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

   1 /*
   2  * GLSL pixel and vertex shader implementation
   3  *
   4  * Copyright 2006 Jason Green
   5  * Copyright 2006-2007 Henri Verbeet
   6  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
   7  *
   8  * This library is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * This library is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with this library; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  21  */
  22
  23 /*
  24  * D3D shader asm has swizzles on source parameters, and write masks for
  25  * destination parameters. GLSL uses swizzles for both. The result of this is
  26  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  27  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  28  * mask for the destination parameter into account.
  29  */
  30
  31 #include "config.h"
  32 #include <limits.h>
  33 #include <stdio.h>
  34 #include "wined3d_private.h"
  35
  36 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  37 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  40
  41 #define GLINFO_LOCATION      (*gl_info)
  42
  43 typedef struct {
  44     char reg_name[150];
  45     char mask_str[6];
  46 } glsl_dst_param_t;
  47
  48 typedef struct {
  49     char reg_name[150];
  50     char param_str[100];
  51 } glsl_src_param_t;
  52
  53 typedef struct {
  54     const char *name;
  55     DWORD coord_mask;
  56 } glsl_sample_function_t;
  57
  58 enum heap_node_op
  59 {
  60     HEAP_NODE_TRAVERSE_LEFT,
  61     HEAP_NODE_TRAVERSE_RIGHT,
  62     HEAP_NODE_POP,
  63 };
  64
  65 struct constant_entry
  66 {
  67     unsigned int idx;
  68     unsigned int version;
  69 };
  70
  71 struct constant_heap
  72 {
  73     struct constant_entry *entries;
  74     unsigned int *positions;
  75     unsigned int size;
  76 };
  77
  78 /* GLSL shader private data */
  79 struct shader_glsl_priv {
  80     struct hash_table_t *glsl_program_lookup;
  81     struct glsl_shader_prog_link *glsl_program;
  82     struct constant_heap vconst_heap;
  83     struct constant_heap pconst_heap;
  84     unsigned char *stack;
  85     GLhandleARB depth_blt_program[tex_type_count];
  86     UINT next_constant_version;
  87 };
  88
  89 /* Struct to maintain data about a linked GLSL program */
  90 struct glsl_shader_prog_link {
  91     struct list                 vshader_entry;
  92     struct list                 pshader_entry;
  93     GLhandleARB                 programId;
  94     GLhandleARB                 *vuniformF_locations;
  95     GLhandleARB                 *puniformF_locations;
  96     GLhandleARB                 vuniformI_locations[MAX_CONST_I];
  97     GLhandleARB                 puniformI_locations[MAX_CONST_I];
  98     GLhandleARB                 posFixup_location;
  99     GLhandleARB                 bumpenvmat_location[MAX_TEXTURES];
 100     GLhandleARB                 luminancescale_location[MAX_TEXTURES];
 101     GLhandleARB                 luminanceoffset_location[MAX_TEXTURES];
 102     GLhandleARB                 ycorrection_location;
 103     GLenum                      vertex_color_clamp;
 104     IWineD3DVertexShader        *vshader;
 105     IWineD3DPixelShader         *pshader;
 106     struct vs_compile_args      vs_args;
 107     struct ps_compile_args      ps_args;
 108     UINT                        constant_version;
 109 };
 110
 111 typedef struct {
 112     IWineD3DVertexShader        *vshader;
 113     IWineD3DPixelShader         *pshader;
 114     struct ps_compile_args      ps_args;
 115     struct vs_compile_args      vs_args;
 116 } glsl_program_key_t;
 117
 118
 119 /** Prints the GLSL info log which will contain error messages if they exist */
 120 static void print_glsl_info_log(const WineD3D_GL_Info *gl_info, GLhandleARB obj)
 121 {
 122     int infologLength = 0;
 123     char *infoLog;
 124     unsigned int i;
 125     BOOL is_spam;
 126
 127     static const char * const spam[] =
 128     {
 129         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 130         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx          */
 131         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 132         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 133         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 134         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 135         "Fragment shader was successfully compiled to run on hardware.\n"
 136         "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported",
 137         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 138         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 139         "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported\n"  /* MacOS ati      */
 140     };
 141
 142     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 143
 144     GL_EXTCALL(glGetObjectParameterivARB(obj,
 145                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 146                &infologLength));
 147
 148     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 149      * that if there are errors. */
 150     if (infologLength > 1)
 151     {
 152         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 153          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 154          */
 155         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 156         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 157         is_spam = FALSE;
 158
 159         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 160             if(strcmp(infoLog, spam[i]) == 0) {
 161                 is_spam = TRUE;
 162                 break;
 163             }
 164         }
 165         if(is_spam) {
 166             TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
 167         } else {
 168             FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
 169         }
 170         HeapFree(GetProcessHeap(), 0, infoLog);
 171     }
 172 }
 173
 174 /**
 175  * Loads (pixel shader) samplers
 176  */
 177 static void shader_glsl_load_psamplers(const WineD3D_GL_Info *gl_info, IWineD3DStateBlock *iface, GLhandleARB programId)
 178 {
 179     IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
 180     GLhandleARB name_loc;
 181     int i;
 182     char sampler_name[20];
 183
 184     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 185         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 186         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 187         if (name_loc != -1) {
 188             int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[i];
 189             if (mapped_unit != -1 && mapped_unit < GL_LIMITS(fragment_samplers)) {
 190                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 191                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 192                 checkGLcall("glUniform1iARB");
 193             } else {
 194                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 195             }
 196         }
 197     }
 198 }
 199
 200 static void shader_glsl_load_vsamplers(const WineD3D_GL_Info *gl_info, IWineD3DStateBlock *iface, GLhandleARB programId)
 201 {
 202     IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
 203     GLhandleARB name_loc;
 204     char sampler_name[20];
 205     int i;
 206
 207     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 208         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 209         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 210         if (name_loc != -1) {
 211             int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[MAX_FRAGMENT_SAMPLERS + i];
 212             if (mapped_unit != -1 && mapped_unit < GL_LIMITS(combined_samplers)) {
 213                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 214                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 215                 checkGLcall("glUniform1iARB");
 216             } else {
 217                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 218             }
 219         }
 220     }
 221 }
 222
 223 static inline void walk_constant_heap(const WineD3D_GL_Info *gl_info, const float *constants,
 224         const GLhandleARB *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 225 {
 226     int stack_idx = 0;
 227     unsigned int heap_idx = 1;
 228     unsigned int idx;
 229
 230     if (heap->entries[heap_idx].version <= version) return;
 231
 232     idx = heap->entries[heap_idx].idx;
 233     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 234     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 235
 236     while (stack_idx >= 0)
 237     {
 238         /* Note that we fall through to the next case statement. */
 239         switch(stack[stack_idx])
 240         {
 241             case HEAP_NODE_TRAVERSE_LEFT:
 242             {
 243                 unsigned int left_idx = heap_idx << 1;
 244                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 245                 {
 246                     heap_idx = left_idx;
 247                     idx = heap->entries[heap_idx].idx;
 248                     if (constant_locations[idx] != -1)
 249                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 250
 251                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 252                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 253                     break;
 254                 }
 255             }
 256
 257             case HEAP_NODE_TRAVERSE_RIGHT:
 258             {
 259                 unsigned int right_idx = (heap_idx << 1) + 1;
 260                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 261                 {
 262                     heap_idx = right_idx;
 263                     idx = heap->entries[heap_idx].idx;
 264                     if (constant_locations[idx] != -1)
 265                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 266
 267                     stack[stack_idx++] = HEAP_NODE_POP;
 268                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 269                     break;
 270                 }
 271             }
 272
 273             case HEAP_NODE_POP:
 274             {
 275                 heap_idx >>= 1;
 276                 --stack_idx;
 277                 break;
 278             }
 279         }
 280     }
 281     checkGLcall("walk_constant_heap()");
 282 }
 283
 284 static inline void apply_clamped_constant(const WineD3D_GL_Info *gl_info, GLint location, const GLfloat *data)
 285 {
 286     GLfloat clamped_constant[4];
 287
 288     if (location == -1) return;
 289
 290     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0 ? 1.0 : data[0];
 291     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0 ? 1.0 : data[1];
 292     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0 ? 1.0 : data[2];
 293     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0 ? 1.0 : data[3];
 294
 295     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 296 }
 297
 298 static inline void walk_constant_heap_clamped(const WineD3D_GL_Info *gl_info, const float *constants,
 299         const GLhandleARB *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 300 {
 301     int stack_idx = 0;
 302     unsigned int heap_idx = 1;
 303     unsigned int idx;
 304
 305     if (heap->entries[heap_idx].version <= version) return;
 306
 307     idx = heap->entries[heap_idx].idx;
 308     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 309     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 310
 311     while (stack_idx >= 0)
 312     {
 313         /* Note that we fall through to the next case statement. */
 314         switch(stack[stack_idx])
 315         {
 316             case HEAP_NODE_TRAVERSE_LEFT:
 317             {
 318                 unsigned int left_idx = heap_idx << 1;
 319                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 320                 {
 321                     heap_idx = left_idx;
 322                     idx = heap->entries[heap_idx].idx;
 323                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 324
 325                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 326                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 327                     break;
 328                 }
 329             }
 330
 331             case HEAP_NODE_TRAVERSE_RIGHT:
 332             {
 333                 unsigned int right_idx = (heap_idx << 1) + 1;
 334                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 335                 {
 336                     heap_idx = right_idx;
 337                     idx = heap->entries[heap_idx].idx;
 338                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 339
 340                     stack[stack_idx++] = HEAP_NODE_POP;
 341                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 342                     break;
 343                 }
 344             }
 345
 346             case HEAP_NODE_POP:
 347             {
 348                 heap_idx >>= 1;
 349                 --stack_idx;
 350                 break;
 351             }
 352         }
 353     }
 354     checkGLcall("walk_constant_heap_clamped()");
 355 }
 356
 357 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 358 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
 359         const float *constants, const GLhandleARB *constant_locations, const struct constant_heap *heap,
 360         unsigned char *stack, UINT version)
 361 {
 362     const local_constant *lconst;
 363
 364     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 365     if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.reg_maps.shader_version) == 1
 366             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version))
 367         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 368     else
 369         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 370
 371     if (!This->baseShader.load_local_constsF)
 372     {
 373         TRACE("No need to load local float constants for this shader\n");
 374         return;
 375     }
 376
 377     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 378     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 379     {
 380         GLhandleARB location = constant_locations[lconst->idx];
 381         /* We found this uniform name in the program - go ahead and send the data */
 382         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 383     }
 384     checkGLcall("glUniform4fvARB()");
 385 }
 386
 387 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 388 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
 389         const GLhandleARB locations[MAX_CONST_I], const int *constants, WORD constants_set)
 390 {
 391     unsigned int i;
 392     struct list* ptr;
 393
 394     for (i = 0; constants_set; constants_set >>= 1, ++i)
 395     {
 396         if (!(constants_set & 1)) continue;
 397
 398         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 399                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 400
 401         /* We found this uniform name in the program - go ahead and send the data */
 402         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 403         checkGLcall("glUniform4ivARB");
 404     }
 405
 406     /* Load immediate constants */
 407     ptr = list_head(&This->baseShader.constantsI);
 408     while (ptr) {
 409         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 410         unsigned int idx = lconst->idx;
 411         const GLint *values = (const GLint *)lconst->value;
 412
 413         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 414             values[0], values[1], values[2], values[3]);
 415
 416         /* We found this uniform name in the program - go ahead and send the data */
 417         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 418         checkGLcall("glUniform4ivARB");
 419         ptr = list_next(&This->baseShader.constantsI, ptr);
 420     }
 421 }
 422
 423 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 424 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
 425         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 426 {
 427     GLhandleARB tmp_loc;
 428     unsigned int i;
 429     char tmp_name[8];
 430     char is_pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version);
 431     const char* prefix = is_pshader? "PB":"VB";
 432     struct list* ptr;
 433
 434     /* TODO: Benchmark and see if it would be beneficial to store the
 435      * locations of the constants to avoid looking up each time */
 436     for (i = 0; constants_set; constants_set >>= 1, ++i)
 437     {
 438         if (!(constants_set & 1)) continue;
 439
 440         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 441
 442         /* TODO: Benchmark and see if it would be beneficial to store the
 443          * locations of the constants to avoid looking up each time */
 444         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 445         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 446         if (tmp_loc != -1)
 447         {
 448             /* We found this uniform name in the program - go ahead and send the data */
 449             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 450             checkGLcall("glUniform1ivARB");
 451         }
 452     }
 453
 454     /* Load immediate constants */
 455     ptr = list_head(&This->baseShader.constantsB);
 456     while (ptr) {
 457         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 458         unsigned int idx = lconst->idx;
 459         const GLint *values = (const GLint *)lconst->value;
 460
 461         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 462
 463         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 464         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 465         if (tmp_loc != -1) {
 466             /* We found this uniform name in the program - go ahead and send the data */
 467             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 468             checkGLcall("glUniform1ivARB");
 469         }
 470         ptr = list_next(&This->baseShader.constantsB, ptr);
 471     }
 472 }
 473
 474 static void reset_program_constant_version(void *value, void *context)
 475 {
 476     struct glsl_shader_prog_link *entry = value;
 477     entry->constant_version = 0;
 478 }
 479
 480 /**
 481  * Loads the app-supplied constants into the currently set GLSL program.
 482  */
 483 static void shader_glsl_load_constants(
 484     IWineD3DDevice* device,
 485     char usePixelShader,
 486     char useVertexShader) {
 487
 488     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
 489     struct shader_glsl_priv *priv = deviceImpl->shader_priv;
 490     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 491     const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 492
 493     GLhandleARB programId;
 494     struct glsl_shader_prog_link *prog = priv->glsl_program;
 495     UINT constant_version;
 496     int i;
 497
 498     if (!prog) {
 499         /* No GLSL program set - nothing to do. */
 500         return;
 501     }
 502     programId = prog->programId;
 503     constant_version = prog->constant_version;
 504
 505     if (useVertexShader) {
 506         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 507
 508         /* Load DirectX 9 float constants/uniforms for vertex shader */
 509         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
 510                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 511
 512         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 513         if(vshader->baseShader.uses_int_consts) {
 514             shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations,
 515                     stateBlock->vertexShaderConstantI, stateBlock->changed.vertexShaderConstantsI);
 516         }
 517
 518         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 519         if(vshader->baseShader.uses_bool_consts) {
 520             shader_glsl_load_constantsB(vshader, gl_info, programId,
 521                     stateBlock->vertexShaderConstantB, stateBlock->changed.vertexShaderConstantsB);
 522         }
 523
 524         /* Upload the position fixup params */
 525         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
 526         checkGLcall("glUniform4fvARB");
 527     }
 528
 529     if (usePixelShader) {
 530
 531         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 532
 533         /* Load DirectX 9 float constants/uniforms for pixel shader */
 534         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
 535                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 536
 537         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 538         if(pshader->baseShader.uses_int_consts) {
 539             shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations,
 540                     stateBlock->pixelShaderConstantI, stateBlock->changed.pixelShaderConstantsI);
 541         }
 542
 543         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 544         if(pshader->baseShader.uses_bool_consts) {
 545             shader_glsl_load_constantsB(pshader, gl_info, programId,
 546                     stateBlock->pixelShaderConstantB, stateBlock->changed.pixelShaderConstantsB);
 547         }
 548
 549         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 550          * It can't be 0 for a valid texbem instruction.
