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

   1 /*
   2  * GLSL pixel and vertex shader implementation
   3  *
   4  * Copyright 2006 Jason Green
   5  * Copyright 2006-2007 Henri Verbeet
   6  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
   7  * Copyright 2009 Henri Verbeet for CodeWeavers
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  22  */
  23
  24 /*
  25  * D3D shader asm has swizzles on source parameters, and write masks for
  26  * destination parameters. GLSL uses swizzles for both. The result of this is
  27  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  28  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  29  * mask for the destination parameter into account.
  30  */
  31
  32 #include "config.h"
  33 #include <limits.h>
  34 #include <stdio.h>
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define GLINFO_LOCATION      (*gl_info)
  43
  44 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  45 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  46 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  47 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  48
  49 typedef struct {
  50     char reg_name[150];
  51     char mask_str[6];
  52 } glsl_dst_param_t;
  53
  54 typedef struct {
  55     char reg_name[150];
  56     char param_str[200];
  57 } glsl_src_param_t;
  58
  59 typedef struct {
  60     const char *name;
  61     DWORD coord_mask;
  62 } glsl_sample_function_t;
  63
  64 enum heap_node_op
  65 {
  66     HEAP_NODE_TRAVERSE_LEFT,
  67     HEAP_NODE_TRAVERSE_RIGHT,
  68     HEAP_NODE_POP,
  69 };
  70
  71 struct constant_entry
  72 {
  73     unsigned int idx;
  74     unsigned int version;
  75 };
  76
  77 struct constant_heap
  78 {
  79     struct constant_entry *entries;
  80     unsigned int *positions;
  81     unsigned int size;
  82 };
  83
  84 /* GLSL shader private data */
  85 struct shader_glsl_priv {
  86     struct wined3d_shader_buffer shader_buffer;
  87     struct wine_rb_tree program_lookup;
  88     struct glsl_shader_prog_link *glsl_program;
  89     struct constant_heap vconst_heap;
  90     struct constant_heap pconst_heap;
  91     unsigned char *stack;
  92     GLhandleARB depth_blt_program[tex_type_count];
  93     UINT next_constant_version;
  94 };
  95
  96 /* Struct to maintain data about a linked GLSL program */
  97 struct glsl_shader_prog_link {
  98     struct wine_rb_entry        program_lookup_entry;
  99     struct list                 vshader_entry;
 100     struct list                 pshader_entry;
 101     GLhandleARB                 programId;
 102     GLint                       *vuniformF_locations;
 103     GLint                       *puniformF_locations;
 104     GLint                       vuniformI_locations[MAX_CONST_I];
 105     GLint                       puniformI_locations[MAX_CONST_I];
 106     GLint                       posFixup_location;
 107     GLint                       np2Fixup_location;
 108     GLint                       bumpenvmat_location[MAX_TEXTURES];
 109     GLint                       luminancescale_location[MAX_TEXTURES];
 110     GLint                       luminanceoffset_location[MAX_TEXTURES];
 111     GLint                       ycorrection_location;
 112     GLenum                      vertex_color_clamp;
 113     IWineD3DVertexShader        *vshader;
 114     IWineD3DPixelShader         *pshader;
 115     struct vs_compile_args      vs_args;
 116     struct ps_compile_args      ps_args;
 117     UINT                        constant_version;
 118     const struct ps_np2fixup_info *np2Fixup_info;
 119 };
 120
 121 typedef struct {
 122     IWineD3DVertexShader        *vshader;
 123     IWineD3DPixelShader         *pshader;
 124     struct ps_compile_args      ps_args;
 125     struct vs_compile_args      vs_args;
 126 } glsl_program_key_t;
 127
 128 struct shader_glsl_ctx_priv {
 129     const struct vs_compile_args    *cur_vs_args;
 130     const struct ps_compile_args    *cur_ps_args;
 131     struct ps_np2fixup_info         *cur_np2fixup_info;
 132 };
 133
 134 struct glsl_ps_compiled_shader
 135 {
 136     struct ps_compile_args          args;
 137     struct ps_np2fixup_info         np2fixup;
 138     GLhandleARB                     prgId;
 139 };
 140
 141 struct glsl_pshader_private
 142 {
 143     struct glsl_ps_compiled_shader  *gl_shaders;
 144     UINT                            num_gl_shaders, shader_array_size;
 145 };
 146
 147 struct glsl_vs_compiled_shader
 148 {
 149     struct vs_compile_args          args;
 150     GLhandleARB                     prgId;
 151 };
 152
 153 struct glsl_vshader_private
 154 {
 155     struct glsl_vs_compiled_shader  *gl_shaders;
 156     UINT                            num_gl_shaders, shader_array_size;
 157 };
 158
 159 /* Extract a line from the info log.
 160  * Note that this modifies the source string. */
 161 static char *get_info_log_line(char **ptr)
 162 {
 163     char *p, *q;
 164
 165     p = *ptr;
 166     if (!(q = strstr(p, "\n")))
 167     {
 168         if (!*p) return NULL;
 169         *ptr += strlen(p);
 170         return p;
 171     }
 172     *q = '\0';
 173     *ptr = q + 1;
 174
 175     return p;
 176 }
 177
 178 /** Prints the GLSL info log which will contain error messages if they exist */
 179 /* GL locking is done by the caller */
 180 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 181 {
 182     int infologLength = 0;
 183     char *infoLog;
 184     unsigned int i;
 185     BOOL is_spam;
 186
 187     static const char * const spam[] =
 188     {
 189         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 190         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx, with \n */
 191         "Fragment shader was successfully compiled to run on hardware.",    /* fglrx, no \n   */
 192         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 193         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 194         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 195         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 196         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 197         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 198     };
 199
 200     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 201
 202     GL_EXTCALL(glGetObjectParameterivARB(obj,
 203                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 204                &infologLength));
 205
 206     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 207      * that if there are errors. */
 208     if (infologLength > 1)
 209     {
 210         char *ptr, *line;
 211
 212         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 213          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 214          */
 215         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 216         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 217         is_spam = FALSE;
 218
 219         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 220             if(strcmp(infoLog, spam[i]) == 0) {
 221                 is_spam = TRUE;
 222                 break;
 223             }
 224         }
 225
 226         ptr = infoLog;
 227         if (is_spam)
 228         {
 229             TRACE("Spam received from GLSL shader #%u:\n", obj);
 230             while ((line = get_info_log_line(&ptr))) TRACE("    %s\n", line);
 231         }
 232         else
 233         {
 234             FIXME("Error received from GLSL shader #%u:\n", obj);
 235             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 236         }
 237         HeapFree(GetProcessHeap(), 0, infoLog);
 238     }
 239 }
 240
 241 /**
 242  * Loads (pixel shader) samplers
 243  */
 244 /* GL locking is done by the caller */
 245 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 246         DWORD *tex_unit_map, GLhandleARB programId)
 247 {
 248     GLint name_loc;
 249     int i;
 250     char sampler_name[20];
 251
 252     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 253         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 254         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 255         if (name_loc != -1) {
 256             DWORD mapped_unit = tex_unit_map[i];
 257             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 258             {
 259                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 260                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 261                 checkGLcall("glUniform1iARB");
 262             } else {
 263                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 264             }
 265         }
 266     }
 267 }
 268
 269 /* GL locking is done by the caller */
 270 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 271         DWORD *tex_unit_map, GLhandleARB programId)
 272 {
 273     GLint name_loc;
 274     char sampler_name[20];
 275     int i;
 276
 277     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 278         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 279         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 280         if (name_loc != -1) {
 281             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 282             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 283             {
 284                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 285                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 286                 checkGLcall("glUniform1iARB");
 287             } else {
 288                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 289             }
 290         }
 291     }
 292 }
 293
 294 /* GL locking is done by the caller */
 295 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 296         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 297 {
 298     int stack_idx = 0;
 299     unsigned int heap_idx = 1;
 300     unsigned int idx;
 301
 302     if (heap->entries[heap_idx].version <= version) return;
 303
 304     idx = heap->entries[heap_idx].idx;
 305     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 306     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 307
 308     while (stack_idx >= 0)
 309     {
 310         /* Note that we fall through to the next case statement. */
 311         switch(stack[stack_idx])
 312         {
 313             case HEAP_NODE_TRAVERSE_LEFT:
 314             {
 315                 unsigned int left_idx = heap_idx << 1;
 316                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 317                 {
 318                     heap_idx = left_idx;
 319                     idx = heap->entries[heap_idx].idx;
 320                     if (constant_locations[idx] != -1)
 321                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 322
 323                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 324                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 325                     break;
 326                 }
 327             }
 328
 329             case HEAP_NODE_TRAVERSE_RIGHT:
 330             {
 331                 unsigned int right_idx = (heap_idx << 1) + 1;
 332                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 333                 {
 334                     heap_idx = right_idx;
 335                     idx = heap->entries[heap_idx].idx;
 336                     if (constant_locations[idx] != -1)
 337                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 338
 339                     stack[stack_idx++] = HEAP_NODE_POP;
 340                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 341                     break;
 342                 }
 343             }
 344
 345             case HEAP_NODE_POP:
 346             {
 347                 heap_idx >>= 1;
 348                 --stack_idx;
 349                 break;
 350             }
 351         }
 352     }
 353     checkGLcall("walk_constant_heap()");
 354 }
 355
 356 /* GL locking is done by the caller */
 357 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 358 {
 359     GLfloat clamped_constant[4];
 360
 361     if (location == -1) return;
 362
 363     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 364     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 365     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 366     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 367
 368     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 369 }
 370
 371 /* GL locking is done by the caller */
 372 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 373         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 374 {
 375     int stack_idx = 0;
 376     unsigned int heap_idx = 1;
 377     unsigned int idx;
 378
 379     if (heap->entries[heap_idx].version <= version) return;
 380
 381     idx = heap->entries[heap_idx].idx;
 382     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 383     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 384
 385     while (stack_idx >= 0)
 386     {
 387         /* Note that we fall through to the next case statement. */
 388         switch(stack[stack_idx])
 389         {
 390             case HEAP_NODE_TRAVERSE_LEFT:
 391             {
 392                 unsigned int left_idx = heap_idx << 1;
 393                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 394                 {
 395                     heap_idx = left_idx;
 396                     idx = heap->entries[heap_idx].idx;
 397                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 398
 399                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 400                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 401                     break;
 402                 }
 403             }
 404
 405             case HEAP_NODE_TRAVERSE_RIGHT:
 406             {
 407                 unsigned int right_idx = (heap_idx << 1) + 1;
 408                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 409                 {
 410                     heap_idx = right_idx;
 411                     idx = heap->entries[heap_idx].idx;
 412                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 413
 414                     stack[stack_idx++] = HEAP_NODE_POP;
 415                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 416                     break;
 417                 }
 418             }
 419
 420             case HEAP_NODE_POP:
 421             {
 422                 heap_idx >>= 1;
 423                 --stack_idx;
 424                 break;
 425             }
 426         }
 427     }
 428     checkGLcall("walk_constant_heap_clamped()");
 429 }
 430
 431 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 432 /* GL locking is done by the caller */
 433 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 434         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 435         unsigned char *stack, UINT version)
 436 {
 437     const local_constant *lconst;
 438
 439     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 440     if (This->baseShader.reg_maps.shader_version.major == 1
 441             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
 442         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 443     else
 444         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 445
