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

   1 /*
   2  * GLSL pixel and vertex shader implementation
   3  *
   4  * Copyright 2006 Jason Green
   5  * Copyright 2006-2007 Henri Verbeet
   6  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
   7  * Copyright 2009 Henri Verbeet for CodeWeavers
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  22  */
  23
  24 /*
  25  * D3D shader asm has swizzles on source parameters, and write masks for
  26  * destination parameters. GLSL uses swizzles for both. The result of this is
  27  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  28  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  29  * mask for the destination parameter into account.
  30  */
  31
  32 #include "config.h"
  33 #include <limits.h>
  34 #include <stdio.h>
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  43 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  44 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  45 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  46
  47 typedef struct {
  48     char reg_name[150];
  49     char mask_str[6];
  50 } glsl_dst_param_t;
  51
  52 typedef struct {
  53     char reg_name[150];
  54     char param_str[200];
  55 } glsl_src_param_t;
  56
  57 typedef struct {
  58     const char *name;
  59     DWORD coord_mask;
  60 } glsl_sample_function_t;
  61
  62 enum heap_node_op
  63 {
  64     HEAP_NODE_TRAVERSE_LEFT,
  65     HEAP_NODE_TRAVERSE_RIGHT,
  66     HEAP_NODE_POP,
  67 };
  68
  69 struct constant_entry
  70 {
  71     unsigned int idx;
  72     unsigned int version;
  73 };
  74
  75 struct constant_heap
  76 {
  77     struct constant_entry *entries;
  78     unsigned int *positions;
  79     unsigned int size;
  80 };
  81
  82 /* GLSL shader private data */
  83 struct shader_glsl_priv {
  84     struct wined3d_shader_buffer shader_buffer;
  85     struct wine_rb_tree program_lookup;
  86     struct glsl_shader_prog_link *glsl_program;
  87     struct constant_heap vconst_heap;
  88     struct constant_heap pconst_heap;
  89     unsigned char *stack;
  90     GLhandleARB depth_blt_program_full[tex_type_count];
  91     GLhandleARB depth_blt_program_masked[tex_type_count];
  92     UINT next_constant_version;
  93 };
  94
  95 /* Struct to maintain data about a linked GLSL program */
  96 struct glsl_shader_prog_link {
  97     struct wine_rb_entry        program_lookup_entry;
  98     struct list                 vshader_entry;
  99     struct list                 pshader_entry;
 100     GLhandleARB                 programId;
 101     GLint                       *vuniformF_locations;
 102     GLint                       *puniformF_locations;
 103     GLint                       vuniformI_locations[MAX_CONST_I];
 104     GLint                       puniformI_locations[MAX_CONST_I];
 105     GLint                       posFixup_location;
 106     GLint                       np2Fixup_location;
 107     GLint                       bumpenvmat_location[MAX_TEXTURES];
 108     GLint                       luminancescale_location[MAX_TEXTURES];
 109     GLint                       luminanceoffset_location[MAX_TEXTURES];
 110     GLint                       ycorrection_location;
 111     GLenum                      vertex_color_clamp;
 112     IWineD3DVertexShader        *vshader;
 113     IWineD3DPixelShader         *pshader;
 114     struct vs_compile_args      vs_args;
 115     struct ps_compile_args      ps_args;
 116     UINT                        constant_version;
 117     const struct ps_np2fixup_info *np2Fixup_info;
 118 };
 119
 120 typedef struct {
 121     IWineD3DVertexShader        *vshader;
 122     IWineD3DPixelShader         *pshader;
 123     struct ps_compile_args      ps_args;
 124     struct vs_compile_args      vs_args;
 125 } glsl_program_key_t;
 126
 127 struct shader_glsl_ctx_priv {
 128     const struct vs_compile_args    *cur_vs_args;
 129     const struct ps_compile_args    *cur_ps_args;
 130     struct ps_np2fixup_info         *cur_np2fixup_info;
 131 };
 132
 133 struct glsl_ps_compiled_shader
 134 {
 135     struct ps_compile_args          args;
 136     struct ps_np2fixup_info         np2fixup;
 137     GLhandleARB                     prgId;
 138 };
 139
 140 struct glsl_pshader_private
 141 {
 142     struct glsl_ps_compiled_shader  *gl_shaders;
 143     UINT                            num_gl_shaders, shader_array_size;
 144 };
 145
 146 struct glsl_vs_compiled_shader
 147 {
 148     struct vs_compile_args          args;
 149     GLhandleARB                     prgId;
 150 };
 151
 152 struct glsl_vshader_private
 153 {
 154     struct glsl_vs_compiled_shader  *gl_shaders;
 155     UINT                            num_gl_shaders, shader_array_size;
 156 };
 157
 158 static const char *debug_gl_shader_type(GLenum type)
 159 {
 160     switch (type)
 161     {
 162 #define WINED3D_TO_STR(u) case u: return #u
 163         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
 164         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
 165         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
 166 #undef WINED3D_TO_STR
 167         default:
 168             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
 169     }
 170 }
 171
 172 /* Extract a line from the info log.
 173  * Note that this modifies the source string. */
 174 static char *get_info_log_line(char **ptr)
 175 {
 176     char *p, *q;
 177
 178     p = *ptr;
 179     if (!(q = strstr(p, "\n")))
 180     {
 181         if (!*p) return NULL;
 182         *ptr += strlen(p);
 183         return p;
 184     }
 185     *q = '\0';
 186     *ptr = q + 1;
 187
 188     return p;
 189 }
 190
 191 /** Prints the GLSL info log which will contain error messages if they exist */
 192 /* GL locking is done by the caller */
 193 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 194 {
 195     int infologLength = 0;
 196     char *infoLog;
 197     unsigned int i;
 198     BOOL is_spam;
 199
 200     static const char * const spam[] =
 201     {
 202         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 203         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx, with \n */
 204         "Fragment shader was successfully compiled to run on hardware.",    /* fglrx, no \n   */
 205         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 206         "Fragment shader(s) linked, vertex shader(s) linked. \n",           /* fglrx, with \n */
 207         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 208         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 209         "Vertex shader(s) linked, no fragment shader(s) defined. \n",       /* fglrx, with \n */
 210         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 211         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 212         "Fragment shader(s) linked, no vertex shader(s) defined. \n",       /* fglrx, with \n */
 213         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 214     };
 215
 216     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 217
 218     GL_EXTCALL(glGetObjectParameterivARB(obj,
 219                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 220                &infologLength));
 221
 222     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 223      * that if there are errors. */
 224     if (infologLength > 1)
 225     {
 226         char *ptr, *line;
 227
 228         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 229          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 230          */
 231         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 232         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 233         is_spam = FALSE;
 234
 235         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 236             if(strcmp(infoLog, spam[i]) == 0) {
 237                 is_spam = TRUE;
 238                 break;
 239             }
 240         }
 241
 242         ptr = infoLog;
 243         if (is_spam)
 244         {
 245             TRACE("Spam received from GLSL shader #%u:\n", obj);
 246             while ((line = get_info_log_line(&ptr))) TRACE("    %s\n", line);
 247         }
 248         else
 249         {
 250             FIXME("Error received from GLSL shader #%u:\n", obj);
 251             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 252         }
 253         HeapFree(GetProcessHeap(), 0, infoLog);
 254     }
 255 }
 256
 257 /* GL locking is done by the caller. */
 258 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 259 {
 260     GLint i, object_count, source_size = -1;
 261     GLhandleARB *objects;
 262     char *source = NULL;
 263
 264     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
 265     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
 266     if (!objects)
 267     {
 268         ERR("Failed to allocate object array memory.\n");
 269         return;
 270     }
 271
 272     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
 273     for (i = 0; i < object_count; ++i)
 274     {
 275         char *ptr, *line;
 276         GLint tmp;
 277
 278         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
 279
 280         if (source_size < tmp)
 281         {
 282             HeapFree(GetProcessHeap(), 0, source);
 283
 284             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
 285             if (!source)
 286             {
 287                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
 288                 HeapFree(GetProcessHeap(), 0, objects);
 289                 return;
 290             }
 291             source_size = tmp;
 292         }
 293
 294         FIXME("Object %u:\n", objects[i]);
 295         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
 296         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
 297         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
 298         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
 299         FIXME("\n");
 300
 301         ptr = source;
 302         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
 303         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 304         FIXME("\n");
 305     }
 306
 307     HeapFree(GetProcessHeap(), 0, source);
 308     HeapFree(GetProcessHeap(), 0, objects);
 309 }
 310
 311 /* GL locking is done by the caller. */
 312 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 313 {
 314     GLint tmp;
 315
 316     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 317
 318     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
 319     if (tmp == GL_PROGRAM_OBJECT_ARB)
 320     {
 321         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
 322         if (!tmp)
 323         {
 324             FIXME("Program %u link status invalid.\n", program);
 325             shader_glsl_dump_program_source(gl_info, program);
 326         }
 327     }
 328
 329     print_glsl_info_log(gl_info, program);
 330 }
 331
 332 /**
 333  * Loads (pixel shader) samplers
 334  */
 335 /* GL locking is done by the caller */
 336 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 337         DWORD *tex_unit_map, GLhandleARB programId)
 338 {
 339     GLint name_loc;
 340     int i;
 341     char sampler_name[20];
 342
 343     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 344         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 345         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 346         if (name_loc != -1) {
 347             DWORD mapped_unit = tex_unit_map[i];
 348             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 349             {
 350                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 351                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 352                 checkGLcall("glUniform1iARB");
 353             } else {
 354                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 355             }
 356         }
 357     }
 358 }
 359
 360 /* GL locking is done by the caller */
 361 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 362         DWORD *tex_unit_map, GLhandleARB programId)
 363 {
 364     GLint name_loc;
 365     char sampler_name[20];
 366     int i;
 367
 368     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 369         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 370         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 371         if (name_loc != -1) {
 372             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 373             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 374             {
 375                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 376                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 377                 checkGLcall("glUniform1iARB");
 378             } else {
 379                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 380             }
 381         }
 382     }
 383 }
 384
 385 /* GL locking is done by the caller */
 386 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 387         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 388 {
 389     int stack_idx = 0;
 390     unsigned int heap_idx = 1;
 391     unsigned int idx;
 392
 393     if (heap->entries[heap_idx].version <= version) return;
 394
 395     idx = heap->entries[heap_idx].idx;
 396     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 397     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 398
 399     while (stack_idx >= 0)
 400     {
 401         /* Note that we fall through to the next case statement. */
 402         switch(stack[stack_idx])
 403         {
 404             case HEAP_NODE_TRAVERSE_LEFT:
 405             {
 406                 unsigned int left_idx = heap_idx << 1;
 407                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 408                 {
 409                     heap_idx = left_idx;
 410                     idx = heap->entries[heap_idx].idx;
 411                     if (constant_locations[idx] != -1)
 412                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 413
 414                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 415                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 416                     break;
 417                 }
 418             }
 419
 420             case HEAP_NODE_TRAVERSE_RIGHT:
 421             {
 422                 unsigned int right_idx = (heap_idx << 1) + 1;
 423                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 424                 {
 425                     heap_idx = right_idx;
 426                     idx = heap->entries[heap_idx].idx;
 427                     if (constant_locations[idx] != -1)
 428                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 429
 430                     stack[stack_idx++] = HEAP_NODE_POP;
 431                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 432                     break;
 433                 }
 434             }
 435
 436             case HEAP_NODE_POP:
 437             {
 438                 heap_idx >>= 1;
 439                 --stack_idx;
 440                 break;
 441             }
 442         }
 443     }
 444     checkGLcall("walk_constant_heap()");
 445 }
 446
 447 /* GL locking is done by the caller */
 448 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 449 {
 450     GLfloat clamped_constant[4];
 451
 452     if (location == -1) return;
 453
 454     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 455     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 456     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 457     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 458
 459     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 460 }
 461
 462 /* GL locking is done by the caller */
 463 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 464         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 465 {
 466     int stack_idx = 0;
 467     unsigned int heap_idx = 1;
 468     unsigned int idx;
 469
 470     if (heap->entries[heap_idx].version <= version) return;
 471
 472     idx = heap->entries[heap_idx].idx;
 473     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 474     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 475
 476     while (stack_idx >= 0)
 477     {
 478         /* Note that we fall through to the next case statement. */
 479         switch(stack[stack_idx])
 480         {
 481             case HEAP_NODE_TRAVERSE_LEFT:
 482             {
 483                 unsigned int left_idx = heap_idx << 1;
 484                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 485                 {
 486                     heap_idx = left_idx;
 487                     idx = heap->entries[heap_idx].idx;
 488                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 489
 490                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 491                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 492                     break;
 493                 }
 494             }
 495
 496             case HEAP_NODE_TRAVERSE_RIGHT:
 497             {
 498                 unsigned int right_idx = (heap_idx << 1) + 1;
 499                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 500                 {
 501                     heap_idx = right_idx;
 502                     idx = heap->entries[heap_idx].idx;
 503                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 504
 505                     stack[stack_idx++] = HEAP_NODE_POP;
 506                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 507                     break;
 508                 }
 509             }
 510
 511             case HEAP_NODE_POP:
 512             {
 513                 heap_idx >>= 1;
 514                 --stack_idx;
 515                 break;
 516             }
 517         }
 518     }
 519     checkGLcall("walk_constant_heap_clamped()");
 520 }
 521
 522 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 523 /* GL locking is done by the caller */
 524 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 525         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 526         unsigned char *stack, UINT version)
 527 {
 528     const local_constant *lconst;
 529
 530     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 531     if (This->baseShader.reg_maps.shader_version.major == 1
 532             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
 533         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 534     else
 535         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 536
 537     if (!This->baseShader.load_local_constsF)
 538     {
 539         TRACE("No need to load local float constants for this shader\n");
 540         return;
 541     }
 542
 543     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 544     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 545     {
 546         GLint location = constant_locations[lconst->idx];
 547         /* We found this uniform name in the program - go ahead and send the data */
 548         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 549     }
 550     checkGLcall("glUniform4fvARB()");
 551 }
 552
 553 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 554 /* GL locking is done by the caller */
 555 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 556         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 557 {
 558     unsigned int i;
 559     struct list* ptr;
 560
 561     for (i = 0; constants_set; constants_set >>= 1, ++i)
 562     {
 563         if (!(constants_set & 1)) continue;
 564
 565         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 566                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 567
 568         /* We found this uniform name in the program - go ahead and send the data */
 569         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 570         checkGLcall("glUniform4ivARB");
 571     }
 572
 573     /* Load immediate constants */
 574     ptr = list_head(&This->baseShader.constantsI);
 575     while (ptr) {
 576         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 577         unsigned int idx = lconst->idx;
 578         const GLint *values = (const GLint *)lconst->value;
 579
 580         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 581             values[0], values[1], values[2], values[3]);
 582
 583         /* We found this uniform name in the program - go ahead and send the data */
 584         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 585         checkGLcall("glUniform4ivARB");
 586         ptr = list_next(&This->baseShader.constantsI, ptr);
 587     }
 588 }
 589
 590 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 591 /* GL locking is done by the caller */
 592 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 593         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 594 {
 595     GLint tmp_loc;
 596     unsigned int i;
 597     char tmp_name[8];
 598     const char *prefix;
 599     struct list* ptr;
 600
 601     switch (This->baseShader.reg_maps.shader_version.type)
 602     {
 603         case WINED3D_SHADER_TYPE_VERTEX:
 604             prefix = "VB";
 605             break;
 606
 607         case WINED3D_SHADER_TYPE_GEOMETRY:
 608             prefix = "GB";
 609             break;
 610
 611         case WINED3D_SHADER_TYPE_PIXEL:
 612             prefix = "PB";
 613             break;
 614
 615         default:
 616             FIXME("Unknown shader type %#x.\n",
 617                     This->baseShader.reg_maps.shader_version.type);
 618             prefix = "UB";
 619             break;
 620     }
 621
 622     /* TODO: Benchmark and see if it would be beneficial to store the
 623      * locations of the constants to avoid looking up each time */
 624     for (i = 0; constants_set; constants_set >>= 1, ++i)
 625     {
 626         if (!(constants_set & 1)) continue;
 627
 628         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 629
 630         /* TODO: Benchmark and see if it would be beneficial to store the
 631          * locations of the constants to avoid looking up each time */
 632         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 633         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 634         if (tmp_loc != -1)
 635         {
 636             /* We found this uniform name in the program - go ahead and send the data */
 637             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 638             checkGLcall("glUniform1ivARB");
 639         }
 640     }
 641
 642     /* Load immediate constants */
 643     ptr = list_head(&This->baseShader.constantsB);
 644     while (ptr) {
 645         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 646         unsigned int idx = lconst->idx;
 647         const GLint *values = (const GLint *)lconst->value;
 648
 649         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 650
 651         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 652         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 653         if (tmp_loc != -1) {
 654             /* We found this uniform name in the program - go ahead and send the data */
 655             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 656             checkGLcall("glUniform1ivARB");
 657         }
 658         ptr = list_next(&This->baseShader.constantsB, ptr);
 659     }
 660 }
 661
 662 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 663 {
 664     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 665 }
 666
 667 /**
 668  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 669  */
 670 /* GL locking is done by the caller (state handler) */
 671 static void shader_glsl_load_np2fixup_constants(
 672     IWineD3DDevice* device,
 673     char usePixelShader,
 674     char useVertexShader) {
 675
 676     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
 677     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
 678
 679     if (!prog) {
 680         /* No GLSL program set - nothing to do. */
 681         return;
 682     }
 683
 684     if (!usePixelShader) {
 685         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 686         return;
 687     }
 688
 689     if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
 690         const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 691         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
 692         UINT i;
 693         UINT fixup = prog->ps_args.np2_fixup;
 694         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 695
 696         for (i = 0; fixup; fixup >>= 1, ++i) {
 697             const unsigned char idx = prog->np2Fixup_info->idx[i];
 698             const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
 699             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 700
 701             if (!tex)
 702             {
 703                 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
 704                 continue;
 705             }
 706
 707             if (idx % 2) {
 708                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 709             } else {
 710                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 711             }
 712         }
 713
 714         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 715     }
 716 }
 717
 718 /**
 719  * Loads the app-supplied constants into the currently set GLSL program.