 551          */
 552         for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
 553             IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
 554             int stage = ps->luminanceconst[i].texunit;
 555
 556             const float *data = (const float *)&stateBlock->textureState[(int)ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
 557             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 558             checkGLcall("glUniformMatrix2fvARB");
 559
 560             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 561              * is set too, so we can check that in the needsbumpmat check
 562              */
 563             if(ps->baseShader.reg_maps.luminanceparams[stage]) {
 564                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
 565                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
 566
 567                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 568                 checkGLcall("glUniform1fvARB");
 569                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 570                 checkGLcall("glUniform1fvARB");
 571             }
 572         }
 573
 574         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 575             float correction_params[4];
 576             if(deviceImpl->render_offscreen) {
 577                 correction_params[0] = 0.0;
 578                 correction_params[1] = 1.0;
 579             } else {
 580                 /* position is window relative, not viewport relative */
 581                 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 582                 correction_params[1] = -1.0;
 583             }
 584             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 585         }
 586     }
 587
 588     if (priv->next_constant_version == UINT_MAX)
 589     {
 590         TRACE("Max constant version reached, resetting to 0.\n");
 591         hash_table_for_each_entry(priv->glsl_program_lookup, reset_program_constant_version, NULL);
 592         priv->next_constant_version = 1;
 593     }
 594     else
 595     {
 596         prog->constant_version = priv->next_constant_version++;
 597     }
 598 }
 599
 600 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 601         unsigned int heap_idx, DWORD new_version)
 602 {
 603     struct constant_entry *entries = heap->entries;
 604     unsigned int *positions = heap->positions;
 605     unsigned int parent_idx;
 606
 607     while (heap_idx > 1)
 608     {
 609         parent_idx = heap_idx >> 1;
 610
 611         if (new_version <= entries[parent_idx].version) break;
 612
 613         entries[heap_idx] = entries[parent_idx];
 614         positions[entries[parent_idx].idx] = heap_idx;
 615         heap_idx = parent_idx;
 616     }
 617
 618     entries[heap_idx].version = new_version;
 619     entries[heap_idx].idx = idx;
 620     positions[idx] = heap_idx;
 621 }
 622
 623 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 624 {
 625     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 626     struct shader_glsl_priv *priv = This->shader_priv;
 627     struct constant_heap *heap = &priv->vconst_heap;
 628     UINT i;
 629
 630     for (i = start; i < count + start; ++i)
 631     {
 632         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
 633             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 634         else
 635             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 636     }
 637 }
 638
 639 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 640 {
 641     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 642     struct shader_glsl_priv *priv = This->shader_priv;
 643     struct constant_heap *heap = &priv->pconst_heap;
 644     UINT i;
 645
 646     for (i = start; i < count + start; ++i)
 647     {
 648         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 649             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 650         else
 651             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 652     }
 653 }
 654
 655 /** Generate the variable & register declarations for the GLSL output target */
 656 static void shader_generate_glsl_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 657         SHADER_BUFFER *buffer, const WineD3D_GL_Info *gl_info,
 658         const struct ps_compile_args *ps_args)
 659 {
 660     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 661     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 662     DWORD shader_version = reg_maps->shader_version;
 663     unsigned int i, extra_constants_needed = 0;
 664     const local_constant *lconst;
 665
 666     /* There are some minor differences between pixel and vertex shaders */
 667     char pshader = shader_is_pshader_version(shader_version);
 668     char prefix = pshader ? 'P' : 'V';
 669
 670     /* Prototype the subroutines */
 671     for (i = 0; i < This->baseShader.limits.label; i++) {
 672         if (reg_maps->labels[i])
 673             shader_addline(buffer, "void subroutine%u();\n", i);
 674     }
 675
 676     /* Declare the constants (aka uniforms) */
 677     if (This->baseShader.limits.constant_float > 0) {
 678         unsigned max_constantsF = min(This->baseShader.limits.constant_float,
 679                 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
 680         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 681     }
 682
 683     if (This->baseShader.limits.constant_int > 0)
 684         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 685
 686     if (This->baseShader.limits.constant_bool > 0)
 687         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 688
 689     if(!pshader) {
 690         shader_addline(buffer, "uniform vec4 posFixup;\n");
 691         /* Predeclaration; This function is added at link time based on the pixel shader.
 692          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 693          * that. We know the input to the reorder function at vertex shader compile time, so
 694          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 695          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 696          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 697          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 698          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 699          * inout.
 700          */
 701         if (shader_version >= WINED3DVS_VERSION(3, 0))
 702         {
 703             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 704         } else {
 705             shader_addline(buffer, "void order_ps_input();\n");
 706         }
 707     } else {
 708         IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
 709
 710         ps_impl->numbumpenvmatconsts = 0;
 711         for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
 712             if(!reg_maps->bumpmat[i]) {
 713                 continue;
 714             }
 715
 716             ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
 717             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 718
 719             if(reg_maps->luminanceparams) {
 720                 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
 721                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
 722                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
 723                 extra_constants_needed++;
 724             } else {
 725                 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
 726             }
 727
 728             extra_constants_needed++;
 729             ps_impl->numbumpenvmatconsts++;
 730         }
 731
 732         if(ps_args->srgb_correction) {
 733             shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
 734                             srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
 735             shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
 736                             srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
 737         }
 738         if(reg_maps->vpos || reg_maps->usesdsy) {
 739             if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
 740                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
 741                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
 742                 extra_constants_needed++;
 743             } else {
 744                 /* This happens because we do not have proper tracking of the constant registers that are
 745                  * actually used, only the max limit of the shader version
 746                  */
 747                 FIXME("Cannot find a free uniform for vpos correction params\n");
 748                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
 749                                device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
 750                                device->render_offscreen ? 1.0 : -1.0);
 751             }
 752             shader_addline(buffer, "vec4 vpos;\n");
 753         }
 754     }
 755
 756     /* Declare texture samplers */
 757     for (i = 0; i < This->baseShader.limits.sampler; i++) {
 758         if (reg_maps->samplers[i]) {
 759
 760             DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
 761             switch (stype) {
 762
 763                 case WINED3DSTT_1D:
 764                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
 765                     break;
 766                 case WINED3DSTT_2D:
 767                     if(device->stateBlock->textures[i] &&
 768                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
 769                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
 770                     } else {
 771                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
 772                     }
 773                     break;
 774                 case WINED3DSTT_CUBE:
 775                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
 776                     break;
 777                 case WINED3DSTT_VOLUME:
 778                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
 779                     break;
 780                 default:
 781                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
 782                     FIXME("Unrecognized sampler type: %#x\n", stype);
 783                     break;
 784             }
 785         }
 786     }
 787
 788     /* Declare address variables */
 789     for (i = 0; i < This->baseShader.limits.address; i++) {
 790         if (reg_maps->address[i])
 791             shader_addline(buffer, "ivec4 A%d;\n", i);
 792     }
 793
 794     /* Declare texture coordinate temporaries and initialize them */
 795     for (i = 0; i < This->baseShader.limits.texcoord; i++) {
 796         if (reg_maps->texcoord[i])
 797             shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
 798     }
 799
 800     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
 801      * helper function shader that is linked in at link time
 802      */
 803     if (pshader && shader_version >= WINED3DPS_VERSION(3, 0))
 804     {
 805         if (use_vs(device->stateBlock))
 806         {
 807             shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 808         } else {
 809             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
 810              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
 811              * pixel shader that reads the fixed function color into the packed input registers.
 812              */
 813             shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 814         }
 815     }
 816
 817     /* Declare output register temporaries */
 818     if(This->baseShader.limits.packed_output) {
 819         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
 820     }
 821
 822     /* Declare temporary variables */
 823     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 824         if (reg_maps->temporary[i])
 825             shader_addline(buffer, "vec4 R%u;\n", i);
 826     }
 827
 828     /* Declare attributes */
 829     for (i = 0; i < This->baseShader.limits.attributes; i++) {
 830         if (reg_maps->attributes[i])
 831             shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
 832     }
 833
 834     /* Declare loop registers aLx */
 835     for (i = 0; i < reg_maps->loop_depth; i++) {
 836         shader_addline(buffer, "int aL%u;\n", i);
 837         shader_addline(buffer, "int tmpInt%u;\n", i);
 838     }
 839
 840     /* Temporary variables for matrix operations */
 841     shader_addline(buffer, "vec4 tmp0;\n");
 842     shader_addline(buffer, "vec4 tmp1;\n");
 843
 844     /* Local constants use a different name so they can be loaded once at shader link time
 845      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
 846      * float -> string conversion can cause precision loss.
 847      */
 848     if(!This->baseShader.load_local_constsF) {
 849         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 850             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
 851         }
 852     }
 853
 854     /* Start the main program */
 855     shader_addline(buffer, "void main() {\n");
 856     if(pshader && reg_maps->vpos) {
 857         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
 858          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
 859          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
 860          * precision troubles when we just substract 0.5.
 861          *
 862          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
 863          *
 864          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
 865          *
 866          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
 867          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
 868          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
 869          * correctly on drivers that returns integer values.
 870          */
 871         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
 872     }
 873 }
 874
 875 /*****************************************************************************
 876  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
 877  *
 878  * For more information, see http://wiki.winehq.org/DirectX-Shaders
 879  ****************************************************************************/
 880
 881 /* Prototypes */
 882 static void shader_glsl_add_src_param(const SHADER_OPCODE_ARG *arg, const DWORD param,
 883         const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
 884
 885 /** Used for opcode modifiers - They multiply the result by the specified amount */
 886 static const char * const shift_glsl_tab[] = {
 887     "",           /*  0 (none) */
 888     "2.0 * ",     /*  1 (x2)   */
 889     "4.0 * ",     /*  2 (x4)   */
 890     "8.0 * ",     /*  3 (x8)   */
 891     "16.0 * ",    /*  4 (x16)  */
 892     "32.0 * ",    /*  5 (x32)  */
 893     "",           /*  6 (x64)  */
 894     "",           /*  7 (x128) */
 895     "",           /*  8 (d256) */
 896     "",           /*  9 (d128) */
 897     "",           /* 10 (d64)  */
 898     "",           /* 11 (d32)  */
 899     "0.0625 * ",  /* 12 (d16)  */
 900     "0.125 * ",   /* 13 (d8)   */
 901     "0.25 * ",    /* 14 (d4)   */
 902     "0.5 * "      /* 15 (d2)   */
 903 };
 904
 905 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
 906 static void shader_glsl_gen_modifier (
 907     const DWORD instr,
 908     const char *in_reg,
 909     const char *in_regswizzle,
 910     char *out_str) {
 911
 912     out_str[0] = 0;
 913
 914     if (instr == WINED3DSIO_TEXKILL)
 915         return;
 916
 917     switch (instr & WINED3DSP_SRCMOD_MASK) {
 918     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
 919     case WINED3DSPSM_DW:
 920     case WINED3DSPSM_NONE:
 921         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
 922         break;
 923     case WINED3DSPSM_NEG:
 924         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
 925         break;
 926     case WINED3DSPSM_NOT:
 927         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
 928         break;
 929     case WINED3DSPSM_BIAS:
 930         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
 931         break;
 932     case WINED3DSPSM_BIASNEG:
 933         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
 934         break;
 935     case WINED3DSPSM_SIGN:
 936         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
 937         break;
 938     case WINED3DSPSM_SIGNNEG:
 939         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
 940         break;
 941     case WINED3DSPSM_COMP:
 942         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
 943         break;
 944     case WINED3DSPSM_X2:
 945         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
 946         break;
 947     case WINED3DSPSM_X2NEG:
 948         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
 949         break;
 950     case WINED3DSPSM_ABS:
 951         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
 952         break;
 953     case WINED3DSPSM_ABSNEG:
 954         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
 955         break;
 956     default:
 957         FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
 958         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
 959     }
 960 }
 961
 962 /** Writes the GLSL variable name that corresponds to the register that the
 963  * DX opcode parameter is trying to access */
 964 static void shader_glsl_get_register_name(const DWORD param, const DWORD addr_token,
 965         char *regstr, BOOL *is_color, const SHADER_OPCODE_ARG *arg)
 966 {
 967     /* oPos, oFog and oPts in D3D */
 968     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
 969
 970     DWORD reg = param & WINED3DSP_REGNUM_MASK;
 971     DWORD regtype = shader_get_regtype(param);
 972     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
 973     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
 974     const WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
 975     DWORD shader_version = This->baseShader.reg_maps.shader_version;
 976     char pshader = shader_is_pshader_version(shader_version);
 977     char tmpStr[150];
 978
 979     *is_color = FALSE;
 980
 981     switch (regtype) {
 982     case WINED3DSPR_TEMP:
 983         sprintf(tmpStr, "R%u", reg);
 984     break;
 985     case WINED3DSPR_INPUT:
 986         if (pshader) {
 987             /* Pixel shaders >= 3.0 */
 988             if (WINED3DSHADER_VERSION_MAJOR(shader_version) >= 3)
 989             {
 990                 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
 991
 992                 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
 993                     glsl_src_param_t rel_param;
 994                     shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
 995
 996                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
 997                      * operation there
 998                      */
 999                     if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
1000                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
1001                             sprintf(tmpStr, "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1002                                     rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count - 1,
1003                                     rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count,
1004                                     rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
1005                         } else {
1006                             sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
1007                         }
1008                     } else {
1009                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
1010                             sprintf(tmpStr, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1011                                     rel_param.param_str, in_count - 1,
1012                                     rel_param.param_str, in_count,
1013                                     rel_param.param_str);
1014                         } else {
1015                             sprintf(tmpStr, "IN[%s]", rel_param.param_str);
1016                         }
1017                     }
1018                 } else {
1019                     DWORD idx = ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg];
1020                     if (idx == in_count) {
1021                         sprintf(tmpStr, "gl_Color");
1022                     } else if (idx == in_count + 1) {
1023                         sprintf(tmpStr, "gl_SecondaryColor");
1024                     } else {
1025                         sprintf(tmpStr, "IN[%u]", idx);
1026                     }
1027                 }
1028             } else {
1029                 if (reg==0)
1030                     strcpy(tmpStr, "gl_Color");
1031                 else
1032                     strcpy(tmpStr, "gl_SecondaryColor");
1033             }
1034         } else {
1035             if (((IWineD3DVertexShaderImpl *)This)->cur_args->swizzle_map & (1 << reg)) *is_color = TRUE;
1036             sprintf(tmpStr, "attrib%u", reg);
1037         }
1038         break;
1039     case WINED3DSPR_CONST:
1040     {
1041         const char prefix = pshader? 'P':'V';
1042
1043         /* Relative addressing */
1044         if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
1045
1046            /* Relative addressing on shaders 2.0+ have a relative address token,
1047             * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
1048            if (WINED3DSHADER_VERSION_MAJOR(shader_version) >= 2)
1049            {
1050                glsl_src_param_t rel_param;
1051                shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
1052                if(reg) {
1053                    sprintf(tmpStr, "%cC[%s + %u]", prefix, rel_param.param_str, reg);
1054                } else {
1055                    sprintf(tmpStr, "%cC[%s]", prefix, rel_param.param_str);
1056                }
1057            } else {
1058                if(reg) {
1059                    sprintf(tmpStr, "%cC[A0.x + %u]", prefix, reg);
1060                } else {
1061                    sprintf(tmpStr, "%cC[A0.x]", prefix);
1062                }
1063            }
1064
1065         } else {
1066             if(shader_constant_is_local(This, reg)) {
1067                 sprintf(tmpStr, "%cLC%u", prefix, reg);
1068             } else {
1069                 sprintf(tmpStr, "%cC[%u]", prefix, reg);
1070             }
1071         }
1072
1073         break;
1074     }
1075     case WINED3DSPR_CONSTINT:
1076         if (pshader)
1077             sprintf(tmpStr, "PI[%u]", reg);
1078         else
1079             sprintf(tmpStr, "VI[%u]", reg);
1080         break;
1081     case WINED3DSPR_CONSTBOOL:
1082         if (pshader)
1083             sprintf(tmpStr, "PB[%u]", reg);
1084         else
1085             sprintf(tmpStr, "VB[%u]", reg);
1086         break;
1087     case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1088         if (pshader) {
1089             sprintf(tmpStr, "T%u", reg);
1090         } else {
1091             sprintf(tmpStr, "A%u", reg);
1092         }
1093     break;
1094     case WINED3DSPR_LOOP:
1095         sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
1096     break;
1097     case WINED3DSPR_SAMPLER:
1098         if (pshader)
1099             sprintf(tmpStr, "Psampler%u", reg);
1100         else
1101             sprintf(tmpStr, "Vsampler%u", reg);
1102     break;
1103     case WINED3DSPR_COLOROUT:
1104         if (reg >= GL_LIMITS(buffers)) {
1105             WARN("Write to render target %u, only %d supported\n", reg, 4);
1106         }
1107         if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
1108             sprintf(tmpStr, "gl_FragData[%u]", reg);
1109         } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
1110             sprintf(tmpStr, "gl_FragColor");
1111         }
1112     break;
1113     case WINED3DSPR_RASTOUT:
1114         sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
1115     break;
1116     case WINED3DSPR_DEPTHOUT:
1117         sprintf(tmpStr, "gl_FragDepth");
1118     break;
1119     case WINED3DSPR_ATTROUT:
1120         if (reg == 0) {
1121             sprintf(tmpStr, "gl_FrontColor");
1122         } else {
1123             sprintf(tmpStr, "gl_FrontSecondaryColor");
1124         }
1125     break;
1126     case WINED3DSPR_TEXCRDOUT:
1127         /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1128         if (WINED3DSHADER_VERSION_MAJOR(shader_version) >= 3) sprintf(tmpStr, "OUT[%u]", reg);
1129         else sprintf(tmpStr, "gl_TexCoord[%u]", reg);
1130     break;
1131     case WINED3DSPR_MISCTYPE:
1132         if (reg == 0) {
1133             /* vPos */
1134             sprintf(tmpStr, "vpos");
1135         } else if (reg == 1){
1136             /* Note that gl_FrontFacing is a bool, while vFace is
1137              * a float for which the sign determines front/back
1138              */
1139             sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
1140         } else {
1141             FIXME("Unhandled misctype register %d\n", reg);
1142             sprintf(tmpStr, "unrecognized_register");
1143         }
1144         break;
1145     default:
1146         FIXME("Unhandled register name Type(%d)\n", regtype);
1147         sprintf(tmpStr, "unrecognized_register");
1148     break;
1149     }
1150
1151     strcat(regstr, tmpStr);
1152 }
1153
1154 /* Get the GLSL write mask for the destination register */
1155 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
1156     char *ptr = write_mask;
1157     DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
1158
1159     if (shader_is_scalar(param)) {
1160         mask = WINED3DSP_WRITEMASK_0;
1161     } else {
1162         *ptr++ = '.';
1163         if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1164         if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1165         if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1166         if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1167     }
1168
1169     *ptr = '\0';
1170
1171     return mask;
1172 }
1173
1174 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1175     unsigned int size = 0;
1176
1177     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1178     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1179     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1180     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1181
1182     return size;
1183 }
1184
1185 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1186     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1187      * but addressed as "rgba". To fix this we need to swap the register's x
1188      * and z components. */
1189     DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1190     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1191     char *ptr = swizzle_str;
1192
1193     if (!shader_is_scalar(param)) {
1194         *ptr++ = '.';
1195         /* swizzle bits fields: wwzzyyxx */
1196         if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1197         if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1198         if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1199         if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1200     }
1201
1202     *ptr = '\0';
1203 }
1204
1205 /* From a given parameter token, generate the corresponding GLSL string.
