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 WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  43 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  44 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  45 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  46
  47 typedef struct {
  48     char reg_name[150];
  49     char mask_str[6];
  50 } glsl_dst_param_t;
  51
  52 typedef struct {
  53     char reg_name[150];
  54     char param_str[200];
  55 } glsl_src_param_t;
  56
  57 typedef struct {
  58     const char *name;
  59     DWORD coord_mask;
  60 } glsl_sample_function_t;
  61
  62 enum heap_node_op
  63 {
  64     HEAP_NODE_TRAVERSE_LEFT,
  65     HEAP_NODE_TRAVERSE_RIGHT,
  66     HEAP_NODE_POP,
  67 };
  68
  69 struct constant_entry
  70 {
  71     unsigned int idx;
  72     unsigned int version;
  73 };
  74
  75 struct constant_heap
  76 {
  77     struct constant_entry *entries;
  78     unsigned int *positions;
  79     unsigned int size;
  80 };
  81
  82 /* GLSL shader private data */
  83 struct shader_glsl_priv {
  84     struct wined3d_shader_buffer shader_buffer;
  85     struct wine_rb_tree program_lookup;
  86     struct glsl_shader_prog_link *glsl_program;
  87     struct constant_heap vconst_heap;
  88     struct constant_heap pconst_heap;
  89     unsigned char *stack;
  90     GLhandleARB depth_blt_program_full[tex_type_count];
  91     GLhandleARB depth_blt_program_masked[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 wine_rb_entry        program_lookup_entry;
  98     struct list                 vshader_entry;
  99     struct list                 pshader_entry;
 100     GLhandleARB                 programId;
 101     GLint                       *vuniformF_locations;
 102     GLint                       *puniformF_locations;
 103     GLint                       vuniformI_locations[MAX_CONST_I];
 104     GLint                       puniformI_locations[MAX_CONST_I];
 105     GLint                       posFixup_location;
 106     GLint                       np2Fixup_location;
 107     GLint                       bumpenvmat_location[MAX_TEXTURES];
 108     GLint                       luminancescale_location[MAX_TEXTURES];
 109     GLint                       luminanceoffset_location[MAX_TEXTURES];
 110     GLint                       ycorrection_location;
 111     GLenum                      vertex_color_clamp;
 112     IWineD3DVertexShader        *vshader;
 113     IWineD3DPixelShader         *pshader;
 114     struct vs_compile_args      vs_args;
 115     struct ps_compile_args      ps_args;
 116     UINT                        constant_version;
 117     const struct ps_np2fixup_info *np2Fixup_info;
 118 };
 119
 120 typedef struct {
 121     IWineD3DVertexShader        *vshader;
 122     IWineD3DPixelShader         *pshader;
 123     struct ps_compile_args      ps_args;
 124     struct vs_compile_args      vs_args;
 125 } glsl_program_key_t;
 126
 127 struct shader_glsl_ctx_priv {
 128     const struct vs_compile_args    *cur_vs_args;
 129     const struct ps_compile_args    *cur_ps_args;
 130     struct ps_np2fixup_info         *cur_np2fixup_info;
 131 };
 132
 133 struct glsl_ps_compiled_shader
 134 {
 135     struct ps_compile_args          args;
 136     struct ps_np2fixup_info         np2fixup;
 137     GLhandleARB                     prgId;
 138 };
 139
 140 struct glsl_pshader_private
 141 {
 142     struct glsl_ps_compiled_shader  *gl_shaders;
 143     UINT                            num_gl_shaders, shader_array_size;
 144 };
 145
 146 struct glsl_vs_compiled_shader
 147 {
 148     struct vs_compile_args          args;
 149     GLhandleARB                     prgId;
 150 };
 151
 152 struct glsl_vshader_private
 153 {
 154     struct glsl_vs_compiled_shader  *gl_shaders;
 155     UINT                            num_gl_shaders, shader_array_size;
 156 };
 157
 158 static const char *debug_gl_shader_type(GLenum type)
 159 {
 160     switch (type)
 161     {
 162 #define WINED3D_TO_STR(u) case u: return #u
 163         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
 164         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
 165         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
 166 #undef WINED3D_TO_STR
 167         default:
 168             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
 169     }
 170 }
 171
 172 /* Extract a line from the info log.
 173  * Note that this modifies the source string. */
 174 static char *get_info_log_line(char **ptr)
 175 {
 176     char *p, *q;
 177
 178     p = *ptr;
 179     if (!(q = strstr(p, "\n")))
 180     {
 181         if (!*p) return NULL;
 182         *ptr += strlen(p);
 183         return p;
 184     }
 185     *q = '\0';
 186     *ptr = q + 1;
 187
 188     return p;
 189 }
 190
 191 /** Prints the GLSL info log which will contain error messages if they exist */
 192 /* GL locking is done by the caller */
 193 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 194 {
 195     int infologLength = 0;
 196     char *infoLog;
 197     unsigned int i;
 198     BOOL is_spam;
 199
 200     static const char * const spam[] =
 201     {
 202         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 203         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx, with \n */
 204         "Fragment shader was successfully compiled to run on hardware.",    /* fglrx, no \n   */
 205         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 206         "Fragment shader(s) linked, vertex shader(s) linked. \n",           /* fglrx, with \n */
 207         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 208         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 209         "Vertex shader(s) linked, no fragment shader(s) defined. \n",       /* fglrx, with \n */
 210         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 211         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 212         "Fragment shader(s) linked, no vertex shader(s) defined. \n",       /* fglrx, with \n */
 213         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 214     };
 215
 216     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 217
 218     GL_EXTCALL(glGetObjectParameterivARB(obj,
 219                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 220                &infologLength));
 221
 222     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 223      * that if there are errors. */
 224     if (infologLength > 1)
 225     {
 226         char *ptr, *line;
 227
 228         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 229          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 230          */
 231         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 232         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 233         is_spam = FALSE;
 234
 235         for (i = 0; i < sizeof(spam) / sizeof(*spam); ++i)
 236         {
 237             if (!strcmp(infoLog, spam[i]))
 238             {
 239                 is_spam = TRUE;
 240                 break;
 241             }
 242         }
 243
 244         ptr = infoLog;
 245         if (is_spam)
 246         {
 247             TRACE("Spam received from GLSL shader #%u:\n", obj);
 248             while ((line = get_info_log_line(&ptr))) TRACE("    %s\n", line);
 249         }
 250         else
 251         {
 252             FIXME("Error received from GLSL shader #%u:\n", obj);
 253             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 254         }
 255         HeapFree(GetProcessHeap(), 0, infoLog);
 256     }
 257 }
 258
 259 /* GL locking is done by the caller. */
 260 static void shader_glsl_compile(const struct wined3d_gl_info *gl_info, GLhandleARB shader, const char *src)
 261 {
 262     TRACE("Compiling shader object %u.\n", shader);
 263     GL_EXTCALL(glShaderSourceARB(shader, 1, &src, NULL));
 264     checkGLcall("glShaderSourceARB");
 265     GL_EXTCALL(glCompileShaderARB(shader));
 266     checkGLcall("glCompileShaderARB");
 267     print_glsl_info_log(gl_info, shader);
 268 }
 269
 270 /* GL locking is done by the caller. */
 271 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 272 {
 273     GLint i, object_count, source_size = -1;
 274     GLhandleARB *objects;
 275     char *source = NULL;
 276
 277     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
 278     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
 279     if (!objects)
 280     {
 281         ERR("Failed to allocate object array memory.\n");
 282         return;
 283     }
 284
 285     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
 286     for (i = 0; i < object_count; ++i)
 287     {
 288         char *ptr, *line;
 289         GLint tmp;
 290
 291         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
 292
 293         if (source_size < tmp)
 294         {
 295             HeapFree(GetProcessHeap(), 0, source);
 296
 297             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
 298             if (!source)
 299             {
 300                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
 301                 HeapFree(GetProcessHeap(), 0, objects);
 302                 return;
 303             }
 304             source_size = tmp;
 305         }
 306
 307         FIXME("Object %u:\n", objects[i]);
 308         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
 309         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
 310         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
 311         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
 312         FIXME("\n");
 313
 314         ptr = source;
 315         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
 316         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 317         FIXME("\n");
 318     }
 319
 320     HeapFree(GetProcessHeap(), 0, source);
 321     HeapFree(GetProcessHeap(), 0, objects);
 322 }
 323
 324 /* GL locking is done by the caller. */
 325 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 326 {
 327     GLint tmp;
 328
 329     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 330
 331     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
 332     if (tmp == GL_PROGRAM_OBJECT_ARB)
 333     {
 334         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
 335         if (!tmp)
 336         {
 337             FIXME("Program %u link status invalid.\n", program);
 338             shader_glsl_dump_program_source(gl_info, program);
 339         }
 340     }
 341
 342     print_glsl_info_log(gl_info, program);
 343 }
 344
 345 /**
 346  * Loads (pixel shader) samplers
 347  */
 348 /* GL locking is done by the caller */
 349 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 350         DWORD *tex_unit_map, GLhandleARB programId)
 351 {
 352     GLint name_loc;
 353     int i;
 354     char sampler_name[20];
 355
 356     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 357         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 358         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 359         if (name_loc != -1) {
 360             DWORD mapped_unit = tex_unit_map[i];
 361             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 362             {
 363                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 364                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 365                 checkGLcall("glUniform1iARB");
 366             } else {
 367                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 368             }
 369         }
 370     }
 371 }
 372
 373 /* GL locking is done by the caller */
 374 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 375         DWORD *tex_unit_map, GLhandleARB programId)
 376 {
 377     GLint name_loc;
 378     char sampler_name[20];
 379     int i;
 380
 381     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 382         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 383         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 384         if (name_loc != -1) {
 385             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 386             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 387             {
 388                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 389                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 390                 checkGLcall("glUniform1iARB");
 391             } else {
 392                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 393             }
 394         }
 395     }
 396 }
 397
 398 /* GL locking is done by the caller */
 399 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 400         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 401 {
 402     int stack_idx = 0;
 403     unsigned int heap_idx = 1;
 404     unsigned int idx;
 405
 406     if (heap->entries[heap_idx].version <= version) return;
 407
 408     idx = heap->entries[heap_idx].idx;
 409     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 410     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 411
 412     while (stack_idx >= 0)
 413     {
 414         /* Note that we fall through to the next case statement. */
 415         switch(stack[stack_idx])
 416         {
 417             case HEAP_NODE_TRAVERSE_LEFT:
 418             {
 419                 unsigned int left_idx = heap_idx << 1;
 420                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 421                 {
 422                     heap_idx = left_idx;
 423                     idx = heap->entries[heap_idx].idx;
 424                     if (constant_locations[idx] != -1)
 425                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 426
 427                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 428                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 429                     break;
 430                 }
 431             }
 432
 433             case HEAP_NODE_TRAVERSE_RIGHT:
 434             {
 435                 unsigned int right_idx = (heap_idx << 1) + 1;
 436                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 437                 {
 438                     heap_idx = right_idx;
 439                     idx = heap->entries[heap_idx].idx;
 440                     if (constant_locations[idx] != -1)
 441                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 442
 443                     stack[stack_idx++] = HEAP_NODE_POP;
 444                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 445                     break;
 446                 }
 447             }
 448
 449             case HEAP_NODE_POP:
 450             {
 451                 heap_idx >>= 1;
 452                 --stack_idx;
 453                 break;
 454             }
 455         }
 456     }
 457     checkGLcall("walk_constant_heap()");
 458 }
 459
 460 /* GL locking is done by the caller */
 461 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 462 {
 463     GLfloat clamped_constant[4];
 464
 465     if (location == -1) return;
 466
 467     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 468     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 469     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 470     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 471
 472     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 473 }
 474
 475 /* GL locking is done by the caller */
 476 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 477         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 478 {
 479     int stack_idx = 0;
 480     unsigned int heap_idx = 1;
 481     unsigned int idx;
 482
 483     if (heap->entries[heap_idx].version <= version) return;
 484
 485     idx = heap->entries[heap_idx].idx;
 486     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 487     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 488
 489     while (stack_idx >= 0)
 490     {
 491         /* Note that we fall through to the next case statement. */
 492         switch(stack[stack_idx])
 493         {
 494             case HEAP_NODE_TRAVERSE_LEFT:
 495             {
 496                 unsigned int left_idx = heap_idx << 1;
 497                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 498                 {
 499                     heap_idx = left_idx;
 500                     idx = heap->entries[heap_idx].idx;
 501                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 502
 503                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 504                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 505                     break;
 506                 }
 507             }
 508
 509             case HEAP_NODE_TRAVERSE_RIGHT:
 510             {
 511                 unsigned int right_idx = (heap_idx << 1) + 1;
 512                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 513                 {
 514                     heap_idx = right_idx;
 515                     idx = heap->entries[heap_idx].idx;
 516                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 517
 518                     stack[stack_idx++] = HEAP_NODE_POP;
 519                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 520                     break;
 521                 }
 522             }
 523
 524             case HEAP_NODE_POP:
 525             {
 526                 heap_idx >>= 1;
 527                 --stack_idx;
 528                 break;
 529             }
 530         }
 531     }
 532     checkGLcall("walk_constant_heap_clamped()");
 533 }
 534
 535 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 536 /* GL locking is done by the caller */
 537 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 538         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 539         unsigned char *stack, UINT version)
 540 {
 541     const local_constant *lconst;
 542
 543     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 544     if (This->baseShader.reg_maps.shader_version.major == 1
 545             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
 546         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 547     else
 548         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 549
 550     if (!This->baseShader.load_local_constsF)
 551     {
 552         TRACE("No need to load local float constants for this shader\n");
 553         return;
 554     }
 555
 556     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 557     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 558     {
 559         GLint location = constant_locations[lconst->idx];
 560         /* We found this uniform name in the program - go ahead and send the data */
 561         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 562     }
 563     checkGLcall("glUniform4fvARB()");
 564 }
 565
 566 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 567 /* GL locking is done by the caller */
 568 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 569         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 570 {
 571     unsigned int i;
 572     struct list* ptr;
 573
 574     for (i = 0; constants_set; constants_set >>= 1, ++i)
 575     {
 576         if (!(constants_set & 1)) continue;
 577
 578         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 579                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 580
 581         /* We found this uniform name in the program - go ahead and send the data */
 582         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 583         checkGLcall("glUniform4ivARB");
 584     }
 585
 586     /* Load immediate constants */
 587     ptr = list_head(&This->baseShader.constantsI);
 588     while (ptr) {
 589         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 590         unsigned int idx = lconst->idx;
 591         const GLint *values = (const GLint *)lconst->value;
 592
 593         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 594             values[0], values[1], values[2], values[3]);
 595
 596         /* We found this uniform name in the program - go ahead and send the data */
 597         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 598         checkGLcall("glUniform4ivARB");
 599         ptr = list_next(&This->baseShader.constantsI, ptr);
 600     }
 601 }
 602
 603 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 604 /* GL locking is done by the caller */
 605 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 606         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 607 {
 608     GLint tmp_loc;
 609     unsigned int i;
 610     char tmp_name[8];
 611     const char *prefix;
 612     struct list* ptr;
 613
 614     switch (This->baseShader.reg_maps.shader_version.type)
 615     {
 616         case WINED3D_SHADER_TYPE_VERTEX:
 617             prefix = "VB";
 618             break;
 619
 620         case WINED3D_SHADER_TYPE_GEOMETRY:
 621             prefix = "GB";
 622             break;
 623
 624         case WINED3D_SHADER_TYPE_PIXEL:
 625             prefix = "PB";
 626             break;
 627
 628         default:
 629             FIXME("Unknown shader type %#x.\n",
 630                     This->baseShader.reg_maps.shader_version.type);
 631             prefix = "UB";
 632             break;
 633     }
 634
 635     /* TODO: Benchmark and see if it would be beneficial to store the
 636      * locations of the constants to avoid looking up each time */
 637     for (i = 0; constants_set; constants_set >>= 1, ++i)
 638     {
 639         if (!(constants_set & 1)) continue;
 640
 641         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 642
 643         /* TODO: Benchmark and see if it would be beneficial to store the
 644          * locations of the constants to avoid looking up each time */
 645         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 646         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 647         if (tmp_loc != -1)
 648         {
 649             /* We found this uniform name in the program - go ahead and send the data */
 650             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 651             checkGLcall("glUniform1ivARB");
 652         }
 653     }
 654
 655     /* Load immediate constants */
 656     ptr = list_head(&This->baseShader.constantsB);
 657     while (ptr) {
 658         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 659         unsigned int idx = lconst->idx;
 660         const GLint *values = (const GLint *)lconst->value;
 661
 662         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 663
 664         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 665         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 666         if (tmp_loc != -1) {
 667             /* We found this uniform name in the program - go ahead and send the data */
 668             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 669             checkGLcall("glUniform1ivARB");
 670         }
 671         ptr = list_next(&This->baseShader.constantsB, ptr);
 672     }
 673 }
 674
 675 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 676 {
 677     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 678 }
 679
 680 /**
 681  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 682  */
 683 /* GL locking is done by the caller (state handler) */
 684 static void shader_glsl_load_np2fixup_constants(
 685     IWineD3DDevice* device,
 686     char usePixelShader,
 687     char useVertexShader) {
 688
 689     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
 690     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
 691
 692     if (!prog) {
 693         /* No GLSL program set - nothing to do. */
 694         return;
 695     }
 696
 697     if (!usePixelShader) {
 698         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 699         return;
 700     }
 701
 702     if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
 703         const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 704         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
 705         UINT i;
 706         UINT fixup = prog->ps_args.np2_fixup;
 707         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 708
 709         for (i = 0; fixup; fixup >>= 1, ++i) {
 710             const unsigned char idx = prog->np2Fixup_info->idx[i];
 711             const IWineD3DBaseTextureImpl *tex = stateBlock->state.textures[i];
 712             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 713
 714             if (!tex)
 715             {
 716                 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
 717                 continue;
 718             }
 719
 720             if (idx % 2) {
 721                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 722             } else {
 723                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 724             }
 725         }
 726
 727         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 728     }
 729 }
 730
 731 /**
 732  * Loads the app-supplied constants into the currently set GLSL program.