 720  */
 721 /* GL locking is done by the caller (state handler) */
 722 static void shader_glsl_load_constants(const struct wined3d_context *context,
 723         char usePixelShader, char useVertexShader)
 724 {
 725     const struct wined3d_gl_info *gl_info = context->gl_info;
 726     IWineD3DDeviceImpl *device = context->swapchain->device;
 727     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 728     struct shader_glsl_priv *priv = device->shader_priv;
 729
 730     GLhandleARB programId;
 731     struct glsl_shader_prog_link *prog = priv->glsl_program;
 732     UINT constant_version;
 733     int i;
 734
 735     if (!prog) {
 736         /* No GLSL program set - nothing to do. */
 737         return;
 738     }
 739     programId = prog->programId;
 740     constant_version = prog->constant_version;
 741
 742     if (useVertexShader) {
 743         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 744
 745         /* Load DirectX 9 float constants/uniforms for vertex shader */
 746         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
 747                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 748
 749         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 750         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
 751                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 752
 753         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 754         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
 755                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 756
 757         /* Upload the position fixup params */
 758         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
 759         checkGLcall("glUniform4fvARB");
 760     }
 761
 762     if (usePixelShader) {
 763
 764         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 765
 766         /* Load DirectX 9 float constants/uniforms for pixel shader */
 767         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
 768                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 769
 770         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 771         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
 772                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 773
 774         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 775         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
 776                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 777
 778         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 779          * It can't be 0 for a valid texbem instruction.
 780          */
 781         for(i = 0; i < MAX_TEXTURES; i++) {
 782             const float *data;
 783
 784             if(prog->bumpenvmat_location[i] == -1) continue;
 785
 786             data = (const float *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVMAT00];
 787             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 788             checkGLcall("glUniformMatrix2fvARB");
 789
 790             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 791              * is set too, so we can check that in the needsbumpmat check
 792              */
 793             if(prog->luminancescale_location[i] != -1) {
 794                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLSCALE];
 795                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLOFFSET];
 796
 797                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 798                 checkGLcall("glUniform1fvARB");
 799                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 800                 checkGLcall("glUniform1fvARB");
 801             }
 802         }
 803
 804         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 805             float correction_params[4];
 806
 807             if (context->render_offscreen)
 808             {
 809                 correction_params[0] = 0.0f;
 810                 correction_params[1] = 1.0f;
 811             } else {
 812                 /* position is window relative, not viewport relative */
 813                 correction_params[0] = context->current_rt->currentDesc.Height;
 814                 correction_params[1] = -1.0f;
 815             }
 816             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 817         }
 818     }
 819
 820     if (priv->next_constant_version == UINT_MAX)
 821     {
 822         TRACE("Max constant version reached, resetting to 0.\n");
 823         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 824         priv->next_constant_version = 1;
 825     }
 826     else
 827     {
 828         prog->constant_version = priv->next_constant_version++;
 829     }
 830 }
 831
 832 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 833         unsigned int heap_idx, DWORD new_version)
 834 {
 835     struct constant_entry *entries = heap->entries;
 836     unsigned int *positions = heap->positions;
 837     unsigned int parent_idx;
 838
 839     while (heap_idx > 1)
 840     {
 841         parent_idx = heap_idx >> 1;
 842
 843         if (new_version <= entries[parent_idx].version) break;
 844
 845         entries[heap_idx] = entries[parent_idx];
 846         positions[entries[parent_idx].idx] = heap_idx;
 847         heap_idx = parent_idx;
 848     }
 849
 850     entries[heap_idx].version = new_version;
 851     entries[heap_idx].idx = idx;
 852     positions[idx] = heap_idx;
 853 }
 854
 855 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 856 {
 857     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 858     struct shader_glsl_priv *priv = This->shader_priv;
 859     struct constant_heap *heap = &priv->vconst_heap;
 860     UINT i;
 861
 862     for (i = start; i < count + start; ++i)
 863     {
 864         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
 865             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 866         else
 867             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 868     }
 869 }
 870
 871 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 872 {
 873     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 874     struct shader_glsl_priv *priv = This->shader_priv;
 875     struct constant_heap *heap = &priv->pconst_heap;
 876     UINT i;
 877
 878     for (i = start; i < count + start; ++i)
 879     {
 880         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 881             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 882         else
 883             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 884     }
 885 }
 886
 887 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 888 {
 889     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 890     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 891     if(shader_major > 3) return ret;
 892
 893     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 894     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 895     return ret;
 896 }
 897
 898 /** Generate the variable & register declarations for the GLSL output target */
 899 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 900         struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
 901         const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
 902 {
 903     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 904     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 905     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 906     const struct wined3d_gl_info *gl_info = context->gl_info;
 907     unsigned int i, extra_constants_needed = 0;
 908     const local_constant *lconst;
 909     DWORD map;
 910
 911     /* There are some minor differences between pixel and vertex shaders */
 912     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 913     char prefix = pshader ? 'P' : 'V';
 914
 915     /* Prototype the subroutines */
 916     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 917     {
 918         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 919     }
 920
 921     /* Declare the constants (aka uniforms) */
 922     if (This->baseShader.limits.constant_float > 0) {
 923         unsigned max_constantsF;
 924         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 925          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 926          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 927          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 928          * a dx9 card, as long as it doesn't also use all the other constants.
 929          *
 930          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 931          * declare only the amount that we're assured to have.
 932          *
 933          * Thus we run into problems in these two cases:
 934          * 1) The shader really uses more uniforms than supported
 935          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 936          */
 937         if (pshader)
 938         {
 939             /* No indirect addressing here. */
 940             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 941         }
 942         else
 943         {
 944             if(This->baseShader.reg_maps.usesrelconstF) {
 945                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
 946                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
 947                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
 948                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
 949                  *
 950                  * Writing gl_ClipVertex requires one uniform for each clipplane as well.
 951                  */
 952                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 953                 if(ctx_priv->cur_vs_args->clip_enabled)
 954                 {
 955                     max_constantsF -= gl_info->limits.clipplanes;
 956                 }
 957                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 958                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 959                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 960                  * for now take this into account when calculating the number of available constants
 961                  */
 962                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
 963                 /* Set by driver quirks in directx.c */
 964                 max_constantsF -= gl_info->reserved_glsl_constants;
 965             }
 966             else
 967             {
 968                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 969             }
 970         }
 971         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
 972         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 973     }
 974
 975     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
 976      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
 977      */
 978     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
 979         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 980
 981     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
 982         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 983
 984     if (!pshader)
 985     {
 986         shader_addline(buffer, "uniform vec4 posFixup;\n");
 987         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 988     }
 989     else
 990     {
 991         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
 992         {
 993             if (!(map & 1)) continue;
 994
 995             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 996
 997             if (reg_maps->luminanceparams & (1 << i))
 998             {
 999                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1000                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1001                 extra_constants_needed++;
1002             }
1003
1004             extra_constants_needed++;
1005         }
1006
1007         if (ps_args->srgb_correction)
1008         {
1009             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1010                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1011             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1012                     srgb_cmp);
1013         }
1014         if (reg_maps->vpos || reg_maps->usesdsy)
1015         {
1016             if (This->baseShader.limits.constant_float + extra_constants_needed
1017                     + 1 < gl_info->limits.glsl_ps_float_constants)
1018             {
1019                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1020                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1021                 extra_constants_needed++;
1022             } else {
1023                 /* This happens because we do not have proper tracking of the constant registers that are
1024                  * actually used, only the max limit of the shader version
1025                  */
1026                 FIXME("Cannot find a free uniform for vpos correction params\n");
1027                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1028                         context->render_offscreen ? 0.0f : device->render_targets[0]->currentDesc.Height,
1029                         context->render_offscreen ? 1.0f : -1.0f);
1030             }
1031             shader_addline(buffer, "vec4 vpos;\n");
1032         }
1033     }
1034
1035     /* Declare texture samplers */
1036     for (i = 0; i < This->baseShader.limits.sampler; i++) {
1037         if (reg_maps->sampler_type[i])
1038         {
1039             switch (reg_maps->sampler_type[i])
1040             {
1041                 case WINED3DSTT_1D:
1042                     if (pshader && ps_args->shadow & (1 << i))
1043                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1044                     else
1045                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1046                     break;
1047                 case WINED3DSTT_2D:
1048                     if (pshader && ps_args->shadow & (1 << i))
1049                     {
1050                         if (device->stateBlock->textures[i]
1051                                 && IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i])
1052                                 == GL_TEXTURE_RECTANGLE_ARB)
1053                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1054                         else
1055                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1056                     }
1057                     else
1058                     {
1059                         if (device->stateBlock->textures[i]
1060                                 && IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i])
1061                                 == GL_TEXTURE_RECTANGLE_ARB)
1062                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1063                         else
1064                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1065                     }
1066                     break;
1067                 case WINED3DSTT_CUBE:
1068                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1069                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1070                     break;
1071                 case WINED3DSTT_VOLUME:
1072                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1073                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1074                     break;
1075                 default:
1076                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1077                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1078                     break;
1079             }
1080         }
1081     }
1082
1083     /* Declare uniforms for NP2 texcoord fixup:
1084      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1085      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1086      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1087     if (pshader && ps_args->np2_fixup) {
1088
1089         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1090         UINT cur = 0;
1091
1092         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1093          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1094          * samplerNP2Fixup stores texture dimensions and is updated through
1095          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1096
1097         for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1098             if (reg_maps->sampler_type[i]) {
1099                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1100
1101                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1102                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1103                     continue;
1104                 }
1105
1106                 fixup->idx[i] = cur++;
1107             }
1108         }
1109
1110         fixup->num_consts = (cur + 1) >> 1;
1111         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1112     }
1113
1114     /* Declare address variables */
1115     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1116     {
1117         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1118     }
1119
1120     /* Declare texture coordinate temporaries and initialize them */
1121     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1122     {
1123         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1124     }
1125
1126     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1127      * helper function shader that is linked in at link time
1128      */
1129     if (pshader && reg_maps->shader_version.major >= 3)
1130     {
1131         if (use_vs(device->stateBlock))
1132         {
1133             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1134         } else {
1135             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1136              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1137              * pixel shader that reads the fixed function color into the packed input registers.
1138              */
1139             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1140         }
1141     }
1142
1143     /* Declare output register temporaries */
1144     if(This->baseShader.limits.packed_output) {
1145         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1146     }
1147
1148     /* Declare temporary variables */
1149     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1150     {
1151         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1152     }
1153
1154     /* Declare attributes */
1155     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1156     {
1157         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1158         {
1159             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1160         }
1161     }
1162
1163     /* Declare loop registers aLx */
1164     for (i = 0; i < reg_maps->loop_depth; i++) {
1165         shader_addline(buffer, "int aL%u;\n", i);
1166         shader_addline(buffer, "int tmpInt%u;\n", i);
1167     }
1168
1169     /* Temporary variables for matrix operations */
1170     shader_addline(buffer, "vec4 tmp0;\n");
1171     shader_addline(buffer, "vec4 tmp1;\n");
1172
1173     /* Local constants use a different name so they can be loaded once at shader link time
1174      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1175      * float -> string conversion can cause precision loss.
1176      */
1177     if(!This->baseShader.load_local_constsF) {
1178         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1179             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1180         }
1181     }
1182
1183     /* Start the main program */
1184     shader_addline(buffer, "void main() {\n");
1185     if(pshader && reg_maps->vpos) {
1186         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1187          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1188          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1189          * precision troubles when we just substract 0.5.
1190          *
1191          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1192          *
1193          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1194          *
1195          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1196          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1197          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1198          * correctly on drivers that returns integer values.
