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 2009-2010 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     IWineD3DVertexShaderImpl    *vshader;
 113     IWineD3DPixelShaderImpl     *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     IWineD3DVertexShaderImpl *vshader;
 122     IWineD3DPixelShaderImpl *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(void *shader_priv,
 685         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 686 {
 687     struct shader_glsl_priv *glsl_priv = shader_priv;
 688     const struct glsl_shader_prog_link *prog = glsl_priv->glsl_program;
 689
 690     /* No GLSL program set - nothing to do. */
 691     if (!prog) return;
 692
 693     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 694     if (!use_ps(state)) return;
 695
 696     if (prog->ps_args.np2_fixup && prog->np2Fixup_location != -1)
 697     {
 698         UINT i;
 699         UINT fixup = prog->ps_args.np2_fixup;
 700         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 701
 702         for (i = 0; fixup; fixup >>= 1, ++i)
 703         {
 704             const struct wined3d_texture *tex = state->textures[i];
 705             const unsigned char idx = prog->np2Fixup_info->idx[i];
 706             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 707
 708             if (!tex)
 709             {
 710                 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
 711                 continue;
 712             }
 713
 714             if (idx % 2) {
 715                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 716             } else {
 717                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 718             }
 719         }
 720
 721         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 722     }
 723 }
 724
 725 /**
 726  * Loads the app-supplied constants into the currently set GLSL program.
 727  */
 728 /* GL locking is done by the caller (state handler) */
 729 static void shader_glsl_load_constants(const struct wined3d_context *context,
 730         char usePixelShader, char useVertexShader)
 731 {
 732     const struct wined3d_gl_info *gl_info = context->gl_info;
 733     IWineD3DDeviceImpl *device = context->swapchain->device;
 734     struct wined3d_stateblock *stateBlock = device->stateBlock;
 735     struct shader_glsl_priv *priv = device->shader_priv;
 736     float position_fixup[4];
 737
 738     GLhandleARB programId;
 739     struct glsl_shader_prog_link *prog = priv->glsl_program;
 740     UINT constant_version;
 741     int i;
 742
 743     if (!prog) {
 744         /* No GLSL program set - nothing to do. */
 745         return;
 746     }
 747     programId = prog->programId;
 748     constant_version = prog->constant_version;
 749
 750     if (useVertexShader)
 751     {
 752         IWineD3DBaseShaderImpl *vshader = (IWineD3DBaseShaderImpl *)stateBlock->state.vertex_shader;
 753
 754         /* Load DirectX 9 float constants/uniforms for vertex shader */
 755         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->state.vs_consts_f,
 756                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 757
 758         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 759         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->state.vs_consts_i,
 760                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 761
 762         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 763         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->state.vs_consts_b,
 764                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 765
 766         /* Upload the position fixup params */
 767         shader_get_position_fixup(context, &stateBlock->state, position_fixup);
 768         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, position_fixup));
 769         checkGLcall("glUniform4fvARB");
 770     }
 771
 772     if (usePixelShader)
 773     {
 774         IWineD3DBaseShaderImpl *pshader = (IWineD3DBaseShaderImpl *)stateBlock->state.pixel_shader;
 775
 776         /* Load DirectX 9 float constants/uniforms for pixel shader */
 777         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->state.ps_consts_f,
 778                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 779
 780         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 781         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->state.ps_consts_i,
 782                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 783
 784         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 785         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->state.ps_consts_b,
 786                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 787
 788         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 789          * It can't be 0 for a valid texbem instruction.
 790          */
 791         for(i = 0; i < MAX_TEXTURES; i++) {
 792             const float *data;
 793
 794             if(prog->bumpenvmat_location[i] == -1) continue;
 795
 796             data = (const float *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVMAT00];
 797             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 798             checkGLcall("glUniformMatrix2fvARB");
 799
 800             /* texbeml needs the luminance scale and offset too. If texbeml
 801              * is used, needsbumpmat is set too, so we can check that in the
 802              * needsbumpmat check. */
 803             if (prog->luminancescale_location[i] != -1)
 804             {
 805                 const GLfloat *scale = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLSCALE];
 806                 const GLfloat *offset = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
 807
 808                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 809                 checkGLcall("glUniform1fvARB");
 810                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 811                 checkGLcall("glUniform1fvARB");
 812             }
 813         }
 814
 815         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 816             float correction_params[4];
 817
 818             if (context->render_offscreen)
 819             {
 820                 correction_params[0] = 0.0f;
 821                 correction_params[1] = 1.0f;
 822             } else {
 823                 /* position is window relative, not viewport relative */
 824                 correction_params[0] = context->current_rt->resource.height;
 825                 correction_params[1] = -1.0f;
 826             }
 827             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 828         }
 829     }
 830
 831     if (priv->next_constant_version == UINT_MAX)
 832     {
 833         TRACE("Max constant version reached, resetting to 0.\n");
 834         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 835         priv->next_constant_version = 1;
 836     }
 837     else
 838     {
 839         prog->constant_version = priv->next_constant_version++;
 840     }
 841 }
 842
 843 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 844         unsigned int heap_idx, DWORD new_version)
 845 {
 846     struct constant_entry *entries = heap->entries;
 847     unsigned int *positions = heap->positions;
 848     unsigned int parent_idx;
 849
 850     while (heap_idx > 1)
 851     {
 852         parent_idx = heap_idx >> 1;
 853
 854         if (new_version <= entries[parent_idx].version) break;
 855
 856         entries[heap_idx] = entries[parent_idx];
 857         positions[entries[parent_idx].idx] = heap_idx;
 858         heap_idx = parent_idx;
 859     }
 860
 861     entries[heap_idx].version = new_version;
 862     entries[heap_idx].idx = idx;
 863     positions[idx] = heap_idx;
 864 }
 865
 866 static void shader_glsl_update_float_vertex_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
 867 {
 868     struct shader_glsl_priv *priv = device->shader_priv;
 869     struct constant_heap *heap = &priv->vconst_heap;
 870     UINT i;
 871
 872     for (i = start; i < count + start; ++i)
 873     {
 874         if (!device->stateBlock->changed.vertexShaderConstantsF[i])
 875             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 876         else
 877             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 878     }
 879 }
 880
 881 static void shader_glsl_update_float_pixel_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
 882 {
 883     struct shader_glsl_priv *priv = device->shader_priv;
 884     struct constant_heap *heap = &priv->pconst_heap;
 885     UINT i;
 886
 887     for (i = start; i < count + start; ++i)
 888     {
 889         if (!device->stateBlock->changed.pixelShaderConstantsF[i])
 890             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 891         else
 892             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 893     }
 894 }
 895
 896 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 897 {
 898     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 899     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 900     if(shader_major > 3) return ret;
 901
 902     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 903     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 904     return ret;
 905 }
 906
 907 /** Generate the variable & register declarations for the GLSL output target */
 908 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 909         struct wined3d_shader_buffer *buffer, IWineD3DBaseShaderImpl *shader,
 910         const struct wined3d_shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
 911 {
 912     IWineD3DDeviceImpl *device = shader->baseShader.device;
 913     const struct wined3d_state *state = &device->stateBlock->state;
 914     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 915     const struct wined3d_gl_info *gl_info = context->gl_info;
 916     unsigned int i, extra_constants_needed = 0;
 917     const local_constant *lconst;
 918     DWORD map;
 919
 920     /* There are some minor differences between pixel and vertex shaders */
 921     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 922     char prefix = pshader ? 'P' : 'V';
 923
 924     /* Prototype the subroutines */
 925     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 926     {
 927         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 928     }
 929
 930     /* Declare the constants (aka uniforms) */
 931     if (shader->baseShader.limits.constant_float > 0)
 932     {
 933         unsigned max_constantsF;
 934         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 935          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 936          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 937          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 938          * a dx9 card, as long as it doesn't also use all the other constants.
 939          *
 940          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 941          * declare only the amount that we're assured to have.
 942          *
 943          * Thus we run into problems in these two cases:
 944          * 1) The shader really uses more uniforms than supported
 945          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 946          */
 947         if (pshader)
 948         {
 949             /* No indirect addressing here. */
 950             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 951         }
 952         else
 953         {
 954             if (reg_maps->usesrelconstF)
 955             {
 956                 /* Subtract the other potential uniforms from the max
 957                  * available (bools, ints, and 1 row of projection matrix).
 958                  * Subtract another uniform for immediate values, which have
 959                  * to be loaded via uniform by the driver as well. The shader
 960                  * code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex
 961                  * shader code, so one vec4 should be enough. (Unfortunately
 962                  * the Nvidia driver doesn't store 128 and -128 in one float).
 963                  *
 964                  * Writing gl_ClipVertex requires one uniform for each
 965                  * clipplane as well. */
 966                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 967                 if(ctx_priv->cur_vs_args->clip_enabled)
 968                 {
 969                     max_constantsF -= gl_info->limits.clipplanes;
 970                 }
 971                 max_constantsF -= count_bits(reg_maps->integer_constants);
 972                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 973                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 974                  * for now take this into account when calculating the number of available constants
 975                  */
 976                 max_constantsF -= count_bits(reg_maps->boolean_constants);
 977                 /* Set by driver quirks in directx.c */
 978                 max_constantsF -= gl_info->reserved_glsl_constants;
 979             }
 980             else
 981             {
 982                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 983             }
 984         }
 985         max_constantsF = min(shader->baseShader.limits.constant_float, max_constantsF);
 986         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 987     }
 988
 989     /* Always declare the full set of constants, the compiler can remove the
 990      * unused ones because d3d doesn't (yet) support indirect int and bool
 991      * constant addressing. This avoids problems if the app uses e.g. i0 and i9. */
 992     if (shader->baseShader.limits.constant_int > 0 && reg_maps->integer_constants)
 993         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, shader->baseShader.limits.constant_int);
 994
 995     if (shader->baseShader.limits.constant_bool > 0 && reg_maps->boolean_constants)
 996         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, shader->baseShader.limits.constant_bool);
 997
 998     if (!pshader)
 999     {
1000         shader_addline(buffer, "uniform vec4 posFixup;\n");
1001         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
1002     }
1003     else
1004     {
1005         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1006         {
1007             if (!(map & 1)) continue;
1008
1009             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1010
1011             if (reg_maps->luminanceparams & (1 << i))
1012             {
1013                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1014                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1015                 extra_constants_needed++;
1016             }
1017
1018             extra_constants_needed++;
1019         }
1020
1021         if (ps_args->srgb_correction)
1022         {
1023             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1024                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1025             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1026                     srgb_cmp);
1027         }
1028         if (reg_maps->vpos || reg_maps->usesdsy)
1029         {
1030             if (shader->baseShader.limits.constant_float + extra_constants_needed
1031                     + 1 < gl_info->limits.glsl_ps_float_constants)
1032             {
1033                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1034                 ((IWineD3DPixelShaderImpl *)shader)->vpos_uniform = 1;
1035                 extra_constants_needed++;
1036             } else {
1037                 /* This happens because we do not have proper tracking of the constant registers that are
1038                  * actually used, only the max limit of the shader version
1039                  */
1040                 FIXME("Cannot find a free uniform for vpos correction params\n");
1041                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1042                         context->render_offscreen ? 0.0f : device->render_targets[0]->resource.height,
1043                         context->render_offscreen ? 1.0f : -1.0f);
1044             }
1045             shader_addline(buffer, "vec4 vpos;\n");
1046         }
1047     }
1048
1049     /* Declare texture samplers */
1050     for (i = 0; i < shader->baseShader.limits.sampler; ++i)
1051     {
1052         if (reg_maps->sampler_type[i])
1053         {
1054             const struct wined3d_texture *texture;
1055
1056             switch (reg_maps->sampler_type[i])
1057             {
1058                 case WINED3DSTT_1D:
1059                     if (pshader && ps_args->shadow & (1 << i))
1060                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1061                     else
1062                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1063                     break;
1064                 case WINED3DSTT_2D:
1065                     texture = state->textures[i];
1066                     if (pshader && ps_args->shadow & (1 << i))
1067                     {
1068                         if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1069                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1070                         else
1071                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1072                     }
1073                     else
1074                     {
1075                         if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1076                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1077                         else
1078                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1079                     }
1080                     break;
1081                 case WINED3DSTT_CUBE:
1082                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1083                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1084                     break;
1085                 case WINED3DSTT_VOLUME:
1086                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1087                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1088                     break;
1089                 default:
1090                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1091                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1092                     break;
1093             }
1094         }
1095     }
1096
1097     /* Declare uniforms for NP2 texcoord fixup:
1098      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1099      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1100      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1101     if (pshader && ps_args->np2_fixup) {
1102
1103         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1104         UINT cur = 0;
1105
1106         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1107          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1108          * samplerNP2Fixup stores texture dimensions and is updated through
1109          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1110
1111         for (i = 0; i < shader->baseShader.limits.sampler; ++i)
1112         {
1113             if (reg_maps->sampler_type[i])
1114             {
1115                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1116
1117                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1118                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1119                     continue;
1120                 }
1121
1122                 fixup->idx[i] = cur++;
1123             }
1124         }
1125
1126         fixup->num_consts = (cur + 1) >> 1;
1127         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1128     }
1129
1130     /* Declare address variables */
1131     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1132     {
1133         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1134     }
1135
1136     /* Declare texture coordinate temporaries and initialize them */
1137     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1138     {
1139         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1140     }
1141
1142     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1143      * helper function shader that is linked in at link time
1144      */
1145     if (pshader && reg_maps->shader_version.major >= 3)
1146     {
1147         if (use_vs(state))
1148         {
1149             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1150         } else {
1151             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1152              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1153              * pixel shader that reads the fixed function color into the packed input registers.
1154              */
1155             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1156         }
1157     }
1158
1159     /* Declare output register temporaries */
1160     if (shader->baseShader.limits.packed_output)
1161         shader_addline(buffer, "vec4 OUT[%u];\n", shader->baseShader.limits.packed_output);
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 (!shader->baseShader.load_local_constsF)
1193     {
1194         LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry)
1195         {
1196             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1197         }
1198     }
1199
1200     /* Start the main program */
1201     shader_addline(buffer, "void main() {\n");
1202     if(pshader && reg_maps->vpos) {
1203         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1204          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1205          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1206          * precision troubles when we just substract 0.5.
