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  * Copyright 2009 Henri Verbeet for CodeWeavers
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  22  */
  23
  24 /*
  25  * D3D shader asm has swizzles on source parameters, and write masks for
  26  * destination parameters. GLSL uses swizzles for both. The result of this is
  27  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  28  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  29  * mask for the destination parameter into account.
  30  */
  31
  32 #include "config.h"
  33 #include <limits.h>
  34 #include <stdio.h>
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define GLINFO_LOCATION      (*gl_info)
  43
  44 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  45 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  46 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  47 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  48
  49 typedef struct {
  50     char reg_name[150];
  51     char mask_str[6];
  52 } glsl_dst_param_t;
  53
  54 typedef struct {
  55     char reg_name[150];
  56     char param_str[200];
  57 } glsl_src_param_t;
  58
  59 typedef struct {
  60     const char *name;
  61     DWORD coord_mask;
  62 } glsl_sample_function_t;
  63
  64 enum heap_node_op
  65 {
  66     HEAP_NODE_TRAVERSE_LEFT,
  67     HEAP_NODE_TRAVERSE_RIGHT,
  68     HEAP_NODE_POP,
  69 };
  70
  71 struct constant_entry
  72 {
  73     unsigned int idx;
  74     unsigned int version;
  75 };
  76
  77 struct constant_heap
  78 {
  79     struct constant_entry *entries;
  80     unsigned int *positions;
  81     unsigned int size;
  82 };
  83
  84 /* GLSL shader private data */
  85 struct shader_glsl_priv {
  86     struct hash_table_t *glsl_program_lookup;
  87     struct glsl_shader_prog_link *glsl_program;
  88     struct constant_heap vconst_heap;
  89     struct constant_heap pconst_heap;
  90     unsigned char *stack;
  91     GLhandleARB depth_blt_program[tex_type_count];
  92     UINT next_constant_version;
  93 };
  94
  95 /* Struct to maintain data about a linked GLSL program */
  96 struct glsl_shader_prog_link {
  97     struct list                 vshader_entry;
  98     struct list                 pshader_entry;
  99     GLhandleARB                 programId;
 100     GLint                       *vuniformF_locations;
 101     GLint                       *puniformF_locations;
 102     GLint                       vuniformI_locations[MAX_CONST_I];
 103     GLint                       puniformI_locations[MAX_CONST_I];
 104     GLint                       posFixup_location;
 105     GLint                       np2Fixup_location[MAX_FRAGMENT_SAMPLERS];
 106     GLint                       bumpenvmat_location[MAX_TEXTURES];
 107     GLint                       luminancescale_location[MAX_TEXTURES];
 108     GLint                       luminanceoffset_location[MAX_TEXTURES];
 109     GLint                       ycorrection_location;
 110     GLenum                      vertex_color_clamp;
 111     IWineD3DVertexShader        *vshader;
 112     IWineD3DPixelShader         *pshader;
 113     struct vs_compile_args      vs_args;
 114     struct ps_compile_args      ps_args;
 115     UINT                        constant_version;
 116 };
 117
 118 typedef struct {
 119     IWineD3DVertexShader        *vshader;
 120     IWineD3DPixelShader         *pshader;
 121     struct ps_compile_args      ps_args;
 122     struct vs_compile_args      vs_args;
 123 } glsl_program_key_t;
 124
 125
 126 /** Prints the GLSL info log which will contain error messages if they exist */
 127 static void print_glsl_info_log(const WineD3D_GL_Info *gl_info, GLhandleARB obj)
 128 {
 129     int infologLength = 0;
 130     char *infoLog;
 131     unsigned int i;
 132     BOOL is_spam;
 133
 134     static const char * const spam[] =
 135     {
 136         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 137         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx          */
 138         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 139         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 140         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 141         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 142         "Fragment shader was successfully compiled to run on hardware.\n"
 143         "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported",
 144         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 145         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 146         "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported\n"  /* MacOS ati      */
 147     };
 148
 149     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 150
 151     GL_EXTCALL(glGetObjectParameterivARB(obj,
 152                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 153                &infologLength));
 154
 155     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 156      * that if there are errors. */
 157     if (infologLength > 1)
 158     {
 159         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 160          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 161          */
 162         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 163         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 164         is_spam = FALSE;
 165
 166         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 167             if(strcmp(infoLog, spam[i]) == 0) {
 168                 is_spam = TRUE;
 169                 break;
 170             }
 171         }
 172         if(is_spam) {
 173             TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
 174         } else {
 175             FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
 176         }
 177         HeapFree(GetProcessHeap(), 0, infoLog);
 178     }
 179 }
 180
 181 /**
 182  * Loads (pixel shader) samplers
 183  */
 184 static void shader_glsl_load_psamplers(const WineD3D_GL_Info *gl_info, DWORD *tex_unit_map, GLhandleARB programId)
 185 {
 186     GLint name_loc;
 187     int i;
 188     char sampler_name[20];
 189
 190     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 191         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 192         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 193         if (name_loc != -1) {
 194             DWORD mapped_unit = tex_unit_map[i];
 195             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < GL_LIMITS(fragment_samplers))
 196             {
 197                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 198                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 199                 checkGLcall("glUniform1iARB");
 200             } else {
 201                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 202             }
 203         }
 204     }
 205 }
 206
 207 static void shader_glsl_load_vsamplers(const WineD3D_GL_Info *gl_info, DWORD *tex_unit_map, GLhandleARB programId)
 208 {
 209     GLint name_loc;
 210     char sampler_name[20];
 211     int i;
 212
 213     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 214         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 215         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 216         if (name_loc != -1) {
 217             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 218             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < GL_LIMITS(combined_samplers))
 219             {
 220                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 221                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 222                 checkGLcall("glUniform1iARB");
 223             } else {
 224                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 225             }
 226         }
 227     }
 228 }
 229
 230 static inline void walk_constant_heap(const WineD3D_GL_Info *gl_info, const float *constants,
 231         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 232 {
 233     int stack_idx = 0;
 234     unsigned int heap_idx = 1;
 235     unsigned int idx;
 236
 237     if (heap->entries[heap_idx].version <= version) return;
 238
 239     idx = heap->entries[heap_idx].idx;
 240     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 241     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 242
 243     while (stack_idx >= 0)
 244     {
 245         /* Note that we fall through to the next case statement. */
 246         switch(stack[stack_idx])
 247         {
 248             case HEAP_NODE_TRAVERSE_LEFT:
 249             {
 250                 unsigned int left_idx = heap_idx << 1;
 251                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 252                 {
 253                     heap_idx = left_idx;
 254                     idx = heap->entries[heap_idx].idx;
 255                     if (constant_locations[idx] != -1)
 256                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 257
 258                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 259                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 260                     break;
 261                 }
 262             }
 263
 264             case HEAP_NODE_TRAVERSE_RIGHT:
 265             {
 266                 unsigned int right_idx = (heap_idx << 1) + 1;
 267                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 268                 {
 269                     heap_idx = right_idx;
 270                     idx = heap->entries[heap_idx].idx;
 271                     if (constant_locations[idx] != -1)
 272                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 273
 274                     stack[stack_idx++] = HEAP_NODE_POP;
 275                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 276                     break;
 277                 }
 278             }
 279
 280             case HEAP_NODE_POP:
 281             {
 282                 heap_idx >>= 1;
 283                 --stack_idx;
 284                 break;
 285             }
 286         }
 287     }
 288     checkGLcall("walk_constant_heap()");
 289 }
 290
 291 static inline void apply_clamped_constant(const WineD3D_GL_Info *gl_info, GLint location, const GLfloat *data)
 292 {
 293     GLfloat clamped_constant[4];
 294
 295     if (location == -1) return;
 296
 297     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0 ? 1.0 : data[0];
 298     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0 ? 1.0 : data[1];
 299     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0 ? 1.0 : data[2];
 300     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0 ? 1.0 : data[3];
 301
 302     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 303 }
 304
 305 static inline void walk_constant_heap_clamped(const WineD3D_GL_Info *gl_info, const float *constants,
 306         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 307 {
 308     int stack_idx = 0;
 309     unsigned int heap_idx = 1;
 310     unsigned int idx;
 311
 312     if (heap->entries[heap_idx].version <= version) return;
 313
 314     idx = heap->entries[heap_idx].idx;
 315     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 316     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 317
 318     while (stack_idx >= 0)
 319     {
 320         /* Note that we fall through to the next case statement. */
 321         switch(stack[stack_idx])
 322         {
 323             case HEAP_NODE_TRAVERSE_LEFT:
 324             {
 325                 unsigned int left_idx = heap_idx << 1;
 326                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 327                 {
 328                     heap_idx = left_idx;
 329                     idx = heap->entries[heap_idx].idx;
 330                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 331
 332                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 333                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 334                     break;
 335                 }
 336             }
 337
 338             case HEAP_NODE_TRAVERSE_RIGHT:
 339             {
 340                 unsigned int right_idx = (heap_idx << 1) + 1;
 341                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 342                 {
 343                     heap_idx = right_idx;
 344                     idx = heap->entries[heap_idx].idx;
 345                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 346
 347                     stack[stack_idx++] = HEAP_NODE_POP;
 348                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 349                     break;
 350                 }
 351             }
 352
 353             case HEAP_NODE_POP:
 354             {
 355                 heap_idx >>= 1;
 356                 --stack_idx;
 357                 break;
 358             }
 359         }
 360     }
 361     checkGLcall("walk_constant_heap_clamped()");
 362 }
 363
 364 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 365 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
 366         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 367         unsigned char *stack, UINT version)
 368 {
 369     const local_constant *lconst;
 370
 371     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 372     if (This->baseShader.reg_maps.shader_version.major == 1
 373             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
 374         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 375     else
 376         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 377
 378     if (!This->baseShader.load_local_constsF)
 379     {
 380         TRACE("No need to load local float constants for this shader\n");
 381         return;
 382     }
 383
 384     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 385     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 386     {
 387         GLint location = constant_locations[lconst->idx];
 388         /* We found this uniform name in the program - go ahead and send the data */
 389         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 390     }
 391     checkGLcall("glUniform4fvARB()");
 392 }
 393
 394 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 395 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
 396         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 397 {
 398     unsigned int i;
 399     struct list* ptr;
 400
 401     for (i = 0; constants_set; constants_set >>= 1, ++i)
 402     {
 403         if (!(constants_set & 1)) continue;
 404
 405         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 406                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 407
 408         /* We found this uniform name in the program - go ahead and send the data */
 409         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 410         checkGLcall("glUniform4ivARB");
 411     }
 412
 413     /* Load immediate constants */
 414     ptr = list_head(&This->baseShader.constantsI);
 415     while (ptr) {
 416         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 417         unsigned int idx = lconst->idx;
 418         const GLint *values = (const GLint *)lconst->value;
 419
 420         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 421             values[0], values[1], values[2], values[3]);
 422
 423         /* We found this uniform name in the program - go ahead and send the data */
 424         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 425         checkGLcall("glUniform4ivARB");
 426         ptr = list_next(&This->baseShader.constantsI, ptr);
 427     }
 428 }
 429
 430 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 431 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
 432         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 433 {
 434     GLint tmp_loc;
 435     unsigned int i;
 436     char tmp_name[8];
 437     char is_pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 438     const char* prefix = is_pshader? "PB":"VB";
 439     struct list* ptr;
 440
 441     /* TODO: Benchmark and see if it would be beneficial to store the
 442      * locations of the constants to avoid looking up each time */
 443     for (i = 0; constants_set; constants_set >>= 1, ++i)
 444     {
 445         if (!(constants_set & 1)) continue;
 446
 447         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 448
 449         /* TODO: Benchmark and see if it would be beneficial to store the
 450          * locations of the constants to avoid looking up each time */
 451         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 452         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 453         if (tmp_loc != -1)
 454         {
 455             /* We found this uniform name in the program - go ahead and send the data */
 456             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 457             checkGLcall("glUniform1ivARB");
 458         }
 459     }
 460
 461     /* Load immediate constants */
 462     ptr = list_head(&This->baseShader.constantsB);
 463     while (ptr) {
 464         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 465         unsigned int idx = lconst->idx;
 466         const GLint *values = (const GLint *)lconst->value;
 467
 468         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 469
 470         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 471         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 472         if (tmp_loc != -1) {
 473             /* We found this uniform name in the program - go ahead and send the data */
 474             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 475             checkGLcall("glUniform1ivARB");
 476         }
 477         ptr = list_next(&This->baseShader.constantsB, ptr);
 478     }
 479 }
 480
 481 static void reset_program_constant_version(void *value, void *context)
 482 {
 483     struct glsl_shader_prog_link *entry = value;
 484     entry->constant_version = 0;
 485 }
 486
 487 /**
 488  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 489  */
 490 static void shader_glsl_load_np2fixup_constants(
 491     IWineD3DDevice* device,
 492     char usePixelShader,
 493     char useVertexShader) {
 494
 495     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
 496     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
 497
 498     if (!prog) {
 499         /* No GLSL program set - nothing to do. */
 500         return;
 501     }
 502
 503     if (!usePixelShader) {
 504         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 505         return;
 506     }
 507
 508     if (prog->ps_args.np2_fixup) {
 509         UINT i;
 510         UINT fixup = prog->ps_args.np2_fixup;
 511         const WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
 512         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
 513
 514         for (i = 0; fixup; fixup >>= 1, ++i) {
 515             if (-1 != prog->np2Fixup_location[i]) {
 516                 const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
 517                 if (!tex) {
 518                     FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 519                     continue;
 520                 } else {
 521                     const float tex_dim[2] = {tex->baseTexture.pow2Matrix[0], tex->baseTexture.pow2Matrix[5]};
 522                     GL_EXTCALL(glUniform2fvARB(prog->np2Fixup_location[i], 1, tex_dim));
 523                 }
 524             }
 525         }
 526     }
 527 }
 528
 529 /**
 530  * Loads the app-supplied constants into the currently set GLSL program.
 531  */
 532 static void shader_glsl_load_constants(
 533     IWineD3DDevice* device,
 534     char usePixelShader,
 535     char useVertexShader) {
 536
 537     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
 538     struct shader_glsl_priv *priv = deviceImpl->shader_priv;
 539     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 540     const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
 541
 542     GLhandleARB programId;
 543     struct glsl_shader_prog_link *prog = priv->glsl_program;
 544     UINT constant_version;
 545     int i;
 546
 547     if (!prog) {
 548         /* No GLSL program set - nothing to do. */
 549         return;
 550     }
 551     programId = prog->programId;
 552     constant_version = prog->constant_version;
 553
 554     if (useVertexShader) {
 555         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 556
 557         /* Load DirectX 9 float constants/uniforms for vertex shader */
 558         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
 559                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 560
 561         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 562         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
 563                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 564
 565         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 566         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
 567                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 568
 569         /* Upload the position fixup params */
 570         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
 571         checkGLcall("glUniform4fvARB");
 572     }
 573
 574     if (usePixelShader) {
 575
 576         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 577
 578         /* Load DirectX 9 float constants/uniforms for pixel shader */
 579         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
 580                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 581
 582         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 583         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
 584                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 585
 586         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 587         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
 588                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 589
 590         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 591          * It can't be 0 for a valid texbem instruction.
