buffer, "IN[%u].%s = vec2(0.0, 0.0);\n", i, reg_mask);
2867                     break;
2868                 case 3:
2869                     shader_addline(buffer, "IN[%u].%s = vec3(0.0, 0.0, 0.0);\n", i, reg_mask);
2870                     break;
2871                 case 4:
2872                     shader_addline(buffer, "IN[%u].%s = vec4(0.0, 0.0, 0.0, 0.0);\n", i, reg_mask);
2873                     break;
2874             }
2875         }
2876     }
2877
2878     HeapFree(GetProcessHeap(), 0, set);
2879 }
2880
2881 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
2882         IWineD3DPixelShader *pixelshader,
2883         WineD3D_GL_Info *gl_info) {
2884     GLhandleARB ret = 0;
2885     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
2886     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
2887     DWORD vs_major = vs ? WINED3DSHADER_VERSION_MAJOR(vs->baseShader.hex_version) : 0;
2888     DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.hex_version) : 0;
2889     unsigned int i;
2890     SHADER_BUFFER buffer;
2891     DWORD usage_token;
2892     DWORD register_token;
2893     DWORD usage, usage_idx, writemask;
2894     char reg_mask[6];
2895     semantic *semantics_out, *semantics_in;
2896
2897     buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
2898     buffer.bsize = 0;
2899     buffer.lineNo = 0;
2900     buffer.newline = TRUE;
2901
2902     if(vs_major < 3 && ps_major < 3) {
2903         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
2904          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
2905          */
2906         if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0) {
2907             shader_addline(&buffer, "void order_ps_input() {\n");
2908             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
2909                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
2910                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
2911                     shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
2912                 }
2913             }
2914             shader_addline(&buffer, "}\n");
2915         } else {
2916             shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
2917         }
2918     } else if(ps_major < 3 && vs_major >= 3) {
2919         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
2920         semantics_out = vs->semantics_out;
2921
2922         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
2923         for(i = 0; i < MAX_REG_OUTPUT; i++) {
2924             usage_token = semantics_out[i].usage;
2925             if (!usage_token) continue;
2926             register_token = semantics_out[i].reg;
2927
2928             usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2929             usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2930             writemask = shader_glsl_get_write_mask(register_token, reg_mask);
2931
2932             switch(usage) {
2933                 case WINED3DDECLUSAGE_COLOR:
2934                     if (usage_idx == 0)
2935                         shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2936                     else if (usage_idx == 1)
2937                         shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2938                     break;
2939
2940                 case WINED3DDECLUSAGE_POSITION:
2941                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2942                     break;
2943
2944                 case WINED3DDECLUSAGE_TEXCOORD:
2945                     if (usage_idx < 8) {
2946                         if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
2947
2948                         shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
2949                                         usage_idx, reg_mask, i, reg_mask);
2950                         if(!(writemask & WINED3DSP_WRITEMASK_3)) {
2951                             shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
2952                         }
2953                     }
2954                     break;
2955
2956                 case WINED3DDECLUSAGE_PSIZE:
2957                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
2958                     break;
2959
2960                 case WINED3DDECLUSAGE_FOG:
2961                     shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
2962                     break;
2963
2964                 default:
2965                     break;
2966             }
2967         }
2968         shader_addline(&buffer, "}\n");
2969
2970     } else if(ps_major >= 3 && vs_major >= 3) {
2971         semantics_out = vs->semantics_out;
2972         semantics_in = ps->semantics_in;
2973
2974         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
2975         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
2976         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
2977
2978         /* First, sort out position and point size. Those are not passed to the pixel shader */
2979         for(i = 0; i < MAX_REG_OUTPUT; i++) {
2980             usage_token = semantics_out[i].usage;
2981             if (!usage_token) continue;
2982             register_token = semantics_out[i].reg;
2983
2984             usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2985             usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2986             shader_glsl_get_write_mask(register_token, reg_mask);
2987
2988             switch(usage) {
2989                 case WINED3DDECLUSAGE_POSITION:
2990                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2991                     break;
2992
2993                 case WINED3DDECLUSAGE_PSIZE:
2994                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
2995                     break;
2996
2997                 default:
2998                     break;
2999             }
3000         }
3001
3002         /* Then, fix the pixel shader input */
3003         handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3004
3005         shader_addline(&buffer, "}\n");
3006     } else if(ps_major >= 3 && vs_major < 3) {
3007         semantics_in = ps->semantics_in;
3008
3009         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3010         shader_addline(&buffer, "void order_ps_input() {\n");
3011         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3012          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3013          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3014          */
3015         handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3016         shader_addline(&buffer, "}\n");
3017     } else {
3018         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3019     }
3020
3021     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3022     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3023     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3024     checkGLcall("glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL)");
3025     GL_EXTCALL(glCompileShaderARB(ret));
3026     checkGLcall("glCompileShaderARB(ret)");
3027
3028     HeapFree(GetProcessHeap(), 0, buffer.buffer);
3029     return ret;
3030 }
3031
3032 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3033  * It sets the programId on the current StateBlock (because it should be called
3034  * inside of the DrawPrimitive() part of the render loop).