 733  */
 734 /* GL locking is done by the caller (state handler) */
 735 static void shader_glsl_load_constants(const struct wined3d_context *context,
 736         char usePixelShader, char useVertexShader)
 737 {
 738     const struct wined3d_gl_info *gl_info = context->gl_info;
 739     IWineD3DDeviceImpl *device = context->swapchain->device;
 740     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 741     struct shader_glsl_priv *priv = device->shader_priv;
 742
 743     GLhandleARB programId;
 744     struct glsl_shader_prog_link *prog = priv->glsl_program;
 745     UINT constant_version;
 746     int i;
 747
 748     if (!prog) {
 749         /* No GLSL program set - nothing to do. */
 750         return;
 751     }
 752     programId = prog->programId;
 753     constant_version = prog->constant_version;
 754
 755     if (useVertexShader)
 756     {
 757         IWineD3DBaseShaderImpl *vshader = (IWineD3DBaseShaderImpl *)stateBlock->state.vertex_shader;
 758
 759         /* Load DirectX 9 float constants/uniforms for vertex shader */
 760         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->state.vs_consts_f,
 761                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 762
 763         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 764         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->state.vs_consts_i,
 765                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 766
 767         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 768         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->state.vs_consts_b,
 769                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 770
 771         /* Upload the position fixup params */
 772         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
 773         checkGLcall("glUniform4fvARB");
 774     }
 775
 776     if (usePixelShader)
 777     {
 778         IWineD3DBaseShaderImpl *pshader = (IWineD3DBaseShaderImpl *)stateBlock->state.pixel_shader;
 779
 780         /* Load DirectX 9 float constants/uniforms for pixel shader */
 781         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->state.ps_consts_f,
 782                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 783
 784         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 785         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->state.ps_consts_i,
 786                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 787
 788         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 789         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->state.ps_consts_b,
 790                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 791
 792         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 793          * It can't be 0 for a valid texbem instruction.
 794          */
 795         for(i = 0; i < MAX_TEXTURES; i++) {
 796             const float *data;
 797
 798             if(prog->bumpenvmat_location[i] == -1) continue;
 799
 800             data = (const float *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVMAT00];
 801             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 802             checkGLcall("glUniformMatrix2fvARB");
 803
 804             /* texbeml needs the luminance scale and offset too. If texbeml
 805              * is used, needsbumpmat is set too, so we can check that in the
 806              * needsbumpmat check. */
 807             if (prog->luminancescale_location[i] != -1)
 808             {
 809                 const GLfloat *scale = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLSCALE];
 810                 const GLfloat *offset = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
 811
 812                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 813                 checkGLcall("glUniform1fvARB");
 814                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 815                 checkGLcall("glUniform1fvARB");
 816             }
 817         }
 818
 819         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 820             float correction_params[4];
 821
 822             if (context->render_offscreen)
 823             {
 824                 correction_params[0] = 0.0f;
 825                 correction_params[1] = 1.0f;
 826             } else {
 827                 /* position is window relative, not viewport relative */
 828                 correction_params[0] = context->current_rt->currentDesc.Height;
 829                 correction_params[1] = -1.0f;
 830             }
 831             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 832         }
 833     }
 834
 835     if (priv->next_constant_version == UINT_MAX)
 836     {
 837         TRACE("Max constant version reached, resetting to 0.\n");
 838         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 839         priv->next_constant_version = 1;
 840     }
 841     else
 842     {
 843         prog->constant_version = priv->next_constant_version++;
 844     }
 845 }
 846
 847 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 848         unsigned int heap_idx, DWORD new_version)
 849 {
 850     struct constant_entry *entries = heap->entries;
 851     unsigned int *positions = heap->positions;
 852     unsigned int parent_idx;
 853
 854     while (heap_idx > 1)
 855     {
 856         parent_idx = heap_idx >> 1;
 857
 858         if (new_version <= entries[parent_idx].version) break;
 859
 860         entries[heap_idx] = entries[parent_idx];
 861         positions[entries[parent_idx].idx] = heap_idx;
 862         heap_idx = parent_idx;
 863     }
 864
 865     entries[heap_idx].version = new_version;
 866     entries[heap_idx].idx = idx;
 867     positions[idx] = heap_idx;
 868 }
 869
 870 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 871 {
 872     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 873     struct shader_glsl_priv *priv = This->shader_priv;
 874     struct constant_heap *heap = &priv->vconst_heap;
 875     UINT i;
 876
 877     for (i = start; i < count + start; ++i)
 878     {
 879         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
 880             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 881         else
 882             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 883     }
 884 }
 885
 886 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 887 {
 888     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 889     struct shader_glsl_priv *priv = This->shader_priv;
 890     struct constant_heap *heap = &priv->pconst_heap;
 891     UINT i;
 892
 893     for (i = start; i < count + start; ++i)
 894     {
 895         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 896             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 897         else
 898             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 899     }
 900 }
 901
 902 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 903 {
 904     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 905     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 906     if(shader_major > 3) return ret;
 907
 908     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 909     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 910     return ret;
 911 }
 912
 913 /** Generate the variable & register declarations for the GLSL output target */
 914 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 915         struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
 916         const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
 917 {
 918     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 919     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 920     const struct wined3d_state *state = &device->stateBlock->state;
 921     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 922     const struct wined3d_gl_info *gl_info = context->gl_info;
 923     unsigned int i, extra_constants_needed = 0;
 924     const local_constant *lconst;
 925     DWORD map;
 926
 927     /* There are some minor differences between pixel and vertex shaders */
 928     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 929     char prefix = pshader ? 'P' : 'V';
 930
 931     /* Prototype the subroutines */
 932     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 933     {
 934         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 935     }
 936
 937     /* Declare the constants (aka uniforms) */
 938     if (This->baseShader.limits.constant_float > 0) {
 939         unsigned max_constantsF;
 940         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 941          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 942          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 943          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 944          * a dx9 card, as long as it doesn't also use all the other constants.
 945          *
 946          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 947          * declare only the amount that we're assured to have.
 948          *
 949          * Thus we run into problems in these two cases:
 950          * 1) The shader really uses more uniforms than supported
 951          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 952          */
 953         if (pshader)
 954         {
 955             /* No indirect addressing here. */
 956             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 957         }
 958         else
 959         {
 960             if(This->baseShader.reg_maps.usesrelconstF) {
 961                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
 962                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
 963                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
 964                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
 965                  *
 966                  * Writing gl_ClipVertex requires one uniform for each clipplane as well.
 967                  */
 968                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 969                 if(ctx_priv->cur_vs_args->clip_enabled)
 970                 {
 971                     max_constantsF -= gl_info->limits.clipplanes;
 972                 }
 973                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 974                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 975                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 976                  * for now take this into account when calculating the number of available constants
 977                  */
 978                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
 979                 /* Set by driver quirks in directx.c */
 980                 max_constantsF -= gl_info->reserved_glsl_constants;
 981             }
 982             else
 983             {
 984                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 985             }
 986         }
 987         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
 988         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 989     }
 990
 991     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
 992      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
 993      */
 994     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
 995         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 996
 997     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
 998         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 999
1000     if (!pshader)
1001     {
1002         shader_addline(buffer, "uniform vec4 posFixup;\n");
1003         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
1004     }
1005     else
1006     {
1007         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1008         {
1009             if (!(map & 1)) continue;
1010
1011             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1012
1013             if (reg_maps->luminanceparams & (1 << i))
1014             {
1015                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1016                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1017                 extra_constants_needed++;
1018             }
1019
1020             extra_constants_needed++;
1021         }
1022
1023         if (ps_args->srgb_correction)
1024         {
1025             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1026                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1027             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1028                     srgb_cmp);
1029         }
1030         if (reg_maps->vpos || reg_maps->usesdsy)
1031         {
1032             if (This->baseShader.limits.constant_float + extra_constants_needed
1033                     + 1 < gl_info->limits.glsl_ps_float_constants)
1034             {
1035                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1036                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1037                 extra_constants_needed++;
1038             } else {
1039                 /* This happens because we do not have proper tracking of the constant registers that are
1040                  * actually used, only the max limit of the shader version
1041                  */
1042                 FIXME("Cannot find a free uniform for vpos correction params\n");
1043                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1044                         context->render_offscreen ? 0.0f : device->render_targets[0]->currentDesc.Height,
1045                         context->render_offscreen ? 1.0f : -1.0f);
1046             }
1047             shader_addline(buffer, "vec4 vpos;\n");
1048         }
1049     }
1050
1051     /* Declare texture samplers */
1052     for (i = 0; i < This->baseShader.limits.sampler; i++) {
1053         if (reg_maps->sampler_type[i])
1054         {
1055             IWineD3DBaseTextureImpl *texture;
1056
1057             switch (reg_maps->sampler_type[i])
1058             {
1059                 case WINED3DSTT_1D:
1060                     if (pshader && ps_args->shadow & (1 << i))
1061                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1062                     else
1063                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1064                     break;
1065                 case WINED3DSTT_2D:
1066                     texture = state->textures[i];
1067                     if (pshader && ps_args->shadow & (1 << i))
1068                     {
1069                         if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1070                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1071                         else
1072                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1073                     }
1074                     else
1075                     {
1076                         if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1077                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1078                         else
1079                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1080                     }
1081                     break;
1082                 case WINED3DSTT_CUBE:
1083                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1084                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1085                     break;
1086                 case WINED3DSTT_VOLUME:
1087                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1088                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1089                     break;
1090                 default:
1091                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1092                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1093                     break;
1094             }
1095         }
1096     }
1097
1098     /* Declare uniforms for NP2 texcoord fixup:
1099      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1100      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1101      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1102     if (pshader && ps_args->np2_fixup) {
1103
1104         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1105         UINT cur = 0;
1106
1107         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1108          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1109          * samplerNP2Fixup stores texture dimensions and is updated through
1110          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1111
1112         for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1113             if (reg_maps->sampler_type[i]) {
1114                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1115
1116                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1117                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1118                     continue;
1119                 }
1120
1121                 fixup->idx[i] = cur++;
1122             }
1123         }
1124
1125         fixup->num_consts = (cur + 1) >> 1;
1126         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1127     }
1128
1129     /* Declare address variables */
1130     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1131     {
1132         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1133     }
1134
1135     /* Declare texture coordinate temporaries and initialize them */
1136     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1137     {
1138         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1139     }
1140
1141     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1142      * helper function shader that is linked in at link time
1143      */
1144     if (pshader && reg_maps->shader_version.major >= 3)
1145     {
1146         if (use_vs(state))
1147         {
1148             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1149         } else {
1150             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1151              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1152              * pixel shader that reads the fixed function color into the packed input registers.
1153              */
1154             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1155         }
1156     }
1157
1158     /* Declare output register temporaries */
1159     if(This->baseShader.limits.packed_output) {
1160         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1161     }
1162
1163     /* Declare temporary variables */
1164     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1165     {
1166         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1167     }
1168
1169     /* Declare attributes */
1170     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1171     {
1172         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1173         {
1174             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1175         }
1176     }
1177
1178     /* Declare loop registers aLx */
1179     for (i = 0; i < reg_maps->loop_depth; i++) {
1180         shader_addline(buffer, "int aL%u;\n", i);
1181         shader_addline(buffer, "int tmpInt%u;\n", i);
1182     }
1183
1184     /* Temporary variables for matrix operations */
1185     shader_addline(buffer, "vec4 tmp0;\n");
1186     shader_addline(buffer, "vec4 tmp1;\n");
1187
1188     /* Local constants use a different name so they can be loaded once at shader link time
1189      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1190      * float -> string conversion can cause precision loss.
1191      */
1192     if(!This->baseShader.load_local_constsF) {
1193         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1194             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1195         }
1196     }
1197
1198     /* Start the main program */
1199     shader_addline(buffer, "void main() {\n");
1200     if(pshader && reg_maps->vpos) {
1201         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1202          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1203          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1204          * precision troubles when we just substract 0.5.
1205          *
1206          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1207          *
1208          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1209          *
1210          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1211          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1212          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1213          * correctly on drivers that returns integer values.