1207          *
1208          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1209          *
1210          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1211          *
1212          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1213          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1214          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1215          * correctly on drivers that returns integer values.
1216          */
1217         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1218     }
1219 }
1220
1221 /*****************************************************************************
1222  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1223  *
1224  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1225  ****************************************************************************/
1226
1227 /* Prototypes */
1228 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1229         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1230
1231 /** Used for opcode modifiers - They multiply the result by the specified amount */
1232 static const char * const shift_glsl_tab[] = {
1233     "",           /*  0 (none) */
1234     "2.0 * ",     /*  1 (x2)   */
1235     "4.0 * ",     /*  2 (x4)   */
1236     "8.0 * ",     /*  3 (x8)   */
1237     "16.0 * ",    /*  4 (x16)  */
1238     "32.0 * ",    /*  5 (x32)  */
1239     "",           /*  6 (x64)  */
1240     "",           /*  7 (x128) */
1241     "",           /*  8 (d256) */
1242     "",           /*  9 (d128) */
1243     "",           /* 10 (d64)  */
1244     "",           /* 11 (d32)  */
1245     "0.0625 * ",  /* 12 (d16)  */
1246     "0.125 * ",   /* 13 (d8)   */
1247     "0.25 * ",    /* 14 (d4)   */
1248     "0.5 * "      /* 15 (d2)   */
1249 };
1250
1251 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1252 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1253 {
1254     out_str[0] = 0;
1255
1256     switch (src_modifier)
1257     {
1258     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1259     case WINED3DSPSM_DW:
1260     case WINED3DSPSM_NONE:
1261         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1262         break;
1263     case WINED3DSPSM_NEG:
1264         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1265         break;
1266     case WINED3DSPSM_NOT:
1267         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1268         break;
1269     case WINED3DSPSM_BIAS:
1270         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1271         break;
1272     case WINED3DSPSM_BIASNEG:
1273         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1274         break;
1275     case WINED3DSPSM_SIGN:
1276         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1277         break;
1278     case WINED3DSPSM_SIGNNEG:
1279         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1280         break;
1281     case WINED3DSPSM_COMP:
1282         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1283         break;
1284     case WINED3DSPSM_X2:
1285         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1286         break;
1287     case WINED3DSPSM_X2NEG:
1288         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1289         break;
1290     case WINED3DSPSM_ABS:
1291         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1292         break;
1293     case WINED3DSPSM_ABSNEG:
1294         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1295         break;
1296     default:
1297         FIXME("Unhandled modifier %u\n", src_modifier);
1298         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1299     }
1300 }
1301
1302 /** Writes the GLSL variable name that corresponds to the register that the
1303  * DX opcode parameter is trying to access */
1304 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1305         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1306 {
1307     /* oPos, oFog and oPts in D3D */
1308     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1309
1310     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
1311     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1312     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1313     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1314
1315     *is_color = FALSE;
1316
1317     switch (reg->type)
1318     {
1319         case WINED3DSPR_TEMP:
1320             sprintf(register_name, "R%u", reg->idx);
1321             break;
1322
1323         case WINED3DSPR_INPUT:
1324             /* vertex shaders */
1325             if (!pshader)
1326             {
1327                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1328                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1329                 sprintf(register_name, "attrib%u", reg->idx);
1330                 break;
1331             }
1332
1333             /* pixel shaders >= 3.0 */
1334             if (reg_maps->shader_version.major >= 3)
1335             {
1336                 DWORD idx = ((IWineD3DPixelShaderImpl *)shader)->input_reg_map[reg->idx];
1337                 unsigned int in_count = vec4_varyings(reg_maps->shader_version.major, gl_info);
1338
1339                 if (reg->rel_addr)
1340                 {
1341                     glsl_src_param_t rel_param;
1342
1343                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1344
1345                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1346                      * operation there */
1347                     if (idx)
1348                     {
1349                         if (((IWineD3DPixelShaderImpl *)shader)->declared_in_count > in_count)
1350                         {
1351                             sprintf(register_name,
1352                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1353                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1354                                     rel_param.param_str, idx);
1355                         }
1356                         else
1357                         {
1358                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1359                         }
1360                     }
1361                     else
1362                     {
1363                         if (((IWineD3DPixelShaderImpl *)shader)->declared_in_count > in_count)
1364                         {
1365                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1366                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1367                                     rel_param.param_str);
1368                         }
1369                         else
1370                         {
1371                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1372                         }
1373                     }
1374                 }
1375                 else
1376                 {
1377                     if (idx == in_count) sprintf(register_name, "gl_Color");
1378                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1379                     else sprintf(register_name, "IN[%u]", idx);
1380                 }
1381             }
1382             else
1383             {
1384                 if (!reg->idx) strcpy(register_name, "gl_Color");
1385                 else strcpy(register_name, "gl_SecondaryColor");
1386                 break;
1387             }
1388             break;
1389
1390         case WINED3DSPR_CONST:
1391             {
1392                 const char prefix = pshader ? 'P' : 'V';
1393
1394                 /* Relative addressing */
1395                 if (reg->rel_addr)
1396                 {
1397                     glsl_src_param_t rel_param;
1398                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1399                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1400                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1401                 }
1402                 else
1403                 {
1404                     if (shader_constant_is_local(shader, reg->idx))
1405                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1406                     else
1407                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1408                 }
1409             }
1410             break;
1411
1412         case WINED3DSPR_CONSTINT:
1413             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1414             else sprintf(register_name, "VI[%u]", reg->idx);
1415             break;
1416
1417         case WINED3DSPR_CONSTBOOL:
1418             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1419             else sprintf(register_name, "VB[%u]", reg->idx);
1420             break;
1421
1422         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1423             if (pshader) sprintf(register_name, "T%u", reg->idx);
1424             else sprintf(register_name, "A%u", reg->idx);
1425             break;
1426
1427         case WINED3DSPR_LOOP:
1428             sprintf(register_name, "aL%u", ins->ctx->loop_state->current_reg - 1);
1429             break;
1430
1431         case WINED3DSPR_SAMPLER:
1432             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1433             else sprintf(register_name, "Vsampler%u", reg->idx);
1434             break;
1435
1436         case WINED3DSPR_COLOROUT:
1437             if (reg->idx >= gl_info->limits.buffers)
1438                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1439
1440             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1441             break;
1442
1443         case WINED3DSPR_RASTOUT:
1444             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1445             break;
1446
1447         case WINED3DSPR_DEPTHOUT:
1448             sprintf(register_name, "gl_FragDepth");
1449             break;
1450
1451         case WINED3DSPR_ATTROUT:
1452             if (!reg->idx) sprintf(register_name, "OUT[8]");
1453             else sprintf(register_name, "OUT[9]");
1454             break;
1455
1456         case WINED3DSPR_TEXCRDOUT:
1457             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1458             sprintf(register_name, "OUT[%u]", reg->idx);
1459             break;
1460
1461         case WINED3DSPR_MISCTYPE:
1462             if (!reg->idx)
1463             {
1464                 /* vPos */
1465                 sprintf(register_name, "vpos");
1466             }
1467             else if (reg->idx == 1)
1468             {
1469                 /* Note that gl_FrontFacing is a bool, while vFace is
1470                  * a float for which the sign determines front/back */
1471                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1472             }
1473             else
1474             {
1475                 FIXME("Unhandled misctype register %d\n", reg->idx);
1476                 sprintf(register_name, "unrecognized_register");
1477             }
1478             break;
1479
1480         case WINED3DSPR_IMMCONST:
1481             switch (reg->immconst_type)
1482             {
1483                 case WINED3D_IMMCONST_SCALAR:
1484                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1485                     break;
1486
1487                 case WINED3D_IMMCONST_VEC4:
1488                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1489                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1490                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1491                     break;
1492
1493                 default:
1494                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1495                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1496             }
1497             break;
1498
1499         default:
1500             FIXME("Unhandled register name Type(%d)\n", reg->type);
1501             sprintf(register_name, "unrecognized_register");
1502             break;
1503     }
1504 }
1505
1506 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1507 {
1508     *str++ = '.';
1509     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1510     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1511     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1512     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1513     *str = '\0';
1514 }
1515
1516 /* Get the GLSL write mask for the destination register */
1517 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1518 {
1519     DWORD mask = param->write_mask;
1520
1521     if (shader_is_scalar(&param->reg))
1522     {
1523         mask = WINED3DSP_WRITEMASK_0;
1524         *write_mask = '\0';
1525     }
1526     else
1527     {
1528         shader_glsl_write_mask_to_str(mask, write_mask);
1529     }
1530
1531     return mask;
1532 }
1533
1534 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1535     unsigned int size = 0;
1536
1537     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1538     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1539     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1540     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1541
1542     return size;
1543 }
1544
1545 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1546 {
1547     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1548      * but addressed as "rgba". To fix this we need to swap the register's x
1549      * and z components. */
1550     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1551
1552     *str++ = '.';
1553     /* swizzle bits fields: wwzzyyxx */
1554     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1555     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1556     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1557     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1558     *str = '\0';
1559 }
1560
1561 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1562         BOOL fixup, DWORD mask, char *swizzle_str)
1563 {
1564     if (shader_is_scalar(&param->reg))
1565         *swizzle_str = '\0';
1566     else
1567         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1568 }
1569
1570 /* From a given parameter token, generate the corresponding GLSL string.
1571  * Also, return the actual register name and swizzle in case the
1572  * caller needs this information as well. */
1573 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1574         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1575 {
1576     BOOL is_color = FALSE;
1577     char swizzle_str[6];
1578
1579     glsl_src->reg_name[0] = '\0';
1580     glsl_src->param_str[0] = '\0';
1581     swizzle_str[0] = '\0';
1582
1583     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1584     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1585     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1586 }
1587
1588 /* From a given parameter token, generate the corresponding GLSL string.