 446     if (!This->baseShader.load_local_constsF)
 447     {
 448         TRACE("No need to load local float constants for this shader\n");
 449         return;
 450     }
 451
 452     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 453     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 454     {
 455         GLint location = constant_locations[lconst->idx];
 456         /* We found this uniform name in the program - go ahead and send the data */
 457         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 458     }
 459     checkGLcall("glUniform4fvARB()");
 460 }
 461
 462 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 463 /* GL locking is done by the caller */
 464 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 465         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 466 {
 467     unsigned int i;
 468     struct list* ptr;
 469
 470     for (i = 0; constants_set; constants_set >>= 1, ++i)
 471     {
 472         if (!(constants_set & 1)) continue;
 473
 474         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 475                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 476
 477         /* We found this uniform name in the program - go ahead and send the data */
 478         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 479         checkGLcall("glUniform4ivARB");
 480     }
 481
 482     /* Load immediate constants */
 483     ptr = list_head(&This->baseShader.constantsI);
 484     while (ptr) {
 485         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 486         unsigned int idx = lconst->idx;
 487         const GLint *values = (const GLint *)lconst->value;
 488
 489         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 490             values[0], values[1], values[2], values[3]);
 491
 492         /* We found this uniform name in the program - go ahead and send the data */
 493         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 494         checkGLcall("glUniform4ivARB");
 495         ptr = list_next(&This->baseShader.constantsI, ptr);
 496     }
 497 }
 498
 499 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 500 /* GL locking is done by the caller */
 501 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 502         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 503 {
 504     GLint tmp_loc;
 505     unsigned int i;
 506     char tmp_name[8];
 507     const char *prefix;
 508     struct list* ptr;
 509
 510     switch (This->baseShader.reg_maps.shader_version.type)
 511     {
 512         case WINED3D_SHADER_TYPE_VERTEX:
 513             prefix = "VB";
 514             break;
 515
 516         case WINED3D_SHADER_TYPE_GEOMETRY:
 517             prefix = "GB";
 518             break;
 519
 520         case WINED3D_SHADER_TYPE_PIXEL:
 521             prefix = "PB";
 522             break;
 523
 524         default:
 525             FIXME("Unknown shader type %#x.\n",
 526                     This->baseShader.reg_maps.shader_version.type);
 527             prefix = "UB";
 528             break;
 529     }
 530
 531     /* TODO: Benchmark and see if it would be beneficial to store the
 532      * locations of the constants to avoid looking up each time */
 533     for (i = 0; constants_set; constants_set >>= 1, ++i)
 534     {
 535         if (!(constants_set & 1)) continue;
 536
 537         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 538
 539         /* TODO: Benchmark and see if it would be beneficial to store the
 540          * locations of the constants to avoid looking up each time */
 541         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 542         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 543         if (tmp_loc != -1)
 544         {
 545             /* We found this uniform name in the program - go ahead and send the data */
 546             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 547             checkGLcall("glUniform1ivARB");
 548         }
 549     }
 550
 551     /* Load immediate constants */
 552     ptr = list_head(&This->baseShader.constantsB);
 553     while (ptr) {
 554         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 555         unsigned int idx = lconst->idx;
 556         const GLint *values = (const GLint *)lconst->value;
 557
 558         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 559
 560         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 561         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 562         if (tmp_loc != -1) {
 563             /* We found this uniform name in the program - go ahead and send the data */
 564             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 565             checkGLcall("glUniform1ivARB");
 566         }
 567         ptr = list_next(&This->baseShader.constantsB, ptr);
 568     }
 569 }
 570
 571 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 572 {
 573     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 574 }
 575
 576 /**
 577  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 578  */
 579 /* GL locking is done by the caller (state handler) */
 580 static void shader_glsl_load_np2fixup_constants(
 581     IWineD3DDevice* device,
 582     char usePixelShader,
 583     char useVertexShader) {
 584
 585     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
 586     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
 587
 588     if (!prog) {
 589         /* No GLSL program set - nothing to do. */
 590         return;
 591     }
 592
 593     if (!usePixelShader) {
 594         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 595         return;
 596     }
 597
 598     if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
 599         const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 600         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
 601         UINT i;
 602         UINT fixup = prog->ps_args.np2_fixup;
 603         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 604
 605         for (i = 0; fixup; fixup >>= 1, ++i) {
 606             const unsigned char idx = prog->np2Fixup_info->idx[i];
 607             const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
 608             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 609
 610             if (!tex) {
 611                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 612                 continue;
 613             }
 614
 615             if (idx % 2) {
 616                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 617             } else {
 618                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 619             }
 620         }
 621
 622         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 623     }
 624 }
 625
 626 /**
 627  * Loads the app-supplied constants into the currently set GLSL program.
 628  */
 629 /* GL locking is done by the caller (state handler) */
 630 static void shader_glsl_load_constants(const struct wined3d_context *context,
 631         char usePixelShader, char useVertexShader)
 632 {
 633     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.device;
 634     const struct wined3d_gl_info *gl_info = context->gl_info;
 635     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 636     struct shader_glsl_priv *priv = device->shader_priv;
 637
 638     GLhandleARB programId;
 639     struct glsl_shader_prog_link *prog = priv->glsl_program;
 640     UINT constant_version;
 641     int i;
 642
 643     if (!prog) {
 644         /* No GLSL program set - nothing to do. */
 645         return;
 646     }
 647     programId = prog->programId;
 648     constant_version = prog->constant_version;
 649
 650     if (useVertexShader) {
 651         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 652
 653         /* Load DirectX 9 float constants/uniforms for vertex shader */
 654         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
 655                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 656
 657         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 658         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
 659                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 660
 661         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 662         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
 663                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 664
 665         /* Upload the position fixup params */
 666         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
 667         checkGLcall("glUniform4fvARB");
 668     }
 669
 670     if (usePixelShader) {
 671
 672         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 673
 674         /* Load DirectX 9 float constants/uniforms for pixel shader */
 675         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
 676                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 677
 678         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 679         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
 680                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 681
 682         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 683         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
 684                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 685
 686         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 687          * It can't be 0 for a valid texbem instruction.
 688          */
 689         for(i = 0; i < MAX_TEXTURES; i++) {
 690             const float *data;
 691
 692             if(prog->bumpenvmat_location[i] == -1) continue;
 693
 694             data = (const float *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVMAT00];
 695             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 696             checkGLcall("glUniformMatrix2fvARB");
 697
 698             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 699              * is set too, so we can check that in the needsbumpmat check
 700              */
 701             if(prog->luminancescale_location[i] != -1) {
 702                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLSCALE];
 703                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLOFFSET];
 704
 705                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 706                 checkGLcall("glUniform1fvARB");
 707                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 708                 checkGLcall("glUniform1fvARB");
 709             }
 710         }
 711
 712         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 713             float correction_params[4];
 714
 715             if (context->render_offscreen)
 716             {
 717                 correction_params[0] = 0.0f;
 718                 correction_params[1] = 1.0f;
 719             } else {
 720                 /* position is window relative, not viewport relative */
 721                 correction_params[0] = ((IWineD3DSurfaceImpl *)context->current_rt)->currentDesc.Height;
 722                 correction_params[1] = -1.0f;
 723             }
 724             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 725         }
 726     }
 727
 728     if (priv->next_constant_version == UINT_MAX)
 729     {
 730         TRACE("Max constant version reached, resetting to 0.\n");
 731         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 732         priv->next_constant_version = 1;
 733     }
 734     else
 735     {
 736         prog->constant_version = priv->next_constant_version++;
 737     }
 738 }
 739
 740 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 741         unsigned int heap_idx, DWORD new_version)
 742 {
 743     struct constant_entry *entries = heap->entries;
 744     unsigned int *positions = heap->positions;
 745     unsigned int parent_idx;
 746
 747     while (heap_idx > 1)
 748     {
 749         parent_idx = heap_idx >> 1;
 750
 751         if (new_version <= entries[parent_idx].version) break;
 752
 753         entries[heap_idx] = entries[parent_idx];
 754         positions[entries[parent_idx].idx] = heap_idx;
 755         heap_idx = parent_idx;
 756     }
 757
 758     entries[heap_idx].version = new_version;
 759     entries[heap_idx].idx = idx;
 760     positions[idx] = heap_idx;
 761 }
 762
 763 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 764 {
 765     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 766     struct shader_glsl_priv *priv = This->shader_priv;
 767     struct constant_heap *heap = &priv->vconst_heap;
 768     UINT i;
 769
 770     for (i = start; i < count + start; ++i)
 771     {
 772         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
 773             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 774         else
 775             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 776     }
 777 }
 778
 779 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 780 {
 781     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 782     struct shader_glsl_priv *priv = This->shader_priv;
 783     struct constant_heap *heap = &priv->pconst_heap;
 784     UINT i;
 785
 786     for (i = start; i < count + start; ++i)
 787     {
 788         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 789             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 790         else
 791             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 792     }
 793 }
 794
 795 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 796 {
 797     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 798     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 799     if(shader_major > 3) return ret;
 800
 801     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 802     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 803     return ret;
 804 }
 805
 806 /** Generate the variable & register declarations for the GLSL output target */
 807 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 808         struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
 809         const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
 810 {
 811     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 812     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 813     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 814     const struct wined3d_gl_info *gl_info = context->gl_info;
 815     unsigned int i, extra_constants_needed = 0;
 816     const local_constant *lconst;
 817     DWORD map;
 818
 819     /* There are some minor differences between pixel and vertex shaders */
 820     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 821     char prefix = pshader ? 'P' : 'V';
 822
 823     /* Prototype the subroutines */
 824     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 825     {
 826         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 827     }
 828
 829     /* Declare the constants (aka uniforms) */
 830     if (This->baseShader.limits.constant_float > 0) {
 831         unsigned max_constantsF;
 832         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 833          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 834          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 835          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 836          * a dx9 card, as long as it doesn't also use all the other constants.