1206  * Also, return the actual register name and swizzle in case the
1207  * caller needs this information as well. */
1208 static void shader_glsl_add_src_param(const SHADER_OPCODE_ARG *arg, const DWORD param,
1209         const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param)
1210 {
1211     BOOL is_color = FALSE;
1212     char swizzle_str[6];
1213
1214     src_param->reg_name[0] = '\0';
1215     src_param->param_str[0] = '\0';
1216     swizzle_str[0] = '\0';
1217
1218     shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1219
1220     shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1221     shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1222 }
1223
1224 /* From a given parameter token, generate the corresponding GLSL string.
1225  * Also, return the actual register name and swizzle in case the
1226  * caller needs this information as well. */
1227 static DWORD shader_glsl_add_dst_param(const SHADER_OPCODE_ARG* arg, const DWORD param,
1228         const DWORD addr_token, glsl_dst_param_t *dst_param)
1229 {
1230     BOOL is_color = FALSE;
1231
1232     dst_param->mask_str[0] = '\0';
1233     dst_param->reg_name[0] = '\0';
1234
1235     shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1236     return shader_glsl_get_write_mask(param, dst_param->mask_str);
1237 }
1238
1239 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1240 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, const SHADER_OPCODE_ARG *arg, const DWORD param)
1241 {
1242     glsl_dst_param_t dst_param;
1243     DWORD mask;
1244     int shift;
1245
1246     mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1247
1248     if(mask) {
1249         shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1250         shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1251     }
1252
1253     return mask;
1254 }
1255
1256 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1257 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, const SHADER_OPCODE_ARG *arg)
1258 {
1259     return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1260 }
1261
1262 /** Process GLSL instruction modifiers */
1263 void shader_glsl_add_instruction_modifiers(const SHADER_OPCODE_ARG* arg)
1264 {
1265     DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1266
1267     if (arg->opcode->dst_token && mask != 0) {
1268         glsl_dst_param_t dst_param;
1269
1270         shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1271
1272         if (mask & WINED3DSPDM_SATURATE) {
1273             /* _SAT means to clamp the value of the register to between 0 and 1 */
1274             shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1275                     dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1276         }
1277         if (mask & WINED3DSPDM_MSAMPCENTROID) {
1278             FIXME("_centroid modifier not handled\n");
1279         }
1280         if (mask & WINED3DSPDM_PARTIALPRECISION) {
1281             /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1282         }
1283     }
1284 }
1285
1286 static inline const char* shader_get_comp_op(
1287     const DWORD opcode) {
1288
1289     DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1290     switch (op) {
1291         case COMPARISON_GT: return ">";
1292         case COMPARISON_EQ: return "==";
1293         case COMPARISON_GE: return ">=";
1294         case COMPARISON_LT: return "<";
1295         case COMPARISON_NE: return "!=";
1296         case COMPARISON_LE: return "<=";
1297         default:
1298             FIXME("Unrecognized comparison value: %u\n", op);
1299             return "(\?\?)";
1300     }
1301 }
1302
1303 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1304     /* Note that there's no such thing as a projected cube texture. */
1305     switch(sampler_type) {
1306         case WINED3DSTT_1D:
1307             sample_function->name = projected ? "texture1DProj" : "texture1D";
1308             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1309             break;
1310         case WINED3DSTT_2D:
1311             if(texrect) {
1312                 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1313             } else {
1314                 sample_function->name = projected ? "texture2DProj" : "texture2D";
1315             }
1316             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1317             break;
1318         case WINED3DSTT_CUBE:
1319             sample_function->name = "textureCube";
1320             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1321             break;
1322         case WINED3DSTT_VOLUME:
1323             sample_function->name = projected ? "texture3DProj" : "texture3D";
1324             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1325             break;
1326         default:
1327             sample_function->name = "";
1328             sample_function->coord_mask = 0;
1329             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1330             break;
1331     }
1332 }
1333
1334 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1335         BOOL sign_fixup, enum fixup_channel_source channel_source)
1336 {
1337     switch(channel_source)
1338     {
1339         case CHANNEL_SOURCE_ZERO:
1340             strcat(arguments, "0.0");
1341             break;
1342
1343         case CHANNEL_SOURCE_ONE:
1344             strcat(arguments, "1.0");
1345             break;
1346
1347         case CHANNEL_SOURCE_X:
1348             strcat(arguments, reg_name);
1349             strcat(arguments, ".x");
1350             break;
1351
1352         case CHANNEL_SOURCE_Y:
1353             strcat(arguments, reg_name);
1354             strcat(arguments, ".y");
1355             break;
1356
1357         case CHANNEL_SOURCE_Z:
1358             strcat(arguments, reg_name);
1359             strcat(arguments, ".z");
1360             break;
1361
1362         case CHANNEL_SOURCE_W:
1363             strcat(arguments, reg_name);
1364             strcat(arguments, ".w");
1365             break;
1366
1367         default:
1368             FIXME("Unhandled channel source %#x\n", channel_source);
1369             strcat(arguments, "undefined");
1370             break;
1371     }
1372
1373     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1374 }
1375
1376 static void shader_glsl_color_correction(const struct SHADER_OPCODE_ARG *arg, struct color_fixup_desc fixup)
1377 {
1378     unsigned int mask_size, remaining;
1379     glsl_dst_param_t dst_param;
1380     char arguments[256];
1381     DWORD mask;
1382     BOOL dummy;
1383
1384     mask = 0;
1385     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1386     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1387     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1388     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1389     mask &= arg->dst;
1390
1391     if (!mask) return; /* Nothing to do */
1392
1393     if (is_yuv_fixup(fixup))
1394     {
1395         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1396         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1397         return;
1398     }
1399
1400     mask_size = shader_glsl_get_write_mask_size(mask);
1401
1402     dst_param.mask_str[0] = '\0';
1403     shader_glsl_get_write_mask(mask, dst_param.mask_str);
1404
1405     dst_param.reg_name[0] = '\0';
1406     shader_glsl_get_register_name(arg->dst, arg->dst_addr, dst_param.reg_name, &dummy, arg);
1407
1408     arguments[0] = '\0';
1409     remaining = mask_size;
1410     if (mask & WINED3DSP_WRITEMASK_0)
1411     {
1412         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1413         if (--remaining) strcat(arguments, ", ");
1414     }
1415     if (mask & WINED3DSP_WRITEMASK_1)
1416     {
1417         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1418         if (--remaining) strcat(arguments, ", ");
1419     }
1420     if (mask & WINED3DSP_WRITEMASK_2)
1421     {
1422         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1423         if (--remaining) strcat(arguments, ", ");
1424     }
1425     if (mask & WINED3DSP_WRITEMASK_3)
1426     {
1427         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1428         if (--remaining) strcat(arguments, ", ");
1429     }
1430
1431     if (mask_size > 1)
1432     {
1433         shader_addline(arg->buffer, "%s%s = vec%u(%s);\n",
1434                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1435     }
1436     else
1437     {
1438         shader_addline(arg->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1439     }
1440 }
1441
1442 /*****************************************************************************
1443  *
1444  * Begin processing individual instruction opcodes
1445  *
1446  ****************************************************************************/
1447
1448 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1449 static void shader_glsl_arith(const SHADER_OPCODE_ARG *arg)
1450 {
1451     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1452     SHADER_BUFFER* buffer = arg->buffer;
1453     glsl_src_param_t src0_param;
1454     glsl_src_param_t src1_param;
1455     DWORD write_mask;
1456     char op;
1457
1458     /* Determine the GLSL operator to use based on the opcode */
1459     switch (curOpcode->opcode) {
1460         case WINED3DSIO_MUL: op = '*'; break;
1461         case WINED3DSIO_ADD: op = '+'; break;
1462         case WINED3DSIO_SUB: op = '-'; break;
1463         default:
1464             op = ' ';
1465             FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1466             break;
1467     }
1468
1469     write_mask = shader_glsl_append_dst(buffer, arg);
1470     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1471     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1472     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1473 }
1474
1475 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1476 static void shader_glsl_mov(const SHADER_OPCODE_ARG *arg)
1477 {
1478     SHADER_BUFFER* buffer = arg->buffer;
1479     glsl_src_param_t src0_param;
1480     DWORD write_mask;
1481
1482     write_mask = shader_glsl_append_dst(buffer, arg);
1483     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1484
1485     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1486      * shader versions WINED3DSIO_MOVA is used for this. */
1487     if ((WINED3DSHADER_VERSION_MAJOR(arg->reg_maps->shader_version) == 1
1488             && !shader_is_pshader_version(arg->reg_maps->shader_version)
1489             && shader_get_regtype(arg->dst) == WINED3DSPR_ADDR))
1490     {
1491         /* This is a simple floor() */
1492         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1493         if (mask_size > 1) {
1494             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1495         } else {
1496             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1497         }
1498     } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1499         /* We need to *round* to the nearest int here. */
1500         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1501         if (mask_size > 1) {
1502             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);
1503         } else {
1504             shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1505         }
1506     } else {
1507         shader_addline(buffer, "%s);\n", src0_param.param_str);
1508     }
1509 }
1510
1511 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1512 static void shader_glsl_dot(const SHADER_OPCODE_ARG *arg)
1513 {
1514     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1515     SHADER_BUFFER* buffer = arg->buffer;
1516     glsl_src_param_t src0_param;
1517     glsl_src_param_t src1_param;
1518     DWORD dst_write_mask, src_write_mask;
1519     unsigned int dst_size = 0;
1520
1521     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1522     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1523
1524     /* dp3 works on vec3, dp4 on vec4 */
1525     if (curOpcode->opcode == WINED3DSIO_DP4) {
1526         src_write_mask = WINED3DSP_WRITEMASK_ALL;
1527     } else {
1528         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1529     }
1530
1531     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1532     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1533
1534     if (dst_size > 1) {
1535         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1536     } else {
1537         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1538     }
1539 }
1540
1541 /* Note that this instruction has some restrictions. The destination write mask
1542  * can't contain the w component, and the source swizzles have to be .xyzw */
1543 static void shader_glsl_cross(const SHADER_OPCODE_ARG *arg)
1544 {
1545     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1546     glsl_src_param_t src0_param;
1547     glsl_src_param_t src1_param;
1548     char dst_mask[6];
1549
1550     shader_glsl_get_write_mask(arg->dst, dst_mask);
1551     shader_glsl_append_dst(arg->buffer, arg);
1552     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1553     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1554     shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1555 }
1556
1557 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1558  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1559  * GLSL uses the value as-is. */
1560 static void shader_glsl_pow(const SHADER_OPCODE_ARG *arg)
1561 {
1562     SHADER_BUFFER *buffer = arg->buffer;
1563     glsl_src_param_t src0_param;
1564     glsl_src_param_t src1_param;
1565     DWORD dst_write_mask;
1566     unsigned int dst_size;
1567
1568     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1569     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1570
1571     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1572     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1573
1574     if (dst_size > 1) {
1575         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1576     } else {
1577         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1578     }
1579 }
1580
1581 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1582  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1583  * GLSL uses the value as-is. */
1584 static void shader_glsl_log(const SHADER_OPCODE_ARG *arg)
1585 {
1586     SHADER_BUFFER *buffer = arg->buffer;
1587     glsl_src_param_t src0_param;
1588     DWORD dst_write_mask;
1589     unsigned int dst_size;
1590
1591     dst_write_mask = shader_glsl_append_dst(buffer, arg);
1592     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1593
1594     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1595
1596     if (dst_size > 1) {
1597         shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1598     } else {
1599         shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1600     }
1601 }
1602
1603 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1604 static void shader_glsl_map2gl(const SHADER_OPCODE_ARG *arg)
1605 {
1606     CONST SHADER_OPCODE* curOpcode = arg->opcode;
1607     SHADER_BUFFER* buffer = arg->buffer;
1608     glsl_src_param_t src_param;
1609     const char *instruction;
1610     char arguments[256];
1611     DWORD write_mask;
1612     unsigned i;
1613
1614     /* Determine the GLSL function to use based on the opcode */
1615     /* TODO: Possibly make this a table for faster lookups */
1616     switch (curOpcode->opcode) {
1617         case WINED3DSIO_MIN: instruction = "min"; break;
1618         case WINED3DSIO_MAX: instruction = "max"; break;
1619         case WINED3DSIO_ABS: instruction = "abs"; break;
1620         case WINED3DSIO_FRC: instruction = "fract"; break;
1621         case WINED3DSIO_NRM: instruction = "normalize"; break;
1622         case WINED3DSIO_EXP: instruction = "exp2"; break;
1623         case WINED3DSIO_SGN: instruction = "sign"; break;
1624         case WINED3DSIO_DSX: instruction = "dFdx"; break;
1625         case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1626         default: instruction = "";
1627             FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1628             break;
1629     }
1630
1631     write_mask = shader_glsl_append_dst(buffer, arg);
1632
1633     arguments[0] = '\0';
1634     if (curOpcode->num_params > 0) {
1635         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1636         strcat(arguments, src_param.param_str);
1637         for (i = 2; i < curOpcode->num_params; ++i) {
1638             strcat(arguments, ", ");
1639             shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1640             strcat(arguments, src_param.param_str);
1641         }
1642     }
1643
1644     shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1645 }
1646
1647 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1648  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1649  *   dst.x = 2^(floor(src))
1650  *   dst.y = src - floor(src)
1651  *   dst.z = 2^src   (partial precision is allowed, but optional)
1652  *   dst.w = 1.0;
1653  * For 2.0 shaders, just do this (honoring writemask and swizzle):