1199          */
1200         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1201     }
1202 }
1203
1204 /*****************************************************************************
1205  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1206  *
1207  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1208  ****************************************************************************/
1209
1210 /* Prototypes */
1211 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1212         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1213
1214 /** Used for opcode modifiers - They multiply the result by the specified amount */
1215 static const char * const shift_glsl_tab[] = {
1216     "",           /*  0 (none) */
1217     "2.0 * ",     /*  1 (x2)   */
1218     "4.0 * ",     /*  2 (x4)   */
1219     "8.0 * ",     /*  3 (x8)   */
1220     "16.0 * ",    /*  4 (x16)  */
1221     "32.0 * ",    /*  5 (x32)  */
1222     "",           /*  6 (x64)  */
1223     "",           /*  7 (x128) */
1224     "",           /*  8 (d256) */
1225     "",           /*  9 (d128) */
1226     "",           /* 10 (d64)  */
1227     "",           /* 11 (d32)  */
1228     "0.0625 * ",  /* 12 (d16)  */
1229     "0.125 * ",   /* 13 (d8)   */
1230     "0.25 * ",    /* 14 (d4)   */
1231     "0.5 * "      /* 15 (d2)   */
1232 };
1233
1234 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1235 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1236 {
1237     out_str[0] = 0;
1238
1239     switch (src_modifier)
1240     {
1241     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1242     case WINED3DSPSM_DW:
1243     case WINED3DSPSM_NONE:
1244         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1245         break;
1246     case WINED3DSPSM_NEG:
1247         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1248         break;
1249     case WINED3DSPSM_NOT:
1250         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1251         break;
1252     case WINED3DSPSM_BIAS:
1253         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1254         break;
1255     case WINED3DSPSM_BIASNEG:
1256         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1257         break;
1258     case WINED3DSPSM_SIGN:
1259         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1260         break;
1261     case WINED3DSPSM_SIGNNEG:
1262         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1263         break;
1264     case WINED3DSPSM_COMP:
1265         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1266         break;
1267     case WINED3DSPSM_X2:
1268         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1269         break;
1270     case WINED3DSPSM_X2NEG:
1271         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1272         break;
1273     case WINED3DSPSM_ABS:
1274         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1275         break;
1276     case WINED3DSPSM_ABSNEG:
1277         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1278         break;
1279     default:
1280         FIXME("Unhandled modifier %u\n", src_modifier);
1281         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1282     }
1283 }
1284
1285 /** Writes the GLSL variable name that corresponds to the register that the
1286  * DX opcode parameter is trying to access */
1287 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1288         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1289 {
1290     /* oPos, oFog and oPts in D3D */
1291     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1292
1293     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1294     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1295     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1296
1297     *is_color = FALSE;
1298
1299     switch (reg->type)
1300     {
1301         case WINED3DSPR_TEMP:
1302             sprintf(register_name, "R%u", reg->idx);
1303             break;
1304
1305         case WINED3DSPR_INPUT:
1306             /* vertex shaders */
1307             if (!pshader)
1308             {
1309                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1310                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1311                 sprintf(register_name, "attrib%u", reg->idx);
1312                 break;
1313             }
1314
1315             /* pixel shaders >= 3.0 */
1316             if (This->baseShader.reg_maps.shader_version.major >= 3)
1317             {
1318                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1319                 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1320
1321                 if (reg->rel_addr)
1322                 {
1323                     glsl_src_param_t rel_param;
1324
1325                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1326
1327                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1328                      * operation there */
1329                     if (idx)
1330                     {
1331                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1332                         {
1333                             sprintf(register_name,
1334                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1335                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1336                                     rel_param.param_str, idx);
1337                         }
1338                         else
1339                         {
1340                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1341                         }
1342                     }
1343                     else
1344                     {
1345                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1346                         {
1347                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1348                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1349                                     rel_param.param_str);
1350                         }
1351                         else
1352                         {
1353                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1354                         }
1355                     }
1356                 }
1357                 else
1358                 {
1359                     if (idx == in_count) sprintf(register_name, "gl_Color");
1360                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1361                     else sprintf(register_name, "IN[%u]", idx);
1362                 }
1363             }
1364             else
1365             {
1366                 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1367                 else strcpy(register_name, "gl_SecondaryColor");
1368                 break;
1369             }
1370             break;
1371
1372         case WINED3DSPR_CONST:
1373             {
1374                 const char prefix = pshader ? 'P' : 'V';
1375
1376                 /* Relative addressing */
1377                 if (reg->rel_addr)
1378                 {
1379                     glsl_src_param_t rel_param;
1380                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1381                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1382                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1383                 }
1384                 else
1385                 {
1386                     if (shader_constant_is_local(This, reg->idx))
1387                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1388                     else
1389                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1390                 }
1391             }
1392             break;
1393
1394         case WINED3DSPR_CONSTINT:
1395             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1396             else sprintf(register_name, "VI[%u]", reg->idx);
1397             break;
1398
1399         case WINED3DSPR_CONSTBOOL:
1400             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1401             else sprintf(register_name, "VB[%u]", reg->idx);
1402             break;
1403
1404         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1405             if (pshader) sprintf(register_name, "T%u", reg->idx);
1406             else sprintf(register_name, "A%u", reg->idx);
1407             break;
1408
1409         case WINED3DSPR_LOOP:
1410             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1411             break;
1412
1413         case WINED3DSPR_SAMPLER:
1414             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1415             else sprintf(register_name, "Vsampler%u", reg->idx);
1416             break;
1417
1418         case WINED3DSPR_COLOROUT:
1419             if (reg->idx >= gl_info->limits.buffers)
1420                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1421
1422             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1423             break;
1424
1425         case WINED3DSPR_RASTOUT:
1426             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1427             break;
1428
1429         case WINED3DSPR_DEPTHOUT:
1430             sprintf(register_name, "gl_FragDepth");
1431             break;
1432
1433         case WINED3DSPR_ATTROUT:
1434             if (reg->idx == 0) sprintf(register_name, "OUT[8]");
1435             else sprintf(register_name, "OUT[9]");
1436             break;
1437
1438         case WINED3DSPR_TEXCRDOUT:
1439             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1440             sprintf(register_name, "OUT[%u]", reg->idx);
1441             break;
1442
1443         case WINED3DSPR_MISCTYPE:
1444             if (reg->idx == 0)
1445             {
1446                 /* vPos */
1447                 sprintf(register_name, "vpos");
1448             }
1449             else if (reg->idx == 1)
1450             {
1451                 /* Note that gl_FrontFacing is a bool, while vFace is
1452                  * a float for which the sign determines front/back */
1453                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1454             }
1455             else
1456             {
1457                 FIXME("Unhandled misctype register %d\n", reg->idx);
1458                 sprintf(register_name, "unrecognized_register");
1459             }
1460             break;
1461
1462         case WINED3DSPR_IMMCONST:
1463             switch (reg->immconst_type)
1464             {
1465                 case WINED3D_IMMCONST_FLOAT:
1466                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1467                     break;
1468
1469                 case WINED3D_IMMCONST_FLOAT4:
1470                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1471                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1472                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1473                     break;
1474
1475                 default:
1476                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1477                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1478             }
1479             break;
1480
1481         default:
1482             FIXME("Unhandled register name Type(%d)\n", reg->type);
1483             sprintf(register_name, "unrecognized_register");
1484             break;
1485     }
1486 }
1487
1488 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1489 {
1490     *str++ = '.';
1491     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1492     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1493     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1494     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1495     *str = '\0';
1496 }
1497
1498 /* Get the GLSL write mask for the destination register */
1499 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1500 {
1501     DWORD mask = param->write_mask;
1502
1503     if (shader_is_scalar(&param->reg))
1504     {
1505         mask = WINED3DSP_WRITEMASK_0;
1506         *write_mask = '\0';
1507     }
1508     else
1509     {
1510         shader_glsl_write_mask_to_str(mask, write_mask);
1511     }
1512
1513     return mask;
1514 }
1515
1516 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1517     unsigned int size = 0;
1518
1519     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1520     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1521     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1522     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1523
1524     return size;
1525 }
1526
1527 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1528 {
1529     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1530      * but addressed as "rgba". To fix this we need to swap the register's x
1531      * and z components. */
1532     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1533
1534     *str++ = '.';
1535     /* swizzle bits fields: wwzzyyxx */
1536     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1537     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1538     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1539     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1540     *str = '\0';
1541 }
1542
1543 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1544         BOOL fixup, DWORD mask, char *swizzle_str)
1545 {
1546     if (shader_is_scalar(&param->reg))
1547         *swizzle_str = '\0';
1548     else
1549         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1550 }
1551
1552 /* From a given parameter token, generate the corresponding GLSL string.
1553  * Also, return the actual register name and swizzle in case the
1554  * caller needs this information as well. */
1555 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1556         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1557 {
1558     BOOL is_color = FALSE;
1559     char swizzle_str[6];
1560
1561     glsl_src->reg_name[0] = '\0';
1562     glsl_src->param_str[0] = '\0';
1563     swizzle_str[0] = '\0';
1564
1565     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1566     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1567     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_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 DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1574         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1575 {
1576     BOOL is_color = FALSE;
1577
1578     glsl_dst->mask_str[0] = '\0';
1579     glsl_dst->reg_name[0] = '\0';
1580
1581     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1582     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1583 }
1584
1585 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1586 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1587         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1588 {
1589     glsl_dst_param_t glsl_dst;
1590     DWORD mask;
1591
1592     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1593     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1594
1595     return mask;
1596 }
1597
1598 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1599 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1600 {
1601     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1602 }
1603
1604 /** Process GLSL instruction modifiers */
1605 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1606 {
1607     glsl_dst_param_t dst_param;
1608     DWORD modifiers;
1609
1610     if (!ins->dst_count) return;
1611
1612     modifiers = ins->dst[0].modifiers;
1613     if (!modifiers) return;
1614
1615     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1616
1617     if (modifiers & WINED3DSPDM_SATURATE)
1618     {
1619         /* _SAT means to clamp the value of the register to between 0 and 1 */
1620         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1621                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1622     }
1623
1624     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1625     {
1626         FIXME("_centroid modifier not handled\n");
1627     }
1628
1629     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1630     {
1631         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1632     }
1633 }
1634
1635 static inline const char *shader_get_comp_op(DWORD op)
1636 {
1637     switch (op) {
1638         case COMPARISON_GT: return ">";
1639         case COMPARISON_EQ: return "==";
1640         case COMPARISON_GE: return ">=";
1641         case COMPARISON_LT: return "<";
1642         case COMPARISON_NE: return "!=";
1643         case COMPARISON_LE: return "<=";
1644         default:
1645             FIXME("Unrecognized comparison value: %u\n", op);
1646             return "(\?\?)";
1647     }
1648 }
1649
1650 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1651         DWORD sampler_idx, DWORD flags, glsl_sample_function_t *sample_function)
1652 {
1653     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1654     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1655     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1656             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1657     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1658     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1659     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1660     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1661
1662     /* Note that there's no such thing as a projected cube texture. */
1663     switch(sampler_type) {
1664         case WINED3DSTT_1D:
1665             if (shadow)
1666             {
1667                 if (lod)
1668                 {
1669                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1670                 }
1671                 else if (grad)
1672                 {
1673                     if (gl_info->supported[EXT_GPU_SHADER4])
1674                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1675                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1676                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1677                     else
1678                     {
1679                         FIXME("Unsupported 1D shadow grad function.\n");
1680                         sample_function->name = "unsupported1DGrad";
1681                     }
1682                 }
1683                 else
1684                 {
1685                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1686                 }
1687                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1688             }
1689             else
1690             {
1691                 if (lod)
1692                 {
1693                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1694                 }
1695                 else if (grad)
1696                 {
1697                     if (gl_info->supported[EXT_GPU_SHADER4])
1698                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1699                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1700                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1701                     else
1702                     {
1703                         FIXME("Unsupported 1D grad function.\n");
1704                         sample_function->name = "unsupported1DGrad";
1705                     }
1706                 }
1707                 else
1708                 {
1709                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1710                 }
1711                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1712             }
1713             break;
1714
1715         case WINED3DSTT_2D:
1716             if (shadow)
1717             {
1718                 if (texrect)
1719                 {
1720                     if (lod)
1721                     {
1722                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1723                     }
1724                     else if (grad)
1725                     {
1726                         if (gl_info->supported[EXT_GPU_SHADER4])
1727                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1728                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1729                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1730                         else
1731                         {
1732                             FIXME("Unsupported RECT shadow grad function.\n");
1733                             sample_function->name = "unsupported2DRectGrad";
1734                         }
1735                     }
1736                     else
1737                     {
1738                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1739                     }
1740                 }
1741                 else
1742                 {
1743                     if (lod)
1744                     {
1745                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1746                     }
1747                     else if (grad)
1748                     {
1749                         if (gl_info->supported[EXT_GPU_SHADER4])
1750                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1751                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1752                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1753                         else
1754                         {
1755                             FIXME("Unsupported 2D shadow grad function.\n");
1756                             sample_function->name = "unsupported2DGrad";
1757                         }
1758                     }
1759                     else
1760                     {
1761                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1762                     }
1763                 }
1764                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1765             }
1766             else
1767             {
1768                 if (texrect)
1769                 {
1770                     if (lod)
1771                     {
1772                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1773                     }
1774                     else if (grad)
1775                     {
1776                         if (gl_info->supported[EXT_GPU_SHADER4])
1777                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1778                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1779                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1780                         else
1781                         {
1782                             FIXME("Unsupported RECT grad function.\n");
1783                             sample_function->name = "unsupported2DRectGrad";
1784                         }
1785                     }
1786                     else
1787                     {
1788                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1789                     }
1790                 }
1791                 else
1792                 {
1793                     if (lod)
1794                     {
1795                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1796                     }
1797                     else if (grad)
1798                     {
1799                         if (gl_info->supported[EXT_GPU_SHADER4])
1800                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1801                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1802                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1803                         else
1804                         {
1805                             FIXME("Unsupported 2D grad function.\n");
1806                             sample_function->name = "unsupported2DGrad";
1807                         }
1808                     }
1809                     else
1810                     {
1811                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1812                     }
1813                 }
1814                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1815             }
1816             break;
1817
1818         case WINED3DSTT_CUBE:
1819             if (shadow)
1820             {
1821                 FIXME("Unsupported Cube shadow function.\n");
1822                 sample_function->name = "unsupportedCubeShadow";
1823                 sample_function->coord_mask = 0;
1824             }
1825             else
1826             {
1827                 if (lod)
1828                 {
1829                     sample_function->name = "textureCubeLod";
1830                 }
1831                 else if (grad)
1832                 {
1833                     if (gl_info->supported[EXT_GPU_SHADER4])
1834                         sample_function->name = "textureCubeGrad";
1835                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1836                         sample_function->name = "textureCubeGradARB";
1837                     else
1838                     {
1839                         FIXME("Unsupported Cube grad function.\n");
1840                         sample_function->name = "unsupportedCubeGrad";
1841                     }
1842                 }
1843                 else
1844                 {
1845                     sample_function->name = "textureCube";
1846                 }
1847                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1848             }
1849             break;
1850
1851         case WINED3DSTT_VOLUME:
1852             if (shadow)
1853             {
1854                 FIXME("Unsupported 3D shadow function.\n");
1855                 sample_function->name = "unsupported3DShadow";
1856                 sample_function->coord_mask = 0;
1857             }
1858             else
1859             {
1860                 if (lod)
1861                 {
1862                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1863                 }
1864                 else  if (grad)
1865                 {
1866                     if (gl_info->supported[EXT_GPU_SHADER4])
1867                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1868                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1869                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1870                     else
1871                     {
1872                         FIXME("Unsupported 3D grad function.\n");
1873                         sample_function->name = "unsupported3DGrad";
1874                     }
1875                 }
1876                 else
1877                 {
1878                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1879                 }
1880                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1881             }
1882             break;
1883
1884         default:
1885             sample_function->name = "";
1886             sample_function->coord_mask = 0;
1887             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1888             break;
1889     }
1890 }
1891
1892 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1893         BOOL sign_fixup, enum fixup_channel_source channel_source)
1894 {
1895     switch(channel_source)
1896     {
1897         case CHANNEL_SOURCE_ZERO:
1898             strcat(arguments, "0.0");
1899             break;
1900
1901         case CHANNEL_SOURCE_ONE:
1902             strcat(arguments, "1.0");
1903             break;
1904
1905         case CHANNEL_SOURCE_X:
1906             strcat(arguments, reg_name);
1907             strcat(arguments, ".x");
1908             break;
1909
1910         case CHANNEL_SOURCE_Y:
1911             strcat(arguments, reg_name);
1912             strcat(arguments, ".y");
1913             break;
1914
1915         case CHANNEL_SOURCE_Z:
1916             strcat(arguments, reg_name);
1917             strcat(arguments, ".z");
1918             break;
1919
1920         case CHANNEL_SOURCE_W:
1921             strcat(arguments, reg_name);
1922             strcat(arguments, ".w");
1923             break;
1924
1925         default:
1926             FIXME("Unhandled channel source %#x\n", channel_source);
1927             strcat(arguments, "undefined");
1928             break;
1929     }
1930
1931     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1932 }
1933
1934 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1935 {
1936     struct wined3d_shader_dst_param dst;
1937     unsigned int mask_size, remaining;
1938     glsl_dst_param_t dst_param;
1939     char arguments[256];
1940     DWORD mask;
1941
1942     mask = 0;
1943     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1944     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1945     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1946     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1947     mask &= ins->dst[0].write_mask;
1948
1949     if (!mask) return; /* Nothing to do */
1950
1951     if (is_complex_fixup(fixup))
1952     {
1953         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1954         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1955         return;
1956     }
1957
1958     mask_size = shader_glsl_get_write_mask_size(mask);
1959
1960     dst = ins->dst[0];
1961     dst.write_mask = mask;
1962     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1963
1964     arguments[0] = '\0';
1965     remaining = mask_size;
1966     if (mask & WINED3DSP_WRITEMASK_0)
1967     {
1968         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1969         if (--remaining) strcat(arguments, ", ");
1970     }
1971     if (mask & WINED3DSP_WRITEMASK_1)
1972     {
1973         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1974         if (--remaining) strcat(arguments, ", ");
1975     }
1976     if (mask & WINED3DSP_WRITEMASK_2)
1977     {
1978         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1979         if (--remaining) strcat(arguments, ", ");
1980     }
1981     if (mask & WINED3DSP_WRITEMASK_3)
1982     {
1983         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1984         if (--remaining) strcat(arguments, ", ");
1985     }
1986
1987     if (mask_size > 1)
1988     {
1989         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1990                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1991     }
1992     else
1993     {
1994         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1995     }
1996 }
1997
1998 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
1999         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
2000         const char *dx, const char *dy,
2001         const char *bias, const char *coord_reg_fmt, ...)