 592          */
 593         for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
 594             IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
 595             int stage = ps->luminanceconst[i].texunit;
 596
 597             const float *data = (const float *)&stateBlock->textureState[(int)ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
 598             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 599             checkGLcall("glUniformMatrix2fvARB");
 600
 601             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 602              * is set too, so we can check that in the needsbumpmat check
 603              */
 604             if(ps->baseShader.reg_maps.luminanceparams[stage]) {
 605                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
 606                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
 607
 608                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 609                 checkGLcall("glUniform1fvARB");
 610                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 611                 checkGLcall("glUniform1fvARB");
 612             }
 613         }
 614
 615         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 616             float correction_params[4];
 617             if(deviceImpl->render_offscreen) {
 618                 correction_params[0] = 0.0;
 619                 correction_params[1] = 1.0;
 620             } else {
 621                 /* position is window relative, not viewport relative */
 622                 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
 623                 correction_params[1] = -1.0;
 624             }
 625             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 626         }
 627     }
 628
 629     if (priv->next_constant_version == UINT_MAX)
 630     {
 631         TRACE("Max constant version reached, resetting to 0.\n");
 632         hash_table_for_each_entry(priv->glsl_program_lookup, reset_program_constant_version, NULL);
 633         priv->next_constant_version = 1;
 634     }
 635     else
 636     {
 637         prog->constant_version = priv->next_constant_version++;
 638     }
 639 }
 640
 641 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 642         unsigned int heap_idx, DWORD new_version)
 643 {
 644     struct constant_entry *entries = heap->entries;
 645     unsigned int *positions = heap->positions;
 646     unsigned int parent_idx;
 647
 648     while (heap_idx > 1)
 649     {
 650         parent_idx = heap_idx >> 1;
 651
 652         if (new_version <= entries[parent_idx].version) break;
 653
 654         entries[heap_idx] = entries[parent_idx];
 655         positions[entries[parent_idx].idx] = heap_idx;
 656         heap_idx = parent_idx;
 657     }
 658
 659     entries[heap_idx].version = new_version;
 660     entries[heap_idx].idx = idx;
 661     positions[idx] = heap_idx;
 662 }
 663
 664 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 665 {
 666     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 667     struct shader_glsl_priv *priv = This->shader_priv;
 668     struct constant_heap *heap = &priv->vconst_heap;
 669     UINT i;
 670
 671     for (i = start; i < count + start; ++i)
 672     {
 673         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
 674             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 675         else
 676             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 677     }
 678 }
 679
 680 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 681 {
 682     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 683     struct shader_glsl_priv *priv = This->shader_priv;
 684     struct constant_heap *heap = &priv->pconst_heap;
 685     UINT i;
 686
 687     for (i = start; i < count + start; ++i)
 688     {
 689         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 690             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 691         else
 692             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 693     }
 694 }
 695
 696 /** Generate the variable & register declarations for the GLSL output target */
 697 static void shader_generate_glsl_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 698         SHADER_BUFFER *buffer, const WineD3D_GL_Info *gl_info,
 699         const struct ps_compile_args *ps_args)
 700 {
 701     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 702     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 703     unsigned int i, extra_constants_needed = 0;
 704     const local_constant *lconst;
 705
 706     /* There are some minor differences between pixel and vertex shaders */
 707     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 708     char prefix = pshader ? 'P' : 'V';
 709
 710     /* Prototype the subroutines */
 711     for (i = 0; i < This->baseShader.limits.label; i++) {
 712         if (reg_maps->labels[i])
 713             shader_addline(buffer, "void subroutine%u();\n", i);
 714     }
 715
 716     /* Declare the constants (aka uniforms) */
 717     if (This->baseShader.limits.constant_float > 0) {
 718         unsigned max_constantsF;
 719         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 720          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 721          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 722          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 723          * a dx9 card, as long as it doesn't also use all the other constants.
 724          *
 725          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 726          * declare only the amount that we're assured to have.
 727          *
 728          * Thus we run into problems in these two cases:
 729          * 1) The shader really uses more uniforms than supported
 730          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 731          */
 732         if(pshader) {
 733             /* No indirect addressing here */
 734             max_constantsF = GL_LIMITS(pshader_constantsF);
 735         } else {
 736             if(This->baseShader.reg_maps.usesrelconstF) {
 737                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
 738                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
 739                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
 740                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float
 741                  */
 742                 max_constantsF = GL_LIMITS(vshader_constantsF) - 3;
 743                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 744                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 745                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 746                  * for now take this into account when calculating the number of available constants
 747                  */
 748                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
 749                 /* Set by driver quirks in directx.c */
 750                 max_constantsF -= GLINFO_LOCATION.reserved_glsl_constants;
 751             } else {
 752                 max_constantsF = GL_LIMITS(vshader_constantsF);
 753             }
 754         }
 755         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
 756         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 757     }
 758
 759     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
 760      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
 761      */
 762     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
 763         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 764
 765     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
 766         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 767
 768     if(!pshader) {
 769         shader_addline(buffer, "uniform vec4 posFixup;\n");
 770         /* Predeclaration; This function is added at link time based on the pixel shader.
 771          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 772          * that. We know the input to the reorder function at vertex shader compile time, so
 773          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 774          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 775          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 776          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 777          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 778          * inout.
 779          */
 780         if (reg_maps->shader_version.major >= 3)
 781         {
 782             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 783         } else {
 784             shader_addline(buffer, "void order_ps_input();\n");
 785         }
 786     } else {
 787         IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
 788
 789         ps_impl->numbumpenvmatconsts = 0;
 790         for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
 791             if(!reg_maps->bumpmat[i]) {
 792                 continue;
 793             }
 794
 795             ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
 796             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 797
 798             if(reg_maps->luminanceparams) {
 799                 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
 800                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
 801                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
 802                 extra_constants_needed++;
 803             } else {
 804                 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
 805             }
 806
 807             extra_constants_needed++;
 808             ps_impl->numbumpenvmatconsts++;
 809         }
 810
 811         if(ps_args->srgb_correction) {
 812             shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
 813                             srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
 814             shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
 815                             srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
 816         }
 817         if(reg_maps->vpos || reg_maps->usesdsy) {
 818             if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
 819                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
 820                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
 821                 extra_constants_needed++;
 822             } else {
 823                 /* This happens because we do not have proper tracking of the constant registers that are
 824                  * actually used, only the max limit of the shader version
 825                  */
 826                 FIXME("Cannot find a free uniform for vpos correction params\n");
 827                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
 828                                device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
 829                                device->render_offscreen ? 1.0 : -1.0);
 830             }
 831             shader_addline(buffer, "vec4 vpos;\n");
 832         }
 833     }
 834
 835     /* Declare texture samplers */
 836     for (i = 0; i < This->baseShader.limits.sampler; i++) {
 837         if (reg_maps->sampler_type[i])
 838         {
 839             switch (reg_maps->sampler_type[i])
 840             {
 841                 case WINED3DSTT_1D:
 842                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
 843                     break;
 844                 case WINED3DSTT_2D:
 845                     if(device->stateBlock->textures[i] &&
 846                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
 847                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
 848                     } else {
 849                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
 850                     }
 851
 852                     if (pshader && ps_args->np2_fixup & (1 << i))
 853                     {
 854                         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
 855                          * while D3D has them in the (normalized) [0,1]x[0,1] range.
 856                          * samplerNP2Fixup stores texture dimensions and is updated through
 857                          * shader_glsl_load_np2fixup_constants when the sampler changes. */
 858                         shader_addline(buffer, "uniform vec2 %csamplerNP2Fixup%u;\n", prefix, i);
 859                     }
 860                     break;
 861                 case WINED3DSTT_CUBE:
 862                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
 863                     break;
 864                 case WINED3DSTT_VOLUME:
 865                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
 866                     break;
 867                 default:
 868                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
 869                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
 870                     break;
 871             }
 872         }
 873     }
 874
 875     /* Declare address variables */
 876     for (i = 0; i < This->baseShader.limits.address; i++) {
 877         if (reg_maps->address[i])
 878             shader_addline(buffer, "ivec4 A%d;\n", i);
 879     }
 880
 881     /* Declare texture coordinate temporaries and initialize them */
 882     for (i = 0; i < This->baseShader.limits.texcoord; i++) {
 883         if (reg_maps->texcoord[i])
 884             shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
 885     }
 886
 887     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
 888      * helper function shader that is linked in at link time
 889      */
 890     if (pshader && reg_maps->shader_version.major >= 3)
 891     {
 892         if (use_vs(device->stateBlock))
 893         {
 894             shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 895         } else {
 896             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
 897              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
 898              * pixel shader that reads the fixed function color into the packed input registers.
 899              */
 900             shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
 901         }
 902     }
 903
 904     /* Declare output register temporaries */
 905     if(This->baseShader.limits.packed_output) {
 906         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
 907     }
 908
 909     /* Declare temporary variables */
 910     for(i = 0; i < This->baseShader.limits.temporary; i++) {
 911         if (reg_maps->temporary[i])
 912             shader_addline(buffer, "vec4 R%u;\n", i);
 913     }
 914
 915     /* Declare attributes */
 916     for (i = 0; i < This->baseShader.limits.attributes; i++) {
 917         if (reg_maps->attributes[i])
 918             shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
 919     }
 920
 921     /* Declare loop registers aLx */
 922     for (i = 0; i < reg_maps->loop_depth; i++) {
 923         shader_addline(buffer, "int aL%u;\n", i);
 924         shader_addline(buffer, "int tmpInt%u;\n", i);
 925     }
 926
 927     /* Temporary variables for matrix operations */
 928     shader_addline(buffer, "vec4 tmp0;\n");
 929     shader_addline(buffer, "vec4 tmp1;\n");
 930
 931     /* Local constants use a different name so they can be loaded once at shader link time
 932      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
 933      * float -> string conversion can cause precision loss.
 934      */
 935     if(!This->baseShader.load_local_constsF) {
 936         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 937             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
 938         }
 939     }
 940
 941     /* Start the main program */
 942     shader_addline(buffer, "void main() {\n");
 943     if(pshader && reg_maps->vpos) {
 944         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
 945          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
 946          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
 947          * precision troubles when we just substract 0.5.
 948          *
 949          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
 950          *
 951          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
 952          *
 953          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
 954          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
 955          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
 956          * correctly on drivers that returns integer values.
 957          */
 958         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
 959     }
 960 }
 961
 962 /*****************************************************************************
 963  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
 964  *
 965  * For more information, see http://wiki.winehq.org/DirectX-Shaders
 966  ****************************************************************************/
 967
 968 /* Prototypes */
 969 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
 970         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
 971
 972 /** Used for opcode modifiers - They multiply the result by the specified amount */
 973 static const char * const shift_glsl_tab[] = {
 974     "",           /*  0 (none) */
 975     "2.0 * ",     /*  1 (x2)   */
 976     "4.0 * ",     /*  2 (x4)   */
 977     "8.0 * ",     /*  3 (x8)   */
 978     "16.0 * ",    /*  4 (x16)  */
 979     "32.0 * ",    /*  5 (x32)  */
 980     "",           /*  6 (x64)  */
 981     "",           /*  7 (x128) */
 982     "",           /*  8 (d256) */
 983     "",           /*  9 (d128) */
 984     "",           /* 10 (d64)  */
 985     "",           /* 11 (d32)  */
 986     "0.0625 * ",  /* 12 (d16)  */
 987     "0.125 * ",   /* 13 (d8)   */
 988     "0.25 * ",    /* 14 (d4)   */
 989     "0.5 * "      /* 15 (d2)   */
 990 };
 991
 992 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
 993 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
 994 {
 995     out_str[0] = 0;
 996
 997     switch (src_modifier)
 998     {
 999     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1000     case WINED3DSPSM_DW:
1001     case WINED3DSPSM_NONE:
1002         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1003         break;
1004     case WINED3DSPSM_NEG:
1005         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1006         break;
1007     case WINED3DSPSM_NOT:
1008         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1009         break;
1010     case WINED3DSPSM_BIAS:
1011         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1012         break;
1013     case WINED3DSPSM_BIASNEG:
1014         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1015         break;
1016     case WINED3DSPSM_SIGN:
1017         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1018         break;
1019     case WINED3DSPSM_SIGNNEG:
1020         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1021         break;
1022     case WINED3DSPSM_COMP:
1023         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1024         break;
1025     case WINED3DSPSM_X2:
1026         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1027         break;
1028     case WINED3DSPSM_X2NEG:
1029         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1030         break;
1031     case WINED3DSPSM_ABS:
1032         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1033         break;
1034     case WINED3DSPSM_ABSNEG:
1035         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1036         break;
1037     default:
1038         FIXME("Unhandled modifier %u\n", src_modifier);
1039         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1040     }
1041 }
1042
1043 /** Writes the GLSL variable name that corresponds to the register that the
1044  * DX opcode parameter is trying to access */
1045 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1046         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1047 {
1048     /* oPos, oFog and oPts in D3D */
1049     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
1050
1051     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1052     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1053     const WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
1054     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1055