1654  *   dst = 2^src;    (partial precision is allowed, but optional)
1655  */
1656 static void shader_glsl_expp(const SHADER_OPCODE_ARG *arg)
1657 {
1658     glsl_src_param_t src_param;
1659
1660     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1661
1662     if (arg->reg_maps->shader_version < WINED3DPS_VERSION(2,0))
1663     {
1664         char dst_mask[6];
1665
1666         shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1667         shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1668         shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1669         shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1670
1671         shader_glsl_append_dst(arg->buffer, arg);
1672         shader_glsl_get_write_mask(arg->dst, dst_mask);
1673         shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1674     } else {
1675         DWORD write_mask;
1676         unsigned int mask_size;
1677
1678         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1679         mask_size = shader_glsl_get_write_mask_size(write_mask);
1680
1681         if (mask_size > 1) {
1682             shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1683         } else {
1684             shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1685         }
1686     }
1687 }
1688
1689 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1690 static void shader_glsl_rcp(const SHADER_OPCODE_ARG *arg)
1691 {
1692     glsl_src_param_t src_param;
1693     DWORD write_mask;
1694     unsigned int mask_size;
1695
1696     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1697     mask_size = shader_glsl_get_write_mask_size(write_mask);
1698     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1699
1700     if (mask_size > 1) {
1701         shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1702     } else {
1703         shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1704     }
1705 }
1706
1707 static void shader_glsl_rsq(const SHADER_OPCODE_ARG *arg)
1708 {
1709     SHADER_BUFFER* buffer = arg->buffer;
1710     glsl_src_param_t src_param;
1711     DWORD write_mask;
1712     unsigned int mask_size;
1713
1714     write_mask = shader_glsl_append_dst(buffer, arg);
1715     mask_size = shader_glsl_get_write_mask_size(write_mask);
1716
1717     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1718
1719     if (mask_size > 1) {
1720         shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1721     } else {
1722         shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1723     }
1724 }
1725
1726 /** Process signed comparison opcodes in GLSL. */
1727 static void shader_glsl_compare(const SHADER_OPCODE_ARG *arg)
1728 {
1729     glsl_src_param_t src0_param;
1730     glsl_src_param_t src1_param;
1731     DWORD write_mask;
1732     unsigned int mask_size;
1733
1734     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1735     mask_size = shader_glsl_get_write_mask_size(write_mask);
1736     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1737     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1738
1739     if (mask_size > 1) {
1740         const char *compare;
1741
1742         switch(arg->opcode->opcode) {
1743             case WINED3DSIO_SLT: compare = "lessThan"; break;
1744             case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1745             default: compare = "";
1746                 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1747         }
1748
1749         shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1750                 src0_param.param_str, src1_param.param_str);
1751     } else {
1752         switch(arg->opcode->opcode) {
1753             case WINED3DSIO_SLT:
1754                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1755                  * to return 0.0 but step returns 1.0 because step is not < x
1756                  * An alternative is a bvec compare padded with an unused second component.
1757                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1758                  * issue. Playing with not() is not possible either because not() does not accept
1759                  * a scalar.
1760                  */
1761                 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1762                 break;
1763             case WINED3DSIO_SGE:
1764                 /* Here we can use the step() function and safe a conditional */
1765                 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1766                 break;
1767             default:
1768                 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1769         }
1770
1771     }
1772 }
1773
1774 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1775 static void shader_glsl_cmp(const SHADER_OPCODE_ARG *arg)
1776 {
1777     glsl_src_param_t src0_param;
1778     glsl_src_param_t src1_param;
1779     glsl_src_param_t src2_param;
1780     DWORD write_mask, cmp_channel = 0;
1781     unsigned int i, j;
1782     char mask_char[6];
1783     BOOL temp_destination = FALSE;
1784
1785     if(shader_is_scalar(arg->src[0])) {
1786         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1787
1788         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1789         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1790         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1791
1792         shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1793                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
1794     } else {
1795         DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1796         DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1797         DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1798         DWORD src0regtype = shader_get_regtype(arg->src[0]);
1799         DWORD src1regtype = shader_get_regtype(arg->src[1]);
1800         DWORD src2regtype = shader_get_regtype(arg->src[2]);
1801         DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1802         DWORD dstregtype = shader_get_regtype(arg->dst);
1803
1804         /* Cycle through all source0 channels */
1805         for (i=0; i<4; i++) {
1806             write_mask = 0;
1807             /* Find the destination channels which use the current source0 channel */
1808             for (j=0; j<4; j++) {
1809                 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1810                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
1811                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1812                 }
1813             }
1814
1815             /* Splitting the cmp instruction up in multiple lines imposes a problem:
1816             * The first lines may overwrite source parameters of the following lines.
1817             * Deal with that by using a temporary destination register if needed
1818             */
1819             if((src0reg == dstreg && src0regtype == dstregtype) ||
1820             (src1reg == dstreg && src1regtype == dstregtype) ||
1821             (src2reg == dstreg && src2regtype == dstregtype)) {
1822
1823                 write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1824                 if (!write_mask) continue;
1825                 shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1826                 temp_destination = TRUE;
1827             } else {
1828                 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1829                 if (!write_mask) continue;
1830             }
1831
1832             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1833             shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1834             shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1835
1836             shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1837                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
1838         }
1839
1840         if(temp_destination) {
1841             shader_glsl_get_write_mask(arg->dst, mask_char);
1842             shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1843             shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1844         }
1845     }
1846
1847 }
1848
1849 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1850 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1851  * the compare is done per component of src0. */
1852 static void shader_glsl_cnd(const SHADER_OPCODE_ARG *arg)
1853 {
1854     glsl_src_param_t src0_param;
1855     glsl_src_param_t src1_param;
1856     glsl_src_param_t src2_param;
1857     DWORD write_mask, cmp_channel = 0;
1858     unsigned int i, j;
1859
1860     if (arg->reg_maps->shader_version < WINED3DPS_VERSION(1, 4))
1861     {
1862         write_mask = shader_glsl_append_dst(arg->buffer, arg);
1863         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1864         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1865         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1866
1867         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1868         if(arg->opcode_token & WINED3DSI_COISSUE) {
1869             shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1870         } else {
1871             shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1872                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
1873         }
1874         return;
1875     }
1876     /* Cycle through all source0 channels */
1877     for (i=0; i<4; i++) {
1878         write_mask = 0;
1879         /* Find the destination channels which use the current source0 channel */
1880         for (j=0; j<4; j++) {
1881             if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1882                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1883                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1884             }
1885         }
1886         write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1887         if (!write_mask) continue;
1888
1889         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1890         shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1891         shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1892
1893         shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1894                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1895     }
1896 }
1897
1898 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1899 static void shader_glsl_mad(const SHADER_OPCODE_ARG *arg)
1900 {
1901     glsl_src_param_t src0_param;
1902     glsl_src_param_t src1_param;
1903     glsl_src_param_t src2_param;
1904     DWORD write_mask;
1905
1906     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1907     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1908     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1909     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1910     shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1911             src0_param.param_str, src1_param.param_str, src2_param.param_str);
1912 }
1913
1914 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1915     Vertex shaders to GLSL codes */
1916 static void shader_glsl_mnxn(const SHADER_OPCODE_ARG *arg)
1917 {
1918     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1919     const SHADER_OPCODE *opcode_table = shader->baseShader.shader_ins;
1920     DWORD shader_version = arg->reg_maps->shader_version;
1921     int i;
1922     int nComponents = 0;
1923     SHADER_OPCODE_ARG tmpArg;
1924
1925     memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1926
1927     /* Set constants for the temporary argument */
1928     tmpArg.shader      = arg->shader;
1929     tmpArg.buffer      = arg->buffer;
1930     tmpArg.src[0]      = arg->src[0];
1931     tmpArg.src_addr[0] = arg->src_addr[0];
1932     tmpArg.src_addr[1] = arg->src_addr[1];
1933     tmpArg.reg_maps = arg->reg_maps;
1934
1935     switch(arg->opcode->opcode) {
1936         case WINED3DSIO_M4x4:
1937             nComponents = 4;
1938             tmpArg.opcode = shader_get_opcode(opcode_table, shader_version, WINED3DSIO_DP4);
1939             break;
1940         case WINED3DSIO_M4x3:
1941             nComponents = 3;
1942             tmpArg.opcode = shader_get_opcode(opcode_table, shader_version, WINED3DSIO_DP4);
1943             break;
1944         case WINED3DSIO_M3x4:
1945             nComponents = 4;
1946             tmpArg.opcode = shader_get_opcode(opcode_table, shader_version, WINED3DSIO_DP3);
1947             break;
1948         case WINED3DSIO_M3x3:
1949             nComponents = 3;
1950             tmpArg.opcode = shader_get_opcode(opcode_table, shader_version, WINED3DSIO_DP3);
1951             break;
1952         case WINED3DSIO_M3x2:
1953             nComponents = 2;
1954             tmpArg.opcode = shader_get_opcode(opcode_table, shader_version, WINED3DSIO_DP3);
1955             break;
1956         default:
1957             break;
1958     }
1959
1960     for (i = 0; i < nComponents; i++) {
1961         tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1962         tmpArg.src[1]      = arg->src[1]+i;
1963         shader_glsl_dot(&tmpArg);
1964     }
1965 }
1966
1967 /**
1968     The LRP instruction performs a component-wise linear interpolation
1969     between the second and third operands using the first operand as the
1970     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
1971     This is equivalent to mix(src2, src1, src0);
1972 */
1973 static void shader_glsl_lrp(const SHADER_OPCODE_ARG *arg)
1974 {
1975     glsl_src_param_t src0_param;
1976     glsl_src_param_t src1_param;
1977     glsl_src_param_t src2_param;
1978     DWORD write_mask;
1979
1980     write_mask = shader_glsl_append_dst(arg->buffer, arg);
1981
1982     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1983     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1984     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1985
1986     shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1987             src2_param.param_str, src1_param.param_str, src0_param.param_str);
1988 }
1989
1990 /** Process the WINED3DSIO_LIT instruction in GLSL:
1991  * dst.x = dst.w = 1.0
1992  * dst.y = (src0.x > 0) ? src0.x
1993  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1994  *                                        where src.w is clamped at +- 128
1995  */
1996 static void shader_glsl_lit(const SHADER_OPCODE_ARG *arg)
1997 {
1998     glsl_src_param_t src0_param;
1999     glsl_src_param_t src1_param;
2000     glsl_src_param_t src3_param;
2001     char dst_mask[6];
2002
2003     shader_glsl_append_dst(arg->buffer, arg);
2004     shader_glsl_get_write_mask(arg->dst, dst_mask);
2005
2006     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2007     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
2008     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
2009
2010     /* The sdk specifies the instruction like this
2011      * dst.x = 1.0;
2012      * if(src.x > 0.0) dst.y = src.x
2013      * else dst.y = 0.0.
2014      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2015      * else dst.z = 0.0;
2016      * dst.w = 1.0;
2017      *
2018      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2019      * dst.x = 1.0                                  ... No further explanation needed
2020      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2021      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2022      * dst.w = 1.0.                                 ... Nothing fancy.
2023      *
2024      * So we still have one conditional in there. So do this:
2025      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2026      *
2027      * 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),
2028      * 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.
2029      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2030      */
2031     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",
2032                    src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2033 }
2034
2035 /** Process the WINED3DSIO_DST instruction in GLSL:
2036  * dst.x = 1.0
2037  * dst.y = src0.x * src0.y
2038  * dst.z = src0.z
2039  * dst.w = src1.w
2040  */
2041 static void shader_glsl_dst(const SHADER_OPCODE_ARG *arg)
2042 {
2043     glsl_src_param_t src0y_param;
2044     glsl_src_param_t src0z_param;
2045     glsl_src_param_t src1y_param;
2046     glsl_src_param_t src1w_param;
2047     char dst_mask[6];
2048
2049     shader_glsl_append_dst(arg->buffer, arg);
2050     shader_glsl_get_write_mask(arg->dst, dst_mask);
2051
2052     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2053     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2054     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2055     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2056
2057     shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2058             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2059 }
2060
2061 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2062  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2063  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2064  *
2065  * dst.x = cos(src0.?)
2066  * dst.y = sin(src0.?)