2002 {
2003     const char *sampler_base;
2004     char dst_swizzle[6];
2005     struct color_fixup_desc fixup;
2006     BOOL np2_fixup = FALSE;
2007     va_list args;
2008
2009     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2010
2011     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2012     {
2013         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2014         fixup = priv->cur_ps_args->color_fixup[sampler];
2015         sampler_base = "Psampler";
2016
2017         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2018             if(bias) {
2019                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2020             } else {
2021                 np2_fixup = TRUE;
2022             }
2023         }
2024     } else {
2025         sampler_base = "Vsampler";
2026         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2027     }
2028
2029     shader_glsl_append_dst(ins->ctx->buffer, ins);
2030
2031     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2032
2033     va_start(args, coord_reg_fmt);
2034     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2035     va_end(args);
2036
2037     if(bias) {
2038         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2039     } else {
2040         if (np2_fixup) {
2041             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2042             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2043
2044             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2045                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2046         } else if(dx && dy) {
2047             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2048         } else {
2049             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2050         }
2051     }
2052
2053     if(!is_identity_fixup(fixup)) {
2054         shader_glsl_color_correction(ins, fixup);
2055     }
2056 }
2057
2058 /*****************************************************************************
2059  * Begin processing individual instruction opcodes
2060  ****************************************************************************/
2061
2062 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2063 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2064 {
2065     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2066     glsl_src_param_t src0_param;
2067     glsl_src_param_t src1_param;
2068     DWORD write_mask;
2069     char op;
2070
2071     /* Determine the GLSL operator to use based on the opcode */
2072     switch (ins->handler_idx)
2073     {
2074         case WINED3DSIH_MUL: op = '*'; break;
2075         case WINED3DSIH_ADD: op = '+'; break;
2076         case WINED3DSIH_SUB: op = '-'; break;
2077         default:
2078             op = ' ';
2079             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2080             break;
2081     }
2082
2083     write_mask = shader_glsl_append_dst(buffer, ins);
2084     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2085     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2086     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2087 }
2088
2089 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2090 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2091 {
2092     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2093     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2094     glsl_src_param_t src0_param;
2095     DWORD write_mask;
2096
2097     write_mask = shader_glsl_append_dst(buffer, ins);
2098     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2099
2100     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2101      * shader versions WINED3DSIO_MOVA is used for this. */
2102     if (ins->ctx->reg_maps->shader_version.major == 1
2103             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2104             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2105     {
2106         /* This is a simple floor() */
2107         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2108         if (mask_size > 1) {
2109             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2110         } else {
2111             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2112         }
2113     }
2114     else if(ins->handler_idx == WINED3DSIH_MOVA)
2115     {
2116         /* We need to *round* to the nearest int here. */
2117         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2118
2119         if (gl_info->supported[EXT_GPU_SHADER4])
2120         {
2121             if (mask_size > 1)
2122                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2123             else
2124                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2125         }
2126         else
2127         {
2128             if (mask_size > 1)
2129                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2130                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2131             else
2132                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2133                         src0_param.param_str, src0_param.param_str);
2134         }
2135     }
2136     else
2137     {
2138         shader_addline(buffer, "%s);\n", src0_param.param_str);
2139     }
2140 }
2141
2142 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2143 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2144 {
2145     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2146     glsl_src_param_t src0_param;
2147     glsl_src_param_t src1_param;
2148     DWORD dst_write_mask, src_write_mask;
2149     unsigned int dst_size = 0;
2150
2151     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2152     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2153
2154     /* dp3 works on vec3, dp4 on vec4 */
2155     if (ins->handler_idx == WINED3DSIH_DP4)
2156     {
2157         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2158     } else {
2159         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2160     }
2161
2162     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2163     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2164
2165     if (dst_size > 1) {
2166         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2167     } else {
2168         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2169     }
2170 }
2171
2172 /* Note that this instruction has some restrictions. The destination write mask
2173  * can't contain the w component, and the source swizzles have to be .xyzw */
2174 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2175 {
2176     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2177     glsl_src_param_t src0_param;
2178     glsl_src_param_t src1_param;
2179     char dst_mask[6];
2180
2181     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2182     shader_glsl_append_dst(ins->ctx->buffer, ins);
2183     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2184     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2185     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2186 }
2187
2188 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2189  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2190  * GLSL uses the value as-is. */
2191 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2192 {
2193     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2194     glsl_src_param_t src0_param;
2195     glsl_src_param_t src1_param;
2196     DWORD dst_write_mask;
2197     unsigned int dst_size;
2198
2199     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2200     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2201
2202     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2203     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2204
2205     if (dst_size > 1)
2206     {
2207         shader_addline(buffer, "vec%u(pow(abs(%s), %s)));\n",
2208                 dst_size, src0_param.param_str, src1_param.param_str);
2209     }
2210     else
2211     {
2212         shader_addline(buffer, "pow(abs(%s), %s));\n",
2213                 src0_param.param_str, src1_param.param_str);
2214     }
2215 }
2216
2217 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2218  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2219  * GLSL uses the value as-is. */
2220 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2221 {
2222     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2223     glsl_src_param_t src0_param;
2224     DWORD dst_write_mask;
2225     unsigned int dst_size;
2226
2227     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2228     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2229
2230     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2231
2232     if (dst_size > 1)
2233     {
2234         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2235                 dst_size, src0_param.param_str);
2236     }
2237     else
2238     {
2239         shader_addline(buffer, "log2(abs(%s)));\n",
2240                 src0_param.param_str);
2241     }
2242 }
2243
2244 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2245 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2246 {
2247     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2248     glsl_src_param_t src_param;
2249     const char *instruction;
2250     DWORD write_mask;
2251     unsigned i;
2252
2253     /* Determine the GLSL function to use based on the opcode */
2254     /* TODO: Possibly make this a table for faster lookups */
2255     switch (ins->handler_idx)
2256     {
2257         case WINED3DSIH_MIN: instruction = "min"; break;
2258         case WINED3DSIH_MAX: instruction = "max"; break;
2259         case WINED3DSIH_ABS: instruction = "abs"; break;
2260         case WINED3DSIH_FRC: instruction = "fract"; break;
2261         case WINED3DSIH_EXP: instruction = "exp2"; break;
2262         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2263         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2264         default: instruction = "";
2265             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2266             break;
2267     }
2268
2269     write_mask = shader_glsl_append_dst(buffer, ins);
2270
2271     shader_addline(buffer, "%s(", instruction);
2272
2273     if (ins->src_count)
2274     {
2275         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2276         shader_addline(buffer, "%s", src_param.param_str);
2277         for (i = 1; i < ins->src_count; ++i)
2278         {
2279             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2280             shader_addline(buffer, ", %s", src_param.param_str);
2281         }
2282     }
2283
2284     shader_addline(buffer, "));\n");
2285 }
2286
2287 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2288 {
2289     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2290     glsl_src_param_t src_param;
2291     unsigned int mask_size;
2292     DWORD write_mask;
2293     char dst_mask[6];
2294
2295     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2296     mask_size = shader_glsl_get_write_mask_size(write_mask);
2297     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2298
2299     shader_addline(buffer, "tmp0.x = length(%s);\n", src_param.param_str);
2300     shader_glsl_append_dst(buffer, ins);
2301     if (mask_size > 1)
2302     {
2303         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s / tmp0.x));\n",
2304                 mask_size, src_param.param_str);
2305     }
2306     else
2307     {
2308         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s / tmp0.x));\n",
2309                 src_param.param_str);
2310     }
2311 }
2312
2313 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2314  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2315  *   dst.x = 2^(floor(src))
2316  *   dst.y = src - floor(src)
2317  *   dst.z = 2^src   (partial precision is allowed, but optional)
2318  *   dst.w = 1.0;
2319  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2320  *   dst = 2^src;    (partial precision is allowed, but optional)
2321  */
2322 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2323 {
2324     glsl_src_param_t src_param;
2325
2326     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2327
2328     if (ins->ctx->reg_maps->shader_version.major < 2)
2329     {
2330         char dst_mask[6];
2331
2332         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2333         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2334         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2335         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2336
2337         shader_glsl_append_dst(ins->ctx->buffer, ins);
2338         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2339         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2340     } else {
2341         DWORD write_mask;
2342         unsigned int mask_size;
2343
2344         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2345         mask_size = shader_glsl_get_write_mask_size(write_mask);
2346
2347         if (mask_size > 1) {
2348             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2349         } else {
2350             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2351         }
2352     }
2353 }
2354
2355 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2356 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2357 {
2358     glsl_src_param_t src_param;
2359     DWORD write_mask;
2360     unsigned int mask_size;
2361
2362     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2363     mask_size = shader_glsl_get_write_mask_size(write_mask);
2364     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2365
2366     if (mask_size > 1)
2367     {
2368         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2369                 mask_size, src_param.param_str);
2370     }
2371     else
2372     {
2373         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2374                 src_param.param_str);
2375     }
2376 }
2377
2378 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2379 {
2380     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2381     glsl_src_param_t src_param;
2382     DWORD write_mask;
2383     unsigned int mask_size;
2384
2385     write_mask = shader_glsl_append_dst(buffer, ins);
2386     mask_size = shader_glsl_get_write_mask_size(write_mask);
2387
2388     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2389
2390     if (mask_size > 1)
2391     {
2392         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2393                 mask_size, src_param.param_str);
2394     }
2395     else
2396     {
2397         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2398                 src_param.param_str);
2399     }
2400 }
2401
2402 /** Process signed comparison opcodes in GLSL. */
2403 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2404 {
2405     glsl_src_param_t src0_param;
2406     glsl_src_param_t src1_param;
2407     DWORD write_mask;
2408     unsigned int mask_size;
2409
2410     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2411     mask_size = shader_glsl_get_write_mask_size(write_mask);
2412     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2413     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2414
2415     if (mask_size > 1) {
2416         const char *compare;
2417
2418         switch(ins->handler_idx)
2419         {
2420             case WINED3DSIH_SLT: compare = "lessThan"; break;
2421             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2422             default: compare = "";
2423                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2424         }
2425
2426         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2427                 src0_param.param_str, src1_param.param_str);
2428     } else {
2429         switch(ins->handler_idx)
2430         {
2431             case WINED3DSIH_SLT:
2432                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2433                  * to return 0.0 but step returns 1.0 because step is not < x
2434                  * An alternative is a bvec compare padded with an unused second component.
2435                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2436                  * issue. Playing with not() is not possible either because not() does not accept
2437                  * a scalar.
2438                  */
2439                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2440                         src0_param.param_str, src1_param.param_str);
2441                 break;
2442             case WINED3DSIH_SGE:
2443                 /* Here we can use the step() function and safe a conditional */
2444                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2445                 break;
2446             default:
2447                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2448         }
2449
2450     }
2451 }
2452
2453 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2454 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2455 {
2456     glsl_src_param_t src0_param;
2457     glsl_src_param_t src1_param;
2458     glsl_src_param_t src2_param;
2459     DWORD write_mask, cmp_channel = 0;
2460     unsigned int i, j;
2461     char mask_char[6];
2462     BOOL temp_destination = FALSE;
2463
2464     if (shader_is_scalar(&ins->src[0].reg))
2465     {
2466         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2467
2468         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2469         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2470         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2471
2472         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2473                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2474     } else {
2475         DWORD dst_mask = ins->dst[0].write_mask;
2476         struct wined3d_shader_dst_param dst = ins->dst[0];
2477
2478         /* Cycle through all source0 channels */
2479         for (i=0; i<4; i++) {
2480             write_mask = 0;
2481             /* Find the destination channels which use the current source0 channel */
2482             for (j=0; j<4; j++) {
2483                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2484                 {
2485                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2486                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2487                 }
2488             }
2489             dst.write_mask = dst_mask & write_mask;
2490
2491             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2492             * The first lines may overwrite source parameters of the following lines.