1589  * Also, return the actual register name and swizzle in case the
1590  * caller needs this information as well. */
1591 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1592         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1593 {
1594     BOOL is_color = FALSE;
1595
1596     glsl_dst->mask_str[0] = '\0';
1597     glsl_dst->reg_name[0] = '\0';
1598
1599     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1600     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1601 }
1602
1603 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1604 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1605         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1606 {
1607     glsl_dst_param_t glsl_dst;
1608     DWORD mask;
1609
1610     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1611     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1612
1613     return mask;
1614 }
1615
1616 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1617 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1618 {
1619     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1620 }
1621
1622 /** Process GLSL instruction modifiers */
1623 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1624 {
1625     glsl_dst_param_t dst_param;
1626     DWORD modifiers;
1627
1628     if (!ins->dst_count) return;
1629
1630     modifiers = ins->dst[0].modifiers;
1631     if (!modifiers) return;
1632
1633     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1634
1635     if (modifiers & WINED3DSPDM_SATURATE)
1636     {
1637         /* _SAT means to clamp the value of the register to between 0 and 1 */
1638         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1639                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1640     }
1641
1642     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1643     {
1644         FIXME("_centroid modifier not handled\n");
1645     }
1646
1647     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1648     {
1649         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1650     }
1651 }
1652
1653 static inline const char *shader_get_comp_op(DWORD op)
1654 {
1655     switch (op) {
1656         case COMPARISON_GT: return ">";
1657         case COMPARISON_EQ: return "==";
1658         case COMPARISON_GE: return ">=";
1659         case COMPARISON_LT: return "<";
1660         case COMPARISON_NE: return "!=";
1661         case COMPARISON_LE: return "<=";
1662         default:
1663             FIXME("Unrecognized comparison value: %u\n", op);
1664             return "(\?\?)";
1665     }
1666 }
1667
1668 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1669         DWORD sampler_idx, DWORD flags, glsl_sample_function_t *sample_function)
1670 {
1671     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1672     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1673     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1674             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1675     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1676     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1677     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1678     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1679
1680     /* Note that there's no such thing as a projected cube texture. */
1681     switch(sampler_type) {
1682         case WINED3DSTT_1D:
1683             if (shadow)
1684             {
1685                 if (lod)
1686                 {
1687                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1688                 }
1689                 else if (grad)
1690                 {
1691                     if (gl_info->supported[EXT_GPU_SHADER4])
1692                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1693                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1694                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1695                     else
1696                     {
1697                         FIXME("Unsupported 1D shadow grad function.\n");
1698                         sample_function->name = "unsupported1DGrad";
1699                     }
1700                 }
1701                 else
1702                 {
1703                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1704                 }
1705                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1706             }
1707             else
1708             {
1709                 if (lod)
1710                 {
1711                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1712                 }
1713                 else if (grad)
1714                 {
1715                     if (gl_info->supported[EXT_GPU_SHADER4])
1716                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1717                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1718                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1719                     else
1720                     {
1721                         FIXME("Unsupported 1D grad function.\n");
1722                         sample_function->name = "unsupported1DGrad";
1723                     }
1724                 }
1725                 else
1726                 {
1727                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1728                 }
1729                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1730             }
1731             break;
1732
1733         case WINED3DSTT_2D:
1734             if (shadow)
1735             {
1736                 if (texrect)
1737                 {
1738                     if (lod)
1739                     {
1740                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1741                     }
1742                     else if (grad)
1743                     {
1744                         if (gl_info->supported[EXT_GPU_SHADER4])
1745                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1746                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1747                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1748                         else
1749                         {
1750                             FIXME("Unsupported RECT shadow grad function.\n");
1751                             sample_function->name = "unsupported2DRectGrad";
1752                         }
1753                     }
1754                     else
1755                     {
1756                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1757                     }
1758                 }
1759                 else
1760                 {
1761                     if (lod)
1762                     {
1763                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1764                     }
1765                     else if (grad)
1766                     {
1767                         if (gl_info->supported[EXT_GPU_SHADER4])
1768                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1769                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1770                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1771                         else
1772                         {
1773                             FIXME("Unsupported 2D shadow grad function.\n");
1774                             sample_function->name = "unsupported2DGrad";
1775                         }
1776                     }
1777                     else
1778                     {
1779                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1780                     }
1781                 }
1782                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1783             }
1784             else
1785             {
1786                 if (texrect)
1787                 {
1788                     if (lod)
1789                     {
1790                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1791                     }
1792                     else if (grad)
1793                     {
1794                         if (gl_info->supported[EXT_GPU_SHADER4])
1795                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1796                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1797                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1798                         else
1799                         {
1800                             FIXME("Unsupported RECT grad function.\n");
1801                             sample_function->name = "unsupported2DRectGrad";
1802                         }
1803                     }
1804                     else
1805                     {
1806                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1807                     }
1808                 }
1809                 else
1810                 {
1811                     if (lod)
1812                     {
1813                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1814                     }
1815                     else if (grad)
1816                     {
1817                         if (gl_info->supported[EXT_GPU_SHADER4])
1818                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1819                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1820                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1821                         else
1822                         {
1823                             FIXME("Unsupported 2D grad function.\n");
1824                             sample_function->name = "unsupported2DGrad";
1825                         }
1826                     }
1827                     else
1828                     {
1829                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1830                     }
1831                 }
1832                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1833             }
1834             break;
1835
1836         case WINED3DSTT_CUBE:
1837             if (shadow)
1838             {
1839                 FIXME("Unsupported Cube shadow function.\n");
1840                 sample_function->name = "unsupportedCubeShadow";
1841                 sample_function->coord_mask = 0;
1842             }
1843             else
1844             {
1845                 if (lod)
1846                 {
1847                     sample_function->name = "textureCubeLod";
1848                 }
1849                 else if (grad)
1850                 {
1851                     if (gl_info->supported[EXT_GPU_SHADER4])
1852                         sample_function->name = "textureCubeGrad";
1853                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1854                         sample_function->name = "textureCubeGradARB";
1855                     else
1856                     {
1857                         FIXME("Unsupported Cube grad function.\n");
1858                         sample_function->name = "unsupportedCubeGrad";
1859                     }
1860                 }
1861                 else
1862                 {
1863                     sample_function->name = "textureCube";
1864                 }
1865                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1866             }
1867             break;
1868
1869         case WINED3DSTT_VOLUME:
1870             if (shadow)
1871             {
1872                 FIXME("Unsupported 3D shadow function.\n");
1873                 sample_function->name = "unsupported3DShadow";
1874                 sample_function->coord_mask = 0;
1875             }
1876             else
1877             {
1878                 if (lod)
1879                 {
1880                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1881                 }
1882                 else  if (grad)
1883                 {
1884                     if (gl_info->supported[EXT_GPU_SHADER4])
1885                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1886                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1887                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1888                     else
1889                     {
1890                         FIXME("Unsupported 3D grad function.\n");
1891                         sample_function->name = "unsupported3DGrad";
1892                     }
1893                 }
1894                 else
1895                 {
1896                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1897                 }
1898                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1899             }
1900             break;
1901
1902         default:
1903             sample_function->name = "";
1904             sample_function->coord_mask = 0;
1905             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1906             break;
1907     }
1908 }
1909
1910 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1911         BOOL sign_fixup, enum fixup_channel_source channel_source)
1912 {
1913     switch(channel_source)
1914     {
1915         case CHANNEL_SOURCE_ZERO:
1916             strcat(arguments, "0.0");
1917             break;
1918
1919         case CHANNEL_SOURCE_ONE:
1920             strcat(arguments, "1.0");
1921             break;
1922
1923         case CHANNEL_SOURCE_X:
1924             strcat(arguments, reg_name);
1925             strcat(arguments, ".x");
1926             break;
1927
1928         case CHANNEL_SOURCE_Y:
1929             strcat(arguments, reg_name);
1930             strcat(arguments, ".y");
1931             break;
1932
1933         case CHANNEL_SOURCE_Z:
1934             strcat(arguments, reg_name);
1935             strcat(arguments, ".z");
1936             break;
1937
1938         case CHANNEL_SOURCE_W:
1939             strcat(arguments, reg_name);
1940             strcat(arguments, ".w");
1941             break;
1942
1943         default:
1944             FIXME("Unhandled channel source %#x\n", channel_source);
1945             strcat(arguments, "undefined");
1946             break;
1947     }
1948
1949     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1950 }
1951
1952 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1953 {
1954     struct wined3d_shader_dst_param dst;
1955     unsigned int mask_size, remaining;
1956     glsl_dst_param_t dst_param;
1957     char arguments[256];
1958     DWORD mask;
1959
1960     mask = 0;
1961     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1962     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1963     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1964     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1965     mask &= ins->dst[0].write_mask;
1966
1967     if (!mask) return; /* Nothing to do */
1968
1969     if (is_complex_fixup(fixup))
1970     {
1971         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1972         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1973         return;
1974     }
1975
1976     mask_size = shader_glsl_get_write_mask_size(mask);
1977
1978     dst = ins->dst[0];
1979     dst.write_mask = mask;
1980     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1981
1982     arguments[0] = '\0';
1983     remaining = mask_size;
1984     if (mask & WINED3DSP_WRITEMASK_0)
1985     {
1986         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1987         if (--remaining) strcat(arguments, ", ");
1988     }
1989     if (mask & WINED3DSP_WRITEMASK_1)
1990     {
1991         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1992         if (--remaining) strcat(arguments, ", ");
1993     }
1994     if (mask & WINED3DSP_WRITEMASK_2)
1995     {
1996         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1997         if (--remaining) strcat(arguments, ", ");
1998     }
1999     if (mask & WINED3DSP_WRITEMASK_3)
2000     {
2001         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
2002         if (--remaining) strcat(arguments, ", ");
2003     }
2004
2005     if (mask_size > 1)
2006     {
2007         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
2008                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
2009     }
2010     else
2011     {
2012         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2013     }
2014 }
2015
2016 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2017         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
2018         const char *dx, const char *dy,
2019         const char *bias, const char *coord_reg_fmt, ...)
2020 {
2021     const char *sampler_base;
2022     char dst_swizzle[6];
2023     struct color_fixup_desc fixup;
2024     BOOL np2_fixup = FALSE;
2025     va_list args;
2026
2027     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2028
2029     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2030     {
2031         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2032         fixup = priv->cur_ps_args->color_fixup[sampler];
2033         sampler_base = "Psampler";
2034
2035         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2036             if(bias) {
2037                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2038             } else {
2039                 np2_fixup = TRUE;
2040             }
2041         }
2042     } else {
2043         sampler_base = "Vsampler";
2044         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2045     }
2046
2047     shader_glsl_append_dst(ins->ctx->buffer, ins);
2048
2049     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2050
2051     va_start(args, coord_reg_fmt);
2052     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2053     va_end(args);
2054
2055     if(bias) {
2056         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2057     } else {
2058         if (np2_fixup) {
2059             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2060             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2061
2062             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2063                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2064         } else if(dx && dy) {
2065             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2066         } else {
2067             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2068         }
2069     }
2070
2071     if(!is_identity_fixup(fixup)) {
2072         shader_glsl_color_correction(ins, fixup);
2073     }
2074 }
2075
2076 /*****************************************************************************
2077  * Begin processing individual instruction opcodes
2078  ****************************************************************************/
2079
2080 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2081 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2082 {
2083     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2084     glsl_src_param_t src0_param;
2085     glsl_src_param_t src1_param;
2086     DWORD write_mask;
2087     char op;
2088
2089     /* Determine the GLSL operator to use based on the opcode */
2090     switch (ins->handler_idx)
2091     {
2092         case WINED3DSIH_MUL: op = '*'; break;
2093         case WINED3DSIH_ADD: op = '+'; break;
2094         case WINED3DSIH_SUB: op = '-'; break;
2095         default:
2096             op = ' ';
2097             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2098             break;
2099     }
2100
2101     write_mask = shader_glsl_append_dst(buffer, ins);
2102     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2103     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2104     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2105 }
2106
2107 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2108 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2109 {
2110     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2111     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2112     glsl_src_param_t src0_param;
2113     DWORD write_mask;
2114
2115     write_mask = shader_glsl_append_dst(buffer, ins);
2116     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2117
2118     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2119      * shader versions WINED3DSIO_MOVA is used for this. */
2120     if (ins->ctx->reg_maps->shader_version.major == 1
2121             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2122             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2123     {
2124         /* This is a simple floor() */
2125         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2126         if (mask_size > 1) {
2127             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2128         } else {
2129             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2130         }
2131     }
2132     else if(ins->handler_idx == WINED3DSIH_MOVA)
2133     {
2134         /* We need to *round* to the nearest int here. */
2135         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2136
2137         if (gl_info->supported[EXT_GPU_SHADER4])
2138         {
2139             if (mask_size > 1)
2140                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2141             else
2142                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2143         }
2144         else
2145         {
2146             if (mask_size > 1)
2147                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2148                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2149             else
2150                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2151                         src0_param.param_str, src0_param.param_str);
2152         }
2153     }
2154     else
2155     {
2156         shader_addline(buffer, "%s);\n", src0_param.param_str);
2157     }
2158 }
2159
2160 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2161 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2162 {
2163     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2164     glsl_src_param_t src0_param;
2165     glsl_src_param_t src1_param;
2166     DWORD dst_write_mask, src_write_mask;
2167     unsigned int dst_size = 0;
2168
2169     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2170     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2171
2172     /* dp3 works on vec3, dp4 on vec4 */
2173     if (ins->handler_idx == WINED3DSIH_DP4)
2174     {
2175         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2176     } else {
2177         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2178     }
2179
2180     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2181     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2182
2183     if (dst_size > 1) {
2184         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2185     } else {
2186         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2187     }
2188 }
2189
2190 /* Note that this instruction has some restrictions. The destination write mask
2191  * can't contain the w component, and the source swizzles have to be .xyzw */
2192 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2193 {
2194     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2195     glsl_src_param_t src0_param;
2196     glsl_src_param_t src1_param;
2197     char dst_mask[6];
2198
2199     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2200     shader_glsl_append_dst(ins->ctx->buffer, ins);
2201     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2202     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2203     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2204 }
2205
2206 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2207  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2208  * GLSL uses the value as-is. */
2209 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2210 {
2211     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2212     glsl_src_param_t src0_param;
2213     glsl_src_param_t src1_param;
2214     DWORD dst_write_mask;
2215     unsigned int dst_size;
2216
2217     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2218     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2219
2220     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2221     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2222
2223     if (dst_size > 1)
2224     {
2225         shader_addline(buffer, "vec%u(%s == 0.0 ? 1.0 : pow(abs(%s), %s)));\n",
2226                 dst_size, src1_param.param_str, src0_param.param_str, src1_param.param_str);
2227     }
2228     else
2229     {
2230         shader_addline(buffer, "%s == 0.0 ? 1.0 : pow(abs(%s), %s));\n",
2231                 src1_param.param_str, src0_param.param_str, src1_param.param_str);
2232     }
2233 }
2234
2235 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2236  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2237  * GLSL uses the value as-is. */
2238 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2239 {
2240     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2241     glsl_src_param_t src0_param;
2242     DWORD dst_write_mask;
2243     unsigned int dst_size;
2244
2245     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2246     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2247
2248     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2249
2250     if (dst_size > 1)
2251     {
2252         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2253                 dst_size, src0_param.param_str);
2254     }
2255     else
2256     {
2257         shader_addline(buffer, "log2(abs(%s)));\n",
2258                 src0_param.param_str);
2259     }
2260 }
2261
2262 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2263 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2264 {
2265     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2266     glsl_src_param_t src_param;
2267     const char *instruction;
2268     DWORD write_mask;
2269     unsigned i;
2270
2271     /* Determine the GLSL function to use based on the opcode */
2272     /* TODO: Possibly make this a table for faster lookups */
2273     switch (ins->handler_idx)
2274     {
2275         case WINED3DSIH_MIN: instruction = "min"; break;
2276         case WINED3DSIH_MAX: instruction = "max"; break;
2277         case WINED3DSIH_ABS: instruction = "abs"; break;
2278         case WINED3DSIH_FRC: instruction = "fract"; break;
2279         case WINED3DSIH_EXP: instruction = "exp2"; break;
2280         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2281         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2282         default: instruction = "";
2283             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2284             break;
2285     }
2286
2287     write_mask = shader_glsl_append_dst(buffer, ins);
2288
2289     shader_addline(buffer, "%s(", instruction);
2290
2291     if (ins->src_count)
2292     {
2293         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2294         shader_addline(buffer, "%s", src_param.param_str);
2295         for (i = 1; i < ins->src_count; ++i)
2296         {
2297             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2298             shader_addline(buffer, ", %s", src_param.param_str);
2299         }
2300     }
2301
2302     shader_addline(buffer, "));\n");
2303 }
2304
2305 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2306 {
2307     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2308     glsl_src_param_t src_param;
2309     unsigned int mask_size;
2310     DWORD write_mask;
2311     char dst_mask[6];
2312
2313     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2314     mask_size = shader_glsl_get_write_mask_size(write_mask);
2315     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2316
2317     shader_addline(buffer, "tmp0.x = dot(%s, %s);\n",
2318             src_param.param_str, src_param.param_str);
2319     shader_glsl_append_dst(buffer, ins);
2320
2321     if (mask_size > 1)
2322     {
2323         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s * inversesqrt(tmp0.x)));\n",
2324                 mask_size, src_param.param_str);
2325     }
2326     else
2327     {
2328         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s * inversesqrt(tmp0.x)));\n",
2329                 src_param.param_str);
2330     }
2331 }
2332
2333 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2334  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2335  *   dst.x = 2^(floor(src))
2336  *   dst.y = src - floor(src)
2337  *   dst.z = 2^src   (partial precision is allowed, but optional)
2338  *   dst.w = 1.0;
2339  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2340  *   dst = 2^src;    (partial precision is allowed, but optional)
2341  */
2342 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2343 {
2344     glsl_src_param_t src_param;
2345
2346     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2347
2348     if (ins->ctx->reg_maps->shader_version.major < 2)
2349     {
2350         char dst_mask[6];
2351
2352         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2353         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2354         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2355         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2356
2357         shader_glsl_append_dst(ins->ctx->buffer, ins);
2358         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2359         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2360     } else {
2361         DWORD write_mask;
2362         unsigned int mask_size;
2363
2364         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2365         mask_size = shader_glsl_get_write_mask_size(write_mask);
2366
2367         if (mask_size > 1) {
2368             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2369         } else {
2370             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2371         }
2372     }
2373 }
2374
2375 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2376 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2377 {
2378     glsl_src_param_t src_param;
2379     DWORD write_mask;
2380     unsigned int mask_size;
2381
2382     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2383     mask_size = shader_glsl_get_write_mask_size(write_mask);
2384     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2385
2386     if (mask_size > 1)
2387     {
2388         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2389                 mask_size, src_param.param_str);
2390     }
2391     else
2392     {
2393         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2394                 src_param.param_str);
2395     }
2396 }
2397
2398 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2399 {
2400     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2401     glsl_src_param_t src_param;
2402     DWORD write_mask;
2403     unsigned int mask_size;
2404
2405     write_mask = shader_glsl_append_dst(buffer, ins);
2406     mask_size = shader_glsl_get_write_mask_size(write_mask);
2407
2408     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2409
2410     if (mask_size > 1)
2411     {
2412         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2413                 mask_size, src_param.param_str);
2414     }
2415     else
2416     {
2417         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2418                 src_param.param_str);
2419     }
2420 }
2421
2422 /** Process signed comparison opcodes in GLSL. */
2423 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2424 {
2425     glsl_src_param_t src0_param;
2426     glsl_src_param_t src1_param;
2427     DWORD write_mask;
2428     unsigned int mask_size;
2429
2430     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2431     mask_size = shader_glsl_get_write_mask_size(write_mask);
2432     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2433     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2434
2435     if (mask_size > 1) {
2436         const char *compare;
2437
2438         switch(ins->handler_idx)
2439         {
2440             case WINED3DSIH_SLT: compare = "lessThan"; break;
2441             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2442             default: compare = "";
2443                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2444         }
2445
2446         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2447                 src0_param.param_str, src1_param.param_str);
2448     } else {
2449         switch(ins->handler_idx)
2450         {
2451             case WINED3DSIH_SLT:
2452                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2453                  * to return 0.0 but step returns 1.0 because step is not < x
2454                  * An alternative is a bvec compare padded with an unused second component.