1056     *is_color = FALSE;
1057
1058     switch (reg->type)
1059     {
1060         case WINED3DSPR_TEMP:
1061             sprintf(register_name, "R%u", reg->idx);
1062             break;
1063
1064         case WINED3DSPR_INPUT:
1065             /* vertex shaders */
1066             if (!pshader)
1067             {
1068                 if (((IWineD3DVertexShaderImpl *)This)->cur_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1069                 sprintf(register_name, "attrib%u", reg->idx);
1070                 break;
1071             }
1072
1073             /* pixel shaders >= 3.0 */
1074             if (This->baseShader.reg_maps.shader_version.major >= 3)
1075             {
1076                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1077                 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
1078
1079                 if (reg->rel_addr)
1080                 {
1081                     glsl_src_param_t rel_param;
1082
1083                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1084
1085                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1086                      * operation there */
1087                     if (idx)
1088                     {
1089                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1090                         {
1091                             sprintf(register_name,
1092                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1093                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1094                                     rel_param.param_str, idx);
1095                         }
1096                         else
1097                         {
1098                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1099                         }
1100                     }
1101                     else
1102                     {
1103                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1104                         {
1105                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1106                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1107                                     rel_param.param_str);
1108                         }
1109                         else
1110                         {
1111                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1112                         }
1113                     }
1114                 }
1115                 else
1116                 {
1117                     if (idx == in_count) sprintf(register_name, "gl_Color");
1118                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1119                     else sprintf(register_name, "IN[%u]", idx);
1120                 }
1121             }
1122             else
1123             {
1124                 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1125                 else strcpy(register_name, "gl_SecondaryColor");
1126                 break;
1127             }
1128             break;
1129
1130         case WINED3DSPR_CONST:
1131             {
1132                 const char prefix = pshader ? 'P' : 'V';
1133
1134                 /* Relative addressing */
1135                 if (reg->rel_addr)
1136                 {
1137                     glsl_src_param_t rel_param;
1138                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1139                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1140                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1141                 }
1142                 else
1143                 {
1144                     if (shader_constant_is_local(This, reg->idx))
1145                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1146                     else
1147                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1148                 }
1149             }
1150             break;
1151
1152         case WINED3DSPR_CONSTINT:
1153             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1154             else sprintf(register_name, "VI[%u]", reg->idx);
1155             break;
1156
1157         case WINED3DSPR_CONSTBOOL:
1158             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1159             else sprintf(register_name, "VB[%u]", reg->idx);
1160             break;
1161
1162         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1163             if (pshader) sprintf(register_name, "T%u", reg->idx);
1164             else sprintf(register_name, "A%u", reg->idx);
1165             break;
1166
1167         case WINED3DSPR_LOOP:
1168             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1169             break;
1170
1171         case WINED3DSPR_SAMPLER:
1172             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1173             else sprintf(register_name, "Vsampler%u", reg->idx);
1174             break;
1175
1176         case WINED3DSPR_COLOROUT:
1177             if (reg->idx >= GL_LIMITS(buffers))
1178                 WARN("Write to render target %u, only %d supported\n", reg->idx, 4);
1179
1180             if (GL_SUPPORT(ARB_DRAW_BUFFERS)) sprintf(register_name, "gl_FragData[%u]", reg->idx);
1181             /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
1182             else sprintf(register_name, "gl_FragColor");
1183             break;
1184
1185         case WINED3DSPR_RASTOUT:
1186             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1187             break;
1188
1189         case WINED3DSPR_DEPTHOUT:
1190             sprintf(register_name, "gl_FragDepth");
1191             break;
1192
1193         case WINED3DSPR_ATTROUT:
1194             if (reg->idx == 0) sprintf(register_name, "gl_FrontColor");
1195             else sprintf(register_name, "gl_FrontSecondaryColor");
1196             break;
1197
1198         case WINED3DSPR_TEXCRDOUT:
1199             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1200             if (This->baseShader.reg_maps.shader_version.major >= 3) sprintf(register_name, "OUT[%u]", reg->idx);
1201             else sprintf(register_name, "gl_TexCoord[%u]", reg->idx);
1202             break;
1203
1204         case WINED3DSPR_MISCTYPE:
1205             if (reg->idx == 0)
1206             {
1207                 /* vPos */
1208                 sprintf(register_name, "vpos");
1209             }
1210             else if (reg->idx == 1)
1211             {
1212                 /* Note that gl_FrontFacing is a bool, while vFace is
1213                  * a float for which the sign determines front/back */
1214                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1215             }
1216             else
1217             {
1218                 FIXME("Unhandled misctype register %d\n", reg->idx);
1219                 sprintf(register_name, "unrecognized_register");
1220             }
1221             break;
1222
1223         case WINED3DSPR_IMMCONST:
1224             switch (reg->immconst_type)
1225             {
1226                 case WINED3D_IMMCONST_FLOAT:
1227                     sprintf(register_name, "%.8e", *(float *)reg->immconst_data);
1228                     break;
1229
1230                 case WINED3D_IMMCONST_FLOAT4:
1231                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1232                             *(float *)&reg->immconst_data[0], *(float *)&reg->immconst_data[1],
1233                             *(float *)&reg->immconst_data[2], *(float *)&reg->immconst_data[3]);
1234                     break;
1235
1236                 default:
1237                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1238                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1239             }
1240             break;
1241
1242         default:
1243             FIXME("Unhandled register name Type(%d)\n", reg->type);
1244             sprintf(register_name, "unrecognized_register");
1245             break;
1246     }
1247 }
1248
1249 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1250 {
1251     *str++ = '.';
1252     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1253     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1254     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1255     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1256     *str = '\0';
1257 }
1258
1259 /* Get the GLSL write mask for the destination register */
1260 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1261 {
1262     DWORD mask = param->write_mask;
1263
1264     if (shader_is_scalar(&param->reg))
1265     {
1266         mask = WINED3DSP_WRITEMASK_0;
1267         *write_mask = '\0';
1268     }
1269     else
1270     {
1271         shader_glsl_write_mask_to_str(mask, write_mask);
1272     }
1273
1274     return mask;
1275 }
1276
1277 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1278     unsigned int size = 0;
1279
1280     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1281     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1282     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1283     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1284
1285     return size;
1286 }
1287
1288 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1289 {
1290     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1291      * but addressed as "rgba". To fix this we need to swap the register's x
1292      * and z components. */
1293     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1294
1295     *str++ = '.';
1296     /* swizzle bits fields: wwzzyyxx */
1297     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1298     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1299     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1300     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1301     *str = '\0';
1302 }
1303
1304 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1305         BOOL fixup, DWORD mask, char *swizzle_str)
1306 {
1307     if (shader_is_scalar(&param->reg))
1308         *swizzle_str = '\0';
1309     else
1310         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1311 }
1312
1313 /* From a given parameter token, generate the corresponding GLSL string.
1314  * Also, return the actual register name and swizzle in case the
1315  * caller needs this information as well. */
1316 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1317         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1318 {
1319     BOOL is_color = FALSE;
1320     char swizzle_str[6];
1321
1322     glsl_src->reg_name[0] = '\0';
1323     glsl_src->param_str[0] = '\0';
1324     swizzle_str[0] = '\0';
1325
1326     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1327     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1328     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1329 }
1330
1331 /* From a given parameter token, generate the corresponding GLSL string.
1332  * Also, return the actual register name and swizzle in case the
1333  * caller needs this information as well. */
1334 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1335         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1336 {
1337     BOOL is_color = FALSE;
1338
1339     glsl_dst->mask_str[0] = '\0';
1340     glsl_dst->reg_name[0] = '\0';
1341
1342     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1343     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1344 }
1345
1346 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1347 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer,
1348         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1349 {
1350     glsl_dst_param_t glsl_dst;
1351     DWORD mask;
1352
1353     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1354     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1355
1356     return mask;
1357 }
1358
1359 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1360 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, const struct wined3d_shader_instruction *ins)
1361 {
1362     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1363 }
1364
1365 /** Process GLSL instruction modifiers */
1366 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1367 {
1368     glsl_dst_param_t dst_param;
1369     DWORD modifiers;
1370
1371     if (!ins->dst_count) return;
1372
1373     modifiers = ins->dst[0].modifiers;
1374     if (!modifiers) return;
1375
1376     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1377
1378     if (modifiers & WINED3DSPDM_SATURATE)
1379     {
1380         /* _SAT means to clamp the value of the register to between 0 and 1 */
1381         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1382                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1383     }
1384
1385     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1386     {
1387         FIXME("_centroid modifier not handled\n");
1388     }
1389
1390     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1391     {
1392         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1393     }
1394 }
1395
1396 static inline const char *shader_get_comp_op(DWORD op)
1397 {
1398     switch (op) {
1399         case COMPARISON_GT: return ">";
1400         case COMPARISON_EQ: return "==";
1401         case COMPARISON_GE: return ">=";
1402         case COMPARISON_LT: return "<";
1403         case COMPARISON_NE: return "!=";
1404         case COMPARISON_LE: return "<=";
1405         default:
1406             FIXME("Unrecognized comparison value: %u\n", op);
1407             return "(\?\?)";
1408     }
1409 }
1410
1411 static void shader_glsl_get_sample_function(DWORD sampler_type, DWORD flags, glsl_sample_function_t *sample_function)
1412 {
1413     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1414     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1415     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1416     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1417
1418     /* Note that there's no such thing as a projected cube texture. */
1419     switch(sampler_type) {
1420         case WINED3DSTT_1D:
1421             if(lod) {
1422                 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1423             } else  if(grad) {
1424                 sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1425             } else {
1426                 sample_function->name = projected ? "texture1DProj" : "texture1D";
1427             }
1428             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1429             break;
1430         case WINED3DSTT_2D:
1431             if(texrect) {
1432                 if(lod) {
1433                     sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1434                 } else  if(grad) {
1435                     /* What good are texrect grad functions? I don't know, but GL_EXT_gpu_shader4 defines them.
1436                     * There is no GL_ARB_shader_texture_lod spec yet, so I don't know if they're defined there
1437                      */
1438                     sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1439                 } else {
1440                     sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1441                 }
1442             } else {
1443                 if(lod) {
1444                     sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1445                 } else  if(grad) {
1446                     sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1447                 } else {
1448                     sample_function->name = projected ? "texture2DProj" : "texture2D";
1449                 }
1450             }
1451             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1452             break;
1453         case WINED3DSTT_CUBE:
1454             if(lod) {
1455                 sample_function->name = "textureCubeLod";
1456             } else if(grad) {
1457                 sample_function->name = "textureCubeGradARB";
1458             } else {
1459                 sample_function->name = "textureCube";
1460             }
1461             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1462             break;
1463         case WINED3DSTT_VOLUME:
1464             if(lod) {
1465                 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1466             } else  if(grad) {
1467                 sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1468             } else {
1469                 sample_function->name = projected ? "texture3DProj" : "texture3D";
1470             }
1471             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1472             break;
1473         default:
1474             sample_function->name = "";
1475             sample_function->coord_mask = 0;
1476             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1477             break;
1478     }
1479 }
1480
1481 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1482         BOOL sign_fixup, enum fixup_channel_source channel_source)
1483 {
1484     switch(channel_source)
1485     {
1486         case CHANNEL_SOURCE_ZERO:
1487             strcat(arguments, "0.0");
1488             break;
1489
1490         case CHANNEL_SOURCE_ONE:
1491             strcat(arguments, "1.0");
1492             break;
1493
1494         case CHANNEL_SOURCE_X:
1495             strcat(arguments, reg_name);
1496             strcat(arguments, ".x");
1497             break;
1498
1499         case CHANNEL_SOURCE_Y:
1500             strcat(arguments, reg_name);
1501             strcat(arguments, ".y");
1502             break;
1503
1504         case CHANNEL_SOURCE_Z:
1505             strcat(arguments, reg_name);
1506             strcat(arguments, ".z");
1507             break;
1508
1509         case CHANNEL_SOURCE_W:
1510             strcat(arguments, reg_name);
1511             strcat(arguments, ".w");
1512             break;
1513
1514         default:
1515             FIXME("Unhandled channel source %#x\n", channel_source);
1516             strcat(arguments, "undefined");
1517             break;
1518     }
1519
1520     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1521 }
1522
1523 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1524 {
1525     struct wined3d_shader_dst_param dst;
1526     unsigned int mask_size, remaining;
1527     glsl_dst_param_t dst_param;
1528     char arguments[256];
1529     DWORD mask;
1530
1531     mask = 0;
1532     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1533     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1534     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1535     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1536     mask &= ins->dst[0].write_mask;
1537
1538     if (!mask) return; /* Nothing to do */
1539
1540     if (is_yuv_fixup(fixup))
1541     {
1542         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1543         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1544         return;
1545     }
1546
1547     mask_size = shader_glsl_get_write_mask_size(mask);
1548
1549     dst = ins->dst[0];
1550     dst.write_mask = mask;
1551     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1552
1553     arguments[0] = '\0';
1554     remaining = mask_size;
1555     if (mask & WINED3DSP_WRITEMASK_0)
1556     {
1557         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1558         if (--remaining) strcat(arguments, ", ");
1559     }
1560     if (mask & WINED3DSP_WRITEMASK_1)
1561     {
1562         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1563         if (--remaining) strcat(arguments, ", ");
1564     }
1565     if (mask & WINED3DSP_WRITEMASK_2)
1566     {
1567         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1568         if (--remaining) strcat(arguments, ", ");
1569     }
1570     if (mask & WINED3DSP_WRITEMASK_3)
1571     {
1572         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1573         if (--remaining) strcat(arguments, ", ");
1574     }
1575
1576     if (mask_size > 1)
1577     {
1578         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1579                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1580     }
1581     else
1582     {
1583         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1584     }
1585 }
1586
1587 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
1588         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
1589         const char *dx, const char *dy,
1590         const char *bias, const char *coord_reg_fmt, ...)