2493             * Deal with that by using a temporary destination register if needed
2494             */
2495             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2496                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2497                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2498                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2499                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2500                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2501             {
2502                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2503                 if (!write_mask) continue;
2504                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2505                 temp_destination = TRUE;
2506             } else {
2507                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2508                 if (!write_mask) continue;
2509             }
2510
2511             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2512             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2513             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2514
2515             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2516                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2517         }
2518
2519         if(temp_destination) {
2520             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2521             shader_glsl_append_dst(ins->ctx->buffer, ins);
2522             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2523         }
2524     }
2525
2526 }
2527
2528 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2529 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2530  * the compare is done per component of src0. */
2531 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2532 {
2533     struct wined3d_shader_dst_param dst;
2534     glsl_src_param_t src0_param;
2535     glsl_src_param_t src1_param;
2536     glsl_src_param_t src2_param;
2537     DWORD write_mask, cmp_channel = 0;
2538     unsigned int i, j;
2539     DWORD dst_mask;
2540     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2541             ins->ctx->reg_maps->shader_version.minor);
2542
2543     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2544     {
2545         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2546         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2547         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2548         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2549
2550         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2551         if (ins->coissue)
2552         {
2553             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2554         } else {
2555             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2556                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2557         }
2558         return;
2559     }
2560     /* Cycle through all source0 channels */
2561     dst_mask = ins->dst[0].write_mask;
2562     dst = ins->dst[0];
2563     for (i=0; i<4; i++) {
2564         write_mask = 0;
2565         /* Find the destination channels which use the current source0 channel */
2566         for (j=0; j<4; j++) {
2567             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2568             {
2569                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2570                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2571             }
2572         }
2573
2574         dst.write_mask = dst_mask & write_mask;
2575         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2576         if (!write_mask) continue;
2577
2578         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2579         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2580         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2581
2582         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2583                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2584     }
2585 }
2586
2587 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2588 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2589 {
2590     glsl_src_param_t src0_param;
2591     glsl_src_param_t src1_param;
2592     glsl_src_param_t src2_param;
2593     DWORD write_mask;
2594
2595     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2596     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2597     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2598     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2599     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2600             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2601 }
2602
2603 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2604    Vertex shaders to GLSL codes */
2605 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2606 {
2607     int i;
2608     int nComponents = 0;
2609     struct wined3d_shader_dst_param tmp_dst = {{0}};
2610     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2611     struct wined3d_shader_instruction tmp_ins;
2612
2613     memset(&tmp_ins, 0, sizeof(tmp_ins));
2614
2615     /* Set constants for the temporary argument */
2616     tmp_ins.ctx = ins->ctx;
2617     tmp_ins.dst_count = 1;
2618     tmp_ins.dst = &tmp_dst;
2619     tmp_ins.src_count = 2;
2620     tmp_ins.src = tmp_src;
2621
2622     switch(ins->handler_idx)
2623     {
2624         case WINED3DSIH_M4x4:
2625             nComponents = 4;
2626             tmp_ins.handler_idx = WINED3DSIH_DP4;
2627             break;
2628         case WINED3DSIH_M4x3:
2629             nComponents = 3;
2630             tmp_ins.handler_idx = WINED3DSIH_DP4;
2631             break;
2632         case WINED3DSIH_M3x4:
2633             nComponents = 4;
2634             tmp_ins.handler_idx = WINED3DSIH_DP3;
2635             break;
2636         case WINED3DSIH_M3x3:
2637             nComponents = 3;
2638             tmp_ins.handler_idx = WINED3DSIH_DP3;
2639             break;
2640         case WINED3DSIH_M3x2:
2641             nComponents = 2;
2642             tmp_ins.handler_idx = WINED3DSIH_DP3;
2643             break;
2644         default:
2645             break;
2646     }
2647
2648     tmp_dst = ins->dst[0];
2649     tmp_src[0] = ins->src[0];
2650     tmp_src[1] = ins->src[1];
2651     for (i = 0; i < nComponents; ++i)
2652     {
2653         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2654         shader_glsl_dot(&tmp_ins);
2655         ++tmp_src[1].reg.idx;
2656     }
2657 }
2658
2659 /**
2660     The LRP instruction performs a component-wise linear interpolation
2661     between the second and third operands using the first operand as the
2662     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2663     This is equivalent to mix(src2, src1, src0);
2664 */
2665 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2666 {
2667     glsl_src_param_t src0_param;
2668     glsl_src_param_t src1_param;
2669     glsl_src_param_t src2_param;
2670     DWORD write_mask;
2671
2672     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2673
2674     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2675     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2676     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2677
2678     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2679             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2680 }
2681
2682 /** Process the WINED3DSIO_LIT instruction in GLSL:
2683  * dst.x = dst.w = 1.0
2684  * dst.y = (src0.x > 0) ? src0.x
2685  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2686  *                                        where src.w is clamped at +- 128
2687  */
2688 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2689 {
2690     glsl_src_param_t src0_param;
2691     glsl_src_param_t src1_param;
2692     glsl_src_param_t src3_param;
2693     char dst_mask[6];
2694
2695     shader_glsl_append_dst(ins->ctx->buffer, ins);
2696     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2697
2698     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2699     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2700     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2701
2702     /* The sdk specifies the instruction like this
2703      * dst.x = 1.0;
2704      * if(src.x > 0.0) dst.y = src.x
2705      * else dst.y = 0.0.
2706      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2707      * else dst.z = 0.0;
2708      * dst.w = 1.0;
2709      *
2710      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2711      * dst.x = 1.0                                  ... No further explanation needed
2712      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2713      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2714      * dst.w = 1.0.                                 ... Nothing fancy.
2715      *
2716      * So we still have one conditional in there. So do this:
2717      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2718      *
2719      * 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),
2720      * 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.
2721      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2722      */
2723     shader_addline(ins->ctx->buffer,
2724             "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",
2725             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2726 }
2727
2728 /** Process the WINED3DSIO_DST instruction in GLSL:
2729  * dst.x = 1.0
2730  * dst.y = src0.x * src0.y
2731  * dst.z = src0.z
2732  * dst.w = src1.w
2733  */
2734 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2735 {
2736     glsl_src_param_t src0y_param;
2737     glsl_src_param_t src0z_param;
2738     glsl_src_param_t src1y_param;
2739     glsl_src_param_t src1w_param;
2740     char dst_mask[6];
2741
2742     shader_glsl_append_dst(ins->ctx->buffer, ins);
2743     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2744
2745     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2746     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2747     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2748     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2749
2750     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2751             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2752 }
2753
2754 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2755  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2756  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2757  *
2758  * dst.x = cos(src0.?)
2759  * dst.y = sin(src0.?)
2760  * dst.z = dst.z
2761  * dst.w = dst.w
2762  */
2763 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2764 {
2765     glsl_src_param_t src0_param;
2766     DWORD write_mask;
2767
2768     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2769     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2770
2771     switch (write_mask) {
2772         case WINED3DSP_WRITEMASK_0:
2773             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2774             break;
2775
2776         case WINED3DSP_WRITEMASK_1:
2777             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2778             break;
2779
2780         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2781             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2782             break;
2783
2784         default:
2785             ERR("Write mask should be .x, .y or .xy\n");
2786             break;
2787     }
2788 }
2789
2790 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2791  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2792  * generate invalid code
2793  */
2794 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2795 {
2796     glsl_src_param_t src0_param;
2797     DWORD write_mask;
2798
2799     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2800     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2801
2802     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2803 }
2804
2805 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2806  * Start a for() loop where src1.y is the initial value of aL,
2807  *  increment aL by src1.z for a total of src1.x iterations.
2808  *  Need to use a temporary variable for this operation.
2809  */
2810 /* FIXME: I don't think nested loops will work correctly this way. */
2811 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2812 {
2813     glsl_src_param_t src1_param;
2814     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2815     const DWORD *control_values = NULL;
2816     const local_constant *constant;
2817
2818     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2819
2820     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2821      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2822      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2823      * addressing.
2824      */
2825     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2826     {
2827         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2828             if (constant->idx == ins->src[1].reg.idx)
2829             {
2830                 control_values = constant->value;
2831                 break;
2832             }
2833         }
2834     }
2835
2836     if (control_values)
2837     {
2838         struct wined3d_shader_loop_control loop_control;
2839         loop_control.count = control_values[0];
2840         loop_control.start = control_values[1];
2841         loop_control.step = (int)control_values[2];
2842
2843         if (loop_control.step > 0)
2844         {
2845             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2846                     shader->baseShader.cur_loop_depth, loop_control.start,
2847                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2848                     shader->baseShader.cur_loop_depth, loop_control.step);
2849         }
2850         else if (loop_control.step < 0)
2851         {
2852             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2853                     shader->baseShader.cur_loop_depth, loop_control.start,
2854                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2855                     shader->baseShader.cur_loop_depth, loop_control.step);
2856         }
2857         else
2858         {
2859             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2860                     shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2861                     shader->baseShader.cur_loop_depth, loop_control.count,
2862                     shader->baseShader.cur_loop_depth);
2863         }
2864     } else {
2865         shader_addline(ins->ctx->buffer,
2866                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2867                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2868                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2869                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2870     }
2871
2872     shader->baseShader.cur_loop_depth++;
2873     shader->baseShader.cur_loop_regno++;
2874 }
2875
2876 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2877 {
2878     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2879
2880     shader_addline(ins->ctx->buffer, "}\n");
2881
2882     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2883     {
2884         shader->baseShader.cur_loop_depth--;
2885         shader->baseShader.cur_loop_regno--;
2886     }
2887
2888     if (ins->handler_idx == WINED3DSIH_ENDREP)
2889     {
2890         shader->baseShader.cur_loop_depth--;
2891     }
2892 }
2893
2894 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2895 {
2896     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2897     glsl_src_param_t src0_param;
2898     const DWORD *control_values = NULL;
2899     const local_constant *constant;
2900
2901     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2902     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2903     {
2904         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2905         {
2906             if (constant->idx == ins->src[0].reg.idx)
2907             {
2908                 control_values = constant->value;
2909                 break;
2910             }
2911         }
2912     }
2913
2914     if(control_values) {
2915         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2916                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2917                        control_values[0], shader->baseShader.cur_loop_depth);
2918     } else {
2919         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2920         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2921                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2922                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2923     }
2924     shader->baseShader.cur_loop_depth++;
2925 }
2926
2927 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2928 {
2929     glsl_src_param_t src0_param;
2930
2931     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2932     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2933 }
2934
2935 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2936 {
2937     glsl_src_param_t src0_param;
2938     glsl_src_param_t src1_param;
2939
2940     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2941     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2942
2943     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2944             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2945 }
2946
2947 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2948 {
2949     shader_addline(ins->ctx->buffer, "} else {\n");
2950 }
2951
2952 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2953 {
2954     shader_addline(ins->ctx->buffer, "break;\n");
2955 }
2956
2957 /* FIXME: According to MSDN the compare is done per component. */
2958 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2959 {
2960     glsl_src_param_t src0_param;
2961     glsl_src_param_t src1_param;
2962
2963     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2964     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2965
2966     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2967             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2968 }
2969
2970 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2971 {
2972     shader_addline(ins->ctx->buffer, "}\n");
2973     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2974 }
2975
2976 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2977 {
2978     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2979 }
2980
2981 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2982 {
2983     glsl_src_param_t src1_param;
2984
2985     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2986     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
2987 }
2988
2989 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
2990 {
2991     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
2992      * function only suppresses the unhandled instruction warning
2993      */
2994 }
2995
2996 /*********************************************
2997  * Pixel Shader Specific Code begins here
2998  ********************************************/
2999 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3000 {
3001     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3002     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3003     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3004             ins->ctx->reg_maps->shader_version.minor);
3005     glsl_sample_function_t sample_function;
3006     DWORD sample_flags = 0;
3007     DWORD sampler_idx;
3008     DWORD mask = 0, swizzle;
3009
3010     /* 1.0-1.4: Use destination register as sampler source.
3011      * 2.0+: Use provided sampler source. */
3012     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3013     else sampler_idx = ins->src[1].reg.idx;
3014
3015     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3016     {
3017         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3018         WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3019
3020         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3021         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3022             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3023             switch (flags & ~WINED3DTTFF_PROJECTED) {
3024                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3025                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3026                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3027                 case WINED3DTTFF_COUNT4:
3028                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3029             }
3030         }
3031     }
3032     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3033     {
3034         DWORD src_mod = ins->src[0].modifiers;
3035
3036         if (src_mod == WINED3DSPSM_DZ) {
3037             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3038             mask = WINED3DSP_WRITEMASK_2;
3039         } else if (src_mod == WINED3DSPSM_DW) {
3040             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3041             mask = WINED3DSP_WRITEMASK_3;
3042         }
3043     } else {
3044         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3045         {
3046             /* ps 2.0 texldp instruction always divides by the fourth component. */
3047             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3048             mask = WINED3DSP_WRITEMASK_3;
3049         }
3050     }
3051
3052     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3053        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3054         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3055     }
3056
3057     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3058     mask |= sample_function.coord_mask;
3059
3060     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3061     else swizzle = ins->src[1].swizzle;
3062
3063     /* 1.0-1.3: Use destination register as coordinate source.
3064        1.4+: Use provided coordinate source register. */
3065     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3066     {
3067         char coord_mask[6];
3068         shader_glsl_write_mask_to_str(mask, coord_mask);
3069         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3070                 "T%u%s", sampler_idx, coord_mask);
3071     } else {
3072         glsl_src_param_t coord_param;
3073         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3074         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3075         {
3076             glsl_src_param_t bias;
3077             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3078             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3079                     "%s", coord_param.param_str);
3080         } else {
3081             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3082                     "%s", coord_param.param_str);
3083         }
3084     }
3085 }
3086
3087 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3088 {
3089     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3090     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3091     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3092     glsl_sample_function_t sample_function;
3093     glsl_src_param_t coord_param, dx_param, dy_param;
3094     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3095     DWORD sampler_idx;
3096     DWORD swizzle = ins->src[1].swizzle;
3097
3098     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3099     {
3100         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3101         shader_glsl_tex(ins);
3102         return;
3103     }
3104
3105     sampler_idx = ins->src[1].reg.idx;
3106     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3107        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3108         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3109     }
3110
3111     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3112     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3113     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3114     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3115
3116     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3117                                 "%s", coord_param.param_str);
3118 }
3119
3120 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3121 {
3122     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3123     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3124     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3125     glsl_sample_function_t sample_function;
3126     glsl_src_param_t coord_param, lod_param;
3127     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3128     DWORD sampler_idx;
3129     DWORD swizzle = ins->src[1].swizzle;
3130
3131     sampler_idx = ins->src[1].reg.idx;
3132     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3133        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3134         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3135     }
3136     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3137     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3138
3139     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3140
3141     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3142             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3143     {
3144         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3145          * However, they seem to work just fine in fragment shaders as well. */
3146         WARN("Using %s in fragment shader.\n", sample_function.name);
3147     }
3148     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3149             "%s", coord_param.param_str);
3150 }
3151
3152 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3153 {
3154     /* FIXME: Make this work for more than just 2D textures */
3155     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3156     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3157
3158     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3159     {
3160         char dst_mask[6];
3161
3162         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3163         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3164                 ins->dst[0].reg.idx, dst_mask);
3165     } else {
3166         DWORD reg = ins->src[0].reg.idx;
3167         DWORD src_mod = ins->src[0].modifiers;
3168         char dst_swizzle[6];
3169
3170         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3171
3172         if (src_mod == WINED3DSPSM_DZ) {
3173             glsl_src_param_t div_param;
3174             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3175             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3176
3177             if (mask_size > 1) {
3178                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3179             } else {
3180                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3181             }
3182         } else if (src_mod == WINED3DSPSM_DW) {
3183             glsl_src_param_t div_param;
3184             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3185             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3186
3187             if (mask_size > 1) {
3188                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3189             } else {
3190                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3191             }
3192         } else {
3193             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3194         }
3195     }
3196 }
3197
3198 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3199  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3200  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3201 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3202 {
3203     glsl_src_param_t src0_param;
3204     glsl_sample_function_t sample_function;
3205     DWORD sampler_idx = ins->dst[0].reg.idx;
3206     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3207     UINT mask_size;
3208
3209     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3210
3211     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3212      * scalar, and projected sampling would require 4.