1214          */
1215         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1216     }
1217 }
1218
1219 /*****************************************************************************
1220  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1221  *
1222  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1223  ****************************************************************************/
1224
1225 /* Prototypes */
1226 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1227         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1228
1229 /** Used for opcode modifiers - They multiply the result by the specified amount */
1230 static const char * const shift_glsl_tab[] = {
1231     "",           /*  0 (none) */
1232     "2.0 * ",     /*  1 (x2)   */
1233     "4.0 * ",     /*  2 (x4)   */
1234     "8.0 * ",     /*  3 (x8)   */
1235     "16.0 * ",    /*  4 (x16)  */
1236     "32.0 * ",    /*  5 (x32)  */
1237     "",           /*  6 (x64)  */
1238     "",           /*  7 (x128) */
1239     "",           /*  8 (d256) */
1240     "",           /*  9 (d128) */
1241     "",           /* 10 (d64)  */
1242     "",           /* 11 (d32)  */
1243     "0.0625 * ",  /* 12 (d16)  */
1244     "0.125 * ",   /* 13 (d8)   */
1245     "0.25 * ",    /* 14 (d4)   */
1246     "0.5 * "      /* 15 (d2)   */
1247 };
1248
1249 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1250 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1251 {
1252     out_str[0] = 0;
1253
1254     switch (src_modifier)
1255     {
1256     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1257     case WINED3DSPSM_DW:
1258     case WINED3DSPSM_NONE:
1259         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1260         break;
1261     case WINED3DSPSM_NEG:
1262         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1263         break;
1264     case WINED3DSPSM_NOT:
1265         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1266         break;
1267     case WINED3DSPSM_BIAS:
1268         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1269         break;
1270     case WINED3DSPSM_BIASNEG:
1271         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1272         break;
1273     case WINED3DSPSM_SIGN:
1274         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1275         break;
1276     case WINED3DSPSM_SIGNNEG:
1277         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1278         break;
1279     case WINED3DSPSM_COMP:
1280         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1281         break;
1282     case WINED3DSPSM_X2:
1283         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1284         break;
1285     case WINED3DSPSM_X2NEG:
1286         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1287         break;
1288     case WINED3DSPSM_ABS:
1289         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1290         break;
1291     case WINED3DSPSM_ABSNEG:
1292         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1293         break;
1294     default:
1295         FIXME("Unhandled modifier %u\n", src_modifier);
1296         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1297     }
1298 }
1299
1300 /** Writes the GLSL variable name that corresponds to the register that the
1301  * DX opcode parameter is trying to access */
1302 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1303         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1304 {
1305     /* oPos, oFog and oPts in D3D */
1306     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1307
1308     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1309     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1310     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1311
1312     *is_color = FALSE;
1313
1314     switch (reg->type)
1315     {
1316         case WINED3DSPR_TEMP:
1317             sprintf(register_name, "R%u", reg->idx);
1318             break;
1319
1320         case WINED3DSPR_INPUT:
1321             /* vertex shaders */
1322             if (!pshader)
1323             {
1324                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1325                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1326                 sprintf(register_name, "attrib%u", reg->idx);
1327                 break;
1328             }
1329
1330             /* pixel shaders >= 3.0 */
1331             if (This->baseShader.reg_maps.shader_version.major >= 3)
1332             {
1333                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1334                 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1335
1336                 if (reg->rel_addr)
1337                 {
1338                     glsl_src_param_t rel_param;
1339
1340                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1341
1342                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1343                      * operation there */
1344                     if (idx)
1345                     {
1346                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1347                         {
1348                             sprintf(register_name,
1349                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1350                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1351                                     rel_param.param_str, idx);
1352                         }
1353                         else
1354                         {
1355                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1356                         }
1357                     }
1358                     else
1359                     {
1360                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1361                         {
1362                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1363                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1364                                     rel_param.param_str);
1365                         }
1366                         else
1367                         {
1368                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1369                         }
1370                     }
1371                 }
1372                 else
1373                 {
1374                     if (idx == in_count) sprintf(register_name, "gl_Color");
1375                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1376                     else sprintf(register_name, "IN[%u]", idx);
1377                 }
1378             }
1379             else
1380             {
1381                 if (!reg->idx) strcpy(register_name, "gl_Color");
1382                 else strcpy(register_name, "gl_SecondaryColor");
1383                 break;
1384             }
1385             break;
1386
1387         case WINED3DSPR_CONST:
1388             {
1389                 const char prefix = pshader ? 'P' : 'V';
1390
1391                 /* Relative addressing */
1392                 if (reg->rel_addr)
1393                 {
1394                     glsl_src_param_t rel_param;
1395                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1396                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1397                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1398                 }
1399                 else
1400                 {
1401                     if (shader_constant_is_local(This, reg->idx))
1402                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1403                     else
1404                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1405                 }
1406             }
1407             break;
1408
1409         case WINED3DSPR_CONSTINT:
1410             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1411             else sprintf(register_name, "VI[%u]", reg->idx);
1412             break;
1413
1414         case WINED3DSPR_CONSTBOOL:
1415             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1416             else sprintf(register_name, "VB[%u]", reg->idx);
1417             break;
1418
1419         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1420             if (pshader) sprintf(register_name, "T%u", reg->idx);
1421             else sprintf(register_name, "A%u", reg->idx);
1422             break;
1423
1424         case WINED3DSPR_LOOP:
1425             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1426             break;
1427
1428         case WINED3DSPR_SAMPLER:
1429             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1430             else sprintf(register_name, "Vsampler%u", reg->idx);
1431             break;
1432
1433         case WINED3DSPR_COLOROUT:
1434             if (reg->idx >= gl_info->limits.buffers)
1435                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1436
1437             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1438             break;
1439
1440         case WINED3DSPR_RASTOUT:
1441             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1442             break;
1443
1444         case WINED3DSPR_DEPTHOUT:
1445             sprintf(register_name, "gl_FragDepth");
1446             break;
1447
1448         case WINED3DSPR_ATTROUT:
1449             if (!reg->idx) sprintf(register_name, "OUT[8]");
1450             else sprintf(register_name, "OUT[9]");
1451             break;
1452
1453         case WINED3DSPR_TEXCRDOUT:
1454             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1455             sprintf(register_name, "OUT[%u]", reg->idx);
1456             break;
1457
1458         case WINED3DSPR_MISCTYPE:
1459             if (!reg->idx)
1460             {
1461                 /* vPos */
1462                 sprintf(register_name, "vpos");
1463             }
1464             else if (reg->idx == 1)
1465             {
1466                 /* Note that gl_FrontFacing is a bool, while vFace is
1467                  * a float for which the sign determines front/back */
1468                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1469             }
1470             else
1471             {
1472                 FIXME("Unhandled misctype register %d\n", reg->idx);
1473                 sprintf(register_name, "unrecognized_register");
1474             }
1475             break;
1476
1477         case WINED3DSPR_IMMCONST:
1478             switch (reg->immconst_type)
1479             {
1480                 case WINED3D_IMMCONST_FLOAT:
1481                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1482                     break;
1483
1484                 case WINED3D_IMMCONST_FLOAT4:
1485                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1486                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1487                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1488                     break;
1489
1490                 default:
1491                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1492                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1493             }
1494             break;
1495
1496         default:
1497             FIXME("Unhandled register name Type(%d)\n", reg->type);
1498             sprintf(register_name, "unrecognized_register");
1499             break;
1500     }
1501 }
1502
1503 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1504 {
1505     *str++ = '.';
1506     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1507     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1508     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1509     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1510     *str = '\0';
1511 }
1512
1513 /* Get the GLSL write mask for the destination register */
1514 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1515 {
1516     DWORD mask = param->write_mask;
1517
1518     if (shader_is_scalar(&param->reg))
1519     {
1520         mask = WINED3DSP_WRITEMASK_0;
1521         *write_mask = '\0';
1522     }
1523     else
1524     {
1525         shader_glsl_write_mask_to_str(mask, write_mask);
1526     }
1527
1528     return mask;
1529 }
1530
1531 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1532     unsigned int size = 0;
1533
1534     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1535     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1536     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1537     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1538
1539     return size;
1540 }
1541
1542 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1543 {
1544     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1545      * but addressed as "rgba". To fix this we need to swap the register's x
1546      * and z components. */
1547     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1548
1549     *str++ = '.';
1550     /* swizzle bits fields: wwzzyyxx */
1551     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1552     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1553     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1554     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1555     *str = '\0';
1556 }
1557
1558 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1559         BOOL fixup, DWORD mask, char *swizzle_str)
1560 {
1561     if (shader_is_scalar(&param->reg))
1562         *swizzle_str = '\0';
1563     else
1564         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1565 }
1566
1567 /* From a given parameter token, generate the corresponding GLSL string.
1568  * Also, return the actual register name and swizzle in case the
1569  * caller needs this information as well. */
1570 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1571         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1572 {
1573     BOOL is_color = FALSE;
1574     char swizzle_str[6];
1575
1576     glsl_src->reg_name[0] = '\0';
1577     glsl_src->param_str[0] = '\0';
1578     swizzle_str[0] = '\0';
1579
1580     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1581     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1582     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1583 }
1584
1585 /* From a given parameter token, generate the corresponding GLSL string.
1586  * Also, return the actual register name and swizzle in case the
1587  * caller needs this information as well. */
1588 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1589         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1590 {
1591     BOOL is_color = FALSE;
1592
1593     glsl_dst->mask_str[0] = '\0';
1594     glsl_dst->reg_name[0] = '\0';
1595
1596     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1597     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1598 }
1599
1600 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1601 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1602         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1603 {
1604     glsl_dst_param_t glsl_dst;
1605     DWORD mask;
1606
1607     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1608     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1609
1610     return mask;
1611 }
1612
1613 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1614 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1615 {
1616     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1617 }
1618
1619 /** Process GLSL instruction modifiers */
1620 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1621 {
1622     glsl_dst_param_t dst_param;
1623     DWORD modifiers;
1624
1625     if (!ins->dst_count) return;
1626
1627     modifiers = ins->dst[0].modifiers;
1628     if (!modifiers) return;
1629
1630     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1631
1632     if (modifiers & WINED3DSPDM_SATURATE)
1633     {
1634         /* _SAT means to clamp the value of the register to between 0 and 1 */
1635         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1636                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1637     }
1638
1639     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1640     {
1641         FIXME("_centroid modifier not handled\n");
1642     }
1643
1644     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1645     {
1646         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1647     }
1648 }
1649
1650 static inline const char *shader_get_comp_op(DWORD op)
1651 {
1652     switch (op) {
1653         case COMPARISON_GT: return ">";
1654         case COMPARISON_EQ: return "==";
1655         case COMPARISON_GE: return ">=";
1656         case COMPARISON_LT: return "<";
1657         case COMPARISON_NE: return "!=";
1658         case COMPARISON_LE: return "<=";
1659         default:
1660             FIXME("Unrecognized comparison value: %u\n", op);
1661             return "(\?\?)";
1662     }
1663 }
1664
1665 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1666         DWORD sampler_idx, DWORD flags, glsl_sample_function_t *sample_function)
1667 {
1668     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1669     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1670     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1671             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1672     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1673     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1674     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1675     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1676
1677     /* Note that there's no such thing as a projected cube texture. */
1678     switch(sampler_type) {
1679         case WINED3DSTT_1D:
1680             if (shadow)
1681             {
1682                 if (lod)
1683                 {
1684                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1685                 }
1686                 else if (grad)
1687                 {
1688                     if (gl_info->supported[EXT_GPU_SHADER4])
1689                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1690                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1691                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1692                     else
1693                     {
1694                         FIXME("Unsupported 1D shadow grad function.\n");
1695                         sample_function->name = "unsupported1DGrad";
1696                     }
1697                 }
1698                 else
1699                 {
1700                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1701                 }
1702                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1703             }
1704             else
1705             {
1706                 if (lod)
1707                 {
1708                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1709                 }
1710                 else if (grad)
1711                 {
1712                     if (gl_info->supported[EXT_GPU_SHADER4])
1713                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1714                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1715                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1716                     else
1717                     {
1718                         FIXME("Unsupported 1D grad function.\n");
1719                         sample_function->name = "unsupported1DGrad";
1720                     }
1721                 }
1722                 else
1723                 {
1724                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1725                 }
1726                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1727             }
1728             break;
1729
1730         case WINED3DSTT_2D:
1731             if (shadow)
1732             {
1733                 if (texrect)
1734                 {
1735                     if (lod)
1736                     {
1737                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1738                     }
1739                     else if (grad)
1740                     {
1741                         if (gl_info->supported[EXT_GPU_SHADER4])
1742                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1743                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1744                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1745                         else
1746                         {
1747                             FIXME("Unsupported RECT shadow grad function.\n");
1748                             sample_function->name = "unsupported2DRectGrad";
1749                         }
1750                     }
1751                     else
1752                     {
1753                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1754                     }
1755                 }
1756                 else
1757                 {
1758                     if (lod)
1759                     {
1760                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1761                     }
1762                     else if (grad)
1763                     {
1764                         if (gl_info->supported[EXT_GPU_SHADER4])
1765                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1766                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1767                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1768                         else
1769                         {
1770                             FIXME("Unsupported 2D shadow grad function.\n");
1771                             sample_function->name = "unsupported2DGrad";
1772                         }
1773                     }
1774                     else
1775                     {
1776                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1777                     }
1778                 }
1779                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1780             }
1781             else
1782             {
1783                 if (texrect)
1784                 {
1785                     if (lod)
1786                     {
1787                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1788                     }
1789                     else if (grad)
1790                     {
1791                         if (gl_info->supported[EXT_GPU_SHADER4])
1792                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1793                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1794                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1795                         else
1796                         {
1797                             FIXME("Unsupported RECT grad function.\n");
1798                             sample_function->name = "unsupported2DRectGrad";
1799                         }
1800                     }
1801                     else
1802                     {
1803                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1804                     }
1805                 }
1806                 else
1807                 {
1808                     if (lod)
1809                     {
1810                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1811                     }
1812                     else if (grad)
1813                     {
1814                         if (gl_info->supported[EXT_GPU_SHADER4])
1815                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1816                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1817                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1818                         else
1819                         {
1820                             FIXME("Unsupported 2D grad function.\n");
1821                             sample_function->name = "unsupported2DGrad";
1822                         }
1823                     }
1824                     else
1825                     {
1826                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1827                     }
1828                 }
1829                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1830             }
1831             break;
1832
1833         case WINED3DSTT_CUBE:
1834             if (shadow)
1835             {
1836                 FIXME("Unsupported Cube shadow function.\n");
1837                 sample_function->name = "unsupportedCubeShadow";
1838                 sample_function->coord_mask = 0;
1839             }
1840             else
1841             {
1842                 if (lod)
1843                 {
1844                     sample_function->name = "textureCubeLod";
1845                 }
1846                 else if (grad)
1847                 {
1848                     if (gl_info->supported[EXT_GPU_SHADER4])
1849                         sample_function->name = "textureCubeGrad";
1850                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1851                         sample_function->name = "textureCubeGradARB";
1852                     else
1853                     {
1854                         FIXME("Unsupported Cube grad function.\n");
1855                         sample_function->name = "unsupportedCubeGrad";
1856                     }
1857                 }
1858                 else
1859                 {
1860                     sample_function->name = "textureCube";
1861                 }
1862                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1863             }
1864             break;
1865
1866         case WINED3DSTT_VOLUME:
1867             if (shadow)
1868             {
1869                 FIXME("Unsupported 3D shadow function.\n");
1870                 sample_function->name = "unsupported3DShadow";
1871                 sample_function->coord_mask = 0;
1872             }
1873             else
1874             {
1875                 if (lod)
1876                 {
1877                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1878                 }
1879                 else  if (grad)
1880                 {
1881                     if (gl_info->supported[EXT_GPU_SHADER4])
1882                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1883                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1884                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1885                     else
1886                     {
1887                         FIXME("Unsupported 3D grad function.\n");
1888                         sample_function->name = "unsupported3DGrad";
1889                     }
1890                 }
1891                 else
1892                 {
1893                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1894                 }
1895                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1896             }
1897             break;
1898
1899         default:
1900             sample_function->name = "";
1901             sample_function->coord_mask = 0;
1902             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1903             break;
1904     }
1905 }
1906
1907 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1908         BOOL sign_fixup, enum fixup_channel_source channel_source)
1909 {
1910     switch(channel_source)
1911     {
1912         case CHANNEL_SOURCE_ZERO:
1913             strcat(arguments, "0.0");
1914             break;
1915
1916         case CHANNEL_SOURCE_ONE:
1917             strcat(arguments, "1.0");
1918             break;
1919
1920         case CHANNEL_SOURCE_X:
1921             strcat(arguments, reg_name);
1922             strcat(arguments, ".x");
1923             break;
1924
1925         case CHANNEL_SOURCE_Y:
1926             strcat(arguments, reg_name);
1927             strcat(arguments, ".y");
1928             break;
1929
1930         case CHANNEL_SOURCE_Z:
1931             strcat(arguments, reg_name);
1932             strcat(arguments, ".z");
1933             break;
1934
1935         case CHANNEL_SOURCE_W:
1936             strcat(arguments, reg_name);
1937             strcat(arguments, ".w");
1938             break;
1939
1940         default:
1941             FIXME("Unhandled channel source %#x\n", channel_source);
1942             strcat(arguments, "undefined");
1943             break;
1944     }
1945
1946     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1947 }
1948
1949 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1950 {
1951     struct wined3d_shader_dst_param dst;
1952     unsigned int mask_size, remaining;
1953     glsl_dst_param_t dst_param;
1954     char arguments[256];
1955     DWORD mask;
1956
1957     mask = 0;
1958     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1959     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1960     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1961     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1962     mask &= ins->dst[0].write_mask;
1963
1964     if (!mask) return; /* Nothing to do */
1965
1966     if (is_complex_fixup(fixup))
1967     {
1968         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1969         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1970         return;
1971     }
1972
1973     mask_size = shader_glsl_get_write_mask_size(mask);
1974
1975     dst = ins->dst[0];
1976     dst.write_mask = mask;
1977     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1978
1979     arguments[0] = '\0';
1980     remaining = mask_size;
1981     if (mask & WINED3DSP_WRITEMASK_0)
1982     {
1983         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1984         if (--remaining) strcat(arguments, ", ");
1985     }
1986     if (mask & WINED3DSP_WRITEMASK_1)
1987     {
1988         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1989         if (--remaining) strcat(arguments, ", ");
1990     }
1991     if (mask & WINED3DSP_WRITEMASK_2)
1992     {
1993         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1994         if (--remaining) strcat(arguments, ", ");
1995     }
1996     if (mask & WINED3DSP_WRITEMASK_3)
1997     {
1998         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1999         if (--remaining) strcat(arguments, ", ");
2000     }
2001
2002     if (mask_size > 1)
2003     {
2004         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
2005                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
2006     }
2007     else
2008     {
2009         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2010     }
2011 }
2012
2013 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2014         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
2015         const char *dx, const char *dy,
2016         const char *bias, const char *coord_reg_fmt, ...)