 837          *
 838          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 839          * declare only the amount that we're assured to have.
 840          *
 841          * Thus we run into problems in these two cases:
 842          * 1) The shader really uses more uniforms than supported
 843          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 844          */
 845         if (pshader)
 846         {
 847             /* No indirect addressing here. */
 848             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 849         }
 850         else
 851         {
 852             if(This->baseShader.reg_maps.usesrelconstF) {
 853                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
 854                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
 855                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
 856                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
 857                  *
 858                  * Writing gl_ClipVertex requires one uniform for each clipplane as well.
 859                  */
 860                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 861                 if(ctx_priv->cur_vs_args->clip_enabled)
 862                 {
 863                     max_constantsF -= gl_info->limits.clipplanes;
 864                 }
 865                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 866                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 867                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 868                  * for now take this into account when calculating the number of available constants
 869                  */
 870                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
 871                 /* Set by driver quirks in directx.c */
 872                 max_constantsF -= gl_info->reserved_glsl_constants;
 873             }
 874             else
 875             {
 876                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 877             }
 878         }
 879         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
 880         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 881     }
 882
 883     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
 884      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
 885      */
 886     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
 887         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 888
 889     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
 890         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 891
 892     if(!pshader) {
 893         shader_addline(buffer, "uniform vec4 posFixup;\n");
 894         /* Predeclaration; This function is added at link time based on the pixel shader.
 895          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 896          * that. We know the input to the reorder function at vertex shader compile time, so
 897          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 898          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 899          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 900          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 901          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 902          * inout.
 903          */
 904         if (reg_maps->shader_version.major >= 3)
 905         {
 906             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 907         } else {
 908             shader_addline(buffer, "void order_ps_input();\n");
 909         }
 910     } else {
 911         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
 912         {
 913             if (!(map & 1)) continue;
 914
 915             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 916
 917             if (reg_maps->luminanceparams & (1 << i))
 918             {
 919                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
 920                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
 921                 extra_constants_needed++;
 922             }
 923
 924             extra_constants_needed++;
 925         }
 926
 927         if (ps_args->srgb_correction)
 928         {
 929             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
 930                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
 931             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
 932                     srgb_cmp);
 933         }
 934         if (reg_maps->vpos || reg_maps->usesdsy)
 935         {
 936             if (This->baseShader.limits.constant_float + extra_constants_needed
 937                     + 1 < gl_info->limits.glsl_ps_float_constants)
 938             {
 939                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
 940                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
 941                 extra_constants_needed++;
 942             } else {
 943                 /* This happens because we do not have proper tracking of the constant registers that are
 944                  * actually used, only the max limit of the shader version
 945                  */
 946                 FIXME("Cannot find a free uniform for vpos correction params\n");
 947                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
 948                         context->render_offscreen ? 0.0f : ((IWineD3DSurfaceImpl *)device->render_targets[0])->currentDesc.Height,
 949                         context->render_offscreen ? 1.0f : -1.0f);
 950             }
 951             shader_addline(buffer, "vec4 vpos;\n");
 952         }
 953     }
 954
 955     /* Declare texture samplers */
 956     for (i = 0; i < This->baseShader.limits.sampler; i++) {
 957         if (reg_maps->sampler_type[i])
 958         {
 959             switch (reg_maps->sampler_type[i])
 960             {
 961                 case WINED3DSTT_1D:
 962                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
 963                     break;
 964                 case WINED3DSTT_2D:
 965                     if(device->stateBlock->textures[i] &&
 966                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
 967                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
 968                     } else {
 969                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
 970                     }
 971                     break;
 972                 case WINED3DSTT_CUBE:
 973                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
 974                     break;
 975                 case WINED3DSTT_VOLUME:
 976                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
 977                     break;
 978                 default:
 979                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
 980                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
 981                     break;
 982             }
 983         }
 984     }
 985
 986     /* Declare uniforms for NP2 texcoord fixup:
 987      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
 988      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
 989      * Modern cards just skip the code anyway, so put it inside a separate loop. */
 990     if (pshader && ps_args->np2_fixup) {
 991
 992         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
 993         UINT cur = 0;
 994
 995         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
 996          * while D3D has them in the (normalized) [0,1]x[0,1] range.
 997          * samplerNP2Fixup stores texture dimensions and is updated through
 998          * shader_glsl_load_np2fixup_constants when the sampler changes. */
 999
1000         for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1001             if (reg_maps->sampler_type[i]) {
1002                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1003
1004                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1005                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1006                     continue;
1007                 }
1008
1009                 fixup->idx[i] = cur++;
1010             }
1011         }
1012
1013         fixup->num_consts = (cur + 1) >> 1;
1014         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1015     }
1016
1017     /* Declare address variables */
1018     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1019     {
1020         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1021     }
1022
1023     /* Declare texture coordinate temporaries and initialize them */
1024     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1025     {
1026         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1027     }
1028
1029     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1030      * helper function shader that is linked in at link time
1031      */
1032     if (pshader && reg_maps->shader_version.major >= 3)
1033     {
1034         if (use_vs(device->stateBlock))
1035         {
1036             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1037         } else {
1038             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1039              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1040              * pixel shader that reads the fixed function color into the packed input registers.
1041              */
1042             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1043         }
1044     }
1045
1046     /* Declare output register temporaries */
1047     if(This->baseShader.limits.packed_output) {
1048         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1049     }
1050
1051     /* Declare temporary variables */
1052     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1053     {
1054         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1055     }
1056
1057     /* Declare attributes */
1058     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1059     {
1060         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1061         {
1062             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1063         }
1064     }
1065
1066     /* Declare loop registers aLx */
1067     for (i = 0; i < reg_maps->loop_depth; i++) {
1068         shader_addline(buffer, "int aL%u;\n", i);
1069         shader_addline(buffer, "int tmpInt%u;\n", i);
1070     }
1071
1072     /* Temporary variables for matrix operations */
1073     shader_addline(buffer, "vec4 tmp0;\n");
1074     shader_addline(buffer, "vec4 tmp1;\n");
1075
1076     /* Local constants use a different name so they can be loaded once at shader link time
1077      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1078      * float -> string conversion can cause precision loss.
1079      */
1080     if(!This->baseShader.load_local_constsF) {
1081         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1082             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1083         }
1084     }
1085
1086     /* Start the main program */
1087     shader_addline(buffer, "void main() {\n");
1088     if(pshader && reg_maps->vpos) {
1089         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1090          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1091          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1092          * precision troubles when we just substract 0.5.
1093          *
1094          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1095          *
1096          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1097          *
1098          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1099          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1100          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1101          * correctly on drivers that returns integer values.