2067  * dst.z = dst.z
2068  * dst.w = dst.w
2069  */
2070 static void shader_glsl_sincos(const SHADER_OPCODE_ARG *arg)
2071 {
2072     glsl_src_param_t src0_param;
2073     DWORD write_mask;
2074
2075     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2076     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2077
2078     switch (write_mask) {
2079         case WINED3DSP_WRITEMASK_0:
2080             shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
2081             break;
2082
2083         case WINED3DSP_WRITEMASK_1:
2084             shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
2085             break;
2086
2087         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2088             shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2089             break;
2090
2091         default:
2092             ERR("Write mask should be .x, .y or .xy\n");
2093             break;
2094     }
2095 }
2096
2097 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2098  * Start a for() loop where src1.y is the initial value of aL,
2099  *  increment aL by src1.z for a total of src1.x iterations.
2100  *  Need to use a temporary variable for this operation.
2101  */
2102 /* FIXME: I don't think nested loops will work correctly this way. */
2103 static void shader_glsl_loop(const SHADER_OPCODE_ARG *arg)
2104 {
2105     glsl_src_param_t src1_param;
2106     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2107     DWORD regtype = shader_get_regtype(arg->src[1]);
2108     DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
2109     const DWORD *control_values = NULL;
2110     const local_constant *constant;
2111
2112     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2113
2114     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2115      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2116      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2117      * addressing.
2118      */
2119     if(regtype == WINED3DSPR_CONSTINT) {
2120         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2121             if(constant->idx == reg) {
2122                 control_values = constant->value;
2123                 break;
2124             }
2125         }
2126     }
2127
2128     if(control_values) {
2129         if(control_values[2] > 0) {
2130             shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
2131                            shader->baseShader.cur_loop_depth, control_values[1],
2132                            shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2133                            shader->baseShader.cur_loop_depth, control_values[2]);
2134         } else if(control_values[2] == 0) {
2135             shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2136                            shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2137                            shader->baseShader.cur_loop_depth, control_values[0],
2138                            shader->baseShader.cur_loop_depth);
2139         } else {
2140             shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2141                            shader->baseShader.cur_loop_depth, control_values[1],
2142                            shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2143                            shader->baseShader.cur_loop_depth, control_values[2]);
2144         }
2145     } else {
2146         shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2147                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2148                        src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2149                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2150     }
2151
2152     shader->baseShader.cur_loop_depth++;
2153     shader->baseShader.cur_loop_regno++;
2154 }
2155
2156 static void shader_glsl_end(const SHADER_OPCODE_ARG *arg)
2157 {
2158     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2159
2160     shader_addline(arg->buffer, "}\n");
2161
2162     if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
2163         shader->baseShader.cur_loop_depth--;
2164         shader->baseShader.cur_loop_regno--;
2165     }
2166     if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
2167         shader->baseShader.cur_loop_depth--;
2168     }
2169 }
2170
2171 static void shader_glsl_rep(const SHADER_OPCODE_ARG *arg)
2172 {
2173     IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2174     glsl_src_param_t src0_param;
2175
2176     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2177     shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2178                    shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2179                    src0_param.param_str, shader->baseShader.cur_loop_depth);
2180     shader->baseShader.cur_loop_depth++;
2181 }
2182
2183 static void shader_glsl_if(const SHADER_OPCODE_ARG *arg)
2184 {
2185     glsl_src_param_t src0_param;
2186
2187     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2188     shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2189 }
2190
2191 static void shader_glsl_ifc(const SHADER_OPCODE_ARG *arg)
2192 {
2193     glsl_src_param_t src0_param;
2194     glsl_src_param_t src1_param;
2195
2196     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2197     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2198
2199     shader_addline(arg->buffer, "if (%s %s %s) {\n",
2200             src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2201 }
2202
2203 static void shader_glsl_else(const SHADER_OPCODE_ARG *arg)
2204 {
2205     shader_addline(arg->buffer, "} else {\n");
2206 }
2207
2208 static void shader_glsl_break(const SHADER_OPCODE_ARG *arg)
2209 {
2210     shader_addline(arg->buffer, "break;\n");
2211 }
2212
2213 /* FIXME: According to MSDN the compare is done per component. */
2214 static void shader_glsl_breakc(const SHADER_OPCODE_ARG *arg)
2215 {
2216     glsl_src_param_t src0_param;
2217     glsl_src_param_t src1_param;
2218
2219     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2220     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2221
2222     shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2223             src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2224 }
2225
2226 static void shader_glsl_label(const SHADER_OPCODE_ARG *arg)
2227 {
2228
2229     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2230     shader_addline(arg->buffer, "}\n");
2231     shader_addline(arg->buffer, "void subroutine%u () {\n",  snum);
2232 }
2233
2234 static void shader_glsl_call(const SHADER_OPCODE_ARG *arg)
2235 {
2236     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2237     shader_addline(arg->buffer, "subroutine%u();\n", snum);
2238 }
2239
2240 static void shader_glsl_callnz(const SHADER_OPCODE_ARG *arg)
2241 {
2242     glsl_src_param_t src1_param;
2243
2244     DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2245     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2246     shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2247 }
2248
2249 /*********************************************
2250  * Pixel Shader Specific Code begins here
2251  ********************************************/
2252 static void pshader_glsl_tex(const SHADER_OPCODE_ARG *arg)
2253 {
2254     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2255     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2256     DWORD shader_version = arg->reg_maps->shader_version;
2257     char dst_swizzle[6];
2258     glsl_sample_function_t sample_function;
2259     DWORD sampler_type;
2260     DWORD sampler_idx;
2261     BOOL projected, texrect = FALSE;
2262     DWORD mask = 0;
2263
2264     /* All versions have a destination register */
2265     shader_glsl_append_dst(arg->buffer, arg);
2266
2267     /* 1.0-1.4: Use destination register as sampler source.
2268      * 2.0+: Use provided sampler source. */
2269     if (shader_version < WINED3DPS_VERSION(2,0)) sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2270     else sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2271     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2272
2273     if (shader_version < WINED3DPS_VERSION(1,4))
2274     {
2275         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2276
2277         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2278         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2279             projected = TRUE;
2280             switch (flags & ~WINED3DTTFF_PROJECTED) {
2281                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2282                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2283                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2284                 case WINED3DTTFF_COUNT4:
2285                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2286             }
2287         } else {
2288             projected = FALSE;
2289         }
2290     }
2291     else if (shader_version < WINED3DPS_VERSION(2,0))
2292     {
2293         DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2294
2295         if (src_mod == WINED3DSPSM_DZ) {
2296             projected = TRUE;
2297             mask = WINED3DSP_WRITEMASK_2;
2298         } else if (src_mod == WINED3DSPSM_DW) {
2299             projected = TRUE;
2300             mask = WINED3DSP_WRITEMASK_3;
2301         } else {
2302             projected = FALSE;
2303         }
2304     } else {
2305         if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2306                 /* ps 2.0 texldp instruction always divides by the fourth component. */
2307                 projected = TRUE;
2308                 mask = WINED3DSP_WRITEMASK_3;
2309         } else {
2310             projected = FALSE;
2311         }
2312     }
2313
2314     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2315        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2316         texrect = TRUE;
2317     }
2318
2319     shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2320     mask |= sample_function.coord_mask;
2321
2322     if (shader_version < WINED3DPS_VERSION(2,0)) shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2323     else shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2324
2325     /* 1.0-1.3: Use destination register as coordinate source.
2326        1.4+: Use provided coordinate source register. */
2327     if (shader_version < WINED3DPS_VERSION(1,4))
2328     {
2329         char coord_mask[6];
2330         shader_glsl_get_write_mask(mask, coord_mask);
2331         shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2332                 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2333     } else {
2334         glsl_src_param_t coord_param;
2335         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2336         if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2337             glsl_src_param_t bias;
2338             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2339
2340             shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2341                     sample_function.name, sampler_idx, coord_param.param_str,
2342                     bias.param_str, dst_swizzle);
2343         } else {
2344             shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2345                     sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2346         }
2347     }
2348 }
2349
2350 static void shader_glsl_texldl(const SHADER_OPCODE_ARG *arg)
2351 {
2352     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2353     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2354     glsl_sample_function_t sample_function;
2355     glsl_src_param_t coord_param, lod_param;
2356     char dst_swizzle[6];
2357     DWORD sampler_type;
2358     DWORD sampler_idx;
2359     BOOL texrect = FALSE;
2360
2361     shader_glsl_append_dst(arg->buffer, arg);
2362     shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2363
2364     sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2365     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2366     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2367        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2368         texrect = TRUE;
2369     }
2370     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);
2371
2372     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2373
2374     if (shader_is_pshader_version(arg->reg_maps->shader_version))
2375     {
2376         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2377          * However, they seem to work just fine in fragment shaders as well. */
2378         WARN("Using %sLod in fragment shader.\n", sample_function.name);
2379         shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2380                 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2381     } else {
2382         shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2383                 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2384     }
2385 }
2386
2387 static void pshader_glsl_texcoord(const SHADER_OPCODE_ARG *arg)
2388 {
2389     /* FIXME: Make this work for more than just 2D textures */
2390     SHADER_BUFFER* buffer = arg->buffer;
2391     DWORD write_mask;
2392     char dst_mask[6];
2393
2394     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2395     shader_glsl_get_write_mask(write_mask, dst_mask);
2396
2397     if (arg->reg_maps->shader_version != WINED3DPS_VERSION(1,4))
2398     {
2399         DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2400         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2401     } else {
2402         DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2403         DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2404         char dst_swizzle[6];
2405
2406         shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2407
2408         if (src_mod == WINED3DSPSM_DZ) {
2409             glsl_src_param_t div_param;
2410             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2411             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2412
2413             if (mask_size > 1) {
2414                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2415             } else {
2416                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2417             }
2418         } else if (src_mod == WINED3DSPSM_DW) {
2419             glsl_src_param_t div_param;
2420             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2421             shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2422
2423             if (mask_size > 1) {
2424                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2425             } else {
2426                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2427             }
2428         } else {
2429             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2430         }
2431     }
2432 }
2433
2434 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2435  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2436  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2437 static void pshader_glsl_texdp3tex(const SHADER_OPCODE_ARG *arg)
2438 {
2439     glsl_src_param_t src0_param;
2440     char dst_mask[6];
2441     glsl_sample_function_t sample_function;
2442     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2443     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2444     DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2445
2446     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2447
2448     shader_glsl_append_dst(arg->buffer, arg);
2449     shader_glsl_get_write_mask(arg->dst, dst_mask);
2450
2451     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2452      * scalar, and projected sampling would require 4.
2453      *
2454      * It is a dependent read - not valid with conditional NP2 textures
2455      */
2456     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2457
2458     switch(count_bits(sample_function.coord_mask)) {
2459         case 1:
2460             shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2461                            sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2462             break;
2463
2464         case 2:
2465             shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2466                           sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2467             break;
2468
2469         case 3:
2470             shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2471                            sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2472             break;
2473         default:
2474             FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2475     }
2476 }
2477
2478 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2479  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2480 static void pshader_glsl_texdp3(const SHADER_OPCODE_ARG *arg)
2481 {
2482     glsl_src_param_t src0_param;
2483     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2484     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2485     DWORD dst_mask;
2486     unsigned int mask_size;
2487
2488     dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2489     mask_size = shader_glsl_get_write_mask_size(dst_mask);
2490     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2491
2492     if (mask_size > 1) {
2493         shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2494     } else {
2495         shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2496     }
2497 }
2498
2499 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2500  * Calculate the depth as dst.x / dst.y   */
2501 static void pshader_glsl_texdepth(const SHADER_OPCODE_ARG *arg)
2502 {
2503     glsl_dst_param_t dst_param;
2504
2505     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2506
2507     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2508      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2509      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2510      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2511      * >= 1.0 or < 0.0
2512      */
2513     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);
2514 }
2515
2516 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2517  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2518  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2519  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2520  */
2521 static void pshader_glsl_texm3x2depth(const SHADER_OPCODE_ARG *arg)
2522 {
2523     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2524     DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2525     glsl_src_param_t src0_param;
2526
2527     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2528
2529     shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2530     shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2531 }
2532
2533 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2534  * Calculate the 1st of a 2-row matrix multiplication. */
2535 static void pshader_glsl_texm3x2pad(const SHADER_OPCODE_ARG *arg)
2536 {
2537     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2538     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2539     SHADER_BUFFER* buffer = arg->buffer;
2540     glsl_src_param_t src0_param;
2541
2542     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2543     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2544 }
2545
2546 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2547  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2548 static void pshader_glsl_texm3x3pad(const SHADER_OPCODE_ARG* arg)
2549 {
2550     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2551     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2552     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2553     SHADER_BUFFER* buffer = arg->buffer;
2554     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2555     glsl_src_param_t src0_param;
2556
2557     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2558     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2559     current_state->texcoord_w[current_state->current_row++] = reg;
2560 }
2561
2562 static void pshader_glsl_texm3x2tex(const SHADER_OPCODE_ARG *arg)
2563 {
2564     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2565     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2566     SHADER_BUFFER* buffer = arg->buffer;
2567     glsl_src_param_t src0_param;
2568     char dst_mask[6];
2569
2570     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2571     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2572
2573     shader_glsl_append_dst(buffer, arg);
2574     shader_glsl_get_write_mask(arg->dst, dst_mask);
2575
2576     /* Sample the texture using the calculated coordinates */
2577     shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2578 }
2579
2580 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2581  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2582 static void pshader_glsl_texm3x3tex(const SHADER_OPCODE_ARG *arg)
2583 {
2584     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2585     glsl_src_param_t src0_param;
2586     char dst_mask[6];
2587     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2588     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2589     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2590     DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2591     glsl_sample_function_t sample_function;
2592
2593     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2594     shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2595
2596     shader_glsl_append_dst(arg->buffer, arg);
2597     shader_glsl_get_write_mask(arg->dst, dst_mask);
2598     /* Dependent read, not valid with conditional NP2 */
2599     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2600
2601     /* Sample the texture using the calculated coordinates */
2602     shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2603
2604     current_state->current_row = 0;
2605 }
2606
2607 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2608  * Perform the 3rd row of a 3x3 matrix multiply */
2609 static void pshader_glsl_texm3x3(const SHADER_OPCODE_ARG *arg)
2610 {
2611     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2612     glsl_src_param_t src0_param;
2613     char dst_mask[6];
2614     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2615     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2616     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2617
2618     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2619
2620     shader_glsl_append_dst(arg->buffer, arg);
2621     shader_glsl_get_write_mask(arg->dst, dst_mask);
2622     shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2623
2624     current_state->current_row = 0;
2625 }
2626
2627 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2628  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2629 static void pshader_glsl_texm3x3spec(const SHADER_OPCODE_ARG *arg)
2630 {
2631     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2632     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2633     glsl_src_param_t src0_param;
2634     glsl_src_param_t src1_param;
2635     char dst_mask[6];
2636     SHADER_BUFFER* buffer = arg->buffer;
2637     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2638     DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2639     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2640     glsl_sample_function_t sample_function;
2641
2642     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2643     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2644
2645     /* Perform the last matrix multiply operation */
2646     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2647     /* Reflection calculation */
2648     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2649
2650     shader_glsl_append_dst(buffer, arg);
2651     shader_glsl_get_write_mask(arg->dst, dst_mask);
2652     /* Dependent read, not valid with conditional NP2 */
2653     shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2654
2655     /* Sample the texture */
2656     shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2657
2658     current_state->current_row = 0;
2659 }
2660
2661 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2662  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2663 static void pshader_glsl_texm3x3vspec(const SHADER_OPCODE_ARG *arg)
2664 {
2665     IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2666     DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2667     SHADER_BUFFER* buffer = arg->buffer;
2668     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2669     glsl_src_param_t src0_param;
2670     char dst_mask[6];
2671     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2672     DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2673     glsl_sample_function_t sample_function;
2674
2675     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2676
2677     /* Perform the last matrix multiply operation */
2678     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2679
2680     /* Construct the eye-ray vector from w coordinates */
2681     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2682             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2683     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2684
2685     shader_glsl_append_dst(buffer, arg);
2686     shader_glsl_get_write_mask(arg->dst, dst_mask);
2687     /* Dependent read, not valid with conditional NP2 */
2688     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2689
2690     /* Sample the texture using the calculated coordinates */
2691     shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2692
2693     current_state->current_row = 0;
2694 }
2695
2696 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2697  * Apply a fake bump map transform.