3213      *
3214      * It is a dependent read - not valid with conditional NP2 textures
3215      */
3216     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3217     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3218
3219     switch(mask_size)
3220     {
3221         case 1:
3222             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3223                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3224             break;
3225
3226         case 2:
3227             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3228                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3229             break;
3230
3231         case 3:
3232             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3233                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3234             break;
3235
3236         default:
3237             FIXME("Unexpected mask size %u\n", mask_size);
3238             break;
3239     }
3240 }
3241
3242 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3243  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3244 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3245 {
3246     glsl_src_param_t src0_param;
3247     DWORD dstreg = ins->dst[0].reg.idx;
3248     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3249     DWORD dst_mask;
3250     unsigned int mask_size;
3251
3252     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3253     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3254     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3255
3256     if (mask_size > 1) {
3257         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3258     } else {
3259         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3260     }
3261 }
3262
3263 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3264  * Calculate the depth as dst.x / dst.y   */
3265 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3266 {
3267     glsl_dst_param_t dst_param;
3268
3269     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3270
3271     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3272      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3273      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3274      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3275      * >= 1.0 or < 0.0
3276      */
3277     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3278             dst_param.reg_name, dst_param.reg_name);
3279 }
3280
3281 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3282  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3283  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3284  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3285  */
3286 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3287 {
3288     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3289     DWORD dstreg = ins->dst[0].reg.idx;
3290     glsl_src_param_t src0_param;
3291
3292     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3293
3294     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3295     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3296 }
3297
3298 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3299  * Calculate the 1st of a 2-row matrix multiplication. */
3300 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3301 {
3302     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3303     DWORD reg = ins->dst[0].reg.idx;
3304     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3305     glsl_src_param_t src0_param;
3306
3307     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3308     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3309 }
3310
3311 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3312  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3313 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3314 {
3315     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3316     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3317     DWORD reg = ins->dst[0].reg.idx;
3318     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3319     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3320     glsl_src_param_t src0_param;
3321
3322     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3323     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3324     current_state->texcoord_w[current_state->current_row++] = reg;
3325 }
3326
3327 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3328 {
3329     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3330     DWORD reg = ins->dst[0].reg.idx;
3331     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3332     glsl_src_param_t src0_param;
3333     glsl_sample_function_t sample_function;
3334
3335     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3336     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3337
3338     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3339
3340     /* Sample the texture using the calculated coordinates */
3341     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3342 }
3343
3344 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3345  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3346 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3347 {
3348     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3349     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3350     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3351     glsl_src_param_t src0_param;
3352     DWORD reg = ins->dst[0].reg.idx;
3353     glsl_sample_function_t sample_function;
3354
3355     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3356     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3357
3358     /* Dependent read, not valid with conditional NP2 */
3359     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3360
3361     /* Sample the texture using the calculated coordinates */
3362     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3363
3364     current_state->current_row = 0;
3365 }
3366
3367 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3368  * Perform the 3rd row of a 3x3 matrix multiply */
3369 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3370 {
3371     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3372     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3373     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3374     glsl_src_param_t src0_param;
3375     char dst_mask[6];
3376     DWORD reg = ins->dst[0].reg.idx;
3377
3378     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3379
3380     shader_glsl_append_dst(ins->ctx->buffer, ins);
3381     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3382     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3383
3384     current_state->current_row = 0;
3385 }
3386
3387 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3388  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3389 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3390 {
3391     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3392     DWORD reg = ins->dst[0].reg.idx;
3393     glsl_src_param_t src0_param;
3394     glsl_src_param_t src1_param;
3395     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3396     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3397     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3398     glsl_sample_function_t sample_function;
3399
3400     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3401     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3402
3403     /* Perform the last matrix multiply operation */
3404     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3405     /* Reflection calculation */
3406     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3407
3408     /* Dependent read, not valid with conditional NP2 */
3409     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3410
3411     /* Sample the texture */
3412     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3413
3414     current_state->current_row = 0;
3415 }
3416
3417 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3418  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3419 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3420 {
3421     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3422     DWORD reg = ins->dst[0].reg.idx;
3423     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3424     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3425     glsl_src_param_t src0_param;
3426     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3427     glsl_sample_function_t sample_function;
3428
3429     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3430
3431     /* Perform the last matrix multiply operation */
3432     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3433
3434     /* Construct the eye-ray vector from w coordinates */
3435     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3436             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3437     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3438
3439     /* Dependent read, not valid with conditional NP2 */
3440     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3441
3442     /* Sample the texture using the calculated coordinates */
3443     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3444
3445     current_state->current_row = 0;
3446 }
3447
3448 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3449  * Apply a fake bump map transform.
3450  * texbem is pshader <= 1.3 only, this saves a few version checks
3451  */
3452 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3453 {
3454     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3455     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3456     glsl_sample_function_t sample_function;
3457     glsl_src_param_t coord_param;
3458     DWORD sampler_idx;
3459     DWORD mask;
3460     DWORD flags;
3461     char coord_mask[6];
3462
3463     sampler_idx = ins->dst[0].reg.idx;
3464     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3465
3466     /* Dependent read, not valid with conditional NP2 */
3467     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3468     mask = sample_function.coord_mask;
3469
3470     shader_glsl_write_mask_to_str(mask, coord_mask);
3471
3472     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3473          * so we can't let the GL handle this.
3474          */
3475     if (flags & WINED3DTTFF_PROJECTED) {
3476         DWORD div_mask=0;
3477         char coord_div_mask[3];
3478         switch (flags & ~WINED3DTTFF_PROJECTED) {
3479             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3480             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3481             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3482             case WINED3DTTFF_COUNT4:
3483             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3484         }
3485         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3486         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3487     }
3488
3489     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3490
3491     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3492             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3493             coord_param.param_str, coord_mask);
3494
3495     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3496     {
3497         glsl_src_param_t luminance_param;
3498         glsl_dst_param_t dst_param;
3499
3500         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3501         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3502
3503         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3504                 dst_param.reg_name, dst_param.mask_str,
3505                 luminance_param.param_str, sampler_idx, sampler_idx);
3506     }
3507 }
3508
3509 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3510 {
3511     glsl_src_param_t src0_param, src1_param;
3512     DWORD sampler_idx = ins->dst[0].reg.idx;
3513
3514     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3515     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3516
3517     shader_glsl_append_dst(ins->ctx->buffer, ins);
3518     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3519             src0_param.param_str, sampler_idx, src1_param.param_str);
3520 }
3521
3522 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3523  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3524 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3525 {
3526     glsl_src_param_t src0_param;
3527     DWORD sampler_idx = ins->dst[0].reg.idx;
3528     glsl_sample_function_t sample_function;
3529
3530     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3531
3532     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3533     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3534             "%s.wx", src0_param.reg_name);
3535 }
3536
3537 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3538  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3539 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3540 {
3541     glsl_src_param_t src0_param;
3542     DWORD sampler_idx = ins->dst[0].reg.idx;
3543     glsl_sample_function_t sample_function;
3544
3545     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3546
3547     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3548     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3549             "%s.yz", src0_param.reg_name);
3550 }
3551
3552 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3553  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3554 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3555 {
3556     glsl_src_param_t src0_param;
3557     DWORD sampler_idx = ins->dst[0].reg.idx;
3558     glsl_sample_function_t sample_function;
3559
3560     /* Dependent read, not valid with conditional NP2 */
3561     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3562     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3563
3564     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3565             "%s", src0_param.param_str);
3566 }
3567
3568 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3569  * If any of the first 3 components are < 0, discard this pixel */
3570 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3571 {
3572     glsl_dst_param_t dst_param;
3573
3574     /* The argument is a destination parameter, and no writemasks are allowed */
3575     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3576     if (ins->ctx->reg_maps->shader_version.major >= 2)
3577     {
3578         /* 2.0 shaders compare all 4 components in texkill */
3579         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3580     } else {
3581         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3582          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3583          * 4 components are defined, only the first 3 are used
3584          */
3585         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3586     }
3587 }
3588
3589 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3590  * dst = dot2(src0, src1) + src2 */
3591 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3592 {
3593     glsl_src_param_t src0_param;
3594     glsl_src_param_t src1_param;
3595     glsl_src_param_t src2_param;
3596     DWORD write_mask;
3597     unsigned int mask_size;
3598
3599     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3600     mask_size = shader_glsl_get_write_mask_size(write_mask);
3601
3602     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3603     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3604     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3605
3606     if (mask_size > 1) {
3607         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3608                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3609     } else {
3610         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3611                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3612     }
3613 }
3614
3615 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3616         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3617         enum vertexprocessing_mode vertexprocessing)
3618 {
3619     unsigned int i;
3620     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3621     WORD map = reg_maps->input_registers;
3622
3623     for (i = 0; map; map >>= 1, ++i)
3624     {
3625         const char *semantic_name;
3626         UINT semantic_idx;
3627         char reg_mask[6];
3628
3629         /* Unused */
3630         if (!(map & 1)) continue;
3631
3632         semantic_name = input_signature[i].semantic_name;
3633         semantic_idx = input_signature[i].semantic_idx;
3634         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3635
3636         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3637         {
3638             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3639                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3640                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3641             else
3642                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3643                         This->input_reg_map[i], reg_mask, reg_mask);
3644         }
3645         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3646         {
3647             if (semantic_idx == 0)
3648                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3649                         This->input_reg_map[i], reg_mask, reg_mask);
3650             else if (semantic_idx == 1)
3651                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3652                         This->input_reg_map[i], reg_mask, reg_mask);
3653             else
3654                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3655                         This->input_reg_map[i], reg_mask, reg_mask);
3656         }
3657         else
3658         {
3659             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3660                     This->input_reg_map[i], reg_mask, reg_mask);
3661         }
3662     }
3663 }
3664
3665 /*********************************************
3666  * Vertex Shader Specific Code begins here
3667  ********************************************/
3668
3669 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3670     glsl_program_key_t key;
3671
3672     key.vshader = entry->vshader;
3673     key.pshader = entry->pshader;
3674     key.vs_args = entry->vs_args;
3675     key.ps_args = entry->ps_args;
3676
3677     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3678     {
3679         ERR("Failed to insert program entry.\n");
3680     }
3681 }
3682
3683 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3684         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3685         struct ps_compile_args *ps_args) {
3686     struct wine_rb_entry *entry;
3687     glsl_program_key_t key;
3688
3689     key.vshader = vshader;
3690     key.pshader = pshader;
3691     key.vs_args = *vs_args;
3692     key.ps_args = *ps_args;
3693
3694     entry = wine_rb_get(&priv->program_lookup, &key);
3695     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3696 }
3697
3698 /* GL locking is done by the caller */
3699 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3700         struct glsl_shader_prog_link *entry)
3701 {
3702     glsl_program_key_t key;
3703
3704     key.vshader = entry->vshader;
3705     key.pshader = entry->pshader;
3706     key.vs_args = entry->vs_args;
3707     key.ps_args = entry->ps_args;
3708     wine_rb_remove(&priv->program_lookup, &key);
3709
3710     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3711     if (entry->vshader) list_remove(&entry->vshader_entry);
3712     if (entry->pshader) list_remove(&entry->pshader_entry);
3713     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3714     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3715     HeapFree(GetProcessHeap(), 0, entry);
3716 }
3717
3718 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3719         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3720         const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3721 {
3722     unsigned int i, j;
3723     const char *semantic_name_in;
3724     UINT semantic_idx_in;
3725     DWORD *set;
3726     DWORD in_idx;
3727     unsigned int in_count = vec4_varyings(3, gl_info);
3728     char reg_mask[6];
3729     char destination[50];
3730     WORD input_map, output_map;
3731
3732     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3733
3734     input_map = reg_maps_in->input_registers;
3735     for (i = 0; input_map; input_map >>= 1, ++i)
3736     {
3737         if (!(input_map & 1)) continue;
3738
3739         in_idx = map[i];
3740         /* Declared, but not read register */
3741         if (in_idx == ~0U) continue;
3742         if (in_idx >= (in_count + 2))
3743         {
3744             FIXME("More input varyings declared than supported, expect issues.\n");
3745             continue;
3746         }
3747
3748         if (in_idx == in_count) {
3749             sprintf(destination, "gl_FrontColor");
3750         } else if (in_idx == in_count + 1) {
3751             sprintf(destination, "gl_FrontSecondaryColor");
3752         } else {
3753             sprintf(destination, "IN[%u]", in_idx);
3754         }
3755
3756         semantic_name_in = input_signature[i].semantic_name;
3757         semantic_idx_in = input_signature[i].semantic_idx;
3758         set[in_idx] = ~0U;
3759
3760         output_map = reg_maps_out->output_registers;
3761         for (j = 0; output_map; output_map >>= 1, ++j)
3762         {
3763             DWORD mask;
3764
3765             if (!(output_map & 1)
3766                     || semantic_idx_in != output_signature[j].semantic_idx
3767                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3768                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3769                 continue;
3770
3771             set[in_idx] = mask;
3772             shader_glsl_write_mask_to_str(mask, reg_mask);
3773
3774             shader_addline(buffer, "%s%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3775                     destination, reg_mask, j, reg_mask);
3776         }
3777     }
3778
3779     for (i = 0; i < in_count + 2; ++i)
3780     {
3781         unsigned int size;
3782
3783         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3784             continue;
3785
3786         if (set[i] == ~0U) set[i] = 0;
3787
3788         size = 0;
3789         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3790         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3791         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3792         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3793         reg_mask[size] = '\0';
3794
3795         if (i == in_count) sprintf(destination, "gl_FrontColor");
3796         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3797         else sprintf(destination, "IN[%u]", i);
3798
3799         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3800         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3801     }
3802
3803     HeapFree(GetProcessHeap(), 0, set);
3804 }
3805
3806 /* GL locking is done by the caller */
3807 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3808         IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3809 {
3810     GLhandleARB ret = 0;
3811     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3812     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3813     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3814     unsigned int i;
3815     const char *semantic_name;
3816     UINT semantic_idx;
3817     char reg_mask[6];
3818     const struct wined3d_shader_signature_element *output_signature = vs->baseShader.output_signature;
3819     WORD map = vs->baseShader.reg_maps.output_registers;
3820
3821     shader_buffer_clear(buffer);
3822
3823     shader_addline(buffer, "#version 120\n");
3824     shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
3825
3826     if (ps_major < 3)
3827     {
3828         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3829
3830         for (i = 0; map; map >>= 1, ++i)
3831         {
3832             DWORD write_mask;
3833
3834             if (!(map & 1)) continue;
3835
3836             semantic_name = output_signature[i].semantic_name;
3837             semantic_idx = output_signature[i].semantic_idx;
3838             write_mask = output_signature[i].mask;
3839             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3840
3841             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3842             {
3843                 if (semantic_idx == 0)
3844                     shader_addline(buffer, "gl_FrontColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3845                             reg_mask, i, reg_mask);
3846                 else if (semantic_idx == 1)
3847                     shader_addline(buffer, "gl_FrontSecondaryColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3848                             reg_mask, i, reg_mask);
3849             }
3850             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3851             {
3852                 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3853                         reg_mask, i, reg_mask);
3854             }
3855             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3856             {
3857                 if (semantic_idx < 8)
3858                 {
3859                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3860                         write_mask |= WINED3DSP_WRITEMASK_3;
3861
3862                     shader_addline(buffer, "gl_TexCoord[%u]%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3863                             semantic_idx, reg_mask, i, reg_mask);
3864                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3865                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3866                 }
3867             }
3868             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3869             {
3870                 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3871             }
3872             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3873             {
3874                 shader_addline(buffer, "gl_FogFragCoord = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3875             }
3876         }
3877         shader_addline(buffer, "}\n");
3878
3879     }
3880     else
3881     {
3882         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3883         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3884         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3885
3886         /* First, sort out position and point size. Those are not passed to the pixel shader */
3887         for (i = 0; map; map >>= 1, ++i)
3888         {
3889             if (!(map & 1)) continue;
3890
3891             semantic_name = output_signature[i].semantic_name;
3892             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3893
3894             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3895             {
3896                 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3897                         reg_mask, i, reg_mask);
3898             }
3899             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3900             {
3901                 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3902             }
3903         }
3904
3905         /* Then, fix the pixel shader input */
3906         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3907                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3908
3909         shader_addline(buffer, "}\n");
3910     }
3911
3912     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3913     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3914     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer->buffer, NULL));
3915     checkGLcall("glShaderSourceARB(ret, 1, &buffer->buffer, NULL)");
3916     GL_EXTCALL(glCompileShaderARB(ret));
3917     checkGLcall("glCompileShaderARB(ret)");
3918
3919     return ret;
3920 }
3921
3922 /* GL locking is done by the caller */
3923 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3924         GLhandleARB programId, char prefix)
3925 {
3926     const local_constant *lconst;
3927     GLint tmp_loc;
3928     const float *value;
3929     char glsl_name[8];
3930
3931     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3932         value = (const float *)lconst->value;
3933         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3934         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3935         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3936     }
3937     checkGLcall("Hardcoding local constants");
3938 }
3939
3940 /* GL locking is done by the caller */
3941 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3942         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3943         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3944 {
3945     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3946     const struct wined3d_gl_info *gl_info = context->gl_info;
3947     CONST DWORD *function = This->baseShader.function;
3948     struct shader_glsl_ctx_priv priv_ctx;
3949
3950     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3951     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3952
3953     memset(&priv_ctx, 0, sizeof(priv_ctx));
3954     priv_ctx.cur_ps_args = args;
3955     priv_ctx.cur_np2fixup_info = np2fixup_info;
3956
3957     shader_addline(buffer, "#version 120\n");
3958
3959     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3960     {
3961         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3962     }
3963     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3964     {
3965         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3966          * drivers write a warning if we don't do so
3967          */
3968         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3969     }
3970     if (gl_info->supported[EXT_GPU_SHADER4])
3971     {
3972         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3973     }
3974
3975     /* Base Declarations */
3976     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3977
3978     /* Pack 3.0 inputs */
3979     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3980     {
3981         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
3982                 This->baseShader.input_signature, reg_maps, args->vp_mode);
3983     }
3984
3985     /* Base Shader Body */
3986     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3987
3988     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3989     if (reg_maps->shader_version.major < 2)
3990     {
3991         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3992         shader_addline(buffer, "gl_FragData[0] = R0;\n");
3993     }
3994
3995     if (args->srgb_correction)
3996     {
3997         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
3998         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
3999         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4000         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4001         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4002         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4003     }
4004     /* Pixel shader < 3.0 do not replace the fog stage.