1102          */
1103         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1104     }
1105 }
1106
1107 /*****************************************************************************
1108  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1109  *
1110  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1111  ****************************************************************************/
1112
1113 /* Prototypes */
1114 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1115         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1116
1117 /** Used for opcode modifiers - They multiply the result by the specified amount */
1118 static const char * const shift_glsl_tab[] = {
1119     "",           /*  0 (none) */
1120     "2.0 * ",     /*  1 (x2)   */
1121     "4.0 * ",     /*  2 (x4)   */
1122     "8.0 * ",     /*  3 (x8)   */
1123     "16.0 * ",    /*  4 (x16)  */
1124     "32.0 * ",    /*  5 (x32)  */
1125     "",           /*  6 (x64)  */
1126     "",           /*  7 (x128) */
1127     "",           /*  8 (d256) */
1128     "",           /*  9 (d128) */
1129     "",           /* 10 (d64)  */
1130     "",           /* 11 (d32)  */
1131     "0.0625 * ",  /* 12 (d16)  */
1132     "0.125 * ",   /* 13 (d8)   */
1133     "0.25 * ",    /* 14 (d4)   */
1134     "0.5 * "      /* 15 (d2)   */
1135 };
1136
1137 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1138 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1139 {
1140     out_str[0] = 0;
1141
1142     switch (src_modifier)
1143     {
1144     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1145     case WINED3DSPSM_DW:
1146     case WINED3DSPSM_NONE:
1147         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1148         break;
1149     case WINED3DSPSM_NEG:
1150         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1151         break;
1152     case WINED3DSPSM_NOT:
1153         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1154         break;
1155     case WINED3DSPSM_BIAS:
1156         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1157         break;
1158     case WINED3DSPSM_BIASNEG:
1159         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1160         break;
1161     case WINED3DSPSM_SIGN:
1162         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1163         break;
1164     case WINED3DSPSM_SIGNNEG:
1165         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1166         break;
1167     case WINED3DSPSM_COMP:
1168         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1169         break;
1170     case WINED3DSPSM_X2:
1171         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1172         break;
1173     case WINED3DSPSM_X2NEG:
1174         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1175         break;
1176     case WINED3DSPSM_ABS:
1177         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1178         break;
1179     case WINED3DSPSM_ABSNEG:
1180         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1181         break;
1182     default:
1183         FIXME("Unhandled modifier %u\n", src_modifier);
1184         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1185     }
1186 }
1187
1188 /** Writes the GLSL variable name that corresponds to the register that the
1189  * DX opcode parameter is trying to access */
1190 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1191         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1192 {
1193     /* oPos, oFog and oPts in D3D */
1194     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
1195
1196     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1197     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1198     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1199
1200     *is_color = FALSE;
1201
1202     switch (reg->type)
1203     {
1204         case WINED3DSPR_TEMP:
1205             sprintf(register_name, "R%u", reg->idx);
1206             break;
1207
1208         case WINED3DSPR_INPUT:
1209             /* vertex shaders */
1210             if (!pshader)
1211             {
1212                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1213                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1214                 sprintf(register_name, "attrib%u", reg->idx);
1215                 break;
1216             }
1217
1218             /* pixel shaders >= 3.0 */
1219             if (This->baseShader.reg_maps.shader_version.major >= 3)
1220             {
1221                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1222                 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1223
1224                 if (reg->rel_addr)
1225                 {
1226                     glsl_src_param_t rel_param;
1227
1228                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1229
1230                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1231                      * operation there */
1232                     if (idx)
1233                     {
1234                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1235                         {
1236                             sprintf(register_name,
1237                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1238                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1239                                     rel_param.param_str, idx);
1240                         }
1241                         else
1242                         {
1243                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1244                         }
1245                     }
1246                     else
1247                     {
1248                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1249                         {
1250                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1251                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1252                                     rel_param.param_str);
1253                         }
1254                         else
1255                         {
1256                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1257                         }
1258                     }
1259                 }
1260                 else
1261                 {
1262                     if (idx == in_count) sprintf(register_name, "gl_Color");
1263                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1264                     else sprintf(register_name, "IN[%u]", idx);
1265                 }
1266             }
1267             else
1268             {
1269                 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1270                 else strcpy(register_name, "gl_SecondaryColor");
1271                 break;
1272             }
1273             break;
1274
1275         case WINED3DSPR_CONST:
1276             {
1277                 const char prefix = pshader ? 'P' : 'V';
1278
1279                 /* Relative addressing */
1280                 if (reg->rel_addr)
1281                 {
1282                     glsl_src_param_t rel_param;
1283                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1284                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1285                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1286                 }
1287                 else
1288                 {
1289                     if (shader_constant_is_local(This, reg->idx))
1290                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1291                     else
1292                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1293                 }
1294             }
1295             break;
1296
1297         case WINED3DSPR_CONSTINT:
1298             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1299             else sprintf(register_name, "VI[%u]", reg->idx);
1300             break;
1301
1302         case WINED3DSPR_CONSTBOOL:
1303             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1304             else sprintf(register_name, "VB[%u]", reg->idx);
1305             break;
1306
1307         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1308             if (pshader) sprintf(register_name, "T%u", reg->idx);
1309             else sprintf(register_name, "A%u", reg->idx);
1310             break;
1311
1312         case WINED3DSPR_LOOP:
1313             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1314             break;
1315
1316         case WINED3DSPR_SAMPLER:
1317             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1318             else sprintf(register_name, "Vsampler%u", reg->idx);
1319             break;
1320
1321         case WINED3DSPR_COLOROUT:
1322             if (reg->idx >= gl_info->limits.buffers)
1323                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1324
1325             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1326             break;
1327
1328         case WINED3DSPR_RASTOUT:
1329             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1330             break;
1331
1332         case WINED3DSPR_DEPTHOUT:
1333             sprintf(register_name, "gl_FragDepth");
1334             break;
1335
1336         case WINED3DSPR_ATTROUT:
1337             if (reg->idx == 0) sprintf(register_name, "gl_FrontColor");
1338             else sprintf(register_name, "gl_FrontSecondaryColor");
1339             break;
1340
1341         case WINED3DSPR_TEXCRDOUT:
1342             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1343             if (This->baseShader.reg_maps.shader_version.major >= 3) sprintf(register_name, "OUT[%u]", reg->idx);
1344             else sprintf(register_name, "gl_TexCoord[%u]", reg->idx);
1345             break;
1346
1347         case WINED3DSPR_MISCTYPE:
1348             if (reg->idx == 0)
1349             {
1350                 /* vPos */
1351                 sprintf(register_name, "vpos");
1352             }
1353             else if (reg->idx == 1)
1354             {
1355                 /* Note that gl_FrontFacing is a bool, while vFace is
1356                  * a float for which the sign determines front/back */
1357                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1358             }
1359             else
1360             {
1361                 FIXME("Unhandled misctype register %d\n", reg->idx);
1362                 sprintf(register_name, "unrecognized_register");
1363             }
1364             break;
1365
1366         case WINED3DSPR_IMMCONST:
1367             switch (reg->immconst_type)
1368             {
1369                 case WINED3D_IMMCONST_FLOAT:
1370                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1371                     break;
1372
1373                 case WINED3D_IMMCONST_FLOAT4:
1374                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1375                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1376                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1377                     break;
1378
1379                 default:
1380                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1381                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1382             }
1383             break;
1384
1385         default:
1386             FIXME("Unhandled register name Type(%d)\n", reg->type);
1387             sprintf(register_name, "unrecognized_register");
1388             break;
1389     }
1390 }
1391
1392 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1393 {
1394     *str++ = '.';
1395     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1396     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1397     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1398     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1399     *str = '\0';
1400 }
1401
1402 /* Get the GLSL write mask for the destination register */
1403 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1404 {
1405     DWORD mask = param->write_mask;
1406
1407     if (shader_is_scalar(&param->reg))
1408     {
1409         mask = WINED3DSP_WRITEMASK_0;
1410         *write_mask = '\0';
1411     }
1412     else
1413     {
1414         shader_glsl_write_mask_to_str(mask, write_mask);
1415     }
1416
1417     return mask;
1418 }
1419
1420 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1421     unsigned int size = 0;
1422
1423     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1424     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1425     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1426     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1427
1428     return size;
1429 }
1430
1431 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1432 {
1433     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1434      * but addressed as "rgba". To fix this we need to swap the register's x
1435      * and z components. */
1436     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1437
1438     *str++ = '.';
1439     /* swizzle bits fields: wwzzyyxx */
1440     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1441     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1442     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1443     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1444     *str = '\0';
1445 }
1446
1447 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1448         BOOL fixup, DWORD mask, char *swizzle_str)
1449 {
1450     if (shader_is_scalar(&param->reg))
1451         *swizzle_str = '\0';
1452     else
1453         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1454 }
1455
1456 /* From a given parameter token, generate the corresponding GLSL string.
1457  * Also, return the actual register name and swizzle in case the
1458  * caller needs this information as well. */
1459 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1460         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1461 {
1462     BOOL is_color = FALSE;
1463     char swizzle_str[6];
1464
1465     glsl_src->reg_name[0] = '\0';
1466     glsl_src->param_str[0] = '\0';
1467     swizzle_str[0] = '\0';
1468
1469     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1470     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1471     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1472 }
1473
1474 /* From a given parameter token, generate the corresponding GLSL string.
1475  * Also, return the actual register name and swizzle in case the
1476  * caller needs this information as well. */
1477 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1478         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1479 {
1480     BOOL is_color = FALSE;
1481
1482     glsl_dst->mask_str[0] = '\0';
1483     glsl_dst->reg_name[0] = '\0';
1484
1485     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1486     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1487 }
1488
1489 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1490 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1491         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1492 {
1493     glsl_dst_param_t glsl_dst;
1494     DWORD mask;
1495
1496     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1497     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1498
1499     return mask;
1500 }
1501
1502 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1503 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1504 {
1505     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1506 }
1507
1508 /** Process GLSL instruction modifiers */
1509 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1510 {
1511     glsl_dst_param_t dst_param;
1512     DWORD modifiers;
1513
1514     if (!ins->dst_count) return;
1515
1516     modifiers = ins->dst[0].modifiers;
1517     if (!modifiers) return;
1518
1519     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1520
1521     if (modifiers & WINED3DSPDM_SATURATE)
1522     {
1523         /* _SAT means to clamp the value of the register to between 0 and 1 */
1524         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1525                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1526     }
1527
1528     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1529     {
1530         FIXME("_centroid modifier not handled\n");
1531     }
1532
1533     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1534     {
1535         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1536     }
1537 }
1538
1539 static inline const char *shader_get_comp_op(DWORD op)
1540 {
1541     switch (op) {
1542         case COMPARISON_GT: return ">";
1543         case COMPARISON_EQ: return "==";
1544         case COMPARISON_GE: return ">=";
1545         case COMPARISON_LT: return "<";
1546         case COMPARISON_NE: return "!=";
1547         case COMPARISON_LE: return "<=";
1548         default:
1549             FIXME("Unrecognized comparison value: %u\n", op);
1550             return "(\?\?)";
1551     }
1552 }
1553
1554 static void shader_glsl_get_sample_function(DWORD sampler_type, DWORD flags, glsl_sample_function_t *sample_function)
1555 {
1556     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1557     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1558     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1559     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1560
1561     /* Note that there's no such thing as a projected cube texture. */
1562     switch(sampler_type) {
1563         case WINED3DSTT_1D:
1564             if(lod) {
1565                 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1566             } else  if(grad) {
1567                 sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1568             } else {
1569                 sample_function->name = projected ? "texture1DProj" : "texture1D";
1570             }
1571             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1572             break;
1573         case WINED3DSTT_2D:
1574             if(texrect) {
1575                 if(lod) {
1576                     sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1577                 } else  if(grad) {
1578                     sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1579                 } else {
1580                     sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1581                 }
1582             } else {
1583                 if(lod) {
1584                     sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1585                 } else  if(grad) {
1586                     sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1587                 } else {
1588                     sample_function->name = projected ? "texture2DProj" : "texture2D";
1589                 }
1590             }
1591             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1592             break;
1593         case WINED3DSTT_CUBE:
1594             if(lod) {
1595                 sample_function->name = "textureCubeLod";
1596             } else if(grad) {
1597                 sample_function->name = "textureCubeGradARB";
1598             } else {
1599                 sample_function->name = "textureCube";
1600             }
1601             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1602             break;
1603         case WINED3DSTT_VOLUME:
1604             if(lod) {
1605                 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1606             } else  if(grad) {
1607                 sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1608             } else {
1609                 sample_function->name = projected ? "texture3DProj" : "texture3D";
1610             }
1611             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1612             break;
1613         default:
1614             sample_function->name = "";
1615             sample_function->coord_mask = 0;
1616             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1617             break;
1618     }
1619 }
1620
1621 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1622         BOOL sign_fixup, enum fixup_channel_source channel_source)
1623 {
1624     switch(channel_source)
1625     {
1626         case CHANNEL_SOURCE_ZERO:
1627             strcat(arguments, "0.0");
1628             break;
1629
1630         case CHANNEL_SOURCE_ONE:
1631             strcat(arguments, "1.0");
1632             break;
1633
1634         case CHANNEL_SOURCE_X:
1635             strcat(arguments, reg_name);
1636             strcat(arguments, ".x");
1637             break;
1638
1639         case CHANNEL_SOURCE_Y:
1640             strcat(arguments, reg_name);
1641             strcat(arguments, ".y");
1642             break;
1643
1644         case CHANNEL_SOURCE_Z:
1645             strcat(arguments, reg_name);
1646             strcat(arguments, ".z");
1647             break;
1648
1649         case CHANNEL_SOURCE_W:
1650             strcat(arguments, reg_name);
1651             strcat(arguments, ".w");
1652             break;
1653
1654         default:
1655             FIXME("Unhandled channel source %#x\n", channel_source);
1656             strcat(arguments, "undefined");
1657             break;
1658     }
1659
1660     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1661 }
1662
1663 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1664 {
1665     struct wined3d_shader_dst_param dst;
1666     unsigned int mask_size, remaining;
1667     glsl_dst_param_t dst_param;
1668     char arguments[256];
1669     DWORD mask;
1670
1671     mask = 0;
1672     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1673     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1674     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1675     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1676     mask &= ins->dst[0].write_mask;
1677
1678     if (!mask) return; /* Nothing to do */
1679
1680     if (is_yuv_fixup(fixup))
1681     {
1682         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1683         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1684         return;
1685     }
1686
1687     mask_size = shader_glsl_get_write_mask_size(mask);
1688
1689     dst = ins->dst[0];
1690     dst.write_mask = mask;
1691     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1692
1693     arguments[0] = '\0';
1694     remaining = mask_size;
1695     if (mask & WINED3DSP_WRITEMASK_0)
1696     {
1697         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1698         if (--remaining) strcat(arguments, ", ");
1699     }
1700     if (mask & WINED3DSP_WRITEMASK_1)
1701     {
1702         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1703         if (--remaining) strcat(arguments, ", ");
1704     }
1705     if (mask & WINED3DSP_WRITEMASK_2)
1706     {
1707         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1708         if (--remaining) strcat(arguments, ", ");
1709     }
1710     if (mask & WINED3DSP_WRITEMASK_3)
1711     {
1712         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1713         if (--remaining) strcat(arguments, ", ");
1714     }
1715
1716     if (mask_size > 1)
1717     {
1718         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1719                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1720     }
1721     else
1722     {
1723         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1724     }
1725 }
1726
1727 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
1728         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
1729         const char *dx, const char *dy,
1730         const char *bias, const char *coord_reg_fmt, ...)