1591 {
1592     const char *sampler_base;
1593     char dst_swizzle[6];
1594     struct color_fixup_desc fixup;
1595     BOOL np2_fixup = FALSE;
1596     va_list args;
1597
1598     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
1599
1600     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1601     {
1602         IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
1603         fixup = This->cur_args->color_fixup[sampler];
1604         sampler_base = "Psampler";
1605
1606         if(This->cur_args->np2_fixup & (1 << sampler)) {
1607             if(bias) {
1608                 FIXME("Biased sampling from NP2 textures is unsupported\n");
1609             } else {
1610                 np2_fixup = TRUE;
1611             }
1612         }
1613     } else {
1614         sampler_base = "Vsampler";
1615         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
1616     }
1617
1618     shader_glsl_append_dst(ins->ctx->buffer, ins);
1619
1620     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
1621
1622     va_start(args, coord_reg_fmt);
1623     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
1624     va_end(args);
1625
1626     if(bias) {
1627         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
1628     } else {
1629         if (np2_fixup) {
1630             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup%u)%s);\n", sampler, dst_swizzle);
1631         } else if(dx && dy) {
1632             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
1633         } else {
1634             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
1635         }
1636     }
1637
1638     if(!is_identity_fixup(fixup)) {
1639         shader_glsl_color_correction(ins, fixup);
1640     }
1641 }
1642
1643 /*****************************************************************************
1644  *
1645  * Begin processing individual instruction opcodes
1646  *
1647  ****************************************************************************/
1648
1649 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1650 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
1651 {
1652     SHADER_BUFFER *buffer = ins->ctx->buffer;
1653     glsl_src_param_t src0_param;
1654     glsl_src_param_t src1_param;
1655     DWORD write_mask;
1656     char op;
1657
1658     /* Determine the GLSL operator to use based on the opcode */
1659     switch (ins->handler_idx)
1660     {
1661         case WINED3DSIH_MUL: op = '*'; break;
1662         case WINED3DSIH_ADD: op = '+'; break;
1663         case WINED3DSIH_SUB: op = '-'; break;
1664         default:
1665             op = ' ';
1666             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1667             break;
1668     }
1669
1670     write_mask = shader_glsl_append_dst(buffer, ins);
1671     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1672     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
1673     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1674 }
1675
1676 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1677 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
1678 {
1679     SHADER_BUFFER *buffer = ins->ctx->buffer;
1680     glsl_src_param_t src0_param;
1681     DWORD write_mask;
1682
1683     write_mask = shader_glsl_append_dst(buffer, ins);
1684     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1685
1686     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1687      * shader versions WINED3DSIO_MOVA is used for this. */
1688     if (ins->ctx->reg_maps->shader_version.major == 1
1689             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1690             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1691     {
1692         /* This is a simple floor() */
1693         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1694         if (mask_size > 1) {
1695             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1696         } else {
1697             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1698         }
1699     }
1700     else if(ins->handler_idx == WINED3DSIH_MOVA)
1701     {
1702         /* We need to *round* to the nearest int here. */
1703         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1704         if (mask_size > 1) {
1705             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);
1706         } else {
1707             shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1708         }
1709     } else {
1710         shader_addline(buffer, "%s);\n", src0_param.param_str);
1711     }
1712 }
1713
1714 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1715 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
1716 {
1717     SHADER_BUFFER *buffer = ins->ctx->buffer;
1718     glsl_src_param_t src0_param;
1719     glsl_src_param_t src1_param;
1720     DWORD dst_write_mask, src_write_mask;
1721     unsigned int dst_size = 0;
1722
1723     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1724     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1725
1726     /* dp3 works on vec3, dp4 on vec4 */
1727     if (ins->handler_idx == WINED3DSIH_DP4)
1728     {
1729         src_write_mask = WINED3DSP_WRITEMASK_ALL;
1730     } else {
1731         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1732     }
1733
1734     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
1735     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
1736
1737     if (dst_size > 1) {
1738         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1739     } else {
1740         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1741     }
1742 }
1743
1744 /* Note that this instruction has some restrictions. The destination write mask
1745  * can't contain the w component, and the source swizzles have to be .xyzw */
1746 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
1747 {
1748     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1749     glsl_src_param_t src0_param;
1750     glsl_src_param_t src1_param;
1751     char dst_mask[6];
1752
1753     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
1754     shader_glsl_append_dst(ins->ctx->buffer, ins);
1755     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
1756     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
1757     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1758 }
1759
1760 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1761  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1762  * GLSL uses the value as-is. */
1763 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
1764 {
1765     SHADER_BUFFER *buffer = ins->ctx->buffer;
1766     glsl_src_param_t src0_param;
1767     glsl_src_param_t src1_param;
1768     DWORD dst_write_mask;
1769     unsigned int dst_size;
1770
1771     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1772     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1773
1774     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
1775     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
1776
1777     if (dst_size > 1) {
1778         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1779     } else {
1780         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1781     }
1782 }
1783
1784 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1785  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1786  * GLSL uses the value as-is. */
1787 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
1788 {
1789     SHADER_BUFFER *buffer = ins->ctx->buffer;
1790     glsl_src_param_t src0_param;
1791     DWORD dst_write_mask;
1792     unsigned int dst_size;
1793
1794     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1795     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1796
1797     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
1798
1799     if (dst_size > 1) {
1800         shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1801     } else {
1802         shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1803     }
1804 }
1805
1806 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1807 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
1808 {
1809     SHADER_BUFFER *buffer = ins->ctx->buffer;
1810     glsl_src_param_t src_param;
1811     const char *instruction;
1812     DWORD write_mask;
1813     unsigned i;
1814
1815     /* Determine the GLSL function to use based on the opcode */
1816     /* TODO: Possibly make this a table for faster lookups */
1817     switch (ins->handler_idx)
1818     {
1819         case WINED3DSIH_MIN: instruction = "min"; break;
1820         case WINED3DSIH_MAX: instruction = "max"; break;
1821         case WINED3DSIH_ABS: instruction = "abs"; break;
1822         case WINED3DSIH_FRC: instruction = "fract"; break;
1823         case WINED3DSIH_NRM: instruction = "normalize"; break;
1824         case WINED3DSIH_EXP: instruction = "exp2"; break;
1825         case WINED3DSIH_SGN: instruction = "sign"; break;
1826         case WINED3DSIH_DSX: instruction = "dFdx"; break;
1827         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
1828         default: instruction = "";
1829             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1830             break;
1831     }
1832
1833     write_mask = shader_glsl_append_dst(buffer, ins);
1834
1835     shader_addline(buffer, "%s(", instruction);
1836
1837     if (ins->src_count)
1838     {
1839         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
1840         shader_addline(buffer, "%s", src_param.param_str);
1841         for (i = 1; i < ins->src_count; ++i)
1842         {
1843             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
1844             shader_addline(buffer, ", %s", src_param.param_str);
1845         }
1846     }
1847
1848     shader_addline(buffer, "));\n");
1849 }
1850
1851 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1852  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1853  *   dst.x = 2^(floor(src))
1854  *   dst.y = src - floor(src)
1855  *   dst.z = 2^src   (partial precision is allowed, but optional)
1856  *   dst.w = 1.0;
1857  * For 2.0 shaders, just do this (honoring writemask and swizzle):
1858  *   dst = 2^src;    (partial precision is allowed, but optional)
1859  */
1860 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
1861 {
1862     glsl_src_param_t src_param;
1863
1864     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
1865
1866     if (ins->ctx->reg_maps->shader_version.major < 2)
1867     {
1868         char dst_mask[6];
1869
1870         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1871         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1872         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1873         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
1874
1875         shader_glsl_append_dst(ins->ctx->buffer, ins);
1876         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
1877         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
1878     } else {
1879         DWORD write_mask;
1880         unsigned int mask_size;
1881
1882         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
1883         mask_size = shader_glsl_get_write_mask_size(write_mask);
1884
1885         if (mask_size > 1) {
1886             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1887         } else {
1888             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
1889         }
1890     }
1891 }
1892
1893 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1894 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
1895 {
1896     glsl_src_param_t src_param;
1897     DWORD write_mask;
1898     unsigned int mask_size;
1899
1900     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
1901     mask_size = shader_glsl_get_write_mask_size(write_mask);
1902     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
1903
1904     if (mask_size > 1) {
1905         shader_addline(ins->ctx->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1906     } else {
1907         shader_addline(ins->ctx->buffer, "1.0 / %s);\n", src_param.param_str);
1908     }
1909 }
1910
1911 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
1912 {
1913     SHADER_BUFFER *buffer = ins->ctx->buffer;
1914     glsl_src_param_t src_param;
1915     DWORD write_mask;
1916     unsigned int mask_size;
1917
1918     write_mask = shader_glsl_append_dst(buffer, ins);
1919     mask_size = shader_glsl_get_write_mask_size(write_mask);
1920
1921     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
1922
1923     if (mask_size > 1) {
1924         shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1925     } else {
1926         shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1927     }
1928 }
1929
1930 /** Process signed comparison opcodes in GLSL. */
1931 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
1932 {
1933     glsl_src_param_t src0_param;
1934     glsl_src_param_t src1_param;
1935     DWORD write_mask;
1936     unsigned int mask_size;
1937
1938     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
1939     mask_size = shader_glsl_get_write_mask_size(write_mask);
1940     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1941     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
1942
1943     if (mask_size > 1) {
1944         const char *compare;
1945
1946         switch(ins->handler_idx)
1947         {
1948             case WINED3DSIH_SLT: compare = "lessThan"; break;
1949             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
1950             default: compare = "";
1951                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
1952         }
1953
1954         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1955                 src0_param.param_str, src1_param.param_str);
1956     } else {
1957         switch(ins->handler_idx)
1958         {
1959             case WINED3DSIH_SLT:
1960                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1961                  * to return 0.0 but step returns 1.0 because step is not < x
1962                  * An alternative is a bvec compare padded with an unused second component.
1963                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1964                  * issue. Playing with not() is not possible either because not() does not accept
1965                  * a scalar.
1966                  */
1967                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
1968                         src0_param.param_str, src1_param.param_str);
1969                 break;
1970             case WINED3DSIH_SGE:
1971                 /* Here we can use the step() function and safe a conditional */
1972                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1973                 break;
1974             default:
1975                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
1976         }
1977
1978     }
1979 }
1980
1981 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1982 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
1983 {
1984     glsl_src_param_t src0_param;
1985     glsl_src_param_t src1_param;
1986     glsl_src_param_t src2_param;
1987     DWORD write_mask, cmp_channel = 0;
1988     unsigned int i, j;
1989     char mask_char[6];
1990     BOOL temp_destination = FALSE;
1991
1992     if (shader_is_scalar(&ins->src[0].reg))
1993     {
1994         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
1995
1996         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1997         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
1998         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
1999
2000         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2001                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2002     } else {
2003         DWORD dst_mask = ins->dst[0].write_mask;
2004         struct wined3d_shader_dst_param dst = ins->dst[0];
2005
2006         /* Cycle through all source0 channels */
2007         for (i=0; i<4; i++) {
2008             write_mask = 0;
2009             /* Find the destination channels which use the current source0 channel */
2010             for (j=0; j<4; j++) {
2011                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2012                 {
2013                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2014                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2015                 }
2016             }
2017             dst.write_mask = dst_mask & write_mask;
2018
2019             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2020             * The first lines may overwrite source parameters of the following lines.
2021             * Deal with that by using a temporary destination register if needed
2022             */
2023             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2024                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2025                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2026                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2027                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2028                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2029             {
2030                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2031                 if (!write_mask) continue;
2032                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2033                 temp_destination = TRUE;
2034             } else {
2035                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2036                 if (!write_mask) continue;
2037             }
2038
2039             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2040             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2041             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2042
2043             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2044                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2045         }
2046
2047         if(temp_destination) {
2048             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2049             shader_glsl_append_dst(ins->ctx->buffer, ins);
2050             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2051         }
2052     }
2053
2054 }
2055
2056 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2057 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2058  * the compare is done per component of src0. */
2059 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2060 {
2061     struct wined3d_shader_dst_param dst;
2062     glsl_src_param_t src0_param;
2063     glsl_src_param_t src1_param;
2064     glsl_src_param_t src2_param;
2065     DWORD write_mask, cmp_channel = 0;
2066     unsigned int i, j;
2067     DWORD dst_mask;
2068     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2069             ins->ctx->reg_maps->shader_version.minor);
2070
2071     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2072     {
2073         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2074         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2075         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2076         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2077
2078         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2079         if (ins->coissue)
2080         {
2081             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2082         } else {
2083             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2084                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2085         }
2086         return;
2087     }
2088     /* Cycle through all source0 channels */
2089     dst_mask = ins->dst[0].write_mask;
2090     dst = ins->dst[0];
2091     for (i=0; i<4; i++) {
2092         write_mask = 0;
2093         /* Find the destination channels which use the current source0 channel */
2094         for (j=0; j<4; j++) {
2095             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2096             {
2097                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2098                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2099             }
2100         }
2101
2102         dst.write_mask = dst_mask & write_mask;
2103         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2104         if (!write_mask) continue;
2105
2106         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2107         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2108         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2109
2110         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2111                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2112     }
2113 }
2114
2115 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2116 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2117 {
2118     glsl_src_param_t src0_param;
2119     glsl_src_param_t src1_param;
2120     glsl_src_param_t src2_param;
2121     DWORD write_mask;
2122
2123     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2124     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2125     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2126     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2127     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2128             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2129 }
2130
2131 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2132     Vertex shaders to GLSL codes */
2133 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2134 {
2135     int i;
2136     int nComponents = 0;
2137     struct wined3d_shader_dst_param tmp_dst = {{0}};
2138     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2139     struct wined3d_shader_instruction tmp_ins;
2140
2141     memset(&tmp_ins, 0, sizeof(tmp_ins));
2142
2143     /* Set constants for the temporary argument */
2144     tmp_ins.ctx = ins->ctx;
2145     tmp_ins.dst_count = 1;
2146     tmp_ins.dst = &tmp_dst;
2147     tmp_ins.src_count = 2;
2148     tmp_ins.src = tmp_src;
2149
2150     switch(ins->handler_idx)
2151     {
2152         case WINED3DSIH_M4x4:
2153             nComponents = 4;
2154             tmp_ins.handler_idx = WINED3DSIH_DP4;
2155             break;
2156         case WINED3DSIH_M4x3:
2157             nComponents = 3;
2158             tmp_ins.handler_idx = WINED3DSIH_DP4;
2159             break;
2160         case WINED3DSIH_M3x4:
2161             nComponents = 4;
2162             tmp_ins.handler_idx = WINED3DSIH_DP3;
2163             break;
2164         case WINED3DSIH_M3x3:
2165             nComponents = 3;
2166             tmp_ins.handler_idx = WINED3DSIH_DP3;
2167             break;
2168         case WINED3DSIH_M3x2:
2169             nComponents = 2;
2170             tmp_ins.handler_idx = WINED3DSIH_DP3;
2171             break;
2172         default:
2173             break;
2174     }
2175
2176     tmp_dst = ins->dst[0];
2177     tmp_src[0] = ins->src[0];
2178     tmp_src[1] = ins->src[1];
2179     for (i = 0; i < nComponents; ++i)
2180     {
2181         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2182         shader_glsl_dot(&tmp_ins);
2183         ++tmp_src[1].reg.idx;
2184     }
2185 }
2186
2187 /**
2188     The LRP instruction performs a component-wise linear interpolation
2189     between the second and third operands using the first operand as the
2190     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2191     This is equivalent to mix(src2, src1, src0);
2192 */
2193 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2194 {
2195     glsl_src_param_t src0_param;
2196     glsl_src_param_t src1_param;
2197     glsl_src_param_t src2_param;
2198     DWORD write_mask;
2199
2200     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2201
2202     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2203     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2204     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2205
2206     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2207             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2208 }
2209
2210 /** Process the WINED3DSIO_LIT instruction in GLSL:
2211  * dst.x = dst.w = 1.0
2212  * dst.y = (src0.x > 0) ? src0.x
2213  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2214  *                                        where src.w is clamped at +- 128
2215  */
2216 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2217 {
2218     glsl_src_param_t src0_param;
2219     glsl_src_param_t src1_param;
2220     glsl_src_param_t src3_param;
2221     char dst_mask[6];
2222
2223     shader_glsl_append_dst(ins->ctx->buffer, ins);
2224     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2225
2226     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2227     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2228     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2229
2230     /* The sdk specifies the instruction like this
2231      * dst.x = 1.0;
2232      * if(src.x > 0.0) dst.y = src.x
2233      * else dst.y = 0.0.
2234      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2235      * else dst.z = 0.0;
2236      * dst.w = 1.0;
2237      *
2238      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2239      * dst.x = 1.0                                  ... No further explanation needed
2240      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2241      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2242      * dst.w = 1.0.                                 ... Nothing fancy.
2243      *
2244      * So we still have one conditional in there. So do this:
2245      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2246      *
2247      * 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),
2248      * 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.
2249      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2250      */
2251     shader_addline(ins->ctx->buffer,
2252             "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",
2253             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2254 }
2255
2256 /** Process the WINED3DSIO_DST instruction in GLSL:
2257  * dst.x = 1.0
2258  * dst.y = src0.x * src0.y
2259  * dst.z = src0.z
2260  * dst.w = src1.w
2261  */
2262 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2263 {
2264     glsl_src_param_t src0y_param;
2265     glsl_src_param_t src0z_param;
2266     glsl_src_param_t src1y_param;
2267     glsl_src_param_t src1w_param;
2268     char dst_mask[6];
2269
2270     shader_glsl_append_dst(ins->ctx->buffer, ins);
2271     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2272
2273     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2274     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2275     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2276     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2277
2278     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2279             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2280 }
2281
2282 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2283  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2284  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2285  *
2286  * dst.x = cos(src0.?)
2287  * dst.y = sin(src0.?)