2017 {
2018     const char *sampler_base;
2019     char dst_swizzle[6];
2020     struct color_fixup_desc fixup;
2021     BOOL np2_fixup = FALSE;
2022     va_list args;
2023
2024     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2025
2026     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2027     {
2028         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2029         fixup = priv->cur_ps_args->color_fixup[sampler];
2030         sampler_base = "Psampler";
2031
2032         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2033             if(bias) {
2034                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2035             } else {
2036                 np2_fixup = TRUE;
2037             }
2038         }
2039     } else {
2040         sampler_base = "Vsampler";
2041         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2042     }
2043
2044     shader_glsl_append_dst(ins->ctx->buffer, ins);
2045
2046     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2047
2048     va_start(args, coord_reg_fmt);
2049     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2050     va_end(args);
2051
2052     if(bias) {
2053         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2054     } else {
2055         if (np2_fixup) {
2056             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2057             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2058
2059             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2060                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2061         } else if(dx && dy) {
2062             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2063         } else {
2064             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2065         }
2066     }
2067
2068     if(!is_identity_fixup(fixup)) {
2069         shader_glsl_color_correction(ins, fixup);
2070     }
2071 }
2072
2073 /*****************************************************************************
2074  * Begin processing individual instruction opcodes
2075  ****************************************************************************/
2076
2077 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2078 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2079 {
2080     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2081     glsl_src_param_t src0_param;
2082     glsl_src_param_t src1_param;
2083     DWORD write_mask;
2084     char op;
2085
2086     /* Determine the GLSL operator to use based on the opcode */
2087     switch (ins->handler_idx)
2088     {
2089         case WINED3DSIH_MUL: op = '*'; break;
2090         case WINED3DSIH_ADD: op = '+'; break;
2091         case WINED3DSIH_SUB: op = '-'; break;
2092         default:
2093             op = ' ';
2094             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2095             break;
2096     }
2097
2098     write_mask = shader_glsl_append_dst(buffer, ins);
2099     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2100     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2101     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2102 }
2103
2104 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2105 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2106 {
2107     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2108     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2109     glsl_src_param_t src0_param;
2110     DWORD write_mask;
2111
2112     write_mask = shader_glsl_append_dst(buffer, ins);
2113     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2114
2115     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2116      * shader versions WINED3DSIO_MOVA is used for this. */
2117     if (ins->ctx->reg_maps->shader_version.major == 1
2118             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2119             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2120     {
2121         /* This is a simple floor() */
2122         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2123         if (mask_size > 1) {
2124             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2125         } else {
2126             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2127         }
2128     }
2129     else if(ins->handler_idx == WINED3DSIH_MOVA)
2130     {
2131         /* We need to *round* to the nearest int here. */
2132         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2133
2134         if (gl_info->supported[EXT_GPU_SHADER4])
2135         {
2136             if (mask_size > 1)
2137                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2138             else
2139                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2140         }
2141         else
2142         {
2143             if (mask_size > 1)
2144                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2145                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2146             else
2147                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2148                         src0_param.param_str, src0_param.param_str);
2149         }
2150     }
2151     else
2152     {
2153         shader_addline(buffer, "%s);\n", src0_param.param_str);
2154     }
2155 }
2156
2157 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2158 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2159 {
2160     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2161     glsl_src_param_t src0_param;
2162     glsl_src_param_t src1_param;
2163     DWORD dst_write_mask, src_write_mask;
2164     unsigned int dst_size = 0;
2165
2166     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2167     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2168
2169     /* dp3 works on vec3, dp4 on vec4 */
2170     if (ins->handler_idx == WINED3DSIH_DP4)
2171     {
2172         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2173     } else {
2174         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2175     }
2176
2177     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2178     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2179
2180     if (dst_size > 1) {
2181         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2182     } else {
2183         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2184     }
2185 }
2186
2187 /* Note that this instruction has some restrictions. The destination write mask
2188  * can't contain the w component, and the source swizzles have to be .xyzw */
2189 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2190 {
2191     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2192     glsl_src_param_t src0_param;
2193     glsl_src_param_t src1_param;
2194     char dst_mask[6];
2195
2196     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2197     shader_glsl_append_dst(ins->ctx->buffer, ins);
2198     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2199     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2200     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2201 }
2202
2203 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2204  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2205  * GLSL uses the value as-is. */
2206 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2207 {
2208     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2209     glsl_src_param_t src0_param;
2210     glsl_src_param_t src1_param;
2211     DWORD dst_write_mask;
2212     unsigned int dst_size;
2213
2214     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2215     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2216
2217     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2218     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2219
2220     if (dst_size > 1)
2221     {
2222         shader_addline(buffer, "vec%u(pow(abs(%s), %s)));\n",
2223                 dst_size, src0_param.param_str, src1_param.param_str);
2224     }
2225     else
2226     {
2227         shader_addline(buffer, "pow(abs(%s), %s));\n",
2228                 src0_param.param_str, src1_param.param_str);
2229     }
2230 }
2231
2232 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2233  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2234  * GLSL uses the value as-is. */
2235 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2236 {
2237     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2238     glsl_src_param_t src0_param;
2239     DWORD dst_write_mask;
2240     unsigned int dst_size;
2241
2242     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2243     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2244
2245     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2246
2247     if (dst_size > 1)
2248     {
2249         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2250                 dst_size, src0_param.param_str);
2251     }
2252     else
2253     {
2254         shader_addline(buffer, "log2(abs(%s)));\n",
2255                 src0_param.param_str);
2256     }
2257 }
2258
2259 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2260 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2261 {
2262     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2263     glsl_src_param_t src_param;
2264     const char *instruction;
2265     DWORD write_mask;
2266     unsigned i;
2267
2268     /* Determine the GLSL function to use based on the opcode */
2269     /* TODO: Possibly make this a table for faster lookups */
2270     switch (ins->handler_idx)
2271     {
2272         case WINED3DSIH_MIN: instruction = "min"; break;
2273         case WINED3DSIH_MAX: instruction = "max"; break;
2274         case WINED3DSIH_ABS: instruction = "abs"; break;
2275         case WINED3DSIH_FRC: instruction = "fract"; break;
2276         case WINED3DSIH_EXP: instruction = "exp2"; break;
2277         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2278         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2279         default: instruction = "";
2280             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2281             break;
2282     }
2283
2284     write_mask = shader_glsl_append_dst(buffer, ins);
2285
2286     shader_addline(buffer, "%s(", instruction);
2287
2288     if (ins->src_count)
2289     {
2290         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2291         shader_addline(buffer, "%s", src_param.param_str);
2292         for (i = 1; i < ins->src_count; ++i)
2293         {
2294             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2295             shader_addline(buffer, ", %s", src_param.param_str);
2296         }
2297     }
2298
2299     shader_addline(buffer, "));\n");
2300 }
2301
2302 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2303 {
2304     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2305     glsl_src_param_t src_param;
2306     unsigned int mask_size;
2307     DWORD write_mask;
2308     char dst_mask[6];
2309
2310     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2311     mask_size = shader_glsl_get_write_mask_size(write_mask);
2312     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2313
2314     shader_addline(buffer, "tmp0.x = length(%s);\n", src_param.param_str);
2315     shader_glsl_append_dst(buffer, ins);
2316     if (mask_size > 1)
2317     {
2318         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s / tmp0.x));\n",
2319                 mask_size, src_param.param_str);
2320     }
2321     else
2322     {
2323         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s / tmp0.x));\n",
2324                 src_param.param_str);
2325     }
2326 }
2327
2328 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2329  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2330  *   dst.x = 2^(floor(src))
2331  *   dst.y = src - floor(src)
2332  *   dst.z = 2^src   (partial precision is allowed, but optional)
2333  *   dst.w = 1.0;
2334  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2335  *   dst = 2^src;    (partial precision is allowed, but optional)
2336  */
2337 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2338 {
2339     glsl_src_param_t src_param;
2340
2341     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2342
2343     if (ins->ctx->reg_maps->shader_version.major < 2)
2344     {
2345         char dst_mask[6];
2346
2347         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2348         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2349         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2350         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2351
2352         shader_glsl_append_dst(ins->ctx->buffer, ins);
2353         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2354         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2355     } else {
2356         DWORD write_mask;
2357         unsigned int mask_size;
2358
2359         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2360         mask_size = shader_glsl_get_write_mask_size(write_mask);
2361
2362         if (mask_size > 1) {
2363             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2364         } else {
2365             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2366         }
2367     }
2368 }
2369
2370 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2371 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2372 {
2373     glsl_src_param_t src_param;
2374     DWORD write_mask;
2375     unsigned int mask_size;
2376
2377     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2378     mask_size = shader_glsl_get_write_mask_size(write_mask);
2379     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2380
2381     if (mask_size > 1)
2382     {
2383         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2384                 mask_size, src_param.param_str);
2385     }
2386     else
2387     {
2388         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2389                 src_param.param_str);
2390     }
2391 }
2392
2393 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2394 {
2395     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2396     glsl_src_param_t src_param;
2397     DWORD write_mask;
2398     unsigned int mask_size;
2399
2400     write_mask = shader_glsl_append_dst(buffer, ins);
2401     mask_size = shader_glsl_get_write_mask_size(write_mask);
2402
2403     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2404
2405     if (mask_size > 1)
2406     {
2407         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2408                 mask_size, src_param.param_str);
2409     }
2410     else
2411     {
2412         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2413                 src_param.param_str);
2414     }
2415 }
2416
2417 /** Process signed comparison opcodes in GLSL. */
2418 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2419 {
2420     glsl_src_param_t src0_param;
2421     glsl_src_param_t src1_param;
2422     DWORD write_mask;
2423     unsigned int mask_size;
2424
2425     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2426     mask_size = shader_glsl_get_write_mask_size(write_mask);
2427     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2428     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2429
2430     if (mask_size > 1) {
2431         const char *compare;
2432
2433         switch(ins->handler_idx)
2434         {
2435             case WINED3DSIH_SLT: compare = "lessThan"; break;
2436             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2437             default: compare = "";
2438                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2439         }
2440
2441         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2442                 src0_param.param_str, src1_param.param_str);
2443     } else {
2444         switch(ins->handler_idx)
2445         {
2446             case WINED3DSIH_SLT:
2447                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2448                  * to return 0.0 but step returns 1.0 because step is not < x
2449                  * An alternative is a bvec compare padded with an unused second component.
2450                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2451                  * issue. Playing with not() is not possible either because not() does not accept
2452                  * a scalar.
2453                  */
2454                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2455                         src0_param.param_str, src1_param.param_str);
2456                 break;
2457             case WINED3DSIH_SGE:
2458                 /* Here we can use the step() function and safe a conditional */
2459                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2460                 break;
2461             default:
2462                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2463         }
2464
2465     }
2466 }
2467
2468 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2469 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2470 {
2471     glsl_src_param_t src0_param;
2472     glsl_src_param_t src1_param;
2473     glsl_src_param_t src2_param;
2474     DWORD write_mask, cmp_channel = 0;
2475     unsigned int i, j;
2476     char mask_char[6];
2477     BOOL temp_destination = FALSE;
2478
2479     if (shader_is_scalar(&ins->src[0].reg))
2480     {
2481         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2482
2483         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2484         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2485         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2486
2487         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2488                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2489     } else {
2490         DWORD dst_mask = ins->dst[0].write_mask;
2491         struct wined3d_shader_dst_param dst = ins->dst[0];
2492
2493         /* Cycle through all source0 channels */
2494         for (i=0; i<4; i++) {
2495             write_mask = 0;
2496             /* Find the destination channels which use the current source0 channel */
2497             for (j=0; j<4; j++) {
2498                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2499                 {
2500                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2501                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2502                 }
2503             }
2504             dst.write_mask = dst_mask & write_mask;
2505
2506             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2507             * The first lines may overwrite source parameters of the following lines.