4005      * This implements linear fog computation and blending.
4006      * TODO: non linear fog
4007      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4008      * -1/(e-s) and e/(e-s) respectively.
4009      */
4010     if (reg_maps->shader_version.major < 3)
4011     {
4012         switch(args->fog) {
4013             case FOG_OFF: break;
4014             case FOG_LINEAR:
4015                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4016                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4017                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4018                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4019                 break;
4020             case FOG_EXP:
4021                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4022                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4023                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4024                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4025                 break;
4026             case FOG_EXP2:
4027                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4028                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4029                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4030                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4031                 break;
4032         }
4033     }
4034
4035     shader_addline(buffer, "}\n");
4036
4037     TRACE("Compiling shader object %u\n", shader_obj);
4038     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4039     GL_EXTCALL(glCompileShaderARB(shader_obj));
4040     print_glsl_info_log(gl_info, shader_obj);
4041
4042     /* Store the shader object */
4043     return shader_obj;
4044 }
4045
4046 /* GL locking is done by the caller */
4047 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4048         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4049         const struct vs_compile_args *args)
4050 {
4051     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4052     const struct wined3d_gl_info *gl_info = context->gl_info;
4053     CONST DWORD *function = This->baseShader.function;
4054     struct shader_glsl_ctx_priv priv_ctx;
4055
4056     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4057     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4058
4059     shader_addline(buffer, "#version 120\n");
4060
4061     if (gl_info->supported[EXT_GPU_SHADER4])
4062     {
4063         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4064     }
4065
4066     memset(&priv_ctx, 0, sizeof(priv_ctx));
4067     priv_ctx.cur_vs_args = args;
4068
4069     /* Base Declarations */
4070     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4071
4072     /* Base Shader Body */
4073     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4074
4075     /* Unpack outputs */
4076     shader_addline(buffer, "order_ps_input(OUT);\n");
4077
4078     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4079      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4080      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4081      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4082      */
4083     if(args->fog_src == VS_FOG_Z) {
4084         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4085     } else if (!reg_maps->fog) {
4086         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4087     }
4088
4089     /* Write the final position.
4090      *
4091      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4092      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4093      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4094      * contains 1.0 to allow a mad.
4095      */
4096     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4097     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4098     if(args->clip_enabled) {
4099         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4100     }
4101
4102     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4103      *
4104      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4105      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4106      * which is the same as z = z * 2 - w.
4107      */
4108     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4109
4110     shader_addline(buffer, "}\n");
4111
4112     TRACE("Compiling shader object %u\n", shader_obj);
4113     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4114     GL_EXTCALL(glCompileShaderARB(shader_obj));
4115     print_glsl_info_log(gl_info, shader_obj);
4116
4117     return shader_obj;
4118 }
4119
4120 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4121         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4122         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4123 {
4124     UINT i;
4125     DWORD new_size;
4126     struct glsl_ps_compiled_shader *new_array;
4127     struct glsl_pshader_private    *shader_data;
4128     struct ps_np2fixup_info        *np2fixup = NULL;
4129     GLhandleARB ret;
4130
4131     if (!shader->baseShader.backend_data)
4132     {
4133         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4134         if (!shader->baseShader.backend_data)
4135         {
4136             ERR("Failed to allocate backend data.\n");
4137             return 0;
4138         }
4139     }
4140     shader_data = shader->baseShader.backend_data;
4141
4142     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4143      * so a linear search is more performant than a hashmap or a binary search
4144      * (cache coherency etc)
4145      */
4146     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4147         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4148             if(args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4149             return shader_data->gl_shaders[i].prgId;
4150         }
4151     }
4152
4153     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4154     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4155         if (shader_data->num_gl_shaders)
4156         {
4157             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4158             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4159                                     new_size * sizeof(*shader_data->gl_shaders));
4160         } else {
4161             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4162             new_size = 1;
4163         }
4164
4165         if(!new_array) {
4166             ERR("Out of memory\n");
4167             return 0;
4168         }
4169         shader_data->gl_shaders = new_array;
4170         shader_data->shader_array_size = new_size;
4171     }
4172
4173     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4174
4175     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4176     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4177
4178     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4179             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4180
4181     shader_buffer_clear(buffer);
4182     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4183     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4184     *np2fixup_info = np2fixup;
4185
4186     return ret;
4187 }
4188
4189 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4190                                  const DWORD use_map) {
4191     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4192     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4193     return stored->fog_src == new->fog_src;
4194 }
4195
4196 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4197         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4198         const struct vs_compile_args *args)
4199 {
4200     UINT i;
4201     DWORD new_size;
4202     struct glsl_vs_compiled_shader *new_array;
4203     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4204     struct glsl_vshader_private *shader_data;
4205     GLhandleARB ret;
4206
4207     if (!shader->baseShader.backend_data)
4208     {
4209         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4210         if (!shader->baseShader.backend_data)
4211         {
4212             ERR("Failed to allocate backend data.\n");
4213             return 0;
4214         }
4215     }
4216     shader_data = shader->baseShader.backend_data;
4217
4218     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4219      * so a linear search is more performant than a hashmap or a binary search
4220      * (cache coherency etc)
4221      */
4222     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4223         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4224             return shader_data->gl_shaders[i].prgId;
4225         }
4226     }
4227
4228     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4229
4230     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4231         if (shader_data->num_gl_shaders)
4232         {
4233             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4234             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4235                                     new_size * sizeof(*shader_data->gl_shaders));
4236         } else {
4237             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4238             new_size = 1;
4239         }
4240
4241         if(!new_array) {
4242             ERR("Out of memory\n");
4243             return 0;
4244         }
4245         shader_data->gl_shaders = new_array;
4246         shader_data->shader_array_size = new_size;
4247     }
4248
4249     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4250
4251     shader_buffer_clear(buffer);
4252     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4253     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4254
4255     return ret;
4256 }
4257
4258 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4259  * It sets the programId on the current StateBlock (because it should be called
4260  * inside of the DrawPrimitive() part of the render loop).
4261  *
4262  * If a program for the given combination does not exist, create one, and store
4263  * the program in the hash table.  If it creates a program, it will link the
4264  * given objects, too.
4265  */
4266
4267 /* GL locking is done by the caller */
4268 static void set_glsl_shader_program(const struct wined3d_context *context,
4269         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4270 {
4271     IWineD3DVertexShader *vshader = use_vs ? device->stateBlock->vertexShader : NULL;
4272     IWineD3DPixelShader *pshader = use_ps ? device->stateBlock->pixelShader : NULL;
4273     const struct wined3d_gl_info *gl_info = context->gl_info;
4274     struct shader_glsl_priv *priv = device->shader_priv;
4275     struct glsl_shader_prog_link *entry    = NULL;
4276     GLhandleARB programId                  = 0;
4277     GLhandleARB reorder_shader_id          = 0;
4278     unsigned int i;
4279     char glsl_name[8];
4280     struct ps_compile_args ps_compile_args;
4281     struct vs_compile_args vs_compile_args;
4282
4283     if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4284     if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4285
4286     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4287     if (entry) {
4288         priv->glsl_program = entry;
4289         return;
4290     }
4291
4292     /* If we get to this point, then no matching program exists, so we create one */
4293     programId = GL_EXTCALL(glCreateProgramObjectARB());
4294     TRACE("Created new GLSL shader program %u\n", programId);
4295
4296     /* Create the entry */
4297     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4298     entry->programId = programId;
4299     entry->vshader = vshader;
4300     entry->pshader = pshader;
4301     entry->vs_args = vs_compile_args;
4302     entry->ps_args = ps_compile_args;
4303     entry->constant_version = 0;
4304     entry->np2Fixup_info = NULL;
4305     /* Add the hash table entry */
4306     add_glsl_program_entry(priv, entry);
4307
4308     /* Set the current program */
4309     priv->glsl_program = entry;
4310
4311     /* Attach GLSL vshader */
4312     if (vshader)
4313     {
4314         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4315                 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4316         WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4317         char tmp_name[10];
4318
4319         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4320         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4321         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4322         checkGLcall("glAttachObjectARB");
4323         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4324          * is destroyed
4325          */
4326         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4327
4328         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4329         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4330         checkGLcall("glAttachObjectARB");
4331
4332         /* Bind vertex attributes to a corresponding index number to match
4333          * the same index numbers as ARB_vertex_programs (makes loading
4334          * vertex attributes simpler).  With this method, we can use the
4335          * exact same code to load the attributes later for both ARB and
4336          * GLSL shaders.
4337          *
4338          * We have to do this here because we need to know the Program ID
4339          * in order to make the bindings work, and it has to be done prior
4340          * to linking the GLSL program. */
4341         for (i = 0; map; map >>= 1, ++i)
4342         {
4343             if (!(map & 1)) continue;
4344
4345             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4346             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4347         }
4348         checkGLcall("glBindAttribLocationARB");
4349
4350         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4351     }
4352
4353     /* Attach GLSL pshader */
4354     if (pshader)
4355     {
4356         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4357                 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4358         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4359         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4360         checkGLcall("glAttachObjectARB");
4361
4362         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4363     }
4364
4365     /* Link the program */
4366     TRACE("Linking GLSL shader program %u\n", programId);
4367     GL_EXTCALL(glLinkProgramARB(programId));
4368     shader_glsl_validate_link(gl_info, programId);
4369
4370     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4371             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4372     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4373     {
4374         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4375         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4376     }
4377     for (i = 0; i < MAX_CONST_I; ++i)
4378     {
4379         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4380         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4381     }
4382     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4383             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4384     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4385     {
4386         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4387         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4388     }
4389     for (i = 0; i < MAX_CONST_I; ++i)
4390     {
4391         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4392         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4393     }
4394
4395     if(pshader) {
4396         char name[32];
4397
4398         for(i = 0; i < MAX_TEXTURES; i++) {
4399             sprintf(name, "bumpenvmat%u", i);
4400             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4401             sprintf(name, "luminancescale%u", i);
4402             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4403             sprintf(name, "luminanceoffset%u", i);
4404             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4405         }
4406
4407         if (ps_compile_args.np2_fixup) {
4408             if (entry->np2Fixup_info) {
4409                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4410             } else {
4411                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4412             }
4413         }
4414     }
4415
4416     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4417     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4418     checkGLcall("Find glsl program uniform locations");
4419
4420     if (pshader
4421             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4422             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4423     {
4424         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4425         entry->vertex_color_clamp = GL_FALSE;
4426     } else {
4427         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4428     }
4429
4430     /* Set the shader to allow uniform loading on it */
4431     GL_EXTCALL(glUseProgramObjectARB(programId));
4432     checkGLcall("glUseProgramObjectARB(programId)");
4433
4434     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4435      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4436      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4437      * vertex shader with fixed function pixel processing is used we make sure that the card
4438      * supports enough samplers to allow the max number of vertex samplers with all possible
4439      * fixed function fragment processing setups. So once the program is linked these samplers
4440      * won't change.