2288  * dst.z = dst.z
2289  * dst.w = dst.w
2290  */
2291 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2292 {
2293     glsl_src_param_t src0_param;
2294     DWORD write_mask;
2295
2296     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2297     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2298
2299     switch (write_mask) {
2300         case WINED3DSP_WRITEMASK_0:
2301             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2302             break;
2303
2304         case WINED3DSP_WRITEMASK_1:
2305             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2306             break;
2307
2308         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2309             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2310             break;
2311
2312         default:
2313             ERR("Write mask should be .x, .y or .xy\n");
2314             break;
2315     }
2316 }
2317
2318 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2319  * Start a for() loop where src1.y is the initial value of aL,
2320  *  increment aL by src1.z for a total of src1.x iterations.
2321  *  Need to use a temporary variable for this operation.
2322  */
2323 /* FIXME: I don't think nested loops will work correctly this way. */
2324 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2325 {
2326     glsl_src_param_t src1_param;
2327     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2328     const DWORD *control_values = NULL;
2329     const local_constant *constant;
2330
2331     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2332
2333     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2334      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2335      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2336      * addressing.
2337      */
2338     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2339     {
2340         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2341             if (constant->idx == ins->src[1].reg.idx)
2342             {
2343                 control_values = constant->value;
2344                 break;
2345             }
2346         }
2347     }
2348
2349     if(control_values) {
2350         if(control_values[2] > 0) {
2351             shader_addline(ins->ctx->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
2352                     shader->baseShader.cur_loop_depth, control_values[1],
2353                     shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2354                     shader->baseShader.cur_loop_depth, control_values[2]);
2355         } else if(control_values[2] == 0) {
2356             shader_addline(ins->ctx->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2357                     shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2358                     shader->baseShader.cur_loop_depth, control_values[0],
2359                     shader->baseShader.cur_loop_depth);
2360         } else {
2361             shader_addline(ins->ctx->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2362                     shader->baseShader.cur_loop_depth, control_values[1],
2363                     shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2364                     shader->baseShader.cur_loop_depth, control_values[2]);
2365         }
2366     } else {
2367         shader_addline(ins->ctx->buffer,
2368                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2369                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2370                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2371                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2372     }
2373
2374     shader->baseShader.cur_loop_depth++;
2375     shader->baseShader.cur_loop_regno++;
2376 }
2377
2378 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2379 {
2380     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2381
2382     shader_addline(ins->ctx->buffer, "}\n");
2383
2384     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2385     {
2386         shader->baseShader.cur_loop_depth--;
2387         shader->baseShader.cur_loop_regno--;
2388     }
2389
2390     if (ins->handler_idx == WINED3DSIH_ENDREP)
2391     {
2392         shader->baseShader.cur_loop_depth--;
2393     }
2394 }
2395
2396 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2397 {
2398     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2399     glsl_src_param_t src0_param;
2400     const DWORD *control_values = NULL;
2401     const local_constant *constant;
2402
2403     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2404     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2405     {
2406         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2407         {
2408             if (constant->idx == ins->src[0].reg.idx)
2409             {
2410                 control_values = constant->value;
2411                 break;
2412             }
2413         }
2414     }
2415
2416     if(control_values) {
2417         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2418                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2419                        control_values[0], shader->baseShader.cur_loop_depth);
2420     } else {
2421         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2422         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2423                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2424                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2425     }
2426     shader->baseShader.cur_loop_depth++;
2427 }
2428
2429 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2430 {
2431     glsl_src_param_t src0_param;
2432
2433     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2434     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2435 }
2436
2437 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2438 {
2439     glsl_src_param_t src0_param;
2440     glsl_src_param_t src1_param;
2441
2442     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2443     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2444
2445     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2446             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2447 }
2448
2449 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2450 {
2451     shader_addline(ins->ctx->buffer, "} else {\n");
2452 }
2453
2454 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2455 {
2456     shader_addline(ins->ctx->buffer, "break;\n");
2457 }
2458
2459 /* FIXME: According to MSDN the compare is done per component. */
2460 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2461 {
2462     glsl_src_param_t src0_param;
2463     glsl_src_param_t src1_param;
2464
2465     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2466     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2467
2468     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2469             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2470 }
2471
2472 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2473 {
2474     shader_addline(ins->ctx->buffer, "}\n");
2475     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2476 }
2477
2478 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2479 {
2480     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2481 }
2482
2483 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2484 {
2485     glsl_src_param_t src1_param;
2486
2487     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2488     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
2489 }
2490
2491 /*********************************************
2492  * Pixel Shader Specific Code begins here
2493  ********************************************/
2494 static void pshader_glsl_tex(const struct wined3d_shader_instruction *ins)
2495 {
2496     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2497     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2498     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2499             ins->ctx->reg_maps->shader_version.minor);
2500     glsl_sample_function_t sample_function;
2501     DWORD sample_flags = 0;
2502     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
2503     DWORD sampler_idx;
2504     DWORD mask = 0, swizzle;
2505
2506     /* 1.0-1.4: Use destination register as sampler source.
2507      * 2.0+: Use provided sampler source. */
2508     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
2509     else sampler_idx = ins->src[1].reg.idx;
2510     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2511
2512     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2513     {
2514         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2515
2516         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2517         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2518             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2519             switch (flags & ~WINED3DTTFF_PROJECTED) {
2520                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2521                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2522                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2523                 case WINED3DTTFF_COUNT4:
2524                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2525             }
2526         }
2527     }
2528     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
2529     {
2530         DWORD src_mod = ins->src[0].modifiers;
2531
2532         if (src_mod == WINED3DSPSM_DZ) {
2533             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2534             mask = WINED3DSP_WRITEMASK_2;
2535         } else if (src_mod == WINED3DSPSM_DW) {
2536             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2537             mask = WINED3DSP_WRITEMASK_3;
2538         }
2539     } else {
2540         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
2541         {
2542             /* ps 2.0 texldp instruction always divides by the fourth component. */
2543             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2544             mask = WINED3DSP_WRITEMASK_3;
2545         }
2546     }
2547
2548     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2549        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2550         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2551     }
2552
2553     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2554     mask |= sample_function.coord_mask;
2555
2556     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
2557     else swizzle = ins->src[1].swizzle;
2558
2559     /* 1.0-1.3: Use destination register as coordinate source.
2560        1.4+: Use provided coordinate source register. */
2561     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2562     {
2563         char coord_mask[6];
2564         shader_glsl_write_mask_to_str(mask, coord_mask);
2565         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2566                 "T%u%s", sampler_idx, coord_mask);
2567     } else {
2568         glsl_src_param_t coord_param;
2569         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
2570         if (ins->flags & WINED3DSI_TEXLD_BIAS)
2571         {
2572             glsl_src_param_t bias;
2573             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
2574             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
2575                     "%s", coord_param.param_str);
2576         } else {
2577             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2578                     "%s", coord_param.param_str);
2579         }
2580     }
2581 }
2582
2583 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
2584 {
2585     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2586     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2587     const WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
2588     glsl_sample_function_t sample_function;
2589     glsl_src_param_t coord_param, dx_param, dy_param;
2590     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
2591     DWORD sampler_type;
2592     DWORD sampler_idx;
2593     DWORD swizzle = ins->src[1].swizzle;
2594
2595     if(!GL_SUPPORT(ARB_SHADER_TEXTURE_LOD)) {
2596         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
2597         return pshader_glsl_tex(ins);
2598     }
2599
2600     sampler_idx = ins->src[1].reg.idx;
2601     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2602     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2603        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2604         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2605     }
2606
2607     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2608     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2609     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
2610     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
2611
2612     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
2613                                 "%s", coord_param.param_str);
2614 }
2615
2616 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
2617 {
2618     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2619     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2620     glsl_sample_function_t sample_function;
2621     glsl_src_param_t coord_param, lod_param;
2622     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
2623     DWORD sampler_type;
2624     DWORD sampler_idx;
2625     DWORD swizzle = ins->src[1].swizzle;
2626
2627     sampler_idx = ins->src[1].reg.idx;
2628     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2629     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2630        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2631         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2632     }
2633     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2634     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2635
2636     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
2637
2638     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2639     {
2640         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2641          * However, they seem to work just fine in fragment shaders as well. */
2642         WARN("Using %s in fragment shader.\n", sample_function.name);
2643     }
2644     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
2645             "%s", coord_param.param_str);
2646 }
2647
2648 static void pshader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
2649 {
2650     /* FIXME: Make this work for more than just 2D textures */
2651     SHADER_BUFFER *buffer = ins->ctx->buffer;
2652     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2653
2654     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
2655     {
2656         char dst_mask[6];
2657
2658         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2659         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
2660                 ins->dst[0].reg.idx, dst_mask);
2661     } else {
2662         DWORD reg = ins->src[0].reg.idx;
2663         DWORD src_mod = ins->src[0].modifiers;
2664         char dst_swizzle[6];
2665
2666         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
2667
2668         if (src_mod == WINED3DSPSM_DZ) {
2669             glsl_src_param_t div_param;
2670             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2671             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
2672
2673             if (mask_size > 1) {
2674                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2675             } else {
2676                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2677             }
2678         } else if (src_mod == WINED3DSPSM_DW) {
2679             glsl_src_param_t div_param;
2680             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2681             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
2682
2683             if (mask_size > 1) {
2684                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2685             } else {
2686                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2687             }
2688         } else {
2689             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2690         }
2691     }
2692 }
2693
2694 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2695  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2696  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2697 static void pshader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
2698 {
2699     glsl_src_param_t src0_param;
2700     glsl_sample_function_t sample_function;
2701     DWORD sampler_idx = ins->dst[0].reg.idx;
2702     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2703     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2704     UINT mask_size;
2705
2706     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2707
2708     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2709      * scalar, and projected sampling would require 4.
2710      *
2711      * It is a dependent read - not valid with conditional NP2 textures
2712      */
2713     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2714     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
2715
2716     switch(mask_size)
2717     {
2718         case 1:
2719             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2720                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
2721             break;
2722
2723         case 2:
2724             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2725                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
2726             break;
2727
2728         case 3:
2729             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2730                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
2731             break;
2732
2733         default:
2734             FIXME("Unexpected mask size %u\n", mask_size);
2735             break;
2736     }
2737 }
2738
2739 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2740  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2741 static void pshader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
2742 {
2743     glsl_src_param_t src0_param;
2744     DWORD dstreg = ins->dst[0].reg.idx;
2745     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2746     DWORD dst_mask;
2747     unsigned int mask_size;
2748
2749     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2750     mask_size = shader_glsl_get_write_mask_size(dst_mask);
2751     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2752
2753     if (mask_size > 1) {
2754         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2755     } else {
2756         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2757     }
2758 }
2759
2760 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2761  * Calculate the depth as dst.x / dst.y   */
2762 static void pshader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
2763 {
2764     glsl_dst_param_t dst_param;
2765
2766     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
2767
2768     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2769      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2770      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2771      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2772      * >= 1.0 or < 0.0
2773      */
2774     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
2775             dst_param.reg_name, dst_param.reg_name);
2776 }
2777
2778 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2779  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2780  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2781  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2782  */
2783 static void pshader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
2784 {
2785     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2786     DWORD dstreg = ins->dst[0].reg.idx;
2787     glsl_src_param_t src0_param;
2788
2789     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2790
2791     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2792     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2793 }
2794
2795 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2796  * Calculate the 1st of a 2-row matrix multiplication. */
2797 static void pshader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
2798 {
2799     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2800     DWORD reg = ins->dst[0].reg.idx;
2801     SHADER_BUFFER *buffer = ins->ctx->buffer;
2802     glsl_src_param_t src0_param;
2803
2804     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2805     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2806 }
2807
2808 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2809  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2810 static void pshader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
2811 {
2812     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2813     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2814     DWORD reg = ins->dst[0].reg.idx;
2815     SHADER_BUFFER *buffer = ins->ctx->buffer;
2816     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2817     glsl_src_param_t src0_param;
2818
2819     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2820     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2821     current_state->texcoord_w[current_state->current_row++] = reg;
2822 }
2823
2824 static void pshader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
2825 {
2826     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2827     DWORD reg = ins->dst[0].reg.idx;
2828     SHADER_BUFFER *buffer = ins->ctx->buffer;
2829     glsl_src_param_t src0_param;
2830     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
2831     glsl_sample_function_t sample_function;
2832
2833     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2834     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2835
2836     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2837
2838     /* Sample the texture using the calculated coordinates */
2839     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
2840 }
2841
2842 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2843  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2844 static void pshader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
2845 {
2846     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2847     glsl_src_param_t src0_param;
2848     DWORD reg = ins->dst[0].reg.idx;
2849     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2850     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2851     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
2852     glsl_sample_function_t sample_function;
2853
2854     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2855     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2856
2857     /* Dependent read, not valid with conditional NP2 */
2858     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2859
2860     /* Sample the texture using the calculated coordinates */
2861     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
2862
2863     current_state->current_row = 0;
2864 }
2865
2866 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2867  * Perform the 3rd row of a 3x3 matrix multiply */
2868 static void pshader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
2869 {
2870     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2871     glsl_src_param_t src0_param;
2872     char dst_mask[6];
2873     DWORD reg = ins->dst[0].reg.idx;
2874     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2875     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2876
2877     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2878
2879     shader_glsl_append_dst(ins->ctx->buffer, ins);
2880     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2881     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2882
2883     current_state->current_row = 0;
2884 }
2885
2886 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2887  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2888 static void pshader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
2889 {
2890     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2891     DWORD reg = ins->dst[0].reg.idx;
2892     glsl_src_param_t src0_param;
2893     glsl_src_param_t src1_param;
2894     SHADER_BUFFER *buffer = ins->ctx->buffer;
2895     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2896     WINED3DSAMPLER_TEXTURE_TYPE stype = ins->ctx->reg_maps->sampler_type[reg];
2897     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2898     glsl_sample_function_t sample_function;
2899
2900     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2901     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2902
2903     /* Perform the last matrix multiply operation */
2904     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2905     /* Reflection calculation */
2906     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2907
2908     /* Dependent read, not valid with conditional NP2 */
2909     shader_glsl_get_sample_function(stype, 0, &sample_function);
2910
2911     /* Sample the texture */
2912     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
2913
2914     current_state->current_row = 0;
2915 }
2916
2917 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2918  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2919 static void pshader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2920 {
2921     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2922     DWORD reg = ins->dst[0].reg.idx;
2923     SHADER_BUFFER *buffer = ins->ctx->buffer;
2924     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2925     glsl_src_param_t src0_param;
2926     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2927     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
2928     glsl_sample_function_t sample_function;
2929
2930     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2931
2932     /* Perform the last matrix multiply operation */
2933     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2934
2935     /* Construct the eye-ray vector from w coordinates */
2936     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2937             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2938     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2939
2940     /* Dependent read, not valid with conditional NP2 */
2941     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2942
2943     /* Sample the texture using the calculated coordinates */
2944     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
2945
2946     current_state->current_row = 0;
2947 }
2948
2949 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2950  * Apply a fake bump map transform.