1731 {
1732     const char *sampler_base;
1733     char dst_swizzle[6];
1734     struct color_fixup_desc fixup;
1735     BOOL np2_fixup = FALSE;
1736     va_list args;
1737
1738     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
1739
1740     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1741     {
1742         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1743         fixup = priv->cur_ps_args->color_fixup[sampler];
1744         sampler_base = "Psampler";
1745
1746         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
1747             if(bias) {
1748                 FIXME("Biased sampling from NP2 textures is unsupported\n");
1749             } else {
1750                 np2_fixup = TRUE;
1751             }
1752         }
1753     } else {
1754         sampler_base = "Vsampler";
1755         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
1756     }
1757
1758     shader_glsl_append_dst(ins->ctx->buffer, ins);
1759
1760     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
1761
1762     va_start(args, coord_reg_fmt);
1763     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
1764     va_end(args);
1765
1766     if(bias) {
1767         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
1768     } else {
1769         if (np2_fixup) {
1770             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1771             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
1772
1773             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
1774                            (idx % 2) ? "zw" : "xy", dst_swizzle);
1775         } else if(dx && dy) {
1776             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
1777         } else {
1778             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
1779         }
1780     }
1781
1782     if(!is_identity_fixup(fixup)) {
1783         shader_glsl_color_correction(ins, fixup);
1784     }
1785 }
1786
1787 /*****************************************************************************
1788  * Begin processing individual instruction opcodes
1789  ****************************************************************************/
1790
1791 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1792 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
1793 {
1794     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1795     glsl_src_param_t src0_param;
1796     glsl_src_param_t src1_param;
1797     DWORD write_mask;
1798     char op;
1799
1800     /* Determine the GLSL operator to use based on the opcode */
1801     switch (ins->handler_idx)
1802     {
1803         case WINED3DSIH_MUL: op = '*'; break;
1804         case WINED3DSIH_ADD: op = '+'; break;
1805         case WINED3DSIH_SUB: op = '-'; break;
1806         default:
1807             op = ' ';
1808             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1809             break;
1810     }
1811
1812     write_mask = shader_glsl_append_dst(buffer, ins);
1813     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1814     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
1815     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1816 }
1817
1818 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1819 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
1820 {
1821     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1822     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1823     glsl_src_param_t src0_param;
1824     DWORD write_mask;
1825
1826     write_mask = shader_glsl_append_dst(buffer, ins);
1827     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1828
1829     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1830      * shader versions WINED3DSIO_MOVA is used for this. */
1831     if (ins->ctx->reg_maps->shader_version.major == 1
1832             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1833             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1834     {
1835         /* This is a simple floor() */
1836         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1837         if (mask_size > 1) {
1838             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1839         } else {
1840             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1841         }
1842     }
1843     else if(ins->handler_idx == WINED3DSIH_MOVA)
1844     {
1845         /* We need to *round* to the nearest int here. */
1846         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1847
1848         if (gl_info->supported[EXT_GPU_SHADER4])
1849         {
1850             if (mask_size > 1)
1851                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
1852             else
1853                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
1854         }
1855         else
1856         {
1857             if (mask_size > 1)
1858                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
1859                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
1860             else
1861                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
1862                         src0_param.param_str, src0_param.param_str);
1863         }
1864     }
1865     else
1866     {
1867         shader_addline(buffer, "%s);\n", src0_param.param_str);
1868     }
1869 }
1870
1871 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1872 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
1873 {
1874     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1875     glsl_src_param_t src0_param;
1876     glsl_src_param_t src1_param;
1877     DWORD dst_write_mask, src_write_mask;
1878     unsigned int dst_size = 0;
1879
1880     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1881     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1882
1883     /* dp3 works on vec3, dp4 on vec4 */
1884     if (ins->handler_idx == WINED3DSIH_DP4)
1885     {
1886         src_write_mask = WINED3DSP_WRITEMASK_ALL;
1887     } else {
1888         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1889     }
1890
1891     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
1892     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
1893
1894     if (dst_size > 1) {
1895         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1896     } else {
1897         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1898     }
1899 }
1900
1901 /* Note that this instruction has some restrictions. The destination write mask
1902  * can't contain the w component, and the source swizzles have to be .xyzw */
1903 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
1904 {
1905     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1906     glsl_src_param_t src0_param;
1907     glsl_src_param_t src1_param;
1908     char dst_mask[6];
1909
1910     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
1911     shader_glsl_append_dst(ins->ctx->buffer, ins);
1912     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
1913     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
1914     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1915 }
1916
1917 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1918  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1919  * GLSL uses the value as-is. */
1920 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
1921 {
1922     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1923     glsl_src_param_t src0_param;
1924     glsl_src_param_t src1_param;
1925     DWORD dst_write_mask;
1926     unsigned int dst_size;
1927
1928     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1929     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1930
1931     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
1932     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
1933
1934     if (dst_size > 1) {
1935         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1936     } else {
1937         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1938     }
1939 }
1940
1941 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1942  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1943  * GLSL uses the value as-is. */
1944 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
1945 {
1946     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1947     glsl_src_param_t src0_param;
1948     DWORD dst_write_mask;
1949     unsigned int dst_size;
1950
1951     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1952     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1953
1954     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
1955
1956     if (dst_size > 1) {
1957         shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1958     } else {
1959         shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1960     }
1961 }
1962
1963 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1964 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
1965 {
1966     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1967     glsl_src_param_t src_param;
1968     const char *instruction;
1969     DWORD write_mask;
1970     unsigned i;
1971
1972     /* Determine the GLSL function to use based on the opcode */
1973     /* TODO: Possibly make this a table for faster lookups */
1974     switch (ins->handler_idx)
1975     {
1976         case WINED3DSIH_MIN: instruction = "min"; break;
1977         case WINED3DSIH_MAX: instruction = "max"; break;
1978         case WINED3DSIH_ABS: instruction = "abs"; break;
1979         case WINED3DSIH_FRC: instruction = "fract"; break;
1980         case WINED3DSIH_NRM: instruction = "normalize"; break;
1981         case WINED3DSIH_EXP: instruction = "exp2"; break;
1982         case WINED3DSIH_DSX: instruction = "dFdx"; break;
1983         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
1984         default: instruction = "";
1985             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1986             break;
1987     }
1988
1989     write_mask = shader_glsl_append_dst(buffer, ins);
1990
1991     shader_addline(buffer, "%s(", instruction);
1992
1993     if (ins->src_count)
1994     {
1995         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
1996         shader_addline(buffer, "%s", src_param.param_str);
1997         for (i = 1; i < ins->src_count; ++i)
1998         {
1999             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2000             shader_addline(buffer, ", %s", src_param.param_str);
2001         }
2002     }
2003
2004     shader_addline(buffer, "));\n");
2005 }
2006
2007 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2008  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2009  *   dst.x = 2^(floor(src))
2010  *   dst.y = src - floor(src)
2011  *   dst.z = 2^src   (partial precision is allowed, but optional)
2012  *   dst.w = 1.0;
2013  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2014  *   dst = 2^src;    (partial precision is allowed, but optional)
2015  */
2016 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2017 {
2018     glsl_src_param_t src_param;
2019
2020     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2021
2022     if (ins->ctx->reg_maps->shader_version.major < 2)
2023     {
2024         char dst_mask[6];
2025
2026         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2027         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2028         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2029         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2030
2031         shader_glsl_append_dst(ins->ctx->buffer, ins);
2032         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2033         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2034     } else {
2035         DWORD write_mask;
2036         unsigned int mask_size;
2037
2038         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2039         mask_size = shader_glsl_get_write_mask_size(write_mask);
2040
2041         if (mask_size > 1) {
2042             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2043         } else {
2044             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2045         }
2046     }
2047 }
2048
2049 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2050 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2051 {
2052     glsl_src_param_t src_param;
2053     DWORD write_mask;
2054     unsigned int mask_size;
2055
2056     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2057     mask_size = shader_glsl_get_write_mask_size(write_mask);
2058     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2059
2060     if (mask_size > 1) {
2061         shader_addline(ins->ctx->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
2062     } else {
2063         shader_addline(ins->ctx->buffer, "1.0 / %s);\n", src_param.param_str);
2064     }
2065 }
2066
2067 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2068 {
2069     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2070     glsl_src_param_t src_param;
2071     DWORD write_mask;
2072     unsigned int mask_size;
2073
2074     write_mask = shader_glsl_append_dst(buffer, ins);
2075     mask_size = shader_glsl_get_write_mask_size(write_mask);
2076
2077     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2078
2079     if (mask_size > 1) {
2080         shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
2081     } else {
2082         shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
2083     }
2084 }
2085
2086 /** Process signed comparison opcodes in GLSL. */
2087 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2088 {
2089     glsl_src_param_t src0_param;
2090     glsl_src_param_t src1_param;
2091     DWORD write_mask;
2092     unsigned int mask_size;
2093
2094     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2095     mask_size = shader_glsl_get_write_mask_size(write_mask);
2096     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2097     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2098
2099     if (mask_size > 1) {
2100         const char *compare;
2101
2102         switch(ins->handler_idx)
2103         {
2104             case WINED3DSIH_SLT: compare = "lessThan"; break;
2105             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2106             default: compare = "";
2107                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2108         }
2109
2110         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2111                 src0_param.param_str, src1_param.param_str);
2112     } else {
2113         switch(ins->handler_idx)
2114         {
2115             case WINED3DSIH_SLT:
2116                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2117                  * to return 0.0 but step returns 1.0 because step is not < x
2118                  * An alternative is a bvec compare padded with an unused second component.