2508             * Deal with that by using a temporary destination register if needed
2509             */
2510             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2511                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2512                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2513                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2514                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2515                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2516             {
2517                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2518                 if (!write_mask) continue;
2519                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2520                 temp_destination = TRUE;
2521             } else {
2522                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2523                 if (!write_mask) continue;
2524             }
2525
2526             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2527             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2528             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2529
2530             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2531                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2532         }
2533
2534         if(temp_destination) {
2535             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2536             shader_glsl_append_dst(ins->ctx->buffer, ins);
2537             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2538         }
2539     }
2540
2541 }
2542
2543 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2544 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2545  * the compare is done per component of src0. */
2546 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2547 {
2548     struct wined3d_shader_dst_param dst;
2549     glsl_src_param_t src0_param;
2550     glsl_src_param_t src1_param;
2551     glsl_src_param_t src2_param;
2552     DWORD write_mask, cmp_channel = 0;
2553     unsigned int i, j;
2554     DWORD dst_mask;
2555     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2556             ins->ctx->reg_maps->shader_version.minor);
2557
2558     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2559     {
2560         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2561         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2562         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2563         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2564
2565         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2566         if (ins->coissue)
2567         {
2568             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2569         } else {
2570             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2571                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2572         }
2573         return;
2574     }
2575     /* Cycle through all source0 channels */
2576     dst_mask = ins->dst[0].write_mask;
2577     dst = ins->dst[0];
2578     for (i=0; i<4; i++) {
2579         write_mask = 0;
2580         /* Find the destination channels which use the current source0 channel */
2581         for (j=0; j<4; j++) {
2582             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2583             {
2584                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2585                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2586             }
2587         }
2588
2589         dst.write_mask = dst_mask & write_mask;
2590         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2591         if (!write_mask) continue;
2592
2593         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2594         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2595         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2596
2597         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2598                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2599     }
2600 }
2601
2602 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2603 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2604 {
2605     glsl_src_param_t src0_param;
2606     glsl_src_param_t src1_param;
2607     glsl_src_param_t src2_param;
2608     DWORD write_mask;
2609
2610     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2611     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2612     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2613     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2614     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2615             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2616 }
2617
2618 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2619    Vertex shaders to GLSL codes */
2620 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2621 {
2622     int i;
2623     int nComponents = 0;
2624     struct wined3d_shader_dst_param tmp_dst = {{0}};
2625     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2626     struct wined3d_shader_instruction tmp_ins;
2627
2628     memset(&tmp_ins, 0, sizeof(tmp_ins));
2629
2630     /* Set constants for the temporary argument */
2631     tmp_ins.ctx = ins->ctx;
2632     tmp_ins.dst_count = 1;
2633     tmp_ins.dst = &tmp_dst;
2634     tmp_ins.src_count = 2;
2635     tmp_ins.src = tmp_src;
2636
2637     switch(ins->handler_idx)
2638     {
2639         case WINED3DSIH_M4x4:
2640             nComponents = 4;
2641             tmp_ins.handler_idx = WINED3DSIH_DP4;
2642             break;
2643         case WINED3DSIH_M4x3:
2644             nComponents = 3;
2645             tmp_ins.handler_idx = WINED3DSIH_DP4;
2646             break;
2647         case WINED3DSIH_M3x4:
2648             nComponents = 4;
2649             tmp_ins.handler_idx = WINED3DSIH_DP3;
2650             break;
2651         case WINED3DSIH_M3x3:
2652             nComponents = 3;
2653             tmp_ins.handler_idx = WINED3DSIH_DP3;
2654             break;
2655         case WINED3DSIH_M3x2:
2656             nComponents = 2;
2657             tmp_ins.handler_idx = WINED3DSIH_DP3;
2658             break;
2659         default:
2660             break;
2661     }
2662
2663     tmp_dst = ins->dst[0];
2664     tmp_src[0] = ins->src[0];
2665     tmp_src[1] = ins->src[1];
2666     for (i = 0; i < nComponents; ++i)
2667     {
2668         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2669         shader_glsl_dot(&tmp_ins);
2670         ++tmp_src[1].reg.idx;
2671     }
2672 }
2673
2674 /**
2675     The LRP instruction performs a component-wise linear interpolation
2676     between the second and third operands using the first operand as the
2677     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2678     This is equivalent to mix(src2, src1, src0);
2679 */
2680 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2681 {
2682     glsl_src_param_t src0_param;
2683     glsl_src_param_t src1_param;
2684     glsl_src_param_t src2_param;
2685     DWORD write_mask;
2686
2687     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2688
2689     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2690     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2691     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2692
2693     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2694             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2695 }
2696
2697 /** Process the WINED3DSIO_LIT instruction in GLSL:
2698  * dst.x = dst.w = 1.0
2699  * dst.y = (src0.x > 0) ? src0.x
2700  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2701  *                                        where src.w is clamped at +- 128
2702  */
2703 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2704 {
2705     glsl_src_param_t src0_param;
2706     glsl_src_param_t src1_param;
2707     glsl_src_param_t src3_param;
2708     char dst_mask[6];
2709
2710     shader_glsl_append_dst(ins->ctx->buffer, ins);
2711     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2712
2713     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2714     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2715     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2716
2717     /* The sdk specifies the instruction like this
2718      * dst.x = 1.0;
2719      * if(src.x > 0.0) dst.y = src.x
2720      * else dst.y = 0.0.
2721      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2722      * else dst.z = 0.0;
2723      * dst.w = 1.0;
2724      *
2725      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2726      * dst.x = 1.0                                  ... No further explanation needed
2727      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2728      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2729      * dst.w = 1.0.                                 ... Nothing fancy.
2730      *
2731      * So we still have one conditional in there. So do this:
2732      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2733      *
2734      * 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),
2735      * 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.
2736      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2737      */
2738     shader_addline(ins->ctx->buffer,
2739             "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",
2740             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2741 }
2742
2743 /** Process the WINED3DSIO_DST instruction in GLSL:
2744  * dst.x = 1.0
2745  * dst.y = src0.x * src0.y
2746  * dst.z = src0.z
2747  * dst.w = src1.w
2748  */
2749 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2750 {
2751     glsl_src_param_t src0y_param;
2752     glsl_src_param_t src0z_param;
2753     glsl_src_param_t src1y_param;
2754     glsl_src_param_t src1w_param;
2755     char dst_mask[6];
2756
2757     shader_glsl_append_dst(ins->ctx->buffer, ins);
2758     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2759
2760     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2761     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2762     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2763     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2764
2765     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2766             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2767 }
2768
2769 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2770  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2771  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2772  *
2773  * dst.x = cos(src0.?)
2774  * dst.y = sin(src0.?)
2775  * dst.z = dst.z
2776  * dst.w = dst.w
2777  */
2778 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2779 {
2780     glsl_src_param_t src0_param;
2781     DWORD write_mask;
2782
2783     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2784     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2785
2786     switch (write_mask) {
2787         case WINED3DSP_WRITEMASK_0:
2788             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2789             break;
2790
2791         case WINED3DSP_WRITEMASK_1:
2792             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2793             break;
2794
2795         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2796             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2797             break;
2798
2799         default:
2800             ERR("Write mask should be .x, .y or .xy\n");
2801             break;
2802     }
2803 }
2804
2805 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2806  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2807  * generate invalid code
2808  */
2809 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2810 {
2811     glsl_src_param_t src0_param;
2812     DWORD write_mask;
2813
2814     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2815     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2816
2817     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2818 }
2819
2820 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2821  * Start a for() loop where src1.y is the initial value of aL,
2822  *  increment aL by src1.z for a total of src1.x iterations.
2823  *  Need to use a temporary variable for this operation.
2824  */
2825 /* FIXME: I don't think nested loops will work correctly this way. */
2826 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2827 {
2828     glsl_src_param_t src1_param;
2829     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2830     const DWORD *control_values = NULL;
2831     const local_constant *constant;
2832
2833     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2834
2835     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2836      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2837      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2838      * addressing.
2839      */
2840     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2841     {
2842         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2843             if (constant->idx == ins->src[1].reg.idx)
2844             {
2845                 control_values = constant->value;
2846                 break;
2847             }
2848         }
2849     }
2850
2851     if (control_values)
2852     {
2853         struct wined3d_shader_loop_control loop_control;
2854         loop_control.count = control_values[0];
2855         loop_control.start = control_values[1];
2856         loop_control.step = (int)control_values[2];
2857
2858         if (loop_control.step > 0)
2859         {
2860             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2861                     shader->baseShader.cur_loop_depth, loop_control.start,
2862                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2863                     shader->baseShader.cur_loop_depth, loop_control.step);
2864         }
2865         else if (loop_control.step < 0)
2866         {
2867             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2868                     shader->baseShader.cur_loop_depth, loop_control.start,
2869                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2870                     shader->baseShader.cur_loop_depth, loop_control.step);
2871         }
2872         else
2873         {
2874             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2875                     shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2876                     shader->baseShader.cur_loop_depth, loop_control.count,
2877                     shader->baseShader.cur_loop_depth);
2878         }
2879     } else {
2880         shader_addline(ins->ctx->buffer,
2881                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2882                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2883                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2884                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2885     }
2886
2887     shader->baseShader.cur_loop_depth++;
2888     shader->baseShader.cur_loop_regno++;
2889 }
2890
2891 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2892 {
2893     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2894
2895     shader_addline(ins->ctx->buffer, "}\n");
2896
2897     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2898     {
2899         shader->baseShader.cur_loop_depth--;
2900         shader->baseShader.cur_loop_regno--;
2901     }
2902
2903     if (ins->handler_idx == WINED3DSIH_ENDREP)
2904     {
2905         shader->baseShader.cur_loop_depth--;
2906     }
2907 }
2908
2909 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2910 {
2911     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2912     glsl_src_param_t src0_param;
2913     const DWORD *control_values = NULL;
2914     const local_constant *constant;
2915
2916     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2917     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2918     {
2919         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2920         {
2921             if (constant->idx == ins->src[0].reg.idx)
2922             {
2923                 control_values = constant->value;
2924                 break;
2925             }
2926         }
2927     }
2928
2929     if(control_values) {
2930         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2931                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2932                        control_values[0], shader->baseShader.cur_loop_depth);
2933     } else {
2934         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2935         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2936                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2937                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2938     }
2939     shader->baseShader.cur_loop_depth++;
2940 }
2941
2942 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2943 {
2944     glsl_src_param_t src0_param;
2945
2946     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2947     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2948 }
2949
2950 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2951 {
2952     glsl_src_param_t src0_param;
2953     glsl_src_param_t src1_param;
2954
2955     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2956     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2957
2958     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2959             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2960 }
2961
2962 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2963 {
2964     shader_addline(ins->ctx->buffer, "} else {\n");
2965 }
2966
2967 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2968 {
2969     shader_addline(ins->ctx->buffer, "break;\n");
2970 }
2971
2972 /* FIXME: According to MSDN the compare is done per component. */
2973 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2974 {
2975     glsl_src_param_t src0_param;
2976     glsl_src_param_t src1_param;
2977
2978     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2979     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2980
2981     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2982             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2983 }
2984
2985 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2986 {
2987     shader_addline(ins->ctx->buffer, "}\n");
2988     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2989 }
2990
2991 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2992 {
2993     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2994 }
2995
2996 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2997 {
2998     glsl_src_param_t src1_param;
2999
3000     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3001     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3002 }
3003
3004 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3005 {
3006     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3007      * function only suppresses the unhandled instruction warning
3008      */
3009 }
3010
3011 /*********************************************
3012  * Pixel Shader Specific Code begins here
3013  ********************************************/
3014 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3015 {
3016     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3017     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3018     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3019             ins->ctx->reg_maps->shader_version.minor);
3020     glsl_sample_function_t sample_function;
3021     IWineD3DBaseTextureImpl *texture;
3022     DWORD sample_flags = 0;
3023     DWORD sampler_idx;
3024     DWORD mask = 0, swizzle;
3025
3026     /* 1.0-1.4: Use destination register as sampler source.
3027      * 2.0+: Use provided sampler source. */
3028     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3029     else sampler_idx = ins->src[1].reg.idx;
3030     texture = deviceImpl->stateBlock->state.textures[sampler_idx];
3031
3032     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3033     {
3034         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3035         DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3036                 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3037         WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3038
3039         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3040         if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3041             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3042             switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3043                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3044                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3045                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3046                 case WINED3DTTFF_COUNT4:
3047                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3048             }
3049         }
3050     }
3051     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3052     {
3053         DWORD src_mod = ins->src[0].modifiers;
3054
3055         if (src_mod == WINED3DSPSM_DZ) {
3056             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3057             mask = WINED3DSP_WRITEMASK_2;
3058         } else if (src_mod == WINED3DSPSM_DW) {
3059             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3060             mask = WINED3DSP_WRITEMASK_3;
3061         }
3062     } else {
3063         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3064         {
3065             /* ps 2.0 texldp instruction always divides by the fourth component. */
3066             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3067             mask = WINED3DSP_WRITEMASK_3;
3068         }
3069     }
3070
3071     if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3072         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3073
3074     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3075     mask |= sample_function.coord_mask;
3076
3077     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3078     else swizzle = ins->src[1].swizzle;
3079
3080     /* 1.0-1.3: Use destination register as coordinate source.
3081        1.4+: Use provided coordinate source register. */
3082     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3083     {
3084         char coord_mask[6];
3085         shader_glsl_write_mask_to_str(mask, coord_mask);
3086         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3087                 "T%u%s", sampler_idx, coord_mask);
3088     } else {
3089         glsl_src_param_t coord_param;
3090         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3091         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3092         {
3093             glsl_src_param_t bias;
3094             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3095             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3096                     "%s", coord_param.param_str);
3097         } else {
3098             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3099                     "%s", coord_param.param_str);
3100         }
3101     }
3102 }
3103
3104 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3105 {
3106     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3107     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3108     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3109     glsl_sample_function_t sample_function;
3110     glsl_src_param_t coord_param, dx_param, dy_param;
3111     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3112     DWORD sampler_idx;
3113     DWORD swizzle = ins->src[1].swizzle;
3114     IWineD3DBaseTextureImpl *texture;
3115
3116     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3117     {
3118         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3119         shader_glsl_tex(ins);
3120         return;
3121     }
3122
3123     sampler_idx = ins->src[1].reg.idx;
3124     texture = deviceImpl->stateBlock->state.textures[sampler_idx];
3125     if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3126         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3127
3128     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3129     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3130     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3131     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3132
3133     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3134                                 "%s", coord_param.param_str);
3135 }
3136
3137 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3138 {
3139     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3140     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3141     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3142     glsl_sample_function_t sample_function;
3143     glsl_src_param_t coord_param, lod_param;
3144     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3145     DWORD sampler_idx;
3146     DWORD swizzle = ins->src[1].swizzle;
3147     IWineD3DBaseTextureImpl *texture;
3148
3149     sampler_idx = ins->src[1].reg.idx;
3150     texture = deviceImpl->stateBlock->state.textures[sampler_idx];
3151     if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3152         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3153
3154     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3155     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3156
3157     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3158
3159     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3160             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3161     {
3162         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3163          * However, they seem to work just fine in fragment shaders as well. */
3164         WARN("Using %s in fragment shader.\n", sample_function.name);
3165     }
3166     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3167             "%s", coord_param.param_str);
3168 }
3169
3170 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3171 {
3172     /* FIXME: Make this work for more than just 2D textures */
3173     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3174     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3175
3176     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3177     {
3178         char dst_mask[6];
3179
3180         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3181         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3182                 ins->dst[0].reg.idx, dst_mask);
3183     } else {
3184         DWORD reg = ins->src[0].reg.idx;
3185         DWORD src_mod = ins->src[0].modifiers;
3186         char dst_swizzle[6];
3187
3188         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3189
3190         if (src_mod == WINED3DSPSM_DZ) {
3191             glsl_src_param_t div_param;
3192             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3193             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3194
3195             if (mask_size > 1) {
3196                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3197             } else {
3198                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3199             }
3200         } else if (src_mod == WINED3DSPSM_DW) {
3201             glsl_src_param_t div_param;
3202             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3203             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3204
3205             if (mask_size > 1) {
3206                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3207             } else {
3208                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3209             }
3210         } else {
3211             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3212         }
3213     }
3214 }
3215
3216 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3217  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3218  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3219 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3220 {
3221     glsl_src_param_t src0_param;
3222     glsl_sample_function_t sample_function;
3223     DWORD sampler_idx = ins->dst[0].reg.idx;
3224     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3225     UINT mask_size;
3226
3227     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3228
3229     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3230      * scalar, and projected sampling would require 4.