4441      */
4442     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4443     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4444
4445     /* If the local constants do not have to be loaded with the environment constants,
4446      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4447      * later
4448      */
4449     if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4450     {
4451         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4452     }
4453     if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4454     {
4455         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4456     }
4457 }
4458
4459 /* GL locking is done by the caller */
4460 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4461 {
4462     GLhandleARB program_id;
4463     GLhandleARB vshader_id, pshader_id;
4464     const char *blt_pshader;
4465
4466     static const char *blt_vshader =
4467         "#version 120\n"
4468         "void main(void)\n"
4469         "{\n"
4470         "    gl_Position = gl_Vertex;\n"
4471         "    gl_FrontColor = vec4(1.0);\n"
4472         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4473         "}\n";
4474
4475     static const char * const blt_pshaders_full[tex_type_count] =
4476     {
4477         /* tex_1d */
4478         NULL,
4479         /* tex_2d */
4480         "#version 120\n"
4481         "uniform sampler2D sampler;\n"
4482         "void main(void)\n"
4483         "{\n"
4484         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4485         "}\n",
4486         /* tex_3d */
4487         NULL,
4488         /* tex_cube */
4489         "#version 120\n"
4490         "uniform samplerCube sampler;\n"
4491         "void main(void)\n"
4492         "{\n"
4493         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4494         "}\n",
4495         /* tex_rect */
4496         "#version 120\n"
4497         "#extension GL_ARB_texture_rectangle : enable\n"
4498         "uniform sampler2DRect sampler;\n"
4499         "void main(void)\n"
4500         "{\n"
4501         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4502         "}\n",
4503     };
4504
4505     static const char * const blt_pshaders_masked[tex_type_count] =
4506     {
4507         /* tex_1d */
4508         NULL,
4509         /* tex_2d */
4510         "#version 120\n"
4511         "uniform sampler2D sampler;\n"
4512         "uniform vec4 mask;\n"
4513         "void main(void)\n"
4514         "{\n"
4515         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4516         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4517         "}\n",
4518         /* tex_3d */
4519         NULL,
4520         /* tex_cube */
4521         "#version 120\n"
4522         "uniform samplerCube sampler;\n"
4523         "uniform vec4 mask;\n"
4524         "void main(void)\n"
4525         "{\n"
4526         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4527         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4528         "}\n",
4529         /* tex_rect */
4530         "#version 120\n"
4531         "#extension GL_ARB_texture_rectangle : enable\n"
4532         "uniform sampler2DRect sampler;\n"
4533         "uniform vec4 mask;\n"
4534         "void main(void)\n"
4535         "{\n"
4536         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4537         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4538         "}\n",
4539     };
4540
4541     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4542     if (!blt_pshader)
4543     {
4544         FIXME("tex_type %#x not supported\n", tex_type);
4545         tex_type = tex_2d;
4546     }
4547
4548     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4549     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, &blt_vshader, NULL));
4550     GL_EXTCALL(glCompileShaderARB(vshader_id));
4551
4552     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4553     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshader, NULL));
4554     GL_EXTCALL(glCompileShaderARB(pshader_id));
4555
4556     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4557     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4558     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4559     GL_EXTCALL(glLinkProgramARB(program_id));
4560
4561     shader_glsl_validate_link(gl_info, program_id);
4562
4563     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4564      * is destroyed
4565      */
4566     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4567     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4568     return program_id;
4569 }
4570
4571 /* GL locking is done by the caller */
4572 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4573 {
4574     const struct wined3d_gl_info *gl_info = context->gl_info;
4575     IWineD3DDeviceImpl *device = context->swapchain->device;
4576     struct shader_glsl_priv *priv = device->shader_priv;
4577     GLhandleARB program_id = 0;
4578     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4579
4580     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4581
4582     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4583     else priv->glsl_program = NULL;
4584
4585     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4586
4587     if (old_vertex_color_clamp != current_vertex_color_clamp)
4588     {
4589         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4590         {
4591             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4592             checkGLcall("glClampColorARB");
4593         }
4594         else
4595         {
4596             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4597         }
4598     }
4599
4600     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4601     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4602     GL_EXTCALL(glUseProgramObjectARB(program_id));
4603     checkGLcall("glUseProgramObjectARB");
4604
4605     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4606      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4607      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4608     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4609     {
4610         shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4611     }
4612 }
4613
4614 /* GL locking is done by the caller */
4615 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface,
4616         enum tex_types tex_type, const SIZE *ds_mask_size)
4617 {
4618     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4619     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4620     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4621     struct shader_glsl_priv *priv = This->shader_priv;
4622     GLhandleARB *blt_program;
4623     GLint loc;
4624
4625     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4626     if (!*blt_program)
4627     {
4628         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4629         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4630         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4631         GL_EXTCALL(glUniform1iARB(loc, 0));
4632     }
4633     else
4634     {
4635         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4636     }
4637
4638     if (masked)
4639     {
4640         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4641         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4642     }
4643 }
4644
4645 /* GL locking is done by the caller */
4646 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4647     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4648     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4649     struct shader_glsl_priv *priv = This->shader_priv;
4650     GLhandleARB program_id;
4651
4652     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4653     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4654
4655     GL_EXTCALL(glUseProgramObjectARB(program_id));
4656     checkGLcall("glUseProgramObjectARB");
4657 }
4658
4659 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4660     const struct list *linked_programs;
4661     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4662     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4663     struct shader_glsl_priv *priv = device->shader_priv;
4664     const struct wined3d_gl_info *gl_info;
4665     struct wined3d_context *context;
4666
4667     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4668      * can be called from IWineD3DBaseShader::Release
4669      */
4670     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4671
4672     if(pshader) {
4673         struct glsl_pshader_private *shader_data;
4674         shader_data = This->baseShader.backend_data;
4675         if(!shader_data || shader_data->num_gl_shaders == 0)
4676         {
4677             HeapFree(GetProcessHeap(), 0, shader_data);
4678             This->baseShader.backend_data = NULL;
4679             return;
4680         }
4681
4682         context = context_acquire(device, NULL);
4683         gl_info = context->gl_info;
4684
4685         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4686         {
4687             ENTER_GL();
4688             shader_glsl_select(context, FALSE, FALSE);
4689             LEAVE_GL();
4690         }
4691     } else {
4692         struct glsl_vshader_private *shader_data;
4693         shader_data = This->baseShader.backend_data;
4694         if(!shader_data || shader_data->num_gl_shaders == 0)
4695         {
4696             HeapFree(GetProcessHeap(), 0, shader_data);
4697             This->baseShader.backend_data = NULL;
4698             return;
4699         }
4700
4701         context = context_acquire(device, NULL);
4702         gl_info = context->gl_info;
4703
4704         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4705         {
4706             ENTER_GL();
4707             shader_glsl_select(context, FALSE, FALSE);
4708             LEAVE_GL();
4709         }
4710     }
4711
4712     linked_programs = &This->baseShader.linked_programs;
4713
4714     TRACE("Deleting linked programs\n");
4715     if (linked_programs->next) {
4716         struct glsl_shader_prog_link *entry, *entry2;
4717
4718         ENTER_GL();
4719         if(pshader) {
4720             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4721                 delete_glsl_program_entry(priv, gl_info, entry);
4722             }
4723         } else {
4724             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4725                 delete_glsl_program_entry(priv, gl_info, entry);
4726             }
4727         }
4728         LEAVE_GL();
4729     }
4730
4731     if(pshader) {
4732         UINT i;
4733         struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
4734
4735         ENTER_GL();
4736         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4737             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4738             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4739             checkGLcall("glDeleteObjectARB");
4740         }
4741         LEAVE_GL();
4742         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4743     }
4744     else
4745     {
4746         UINT i;
4747         struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
4748
4749         ENTER_GL();
4750         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4751             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4752             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4753             checkGLcall("glDeleteObjectARB");
4754         }
4755         LEAVE_GL();
4756         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4757     }
4758
4759     HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
4760     This->baseShader.backend_data = NULL;
4761
4762     context_release(context);
4763 }
4764
4765 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4766 {
4767     const glsl_program_key_t *k = key;
4768     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4769             const struct glsl_shader_prog_link, program_lookup_entry);
4770     int cmp;
4771
4772     if (k->vshader > prog->vshader) return 1;
4773     else if (k->vshader < prog->vshader) return -1;
4774
4775     if (k->pshader > prog->pshader) return 1;
4776     else if (k->pshader < prog->pshader) return -1;
4777
4778     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4779     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4780
4781     return 0;
4782 }
4783
4784 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4785 {
4786     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4787     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4788
4789     if (!mem)
4790     {
4791         ERR("Failed to allocate memory\n");
4792         return FALSE;
4793     }
4794
4795     heap->entries = mem;
4796     heap->entries[1].version = 0;
4797     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4798     heap->size = 1;
4799
4800     return TRUE;
4801 }
4802
4803 static void constant_heap_free(struct constant_heap *heap)
4804 {
4805     HeapFree(GetProcessHeap(), 0, heap->entries);
4806 }
4807
4808 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4809 {
4810     wined3d_rb_alloc,
4811     wined3d_rb_realloc,
4812     wined3d_rb_free,
4813     glsl_program_key_compare,
4814 };
4815
4816 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4817     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4818     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4819     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4820     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4821             gl_info->limits.glsl_ps_float_constants)) + 1;
4822
4823     if (!shader_buffer_init(&priv->shader_buffer))
4824     {
4825         ERR("Failed to initialize shader buffer.\n");
4826         goto fail;
4827     }
4828
4829     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4830     if (!priv->stack)
4831     {
4832         ERR("Failed to allocate memory.\n");
4833         goto fail;
4834     }
4835
4836     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4837     {
4838         ERR("Failed to initialize vertex shader constant heap\n");
4839         goto fail;
4840     }
4841
4842     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4843     {
4844         ERR("Failed to initialize pixel shader constant heap\n");
4845         goto fail;
4846     }
4847
4848     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4849     {
4850         ERR("Failed to initialize rbtree.\n");
4851         goto fail;
4852     }
4853
4854     priv->next_constant_version = 1;
4855
4856     This->shader_priv = priv;
4857     return WINED3D_OK;
4858
4859 fail:
4860     constant_heap_free(&priv->pconst_heap);
4861     constant_heap_free(&priv->vconst_heap);
4862     HeapFree(GetProcessHeap(), 0, priv->stack);
4863     shader_buffer_free(&priv->shader_buffer);
4864     HeapFree(GetProcessHeap(), 0, priv);
4865     return E_OUTOFMEMORY;
4866 }
4867
4868 /* Context activation is done by the caller. */
4869 static void shader_glsl_free(IWineD3DDevice *iface) {
4870     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4871     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4872     struct shader_glsl_priv *priv = This->shader_priv;
4873     int i;
4874
4875     ENTER_GL();
4876     for (i = 0; i < tex_type_count; ++i)
4877     {
4878         if (priv->depth_blt_program_full[i])
4879         {
4880             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4881         }
4882         if (priv->depth_blt_program_masked[i])
4883         {
4884             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4885         }
4886     }
4887     LEAVE_GL();
4888
4889     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4890     constant_heap_free(&priv->pconst_heap);
4891     constant_heap_free(&priv->vconst_heap);
4892     HeapFree(GetProcessHeap(), 0, priv->stack);
4893     shader_buffer_free(&priv->shader_buffer);
4894
4895     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4896     This->shader_priv = NULL;
4897 }
4898
4899 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4900     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4901     return FALSE;
4902 }
4903
4904 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4905 {
4906     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4907      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4908      * on the version of NV_vertex_program.
4909      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4910      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4911      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4912      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4913      */
4914     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4915             || gl_info->limits.arb_ps_instructions <= 512)
4916         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4917     else
4918         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4919     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4920     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4921
4922     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4923      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4924      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4925      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4926      * in max native instructions. Intel and others also offer the info in this extension but they
4927      * don't support GLSL (at least on Windows).
4928      *
4929      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4930      * of instructions is 512 or less we have to do with ps2.0 hardware.
4931      * 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.
4932      */
4933     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4934             || gl_info->limits.arb_ps_instructions <= 512)
4935         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4936     else
4937         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4938
4939     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4940
4941     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4942      * Direct3D minimum requirement.
4943      *
4944      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4945      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4946      *
4947      * The problem is that the refrast clamps temporary results in the shader to
4948      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4949      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4950      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4951      * offer a way to query this.
4952      */
4953     pCaps->PixelShader1xMaxValue = 8.0;
4954     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4955
4956     pCaps->VSClipping = TRUE;
4957 }
4958
4959 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4960 {
4961     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4962     {
4963         TRACE("Checking support for fixup:\n");
4964         dump_color_fixup_desc(fixup);
4965     }
4966
4967     /* We support everything except YUV conversions. */
4968     if (!is_complex_fixup(fixup))
4969     {
4970         TRACE("[OK]\n");
4971         return TRUE;
4972     }
4973
4974     TRACE("[FAILED]\n");
4975     return FALSE;
4976 }
4977
4978 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4979 {
4980     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4981     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4982     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4983     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4984     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4985     /* WINED3DSIH_BREAKP        */ NULL,
4986     /* WINED3DSIH_CALL          */ shader_glsl_call,
4987     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4988     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4989     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4990     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4991     /* WINED3DSIH_CUT           */ NULL,
4992     /* WINED3DSIH_DCL           */ NULL,
4993     /* WINED3DSIH_DEF           */ NULL,
4994     /* WINED3DSIH_DEFB          */ NULL,
4995     /* WINED3DSIH_DEFI          */ NULL,
4996     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
4997     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4998     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4999     /* WINED3DSIH_DST           */ shader_glsl_dst,
5000     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5001     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5002     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5003     /* WINED3DSIH_EMIT          */ NULL,
5004     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5005     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5006     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5007     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5008     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5009     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5010     /* WINED3DSIH_IADD          */ NULL,
5011     /* WINED3DSIH_IF            */ shader_glsl_if,
5012     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5013     /* WINED3DSIH_IGE           */ NULL,
5014     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5015     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5016     /* WINED3DSIH_LOG           */ shader_glsl_log,
5017     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5018     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5019     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5020     /* WINED3DSIH_LT            */ NULL,
5021     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5022     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5023     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5024     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5025     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5026     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5027     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5028     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5029     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5030     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5031     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5032     /* WINED3DSIH_NOP           */ NULL,
5033     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5034     /* WINED3DSIH_PHASE         */ NULL,
5035     /* WINED3DSIH_POW           */ shader_glsl_pow,
5036     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5037     /* WINED3DSIH_REP           */ shader_glsl_rep,
5038     /* WINED3DSIH_RET           */ shader_glsl_ret,
5039     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5040     /* WINED3DSIH_SETP          */ NULL,
5041     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5042     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5043     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5044     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5045     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5046     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5047     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5048     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5049     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5050     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5051     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5052     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5053     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5054     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5055     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5056     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5057     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5058     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5059     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5060     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5061     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5062     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5063     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5064     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5065     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5066     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5067     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5068 };
5069
5070 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5071     SHADER_HANDLER hw_fct;
5072
5073     /* Select handler */
5074     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5075
5076     /* Unhandled opcode */
5077     if (!hw_fct)
5078     {
5079         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5080         return;
5081     }
5082     hw_fct(ins);
5083
5084     shader_glsl_add_instruction_modifiers(ins);
5085 }
5086
5087 const shader_backend_t glsl_shader_backend = {
5088     shader_glsl_handle_instruction,
5089     shader_glsl_select,
5090     shader_glsl_select_depth_blt,
5091     shader_glsl_deselect_depth_blt,
5092     shader_glsl_update_float_vertex_constants,
5093     shader_glsl_update_float_pixel_constants,
5094     shader_glsl_load_constants,
5095     shader_glsl_load_np2fixup_constants,
5096     shader_glsl_destroy,
5097     shader_glsl_alloc,
5098     shader_glsl_free,
5099     shader_glsl_dirty_const,
5100     shader_glsl_get_caps,
5101     shader_glsl_color_fixup_supported,
5102 };