2698  * texbem is pshader <= 1.3 only, this saves a few version checks
2699  */
2700 static void pshader_glsl_texbem(const SHADER_OPCODE_ARG *arg)
2701 {
2702     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2703     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2704     char dst_swizzle[6];
2705     glsl_sample_function_t sample_function;
2706     glsl_src_param_t coord_param;
2707     DWORD sampler_type;
2708     DWORD sampler_idx;
2709     DWORD mask;
2710     DWORD flags;
2711     char coord_mask[6];
2712
2713     sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2714     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2715
2716     sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2717     /* Dependent read, not valid with conditional NP2 */
2718     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2719     mask = sample_function.coord_mask;
2720
2721     shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2722
2723     shader_glsl_get_write_mask(mask, coord_mask);
2724
2725     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2726          * so we can't let the GL handle this.
2727          */
2728     if (flags & WINED3DTTFF_PROJECTED) {
2729         DWORD div_mask=0;
2730         char coord_div_mask[3];
2731         switch (flags & ~WINED3DTTFF_PROJECTED) {
2732             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2733             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2734             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2735             case WINED3DTTFF_COUNT4:
2736             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2737         }
2738         shader_glsl_get_write_mask(div_mask, coord_div_mask);
2739         shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2740     }
2741
2742     shader_glsl_append_dst(arg->buffer, arg);
2743     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2744     if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2745         glsl_src_param_t luminance_param;
2746         shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2747         shader_addline(arg->buffer, "(%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )*(%s * luminancescale%d + luminanceoffset%d))%s);\n",
2748                        sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask,
2749                        luminance_param.param_str, sampler_idx, sampler_idx, dst_swizzle);
2750     } else {
2751         shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )%s);\n",
2752                        sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask, dst_swizzle);
2753     }
2754 }
2755
2756 static void pshader_glsl_bem(const SHADER_OPCODE_ARG *arg)
2757 {
2758     glsl_src_param_t src0_param, src1_param;
2759     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2760
2761     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2762     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2763
2764     shader_glsl_append_dst(arg->buffer, arg);
2765     shader_addline(arg->buffer, "%s + bumpenvmat%d * %s);\n",
2766                    src0_param.param_str, sampler_idx, src1_param.param_str);
2767 }
2768
2769 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2770  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2771 static void pshader_glsl_texreg2ar(const SHADER_OPCODE_ARG *arg)
2772 {
2773     glsl_src_param_t src0_param;
2774     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2775     char dst_mask[6];
2776
2777     shader_glsl_append_dst(arg->buffer, arg);
2778     shader_glsl_get_write_mask(arg->dst, dst_mask);
2779     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2780
2781     shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2782 }
2783
2784 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2785  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2786 static void pshader_glsl_texreg2gb(const SHADER_OPCODE_ARG *arg)
2787 {
2788     glsl_src_param_t src0_param;
2789     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2790     char dst_mask[6];
2791
2792     shader_glsl_append_dst(arg->buffer, arg);
2793     shader_glsl_get_write_mask(arg->dst, dst_mask);
2794     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2795
2796     shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2797 }
2798
2799 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2800  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2801 static void pshader_glsl_texreg2rgb(const SHADER_OPCODE_ARG *arg)
2802 {
2803     glsl_src_param_t src0_param;
2804     char dst_mask[6];
2805     DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2806     DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2807     glsl_sample_function_t sample_function;
2808
2809     shader_glsl_append_dst(arg->buffer, arg);
2810     shader_glsl_get_write_mask(arg->dst, dst_mask);
2811     /* Dependent read, not valid with conditional NP2 */
2812     shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2813     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2814
2815     shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2816 }
2817
2818 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2819  * If any of the first 3 components are < 0, discard this pixel */
2820 static void pshader_glsl_texkill(const SHADER_OPCODE_ARG *arg)
2821 {
2822     glsl_dst_param_t dst_param;
2823
2824     /* The argument is a destination parameter, and no writemasks are allowed */
2825     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2826     if ((arg->reg_maps->shader_version >= WINED3DPS_VERSION(2,0)))
2827     {
2828         /* 2.0 shaders compare all 4 components in texkill */
2829         shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2830     } else {
2831         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2832          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2833          * 4 components are defined, only the first 3 are used
2834          */
2835         shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2836     }
2837 }
2838
2839 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2840  * dst = dot2(src0, src1) + src2 */
2841 static void pshader_glsl_dp2add(const SHADER_OPCODE_ARG *arg)
2842 {
2843     glsl_src_param_t src0_param;
2844     glsl_src_param_t src1_param;
2845     glsl_src_param_t src2_param;
2846     DWORD write_mask;
2847     unsigned int mask_size;
2848
2849     write_mask = shader_glsl_append_dst(arg->buffer, arg);
2850     mask_size = shader_glsl_get_write_mask_size(write_mask);
2851
2852     shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2853     shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2854     shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2855
2856     if (mask_size > 1) {
2857         shader_addline(arg->buffer, "vec%d(dot(%s, %s) + %s));\n", mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
2858     } else {
2859         shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2860     }
2861 }
2862
2863 static void pshader_glsl_input_pack(SHADER_BUFFER* buffer, const struct semantic* semantics_in,
2864         IWineD3DPixelShader *iface, enum vertexprocessing_mode vertexprocessing)
2865 {
2866    unsigned int i;
2867    IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2868
2869    for (i = 0; i < MAX_REG_INPUT; i++) {
2870
2871        DWORD usage_token = semantics_in[i].usage;
2872        DWORD register_token = semantics_in[i].reg;
2873        DWORD usage, usage_idx;
2874        char reg_mask[6];
2875
2876        /* Uninitialized */
2877        if (!usage_token) continue;
2878        usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2879        usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2880        shader_glsl_get_write_mask(register_token, reg_mask);
2881
2882        switch(usage) {
2883
2884            case WINED3DDECLUSAGE_TEXCOORD:
2885                if(usage_idx < 8 && vertexprocessing == pretransformed) {
2886                    shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2887                                   This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2888                } else {
2889                    shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2890                                   This->input_reg_map[i], reg_mask, reg_mask);
2891                }
2892                break;
2893
2894            case WINED3DDECLUSAGE_COLOR:
2895                if (usage_idx == 0)
2896                    shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2897                        This->input_reg_map[i], reg_mask, reg_mask);
2898                else if (usage_idx == 1)
2899                    shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2900                        This->input_reg_map[i], reg_mask, reg_mask);
2901                else
2902                    shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2903                        This->input_reg_map[i], reg_mask, reg_mask);
2904                break;
2905
2906            default:
2907                shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2908                    This->input_reg_map[i], reg_mask, reg_mask);
2909         }
2910     }
2911 }
2912
2913 /*********************************************
2914  * Vertex Shader Specific Code begins here
2915  ********************************************/
2916
2917 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
2918     glsl_program_key_t *key;
2919
2920     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2921     key->vshader = entry->vshader;
2922     key->pshader = entry->pshader;
2923     key->vs_args = entry->vs_args;
2924     key->ps_args = entry->ps_args;
2925
2926     hash_table_put(priv->glsl_program_lookup, key, entry);
2927 }
2928
2929 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
2930         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
2931         struct ps_compile_args *ps_args) {
2932     glsl_program_key_t key;
2933
2934     key.vshader = vshader;
2935     key.pshader = pshader;
2936     key.vs_args = *vs_args;
2937     key.ps_args = *ps_args;
2938
2939     return hash_table_get(priv->glsl_program_lookup, &key);
2940 }
2941
2942 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const WineD3D_GL_Info *gl_info,
2943         struct glsl_shader_prog_link *entry)
2944 {
2945     glsl_program_key_t *key;
2946
2947     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2948     key->vshader = entry->vshader;
2949     key->pshader = entry->pshader;
2950     key->vs_args = entry->vs_args;
2951     key->ps_args = entry->ps_args;
2952     hash_table_remove(priv->glsl_program_lookup, key);
2953
2954     GL_EXTCALL(glDeleteObjectARB(entry->programId));
2955     if (entry->vshader) list_remove(&entry->vshader_entry);
2956     if (entry->pshader) list_remove(&entry->pshader_entry);
2957     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2958     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2959     HeapFree(GetProcessHeap(), 0, entry);
2960 }
2961
2962 static void handle_ps3_input(SHADER_BUFFER *buffer, const struct semantic *semantics_in,
2963         const struct semantic *semantics_out, const WineD3D_GL_Info *gl_info, const DWORD *map)
2964 {
2965     unsigned int i, j;
2966     DWORD usage_token, usage_token_out;
2967     DWORD register_token, register_token_out;
2968     DWORD usage, usage_idx, usage_out, usage_idx_out;
2969     DWORD *set;
2970     DWORD in_idx;
2971     DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
2972     char reg_mask[6], reg_mask_out[6];
2973     char destination[50];
2974
2975     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
2976
2977     if (!semantics_out) {
2978         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
2979         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
2980         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
2981     }
2982
2983     for(i = 0; i < MAX_REG_INPUT; i++) {
2984         usage_token = semantics_in[i].usage;
2985         if (!usage_token) continue;
2986
2987         in_idx = map[i];
2988         if (in_idx >= (in_count + 2)) {
2989             FIXME("More input varyings declared than supported, expect issues\n");
2990             continue;
2991         } else if(map[i] == -1) {
2992             /* Declared, but not read register */
2993             continue;
2994         }
2995
2996         if (in_idx == in_count) {
2997             sprintf(destination, "gl_FrontColor");
2998         } else if (in_idx == in_count + 1) {
2999             sprintf(destination, "gl_FrontSecondaryColor");
3000         } else {
3001             sprintf(destination, "IN[%u]", in_idx);
3002         }
3003
3004         register_token = semantics_in[i].reg;
3005
3006         usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3007         usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3008         set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
3009
3010         if(!semantics_out) {
3011             switch(usage) {
3012                 case WINED3DDECLUSAGE_COLOR:
3013                     if (usage_idx == 0)
3014                         shader_addline(buffer, "%s%s = front_color%s;\n",
3015                                        destination, reg_mask, reg_mask);
3016                     else if (usage_idx == 1)
3017                         shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3018                                        destination, reg_mask, reg_mask);
3019                     else
3020                         shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3021                                        destination, reg_mask, reg_mask);
3022                     break;
3023
3024                 case WINED3DDECLUSAGE_TEXCOORD:
3025                     if (usage_idx < 8) {
3026                         shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3027                                        destination, reg_mask, usage_idx, reg_mask);
3028                     } else {
3029                         shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3030                                        destination, reg_mask, reg_mask);
3031                     }
3032                     break;
3033
3034                 case WINED3DDECLUSAGE_FOG:
3035                     shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3036                                    destination, reg_mask, reg_mask);
3037                     break;
3038
3039                 default:
3040                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3041                                    destination, reg_mask, reg_mask);
3042             }
3043         } else {
3044             BOOL found = FALSE;
3045             for(j = 0; j < MAX_REG_OUTPUT; j++) {
3046                 usage_token_out = semantics_out[j].usage;
3047                 if (!usage_token_out) continue;
3048                 register_token_out = semantics_out[j].reg;
3049
3050                 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3051                 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3052                 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
3053
3054                 if(usage == usage_out &&
3055                    usage_idx == usage_idx_out) {
3056                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3057                                    destination, reg_mask, j, reg_mask);
3058                     found = TRUE;
3059                 }
3060             }
3061             if(!found) {
3062                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3063                                destination, reg_mask, reg_mask);
3064             }
3065         }
3066     }
3067
3068     /* This is solely to make the compiler / linker happy and avoid warning about undefined
3069      * varyings. It shouldn't result in any real code executed on the GPU, since all read
3070      * input varyings are assigned above, if the optimizer works properly.