2455                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2456                  * issue. Playing with not() is not possible either because not() does not accept
2457                  * a scalar.
2458                  */
2459                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2460                         src0_param.param_str, src1_param.param_str);
2461                 break;
2462             case WINED3DSIH_SGE:
2463                 /* Here we can use the step() function and safe a conditional */
2464                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2465                 break;
2466             default:
2467                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2468         }
2469
2470     }
2471 }
2472
2473 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2474 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2475 {
2476     glsl_src_param_t src0_param;
2477     glsl_src_param_t src1_param;
2478     glsl_src_param_t src2_param;
2479     DWORD write_mask, cmp_channel = 0;
2480     unsigned int i, j;
2481     char mask_char[6];
2482     BOOL temp_destination = FALSE;
2483
2484     if (shader_is_scalar(&ins->src[0].reg))
2485     {
2486         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2487
2488         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2489         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2490         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2491
2492         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2493                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2494     } else {
2495         DWORD dst_mask = ins->dst[0].write_mask;
2496         struct wined3d_shader_dst_param dst = ins->dst[0];
2497
2498         /* Cycle through all source0 channels */
2499         for (i=0; i<4; i++) {
2500             write_mask = 0;
2501             /* Find the destination channels which use the current source0 channel */
2502             for (j=0; j<4; j++) {
2503                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2504                 {
2505                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2506                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2507                 }
2508             }
2509             dst.write_mask = dst_mask & write_mask;
2510
2511             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2512             * The first lines may overwrite source parameters of the following lines.
2513             * Deal with that by using a temporary destination register if needed
2514             */
2515             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2516                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2517                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2518                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2519                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2520                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2521             {
2522                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2523                 if (!write_mask) continue;
2524                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2525                 temp_destination = TRUE;
2526             } else {
2527                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2528                 if (!write_mask) continue;
2529             }
2530
2531             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2532             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2533             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2534
2535             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2536                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2537         }
2538
2539         if(temp_destination) {
2540             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2541             shader_glsl_append_dst(ins->ctx->buffer, ins);
2542             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2543         }
2544     }
2545
2546 }
2547
2548 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2549 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2550  * the compare is done per component of src0. */
2551 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2552 {
2553     struct wined3d_shader_dst_param dst;
2554     glsl_src_param_t src0_param;
2555     glsl_src_param_t src1_param;
2556     glsl_src_param_t src2_param;
2557     DWORD write_mask, cmp_channel = 0;
2558     unsigned int i, j;
2559     DWORD dst_mask;
2560     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2561             ins->ctx->reg_maps->shader_version.minor);
2562
2563     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2564     {
2565         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2566         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2567         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2568         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2569
2570         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2571         if (ins->coissue)
2572         {
2573             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2574         } else {
2575             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2576                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2577         }
2578         return;
2579     }
2580     /* Cycle through all source0 channels */
2581     dst_mask = ins->dst[0].write_mask;
2582     dst = ins->dst[0];
2583     for (i=0; i<4; i++) {
2584         write_mask = 0;
2585         /* Find the destination channels which use the current source0 channel */
2586         for (j=0; j<4; j++) {
2587             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2588             {
2589                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2590                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2591             }
2592         }
2593
2594         dst.write_mask = dst_mask & write_mask;
2595         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2596         if (!write_mask) continue;
2597
2598         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2599         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2600         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2601
2602         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2603                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2604     }
2605 }
2606
2607 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2608 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2609 {
2610     glsl_src_param_t src0_param;
2611     glsl_src_param_t src1_param;
2612     glsl_src_param_t src2_param;
2613     DWORD write_mask;
2614
2615     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2616     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2617     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2618     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2619     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2620             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2621 }
2622
2623 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2624    Vertex shaders to GLSL codes */
2625 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2626 {
2627     int i;
2628     int nComponents = 0;
2629     struct wined3d_shader_dst_param tmp_dst = {{0}};
2630     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2631     struct wined3d_shader_instruction tmp_ins;
2632
2633     memset(&tmp_ins, 0, sizeof(tmp_ins));
2634
2635     /* Set constants for the temporary argument */
2636     tmp_ins.ctx = ins->ctx;
2637     tmp_ins.dst_count = 1;
2638     tmp_ins.dst = &tmp_dst;
2639     tmp_ins.src_count = 2;
2640     tmp_ins.src = tmp_src;
2641
2642     switch(ins->handler_idx)
2643     {
2644         case WINED3DSIH_M4x4:
2645             nComponents = 4;
2646             tmp_ins.handler_idx = WINED3DSIH_DP4;
2647             break;
2648         case WINED3DSIH_M4x3:
2649             nComponents = 3;
2650             tmp_ins.handler_idx = WINED3DSIH_DP4;
2651             break;
2652         case WINED3DSIH_M3x4:
2653             nComponents = 4;
2654             tmp_ins.handler_idx = WINED3DSIH_DP3;
2655             break;
2656         case WINED3DSIH_M3x3:
2657             nComponents = 3;
2658             tmp_ins.handler_idx = WINED3DSIH_DP3;
2659             break;
2660         case WINED3DSIH_M3x2:
2661             nComponents = 2;
2662             tmp_ins.handler_idx = WINED3DSIH_DP3;
2663             break;
2664         default:
2665             break;
2666     }
2667
2668     tmp_dst = ins->dst[0];
2669     tmp_src[0] = ins->src[0];
2670     tmp_src[1] = ins->src[1];
2671     for (i = 0; i < nComponents; ++i)
2672     {
2673         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2674         shader_glsl_dot(&tmp_ins);
2675         ++tmp_src[1].reg.idx;
2676     }
2677 }
2678
2679 /**
2680     The LRP instruction performs a component-wise linear interpolation
2681     between the second and third operands using the first operand as the
2682     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2683     This is equivalent to mix(src2, src1, src0);
2684 */
2685 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2686 {
2687     glsl_src_param_t src0_param;
2688     glsl_src_param_t src1_param;
2689     glsl_src_param_t src2_param;
2690     DWORD write_mask;
2691
2692     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2693
2694     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2695     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2696     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2697
2698     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2699             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2700 }
2701
2702 /** Process the WINED3DSIO_LIT instruction in GLSL:
2703  * dst.x = dst.w = 1.0
2704  * dst.y = (src0.x > 0) ? src0.x
2705  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2706  *                                        where src.w is clamped at +- 128
2707  */
2708 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2709 {
2710     glsl_src_param_t src0_param;
2711     glsl_src_param_t src1_param;
2712     glsl_src_param_t src3_param;
2713     char dst_mask[6];
2714
2715     shader_glsl_append_dst(ins->ctx->buffer, ins);
2716     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2717
2718     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2719     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2720     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2721
2722     /* The sdk specifies the instruction like this
2723      * dst.x = 1.0;
2724      * if(src.x > 0.0) dst.y = src.x
2725      * else dst.y = 0.0.
2726      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2727      * else dst.z = 0.0;
2728      * dst.w = 1.0;
2729      *
2730      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2731      * dst.x = 1.0                                  ... No further explanation needed
2732      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2733      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2734      * dst.w = 1.0.                                 ... Nothing fancy.
2735      *
2736      * So we still have one conditional in there. So do this:
2737      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2738      *
2739      * 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),
2740      * 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.
2741      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2742      */
2743     shader_addline(ins->ctx->buffer,
2744             "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",
2745             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2746 }
2747
2748 /** Process the WINED3DSIO_DST instruction in GLSL:
2749  * dst.x = 1.0
2750  * dst.y = src0.x * src0.y
2751  * dst.z = src0.z
2752  * dst.w = src1.w
2753  */
2754 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2755 {
2756     glsl_src_param_t src0y_param;
2757     glsl_src_param_t src0z_param;
2758     glsl_src_param_t src1y_param;
2759     glsl_src_param_t src1w_param;
2760     char dst_mask[6];
2761
2762     shader_glsl_append_dst(ins->ctx->buffer, ins);
2763     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2764
2765     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2766     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2767     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2768     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2769
2770     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2771             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2772 }
2773
2774 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2775  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2776  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2777  *
2778  * dst.x = cos(src0.?)
2779  * dst.y = sin(src0.?)
2780  * dst.z = dst.z
2781  * dst.w = dst.w
2782  */
2783 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2784 {
2785     glsl_src_param_t src0_param;
2786     DWORD write_mask;
2787
2788     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2789     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2790
2791     switch (write_mask) {
2792         case WINED3DSP_WRITEMASK_0:
2793             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2794             break;
2795
2796         case WINED3DSP_WRITEMASK_1:
2797             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2798             break;
2799
2800         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2801             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2802             break;
2803
2804         default:
2805             ERR("Write mask should be .x, .y or .xy\n");
2806             break;
2807     }
2808 }
2809
2810 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2811  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2812  * generate invalid code
2813  */
2814 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2815 {
2816     glsl_src_param_t src0_param;
2817     DWORD write_mask;
2818
2819     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2820     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2821
2822     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2823 }
2824
2825 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2826  * Start a for() loop where src1.y is the initial value of aL,
2827  *  increment aL by src1.z for a total of src1.x iterations.
2828  *  Need to use a temporary variable for this operation.
2829  */
2830 /* FIXME: I don't think nested loops will work correctly this way. */
2831 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2832 {
2833     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2834     glsl_src_param_t src1_param;
2835     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
2836     const DWORD *control_values = NULL;
2837     const local_constant *constant;
2838
2839     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2840
2841     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2842      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2843      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2844      * addressing.