2119                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2120                  * issue. Playing with not() is not possible either because not() does not accept
2121                  * a scalar.
2122                  */
2123                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2124                         src0_param.param_str, src1_param.param_str);
2125                 break;
2126             case WINED3DSIH_SGE:
2127                 /* Here we can use the step() function and safe a conditional */
2128                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2129                 break;
2130             default:
2131                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2132         }
2133
2134     }
2135 }
2136
2137 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2138 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2139 {
2140     glsl_src_param_t src0_param;
2141     glsl_src_param_t src1_param;
2142     glsl_src_param_t src2_param;
2143     DWORD write_mask, cmp_channel = 0;
2144     unsigned int i, j;
2145     char mask_char[6];
2146     BOOL temp_destination = FALSE;
2147
2148     if (shader_is_scalar(&ins->src[0].reg))
2149     {
2150         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2151
2152         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2153         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2154         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2155
2156         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2157                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2158     } else {
2159         DWORD dst_mask = ins->dst[0].write_mask;
2160         struct wined3d_shader_dst_param dst = ins->dst[0];
2161
2162         /* Cycle through all source0 channels */
2163         for (i=0; i<4; i++) {
2164             write_mask = 0;
2165             /* Find the destination channels which use the current source0 channel */
2166             for (j=0; j<4; j++) {
2167                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2168                 {
2169                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2170                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2171                 }
2172             }
2173             dst.write_mask = dst_mask & write_mask;
2174
2175             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2176             * The first lines may overwrite source parameters of the following lines.
2177             * Deal with that by using a temporary destination register if needed
2178             */
2179             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2180                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2181                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2182                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2183                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2184                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2185             {
2186                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2187                 if (!write_mask) continue;
2188                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2189                 temp_destination = TRUE;
2190             } else {
2191                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2192                 if (!write_mask) continue;
2193             }
2194
2195             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2196             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2197             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2198
2199             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2200                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2201         }
2202
2203         if(temp_destination) {
2204             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2205             shader_glsl_append_dst(ins->ctx->buffer, ins);
2206             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2207         }
2208     }
2209
2210 }
2211
2212 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2213 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2214  * the compare is done per component of src0. */
2215 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2216 {
2217     struct wined3d_shader_dst_param dst;
2218     glsl_src_param_t src0_param;
2219     glsl_src_param_t src1_param;
2220     glsl_src_param_t src2_param;
2221     DWORD write_mask, cmp_channel = 0;
2222     unsigned int i, j;
2223     DWORD dst_mask;
2224     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2225             ins->ctx->reg_maps->shader_version.minor);
2226
2227     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2228     {
2229         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2230         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2231         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2232         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2233
2234         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2235         if (ins->coissue)
2236         {
2237             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2238         } else {
2239             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2240                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2241         }
2242         return;
2243     }
2244     /* Cycle through all source0 channels */
2245     dst_mask = ins->dst[0].write_mask;
2246     dst = ins->dst[0];
2247     for (i=0; i<4; i++) {
2248         write_mask = 0;
2249         /* Find the destination channels which use the current source0 channel */
2250         for (j=0; j<4; j++) {
2251             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2252             {
2253                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2254                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2255             }
2256         }
2257
2258         dst.write_mask = dst_mask & write_mask;
2259         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2260         if (!write_mask) continue;
2261
2262         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2263         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2264         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2265
2266         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2267                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2268     }
2269 }
2270
2271 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2272 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2273 {
2274     glsl_src_param_t src0_param;
2275     glsl_src_param_t src1_param;
2276     glsl_src_param_t src2_param;
2277     DWORD write_mask;
2278
2279     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2280     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2281     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2282     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2283     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2284             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2285 }
2286
2287 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2288    Vertex shaders to GLSL codes */
2289 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2290 {
2291     int i;
2292     int nComponents = 0;
2293     struct wined3d_shader_dst_param tmp_dst = {{0}};
2294     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2295     struct wined3d_shader_instruction tmp_ins;
2296
2297     memset(&tmp_ins, 0, sizeof(tmp_ins));
2298
2299     /* Set constants for the temporary argument */
2300     tmp_ins.ctx = ins->ctx;
2301     tmp_ins.dst_count = 1;
2302     tmp_ins.dst = &tmp_dst;
2303     tmp_ins.src_count = 2;
2304     tmp_ins.src = tmp_src;
2305
2306     switch(ins->handler_idx)
2307     {
2308         case WINED3DSIH_M4x4:
2309             nComponents = 4;
2310             tmp_ins.handler_idx = WINED3DSIH_DP4;
2311             break;
2312         case WINED3DSIH_M4x3:
2313             nComponents = 3;
2314             tmp_ins.handler_idx = WINED3DSIH_DP4;
2315             break;
2316         case WINED3DSIH_M3x4:
2317             nComponents = 4;
2318             tmp_ins.handler_idx = WINED3DSIH_DP3;
2319             break;
2320         case WINED3DSIH_M3x3:
2321             nComponents = 3;
2322             tmp_ins.handler_idx = WINED3DSIH_DP3;
2323             break;
2324         case WINED3DSIH_M3x2:
2325             nComponents = 2;
2326             tmp_ins.handler_idx = WINED3DSIH_DP3;
2327             break;
2328         default:
2329             break;
2330     }
2331
2332     tmp_dst = ins->dst[0];
2333     tmp_src[0] = ins->src[0];
2334     tmp_src[1] = ins->src[1];
2335     for (i = 0; i < nComponents; ++i)
2336     {
2337         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2338         shader_glsl_dot(&tmp_ins);
2339         ++tmp_src[1].reg.idx;
2340     }
2341 }
2342
2343 /**
2344     The LRP instruction performs a component-wise linear interpolation
2345     between the second and third operands using the first operand as the
2346     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2347     This is equivalent to mix(src2, src1, src0);
2348 */
2349 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2350 {
2351     glsl_src_param_t src0_param;
2352     glsl_src_param_t src1_param;
2353     glsl_src_param_t src2_param;
2354     DWORD write_mask;
2355
2356     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2357
2358     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2359     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2360     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2361
2362     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2363             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2364 }
2365
2366 /** Process the WINED3DSIO_LIT instruction in GLSL:
2367  * dst.x = dst.w = 1.0
2368  * dst.y = (src0.x > 0) ? src0.x
2369  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2370  *                                        where src.w is clamped at +- 128
2371  */
2372 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2373 {
2374     glsl_src_param_t src0_param;
2375     glsl_src_param_t src1_param;
2376     glsl_src_param_t src3_param;
2377     char dst_mask[6];
2378
2379     shader_glsl_append_dst(ins->ctx->buffer, ins);
2380     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2381
2382     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2383     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2384     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2385
2386     /* The sdk specifies the instruction like this
2387      * dst.x = 1.0;
2388      * if(src.x > 0.0) dst.y = src.x
2389      * else dst.y = 0.0.
2390      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2391      * else dst.z = 0.0;
2392      * dst.w = 1.0;
2393      *
2394      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2395      * dst.x = 1.0                                  ... No further explanation needed
2396      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2397      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2398      * dst.w = 1.0.                                 ... Nothing fancy.
2399      *
2400      * So we still have one conditional in there. So do this:
2401      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2402      *
2403      * 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),
2404      * 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.
2405      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2406      */
2407     shader_addline(ins->ctx->buffer,
2408             "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",
2409             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2410 }
2411
2412 /** Process the WINED3DSIO_DST instruction in GLSL:
2413  * dst.x = 1.0
2414  * dst.y = src0.x * src0.y
2415  * dst.z = src0.z
2416  * dst.w = src1.w
2417  */
2418 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2419 {
2420     glsl_src_param_t src0y_param;
2421     glsl_src_param_t src0z_param;
2422     glsl_src_param_t src1y_param;
2423     glsl_src_param_t src1w_param;
2424     char dst_mask[6];
2425
2426     shader_glsl_append_dst(ins->ctx->buffer, ins);
2427     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2428
2429     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2430     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2431     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2432     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2433
2434     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2435             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2436 }
2437
2438 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2439  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2440  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2441  *
2442  * dst.x = cos(src0.?)