3231      *
3232      * It is a dependent read - not valid with conditional NP2 textures
3233      */
3234     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3235     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3236
3237     switch(mask_size)
3238     {
3239         case 1:
3240             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3241                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3242             break;
3243
3244         case 2:
3245             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3246                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3247             break;
3248
3249         case 3:
3250             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3251                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3252             break;
3253
3254         default:
3255             FIXME("Unexpected mask size %u\n", mask_size);
3256             break;
3257     }
3258 }
3259
3260 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3261  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3262 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3263 {
3264     glsl_src_param_t src0_param;
3265     DWORD dstreg = ins->dst[0].reg.idx;
3266     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3267     DWORD dst_mask;
3268     unsigned int mask_size;
3269
3270     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3271     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3272     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3273
3274     if (mask_size > 1) {
3275         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3276     } else {
3277         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3278     }
3279 }
3280
3281 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3282  * Calculate the depth as dst.x / dst.y   */
3283 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3284 {
3285     glsl_dst_param_t dst_param;
3286
3287     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3288
3289     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3290      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3291      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3292      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3293      * >= 1.0 or < 0.0
3294      */
3295     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3296             dst_param.reg_name, dst_param.reg_name);
3297 }
3298
3299 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3300  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3301  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3302  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3303  */
3304 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3305 {
3306     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3307     DWORD dstreg = ins->dst[0].reg.idx;
3308     glsl_src_param_t src0_param;
3309
3310     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3311
3312     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3313     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3314 }
3315
3316 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3317  * Calculate the 1st of a 2-row matrix multiplication. */
3318 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3319 {
3320     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3321     DWORD reg = ins->dst[0].reg.idx;
3322     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3323     glsl_src_param_t src0_param;
3324
3325     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3326     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3327 }
3328
3329 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3330  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3331 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3332 {
3333     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3334     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3335     DWORD reg = ins->dst[0].reg.idx;
3336     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3337     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3338     glsl_src_param_t src0_param;
3339
3340     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3341     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3342     current_state->texcoord_w[current_state->current_row++] = reg;
3343 }
3344
3345 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3346 {
3347     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3348     DWORD reg = ins->dst[0].reg.idx;
3349     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3350     glsl_src_param_t src0_param;
3351     glsl_sample_function_t sample_function;
3352
3353     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3354     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3355
3356     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3357
3358     /* Sample the texture using the calculated coordinates */
3359     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3360 }
3361
3362 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3363  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3364 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3365 {
3366     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3367     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3368     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3369     glsl_src_param_t src0_param;
3370     DWORD reg = ins->dst[0].reg.idx;
3371     glsl_sample_function_t sample_function;
3372
3373     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3374     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3375
3376     /* Dependent read, not valid with conditional NP2 */
3377     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3378
3379     /* Sample the texture using the calculated coordinates */
3380     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3381
3382     current_state->current_row = 0;
3383 }
3384
3385 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3386  * Perform the 3rd row of a 3x3 matrix multiply */
3387 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3388 {
3389     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3390     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3391     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3392     glsl_src_param_t src0_param;
3393     char dst_mask[6];
3394     DWORD reg = ins->dst[0].reg.idx;
3395
3396     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3397
3398     shader_glsl_append_dst(ins->ctx->buffer, ins);
3399     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3400     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3401
3402     current_state->current_row = 0;
3403 }
3404
3405 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3406  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3407 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3408 {
3409     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3410     DWORD reg = ins->dst[0].reg.idx;
3411     glsl_src_param_t src0_param;
3412     glsl_src_param_t src1_param;
3413     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3414     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3415     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3416     glsl_sample_function_t sample_function;
3417
3418     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3419     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3420
3421     /* Perform the last matrix multiply operation */
3422     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3423     /* Reflection calculation */
3424     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3425
3426     /* Dependent read, not valid with conditional NP2 */
3427     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3428
3429     /* Sample the texture */
3430     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3431
3432     current_state->current_row = 0;
3433 }
3434
3435 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3436  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3437 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3438 {
3439     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3440     DWORD reg = ins->dst[0].reg.idx;
3441     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3442     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3443     glsl_src_param_t src0_param;
3444     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3445     glsl_sample_function_t sample_function;
3446
3447     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3448
3449     /* Perform the last matrix multiply operation */
3450     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3451
3452     /* Construct the eye-ray vector from w coordinates */
3453     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3454             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3455     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3456
3457     /* Dependent read, not valid with conditional NP2 */
3458     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3459
3460     /* Sample the texture using the calculated coordinates */
3461     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3462
3463     current_state->current_row = 0;
3464 }
3465
3466 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3467  * Apply a fake bump map transform.
3468  * texbem is pshader <= 1.3 only, this saves a few version checks
3469  */
3470 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3471 {
3472     const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3473     glsl_sample_function_t sample_function;
3474     glsl_src_param_t coord_param;
3475     DWORD sampler_idx;
3476     DWORD mask;
3477     DWORD flags;
3478     char coord_mask[6];
3479
3480     sampler_idx = ins->dst[0].reg.idx;
3481     flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3482             & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3483
3484     /* Dependent read, not valid with conditional NP2 */
3485     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3486     mask = sample_function.coord_mask;
3487
3488     shader_glsl_write_mask_to_str(mask, coord_mask);
3489
3490     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3491          * so we can't let the GL handle this.
3492          */
3493     if (flags & WINED3D_PSARGS_PROJECTED) {
3494         DWORD div_mask=0;
3495         char coord_div_mask[3];
3496         switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3497             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3498             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3499             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3500             case WINED3DTTFF_COUNT4:
3501             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3502         }
3503         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3504         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3505     }
3506
3507     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3508
3509     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3510             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3511             coord_param.param_str, coord_mask);
3512
3513     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3514     {
3515         glsl_src_param_t luminance_param;
3516         glsl_dst_param_t dst_param;
3517
3518         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3519         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3520
3521         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3522                 dst_param.reg_name, dst_param.mask_str,
3523                 luminance_param.param_str, sampler_idx, sampler_idx);
3524     }
3525 }
3526
3527 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3528 {
3529     glsl_src_param_t src0_param, src1_param;
3530     DWORD sampler_idx = ins->dst[0].reg.idx;
3531
3532     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3533     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3534
3535     shader_glsl_append_dst(ins->ctx->buffer, ins);
3536     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3537             src0_param.param_str, sampler_idx, src1_param.param_str);
3538 }
3539
3540 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3541  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3542 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3543 {
3544     glsl_src_param_t src0_param;
3545     DWORD sampler_idx = ins->dst[0].reg.idx;
3546     glsl_sample_function_t sample_function;
3547
3548     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3549
3550     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3551     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3552             "%s.wx", src0_param.reg_name);
3553 }
3554
3555 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3556  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3557 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3558 {
3559     glsl_src_param_t src0_param;
3560     DWORD sampler_idx = ins->dst[0].reg.idx;
3561     glsl_sample_function_t sample_function;
3562
3563     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3564
3565     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3566     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3567             "%s.yz", src0_param.reg_name);
3568 }
3569
3570 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3571  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3572 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3573 {
3574     glsl_src_param_t src0_param;
3575     DWORD sampler_idx = ins->dst[0].reg.idx;
3576     glsl_sample_function_t sample_function;
3577
3578     /* Dependent read, not valid with conditional NP2 */
3579     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3580     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3581
3582     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3583             "%s", src0_param.param_str);
3584 }
3585
3586 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3587  * If any of the first 3 components are < 0, discard this pixel */
3588 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3589 {
3590     glsl_dst_param_t dst_param;
3591
3592     /* The argument is a destination parameter, and no writemasks are allowed */
3593     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3594     if (ins->ctx->reg_maps->shader_version.major >= 2)
3595     {
3596         /* 2.0 shaders compare all 4 components in texkill */
3597         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3598     } else {
3599         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3600          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3601          * 4 components are defined, only the first 3 are used
3602          */
3603         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3604     }
3605 }
3606
3607 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3608  * dst = dot2(src0, src1) + src2 */
3609 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3610 {
3611     glsl_src_param_t src0_param;
3612     glsl_src_param_t src1_param;
3613     glsl_src_param_t src2_param;
3614     DWORD write_mask;
3615     unsigned int mask_size;
3616
3617     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3618     mask_size = shader_glsl_get_write_mask_size(write_mask);
3619
3620     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3621     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3622     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3623
3624     if (mask_size > 1) {
3625         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3626                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3627     } else {
3628         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3629                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3630     }
3631 }
3632
3633 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3634         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3635         enum vertexprocessing_mode vertexprocessing)
3636 {
3637     unsigned int i;
3638     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3639     WORD map = reg_maps->input_registers;
3640
3641     for (i = 0; map; map >>= 1, ++i)
3642     {
3643         const char *semantic_name;
3644         UINT semantic_idx;
3645         char reg_mask[6];
3646
3647         /* Unused */
3648         if (!(map & 1)) continue;
3649
3650         semantic_name = input_signature[i].semantic_name;
3651         semantic_idx = input_signature[i].semantic_idx;
3652         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3653
3654         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3655         {
3656             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3657                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3658                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3659             else
3660                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3661                         This->input_reg_map[i], reg_mask, reg_mask);
3662         }
3663         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3664         {
3665             if (!semantic_idx)
3666                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3667                         This->input_reg_map[i], reg_mask, reg_mask);
3668             else if (semantic_idx == 1)
3669                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3670                         This->input_reg_map[i], reg_mask, reg_mask);
3671             else
3672                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3673                         This->input_reg_map[i], reg_mask, reg_mask);
3674         }
3675         else
3676         {
3677             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3678                     This->input_reg_map[i], reg_mask, reg_mask);
3679         }
3680     }
3681 }
3682
3683 /*********************************************
3684  * Vertex Shader Specific Code begins here
3685  ********************************************/
3686
3687 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3688     glsl_program_key_t key;
3689
3690     key.vshader = entry->vshader;
3691     key.pshader = entry->pshader;
3692     key.vs_args = entry->vs_args;
3693     key.ps_args = entry->ps_args;
3694
3695     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3696     {
3697         ERR("Failed to insert program entry.\n");
3698     }
3699 }
3700
3701 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3702         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3703         struct ps_compile_args *ps_args) {
3704     struct wine_rb_entry *entry;
3705     glsl_program_key_t key;
3706
3707     key.vshader = vshader;
3708     key.pshader = pshader;
3709     key.vs_args = *vs_args;
3710     key.ps_args = *ps_args;
3711
3712     entry = wine_rb_get(&priv->program_lookup, &key);
3713     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3714 }
3715
3716 /* GL locking is done by the caller */
3717 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3718         struct glsl_shader_prog_link *entry)
3719 {
3720     glsl_program_key_t key;
3721
3722     key.vshader = entry->vshader;
3723     key.pshader = entry->pshader;
3724     key.vs_args = entry->vs_args;
3725     key.ps_args = entry->ps_args;
3726     wine_rb_remove(&priv->program_lookup, &key);
3727
3728     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3729     if (entry->vshader) list_remove(&entry->vshader_entry);
3730     if (entry->pshader) list_remove(&entry->pshader_entry);
3731     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3732     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3733     HeapFree(GetProcessHeap(), 0, entry);
3734 }
3735
3736 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3737         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3738         const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3739 {
3740     unsigned int i, j;
3741     const char *semantic_name_in;
3742     UINT semantic_idx_in;
3743     DWORD *set;
3744     DWORD in_idx;
3745     unsigned int in_count = vec4_varyings(3, gl_info);
3746     char reg_mask[6];
3747     char destination[50];
3748     WORD input_map, output_map;
3749
3750     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3751
3752     input_map = reg_maps_in->input_registers;
3753     for (i = 0; input_map; input_map >>= 1, ++i)
3754     {
3755         if (!(input_map & 1)) continue;
3756
3757         in_idx = map[i];
3758         /* Declared, but not read register */
3759         if (in_idx == ~0U) continue;
3760         if (in_idx >= (in_count + 2))
3761         {
3762             FIXME("More input varyings declared than supported, expect issues.\n");
3763             continue;
3764         }
3765
3766         if (in_idx == in_count) {
3767             sprintf(destination, "gl_FrontColor");
3768         } else if (in_idx == in_count + 1) {
3769             sprintf(destination, "gl_FrontSecondaryColor");
3770         } else {
3771             sprintf(destination, "IN[%u]", in_idx);
3772         }
3773
3774         semantic_name_in = input_signature[i].semantic_name;
3775         semantic_idx_in = input_signature[i].semantic_idx;
3776         set[in_idx] = ~0U;
3777
3778         output_map = reg_maps_out->output_registers;
3779         for (j = 0; output_map; output_map >>= 1, ++j)
3780         {
3781             DWORD mask;
3782
3783             if (!(output_map & 1)
3784                     || semantic_idx_in != output_signature[j].semantic_idx
3785                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3786                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3787                 continue;
3788
3789             set[in_idx] = mask;
3790             shader_glsl_write_mask_to_str(mask, reg_mask);
3791
3792             shader_addline(buffer, "%s%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3793                     destination, reg_mask, j, reg_mask);
3794         }
3795     }
3796
3797     for (i = 0; i < in_count + 2; ++i)
3798     {
3799         unsigned int size;
3800
3801         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3802             continue;
3803
3804         if (set[i] == ~0U) set[i] = 0;
3805
3806         size = 0;
3807         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3808         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3809         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3810         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3811         reg_mask[size] = '\0';
3812
3813         if (i == in_count) sprintf(destination, "gl_FrontColor");
3814         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3815         else sprintf(destination, "IN[%u]", i);
3816
3817         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3818         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3819     }
3820
3821     HeapFree(GetProcessHeap(), 0, set);
3822 }
3823
3824 /* GL locking is done by the caller */
3825 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3826         IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3827 {
3828     GLhandleARB ret = 0;
3829     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3830     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3831     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3832     unsigned int i;
3833     const char *semantic_name;
3834     UINT semantic_idx;
3835     char reg_mask[6];
3836     const struct wined3d_shader_signature_element *output_signature = vs->baseShader.output_signature;
3837     WORD map = vs->baseShader.reg_maps.output_registers;
3838
3839     shader_buffer_clear(buffer);
3840
3841     shader_addline(buffer, "#version 120\n");
3842     shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
3843
3844     if (ps_major < 3)
3845     {
3846         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3847
3848         for (i = 0; map; map >>= 1, ++i)
3849         {
3850             DWORD write_mask;
3851
3852             if (!(map & 1)) continue;
3853
3854             semantic_name = output_signature[i].semantic_name;
3855             semantic_idx = output_signature[i].semantic_idx;
3856             write_mask = output_signature[i].mask;
3857             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3858
3859             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3860             {
3861                 if (!semantic_idx)
3862                     shader_addline(buffer, "gl_FrontColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3863                             reg_mask, i, reg_mask);
3864                 else if (semantic_idx == 1)
3865                     shader_addline(buffer, "gl_FrontSecondaryColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3866                             reg_mask, i, reg_mask);
3867             }
3868             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3869             {
3870                 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3871                         reg_mask, i, reg_mask);
3872             }
3873             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3874             {
3875                 if (semantic_idx < 8)
3876                 {
3877                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3878                         write_mask |= WINED3DSP_WRITEMASK_3;
3879
3880                     shader_addline(buffer, "gl_TexCoord[%u]%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3881                             semantic_idx, reg_mask, i, reg_mask);
3882                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3883                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3884                 }
3885             }
3886             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3887             {
3888                 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3889             }
3890             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3891             {
3892                 shader_addline(buffer, "gl_FogFragCoord = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3893             }
3894         }
3895         shader_addline(buffer, "}\n");
3896
3897     }
3898     else
3899     {
3900         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3901         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3902         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3903
3904         /* First, sort out position and point size. Those are not passed to the pixel shader */
3905         for (i = 0; map; map >>= 1, ++i)
3906         {
3907             if (!(map & 1)) continue;
3908
3909             semantic_name = output_signature[i].semantic_name;
3910             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3911
3912             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3913             {
3914                 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3915                         reg_mask, i, reg_mask);
3916             }
3917             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3918             {
3919                 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3920             }
3921         }
3922
3923         /* Then, fix the pixel shader input */
3924         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3925                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3926
3927         shader_addline(buffer, "}\n");
3928     }
3929
3930     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3931     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3932     shader_glsl_compile(gl_info, ret, buffer->buffer);
3933
3934     return ret;
3935 }
3936
3937 /* GL locking is done by the caller */
3938 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3939         GLhandleARB programId, char prefix)
3940 {
3941     const local_constant *lconst;
3942     GLint tmp_loc;
3943     const float *value;
3944     char glsl_name[8];
3945
3946     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3947         value = (const float *)lconst->value;
3948         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3949         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3950         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3951     }
3952     checkGLcall("Hardcoding local constants");
3953 }
3954
3955 /* GL locking is done by the caller */
3956 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3957         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3958         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3959 {
3960     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3961     const struct wined3d_gl_info *gl_info = context->gl_info;
3962     CONST DWORD *function = This->baseShader.function;
3963     struct shader_glsl_ctx_priv priv_ctx;
3964
3965     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3966     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3967
3968     memset(&priv_ctx, 0, sizeof(priv_ctx));
3969     priv_ctx.cur_ps_args = args;
3970     priv_ctx.cur_np2fixup_info = np2fixup_info;
3971
3972     shader_addline(buffer, "#version 120\n");
3973
3974     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3975     {
3976         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3977     }
3978     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3979     {
3980         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3981          * drivers write a warning if we don't do so
3982          */
3983         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3984     }
3985     if (gl_info->supported[EXT_GPU_SHADER4])
3986     {
3987         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3988     }
3989
3990     /* Base Declarations */
3991     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3992
3993     /* Pack 3.0 inputs */
3994     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3995     {
3996         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
3997                 This->baseShader.input_signature, reg_maps, args->vp_mode);
3998     }
3999
4000     /* Base Shader Body */
4001     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4002
4003     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4004     if (reg_maps->shader_version.major < 2)
4005     {
4006         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4007         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4008     }
4009
4010     if (args->srgb_correction)
4011     {
4012         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4013         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4014         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4015         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4016         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4017         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4018     }
4019     /* Pixel shader < 3.0 do not replace the fog stage.