2845      */
2846     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2847     {
2848         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2849             if (constant->idx == ins->src[1].reg.idx)
2850             {
2851                 control_values = constant->value;
2852                 break;
2853             }
2854         }
2855     }
2856
2857     if (control_values)
2858     {
2859         struct wined3d_shader_loop_control loop_control;
2860         loop_control.count = control_values[0];
2861         loop_control.start = control_values[1];
2862         loop_control.step = (int)control_values[2];
2863
2864         if (loop_control.step > 0)
2865         {
2866             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2867                     loop_state->current_depth, loop_control.start,
2868                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2869                     loop_state->current_depth, loop_control.step);
2870         }
2871         else if (loop_control.step < 0)
2872         {
2873             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2874                     loop_state->current_depth, loop_control.start,
2875                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2876                     loop_state->current_depth, loop_control.step);
2877         }
2878         else
2879         {
2880             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2881                     loop_state->current_depth, loop_control.start, loop_state->current_depth,
2882                     loop_state->current_depth, loop_control.count,
2883                     loop_state->current_depth);
2884         }
2885     } else {
2886         shader_addline(ins->ctx->buffer,
2887                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2888                 loop_state->current_depth, loop_state->current_reg,
2889                 src1_param.reg_name, loop_state->current_depth, src1_param.reg_name,
2890                 loop_state->current_depth, loop_state->current_reg, src1_param.reg_name);
2891     }
2892
2893     ++loop_state->current_depth;
2894     ++loop_state->current_reg;
2895 }
2896
2897 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2898 {
2899     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2900
2901     shader_addline(ins->ctx->buffer, "}\n");
2902
2903     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2904     {
2905         --loop_state->current_depth;
2906         --loop_state->current_reg;
2907     }
2908
2909     if (ins->handler_idx == WINED3DSIH_ENDREP)
2910     {
2911         --loop_state->current_depth;
2912     }
2913 }
2914
2915 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2916 {
2917     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
2918     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2919     glsl_src_param_t src0_param;
2920     const DWORD *control_values = NULL;
2921     const local_constant *constant;
2922
2923     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2924     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2925     {
2926         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2927         {
2928             if (constant->idx == ins->src[0].reg.idx)
2929             {
2930                 control_values = constant->value;
2931                 break;
2932             }
2933         }
2934     }
2935
2936     if (control_values)
2937     {
2938         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2939                 loop_state->current_depth, loop_state->current_depth,
2940                 control_values[0], loop_state->current_depth);
2941     }
2942     else
2943     {
2944         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2945         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2946                 loop_state->current_depth, loop_state->current_depth,
2947                 src0_param.param_str, loop_state->current_depth);
2948     }
2949
2950     ++loop_state->current_depth;
2951 }
2952
2953 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2954 {
2955     glsl_src_param_t src0_param;
2956
2957     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2958     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2959 }
2960
2961 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2962 {
2963     glsl_src_param_t src0_param;
2964     glsl_src_param_t src1_param;
2965
2966     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2967     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2968
2969     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2970             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2971 }
2972
2973 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2974 {
2975     shader_addline(ins->ctx->buffer, "} else {\n");
2976 }
2977
2978 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2979 {
2980     shader_addline(ins->ctx->buffer, "break;\n");
2981 }
2982
2983 /* FIXME: According to MSDN the compare is done per component. */
2984 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2985 {
2986     glsl_src_param_t src0_param;
2987     glsl_src_param_t src1_param;
2988
2989     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2990     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2991
2992     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2993             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2994 }
2995
2996 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2997 {
2998     shader_addline(ins->ctx->buffer, "}\n");
2999     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
3000 }
3001
3002 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
3003 {
3004     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
3005 }
3006
3007 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3008 {
3009     glsl_src_param_t src1_param;
3010
3011     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3012     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3013 }
3014
3015 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3016 {
3017     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3018      * function only suppresses the unhandled instruction warning
3019      */
3020 }
3021
3022 /*********************************************
3023  * Pixel Shader Specific Code begins here
3024  ********************************************/
3025 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3026 {
3027     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
3028     IWineD3DDeviceImpl *device = shader->baseShader.device;
3029     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3030             ins->ctx->reg_maps->shader_version.minor);
3031     glsl_sample_function_t sample_function;
3032     const struct wined3d_texture *texture;
3033     DWORD sample_flags = 0;
3034     DWORD sampler_idx;
3035     DWORD mask = 0, swizzle;
3036
3037     /* 1.0-1.4: Use destination register as sampler source.
3038      * 2.0+: Use provided sampler source. */
3039     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3040     else sampler_idx = ins->src[1].reg.idx;
3041     texture = device->stateBlock->state.textures[sampler_idx];
3042
3043     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3044     {
3045         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3046         DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3047                 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3048         WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3049
3050         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3051         if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3052             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3053             switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3054                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3055                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3056                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3057                 case WINED3DTTFF_COUNT4:
3058                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3059             }
3060         }
3061     }
3062     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3063     {
3064         DWORD src_mod = ins->src[0].modifiers;
3065
3066         if (src_mod == WINED3DSPSM_DZ) {
3067             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3068             mask = WINED3DSP_WRITEMASK_2;
3069         } else if (src_mod == WINED3DSPSM_DW) {
3070             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3071             mask = WINED3DSP_WRITEMASK_3;
3072         }
3073     } else {
3074         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3075         {
3076             /* ps 2.0 texldp instruction always divides by the fourth component. */
3077             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3078             mask = WINED3DSP_WRITEMASK_3;
3079         }
3080     }
3081
3082     if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3083         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3084
3085     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3086     mask |= sample_function.coord_mask;
3087
3088     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3089     else swizzle = ins->src[1].swizzle;
3090
3091     /* 1.0-1.3: Use destination register as coordinate source.
3092        1.4+: Use provided coordinate source register. */
3093     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3094     {
3095         char coord_mask[6];
3096         shader_glsl_write_mask_to_str(mask, coord_mask);
3097         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3098                 "T%u%s", sampler_idx, coord_mask);
3099     } else {
3100         glsl_src_param_t coord_param;
3101         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3102         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3103         {
3104             glsl_src_param_t bias;
3105             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3106             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3107                     "%s", coord_param.param_str);
3108         } else {
3109             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3110                     "%s", coord_param.param_str);
3111         }
3112     }
3113 }
3114
3115 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3116 {
3117     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
3118     IWineD3DDeviceImpl *device = shader->baseShader.device;
3119     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3120     glsl_sample_function_t sample_function;
3121     glsl_src_param_t coord_param, dx_param, dy_param;
3122     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3123     DWORD sampler_idx;
3124     DWORD swizzle = ins->src[1].swizzle;
3125     const struct wined3d_texture *texture;
3126
3127     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3128     {
3129         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3130         shader_glsl_tex(ins);
3131         return;
3132     }
3133
3134     sampler_idx = ins->src[1].reg.idx;
3135     texture = device->stateBlock->state.textures[sampler_idx];
3136     if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3137         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3138
3139     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3140     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3141     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3142     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3143
3144     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3145                                 "%s", coord_param.param_str);
3146 }
3147
3148 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3149 {
3150     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
3151     IWineD3DDeviceImpl *device = shader->baseShader.device;
3152     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3153     glsl_sample_function_t sample_function;
3154     glsl_src_param_t coord_param, lod_param;
3155     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3156     DWORD sampler_idx;
3157     DWORD swizzle = ins->src[1].swizzle;
3158     const struct wined3d_texture *texture;
3159
3160     sampler_idx = ins->src[1].reg.idx;
3161     texture = device->stateBlock->state.textures[sampler_idx];
3162     if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3163         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3164
3165     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3166     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3167
3168     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3169
3170     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3171             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3172     {
3173         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3174          * However, they seem to work just fine in fragment shaders as well. */
3175         WARN("Using %s in fragment shader.\n", sample_function.name);
3176     }
3177     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3178             "%s", coord_param.param_str);
3179 }
3180
3181 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3182 {
3183     /* FIXME: Make this work for more than just 2D textures */
3184     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3185     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3186
3187     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3188     {
3189         char dst_mask[6];
3190
3191         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3192         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3193                 ins->dst[0].reg.idx, dst_mask);
3194     } else {
3195         DWORD reg = ins->src[0].reg.idx;
3196         DWORD src_mod = ins->src[0].modifiers;
3197         char dst_swizzle[6];
3198
3199         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3200
3201         if (src_mod == WINED3DSPSM_DZ) {
3202             glsl_src_param_t div_param;
3203             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3204             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3205
3206             if (mask_size > 1) {
3207                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3208             } else {
3209                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3210             }
3211         } else if (src_mod == WINED3DSPSM_DW) {
3212             glsl_src_param_t div_param;
3213             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3214             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3215
3216             if (mask_size > 1) {
3217                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3218             } else {
3219                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3220             }
3221         } else {
3222             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3223         }
3224     }
3225 }
3226
3227 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3228  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3229  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3230 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3231 {
3232     glsl_src_param_t src0_param;
3233     glsl_sample_function_t sample_function;
3234     DWORD sampler_idx = ins->dst[0].reg.idx;
3235     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3236     UINT mask_size;
3237
3238     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3239
3240     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3241      * scalar, and projected sampling would require 4.
3242      *
3243      * It is a dependent read - not valid with conditional NP2 textures
3244      */
3245     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3246     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3247
3248     switch(mask_size)
3249     {
3250         case 1:
3251             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3252                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3253             break;
3254
3255         case 2:
3256             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3257                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3258             break;
3259
3260         case 3:
3261             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3262                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3263             break;
3264
3265         default:
3266             FIXME("Unexpected mask size %u\n", mask_size);
3267             break;
3268     }
3269 }
3270
3271 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3272  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3273 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3274 {
3275     glsl_src_param_t src0_param;
3276     DWORD dstreg = ins->dst[0].reg.idx;
3277     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3278     DWORD dst_mask;
3279     unsigned int mask_size;
3280
3281     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3282     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3283     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3284
3285     if (mask_size > 1) {
3286         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3287     } else {
3288         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3289     }
3290 }
3291
3292 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3293  * Calculate the depth as dst.x / dst.y   */
3294 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3295 {
3296     glsl_dst_param_t dst_param;
3297
3298     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3299
3300     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3301      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3302      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3303      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3304      * >= 1.0 or < 0.0
3305      */
3306     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3307             dst_param.reg_name, dst_param.reg_name);
3308 }
3309
3310 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3311  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3312  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3313  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3314  */
3315 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3316 {
3317     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3318     DWORD dstreg = ins->dst[0].reg.idx;
3319     glsl_src_param_t src0_param;
3320
3321     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3322
3323     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3324     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3325 }
3326
3327 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3328  * Calculate the 1st of a 2-row matrix multiplication. */
3329 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3330 {
3331     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3332     DWORD reg = ins->dst[0].reg.idx;
3333     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3334     glsl_src_param_t src0_param;
3335
3336     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3337     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3338 }
3339
3340 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3341  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3342 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3343 {
3344     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3345     DWORD reg = ins->dst[0].reg.idx;
3346     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3347     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3348     glsl_src_param_t src0_param;
3349
3350     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3351     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + tex_mx->current_row, reg, src0_param.param_str);
3352     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
3353 }
3354
3355 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3356 {
3357     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3358     DWORD reg = ins->dst[0].reg.idx;
3359     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3360     glsl_src_param_t src0_param;
3361     glsl_sample_function_t sample_function;
3362
3363     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3364     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3365
3366     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3367
3368     /* Sample the texture using the calculated coordinates */
3369     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3370 }
3371
3372 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3373  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3374 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3375 {
3376     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3377     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3378     glsl_src_param_t src0_param;
3379     DWORD reg = ins->dst[0].reg.idx;
3380     glsl_sample_function_t sample_function;
3381
3382     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3383     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3384
3385     /* Dependent read, not valid with conditional NP2 */
3386     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3387
3388     /* Sample the texture using the calculated coordinates */
3389     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3390
3391     tex_mx->current_row = 0;
3392 }
3393
3394 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3395  * Perform the 3rd row of a 3x3 matrix multiply */
3396 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3397 {
3398     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3399     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3400     glsl_src_param_t src0_param;
3401     char dst_mask[6];
3402     DWORD reg = ins->dst[0].reg.idx;
3403
3404     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3405
3406     shader_glsl_append_dst(ins->ctx->buffer, ins);
3407     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3408     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3409
3410     tex_mx->current_row = 0;
3411 }
3412
3413 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3414  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3415 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3416 {
3417     DWORD reg = ins->dst[0].reg.idx;
3418     glsl_src_param_t src0_param;
3419     glsl_src_param_t src1_param;
3420     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3421     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3422     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3423     glsl_sample_function_t sample_function;
3424
3425     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3426     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3427
3428     /* Perform the last matrix multiply operation */
3429     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3430     /* Reflection calculation */
3431     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3432
3433     /* Dependent read, not valid with conditional NP2 */
3434     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3435
3436     /* Sample the texture */
3437     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3438
3439     tex_mx->current_row = 0;
3440 }
3441
3442 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3443  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3444 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3445 {
3446     DWORD reg = ins->dst[0].reg.idx;
3447     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3448     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3449     glsl_src_param_t src0_param;
3450     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3451     glsl_sample_function_t sample_function;
3452
3453     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3454
3455     /* Perform the last matrix multiply operation */
3456     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3457
3458     /* Construct the eye-ray vector from w coordinates */
3459     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3460             tex_mx->texcoord_w[0], tex_mx->texcoord_w[1], reg);
3461     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3462
3463     /* Dependent read, not valid with conditional NP2 */
3464     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3465
3466     /* Sample the texture using the calculated coordinates */
3467     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3468
3469     tex_mx->current_row = 0;
3470 }
3471
3472 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3473  * Apply a fake bump map transform.
3474  * texbem is pshader <= 1.3 only, this saves a few version checks
3475  */
3476 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3477 {
3478     const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3479     glsl_sample_function_t sample_function;
3480     glsl_src_param_t coord_param;
3481     DWORD sampler_idx;
3482     DWORD mask;
3483     DWORD flags;
3484     char coord_mask[6];
3485
3486     sampler_idx = ins->dst[0].reg.idx;
3487     flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3488             & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3489
3490     /* Dependent read, not valid with conditional NP2 */
3491     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3492     mask = sample_function.coord_mask;
3493
3494     shader_glsl_write_mask_to_str(mask, coord_mask);
3495
3496     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3497          * so we can't let the GL handle this.