3071      */
3072     for(i = 0; i < in_count + 2; i++) {
3073         if(set[i] != WINED3DSP_WRITEMASK_ALL) {
3074             unsigned int size = 0;
3075             memset(reg_mask, 0, sizeof(reg_mask));
3076             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3077                 reg_mask[size] = 'x';
3078                 size++;
3079             }
3080             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3081                 reg_mask[size] = 'y';
3082                 size++;
3083             }
3084             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3085                 reg_mask[size] = 'z';
3086                 size++;
3087             }
3088             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3089                 reg_mask[size] = 'w';
3090                 size++;
3091             }
3092
3093             if (i == in_count) {
3094                 sprintf(destination, "gl_FrontColor");
3095             } else if (i == in_count + 1) {
3096                 sprintf(destination, "gl_FrontSecondaryColor");
3097             } else {
3098                 sprintf(destination, "IN[%u]", i);
3099             }
3100
3101             if (size == 1) {
3102                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3103             } else {
3104                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3105             }
3106         }
3107     }
3108
3109     HeapFree(GetProcessHeap(), 0, set);
3110 }
3111
3112 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
3113         IWineD3DPixelShader *pixelshader, const WineD3D_GL_Info *gl_info)
3114 {
3115     GLhandleARB ret = 0;
3116     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3117     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3118     IWineD3DDeviceImpl *device;
3119     DWORD vs_major = WINED3DSHADER_VERSION_MAJOR(vs->baseShader.reg_maps.shader_version);
3120     DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.reg_maps.shader_version) : 0;
3121     unsigned int i;
3122     SHADER_BUFFER buffer;
3123     DWORD usage_token;
3124     DWORD register_token;
3125     DWORD usage, usage_idx, writemask;
3126     char reg_mask[6];
3127     const struct semantic *semantics_out, *semantics_in;
3128
3129     shader_buffer_init(&buffer);
3130
3131     shader_addline(&buffer, "#version 120\n");
3132
3133     if(vs_major < 3 && ps_major < 3) {
3134         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3135          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3136          */
3137         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3138         if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0 &&
3139             !device->frag_pipe->ffp_proj_control) {
3140             shader_addline(&buffer, "void order_ps_input() {\n");
3141             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3142                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3143                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3144                     shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3145                 }
3146             }
3147             shader_addline(&buffer, "}\n");
3148         } else {
3149             shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
3150         }
3151     } else if(ps_major < 3 && vs_major >= 3) {
3152         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3153         semantics_out = vs->semantics_out;
3154
3155         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3156         for(i = 0; i < MAX_REG_OUTPUT; i++) {
3157             usage_token = semantics_out[i].usage;
3158             if (!usage_token) continue;
3159             register_token = semantics_out[i].reg;
3160
3161             usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3162             usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3163             writemask = shader_glsl_get_write_mask(register_token, reg_mask);
3164
3165             switch(usage) {
3166                 case WINED3DDECLUSAGE_COLOR:
3167                     if (usage_idx == 0)
3168                         shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3169                     else if (usage_idx == 1)
3170                         shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3171                     break;
3172
3173                 case WINED3DDECLUSAGE_POSITION:
3174                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3175                     break;
3176
3177                 case WINED3DDECLUSAGE_TEXCOORD:
3178                     if (usage_idx < 8) {
3179                         if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
3180
3181                         shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3182                                         usage_idx, reg_mask, i, reg_mask);
3183                         if(!(writemask & WINED3DSP_WRITEMASK_3)) {
3184                             shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
3185                         }
3186                     }
3187                     break;
3188
3189                 case WINED3DDECLUSAGE_PSIZE:
3190                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3191                     break;
3192
3193                 case WINED3DDECLUSAGE_FOG:
3194                     shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3195                     break;
3196
3197                 default:
3198                     break;
3199             }
3200         }
3201         shader_addline(&buffer, "}\n");
3202
3203     } else if(ps_major >= 3 && vs_major >= 3) {
3204         semantics_out = vs->semantics_out;
3205         semantics_in = ps->semantics_in;
3206
3207         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3208         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3209         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3210
3211         /* First, sort out position and point size. Those are not passed to the pixel shader */
3212         for(i = 0; i < MAX_REG_OUTPUT; i++) {
3213             usage_token = semantics_out[i].usage;
3214             if (!usage_token) continue;
3215             register_token = semantics_out[i].reg;
3216
3217             usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3218             usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3219             shader_glsl_get_write_mask(register_token, reg_mask);
3220
3221             switch(usage) {
3222                 case WINED3DDECLUSAGE_POSITION:
3223                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3224                     break;
3225
3226                 case WINED3DDECLUSAGE_PSIZE:
3227                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3228                     break;
3229
3230                 default:
3231                     break;
3232             }
3233         }
3234
3235         /* Then, fix the pixel shader input */
3236         handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3237
3238         shader_addline(&buffer, "}\n");
3239     } else if(ps_major >= 3 && vs_major < 3) {
3240         semantics_in = ps->semantics_in;
3241
3242         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3243         shader_addline(&buffer, "void order_ps_input() {\n");
3244         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3245          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3246          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3247          */
3248         handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3249         shader_addline(&buffer, "}\n");
3250     } else {
3251         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3252     }
3253
3254     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3255     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3256     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3257     checkGLcall("glShaderSourceARB(ret, 1, &buffer.buffer, NULL)");
3258     GL_EXTCALL(glCompileShaderARB(ret));
3259     checkGLcall("glCompileShaderARB(ret)");
3260
3261     shader_buffer_free(&buffer);
3262     return ret;
3263 }
3264
3265 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const WineD3D_GL_Info *gl_info,
3266         GLhandleARB programId, char prefix)
3267 {
3268     const local_constant *lconst;
3269     GLuint tmp_loc;
3270     const float *value;
3271     char glsl_name[8];
3272
3273     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3274         value = (const float *)lconst->value;
3275         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3276         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3277         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3278     }
3279     checkGLcall("Hardcoding local constants\n");
3280 }
3281
3282 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3283  * It sets the programId on the current StateBlock (because it should be called
3284  * inside of the DrawPrimitive() part of the render loop).
3285  *
3286  * If a program for the given combination does not exist, create one, and store
3287  * the program in the hash table.  If it creates a program, it will link the
3288  * given objects, too.
3289  */
3290 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3291     IWineD3DDeviceImpl *This               = (IWineD3DDeviceImpl *)iface;
3292     struct shader_glsl_priv *priv          = This->shader_priv;
3293     const WineD3D_GL_Info *gl_info         = &This->adapter->gl_info;
3294     IWineD3DPixelShader  *pshader          = This->stateBlock->pixelShader;
3295     IWineD3DVertexShader *vshader          = This->stateBlock->vertexShader;
3296     struct glsl_shader_prog_link *entry    = NULL;
3297     GLhandleARB programId                  = 0;
3298     GLhandleARB reorder_shader_id          = 0;
3299     int i;
3300     char glsl_name[8];
3301     GLhandleARB vshader_id, pshader_id;
3302     struct ps_compile_args ps_compile_args;
3303     struct vs_compile_args vs_compile_args;
3304
3305     if(use_vs) {
3306         find_vs_compile_args((IWineD3DVertexShaderImpl*)This->stateBlock->vertexShader, This->stateBlock, &vs_compile_args);
3307     } else {
3308         /* FIXME: Do we really have to spend CPU cycles to generate a few zeroed bytes? */
3309         memset(&vs_compile_args, 0, sizeof(vs_compile_args));
3310     }
3311     if(use_ps) {
3312         find_ps_compile_args((IWineD3DPixelShaderImpl*)This->stateBlock->pixelShader, This->stateBlock, &ps_compile_args);
3313     } else {
3314         /* FIXME: Do we really have to spend CPU cycles to generate a few zeroed bytes? */
3315         memset(&ps_compile_args, 0, sizeof(ps_compile_args));
3316     }
3317     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
3318     if (entry) {
3319         priv->glsl_program = entry;
3320         return;
3321     }
3322
3323     /* If we get to this point, then no matching program exists, so we create one */
3324     programId = GL_EXTCALL(glCreateProgramObjectARB());
3325     TRACE("Created new GLSL shader program %u\n", programId);
3326
3327     /* Create the entry */
3328     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3329     entry->programId = programId;
3330     entry->vshader = vshader;
3331     entry->pshader = pshader;
3332     entry->vs_args = vs_compile_args;
3333     entry->ps_args = ps_compile_args;
3334     entry->constant_version = 0;
3335     /* Add the hash table entry */
3336     add_glsl_program_entry(priv, entry);
3337
3338     /* Set the current program */
3339     priv->glsl_program = entry;
3340
3341     if(use_vs) {
3342         vshader_id = find_gl_vshader((IWineD3DVertexShaderImpl *) vshader, &vs_compile_args);
3343     } else {
3344         vshader_id = 0;
3345     }
3346
3347     /* Attach GLSL vshader */
3348     if (vshader_id) {
3349         int max_attribs = 16;   /* TODO: Will this always be the case? It is at the moment... */
3350         char tmp_name[10];
3351
3352         reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3353         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3354         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3355         checkGLcall("glAttachObjectARB");
3356         /* Flag the reorder function for deletion, then it will be freed automatically when the program
3357          * is destroyed
3358          */
3359         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3360
3361         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3362         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3363         checkGLcall("glAttachObjectARB");
3364
3365         /* Bind vertex attributes to a corresponding index number to match
3366          * the same index numbers as ARB_vertex_programs (makes loading
3367          * vertex attributes simpler).  With this method, we can use the
3368          * exact same code to load the attributes later for both ARB and
3369          * GLSL shaders.
3370          *
3371          * We have to do this here because we need to know the Program ID
3372          * in order to make the bindings work, and it has to be done prior
3373          * to linking the GLSL program. */
3374         for (i = 0; i < max_attribs; ++i) {
3375             if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3376                 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3377                 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3378             }
3379         }
3380         checkGLcall("glBindAttribLocationARB");
3381
3382         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3383     }
3384
3385     if(use_ps) {
3386         pshader_id = find_gl_pshader((IWineD3DPixelShaderImpl *) pshader, &ps_compile_args);
3387     } else {
3388         pshader_id = 0;
3389     }
3390
3391     /* Attach GLSL pshader */
3392     if (pshader_id) {
3393         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3394         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3395         checkGLcall("glAttachObjectARB");
3396
3397         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3398     }
3399
3400     /* Link the program */
3401     TRACE("Linking GLSL shader program %u\n", programId);
3402     GL_EXTCALL(glLinkProgramARB(programId));
3403     print_glsl_info_log(&GLINFO_LOCATION, programId);
3404
3405     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3406     for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3407         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3408         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3409     }
3410     for (i = 0; i < MAX_CONST_I; ++i) {
3411         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3412         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3413     }
3414     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3415     for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3416         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3417         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3418     }
3419     for (i = 0; i < MAX_CONST_I; ++i) {
3420         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3421         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3422     }
3423
3424     if(pshader) {
3425         for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3426             char name[32];
3427             sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3428             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3429             sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3430             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3431             sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3432             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3433         }
3434     }
3435
3436
3437     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3438     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3439     checkGLcall("Find glsl program uniform locations");
3440
3441     if (pshader
3442             && WINED3DSHADER_VERSION_MAJOR(((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version) >= 3
3443             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > GL_LIMITS(glsl_varyings) / 4)
3444     {
3445         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
3446         entry->vertex_color_clamp = GL_FALSE;
3447     } else {
3448         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
3449     }
3450
3451     /* Set the shader to allow uniform loading on it */
3452     GL_EXTCALL(glUseProgramObjectARB(programId));
3453     checkGLcall("glUseProgramObjectARB(programId)");
3454
3455     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3456      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3457      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3458      * vertex shader with fixed function pixel processing is used we make sure that the card
3459      * supports enough samplers to allow the max number of vertex samplers with all possible
3460      * fixed function fragment processing setups. So once the program is linked these samplers
3461      * won't change.
3462      */
3463     if(vshader_id) {
3464         /* Load vertex shader samplers */
3465         shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3466     }
3467     if(pshader_id) {
3468         /* Load pixel shader samplers */
3469         shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3470     }
3471
3472     /* If the local constants do not have to be loaded with the environment constants,
3473      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
3474      * later
3475      */
3476     if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
3477         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
3478     }
3479     if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
3480         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
3481     }
3482 }
3483
3484 static GLhandleARB create_glsl_blt_shader(const WineD3D_GL_Info *gl_info, enum tex_types tex_type)
3485 {
3486     GLhandleARB program_id;
3487     GLhandleARB vshader_id, pshader_id;
3488     static const char *blt_vshader[] =
3489     {
3490         "#version 120\n"
3491         "void main(void)\n"
3492         "{\n"
3493         "    gl_Position = gl_Vertex;\n"
3494         "    gl_FrontColor = vec4(1.0);\n"
3495         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
3496         "}\n"
3497     };
3498
3499     static const char *blt_pshaders[tex_type_count] =
3500     {
3501         /* tex_1d */
3502         NULL,
3503         /* tex_2d */
3504         "#version 120\n"
3505         "uniform sampler2D sampler;\n"
3506         "void main(void)\n"
3507         "{\n"
3508         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3509         "}\n",
3510         /* tex_3d */
3511         NULL,
3512         /* tex_cube */
3513         "#version 120\n"
3514         "uniform samplerCube sampler;\n"
3515         "void main(void)\n"
3516         "{\n"
3517         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
3518         "}\n",
3519         /* tex_rect */
3520         "#version 120\n"
3521         "#extension GL_ARB_texture_rectangle : enable\n"
3522         "uniform sampler2DRect sampler;\n"
3523         "void main(void)\n"
3524         "{\n"
3525         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
3526         "}\n",
3527     };
3528
3529     if (!blt_pshaders[tex_type])
3530     {
3531         FIXME("tex_type %#x not supported\n", tex_type);
3532         tex_type = tex_2d;
3533     }
3534
3535     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3536     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3537     GL_EXTCALL(glCompileShaderARB(vshader_id));
3538
3539     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3540     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
3541     GL_EXTCALL(glCompileShaderARB(pshader_id));
3542
3543     program_id = GL_EXTCALL(glCreateProgramObjectARB());
3544     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3545     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3546     GL_EXTCALL(glLinkProgramARB(program_id));
3547
3548     print_glsl_info_log(&GLINFO_LOCATION, program_id);
3549
3550     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3551      * is destroyed
3552      */
3553     GL_EXTCALL(glDeleteObjectARB(vshader_id));
3554     GL_EXTCALL(glDeleteObjectARB(pshader_id));
3555     return program_id;
3556 }
3557
3558 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3559     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3560     struct shader_glsl_priv *priv = This->shader_priv;
3561     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3562     GLhandleARB program_id = 0;
3563     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
3564
3565     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3566
3567     if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3568     else priv->glsl_program = NULL;
3569
3570     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3571
3572     if (old_vertex_color_clamp != current_vertex_color_clamp) {
3573         if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
3574             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
3575             checkGLcall("glClampColorARB");
3576         } else {
3577             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
3578         }
3579     }
3580
3581     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3582     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3583     GL_EXTCALL(glUseProgramObjectARB(program_id));
3584     checkGLcall("glUseProgramObjectARB");
3585 }
3586
3587 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
3588     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3589     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3590     struct shader_glsl_priv *priv = This->shader_priv;
3591     GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
3592
3593     if (!*blt_program) {
3594         GLhandleARB loc;
3595         *blt_program = create_glsl_blt_shader(gl_info, tex_type);
3596         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
3597         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3598         GL_EXTCALL(glUniform1iARB(loc, 0));