2951  * texbem is pshader <= 1.3 only, this saves a few version checks
2952  */
2953 static void pshader_glsl_texbem(const struct wined3d_shader_instruction *ins)
2954 {
2955     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2956     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2957     glsl_sample_function_t sample_function;
2958     glsl_src_param_t coord_param;
2959     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
2960     DWORD sampler_idx;
2961     DWORD mask;
2962     DWORD flags;
2963     char coord_mask[6];
2964
2965     sampler_idx = ins->dst[0].reg.idx;
2966     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2967
2968     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2969     /* Dependent read, not valid with conditional NP2 */
2970     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2971     mask = sample_function.coord_mask;
2972
2973     shader_glsl_write_mask_to_str(mask, coord_mask);
2974
2975     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2976          * so we can't let the GL handle this.
2977          */
2978     if (flags & WINED3DTTFF_PROJECTED) {
2979         DWORD div_mask=0;
2980         char coord_div_mask[3];
2981         switch (flags & ~WINED3DTTFF_PROJECTED) {
2982             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2983             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2984             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2985             case WINED3DTTFF_COUNT4:
2986             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2987         }
2988         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
2989         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2990     }
2991
2992     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
2993
2994     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2995             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
2996             coord_param.param_str, coord_mask);
2997
2998     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2999     {
3000         glsl_src_param_t luminance_param;
3001         glsl_dst_param_t dst_param;
3002
3003         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3004         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3005
3006         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3007                 dst_param.reg_name, dst_param.mask_str,
3008                 luminance_param.param_str, sampler_idx, sampler_idx);
3009     }
3010 }
3011
3012 static void pshader_glsl_bem(const struct wined3d_shader_instruction *ins)
3013 {
3014     glsl_src_param_t src0_param, src1_param;
3015     DWORD sampler_idx = ins->dst[0].reg.idx;
3016
3017     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3018     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3019
3020     shader_glsl_append_dst(ins->ctx->buffer, ins);
3021     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3022             src0_param.param_str, sampler_idx, src1_param.param_str);
3023 }
3024
3025 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3026  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3027 static void pshader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3028 {
3029     glsl_src_param_t src0_param;
3030     DWORD sampler_idx = ins->dst[0].reg.idx;
3031     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3032     glsl_sample_function_t sample_function;
3033
3034     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3035
3036     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3037     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3038             "%s.wx", src0_param.reg_name);
3039 }
3040
3041 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3042  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3043 static void pshader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3044 {
3045     glsl_src_param_t src0_param;
3046     DWORD sampler_idx = ins->dst[0].reg.idx;
3047     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3048     glsl_sample_function_t sample_function;
3049
3050     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3051
3052     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3053     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3054             "%s.yz", src0_param.reg_name);
3055 }
3056
3057 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3058  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3059 static void pshader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3060 {
3061     glsl_src_param_t src0_param;
3062     DWORD sampler_idx = ins->dst[0].reg.idx;
3063     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3064     glsl_sample_function_t sample_function;
3065
3066     /* Dependent read, not valid with conditional NP2 */
3067     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3068     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3069
3070     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3071             "%s", src0_param.param_str);
3072 }
3073
3074 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3075  * If any of the first 3 components are < 0, discard this pixel */
3076 static void pshader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3077 {
3078     glsl_dst_param_t dst_param;
3079
3080     /* The argument is a destination parameter, and no writemasks are allowed */
3081     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3082     if (ins->ctx->reg_maps->shader_version.major >= 2)
3083     {
3084         /* 2.0 shaders compare all 4 components in texkill */
3085         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3086     } else {
3087         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3088          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3089          * 4 components are defined, only the first 3 are used
3090          */
3091         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3092     }
3093 }
3094
3095 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3096  * dst = dot2(src0, src1) + src2 */
3097 static void pshader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3098 {
3099     glsl_src_param_t src0_param;
3100     glsl_src_param_t src1_param;
3101     glsl_src_param_t src2_param;
3102     DWORD write_mask;
3103     unsigned int mask_size;
3104
3105     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3106     mask_size = shader_glsl_get_write_mask_size(write_mask);
3107
3108     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3109     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3110     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3111
3112     if (mask_size > 1) {
3113         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3114                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3115     } else {
3116         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3117                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3118     }
3119 }
3120
3121 static void pshader_glsl_input_pack(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer,
3122         const struct wined3d_shader_semantic *semantics_in, const struct shader_reg_maps *reg_maps,
3123         enum vertexprocessing_mode vertexprocessing)
3124 {
3125     unsigned int i;
3126     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3127
3128     for (i = 0; i < MAX_REG_INPUT; ++i)
3129     {
3130         DWORD usage, usage_idx;
3131         char reg_mask[6];
3132
3133         /* Unused */
3134         if (!reg_maps->packed_input[i]) continue;
3135
3136         usage = semantics_in[i].usage;
3137         usage_idx = semantics_in[i].usage_idx;
3138         shader_glsl_get_write_mask(&semantics_in[i].reg, reg_mask);
3139
3140         switch (usage)
3141         {
3142             case WINED3DDECLUSAGE_TEXCOORD:
3143                 if (usage_idx < 8 && vertexprocessing == pretransformed)
3144                     shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3145                             This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
3146                 else
3147                     shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3148                             This->input_reg_map[i], reg_mask, reg_mask);
3149                 break;
3150
3151             case WINED3DDECLUSAGE_COLOR:
3152                 if (usage_idx == 0)
3153                     shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3154                             This->input_reg_map[i], reg_mask, reg_mask);
3155                 else if (usage_idx == 1)
3156                     shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3157                             This->input_reg_map[i], reg_mask, reg_mask);
3158                 else
3159                     shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3160                             This->input_reg_map[i], reg_mask, reg_mask);
3161                 break;
3162
3163             default:
3164                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3165                         This->input_reg_map[i], reg_mask, reg_mask);
3166                 break;
3167         }
3168     }
3169 }
3170
3171 /*********************************************
3172  * Vertex Shader Specific Code begins here
3173  ********************************************/
3174
3175 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3176     glsl_program_key_t *key;
3177
3178     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
3179     key->vshader = entry->vshader;
3180     key->pshader = entry->pshader;
3181     key->vs_args = entry->vs_args;
3182     key->ps_args = entry->ps_args;
3183
3184     hash_table_put(priv->glsl_program_lookup, key, entry);
3185 }
3186
3187 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3188         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3189         struct ps_compile_args *ps_args) {
3190     glsl_program_key_t key;
3191
3192     key.vshader = vshader;
3193     key.pshader = pshader;
3194     key.vs_args = *vs_args;
3195     key.ps_args = *ps_args;
3196
3197     return hash_table_get(priv->glsl_program_lookup, &key);
3198 }
3199
3200 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const WineD3D_GL_Info *gl_info,
3201         struct glsl_shader_prog_link *entry)
3202 {
3203     glsl_program_key_t *key;
3204
3205     key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
3206     key->vshader = entry->vshader;
3207     key->pshader = entry->pshader;
3208     key->vs_args = entry->vs_args;
3209     key->ps_args = entry->ps_args;
3210     hash_table_remove(priv->glsl_program_lookup, key);
3211
3212     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3213     if (entry->vshader) list_remove(&entry->vshader_entry);
3214     if (entry->pshader) list_remove(&entry->pshader_entry);
3215     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3216     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3217     HeapFree(GetProcessHeap(), 0, entry);
3218 }
3219
3220 static void handle_ps3_input(SHADER_BUFFER *buffer, const WineD3D_GL_Info *gl_info, const DWORD *map,
3221         const struct wined3d_shader_semantic *semantics_in, const struct shader_reg_maps *reg_maps_in,
3222         const struct wined3d_shader_semantic *semantics_out, const struct shader_reg_maps *reg_maps_out)
3223 {
3224     unsigned int i, j;
3225     DWORD usage, usage_idx, usage_out, usage_idx_out;
3226     DWORD *set;
3227     DWORD in_idx;
3228     DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
3229     char reg_mask[6], reg_mask_out[6];
3230     char destination[50];
3231
3232     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3233
3234     if (!semantics_out) {
3235         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3236         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3237         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3238     }
3239
3240     for(i = 0; i < MAX_REG_INPUT; i++) {
3241         if (!reg_maps_in->packed_input[i]) continue;
3242
3243         in_idx = map[i];
3244         if (in_idx >= (in_count + 2)) {
3245             FIXME("More input varyings declared than supported, expect issues\n");
3246             continue;
3247         }
3248         else if (map[i] == ~0U)
3249         {
3250             /* Declared, but not read register */
3251             continue;
3252         }
3253
3254         if (in_idx == in_count) {
3255             sprintf(destination, "gl_FrontColor");
3256         } else if (in_idx == in_count + 1) {
3257             sprintf(destination, "gl_FrontSecondaryColor");
3258         } else {
3259             sprintf(destination, "IN[%u]", in_idx);
3260         }
3261
3262         usage = semantics_in[i].usage;
3263         usage_idx = semantics_in[i].usage_idx;
3264         set[map[i]] = shader_glsl_get_write_mask(&semantics_in[i].reg, reg_mask);
3265
3266         if(!semantics_out) {
3267             switch(usage) {
3268                 case WINED3DDECLUSAGE_COLOR:
3269                     if (usage_idx == 0)
3270                         shader_addline(buffer, "%s%s = front_color%s;\n",
3271                                        destination, reg_mask, reg_mask);
3272                     else if (usage_idx == 1)
3273                         shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3274                                        destination, reg_mask, reg_mask);
3275                     else
3276                         shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3277                                        destination, reg_mask, reg_mask);
3278                     break;
3279
3280                 case WINED3DDECLUSAGE_TEXCOORD:
3281                     if (usage_idx < 8) {
3282                         shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3283                                        destination, reg_mask, usage_idx, reg_mask);
3284                     } else {
3285                         shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3286                                        destination, reg_mask, reg_mask);
3287                     }
3288                     break;
3289
3290                 case WINED3DDECLUSAGE_FOG:
3291                     shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3292                                    destination, reg_mask, reg_mask);
3293                     break;
3294
3295                 default:
3296                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3297                                    destination, reg_mask, reg_mask);
3298             }
3299         } else {
3300             BOOL found = FALSE;
3301             for(j = 0; j < MAX_REG_OUTPUT; j++) {
3302                 if (!reg_maps_out->packed_output[j]) continue;
3303
3304                 usage_out = semantics_out[j].usage;
3305                 usage_idx_out = semantics_out[j].usage_idx;
3306                 shader_glsl_get_write_mask(&semantics_out[j].reg, reg_mask_out);
3307
3308                 if(usage == usage_out &&
3309                    usage_idx == usage_idx_out) {
3310                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3311                                    destination, reg_mask, j, reg_mask);
3312                     found = TRUE;
3313                 }
3314             }
3315             if(!found) {
3316                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3317                                destination, reg_mask, reg_mask);
3318             }
3319         }
3320     }
3321
3322     /* This is solely to make the compiler / linker happy and avoid warning about undefined
3323      * varyings. It shouldn't result in any real code executed on the GPU, since all read
3324      * input varyings are assigned above, if the optimizer works properly.
3325      */
3326     for(i = 0; i < in_count + 2; i++) {
3327         if(set[i] != WINED3DSP_WRITEMASK_ALL) {
3328             unsigned int size = 0;
3329             memset(reg_mask, 0, sizeof(reg_mask));
3330             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3331                 reg_mask[size] = 'x';
3332                 size++;
3333             }
3334             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3335                 reg_mask[size] = 'y';
3336                 size++;
3337             }
3338             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3339                 reg_mask[size] = 'z';
3340                 size++;
3341             }
3342             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3343                 reg_mask[size] = 'w';
3344                 size++;
3345             }
3346
3347             if (i == in_count) {
3348                 sprintf(destination, "gl_FrontColor");
3349             } else if (i == in_count + 1) {
3350                 sprintf(destination, "gl_FrontSecondaryColor");
3351             } else {
3352                 sprintf(destination, "IN[%u]", i);
3353             }
3354
3355             if (size == 1) {
3356                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3357             } else {
3358                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3359             }
3360         }
3361     }
3362
3363     HeapFree(GetProcessHeap(), 0, set);
3364 }
3365
3366 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
3367         IWineD3DPixelShader *pixelshader, const WineD3D_GL_Info *gl_info)
3368 {
3369     GLhandleARB ret = 0;
3370     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3371     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3372     IWineD3DDeviceImpl *device;
3373     DWORD vs_major = vs->baseShader.reg_maps.shader_version.major;
3374     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3375     unsigned int i;
3376     SHADER_BUFFER buffer;
3377     DWORD usage, usage_idx, writemask;
3378     char reg_mask[6];
3379     const struct wined3d_shader_semantic *semantics_out;
3380
3381     shader_buffer_init(&buffer);
3382
3383     shader_addline(&buffer, "#version 120\n");
3384
3385     if(vs_major < 3 && ps_major < 3) {
3386         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3387          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3388          */
3389         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3390         if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0 &&
3391             !device->frag_pipe->ffp_proj_control) {
3392             shader_addline(&buffer, "void order_ps_input() {\n");
3393             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3394                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3395                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3396                     shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3397                 }
3398             }
3399             shader_addline(&buffer, "}\n");
3400         } else {
3401             shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
3402         }
3403     } else if(ps_major < 3 && vs_major >= 3) {
3404         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3405         semantics_out = vs->semantics_out;
3406
3407         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3408         for(i = 0; i < MAX_REG_OUTPUT; i++) {
3409             if (!vs->baseShader.reg_maps.packed_output[i]) continue;
3410
3411             usage = semantics_out[i].usage;
3412             usage_idx = semantics_out[i].usage_idx;
3413             writemask = shader_glsl_get_write_mask(&semantics_out[i].reg, reg_mask);
3414
3415             switch(usage) {
3416                 case WINED3DDECLUSAGE_COLOR:
3417                     if (usage_idx == 0)
3418                         shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3419                     else if (usage_idx == 1)
3420                         shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3421                     break;
3422
3423                 case WINED3DDECLUSAGE_POSITION:
3424                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3425                     break;
3426
3427                 case WINED3DDECLUSAGE_TEXCOORD:
3428                     if (usage_idx < 8) {
3429                         if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
3430
3431                         shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3432                                         usage_idx, reg_mask, i, reg_mask);
3433                         if(!(writemask & WINED3DSP_WRITEMASK_3)) {
3434                             shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
3435                         }
3436                     }
3437                     break;
3438
3439                 case WINED3DDECLUSAGE_PSIZE:
3440                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3441                     break;
3442
3443                 case WINED3DDECLUSAGE_FOG:
3444                     shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3445                     break;
3446
3447                 default:
3448                     break;
3449             }
3450         }
3451         shader_addline(&buffer, "}\n");
3452
3453     } else if(ps_major >= 3 && vs_major >= 3) {
3454         semantics_out = vs->semantics_out;
3455
3456         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3457         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3458         shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3459
3460         /* First, sort out position and point size. Those are not passed to the pixel shader */
3461         for(i = 0; i < MAX_REG_OUTPUT; i++) {
3462             if (!vs->baseShader.reg_maps.packed_output[i]) continue;
3463
3464             usage = semantics_out[i].usage;
3465             usage_idx = semantics_out[i].usage_idx;
3466             shader_glsl_get_write_mask(&semantics_out[i].reg, reg_mask);
3467
3468             switch(usage) {
3469                 case WINED3DDECLUSAGE_POSITION:
3470                     shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3471                     break;
3472
3473                 case WINED3DDECLUSAGE_PSIZE:
3474                     shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3475                     break;
3476
3477                 default:
3478                     break;
3479             }
3480         }
3481
3482         /* Then, fix the pixel shader input */
3483         handle_ps3_input(&buffer, gl_info, ps->input_reg_map,
3484                 ps->semantics_in, &ps->baseShader.reg_maps, semantics_out, &vs->baseShader.reg_maps);
3485
3486         shader_addline(&buffer, "}\n");
3487     } else if(ps_major >= 3 && vs_major < 3) {
3488         shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3489         shader_addline(&buffer, "void order_ps_input() {\n");
3490         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3491          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3492          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3493          */
3494         handle_ps3_input(&buffer, gl_info, ps->input_reg_map, ps->semantics_in, &ps->baseShader.reg_maps, NULL, NULL);
3495         shader_addline(&buffer, "}\n");
3496     } else {
3497         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3498     }
3499
3500     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3501     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3502     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3503     checkGLcall("glShaderSourceARB(ret, 1, &buffer.buffer, NULL)");
3504     GL_EXTCALL(glCompileShaderARB(ret));
3505     checkGLcall("glCompileShaderARB(ret)");
3506
3507     shader_buffer_free(&buffer);
3508     return ret;
3509 }
3510
3511 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const WineD3D_GL_Info *gl_info,
3512         GLhandleARB programId, char prefix)
3513 {
3514     const local_constant *lconst;
3515     GLint tmp_loc;
3516     const float *value;
3517     char glsl_name[8];
3518
3519     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3520         value = (const float *)lconst->value;
3521         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3522         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3523         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3524     }
3525     checkGLcall("Hardcoding local constants\n");
3526 }
3527
3528 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3529  * It sets the programId on the current StateBlock (because it should be called
3530  * inside of the DrawPrimitive() part of the render loop).