3035  *
3036  * If a program for the given combination does not exist, create one, and store
3037  * the program in the hash table.  If it creates a program, it will link the
3038  * given objects, too.
3039  */
3040 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3041     IWineD3DDeviceImpl *This               = (IWineD3DDeviceImpl *)iface;
3042     WineD3D_GL_Info *gl_info               = &This->adapter->gl_info;
3043     IWineD3DPixelShader  *pshader          = This->stateBlock->pixelShader;
3044     IWineD3DVertexShader *vshader          = This->stateBlock->vertexShader;
3045     struct glsl_shader_prog_link *entry    = NULL;
3046     GLhandleARB programId                  = 0;
3047     GLhandleARB reorder_shader_id          = 0;
3048     int i;
3049     char glsl_name[8];
3050
3051     GLhandleARB vshader_id = use_vs ? ((IWineD3DBaseShaderImpl*)vshader)->baseShader.prgId : 0;
3052     GLhandleARB pshader_id = use_ps ? ((IWineD3DBaseShaderImpl*)pshader)->baseShader.prgId : 0;
3053     entry = get_glsl_program_entry(This, vshader_id, pshader_id);
3054     if (entry) {
3055         This->stateBlock->glsl_program = entry;
3056         return;
3057     }
3058
3059     /* If we get to this point, then no matching program exists, so we create one */
3060     programId = GL_EXTCALL(glCreateProgramObjectARB());
3061     TRACE("Created new GLSL shader program %u\n", programId);
3062
3063     /* Create the entry */
3064     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3065     entry->programId = programId;
3066     entry->vshader = vshader_id;
3067     entry->pshader = pshader_id;
3068     /* Add the hash table entry */
3069     add_glsl_program_entry(This, entry);
3070
3071     /* Set the current program */
3072     This->stateBlock->glsl_program = entry;
3073
3074     /* Attach GLSL vshader */
3075     if (vshader_id) {
3076         int max_attribs = 16;   /* TODO: Will this always be the case? It is at the moment... */
3077         char tmp_name[10];
3078
3079         reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3080         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3081         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3082         checkGLcall("glAttachObjectARB");
3083         /* Flag the reorder function for deletion, then it will be freed automatically when the program
3084          * is destroyed
3085          */
3086         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3087
3088         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3089         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3090         checkGLcall("glAttachObjectARB");
3091
3092         /* Bind vertex attributes to a corresponding index number to match
3093          * the same index numbers as ARB_vertex_programs (makes loading
3094          * vertex attributes simpler).  With this method, we can use the
3095          * exact same code to load the attributes later for both ARB and
3096          * GLSL shaders.
3097          *
3098          * We have to do this here because we need to know the Program ID
3099          * in order to make the bindings work, and it has to be done prior
3100          * to linking the GLSL program. */
3101         for (i = 0; i < max_attribs; ++i) {
3102             if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3103                 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3104                 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3105             }
3106         }
3107         checkGLcall("glBindAttribLocationARB");
3108
3109         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3110     }
3111
3112     /* Attach GLSL pshader */
3113     if (pshader_id) {
3114         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3115         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3116         checkGLcall("glAttachObjectARB");
3117
3118         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3119     }
3120
3121     /* Link the program */
3122     TRACE("Linking GLSL shader program %u\n", programId);
3123     GL_EXTCALL(glLinkProgramARB(programId));
3124     print_glsl_info_log(&GLINFO_LOCATION, programId);
3125
3126     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3127     for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3128         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3129         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3130     }
3131     for (i = 0; i < MAX_CONST_I; ++i) {
3132         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3133         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3134     }
3135     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3136     for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3137         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3138         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3139     }
3140     for (i = 0; i < MAX_CONST_I; ++i) {
3141         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3142         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3143     }
3144
3145     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3146     entry->bumpenvmat_location = GL_EXTCALL(glGetUniformLocationARB(programId, "bumpenvmat"));
3147     entry->luminancescale_location = GL_EXTCALL(glGetUniformLocationARB(programId, "luminancescale"));
3148     entry->luminanceoffset_location = GL_EXTCALL(glGetUniformLocationARB(programId, "luminanceoffset"));
3149     entry->srgb_comparison_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_comparison"));
3150     entry->srgb_mul_low_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_mul_low"));
3151     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3152     checkGLcall("Find glsl program uniform locations");
3153
3154     /* Set the shader to allow uniform loading on it */
3155     GL_EXTCALL(glUseProgramObjectARB(programId));
3156     checkGLcall("glUseProgramObjectARB(programId)");
3157
3158     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3159      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3160      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3161      * vertex shader with fixed function pixel processing is used we make sure that the card
3162      * supports enough samplers to allow the max number of vertex samplers with all possible
3163      * fixed function fragment processing setups. So once the program is linked these samplers
3164      * won't change.