3599     } else {
3600         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3601     }
3602 }
3603
3604 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
3605     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3606     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3607     struct shader_glsl_priv *priv = This->shader_priv;
3608     GLhandleARB program_id;
3609
3610     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3611     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3612
3613     GL_EXTCALL(glUseProgramObjectARB(program_id));
3614     checkGLcall("glUseProgramObjectARB");
3615 }
3616
3617 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3618     const struct list *linked_programs;
3619     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3620     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3621     struct shader_glsl_priv *priv = device->shader_priv;
3622     const WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
3623     IWineD3DPixelShaderImpl *ps = NULL;
3624     IWineD3DVertexShaderImpl *vs = NULL;
3625
3626     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3627      * can be called from IWineD3DBaseShader::Release
3628      */
3629     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version);
3630
3631     if(pshader) {
3632         ps = (IWineD3DPixelShaderImpl *) This;
3633         if(ps->num_gl_shaders == 0) return;
3634     } else {
3635         vs = (IWineD3DVertexShaderImpl *) This;
3636         if(vs->num_gl_shaders == 0) return;
3637     }
3638
3639     linked_programs = &This->baseShader.linked_programs;
3640
3641     TRACE("Deleting linked programs\n");
3642     if (linked_programs->next) {
3643         struct glsl_shader_prog_link *entry, *entry2;
3644
3645         if(pshader) {
3646             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3647                 delete_glsl_program_entry(priv, gl_info, entry);
3648             }
3649         } else {
3650             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3651                 delete_glsl_program_entry(priv, gl_info, entry);
3652             }
3653         }
3654     }
3655
3656     if(pshader) {
3657         UINT i;
3658
3659         ENTER_GL();
3660         for(i = 0; i < ps->num_gl_shaders; i++) {
3661             TRACE("deleting pshader %u\n", ps->gl_shaders[i].prgId);
3662             GL_EXTCALL(glDeleteObjectARB(ps->gl_shaders[i].prgId));
3663             checkGLcall("glDeleteObjectARB");
3664         }
3665         LEAVE_GL();
3666         HeapFree(GetProcessHeap(), 0, ps->gl_shaders);
3667         ps->gl_shaders = NULL;
3668         ps->num_gl_shaders = 0;
3669         ps->shader_array_size = 0;
3670     } else {
3671         UINT i;
3672
3673         ENTER_GL();
3674         for(i = 0; i < vs->num_gl_shaders; i++) {
3675             TRACE("deleting vshader %u\n", vs->gl_shaders[i].prgId);
3676             GL_EXTCALL(glDeleteObjectARB(vs->gl_shaders[i].prgId));
3677             checkGLcall("glDeleteObjectARB");
3678         }
3679         LEAVE_GL();
3680         HeapFree(GetProcessHeap(), 0, vs->gl_shaders);
3681         vs->gl_shaders = NULL;
3682         vs->num_gl_shaders = 0;
3683         vs->shader_array_size = 0;
3684     }
3685 }
3686
3687 static unsigned int glsl_program_key_hash(const void *key)
3688 {
3689     const glsl_program_key_t *k = key;
3690
3691     unsigned int hash = ((DWORD_PTR) k->vshader) | ((DWORD_PTR) k->pshader) << 16;
3692     hash += ~(hash << 15);
3693     hash ^=  (hash >> 10);
3694     hash +=  (hash << 3);
3695     hash ^=  (hash >> 6);
3696     hash += ~(hash << 11);
3697     hash ^=  (hash >> 16);
3698
3699     return hash;
3700 }
3701
3702 static BOOL glsl_program_key_compare(const void *keya, const void *keyb)
3703 {
3704     const glsl_program_key_t *ka = keya;
3705     const glsl_program_key_t *kb = keyb;
3706
3707     return ka->vshader == kb->vshader && ka->pshader == kb->pshader &&
3708            (memcmp(&ka->ps_args, &kb->ps_args, sizeof(kb->ps_args)) == 0) &&
3709            (memcmp(&ka->vs_args, &kb->vs_args, sizeof(kb->vs_args)) == 0);
3710 }
3711
3712 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
3713 {
3714     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
3715     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
3716
3717     if (!mem)
3718     {
3719         ERR("Failed to allocate memory\n");
3720         return FALSE;
3721     }
3722
3723     heap->entries = mem;
3724     heap->entries[1].version = 0;
3725     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
3726     heap->size = 1;
3727
3728     return TRUE;
3729 }
3730
3731 static void constant_heap_free(struct constant_heap *heap)
3732 {
3733     HeapFree(GetProcessHeap(), 0, heap->entries);
3734 }
3735
3736 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3737     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3738     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3739     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3740     SIZE_T stack_size = wined3d_log2i(max(GL_LIMITS(vshader_constantsF), GL_LIMITS(pshader_constantsF))) + 1;
3741
3742     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
3743     if (!priv->stack)
3744     {
3745         ERR("Failed to allocate memory.\n");
3746         HeapFree(GetProcessHeap(), 0, priv);
3747         return E_OUTOFMEMORY;
3748     }
3749
3750     if (!constant_heap_init(&priv->vconst_heap, GL_LIMITS(vshader_constantsF)))
3751     {
3752         ERR("Failed to initialize vertex shader constant heap\n");
3753         HeapFree(GetProcessHeap(), 0, priv->stack);
3754         HeapFree(GetProcessHeap(), 0, priv);
3755         return E_OUTOFMEMORY;
3756     }
3757
3758     if (!constant_heap_init(&priv->pconst_heap, GL_LIMITS(pshader_constantsF)))
3759     {
3760         ERR("Failed to initialize pixel shader constant heap\n");
3761         constant_heap_free(&priv->vconst_heap);
3762         HeapFree(GetProcessHeap(), 0, priv->stack);
3763         HeapFree(GetProcessHeap(), 0, priv);
3764         return E_OUTOFMEMORY;
3765     }
3766
3767     priv->glsl_program_lookup = hash_table_create(glsl_program_key_hash, glsl_program_key_compare);
3768     priv->next_constant_version = 1;
3769
3770     This->shader_priv = priv;
3771     return WINED3D_OK;
3772 }
3773
3774 static void shader_glsl_free(IWineD3DDevice *iface) {
3775     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3776     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3777     struct shader_glsl_priv *priv = This->shader_priv;
3778     int i;
3779
3780     for (i = 0; i < tex_type_count; ++i)
3781     {
3782         if (priv->depth_blt_program[i])
3783         {
3784             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
3785         }
3786     }
3787
3788     hash_table_destroy(priv->glsl_program_lookup, NULL, NULL);
3789     constant_heap_free(&priv->pconst_heap);
3790     constant_heap_free(&priv->vconst_heap);
3791
3792     HeapFree(GetProcessHeap(), 0, This->shader_priv);
3793     This->shader_priv = NULL;
3794 }
3795
3796 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3797     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3798     return FALSE;
3799 }
3800
3801 static GLuint shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer, const struct ps_compile_args *args) {
3802     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3803     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3804     CONST DWORD *function = This->baseShader.function;
3805     const char *fragcolor;
3806     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3807
3808     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3809     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3810
3811     shader_addline(buffer, "#version 120\n");
3812
3813     if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3814         shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3815     }
3816     if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3817         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3818          * drivers write a warning if we don't do so
3819          */
3820         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3821     }
3822
3823     /* Base Declarations */
3824     shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION, args);
3825
3826     /* Pack 3.0 inputs */
3827     if (reg_maps->shader_version >= WINED3DPS_VERSION(3,0) && args->vp_mode != vertexshader) {
3828         pshader_glsl_input_pack(buffer, This->semantics_in, iface, args->vp_mode);
3829     }
3830
3831     /* Base Shader Body */
3832     shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3833
3834     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3835     if (reg_maps->shader_version < WINED3DPS_VERSION(2,0))
3836     {
3837         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3838         if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3839             shader_addline(buffer, "gl_FragData[0] = R0;\n");
3840         else
3841             shader_addline(buffer, "gl_FragColor = R0;\n");
3842     }
3843
3844     if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3845         fragcolor = "gl_FragData[0]";
3846     } else {
3847         fragcolor = "gl_FragColor";
3848     }
3849     if(args->srgb_correction) {
3850         shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3851                         fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3852                         srgb_sub_high, srgb_sub_high, srgb_sub_high);
3853         shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3854         shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3855         shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3856         shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3857         shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3858     }
3859     /* Pixel shader < 3.0 do not replace the fog stage.
3860      * This implements linear fog computation and blending.
3861      * TODO: non linear fog
3862      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3863      * -1/(e-s) and e/(e-s) respectively.
3864      */
3865     if(reg_maps->shader_version < WINED3DPS_VERSION(3,0)) {
3866         switch(args->fog) {
3867             case FOG_OFF: break;
3868             case FOG_LINEAR:
3869                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
3870                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
3871                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
3872                 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
3873                 break;
3874             case FOG_EXP:
3875                 FIXME("Implement EXP fog in glsl\n");
3876                 break;
3877             case FOG_EXP2:
3878                 FIXME("Implement EXP2 fog in glsl\n");
3879                 break;
3880         }
3881     }
3882
3883     shader_addline(buffer, "}\n");
3884
3885     TRACE("Compiling shader object %u\n", shader_obj);
3886     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3887     GL_EXTCALL(glCompileShaderARB(shader_obj));
3888     print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3889
3890     /* Store the shader object */
3891     return shader_obj;
3892 }
3893
3894 static GLuint shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer, const struct vs_compile_args *args) {
3895     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
3896     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3897     CONST DWORD *function = This->baseShader.function;
3898     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3899
3900     /* Create the hw GLSL shader program and assign it as the shader->prgId */
3901     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3902
3903     shader_addline(buffer, "#version 120\n");
3904
3905     /* Base Declarations */
3906     shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION, NULL);
3907
3908     /* Base Shader Body */
3909     shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3910
3911     /* Unpack 3.0 outputs */
3912     if (reg_maps->shader_version >= WINED3DVS_VERSION(3,0)) shader_addline(buffer, "order_ps_input(OUT);\n");
3913     else shader_addline(buffer, "order_ps_input();\n");
3914
3915     /* If this shader doesn't use fog copy the z coord to the fog coord so that we can use table fog */
3916     if (!reg_maps->fog)
3917         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3918
3919     /* Write the final position.
3920      *
3921      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3922      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3923      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3924      * contains 1.0 to allow a mad.
3925      */
3926     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3927     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3928
3929     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3930      *
3931      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3932      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3933      * which is the same as z = z * 2 - w.
3934      */
3935     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3936
3937     shader_addline(buffer, "}\n");
3938
3939     TRACE("Compiling shader object %u\n", shader_obj);
3940     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3941     GL_EXTCALL(glCompileShaderARB(shader_obj));
3942     print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3943
3944     return shader_obj;
3945 }
3946
3947 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, const WineD3D_GL_Info *gl_info, struct shader_caps *pCaps)
3948 {
3949     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
3950      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
3951      * vs_nv_version which is based on NV_vertex_program.
3952      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
3953      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
3954      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
3955      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
3956      */
3957     if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3958         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
3959     else
3960         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
3961     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
3962     pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
3963
3964     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
3965      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
3966      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
3967      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
3968      * in max native instructions. Intel and others also offer the info in this extension but they
3969      * don't support GLSL (at least on Windows).
3970      *
3971      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
3972      * of instructions is 512 or less we have to do with ps2.0 hardware.
3973      * NOTE: ps3.0 hardware requires 512 or more instructions but ati and nvidia offer 'enough' (1024 vs 4096) on their most basic ps3.0 hardware.
3974      */
3975     if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3976         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
3977     else
3978         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
3979
3980     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
3981      * Direct3D minimum requirement.
3982      *
3983      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
3984      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
3985      *
3986      * The problem is that the refrast clamps temporary results in the shader to
3987      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
3988      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
3989      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
3990      * offer a way to query this.
3991      */
3992     pCaps->PixelShader1xMaxValue = 8.0;
3993     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
3994 }
3995
3996 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
3997 {
3998     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
3999     {
4000         TRACE("Checking support for fixup:\n");
4001         dump_color_fixup_desc(fixup);
4002     }
4003
4004     /* We support everything except YUV conversions. */
4005     if (!is_yuv_fixup(fixup))
4006     {
4007         TRACE("[OK]\n");
4008         return TRUE;
4009     }
4010
4011     TRACE("[FAILED]\n");
4012     return FALSE;
4013 }
4014
4015 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4016 {
4017     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4018     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4019     /* WINED3DSIH_BEM           */ pshader_glsl_bem,
4020     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4021     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4022     /* WINED3DSIH_BREAKP        */ NULL,
4023     /* WINED3DSIH_CALL          */ shader_glsl_call,
4024     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4025     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4026     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4027     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4028     /* WINED3DSIH_DCL           */ NULL,
4029     /* WINED3DSIH_DEF           */ NULL,
4030     /* WINED3DSIH_DEFB          */ NULL,
4031     /* WINED3DSIH_DEFI          */ NULL,
4032     /* WINED3DSIH_DP2ADD        */ pshader_glsl_dp2add,
4033     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4034     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4035     /* WINED3DSIH_DST           */ shader_glsl_dst,
4036     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
4037     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
4038     /* WINED3DSIH_ELSE          */ shader_glsl_else,
4039     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
4040     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
4041     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
4042     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
4043     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
4044     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
4045     /* WINED3DSIH_IF            */ shader_glsl_if,
4046     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
4047     /* WINED3DSIH_LABEL         */ shader_glsl_label,
4048     /* WINED3DSIH_LIT           */ shader_glsl_lit,
4049     /* WINED3DSIH_LOG           */ shader_glsl_log,
4050     /* WINED3DSIH_LOGP          */ shader_glsl_log,
4051     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
4052     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
4053     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
4054     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
4055     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
4056     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
4057     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
4058     /* WINED3DSIH_MAD           */ shader_glsl_mad,
4059     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
4060     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
4061     /* WINED3DSIH_MOV           */ shader_glsl_mov,
4062     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
4063     /* WINED3DSIH_MUL           */ shader_glsl_arith,
4064     /* WINED3DSIH_NOP           */ NULL,
4065     /* WINED3DSIH_NRM           */ shader_glsl_map2gl,
4066     /* WINED3DSIH_PHASE         */ NULL,
4067     /* WINED3DSIH_POW           */ shader_glsl_pow,
4068     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
4069     /* WINED3DSIH_REP           */ shader_glsl_rep,
4070     /* WINED3DSIH_RET           */ NULL,
4071     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
4072     /* WINED3DSIH_SETP          */ NULL,
4073     /* WINED3DSIH_SGE           */ shader_glsl_compare,
4074     /* WINED3DSIH_SGN           */ shader_glsl_map2gl,
4075     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
4076     /* WINED3DSIH_SLT           */ shader_glsl_compare,
4077     /* WINED3DSIH_SUB           */ shader_glsl_arith,
4078     /* WINED3DSIH_TEX           */ pshader_glsl_tex,
4079     /* WINED3DSIH_TEXBEM        */ pshader_glsl_texbem,
4080     /* WINED3DSIH_TEXBEML       */ pshader_glsl_texbem,
4081     /* WINED3DSIH_TEXCOORD      */ pshader_glsl_texcoord,
4082     /* WINED3DSIH_TEXDEPTH      */ pshader_glsl_texdepth,
4083     /* WINED3DSIH_TEXDP3        */ pshader_glsl_texdp3,
4084     /* WINED3DSIH_TEXDP3TEX     */ pshader_glsl_texdp3tex,
4085     /* WINED3DSIH_TEXKILL       */ pshader_glsl_texkill,
4086     /* WINED3DSIH_TEXLDD        */ NULL,
4087     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
4088     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_glsl_texm3x2depth,
4089     /* WINED3DSIH_TEXM3x2PAD    */ pshader_glsl_texm3x2pad,
4090     /* WINED3DSIH_TEXM3x2TEX    */ pshader_glsl_texm3x2tex,
4091     /* WINED3DSIH_TEXM3x3       */ pshader_glsl_texm3x3,
4092     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4093     /* WINED3DSIH_TEXM3x3PAD    */ pshader_glsl_texm3x3pad,
4094     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_glsl_texm3x3spec,
4095     /* WINED3DSIH_TEXM3x3TEX    */ pshader_glsl_texm3x3tex,
4096     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_glsl_texm3x3vspec,
4097     /* WINED3DSIH_TEXREG2AR     */ pshader_glsl_texreg2ar,
4098     /* WINED3DSIH_TEXREG2GB     */ pshader_glsl_texreg2gb,
4099     /* WINED3DSIH_TEXREG2RGB    */ pshader_glsl_texreg2rgb,
4100 };
4101
4102 const shader_backend_t glsl_shader_backend = {
4103     shader_glsl_instruction_handler_table,
4104     shader_glsl_select,
4105     shader_glsl_select_depth_blt,
4106     shader_glsl_deselect_depth_blt,
4107     shader_glsl_update_float_vertex_constants,
4108     shader_glsl_update_float_pixel_constants,
4109     shader_glsl_load_constants,
4110     shader_glsl_color_correction,
4111     shader_glsl_destroy,
4112     shader_glsl_alloc,
4113     shader_glsl_free,
4114     shader_glsl_dirty_const,
4115     shader_glsl_generate_pshader,
4116     shader_glsl_generate_vshader,
4117     shader_glsl_get_caps,
4118     shader_glsl_color_fixup_supported,
4119 };