3498          */
3499     if (flags & WINED3D_PSARGS_PROJECTED) {
3500         DWORD div_mask=0;
3501         char coord_div_mask[3];
3502         switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3503             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3504             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3505             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3506             case WINED3DTTFF_COUNT4:
3507             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3508         }
3509         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3510         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3511     }
3512
3513     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3514
3515     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3516             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3517             coord_param.param_str, coord_mask);
3518
3519     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3520     {
3521         glsl_src_param_t luminance_param;
3522         glsl_dst_param_t dst_param;
3523
3524         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3525         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3526
3527         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3528                 dst_param.reg_name, dst_param.mask_str,
3529                 luminance_param.param_str, sampler_idx, sampler_idx);
3530     }
3531 }
3532
3533 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3534 {
3535     glsl_src_param_t src0_param, src1_param;
3536     DWORD sampler_idx = ins->dst[0].reg.idx;
3537
3538     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3539     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3540
3541     shader_glsl_append_dst(ins->ctx->buffer, ins);
3542     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3543             src0_param.param_str, sampler_idx, src1_param.param_str);
3544 }
3545
3546 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3547  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3548 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3549 {
3550     glsl_src_param_t src0_param;
3551     DWORD sampler_idx = ins->dst[0].reg.idx;
3552     glsl_sample_function_t sample_function;
3553
3554     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3555
3556     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3557     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3558             "%s.wx", src0_param.reg_name);
3559 }
3560
3561 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3562  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3563 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3564 {
3565     glsl_src_param_t src0_param;
3566     DWORD sampler_idx = ins->dst[0].reg.idx;
3567     glsl_sample_function_t sample_function;
3568
3569     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3570
3571     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3572     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3573             "%s.yz", src0_param.reg_name);
3574 }
3575
3576 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3577  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3578 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3579 {
3580     glsl_src_param_t src0_param;
3581     DWORD sampler_idx = ins->dst[0].reg.idx;
3582     glsl_sample_function_t sample_function;
3583
3584     /* Dependent read, not valid with conditional NP2 */
3585     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3586     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3587
3588     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3589             "%s", src0_param.param_str);
3590 }
3591
3592 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3593  * If any of the first 3 components are < 0, discard this pixel */
3594 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3595 {
3596     glsl_dst_param_t dst_param;
3597
3598     /* The argument is a destination parameter, and no writemasks are allowed */
3599     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3600     if (ins->ctx->reg_maps->shader_version.major >= 2)
3601     {
3602         /* 2.0 shaders compare all 4 components in texkill */
3603         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3604     } else {
3605         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3606          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3607          * 4 components are defined, only the first 3 are used
3608          */
3609         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3610     }
3611 }
3612
3613 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3614  * dst = dot2(src0, src1) + src2 */
3615 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3616 {
3617     glsl_src_param_t src0_param;
3618     glsl_src_param_t src1_param;
3619     glsl_src_param_t src2_param;
3620     DWORD write_mask;
3621     unsigned int mask_size;
3622
3623     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3624     mask_size = shader_glsl_get_write_mask_size(write_mask);
3625
3626     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3627     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3628     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3629
3630     if (mask_size > 1) {
3631         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3632                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3633     } else {
3634         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3635                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3636     }
3637 }
3638
3639 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3640         const struct wined3d_shader_signature_element *input_signature, const struct wined3d_shader_reg_maps *reg_maps,
3641         enum vertexprocessing_mode vertexprocessing)
3642 {
3643     unsigned int i;
3644     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3645     WORD map = reg_maps->input_registers;
3646
3647     for (i = 0; map; map >>= 1, ++i)
3648     {
3649         const char *semantic_name;
3650         UINT semantic_idx;
3651         char reg_mask[6];
3652
3653         /* Unused */
3654         if (!(map & 1)) continue;
3655
3656         semantic_name = input_signature[i].semantic_name;
3657         semantic_idx = input_signature[i].semantic_idx;
3658         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3659
3660         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3661         {
3662             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3663                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3664                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3665             else
3666                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3667                         This->input_reg_map[i], reg_mask, reg_mask);
3668         }
3669         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3670         {
3671             if (!semantic_idx)
3672                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3673                         This->input_reg_map[i], reg_mask, reg_mask);
3674             else if (semantic_idx == 1)
3675                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3676                         This->input_reg_map[i], reg_mask, reg_mask);
3677             else
3678                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3679                         This->input_reg_map[i], reg_mask, reg_mask);
3680         }
3681         else
3682         {
3683             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3684                     This->input_reg_map[i], reg_mask, reg_mask);
3685         }
3686     }
3687 }
3688
3689 /*********************************************
3690  * Vertex Shader Specific Code begins here
3691  ********************************************/
3692
3693 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3694     glsl_program_key_t key;
3695
3696     key.vshader = entry->vshader;
3697     key.pshader = entry->pshader;
3698     key.vs_args = entry->vs_args;
3699     key.ps_args = entry->ps_args;
3700
3701     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3702     {
3703         ERR("Failed to insert program entry.\n");
3704     }
3705 }
3706
3707 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3708         IWineD3DVertexShaderImpl *vshader, IWineD3DPixelShaderImpl *pshader,
3709         struct vs_compile_args *vs_args, struct ps_compile_args *ps_args)
3710 {
3711     struct wine_rb_entry *entry;
3712     glsl_program_key_t key;
3713
3714     key.vshader = vshader;
3715     key.pshader = pshader;
3716     key.vs_args = *vs_args;
3717     key.ps_args = *ps_args;
3718
3719     entry = wine_rb_get(&priv->program_lookup, &key);
3720     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3721 }
3722
3723 /* GL locking is done by the caller */
3724 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3725         struct glsl_shader_prog_link *entry)
3726 {
3727     glsl_program_key_t key;
3728
3729     key.vshader = entry->vshader;
3730     key.pshader = entry->pshader;
3731     key.vs_args = entry->vs_args;
3732     key.ps_args = entry->ps_args;
3733     wine_rb_remove(&priv->program_lookup, &key);
3734
3735     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3736     if (entry->vshader) list_remove(&entry->vshader_entry);
3737     if (entry->pshader) list_remove(&entry->pshader_entry);
3738     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3739     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3740     HeapFree(GetProcessHeap(), 0, entry);
3741 }
3742
3743 static void handle_ps3_input(struct wined3d_shader_buffer *buffer,
3744         const struct wined3d_gl_info *gl_info, const DWORD *map,
3745         const struct wined3d_shader_signature_element *input_signature,
3746         const struct wined3d_shader_reg_maps *reg_maps_in,
3747         const struct wined3d_shader_signature_element *output_signature,
3748         const struct wined3d_shader_reg_maps *reg_maps_out)
3749 {
3750     unsigned int i, j;
3751     const char *semantic_name_in;
3752     UINT semantic_idx_in;
3753     DWORD *set;
3754     DWORD in_idx;
3755     unsigned int in_count = vec4_varyings(3, gl_info);
3756     char reg_mask[6];
3757     char destination[50];
3758     WORD input_map, output_map;
3759
3760     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3761
3762     input_map = reg_maps_in->input_registers;
3763     for (i = 0; input_map; input_map >>= 1, ++i)
3764     {
3765         if (!(input_map & 1)) continue;
3766
3767         in_idx = map[i];
3768         /* Declared, but not read register */
3769         if (in_idx == ~0U) continue;
3770         if (in_idx >= (in_count + 2))
3771         {
3772             FIXME("More input varyings declared than supported, expect issues.\n");
3773             continue;
3774         }
3775
3776         if (in_idx == in_count) {
3777             sprintf(destination, "gl_FrontColor");
3778         } else if (in_idx == in_count + 1) {
3779             sprintf(destination, "gl_FrontSecondaryColor");
3780         } else {
3781             sprintf(destination, "IN[%u]", in_idx);
3782         }
3783
3784         semantic_name_in = input_signature[i].semantic_name;
3785         semantic_idx_in = input_signature[i].semantic_idx;
3786         set[in_idx] = ~0U;
3787
3788         output_map = reg_maps_out->output_registers;
3789         for (j = 0; output_map; output_map >>= 1, ++j)
3790         {
3791             DWORD mask;
3792
3793             if (!(output_map & 1)
3794                     || semantic_idx_in != output_signature[j].semantic_idx
3795                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3796                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3797                 continue;
3798
3799             set[in_idx] = mask;
3800             shader_glsl_write_mask_to_str(mask, reg_mask);
3801
3802             shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3803                     destination, reg_mask, j, reg_mask);
3804         }
3805     }
3806
3807     for (i = 0; i < in_count + 2; ++i)
3808     {
3809         unsigned int size;
3810
3811         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3812             continue;
3813
3814         if (set[i] == ~0U) set[i] = 0;
3815
3816         size = 0;
3817         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3818         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3819         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3820         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3821         reg_mask[size] = '\0';
3822
3823         if (i == in_count) sprintf(destination, "gl_FrontColor");
3824         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3825         else sprintf(destination, "IN[%u]", i);
3826
3827         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3828         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3829     }
3830
3831     HeapFree(GetProcessHeap(), 0, set);
3832 }
3833
3834 /* GL locking is done by the caller */
3835 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3836         IWineD3DVertexShaderImpl *vs, IWineD3DPixelShaderImpl *ps, const struct wined3d_gl_info *gl_info)
3837 {
3838     GLhandleARB ret = 0;
3839     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3840     unsigned int i;
3841     const char *semantic_name;
3842     UINT semantic_idx;
3843     char reg_mask[6];
3844     const struct wined3d_shader_signature_element *output_signature = vs->baseShader.output_signature;
3845     WORD map = vs->baseShader.reg_maps.output_registers;
3846
3847     shader_buffer_clear(buffer);
3848
3849     shader_addline(buffer, "#version 120\n");
3850
3851     if (ps_major < 3)
3852     {
3853         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3854
3855         for (i = 0; map; map >>= 1, ++i)
3856         {
3857             DWORD write_mask;
3858
3859             if (!(map & 1)) continue;
3860
3861             semantic_name = output_signature[i].semantic_name;
3862             semantic_idx = output_signature[i].semantic_idx;
3863             write_mask = output_signature[i].mask;
3864             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3865
3866             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3867             {
3868                 if (!semantic_idx)
3869                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n",
3870                             reg_mask, i, reg_mask);
3871                 else if (semantic_idx == 1)
3872                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n",
3873                             reg_mask, i, reg_mask);
3874             }
3875             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3876             {
3877                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3878                         reg_mask, i, reg_mask);
3879             }
3880             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3881             {
3882                 if (semantic_idx < 8)
3883                 {
3884                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3885                         write_mask |= WINED3DSP_WRITEMASK_3;
3886
3887                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3888                             semantic_idx, reg_mask, i, reg_mask);
3889                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3890                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3891                 }
3892             }
3893             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3894             {
3895                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3896             }
3897             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3898             {
3899                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3900             }
3901         }
3902         shader_addline(buffer, "}\n");
3903
3904     }
3905     else
3906     {
3907         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3908         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3909         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3910
3911         /* First, sort out position and point size. Those are not passed to the pixel shader */
3912         for (i = 0; map; map >>= 1, ++i)
3913         {
3914             if (!(map & 1)) continue;
3915
3916             semantic_name = output_signature[i].semantic_name;
3917             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3918
3919             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3920             {
3921                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3922                         reg_mask, i, reg_mask);
3923             }
3924             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3925             {
3926                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3927             }
3928         }
3929
3930         /* Then, fix the pixel shader input */
3931         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3932                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3933
3934         shader_addline(buffer, "}\n");
3935     }
3936
3937     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3938     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3939     shader_glsl_compile(gl_info, ret, buffer->buffer);
3940
3941     return ret;
3942 }
3943
3944 /* GL locking is done by the caller */
3945 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3946         GLhandleARB programId, char prefix)
3947 {
3948     const local_constant *lconst;
3949     GLint tmp_loc;
3950     const float *value;
3951     char glsl_name[8];
3952
3953     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3954         value = (const float *)lconst->value;
3955         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3956         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3957         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3958     }
3959     checkGLcall("Hardcoding local constants");
3960 }
3961
3962 /* GL locking is done by the caller */
3963 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3964         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3965         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3966 {
3967     const struct wined3d_shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3968     const struct wined3d_gl_info *gl_info = context->gl_info;
3969     CONST DWORD *function = This->baseShader.function;
3970     struct shader_glsl_ctx_priv priv_ctx;
3971
3972     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3973     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3974
3975     memset(&priv_ctx, 0, sizeof(priv_ctx));
3976     priv_ctx.cur_ps_args = args;
3977     priv_ctx.cur_np2fixup_info = np2fixup_info;
3978
3979     shader_addline(buffer, "#version 120\n");
3980
3981     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3982     {
3983         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3984     }
3985     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3986     {
3987         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3988          * drivers write a warning if we don't do so
3989          */
3990         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3991     }
3992     if (gl_info->supported[EXT_GPU_SHADER4])
3993     {
3994         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3995     }
3996
3997     /* Base Declarations */
3998     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShaderImpl *)This, reg_maps, &priv_ctx);
3999
4000     /* Pack 3.0 inputs */
4001     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4002     {
4003         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
4004                 This->baseShader.input_signature, reg_maps, args->vp_mode);
4005     }
4006
4007     /* Base Shader Body */
4008     shader_generate_main((IWineD3DBaseShaderImpl *)This, buffer, reg_maps, function, &priv_ctx);
4009
4010     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4011     if (reg_maps->shader_version.major < 2)
4012     {
4013         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4014         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4015     }
4016
4017     if (args->srgb_correction)
4018     {
4019         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4020         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4021         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4022         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4023         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4024         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4025     }
4026     /* Pixel shader < 3.0 do not replace the fog stage.
4027      * This implements linear fog computation and blending.
4028      * TODO: non linear fog
4029      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4030      * -1/(e-s) and e/(e-s) respectively.
4031      */
4032     if (reg_maps->shader_version.major < 3)
4033     {
4034         switch(args->fog) {
4035             case FOG_OFF: break;
4036             case FOG_LINEAR:
4037                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4038                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4039                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4040                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4041                 break;
4042             case FOG_EXP:
4043                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4044                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4045                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4046                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4047                 break;
4048             case FOG_EXP2:
4049                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4050                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4051                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4052                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4053                 break;
4054         }
4055     }
4056
4057     shader_addline(buffer, "}\n");
4058
4059     TRACE("Compiling shader object %u\n", shader_obj);
4060     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4061
4062     /* Store the shader object */
4063     return shader_obj;
4064 }
4065
4066 /* GL locking is done by the caller */
4067 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4068         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4069         const struct vs_compile_args *args)
4070 {
4071     const struct wined3d_shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4072     const struct wined3d_gl_info *gl_info = context->gl_info;
4073     CONST DWORD *function = This->baseShader.function;
4074     struct shader_glsl_ctx_priv priv_ctx;
4075
4076     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4077     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4078
4079     shader_addline(buffer, "#version 120\n");
4080
4081     if (gl_info->supported[EXT_GPU_SHADER4])
4082         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4083
4084     memset(&priv_ctx, 0, sizeof(priv_ctx));
4085     priv_ctx.cur_vs_args = args;
4086
4087     /* Base Declarations */
4088     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShaderImpl *)This, reg_maps, &priv_ctx);
4089
4090     /* Base Shader Body */
4091     shader_generate_main((IWineD3DBaseShaderImpl *)This, buffer, reg_maps, function, &priv_ctx);
4092
4093     /* Unpack outputs */
4094     shader_addline(buffer, "order_ps_input(OUT);\n");
4095
4096     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4097      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4098      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4099      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4100      */
4101     if (args->fog_src == VS_FOG_Z)
4102         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4103     else if (!reg_maps->fog)
4104         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4105
4106     /* We always store the clipplanes without y inversion */
4107     if (args->clip_enabled)
4108         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4109
4110     /* Write the final position.