3531  *
3532  * If a program for the given combination does not exist, create one, and store
3533  * the program in the hash table.  If it creates a program, it will link the
3534  * given objects, too.
3535  */
3536 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3537     IWineD3DDeviceImpl *This               = (IWineD3DDeviceImpl *)iface;
3538     struct shader_glsl_priv *priv          = This->shader_priv;
3539     const WineD3D_GL_Info *gl_info         = &This->adapter->gl_info;
3540     IWineD3DPixelShader  *pshader          = This->stateBlock->pixelShader;
3541     IWineD3DVertexShader *vshader          = This->stateBlock->vertexShader;
3542     struct glsl_shader_prog_link *entry    = NULL;
3543     GLhandleARB programId                  = 0;
3544     GLhandleARB reorder_shader_id          = 0;
3545     unsigned int i;
3546     char glsl_name[8];
3547     GLhandleARB vshader_id, pshader_id;
3548     struct ps_compile_args ps_compile_args;
3549     struct vs_compile_args vs_compile_args;
3550
3551     if(use_vs) {
3552         find_vs_compile_args((IWineD3DVertexShaderImpl*)This->stateBlock->vertexShader, This->stateBlock, &vs_compile_args);
3553     } else {
3554         /* FIXME: Do we really have to spend CPU cycles to generate a few zeroed bytes? */
3555         memset(&vs_compile_args, 0, sizeof(vs_compile_args));
3556     }
3557     if(use_ps) {
3558         find_ps_compile_args((IWineD3DPixelShaderImpl*)This->stateBlock->pixelShader, This->stateBlock, &ps_compile_args);
3559     } else {
3560         /* FIXME: Do we really have to spend CPU cycles to generate a few zeroed bytes? */
3561         memset(&ps_compile_args, 0, sizeof(ps_compile_args));
3562     }
3563     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
3564     if (entry) {
3565         priv->glsl_program = entry;
3566         return;
3567     }
3568
3569     /* If we get to this point, then no matching program exists, so we create one */
3570     programId = GL_EXTCALL(glCreateProgramObjectARB());
3571     TRACE("Created new GLSL shader program %u\n", programId);
3572
3573     /* Create the entry */
3574     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3575     entry->programId = programId;
3576     entry->vshader = vshader;
3577     entry->pshader = pshader;
3578     entry->vs_args = vs_compile_args;
3579     entry->ps_args = ps_compile_args;
3580     entry->constant_version = 0;
3581     /* Add the hash table entry */
3582     add_glsl_program_entry(priv, entry);
3583
3584     /* Set the current program */
3585     priv->glsl_program = entry;
3586
3587     if(use_vs) {
3588         vshader_id = find_gl_vshader((IWineD3DVertexShaderImpl *) vshader, &vs_compile_args);
3589     } else {
3590         vshader_id = 0;
3591     }
3592
3593     /* Attach GLSL vshader */
3594     if (vshader_id) {
3595         const unsigned int max_attribs = 16;   /* TODO: Will this always be the case? It is at the moment... */
3596         char tmp_name[10];
3597
3598         reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3599         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3600         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3601         checkGLcall("glAttachObjectARB");
3602         /* Flag the reorder function for deletion, then it will be freed automatically when the program
3603          * is destroyed
3604          */
3605         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3606
3607         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3608         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3609         checkGLcall("glAttachObjectARB");
3610
3611         /* Bind vertex attributes to a corresponding index number to match
3612          * the same index numbers as ARB_vertex_programs (makes loading
3613          * vertex attributes simpler).  With this method, we can use the
3614          * exact same code to load the attributes later for both ARB and
3615          * GLSL shaders.
3616          *
3617          * We have to do this here because we need to know the Program ID
3618          * in order to make the bindings work, and it has to be done prior
3619          * to linking the GLSL program. */
3620         for (i = 0; i < max_attribs; ++i) {
3621             if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3622                 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3623                 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3624             }
3625         }
3626         checkGLcall("glBindAttribLocationARB");
3627
3628         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3629     }
3630
3631     if(use_ps) {
3632         pshader_id = find_gl_pshader((IWineD3DPixelShaderImpl *) pshader, &ps_compile_args);
3633     } else {
3634         pshader_id = 0;
3635     }
3636
3637     /* Attach GLSL pshader */
3638     if (pshader_id) {
3639         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3640         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3641         checkGLcall("glAttachObjectARB");
3642
3643         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3644     }
3645
3646     /* Link the program */
3647     TRACE("Linking GLSL shader program %u\n", programId);
3648     GL_EXTCALL(glLinkProgramARB(programId));
3649     print_glsl_info_log(&GLINFO_LOCATION, programId);
3650
3651     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3652     for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3653         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3654         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3655     }
3656     for (i = 0; i < MAX_CONST_I; ++i) {
3657         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3658         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3659     }
3660     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3661     for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3662         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3663         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3664     }
3665     for (i = 0; i < MAX_CONST_I; ++i) {
3666         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3667         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3668     }
3669
3670     if(pshader) {
3671         for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3672             char name[32];
3673             sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3674             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3675             sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3676             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3677             sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3678             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3679         }
3680     }
3681
3682     if (use_ps && ps_compile_args.np2_fixup) {
3683         char name[32];
3684         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3685             if (ps_compile_args.np2_fixup & (1 << i)) {
3686                 sprintf(name, "PsamplerNP2Fixup%u", i);
3687                 entry->np2Fixup_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3688             } else {
3689                 entry->np2Fixup_location[i] = -1;
3690             }
3691         }
3692     }
3693
3694     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3695     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3696     checkGLcall("Find glsl program uniform locations");
3697
3698     if (pshader
3699             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
3700             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > GL_LIMITS(glsl_varyings) / 4)
3701     {
3702         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
3703         entry->vertex_color_clamp = GL_FALSE;
3704     } else {
3705         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
3706     }
3707
3708     /* Set the shader to allow uniform loading on it */
3709     GL_EXTCALL(glUseProgramObjectARB(programId));
3710     checkGLcall("glUseProgramObjectARB(programId)");
3711
3712     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3713      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3714      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3715      * vertex shader with fixed function pixel processing is used we make sure that the card
3716      * supports enough samplers to allow the max number of vertex samplers with all possible
3717      * fixed function fragment processing setups. So once the program is linked these samplers
3718      * won't change.
3719      */
3720     if(vshader_id) {
3721         /* Load vertex shader samplers */
3722         shader_glsl_load_vsamplers(gl_info, This->texUnitMap, programId);
3723     }
3724     if(pshader_id) {
3725         /* Load pixel shader samplers */
3726         shader_glsl_load_psamplers(gl_info, This->texUnitMap, programId);
3727     }
3728
3729     /* If the local constants do not have to be loaded with the environment constants,
3730      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
3731      * later
3732      */
3733     if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
3734         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
3735     }
3736     if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
3737         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
3738     }
3739 }
3740
3741 static GLhandleARB create_glsl_blt_shader(const WineD3D_GL_Info *gl_info, enum tex_types tex_type)
3742 {
3743     GLhandleARB program_id;
3744     GLhandleARB vshader_id, pshader_id;
3745     static const char *blt_vshader[] =
3746     {
3747         "#version 120\n"
3748         "void main(void)\n"
3749         "{\n"
3750         "    gl_Position = gl_Vertex;\n"
3751         "    gl_FrontColor = vec4(1.0);\n"
3752         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
3753         "}\n"
3754     };
3755
3756     static const char *blt_pshaders[tex_type_count] =
3757     {
3758         /* tex_1d */
3759         NULL,
3760         /* tex_2d */
3761         "#version 120\n"
3762         "uniform sampler2D sampler;\n"
3763         "void main(void)\n"
3764         "{\n"
3765         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3766         "}\n",
3767         /* tex_3d */
3768         NULL,
3769         /* tex_cube */
3770         "#version 120\n"
3771         "uniform samplerCube sampler;\n"
3772         "void main(void)\n"
3773         "{\n"
3774         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
3775         "}\n",
3776         /* tex_rect */
3777         "#version 120\n"
3778         "#extension GL_ARB_texture_rectangle : enable\n"
3779         "uniform sampler2DRect sampler;\n"
3780         "void main(void)\n"
3781         "{\n"
3782         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
3783         "}\n",
3784     };
3785
3786     if (!blt_pshaders[tex_type])
3787     {
3788         FIXME("tex_type %#x not supported\n", tex_type);
3789         tex_type = tex_2d;
3790     }
3791
3792     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3793     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3794     GL_EXTCALL(glCompileShaderARB(vshader_id));
3795
3796     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3797     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
3798     GL_EXTCALL(glCompileShaderARB(pshader_id));
3799
3800     program_id = GL_EXTCALL(glCreateProgramObjectARB());
3801     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3802     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3803     GL_EXTCALL(glLinkProgramARB(program_id));
3804
3805     print_glsl_info_log(&GLINFO_LOCATION, program_id);
3806
3807     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3808      * is destroyed
3809      */
3810     GL_EXTCALL(glDeleteObjectARB(vshader_id));
3811     GL_EXTCALL(glDeleteObjectARB(pshader_id));
3812     return program_id;
3813 }
3814
3815 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3816     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3817     struct shader_glsl_priv *priv = This->shader_priv;
3818     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3819     GLhandleARB program_id = 0;
3820     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
3821
3822     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3823
3824     if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3825     else priv->glsl_program = NULL;
3826
3827     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3828
3829     if (old_vertex_color_clamp != current_vertex_color_clamp) {
3830         if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
3831             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
3832             checkGLcall("glClampColorARB");
3833         } else {
3834             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
3835         }
3836     }
3837
3838     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3839     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3840     GL_EXTCALL(glUseProgramObjectARB(program_id));
3841     checkGLcall("glUseProgramObjectARB");
3842 }
3843
3844 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
3845     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3846     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3847     struct shader_glsl_priv *priv = This->shader_priv;
3848     GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
3849
3850     if (!*blt_program) {
3851         GLint loc;
3852         *blt_program = create_glsl_blt_shader(gl_info, tex_type);
3853         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
3854         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3855         GL_EXTCALL(glUniform1iARB(loc, 0));
3856     } else {
3857         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3858     }
3859 }
3860
3861 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
3862     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3863     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3864     struct shader_glsl_priv *priv = This->shader_priv;
3865     GLhandleARB program_id;
3866
3867     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3868     if (program_id) TRACE("Using GLSL program %u\n", program_id);
3869
3870     GL_EXTCALL(glUseProgramObjectARB(program_id));
3871     checkGLcall("glUseProgramObjectARB");
3872 }
3873
3874 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3875     const struct list *linked_programs;
3876     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3877     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3878     struct shader_glsl_priv *priv = device->shader_priv;
3879     const WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
3880     IWineD3DPixelShaderImpl *ps = NULL;
3881     IWineD3DVertexShaderImpl *vs = NULL;
3882
3883     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3884      * can be called from IWineD3DBaseShader::Release
3885      */
3886     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
3887
3888     if(pshader) {
3889         ps = (IWineD3DPixelShaderImpl *) This;
3890         if(ps->num_gl_shaders == 0) return;
3891         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
3892             shader_glsl_select(This->baseShader.device, FALSE, FALSE);
3893     } else {
3894         vs = (IWineD3DVertexShaderImpl *) This;
3895         if(vs->num_gl_shaders == 0) return;
3896         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
3897             shader_glsl_select(This->baseShader.device, FALSE, FALSE);
3898     }
3899
3900     linked_programs = &This->baseShader.linked_programs;
3901
3902     TRACE("Deleting linked programs\n");
3903     if (linked_programs->next) {
3904         struct glsl_shader_prog_link *entry, *entry2;
3905
3906         if(pshader) {
3907             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3908                 delete_glsl_program_entry(priv, gl_info, entry);
3909             }
3910         } else {
3911             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3912                 delete_glsl_program_entry(priv, gl_info, entry);
3913             }
3914         }
3915     }
3916
3917     if(pshader) {
3918         UINT i;
3919
3920         ENTER_GL();
3921         for(i = 0; i < ps->num_gl_shaders; i++) {
3922             TRACE("deleting pshader %u\n", ps->gl_shaders[i].prgId);
3923             GL_EXTCALL(glDeleteObjectARB(ps->gl_shaders[i].prgId));
3924             checkGLcall("glDeleteObjectARB");
3925         }
3926         LEAVE_GL();
3927         HeapFree(GetProcessHeap(), 0, ps->gl_shaders);
3928         ps->gl_shaders = NULL;
3929         ps->num_gl_shaders = 0;
3930         ps->shader_array_size = 0;
3931     } else {
3932         UINT i;
3933
3934         ENTER_GL();
3935         for(i = 0; i < vs->num_gl_shaders; i++) {
3936             TRACE("deleting vshader %u\n", vs->gl_shaders[i].prgId);
3937             GL_EXTCALL(glDeleteObjectARB(vs->gl_shaders[i].prgId));
3938             checkGLcall("glDeleteObjectARB");
3939         }
3940         LEAVE_GL();
3941         HeapFree(GetProcessHeap(), 0, vs->gl_shaders);
3942         vs->gl_shaders = NULL;
3943         vs->num_gl_shaders = 0;
3944         vs->shader_array_size = 0;
3945     }
3946 }
3947
3948 static unsigned int glsl_program_key_hash(const void *key)
3949 {
3950     const glsl_program_key_t *k = key;
3951
3952     unsigned int hash = ((DWORD_PTR) k->vshader) | ((DWORD_PTR) k->pshader) << 16;
3953     hash += ~(hash << 15);
3954     hash ^=  (hash >> 10);
3955     hash +=  (hash << 3);
3956     hash ^=  (hash >> 6);
3957     hash += ~(hash << 11);
3958     hash ^=  (hash >> 16);
3959
3960     return hash;
3961 }
3962
3963 static BOOL glsl_program_key_compare(const void *keya, const void *keyb)
3964 {
3965     const glsl_program_key_t *ka = keya;
3966     const glsl_program_key_t *kb = keyb;
3967
3968     return ka->vshader == kb->vshader && ka->pshader == kb->pshader &&
3969            (memcmp(&ka->ps_args, &kb->ps_args, sizeof(kb->ps_args)) == 0) &&
3970            (memcmp(&ka->vs_args, &kb->vs_args, sizeof(kb->vs_args)) == 0);
3971 }
3972
3973 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
3974 {
3975     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
3976     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
3977
3978     if (!mem)
3979     {
3980         ERR("Failed to allocate memory\n");
3981         return FALSE;
3982     }
3983
3984     heap->entries = mem;
3985     heap->entries[1].version = 0;
3986     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
3987     heap->size = 1;
3988
3989     return TRUE;
3990 }
3991