3165      */
3166     if(vshader_id) {
3167         /* Load vertex shader samplers */
3168         shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3169     }
3170     if(pshader_id) {
3171         /* Load pixel shader samplers */
3172         shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3173     }
3174 }
3175
3176 static GLhandleARB create_glsl_blt_shader(WineD3D_GL_Info *gl_info) {
3177     GLhandleARB program_id;
3178     GLhandleARB vshader_id, pshader_id;
3179     const char *blt_vshader[] = {
3180         "void main(void)\n"
3181         "{\n"
3182         "    gl_Position = gl_Vertex;\n"
3183         "    gl_FrontColor = vec4(1.0);\n"
3184         "    gl_TexCoord[0].x = (gl_Vertex.x * 0.5) + 0.5;\n"
3185         "    gl_TexCoord[0].y = (-gl_Vertex.y * 0.5) + 0.5;\n"
3186         "}\n"
3187     };
3188
3189     const char *blt_pshader[] = {
3190         "uniform sampler2D sampler;\n"
3191         "void main(void)\n"
3192         "{\n"
3193         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3194         "}\n"
3195     };
3196
3197     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3198     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3199     GL_EXTCALL(glCompileShaderARB(vshader_id));
3200
3201     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3202     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, blt_pshader, NULL));
3203     GL_EXTCALL(glCompileShaderARB(pshader_id));
3204
3205     program_id = GL_EXTCALL(glCreateProgramObjectARB());
3206     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3207     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3208     GL_EXTCALL(glLinkProgramARB(program_id));
3209
3210     print_glsl_info_log(&GLINFO_LOCATION, program_id);
3211
3212     return program_id;
3213 }
3214
3215 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3216     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3217     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3218     GLhandleARB program_id = 0;
3219
3220     if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3221     else This->stateBlock->glsl_program = NULL;
3222
3223     program_id = This->stateBlock->glsl_program ? This->stateBlock->glsl_program->programId : 0;
3224     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3225     GL_EXTCALL(glUseProgramObjectARB(program_id));
3226     checkGLcall("glUseProgramObjectARB");
3227 }
3228
3229 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface) {
3230     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3231     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3232     static GLhandleARB program_id = 0;
3233     static GLhandleARB loc = -1;
3234
3235     if (!program_id) {
3236         program_id = create_glsl_blt_shader(gl_info);
3237         loc = GL_EXTCALL(glGetUniformLocationARB(program_id, "sampler"));
3238     }
3239
3240     GL_EXTCALL(glUseProgramObjectARB(program_id));
3241     GL_EXTCALL(glUniform1iARB(loc, 0));
3242 }
3243
3244 static void shader_glsl_cleanup(IWineD3DDevice *iface) {
3245     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3246     WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3247     GL_EXTCALL(glUseProgramObjectARB(0));
3248 }
3249
3250 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3251     struct list *linked_programs;
3252     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3253     WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *) This->baseShader.device)->adapter->gl_info;
3254
3255     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3256      * can be called from IWineD3DBaseShader::Release
3257      */
3258     char pshader = shader_is_pshader_version(This->baseShader.hex_version);
3259
3260     if(This->baseShader.prgId == 0) return;
3261     linked_programs = &This->baseShader.linked_programs;
3262
3263     TRACE("Deleting linked programs\n");
3264     if (linked_programs->next) {
3265         struct glsl_shader_prog_link *entry, *entry2;
3266
3267         if(pshader) {
3268             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3269                 delete_glsl_program_entry(This->baseShader.device, entry);
3270             }
3271         } else {
3272             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3273                 delete_glsl_program_entry(This->baseShader.device, entry);
3274             }
3275         }
3276     }
3277
3278     TRACE("Deleting shader object %u\n", This->baseShader.prgId);
3279     GL_EXTCALL(glDeleteObjectARB(This->baseShader.prgId));
3280     checkGLcall("glDeleteObjectARB");
3281 }
3282
3283 const shader_backend_t glsl_shader_backend = {
3284     &shader_glsl_select,
3285     &shader_glsl_select_depth_blt,
3286     &shader_glsl_load_constants,
3287     &shader_glsl_cleanup,
3288     &shader_glsl_color_correction,
3289     &shader_glsl_destroy
3290 };