4111      *
4112      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4113      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4114      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4115      * contains 1.0 to allow a mad.
4116      */
4117     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4118     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4119
4120     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4121      *
4122      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4123      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4124      * which is the same as z = z * 2 - w.
4125      */
4126     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4127
4128     shader_addline(buffer, "}\n");
4129
4130     TRACE("Compiling shader object %u\n", shader_obj);
4131     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4132
4133     return shader_obj;
4134 }
4135
4136 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4137         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4138         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4139 {
4140     struct wined3d_state *state = &shader->baseShader.device->stateBlock->state;
4141     UINT i;
4142     DWORD new_size;
4143     struct glsl_ps_compiled_shader *new_array;
4144     struct glsl_pshader_private    *shader_data;
4145     struct ps_np2fixup_info        *np2fixup = NULL;
4146     GLhandleARB ret;
4147
4148     if (!shader->baseShader.backend_data)
4149     {
4150         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4151         if (!shader->baseShader.backend_data)
4152         {
4153             ERR("Failed to allocate backend data.\n");
4154             return 0;
4155         }
4156     }
4157     shader_data = shader->baseShader.backend_data;
4158
4159     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4160      * so a linear search is more performant than a hashmap or a binary search
4161      * (cache coherency etc)
4162      */
4163     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4164     {
4165         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4166         {
4167             if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4168             return shader_data->gl_shaders[i].prgId;
4169         }
4170     }
4171
4172     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4173     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4174         if (shader_data->num_gl_shaders)
4175         {
4176             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4177             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4178                                     new_size * sizeof(*shader_data->gl_shaders));
4179         } else {
4180             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4181             new_size = 1;
4182         }
4183
4184         if(!new_array) {
4185             ERR("Out of memory\n");
4186             return 0;
4187         }
4188         shader_data->gl_shaders = new_array;
4189         shader_data->shader_array_size = new_size;
4190     }
4191
4192     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4193
4194     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4195     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4196
4197     pixelshader_update_samplers(&shader->baseShader.reg_maps, state->textures);
4198
4199     shader_buffer_clear(buffer);
4200     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4201     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4202     *np2fixup_info = np2fixup;
4203
4204     return ret;
4205 }
4206
4207 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4208                                  const DWORD use_map) {
4209     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4210     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4211     return stored->fog_src == new->fog_src;
4212 }
4213
4214 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4215         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4216         const struct vs_compile_args *args)
4217 {
4218     UINT i;
4219     DWORD new_size;
4220     struct glsl_vs_compiled_shader *new_array;
4221     DWORD use_map = shader->baseShader.device->strided_streams.use_map;
4222     struct glsl_vshader_private *shader_data;
4223     GLhandleARB ret;
4224
4225     if (!shader->baseShader.backend_data)
4226     {
4227         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4228         if (!shader->baseShader.backend_data)
4229         {
4230             ERR("Failed to allocate backend data.\n");
4231             return 0;
4232         }
4233     }
4234     shader_data = shader->baseShader.backend_data;
4235
4236     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4237      * so a linear search is more performant than a hashmap or a binary search
4238      * (cache coherency etc)
4239      */
4240     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4241         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4242             return shader_data->gl_shaders[i].prgId;
4243         }
4244     }
4245
4246     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4247
4248     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4249         if (shader_data->num_gl_shaders)
4250         {
4251             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4252             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4253                                     new_size * sizeof(*shader_data->gl_shaders));
4254         } else {
4255             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4256             new_size = 1;
4257         }
4258
4259         if(!new_array) {
4260             ERR("Out of memory\n");
4261             return 0;
4262         }
4263         shader_data->gl_shaders = new_array;
4264         shader_data->shader_array_size = new_size;
4265     }
4266
4267     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4268
4269     shader_buffer_clear(buffer);
4270     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4271     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4272
4273     return ret;
4274 }
4275
4276 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4277  * It sets the programId on the current StateBlock (because it should be called
4278  * inside of the DrawPrimitive() part of the render loop).
4279  *
4280  * If a program for the given combination does not exist, create one, and store
4281  * the program in the hash table.  If it creates a program, it will link the
4282  * given objects, too.
4283  */
4284
4285 /* GL locking is done by the caller */
4286 static void set_glsl_shader_program(const struct wined3d_context *context,
4287         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4288 {
4289     const struct wined3d_state *state = &device->stateBlock->state;
4290     IWineD3DVertexShaderImpl *vshader = use_vs ? state->vertex_shader : NULL;
4291     IWineD3DPixelShaderImpl *pshader = use_ps ? state->pixel_shader : NULL;
4292     const struct wined3d_gl_info *gl_info = context->gl_info;
4293     struct shader_glsl_priv *priv = device->shader_priv;
4294     struct glsl_shader_prog_link *entry    = NULL;
4295     GLhandleARB programId                  = 0;
4296     GLhandleARB reorder_shader_id          = 0;
4297     unsigned int i;
4298     char glsl_name[8];
4299     struct ps_compile_args ps_compile_args;
4300     struct vs_compile_args vs_compile_args;
4301
4302     if (vshader) find_vs_compile_args(state, vshader, &vs_compile_args);
4303     if (pshader) find_ps_compile_args(state, pshader, &ps_compile_args);
4304
4305     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4306     if (entry)
4307     {
4308         priv->glsl_program = entry;
4309         return;
4310     }
4311
4312     /* If we get to this point, then no matching program exists, so we create one */
4313     programId = GL_EXTCALL(glCreateProgramObjectARB());
4314     TRACE("Created new GLSL shader program %u\n", programId);
4315
4316     /* Create the entry */
4317     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4318     entry->programId = programId;
4319     entry->vshader = vshader;
4320     entry->pshader = pshader;
4321     entry->vs_args = vs_compile_args;
4322     entry->ps_args = ps_compile_args;
4323     entry->constant_version = 0;
4324     entry->np2Fixup_info = NULL;
4325     /* Add the hash table entry */
4326     add_glsl_program_entry(priv, entry);
4327
4328     /* Set the current program */
4329     priv->glsl_program = entry;
4330
4331     /* Attach GLSL vshader */
4332     if (vshader)
4333     {
4334         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer, vshader, &vs_compile_args);
4335         WORD map = vshader->baseShader.reg_maps.input_registers;
4336         char tmp_name[10];
4337
4338         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4339         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4340         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4341         checkGLcall("glAttachObjectARB");
4342         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4343          * is destroyed
4344          */
4345         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4346
4347         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4348         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4349         checkGLcall("glAttachObjectARB");
4350
4351         /* Bind vertex attributes to a corresponding index number to match
4352          * the same index numbers as ARB_vertex_programs (makes loading
4353          * vertex attributes simpler).  With this method, we can use the
4354          * exact same code to load the attributes later for both ARB and
4355          * GLSL shaders.
4356          *
4357          * We have to do this here because we need to know the Program ID
4358          * in order to make the bindings work, and it has to be done prior
4359          * to linking the GLSL program. */
4360         for (i = 0; map; map >>= 1, ++i)
4361         {
4362             if (!(map & 1)) continue;
4363
4364             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4365             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4366         }
4367         checkGLcall("glBindAttribLocationARB");
4368
4369         list_add_head(&vshader->baseShader.linked_programs, &entry->vshader_entry);
4370     }
4371
4372     /* Attach GLSL pshader */
4373     if (pshader)
4374     {
4375         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4376                 pshader, &ps_compile_args, &entry->np2Fixup_info);
4377         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4378         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4379         checkGLcall("glAttachObjectARB");
4380
4381         list_add_head(&pshader->baseShader.linked_programs, &entry->pshader_entry);
4382     }
4383
4384     /* Link the program */
4385     TRACE("Linking GLSL shader program %u\n", programId);
4386     GL_EXTCALL(glLinkProgramARB(programId));
4387     shader_glsl_validate_link(gl_info, programId);
4388
4389     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4390             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4391     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4392     {
4393         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4394         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4395     }
4396     for (i = 0; i < MAX_CONST_I; ++i)
4397     {
4398         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4399         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4400     }
4401     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4402             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4403     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4404     {
4405         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4406         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4407     }
4408     for (i = 0; i < MAX_CONST_I; ++i)
4409     {
4410         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4411         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4412     }
4413
4414     if(pshader) {
4415         char name[32];
4416
4417         for(i = 0; i < MAX_TEXTURES; i++) {
4418             sprintf(name, "bumpenvmat%u", i);
4419             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4420             sprintf(name, "luminancescale%u", i);
4421             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4422             sprintf(name, "luminanceoffset%u", i);
4423             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4424         }
4425
4426         if (ps_compile_args.np2_fixup) {
4427             if (entry->np2Fixup_info) {
4428                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4429             } else {
4430                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4431             }
4432         }
4433     }
4434
4435     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4436     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4437     checkGLcall("Find glsl program uniform locations");
4438
4439     if (pshader
4440             && pshader->baseShader.reg_maps.shader_version.major >= 3
4441             && pshader->declared_in_count > vec4_varyings(3, gl_info))
4442     {
4443         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4444         entry->vertex_color_clamp = GL_FALSE;
4445     } else {
4446         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4447     }
4448
4449     /* Set the shader to allow uniform loading on it */
4450     GL_EXTCALL(glUseProgramObjectARB(programId));
4451     checkGLcall("glUseProgramObjectARB(programId)");
4452
4453     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4454      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4455      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4456      * vertex shader with fixed function pixel processing is used we make sure that the card
4457      * supports enough samplers to allow the max number of vertex samplers with all possible
4458      * fixed function fragment processing setups. So once the program is linked these samplers
4459      * won't change.