2443  * dst.y = sin(src0.?)
2444  * dst.z = dst.z
2445  * dst.w = dst.w
2446  */
2447 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2448 {
2449     glsl_src_param_t src0_param;
2450     DWORD write_mask;
2451
2452     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2453     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2454
2455     switch (write_mask) {
2456         case WINED3DSP_WRITEMASK_0:
2457             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2458             break;
2459
2460         case WINED3DSP_WRITEMASK_1:
2461             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2462             break;
2463
2464         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2465             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2466             break;
2467
2468         default:
2469             ERR("Write mask should be .x, .y or .xy\n");
2470             break;
2471     }
2472 }
2473
2474 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2475  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2476  * generate invalid code
2477  */
2478 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2479 {
2480     glsl_src_param_t src0_param;
2481     DWORD write_mask;
2482
2483     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2484     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2485
2486     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2487 }
2488
2489 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2490  * Start a for() loop where src1.y is the initial value of aL,
2491  *  increment aL by src1.z for a total of src1.x iterations.
2492  *  Need to use a temporary variable for this operation.
2493  */
2494 /* FIXME: I don't think nested loops will work correctly this way. */
2495 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2496 {
2497     glsl_src_param_t src1_param;
2498     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2499     const DWORD *control_values = NULL;
2500     const local_constant *constant;
2501
2502     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2503
2504     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2505      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2506      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2507      * addressing.
2508      */
2509     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2510     {
2511         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2512             if (constant->idx == ins->src[1].reg.idx)
2513             {
2514                 control_values = constant->value;
2515                 break;
2516             }
2517         }
2518     }
2519
2520     if (control_values)
2521     {
2522         struct wined3d_shader_loop_control loop_control;
2523         loop_control.count = control_values[0];
2524         loop_control.start = control_values[1];
2525         loop_control.step = (int)control_values[2];
2526
2527         if (loop_control.step > 0)
2528         {
2529             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2530                     shader->baseShader.cur_loop_depth, loop_control.start,
2531                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2532                     shader->baseShader.cur_loop_depth, loop_control.step);
2533         }
2534         else if (loop_control.step < 0)
2535         {
2536             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2537                     shader->baseShader.cur_loop_depth, loop_control.start,
2538                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2539                     shader->baseShader.cur_loop_depth, loop_control.step);
2540         }
2541         else
2542         {
2543             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2544                     shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2545                     shader->baseShader.cur_loop_depth, loop_control.count,
2546                     shader->baseShader.cur_loop_depth);
2547         }
2548     } else {
2549         shader_addline(ins->ctx->buffer,
2550                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2551                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2552                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2553                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2554     }
2555
2556     shader->baseShader.cur_loop_depth++;
2557     shader->baseShader.cur_loop_regno++;
2558 }
2559
2560 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2561 {
2562     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2563
2564     shader_addline(ins->ctx->buffer, "}\n");
2565
2566     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2567     {
2568         shader->baseShader.cur_loop_depth--;
2569         shader->baseShader.cur_loop_regno--;
2570     }
2571
2572     if (ins->handler_idx == WINED3DSIH_ENDREP)
2573     {
2574         shader->baseShader.cur_loop_depth--;
2575     }
2576 }
2577
2578 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2579 {
2580     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2581     glsl_src_param_t src0_param;
2582     const DWORD *control_values = NULL;
2583     const local_constant *constant;
2584
2585     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2586     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2587     {
2588         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2589         {
2590             if (constant->idx == ins->src[0].reg.idx)
2591             {
2592                 control_values = constant->value;
2593                 break;
2594             }
2595         }
2596     }
2597
2598     if(control_values) {
2599         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2600                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2601                        control_values[0], shader->baseShader.cur_loop_depth);
2602     } else {
2603         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2604         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2605                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2606                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2607     }
2608     shader->baseShader.cur_loop_depth++;
2609 }
2610
2611 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2612 {
2613     glsl_src_param_t src0_param;
2614
2615     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2616     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2617 }
2618
2619 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2620 {
2621     glsl_src_param_t src0_param;
2622     glsl_src_param_t src1_param;
2623
2624     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2625     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2626
2627     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2628             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2629 }
2630
2631 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2632 {
2633     shader_addline(ins->ctx->buffer, "} else {\n");
2634 }
2635
2636 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2637 {
2638     shader_addline(ins->ctx->buffer, "break;\n");
2639 }
2640
2641 /* FIXME: According to MSDN the compare is done per component. */
2642 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2643 {
2644     glsl_src_param_t src0_param;
2645     glsl_src_param_t src1_param;
2646
2647     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2648     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2649
2650     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2651             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2652 }
2653
2654 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2655 {
2656     shader_addline(ins->ctx->buffer, "}\n");
2657     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2658 }
2659
2660 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2661 {
2662     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2663 }
2664
2665 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2666 {
2667     glsl_src_param_t src1_param;
2668
2669     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2670     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
2671 }
2672
2673 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
2674 {
2675     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
2676      * function only suppresses the unhandled instruction warning
2677      */
2678 }
2679
2680 /*********************************************
2681  * Pixel Shader Specific Code begins here
2682  ********************************************/
2683 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
2684 {
2685     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2686     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2687     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2688             ins->ctx->reg_maps->shader_version.minor);
2689     glsl_sample_function_t sample_function;
2690     DWORD sample_flags = 0;
2691     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
2692     DWORD sampler_idx;
2693     DWORD mask = 0, swizzle;
2694
2695     /* 1.0-1.4: Use destination register as sampler source.
2696      * 2.0+: Use provided sampler source. */
2697     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
2698     else sampler_idx = ins->src[1].reg.idx;
2699     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2700
2701     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2702     {
2703         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2704
2705         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2706         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2707             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2708             switch (flags & ~WINED3DTTFF_PROJECTED) {
2709                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2710                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2711                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2712                 case WINED3DTTFF_COUNT4:
2713                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2714             }
2715         }
2716     }
2717     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
2718     {
2719         DWORD src_mod = ins->src[0].modifiers;
2720
2721         if (src_mod == WINED3DSPSM_DZ) {
2722             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2723             mask = WINED3DSP_WRITEMASK_2;
2724         } else if (src_mod == WINED3DSPSM_DW) {
2725             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2726             mask = WINED3DSP_WRITEMASK_3;
2727         }
2728     } else {
2729         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
2730         {
2731             /* ps 2.0 texldp instruction always divides by the fourth component. */
2732             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2733             mask = WINED3DSP_WRITEMASK_3;
2734         }
2735     }
2736
2737     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2738        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2739         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2740     }
2741
2742     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2743     mask |= sample_function.coord_mask;
2744
2745     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
2746     else swizzle = ins->src[1].swizzle;
2747
2748     /* 1.0-1.3: Use destination register as coordinate source.
2749        1.4+: Use provided coordinate source register. */
2750     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2751     {
2752         char coord_mask[6];
2753         shader_glsl_write_mask_to_str(mask, coord_mask);
2754         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2755                 "T%u%s", sampler_idx, coord_mask);
2756     } else {
2757         glsl_src_param_t coord_param;
2758         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
2759         if (ins->flags & WINED3DSI_TEXLD_BIAS)
2760         {
2761             glsl_src_param_t bias;
2762             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
2763             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
2764                     "%s", coord_param.param_str);
2765         } else {
2766             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2767                     "%s", coord_param.param_str);
2768         }
2769     }
2770 }
2771
2772 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
2773 {
2774     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2775     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2776     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2777     glsl_sample_function_t sample_function;
2778     glsl_src_param_t coord_param, dx_param, dy_param;
2779     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
2780     DWORD sampler_type;
2781     DWORD sampler_idx;
2782     DWORD swizzle = ins->src[1].swizzle;
2783
2784     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD])
2785     {
2786         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
2787         return shader_glsl_tex(ins);
2788     }
2789
2790     sampler_idx = ins->src[1].reg.idx;
2791     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2792     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2793        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2794         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2795     }
2796
2797     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2798     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2799     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
2800     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
2801
2802     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
2803                                 "%s", coord_param.param_str);
2804 }
2805
2806 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
2807 {
2808     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2809     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2810     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2811     glsl_sample_function_t sample_function;
2812     glsl_src_param_t coord_param, lod_param;
2813     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
2814     DWORD sampler_type;
2815     DWORD sampler_idx;
2816     DWORD swizzle = ins->src[1].swizzle;
2817
2818     sampler_idx = ins->src[1].reg.idx;
2819     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2820     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2821        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2822         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2823     }
2824     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2825     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2826
2827     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
2828
2829     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD]
2830             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2831     {
2832         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2833          * However, they seem to work just fine in fragment shaders as well. */
2834         WARN("Using %s in fragment shader.\n", sample_function.name);
2835     }
2836     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
2837             "%s", coord_param.param_str);
2838 }
2839
2840 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
2841 {
2842     /* FIXME: Make this work for more than just 2D textures */
2843     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2844     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2845
2846     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
2847     {
2848         char dst_mask[6];
2849
2850         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2851         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
2852                 ins->dst[0].reg.idx, dst_mask);
2853     } else {
2854         DWORD reg = ins->src[0].reg.idx;
2855         DWORD src_mod = ins->src[0].modifiers;
2856         char dst_swizzle[6];
2857
2858         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
2859
2860         if (src_mod == WINED3DSPSM_DZ) {
2861             glsl_src_param_t div_param;
2862             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2863             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
2864
2865             if (mask_size > 1) {
2866                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2867             } else {
2868                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2869             }
2870         } else if (src_mod == WINED3DSPSM_DW) {
2871             glsl_src_param_t div_param;
2872             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2873             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
2874
2875             if (mask_size > 1) {
2876                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2877             } else {
2878                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2879             }
2880         } else {
2881             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2882         }
2883     }
2884 }
2885
2886 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2887  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2888  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2889 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
2890 {
2891     glsl_src_param_t src0_param;
2892     glsl_sample_function_t sample_function;
2893     DWORD sampler_idx = ins->dst[0].reg.idx;
2894     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2895     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2896     UINT mask_size;
2897
2898     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2899
2900     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2901      * scalar, and projected sampling would require 4.