4020      * This implements linear fog computation and blending.
4021      * TODO: non linear fog
4022      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4023      * -1/(e-s) and e/(e-s) respectively.
4024      */
4025     if (reg_maps->shader_version.major < 3)
4026     {
4027         switch(args->fog) {
4028             case FOG_OFF: break;
4029             case FOG_LINEAR:
4030                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4031                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4032                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4033                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4034                 break;
4035             case FOG_EXP:
4036                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4037                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4038                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4039                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4040                 break;
4041             case FOG_EXP2:
4042                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4043                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4044                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4045                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4046                 break;
4047         }
4048     }
4049
4050     shader_addline(buffer, "}\n");
4051
4052     TRACE("Compiling shader object %u\n", shader_obj);
4053     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4054
4055     /* Store the shader object */
4056     return shader_obj;
4057 }
4058
4059 /* GL locking is done by the caller */
4060 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4061         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4062         const struct vs_compile_args *args)
4063 {
4064     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4065     const struct wined3d_gl_info *gl_info = context->gl_info;
4066     CONST DWORD *function = This->baseShader.function;
4067     struct shader_glsl_ctx_priv priv_ctx;
4068
4069     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4070     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4071
4072     shader_addline(buffer, "#version 120\n");
4073
4074     if (gl_info->supported[EXT_GPU_SHADER4])
4075     {
4076         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4077     }
4078
4079     memset(&priv_ctx, 0, sizeof(priv_ctx));
4080     priv_ctx.cur_vs_args = args;
4081
4082     /* Base Declarations */
4083     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4084
4085     /* Base Shader Body */
4086     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4087
4088     /* Unpack outputs */
4089     shader_addline(buffer, "order_ps_input(OUT);\n");
4090
4091     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4092      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4093      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4094      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4095      */
4096     if(args->fog_src == VS_FOG_Z) {
4097         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4098     } else if (!reg_maps->fog) {
4099         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4100     }
4101
4102     /* Write the final position.
4103      *
4104      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4105      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4106      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4107      * contains 1.0 to allow a mad.
4108      */
4109     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4110     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4111     if(args->clip_enabled) {
4112         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4113     }
4114
4115     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4116      *
4117      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4118      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4119      * which is the same as z = z * 2 - w.
4120      */
4121     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4122
4123     shader_addline(buffer, "}\n");
4124
4125     TRACE("Compiling shader object %u\n", shader_obj);
4126     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4127
4128     return shader_obj;
4129 }
4130
4131 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4132         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4133         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4134 {
4135     struct wined3d_state *state = &((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->state;
4136     UINT i;
4137     DWORD new_size;
4138     struct glsl_ps_compiled_shader *new_array;
4139     struct glsl_pshader_private    *shader_data;
4140     struct ps_np2fixup_info        *np2fixup = NULL;
4141     GLhandleARB ret;
4142
4143     if (!shader->baseShader.backend_data)
4144     {
4145         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4146         if (!shader->baseShader.backend_data)
4147         {
4148             ERR("Failed to allocate backend data.\n");
4149             return 0;
4150         }
4151     }
4152     shader_data = shader->baseShader.backend_data;
4153
4154     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4155      * so a linear search is more performant than a hashmap or a binary search
4156      * (cache coherency etc)
4157      */
4158     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4159     {
4160         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4161         {
4162             if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4163             return shader_data->gl_shaders[i].prgId;
4164         }
4165     }
4166
4167     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4168     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4169         if (shader_data->num_gl_shaders)
4170         {
4171             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4172             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4173                                     new_size * sizeof(*shader_data->gl_shaders));
4174         } else {
4175             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4176             new_size = 1;
4177         }
4178
4179         if(!new_array) {
4180             ERR("Out of memory\n");
4181             return 0;
4182         }
4183         shader_data->gl_shaders = new_array;
4184         shader_data->shader_array_size = new_size;
4185     }
4186
4187     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4188
4189     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4190     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4191
4192     pixelshader_update_samplers(&shader->baseShader.reg_maps, (IWineD3DBaseTexture **)state->textures);
4193
4194     shader_buffer_clear(buffer);
4195     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4196     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4197     *np2fixup_info = np2fixup;
4198
4199     return ret;
4200 }
4201
4202 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4203                                  const DWORD use_map) {
4204     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4205     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4206     return stored->fog_src == new->fog_src;
4207 }
4208
4209 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4210         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4211         const struct vs_compile_args *args)
4212 {
4213     UINT i;
4214     DWORD new_size;
4215     struct glsl_vs_compiled_shader *new_array;
4216     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4217     struct glsl_vshader_private *shader_data;
4218     GLhandleARB ret;
4219
4220     if (!shader->baseShader.backend_data)
4221     {
4222         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4223         if (!shader->baseShader.backend_data)
4224         {
4225             ERR("Failed to allocate backend data.\n");
4226             return 0;
4227         }
4228     }
4229     shader_data = shader->baseShader.backend_data;
4230
4231     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4232      * so a linear search is more performant than a hashmap or a binary search
4233      * (cache coherency etc)
4234      */
4235     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4236         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4237             return shader_data->gl_shaders[i].prgId;
4238         }
4239     }
4240
4241     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4242
4243     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4244         if (shader_data->num_gl_shaders)
4245         {
4246             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4247             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4248                                     new_size * sizeof(*shader_data->gl_shaders));
4249         } else {
4250             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4251             new_size = 1;
4252         }
4253
4254         if(!new_array) {
4255             ERR("Out of memory\n");
4256             return 0;
4257         }
4258         shader_data->gl_shaders = new_array;
4259         shader_data->shader_array_size = new_size;
4260     }
4261
4262     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4263
4264     shader_buffer_clear(buffer);
4265     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4266     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4267
4268     return ret;
4269 }
4270
4271 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4272  * It sets the programId on the current StateBlock (because it should be called
4273  * inside of the DrawPrimitive() part of the render loop).
4274  *
4275  * If a program for the given combination does not exist, create one, and store
4276  * the program in the hash table.  If it creates a program, it will link the
4277  * given objects, too.
4278  */
4279
4280 /* GL locking is done by the caller */
4281 static void set_glsl_shader_program(const struct wined3d_context *context,
4282         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4283 {
4284     const struct wined3d_state *state = &device->stateBlock->state;
4285     IWineD3DVertexShader *vshader = use_vs ? (IWineD3DVertexShader *)state->vertex_shader : NULL;
4286     IWineD3DPixelShader *pshader = use_ps ? (IWineD3DPixelShader *)state->pixel_shader : NULL;
4287     const struct wined3d_gl_info *gl_info = context->gl_info;
4288     struct shader_glsl_priv *priv = device->shader_priv;
4289     struct glsl_shader_prog_link *entry    = NULL;
4290     GLhandleARB programId                  = 0;
4291     GLhandleARB reorder_shader_id          = 0;
4292     unsigned int i;
4293     char glsl_name[8];
4294     struct ps_compile_args ps_compile_args;
4295     struct vs_compile_args vs_compile_args;
4296
4297     if (vshader) find_vs_compile_args(state, (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4298     if (pshader) find_ps_compile_args(state, (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args);
4299
4300     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4301     if (entry) {
4302         priv->glsl_program = entry;
4303         return;
4304     }
4305
4306     /* If we get to this point, then no matching program exists, so we create one */
4307     programId = GL_EXTCALL(glCreateProgramObjectARB());
4308     TRACE("Created new GLSL shader program %u\n", programId);
4309
4310     /* Create the entry */
4311     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4312     entry->programId = programId;
4313     entry->vshader = vshader;
4314     entry->pshader = pshader;
4315     entry->vs_args = vs_compile_args;
4316     entry->ps_args = ps_compile_args;
4317     entry->constant_version = 0;
4318     entry->np2Fixup_info = NULL;
4319     /* Add the hash table entry */
4320     add_glsl_program_entry(priv, entry);
4321
4322     /* Set the current program */
4323     priv->glsl_program = entry;
4324
4325     /* Attach GLSL vshader */
4326     if (vshader)
4327     {
4328         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4329                 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4330         WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4331         char tmp_name[10];
4332
4333         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4334         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4335         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4336         checkGLcall("glAttachObjectARB");
4337         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4338          * is destroyed
4339          */
4340         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4341
4342         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4343         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4344         checkGLcall("glAttachObjectARB");
4345
4346         /* Bind vertex attributes to a corresponding index number to match
4347          * the same index numbers as ARB_vertex_programs (makes loading
4348          * vertex attributes simpler).  With this method, we can use the
4349          * exact same code to load the attributes later for both ARB and
4350          * GLSL shaders.
4351          *
4352          * We have to do this here because we need to know the Program ID
4353          * in order to make the bindings work, and it has to be done prior
4354          * to linking the GLSL program. */
4355         for (i = 0; map; map >>= 1, ++i)
4356         {
4357             if (!(map & 1)) continue;
4358
4359             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4360             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4361         }
4362         checkGLcall("glBindAttribLocationARB");
4363
4364         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4365     }
4366
4367     /* Attach GLSL pshader */
4368     if (pshader)
4369     {
4370         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4371                 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4372         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4373         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4374         checkGLcall("glAttachObjectARB");
4375
4376         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4377     }
4378
4379     /* Link the program */
4380     TRACE("Linking GLSL shader program %u\n", programId);
4381     GL_EXTCALL(glLinkProgramARB(programId));
4382     shader_glsl_validate_link(gl_info, programId);
4383
4384     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4385             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4386     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4387     {
4388         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4389         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4390     }
4391     for (i = 0; i < MAX_CONST_I; ++i)
4392     {
4393         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4394         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4395     }
4396     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4397             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4398     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4399     {
4400         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4401         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4402     }
4403     for (i = 0; i < MAX_CONST_I; ++i)
4404     {
4405         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4406         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4407     }
4408
4409     if(pshader) {
4410         char name[32];
4411
4412         for(i = 0; i < MAX_TEXTURES; i++) {
4413             sprintf(name, "bumpenvmat%u", i);
4414             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4415             sprintf(name, "luminancescale%u", i);
4416             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4417             sprintf(name, "luminanceoffset%u", i);
4418             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4419         }
4420
4421         if (ps_compile_args.np2_fixup) {
4422             if (entry->np2Fixup_info) {
4423                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4424             } else {
4425                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4426             }
4427         }
4428     }
4429
4430     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4431     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4432     checkGLcall("Find glsl program uniform locations");
4433
4434     if (pshader
4435             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4436             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4437     {
4438         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4439         entry->vertex_color_clamp = GL_FALSE;
4440     } else {
4441         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4442     }
4443
4444     /* Set the shader to allow uniform loading on it */
4445     GL_EXTCALL(glUseProgramObjectARB(programId));
4446     checkGLcall("glUseProgramObjectARB(programId)");
4447
4448     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4449      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4450      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4451      * vertex shader with fixed function pixel processing is used we make sure that the card
4452      * supports enough samplers to allow the max number of vertex samplers with all possible
4453      * fixed function fragment processing setups. So once the program is linked these samplers
4454      * won't change.