4460      */
4461     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4462     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4463
4464     /* If the local constants do not have to be loaded with the environment constants,
4465      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4466      * later
4467      */
4468     if (pshader && !pshader->baseShader.load_local_constsF)
4469     {
4470         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4471     }
4472     if (vshader && !vshader->baseShader.load_local_constsF)
4473     {
4474         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4475     }
4476 }
4477
4478 /* GL locking is done by the caller */
4479 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4480 {
4481     GLhandleARB program_id;
4482     GLhandleARB vshader_id, pshader_id;
4483     const char *blt_pshader;
4484
4485     static const char *blt_vshader =
4486         "#version 120\n"
4487         "void main(void)\n"
4488         "{\n"
4489         "    gl_Position = gl_Vertex;\n"
4490         "    gl_FrontColor = vec4(1.0);\n"
4491         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4492         "}\n";
4493
4494     static const char * const blt_pshaders_full[tex_type_count] =
4495     {
4496         /* tex_1d */
4497         NULL,
4498         /* tex_2d */
4499         "#version 120\n"
4500         "uniform sampler2D sampler;\n"
4501         "void main(void)\n"
4502         "{\n"
4503         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4504         "}\n",
4505         /* tex_3d */
4506         NULL,
4507         /* tex_cube */
4508         "#version 120\n"
4509         "uniform samplerCube sampler;\n"
4510         "void main(void)\n"
4511         "{\n"
4512         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4513         "}\n",
4514         /* tex_rect */
4515         "#version 120\n"
4516         "#extension GL_ARB_texture_rectangle : enable\n"
4517         "uniform sampler2DRect sampler;\n"
4518         "void main(void)\n"
4519         "{\n"
4520         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4521         "}\n",
4522     };
4523
4524     static const char * const blt_pshaders_masked[tex_type_count] =
4525     {
4526         /* tex_1d */
4527         NULL,
4528         /* tex_2d */
4529         "#version 120\n"
4530         "uniform sampler2D sampler;\n"
4531         "uniform vec4 mask;\n"
4532         "void main(void)\n"
4533         "{\n"
4534         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4535         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4536         "}\n",
4537         /* tex_3d */
4538         NULL,
4539         /* tex_cube */
4540         "#version 120\n"
4541         "uniform samplerCube sampler;\n"
4542         "uniform vec4 mask;\n"
4543         "void main(void)\n"
4544         "{\n"
4545         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4546         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4547         "}\n",
4548         /* tex_rect */
4549         "#version 120\n"
4550         "#extension GL_ARB_texture_rectangle : enable\n"
4551         "uniform sampler2DRect sampler;\n"
4552         "uniform vec4 mask;\n"
4553         "void main(void)\n"
4554         "{\n"
4555         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4556         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4557         "}\n",
4558     };
4559
4560     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4561     if (!blt_pshader)
4562     {
4563         FIXME("tex_type %#x not supported\n", tex_type);
4564         return 0;
4565     }
4566
4567     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4568     shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4569
4570     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4571     shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4572
4573     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4574     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4575     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4576     GL_EXTCALL(glLinkProgramARB(program_id));
4577
4578     shader_glsl_validate_link(gl_info, program_id);
4579
4580     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4581      * is destroyed
4582      */
4583     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4584     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4585     return program_id;
4586 }
4587
4588 /* GL locking is done by the caller */
4589 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4590 {
4591     const struct wined3d_gl_info *gl_info = context->gl_info;
4592     IWineD3DDeviceImpl *device = context->swapchain->device;
4593     struct shader_glsl_priv *priv = device->shader_priv;
4594     GLhandleARB program_id = 0;
4595     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4596
4597     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4598
4599     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4600     else priv->glsl_program = NULL;
4601
4602     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4603
4604     if (old_vertex_color_clamp != current_vertex_color_clamp)
4605     {
4606         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4607         {
4608             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4609             checkGLcall("glClampColorARB");
4610         }
4611         else
4612         {
4613             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4614         }
4615     }
4616
4617     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4618     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4619     GL_EXTCALL(glUseProgramObjectARB(program_id));
4620     checkGLcall("glUseProgramObjectARB");
4621
4622     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4623      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4624      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4625     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4626     {
4627         shader_glsl_load_np2fixup_constants(priv, gl_info, &device->stateBlock->state);
4628     }
4629 }
4630
4631 /* GL locking is done by the caller */
4632 static void shader_glsl_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4633         enum tex_types tex_type, const SIZE *ds_mask_size)
4634 {
4635     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4636     struct shader_glsl_priv *priv = shader_priv;
4637     GLhandleARB *blt_program;
4638     GLint loc;
4639
4640     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4641     if (!*blt_program)
4642     {
4643         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4644         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4645         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4646         GL_EXTCALL(glUniform1iARB(loc, 0));
4647     }
4648     else
4649     {
4650         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4651     }
4652
4653     if (masked)
4654     {
4655         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4656         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4657     }
4658 }
4659
4660 /* GL locking is done by the caller */
4661 static void shader_glsl_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4662 {
4663     struct shader_glsl_priv *priv = shader_priv;
4664     GLhandleARB program_id;
4665
4666     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4667     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4668
4669     GL_EXTCALL(glUseProgramObjectARB(program_id));
4670     checkGLcall("glUseProgramObjectARB");
4671 }
4672
4673 static void shader_glsl_destroy(IWineD3DBaseShaderImpl *shader)
4674 {
4675     const struct list *linked_programs;
4676     IWineD3DDeviceImpl *device = shader->baseShader.device;
4677     struct shader_glsl_priv *priv = device->shader_priv;
4678     const struct wined3d_gl_info *gl_info;
4679     struct wined3d_context *context;
4680
4681     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4682      * can be called from IWineD3DBaseShader::Release
4683      */
4684     char pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
4685
4686     if (pshader)
4687     {
4688         struct glsl_pshader_private *shader_data = shader->baseShader.backend_data;
4689
4690         if (!shader_data || !shader_data->num_gl_shaders)
4691         {
4692             HeapFree(GetProcessHeap(), 0, shader_data);
4693             shader->baseShader.backend_data = NULL;
4694             return;
4695         }
4696
4697         context = context_acquire(device, NULL);
4698         gl_info = context->gl_info;
4699
4700         if (priv->glsl_program && (IWineD3DBaseShaderImpl *)priv->glsl_program->pshader == shader)
4701         {
4702             ENTER_GL();
4703             shader_glsl_select(context, FALSE, FALSE);
4704             LEAVE_GL();
4705         }
4706     }
4707     else
4708     {
4709         struct glsl_vshader_private *shader_data = shader->baseShader.backend_data;
4710
4711         if (!shader_data || !shader_data->num_gl_shaders)
4712         {
4713             HeapFree(GetProcessHeap(), 0, shader_data);
4714             shader->baseShader.backend_data = NULL;
4715             return;
4716         }
4717
4718         context = context_acquire(device, NULL);
4719         gl_info = context->gl_info;
4720
4721         if (priv->glsl_program && (IWineD3DBaseShaderImpl *)priv->glsl_program->vshader == shader)
4722         {
4723             ENTER_GL();
4724             shader_glsl_select(context, FALSE, FALSE);
4725             LEAVE_GL();
4726         }
4727     }
4728
4729     linked_programs = &shader->baseShader.linked_programs;
4730
4731     TRACE("Deleting linked programs\n");
4732     if (linked_programs->next) {
4733         struct glsl_shader_prog_link *entry, *entry2;
4734
4735         ENTER_GL();
4736         if(pshader) {
4737             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4738                 delete_glsl_program_entry(priv, gl_info, entry);
4739             }
4740         } else {
4741             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4742                 delete_glsl_program_entry(priv, gl_info, entry);
4743             }
4744         }
4745         LEAVE_GL();
4746     }
4747
4748     if (pshader)
4749     {
4750         struct glsl_pshader_private *shader_data = shader->baseShader.backend_data;
4751         UINT i;
4752
4753         ENTER_GL();
4754         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4755             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4756             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4757             checkGLcall("glDeleteObjectARB");
4758         }
4759         LEAVE_GL();
4760         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4761     }
4762     else
4763     {
4764         struct glsl_vshader_private *shader_data = shader->baseShader.backend_data;
4765         UINT i;
4766
4767         ENTER_GL();
4768         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4769             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4770             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4771             checkGLcall("glDeleteObjectARB");
4772         }
4773         LEAVE_GL();
4774         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4775     }
4776
4777     HeapFree(GetProcessHeap(), 0, shader->baseShader.backend_data);
4778     shader->baseShader.backend_data = NULL;
4779
4780     context_release(context);
4781 }
4782
4783 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4784 {
4785     const glsl_program_key_t *k = key;
4786     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4787             const struct glsl_shader_prog_link, program_lookup_entry);
4788     int cmp;
4789
4790     if (k->vshader > prog->vshader) return 1;
4791     else if (k->vshader < prog->vshader) return -1;
4792
4793     if (k->pshader > prog->pshader) return 1;
4794     else if (k->pshader < prog->pshader) return -1;
4795
4796     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4797     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4798
4799     return 0;
4800 }
4801
4802 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4803 {
4804     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4805     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4806
4807     if (!mem)
4808     {
4809         ERR("Failed to allocate memory\n");
4810         return FALSE;
4811     }
4812
4813     heap->entries = mem;
4814     heap->entries[1].version = 0;
4815     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4816     heap->size = 1;
4817
4818     return TRUE;
4819 }
4820
4821 static void constant_heap_free(struct constant_heap *heap)
4822 {
4823     HeapFree(GetProcessHeap(), 0, heap->entries);
4824 }
4825
4826 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4827 {
4828     wined3d_rb_alloc,
4829     wined3d_rb_realloc,
4830     wined3d_rb_free,
4831     glsl_program_key_compare,
4832 };
4833
4834 static HRESULT shader_glsl_alloc(IWineD3DDeviceImpl *device)
4835 {
4836     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4837     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4838     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4839             gl_info->limits.glsl_ps_float_constants)) + 1;
4840
4841     if (!shader_buffer_init(&priv->shader_buffer))
4842     {
4843         ERR("Failed to initialize shader buffer.\n");
4844         goto fail;
4845     }
4846
4847     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4848     if (!priv->stack)
4849     {
4850         ERR("Failed to allocate memory.\n");
4851         goto fail;
4852     }
4853
4854     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4855     {
4856         ERR("Failed to initialize vertex shader constant heap\n");
4857         goto fail;
4858     }
4859
4860     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4861     {
4862         ERR("Failed to initialize pixel shader constant heap\n");
4863         goto fail;
4864     }
4865
4866     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4867     {
4868         ERR("Failed to initialize rbtree.\n");
4869         goto fail;
4870     }
4871
4872     priv->next_constant_version = 1;
4873
4874     device->shader_priv = priv;
4875     return WINED3D_OK;
4876
4877 fail:
4878     constant_heap_free(&priv->pconst_heap);
4879     constant_heap_free(&priv->vconst_heap);
4880     HeapFree(GetProcessHeap(), 0, priv->stack);
4881     shader_buffer_free(&priv->shader_buffer);
4882     HeapFree(GetProcessHeap(), 0, priv);
4883     return E_OUTOFMEMORY;
4884 }
4885
4886 /* Context activation is done by the caller. */
4887 static void shader_glsl_free(IWineD3DDeviceImpl *device)
4888 {
4889     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4890     struct shader_glsl_priv *priv = device->shader_priv;
4891     int i;
4892
4893     ENTER_GL();
4894     for (i = 0; i < tex_type_count; ++i)
4895     {
4896         if (priv->depth_blt_program_full[i])
4897         {
4898             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4899         }
4900         if (priv->depth_blt_program_masked[i])
4901         {
4902             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4903         }
4904     }
4905     LEAVE_GL();
4906
4907     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4908     constant_heap_free(&priv->pconst_heap);
4909     constant_heap_free(&priv->vconst_heap);
4910     HeapFree(GetProcessHeap(), 0, priv->stack);
4911     shader_buffer_free(&priv->shader_buffer);
4912
4913     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4914     device->shader_priv = NULL;
4915 }
4916
4917 static BOOL shader_glsl_dirty_const(void)
4918 {
4919     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4920     return FALSE;
4921 }
4922
4923 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4924 {
4925     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4926      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4927      * on the version of NV_vertex_program.
4928      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4929      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4930      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4931      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4932      */
4933     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4934             || gl_info->limits.arb_ps_instructions <= 512)
4935         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4936     else
4937         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4938     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4939     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4940
4941     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4942      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4943      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4944      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4945      * in max native instructions. Intel and others also offer the info in this extension but they
4946      * don't support GLSL (at least on Windows).
4947      *
4948      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4949      * of instructions is 512 or less we have to do with ps2.0 hardware.
4950      * 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.
4951      */
4952     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4953             || gl_info->limits.arb_ps_instructions <= 512)
4954         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4955     else
4956         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4957
4958     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4959
4960     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4961      * Direct3D minimum requirement.
4962      *
4963      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4964      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4965      *
4966      * The problem is that the refrast clamps temporary results in the shader to
4967      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4968      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4969      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4970      * offer a way to query this.
4971      */
4972     pCaps->PixelShader1xMaxValue = 8.0;
4973     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4974
4975     pCaps->VSClipping = TRUE;
4976 }
4977
4978 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4979 {
4980     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4981     {
4982         TRACE("Checking support for fixup:\n");
4983         dump_color_fixup_desc(fixup);
4984     }
4985
4986     /* We support everything except YUV conversions. */
4987     if (!is_complex_fixup(fixup))
4988     {
4989         TRACE("[OK]\n");
4990         return TRUE;
4991     }
4992
4993     TRACE("[FAILED]\n");
4994     return FALSE;
4995 }
4996
4997 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4998 {
4999     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
5000     /* WINED3DSIH_ADD           */ shader_glsl_arith,
5001     /* WINED3DSIH_AND           */ NULL,
5002     /* WINED3DSIH_BEM           */ shader_glsl_bem,
5003     /* WINED3DSIH_BREAK         */ shader_glsl_break,
5004     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
5005     /* WINED3DSIH_BREAKP        */ NULL,
5006     /* WINED3DSIH_CALL          */ shader_glsl_call,
5007     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
5008     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
5009     /* WINED3DSIH_CND           */ shader_glsl_cnd,
5010     /* WINED3DSIH_CRS           */ shader_glsl_cross,
5011     /* WINED3DSIH_CUT           */ NULL,
5012     /* WINED3DSIH_DCL           */ NULL,
5013     /* WINED3DSIH_DEF           */ NULL,
5014     /* WINED3DSIH_DEFB          */ NULL,
5015     /* WINED3DSIH_DEFI          */ NULL,
5016     /* WINED3DSIH_DIV           */ NULL,
5017     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
5018     /* WINED3DSIH_DP3           */ shader_glsl_dot,
5019     /* WINED3DSIH_DP4           */ shader_glsl_dot,
5020     /* WINED3DSIH_DST           */ shader_glsl_dst,
5021     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5022     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5023     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5024     /* WINED3DSIH_EMIT          */ NULL,
5025     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5026     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5027     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5028     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5029     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5030     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5031     /* WINED3DSIH_FTOI          */ NULL,
5032     /* WINED3DSIH_IADD          */ NULL,
5033     /* WINED3DSIH_IEQ           */ NULL,
5034     /* WINED3DSIH_IF            */ shader_glsl_if,
5035     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5036     /* WINED3DSIH_IGE           */ NULL,
5037     /* WINED3DSIH_IMUL          */ NULL,
5038     /* WINED3DSIH_ITOF          */ NULL,
5039     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5040     /* WINED3DSIH_LD            */ NULL,
5041     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5042     /* WINED3DSIH_LOG           */ shader_glsl_log,
5043     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5044     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5045     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5046     /* WINED3DSIH_LT            */ NULL,
5047     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5048     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5049     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5050     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5051     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5052     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5053     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5054     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5055     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5056     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5057     /* WINED3DSIH_MOVC          */ NULL,
5058     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5059     /* WINED3DSIH_NOP           */ NULL,
5060     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5061     /* WINED3DSIH_PHASE         */ NULL,
5062     /* WINED3DSIH_POW           */ shader_glsl_pow,
5063     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5064     /* WINED3DSIH_REP           */ shader_glsl_rep,
5065     /* WINED3DSIH_RET           */ shader_glsl_ret,
5066     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5067     /* WINED3DSIH_SAMPLE        */ NULL,
5068     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5069     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5070     /* WINED3DSIH_SETP          */ NULL,
5071     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5072     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5073     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5074     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5075     /* WINED3DSIH_SQRT          */ NULL,
5076     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5077     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5078     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5079     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5080     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5081     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5082     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5083     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5084     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5085     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5086     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5087     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5088     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5089     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5090     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5091     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5092     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5093     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5094     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5095     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5096     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5097     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5098     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5099     /* WINED3DSIH_UTOF          */ NULL,
5100 };
5101
5102 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5103     SHADER_HANDLER hw_fct;
5104
5105     /* Select handler */
5106     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5107
5108     /* Unhandled opcode */
5109     if (!hw_fct)
5110     {
5111         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5112         return;
5113     }
5114     hw_fct(ins);
5115
5116     shader_glsl_add_instruction_modifiers(ins);
5117 }
5118
5119 const shader_backend_t glsl_shader_backend = {
5120     shader_glsl_handle_instruction,
5121     shader_glsl_select,
5122     shader_glsl_select_depth_blt,
5123     shader_glsl_deselect_depth_blt,
5124     shader_glsl_update_float_vertex_constants,
5125     shader_glsl_update_float_pixel_constants,
5126     shader_glsl_load_constants,
5127     shader_glsl_load_np2fixup_constants,
5128     shader_glsl_destroy,
5129     shader_glsl_alloc,
5130     shader_glsl_free,
5131     shader_glsl_dirty_const,
5132     shader_glsl_get_caps,
5133     shader_glsl_color_fixup_supported,
5134 };