3992 static void constant_heap_free(struct constant_heap *heap)
3993 {
3994     HeapFree(GetProcessHeap(), 0, heap->entries);
3995 }
3996
3997 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3998     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3999     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
4000     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4001     SIZE_T stack_size = wined3d_log2i(max(GL_LIMITS(vshader_constantsF), GL_LIMITS(pshader_constantsF))) + 1;
4002
4003     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4004     if (!priv->stack)
4005     {
4006         ERR("Failed to allocate memory.\n");
4007         HeapFree(GetProcessHeap(), 0, priv);
4008         return E_OUTOFMEMORY;
4009     }
4010
4011     if (!constant_heap_init(&priv->vconst_heap, GL_LIMITS(vshader_constantsF)))
4012     {
4013         ERR("Failed to initialize vertex shader constant heap\n");
4014         HeapFree(GetProcessHeap(), 0, priv->stack);
4015         HeapFree(GetProcessHeap(), 0, priv);
4016         return E_OUTOFMEMORY;
4017     }
4018
4019     if (!constant_heap_init(&priv->pconst_heap, GL_LIMITS(pshader_constantsF)))
4020     {
4021         ERR("Failed to initialize pixel shader constant heap\n");
4022         constant_heap_free(&priv->vconst_heap);
4023         HeapFree(GetProcessHeap(), 0, priv->stack);
4024         HeapFree(GetProcessHeap(), 0, priv);
4025         return E_OUTOFMEMORY;
4026     }
4027
4028     priv->glsl_program_lookup = hash_table_create(glsl_program_key_hash, glsl_program_key_compare);
4029     priv->next_constant_version = 1;
4030
4031     This->shader_priv = priv;
4032     return WINED3D_OK;
4033 }
4034
4035 static void shader_glsl_free(IWineD3DDevice *iface) {
4036     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4037     const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
4038     struct shader_glsl_priv *priv = This->shader_priv;
4039     int i;
4040
4041     for (i = 0; i < tex_type_count; ++i)
4042     {
4043         if (priv->depth_blt_program[i])
4044         {
4045             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
4046         }
4047     }
4048
4049     hash_table_destroy(priv->glsl_program_lookup, NULL, NULL);
4050     constant_heap_free(&priv->pconst_heap);
4051     constant_heap_free(&priv->vconst_heap);
4052
4053     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4054     This->shader_priv = NULL;
4055 }
4056
4057 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4058     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4059     return FALSE;
4060 }
4061
4062 static GLuint shader_glsl_generate_pshader(IWineD3DPixelShader *iface,
4063         SHADER_BUFFER *buffer, const struct ps_compile_args *args)
4064 {
4065     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
4066     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4067     CONST DWORD *function = This->baseShader.function;
4068     const char *fragcolor;
4069     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
4070
4071     /* Create the hw GLSL shader object and assign it as the shader->prgId */
4072     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4073
4074     shader_addline(buffer, "#version 120\n");
4075
4076     if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
4077         shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
4078     }
4079     if(GL_SUPPORT(ARB_SHADER_TEXTURE_LOD) && reg_maps->usestexldd) {
4080         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
4081     }
4082     if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
4083         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
4084          * drivers write a warning if we don't do so
4085          */
4086         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
4087     }
4088
4089     /* Base Declarations */
4090     shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION, args);
4091
4092     /* Pack 3.0 inputs */
4093     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4094     {
4095         pshader_glsl_input_pack(iface, buffer, This->semantics_in, reg_maps, args->vp_mode);
4096     }
4097
4098     /* Base Shader Body */
4099     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function);
4100
4101     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4102     if (reg_maps->shader_version.major < 2)
4103     {
4104         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4105         if(GL_SUPPORT(ARB_DRAW_BUFFERS))
4106             shader_addline(buffer, "gl_FragData[0] = R0;\n");
4107         else
4108             shader_addline(buffer, "gl_FragColor = R0;\n");
4109     }
4110
4111     if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
4112         fragcolor = "gl_FragData[0]";
4113     } else {
4114         fragcolor = "gl_FragColor";
4115     }
4116     if(args->srgb_correction) {
4117         shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
4118                         fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
4119                         srgb_sub_high, srgb_sub_high, srgb_sub_high);
4120         shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
4121         shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
4122         shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
4123         shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
4124         shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
4125     }
4126     /* Pixel shader < 3.0 do not replace the fog stage.
4127      * This implements linear fog computation and blending.
4128      * TODO: non linear fog
4129      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4130      * -1/(e-s) and e/(e-s) respectively.
4131      */
4132     if (reg_maps->shader_version.major < 3)
4133     {
4134         switch(args->fog) {
4135             case FOG_OFF: break;
4136             case FOG_LINEAR:
4137                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4138                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4139                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4140                 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
4141                 break;
4142             case FOG_EXP:
4143                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4144                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4145                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4146                 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
4147                 break;
4148             case FOG_EXP2:
4149                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4150                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4151                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4152                 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
4153                 break;
4154         }
4155     }
4156
4157     shader_addline(buffer, "}\n");
4158
4159     TRACE("Compiling shader object %u\n", shader_obj);
4160     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4161     GL_EXTCALL(glCompileShaderARB(shader_obj));
4162     print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
4163
4164     /* Store the shader object */
4165     return shader_obj;
4166 }
4167
4168 static GLuint shader_glsl_generate_vshader(IWineD3DVertexShader *iface,
4169         SHADER_BUFFER *buffer, const struct vs_compile_args *args)
4170 {
4171     IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
4172     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4173     CONST DWORD *function = This->baseShader.function;
4174     const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
4175
4176     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4177     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4178
4179     shader_addline(buffer, "#version 120\n");
4180
4181     /* Base Declarations */
4182     shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION, NULL);
4183
4184     /* Base Shader Body */
4185     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function);
4186
4187     /* Unpack 3.0 outputs */
4188     if (reg_maps->shader_version.major >= 3) shader_addline(buffer, "order_ps_input(OUT);\n");
4189     else shader_addline(buffer, "order_ps_input();\n");
4190
4191     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4192      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4193      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4194      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4195      */
4196     if(args->fog_src == VS_FOG_Z) {
4197         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4198     } else if (!reg_maps->fog) {
4199         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4200     }
4201
4202     /* Write the final position.
4203      *
4204      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4205      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4206      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4207      * contains 1.0 to allow a mad.
4208      */
4209     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4210     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4211
4212     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4213      *
4214      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4215      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4216      * which is the same as z = z * 2 - w.
4217      */
4218     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4219
4220     shader_addline(buffer, "}\n");
4221
4222     TRACE("Compiling shader object %u\n", shader_obj);
4223     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4224     GL_EXTCALL(glCompileShaderARB(shader_obj));
4225     print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
4226
4227     return shader_obj;
4228 }
4229
4230 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, const WineD3D_GL_Info *gl_info, struct shader_caps *pCaps)
4231 {
4232     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4233      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
4234      * vs_nv_version which is based on NV_vertex_program.
4235      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4236      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4237      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4238      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4239      */
4240     if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
4241         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4242     else
4243         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4244     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4245     pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4246
4247     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4248      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4249      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4250      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4251      * in max native instructions. Intel and others also offer the info in this extension but they
4252      * don't support GLSL (at least on Windows).
4253      *
4254      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4255      * of instructions is 512 or less we have to do with ps2.0 hardware.
4256      * 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.
4257      */
4258     if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
4259         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4260     else
4261         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4262
4263     pCaps->MaxPixelShaderConst = GL_LIMITS(pshader_constantsF);
4264
4265     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4266      * Direct3D minimum requirement.
4267      *
4268      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4269      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4270      *
4271      * The problem is that the refrast clamps temporary results in the shader to
4272      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4273      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4274      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4275      * offer a way to query this.
4276      */
4277     pCaps->PixelShader1xMaxValue = 8.0;
4278     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4279 }
4280
4281 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4282 {
4283     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4284     {
4285         TRACE("Checking support for fixup:\n");
4286         dump_color_fixup_desc(fixup);
4287     }
4288
4289     /* We support everything except YUV conversions. */
4290     if (!is_yuv_fixup(fixup))
4291     {
4292         TRACE("[OK]\n");
4293         return TRUE;
4294     }
4295
4296     TRACE("[FAILED]\n");
4297     return FALSE;
4298 }
4299
4300 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4301 {
4302     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4303     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4304     /* WINED3DSIH_BEM           */ pshader_glsl_bem,
4305     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4306     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4307     /* WINED3DSIH_BREAKP        */ NULL,
4308     /* WINED3DSIH_CALL          */ shader_glsl_call,
4309     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4310     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4311     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4312     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4313     /* WINED3DSIH_DCL           */ NULL,
4314     /* WINED3DSIH_DEF           */ NULL,
4315     /* WINED3DSIH_DEFB          */ NULL,
4316     /* WINED3DSIH_DEFI          */ NULL,
4317     /* WINED3DSIH_DP2ADD        */ pshader_glsl_dp2add,
4318     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4319     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4320     /* WINED3DSIH_DST           */ shader_glsl_dst,
4321     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
4322     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
4323     /* WINED3DSIH_ELSE          */ shader_glsl_else,
4324     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
4325     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
4326     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
4327     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
4328     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
4329     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
4330     /* WINED3DSIH_IF            */ shader_glsl_if,
4331     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
4332     /* WINED3DSIH_LABEL         */ shader_glsl_label,
4333     /* WINED3DSIH_LIT           */ shader_glsl_lit,
4334     /* WINED3DSIH_LOG           */ shader_glsl_log,
4335     /* WINED3DSIH_LOGP          */ shader_glsl_log,
4336     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
4337     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
4338     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
4339     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
4340     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
4341     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
4342     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
4343     /* WINED3DSIH_MAD           */ shader_glsl_mad,
4344     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
4345     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
4346     /* WINED3DSIH_MOV           */ shader_glsl_mov,
4347     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
4348     /* WINED3DSIH_MUL           */ shader_glsl_arith,
4349     /* WINED3DSIH_NOP           */ NULL,
4350     /* WINED3DSIH_NRM           */ shader_glsl_map2gl,
4351     /* WINED3DSIH_PHASE         */ NULL,
4352     /* WINED3DSIH_POW           */ shader_glsl_pow,
4353     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
4354     /* WINED3DSIH_REP           */ shader_glsl_rep,
4355     /* WINED3DSIH_RET           */ NULL,
4356     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
4357     /* WINED3DSIH_SETP          */ NULL,
4358     /* WINED3DSIH_SGE           */ shader_glsl_compare,
4359     /* WINED3DSIH_SGN           */ shader_glsl_map2gl,
4360     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
4361     /* WINED3DSIH_SLT           */ shader_glsl_compare,
4362     /* WINED3DSIH_SUB           */ shader_glsl_arith,
4363     /* WINED3DSIH_TEX           */ pshader_glsl_tex,
4364     /* WINED3DSIH_TEXBEM        */ pshader_glsl_texbem,
4365     /* WINED3DSIH_TEXBEML       */ pshader_glsl_texbem,
4366     /* WINED3DSIH_TEXCOORD      */ pshader_glsl_texcoord,
4367     /* WINED3DSIH_TEXDEPTH      */ pshader_glsl_texdepth,
4368     /* WINED3DSIH_TEXDP3        */ pshader_glsl_texdp3,
4369     /* WINED3DSIH_TEXDP3TEX     */ pshader_glsl_texdp3tex,
4370     /* WINED3DSIH_TEXKILL       */ pshader_glsl_texkill,
4371     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
4372     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
4373     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_glsl_texm3x2depth,
4374     /* WINED3DSIH_TEXM3x2PAD    */ pshader_glsl_texm3x2pad,
4375     /* WINED3DSIH_TEXM3x2TEX    */ pshader_glsl_texm3x2tex,
4376     /* WINED3DSIH_TEXM3x3       */ pshader_glsl_texm3x3,
4377     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4378     /* WINED3DSIH_TEXM3x3PAD    */ pshader_glsl_texm3x3pad,
4379     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_glsl_texm3x3spec,
4380     /* WINED3DSIH_TEXM3x3TEX    */ pshader_glsl_texm3x3tex,
4381     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_glsl_texm3x3vspec,
4382     /* WINED3DSIH_TEXREG2AR     */ pshader_glsl_texreg2ar,
4383     /* WINED3DSIH_TEXREG2GB     */ pshader_glsl_texreg2gb,
4384     /* WINED3DSIH_TEXREG2RGB    */ pshader_glsl_texreg2rgb,
4385 };
4386
4387 const shader_backend_t glsl_shader_backend = {
4388     shader_glsl_instruction_handler_table,
4389     shader_glsl_select,
4390     shader_glsl_select_depth_blt,
4391     shader_glsl_deselect_depth_blt,
4392     shader_glsl_update_float_vertex_constants,
4393     shader_glsl_update_float_pixel_constants,
4394     shader_glsl_load_constants,
4395     shader_glsl_load_np2fixup_constants,
4396     shader_glsl_destroy,
4397     shader_glsl_alloc,
4398     shader_glsl_free,
4399     shader_glsl_dirty_const,
4400     shader_glsl_generate_pshader,
4401     shader_glsl_generate_vshader,
4402     shader_glsl_get_caps,
4403     shader_glsl_color_fixup_supported,
4404     shader_glsl_add_instruction_modifiers,
4405 };