4455      */
4456     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4457     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4458
4459     /* If the local constants do not have to be loaded with the environment constants,
4460      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4461      * later
4462      */
4463     if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4464     {
4465         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4466     }
4467     if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4468     {
4469         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4470     }
4471 }
4472
4473 /* GL locking is done by the caller */
4474 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4475 {
4476     GLhandleARB program_id;
4477     GLhandleARB vshader_id, pshader_id;
4478     const char *blt_pshader;
4479
4480     static const char *blt_vshader =
4481         "#version 120\n"
4482         "void main(void)\n"
4483         "{\n"
4484         "    gl_Position = gl_Vertex;\n"
4485         "    gl_FrontColor = vec4(1.0);\n"
4486         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4487         "}\n";
4488
4489     static const char * const blt_pshaders_full[tex_type_count] =
4490     {
4491         /* tex_1d */
4492         NULL,
4493         /* tex_2d */
4494         "#version 120\n"
4495         "uniform sampler2D sampler;\n"
4496         "void main(void)\n"
4497         "{\n"
4498         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4499         "}\n",
4500         /* tex_3d */
4501         NULL,
4502         /* tex_cube */
4503         "#version 120\n"
4504         "uniform samplerCube sampler;\n"
4505         "void main(void)\n"
4506         "{\n"
4507         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4508         "}\n",
4509         /* tex_rect */
4510         "#version 120\n"
4511         "#extension GL_ARB_texture_rectangle : enable\n"
4512         "uniform sampler2DRect sampler;\n"
4513         "void main(void)\n"
4514         "{\n"
4515         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4516         "}\n",
4517     };
4518
4519     static const char * const blt_pshaders_masked[tex_type_count] =
4520     {
4521         /* tex_1d */
4522         NULL,
4523         /* tex_2d */
4524         "#version 120\n"
4525         "uniform sampler2D sampler;\n"
4526         "uniform vec4 mask;\n"
4527         "void main(void)\n"
4528         "{\n"
4529         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4530         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4531         "}\n",
4532         /* tex_3d */
4533         NULL,
4534         /* tex_cube */
4535         "#version 120\n"
4536         "uniform samplerCube sampler;\n"
4537         "uniform vec4 mask;\n"
4538         "void main(void)\n"
4539         "{\n"
4540         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4541         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4542         "}\n",
4543         /* tex_rect */
4544         "#version 120\n"
4545         "#extension GL_ARB_texture_rectangle : enable\n"
4546         "uniform sampler2DRect sampler;\n"
4547         "uniform vec4 mask;\n"
4548         "void main(void)\n"
4549         "{\n"
4550         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4551         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4552         "}\n",
4553     };
4554
4555     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4556     if (!blt_pshader)
4557     {
4558         FIXME("tex_type %#x not supported\n", tex_type);
4559         tex_type = tex_2d;
4560     }
4561
4562     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4563     shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4564
4565     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4566     shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4567
4568     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4569     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4570     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4571     GL_EXTCALL(glLinkProgramARB(program_id));
4572
4573     shader_glsl_validate_link(gl_info, program_id);
4574
4575     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4576      * is destroyed
4577      */
4578     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4579     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4580     return program_id;
4581 }
4582
4583 /* GL locking is done by the caller */
4584 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4585 {
4586     const struct wined3d_gl_info *gl_info = context->gl_info;
4587     IWineD3DDeviceImpl *device = context->swapchain->device;
4588     struct shader_glsl_priv *priv = device->shader_priv;
4589     GLhandleARB program_id = 0;
4590     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4591
4592     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4593
4594     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4595     else priv->glsl_program = NULL;
4596
4597     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4598
4599     if (old_vertex_color_clamp != current_vertex_color_clamp)
4600     {
4601         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4602         {
4603             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4604             checkGLcall("glClampColorARB");
4605         }
4606         else
4607         {
4608             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4609         }
4610     }
4611
4612     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4613     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4614     GL_EXTCALL(glUseProgramObjectARB(program_id));
4615     checkGLcall("glUseProgramObjectARB");
4616
4617     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4618      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4619      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4620     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4621     {
4622         shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4623     }
4624 }
4625
4626 /* GL locking is done by the caller */
4627 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface,
4628         enum tex_types tex_type, const SIZE *ds_mask_size)
4629 {
4630     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4631     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4632     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4633     struct shader_glsl_priv *priv = This->shader_priv;
4634     GLhandleARB *blt_program;
4635     GLint loc;
4636
4637     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4638     if (!*blt_program)
4639     {
4640         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4641         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4642         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4643         GL_EXTCALL(glUniform1iARB(loc, 0));
4644     }
4645     else
4646     {
4647         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4648     }
4649
4650     if (masked)
4651     {
4652         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4653         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4654     }
4655 }
4656
4657 /* GL locking is done by the caller */
4658 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4659     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4660     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4661     struct shader_glsl_priv *priv = This->shader_priv;
4662     GLhandleARB program_id;
4663
4664     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4665     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4666
4667     GL_EXTCALL(glUseProgramObjectARB(program_id));
4668     checkGLcall("glUseProgramObjectARB");
4669 }
4670
4671 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4672     const struct list *linked_programs;
4673     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4674     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4675     struct shader_glsl_priv *priv = device->shader_priv;
4676     const struct wined3d_gl_info *gl_info;
4677     struct wined3d_context *context;
4678
4679     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4680      * can be called from IWineD3DBaseShader::Release
4681      */
4682     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4683
4684     if(pshader) {
4685         struct glsl_pshader_private *shader_data;
4686         shader_data = This->baseShader.backend_data;
4687         if (!shader_data || !shader_data->num_gl_shaders)
4688         {
4689             HeapFree(GetProcessHeap(), 0, shader_data);
4690             This->baseShader.backend_data = NULL;
4691             return;
4692         }
4693
4694         context = context_acquire(device, NULL);
4695         gl_info = context->gl_info;
4696
4697         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4698         {
4699             ENTER_GL();
4700             shader_glsl_select(context, FALSE, FALSE);
4701             LEAVE_GL();
4702         }
4703     } else {
4704         struct glsl_vshader_private *shader_data;
4705         shader_data = This->baseShader.backend_data;
4706         if (!shader_data || !shader_data->num_gl_shaders)
4707         {
4708             HeapFree(GetProcessHeap(), 0, shader_data);
4709             This->baseShader.backend_data = NULL;
4710             return;
4711         }
4712
4713         context = context_acquire(device, NULL);
4714         gl_info = context->gl_info;
4715
4716         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4717         {
4718             ENTER_GL();
4719             shader_glsl_select(context, FALSE, FALSE);
4720             LEAVE_GL();
4721         }
4722     }
4723
4724     linked_programs = &This->baseShader.linked_programs;
4725
4726     TRACE("Deleting linked programs\n");
4727     if (linked_programs->next) {
4728         struct glsl_shader_prog_link *entry, *entry2;
4729
4730         ENTER_GL();
4731         if(pshader) {
4732             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4733                 delete_glsl_program_entry(priv, gl_info, entry);
4734             }
4735         } else {
4736             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4737                 delete_glsl_program_entry(priv, gl_info, entry);
4738             }
4739         }
4740         LEAVE_GL();
4741     }
4742
4743     if(pshader) {
4744         UINT i;
4745         struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
4746
4747         ENTER_GL();
4748         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4749             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4750             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4751             checkGLcall("glDeleteObjectARB");
4752         }
4753         LEAVE_GL();
4754         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4755     }
4756     else
4757     {
4758         UINT i;
4759         struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
4760
4761         ENTER_GL();
4762         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4763             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4764             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4765             checkGLcall("glDeleteObjectARB");
4766         }
4767         LEAVE_GL();
4768         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4769     }
4770
4771     HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
4772     This->baseShader.backend_data = NULL;
4773
4774     context_release(context);
4775 }
4776
4777 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4778 {
4779     const glsl_program_key_t *k = key;
4780     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4781             const struct glsl_shader_prog_link, program_lookup_entry);
4782     int cmp;
4783
4784     if (k->vshader > prog->vshader) return 1;
4785     else if (k->vshader < prog->vshader) return -1;
4786
4787     if (k->pshader > prog->pshader) return 1;
4788     else if (k->pshader < prog->pshader) return -1;
4789
4790     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4791     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4792
4793     return 0;
4794 }
4795
4796 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4797 {
4798     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4799     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4800
4801     if (!mem)
4802     {
4803         ERR("Failed to allocate memory\n");
4804         return FALSE;
4805     }
4806
4807     heap->entries = mem;
4808     heap->entries[1].version = 0;
4809     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4810     heap->size = 1;
4811
4812     return TRUE;
4813 }
4814
4815 static void constant_heap_free(struct constant_heap *heap)
4816 {
4817     HeapFree(GetProcessHeap(), 0, heap->entries);
4818 }
4819
4820 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4821 {
4822     wined3d_rb_alloc,
4823     wined3d_rb_realloc,
4824     wined3d_rb_free,
4825     glsl_program_key_compare,
4826 };
4827
4828 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4829     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4830     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4831     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4832     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4833             gl_info->limits.glsl_ps_float_constants)) + 1;
4834
4835     if (!shader_buffer_init(&priv->shader_buffer))
4836     {
4837         ERR("Failed to initialize shader buffer.\n");
4838         goto fail;
4839     }
4840
4841     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4842     if (!priv->stack)
4843     {
4844         ERR("Failed to allocate memory.\n");
4845         goto fail;
4846     }
4847
4848     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4849     {
4850         ERR("Failed to initialize vertex shader constant heap\n");
4851         goto fail;
4852     }
4853
4854     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4855     {
4856         ERR("Failed to initialize pixel shader constant heap\n");
4857         goto fail;
4858     }
4859
4860     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4861     {
4862         ERR("Failed to initialize rbtree.\n");
4863         goto fail;
4864     }
4865
4866     priv->next_constant_version = 1;
4867
4868     This->shader_priv = priv;
4869     return WINED3D_OK;
4870
4871 fail:
4872     constant_heap_free(&priv->pconst_heap);
4873     constant_heap_free(&priv->vconst_heap);
4874     HeapFree(GetProcessHeap(), 0, priv->stack);
4875     shader_buffer_free(&priv->shader_buffer);
4876     HeapFree(GetProcessHeap(), 0, priv);
4877     return E_OUTOFMEMORY;
4878 }
4879
4880 /* Context activation is done by the caller. */
4881 static void shader_glsl_free(IWineD3DDevice *iface) {
4882     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4883     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4884     struct shader_glsl_priv *priv = This->shader_priv;
4885     int i;
4886
4887     ENTER_GL();
4888     for (i = 0; i < tex_type_count; ++i)
4889     {
4890         if (priv->depth_blt_program_full[i])
4891         {
4892             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4893         }
4894         if (priv->depth_blt_program_masked[i])
4895         {
4896             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4897         }
4898     }
4899     LEAVE_GL();
4900
4901     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4902     constant_heap_free(&priv->pconst_heap);
4903     constant_heap_free(&priv->vconst_heap);
4904     HeapFree(GetProcessHeap(), 0, priv->stack);
4905     shader_buffer_free(&priv->shader_buffer);
4906
4907     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4908     This->shader_priv = NULL;
4909 }
4910
4911 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4912     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4913     return FALSE;
4914 }
4915
4916 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4917 {
4918     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4919      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4920      * on the version of NV_vertex_program.
4921      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4922      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4923      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4924      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4925      */
4926     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4927             || gl_info->limits.arb_ps_instructions <= 512)
4928         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4929     else
4930         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4931     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4932     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4933
4934     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4935      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4936      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4937      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4938      * in max native instructions. Intel and others also offer the info in this extension but they
4939      * don't support GLSL (at least on Windows).
4940      *
4941      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4942      * of instructions is 512 or less we have to do with ps2.0 hardware.
4943      * 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.
4944      */
4945     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4946             || gl_info->limits.arb_ps_instructions <= 512)
4947         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4948     else
4949         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4950
4951     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4952
4953     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4954      * Direct3D minimum requirement.
4955      *
4956      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4957      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4958      *
4959      * The problem is that the refrast clamps temporary results in the shader to
4960      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4961      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4962      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4963      * offer a way to query this.
4964      */
4965     pCaps->PixelShader1xMaxValue = 8.0;
4966     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4967
4968     pCaps->VSClipping = TRUE;
4969 }
4970
4971 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4972 {
4973     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4974     {
4975         TRACE("Checking support for fixup:\n");
4976         dump_color_fixup_desc(fixup);
4977     }
4978
4979     /* We support everything except YUV conversions. */
4980     if (!is_complex_fixup(fixup))
4981     {
4982         TRACE("[OK]\n");
4983         return TRUE;
4984     }
4985
4986     TRACE("[FAILED]\n");
4987     return FALSE;
4988 }
4989
4990 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4991 {
4992     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4993     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4994     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4995     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4996     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4997     /* WINED3DSIH_BREAKP        */ NULL,
4998     /* WINED3DSIH_CALL          */ shader_glsl_call,
4999     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
5000     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
5001     /* WINED3DSIH_CND           */ shader_glsl_cnd,
5002     /* WINED3DSIH_CRS           */ shader_glsl_cross,
5003     /* WINED3DSIH_CUT           */ NULL,
5004     /* WINED3DSIH_DCL           */ NULL,
5005     /* WINED3DSIH_DEF           */ NULL,
5006     /* WINED3DSIH_DEFB          */ NULL,
5007     /* WINED3DSIH_DEFI          */ NULL,
5008     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
5009     /* WINED3DSIH_DP3           */ shader_glsl_dot,
5010     /* WINED3DSIH_DP4           */ shader_glsl_dot,
5011     /* WINED3DSIH_DST           */ shader_glsl_dst,
5012     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5013     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5014     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5015     /* WINED3DSIH_EMIT          */ NULL,
5016     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5017     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5018     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5019     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5020     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5021     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5022     /* WINED3DSIH_IADD          */ NULL,
5023     /* WINED3DSIH_IF            */ shader_glsl_if,
5024     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5025     /* WINED3DSIH_IGE           */ NULL,
5026     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5027     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5028     /* WINED3DSIH_LOG           */ shader_glsl_log,
5029     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5030     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5031     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5032     /* WINED3DSIH_LT            */ NULL,
5033     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5034     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5035     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5036     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5037     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5038     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5039     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5040     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5041     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5042     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5043     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5044     /* WINED3DSIH_NOP           */ NULL,
5045     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5046     /* WINED3DSIH_PHASE         */ NULL,
5047     /* WINED3DSIH_POW           */ shader_glsl_pow,
5048     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5049     /* WINED3DSIH_REP           */ shader_glsl_rep,
5050     /* WINED3DSIH_RET           */ shader_glsl_ret,
5051     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5052     /* WINED3DSIH_SETP          */ NULL,
5053     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5054     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5055     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5056     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5057     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5058     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5059     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5060     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5061     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5062     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5063     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5064     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5065     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5066     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5067     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5068     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5069     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5070     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5071     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5072     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5073     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5074     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5075     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5076     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5077     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5078     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5079     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5080 };
5081
5082 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5083     SHADER_HANDLER hw_fct;
5084
5085     /* Select handler */
5086     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5087
5088     /* Unhandled opcode */
5089     if (!hw_fct)
5090     {
5091         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5092         return;
5093     }
5094     hw_fct(ins);
5095
5096     shader_glsl_add_instruction_modifiers(ins);
5097 }
5098
5099 const shader_backend_t glsl_shader_backend = {
5100     shader_glsl_handle_instruction,
5101     shader_glsl_select,
5102     shader_glsl_select_depth_blt,
5103     shader_glsl_deselect_depth_blt,
5104     shader_glsl_update_float_vertex_constants,
5105     shader_glsl_update_float_pixel_constants,
5106     shader_glsl_load_constants,
5107     shader_glsl_load_np2fixup_constants,
5108     shader_glsl_destroy,
5109     shader_glsl_alloc,
5110     shader_glsl_free,
5111     shader_glsl_dirty_const,
5112     shader_glsl_get_caps,
5113     shader_glsl_color_fixup_supported,
5114 };