2902      *
2903      * It is a dependent read - not valid with conditional NP2 textures
2904      */
2905     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2906     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
2907
2908     switch(mask_size)
2909     {
2910         case 1:
2911             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2912                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
2913             break;
2914
2915         case 2:
2916             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2917                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
2918             break;
2919
2920         case 3:
2921             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2922                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
2923             break;
2924
2925         default:
2926             FIXME("Unexpected mask size %u\n", mask_size);
2927             break;
2928     }
2929 }
2930
2931 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2932  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2933 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
2934 {
2935     glsl_src_param_t src0_param;
2936     DWORD dstreg = ins->dst[0].reg.idx;
2937     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2938     DWORD dst_mask;
2939     unsigned int mask_size;
2940
2941     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2942     mask_size = shader_glsl_get_write_mask_size(dst_mask);
2943     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2944
2945     if (mask_size > 1) {
2946         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2947     } else {
2948         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2949     }
2950 }
2951
2952 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2953  * Calculate the depth as dst.x / dst.y   */
2954 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
2955 {
2956     glsl_dst_param_t dst_param;
2957
2958     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
2959
2960     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2961      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2962      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2963      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2964      * >= 1.0 or < 0.0
2965      */
2966     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
2967             dst_param.reg_name, dst_param.reg_name);
2968 }
2969
2970 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2971  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2972  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2973  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2974  */
2975 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
2976 {
2977     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2978     DWORD dstreg = ins->dst[0].reg.idx;
2979     glsl_src_param_t src0_param;
2980
2981     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2982
2983     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2984     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2985 }
2986
2987 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2988  * Calculate the 1st of a 2-row matrix multiplication. */
2989 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
2990 {
2991     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2992     DWORD reg = ins->dst[0].reg.idx;
2993     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2994     glsl_src_param_t src0_param;
2995
2996     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2997     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2998 }
2999
3000 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3001  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3002 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3003 {
3004     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3005     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3006     DWORD reg = ins->dst[0].reg.idx;
3007     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3008     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3009     glsl_src_param_t src0_param;
3010
3011     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3012     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3013     current_state->texcoord_w[current_state->current_row++] = reg;
3014 }
3015
3016 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3017 {
3018     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3019     DWORD reg = ins->dst[0].reg.idx;
3020     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3021     glsl_src_param_t src0_param;
3022     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3023     glsl_sample_function_t sample_function;
3024
3025     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3026     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3027
3028     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3029
3030     /* Sample the texture using the calculated coordinates */
3031     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3032 }
3033
3034 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3035  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3036 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3037 {
3038     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3039     glsl_src_param_t src0_param;
3040     DWORD reg = ins->dst[0].reg.idx;
3041     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3042     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
3043     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3044     glsl_sample_function_t sample_function;
3045
3046     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3047     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3048
3049     /* Dependent read, not valid with conditional NP2 */
3050     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3051
3052     /* Sample the texture using the calculated coordinates */
3053     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3054
3055     current_state->current_row = 0;
3056 }
3057
3058 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3059  * Perform the 3rd row of a 3x3 matrix multiply */
3060 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3061 {
3062     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3063     glsl_src_param_t src0_param;
3064     char dst_mask[6];
3065     DWORD reg = ins->dst[0].reg.idx;
3066     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3067     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
3068
3069     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3070
3071     shader_glsl_append_dst(ins->ctx->buffer, ins);
3072     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3073     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3074
3075     current_state->current_row = 0;
3076 }
3077
3078 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3079  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3080 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3081 {
3082     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3083     DWORD reg = ins->dst[0].reg.idx;
3084     glsl_src_param_t src0_param;
3085     glsl_src_param_t src1_param;
3086     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3087     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3088     WINED3DSAMPLER_TEXTURE_TYPE stype = ins->ctx->reg_maps->sampler_type[reg];
3089     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3090     glsl_sample_function_t sample_function;
3091
3092     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3093     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3094
3095     /* Perform the last matrix multiply operation */
3096     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3097     /* Reflection calculation */
3098     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3099
3100     /* Dependent read, not valid with conditional NP2 */
3101     shader_glsl_get_sample_function(stype, 0, &sample_function);
3102
3103     /* Sample the texture */
3104     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3105
3106     current_state->current_row = 0;
3107 }
3108
3109 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3110  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3111 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3112 {
3113     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3114     DWORD reg = ins->dst[0].reg.idx;
3115     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3116     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3117     glsl_src_param_t src0_param;
3118     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3119     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3120     glsl_sample_function_t sample_function;
3121
3122     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3123
3124     /* Perform the last matrix multiply operation */
3125     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3126
3127     /* Construct the eye-ray vector from w coordinates */
3128     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3129             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3130     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3131
3132     /* Dependent read, not valid with conditional NP2 */
3133     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3134
3135     /* Sample the texture using the calculated coordinates */
3136     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3137
3138     current_state->current_row = 0;
3139 }
3140
3141 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3142  * Apply a fake bump map transform.
3143  * texbem is pshader <= 1.3 only, this saves a few version checks
3144  */
3145 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3146 {
3147     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3148     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3149     glsl_sample_function_t sample_function;
3150     glsl_src_param_t coord_param;
3151     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
3152     DWORD sampler_idx;
3153     DWORD mask;
3154     DWORD flags;
3155     char coord_mask[6];
3156
3157     sampler_idx = ins->dst[0].reg.idx;
3158     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3159
3160     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3161     /* Dependent read, not valid with conditional NP2 */
3162     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3163     mask = sample_function.coord_mask;
3164
3165     shader_glsl_write_mask_to_str(mask, coord_mask);
3166
3167     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3168          * so we can't let the GL handle this.
3169          */
3170     if (flags & WINED3DTTFF_PROJECTED) {
3171         DWORD div_mask=0;
3172         char coord_div_mask[3];
3173         switch (flags & ~WINED3DTTFF_PROJECTED) {
3174             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3175             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3176             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3177             case WINED3DTTFF_COUNT4:
3178             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3179         }
3180         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3181         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3182     }
3183
3184     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3185
3186     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3187             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3188             coord_param.param_str, coord_mask);
3189
3190     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3191     {
3192         glsl_src_param_t luminance_param;
3193         glsl_dst_param_t dst_param;
3194
3195         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3196         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3197
3198         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3199                 dst_param.reg_name, dst_param.mask_str,
3200                 luminance_param.param_str, sampler_idx, sampler_idx);
3201     }
3202 }
3203
3204 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3205 {
3206     glsl_src_param_t src0_param, src1_param;
3207     DWORD sampler_idx = ins->dst[0].reg.idx;
3208
3209     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3210     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3211
3212     shader_glsl_append_dst(ins->ctx->buffer, ins);
3213     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3214             src0_param.param_str, sampler_idx, src1_param.param_str);
3215 }
3216
3217 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3218  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3219 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3220 {
3221     glsl_src_param_t src0_param;
3222     DWORD sampler_idx = ins->dst[0].reg.idx;
3223     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3224     glsl_sample_function_t sample_function;
3225
3226     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3227
3228     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3229     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3230             "%s.wx", src0_param.reg_name);
3231 }
3232
3233 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3234  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3235 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3236 {
3237     glsl_src_param_t src0_param;
3238     DWORD sampler_idx = ins->dst[0].reg.idx;
3239     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3240     glsl_sample_function_t sample_function;
3241
3242     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3243
3244     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3245     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3246             "%s.yz", src0_param.reg_name);
3247 }
3248
3249 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3250  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3251 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3252 {
3253     glsl_src_param_t src0_param;
3254     DWORD sampler_idx = ins->dst[0].reg.idx;
3255     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3256     glsl_sample_function_t sample_function;
3257
3258     /* Dependent read, not valid with conditional NP2 */
3259     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3260     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3261
3262     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3263             "%s", src0_param.param_str);
3264 }
3265
3266 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3267  * If any of the first 3 components are < 0, discard this pixel */
3268 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3269 {
3270     glsl_dst_param_t dst_param;
3271
3272     /* The argument is a destination parameter, and no writemasks are allowed */
3273     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3274     if (ins->ctx->reg_maps->shader_version.major >= 2)
3275     {
3276         /* 2.0 shaders compare all 4 components in texkill */
3277         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3278     } else {
3279         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3280          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3281          * 4 components are defined, only the first 3 are used
3282          */
3283         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3284     }
3285 }
3286
3287 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3288  * dst = dot2(src0, src1) + src2 */
3289 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3290 {
3291     glsl_src_param_t src0_param;
3292     glsl_src_param_t src1_param;
3293     glsl_src_param_t src2_param;
3294     DWORD write_mask;
3295     unsigned int mask_size;
3296
3297     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3298     mask_size = shader_glsl_get_write_mask_size(write_mask);
3299
3300     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3301     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3302     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3303
3304     if (mask_size > 1) {
3305         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3306                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3307     } else {
3308         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3309                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3310     }
3311 }
3312
3313 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3314         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3315         enum vertexprocessing_mode vertexprocessing)
3316 {
3317     unsigned int i;
3318     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3319     WORD map = reg_maps->input_registers;
3320
3321     for (i = 0; map; map >>= 1, ++i)
3322     {
3323         const char *semantic_name;
3324         UINT semantic_idx;
3325         char reg_mask[6];
3326
3327         /* Unused */
3328         if (!(map & 1)) continue;
3329
3330         semantic_name = input_signature[i].semantic_name;
3331         semantic_idx = input_signature[i].semantic_idx;
3332         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3333
3334         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3335         {
3336             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3337                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3338                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3339             else
3340                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3341                         This->input_reg_map[i], reg_mask, reg_mask);
3342         }
3343         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3344         {
3345             if (semantic_idx == 0)
3346                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3347                         This->input_reg_map[i], reg_mask, reg_mask);
3348             else if (semantic_idx == 1)
3349                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3350                         This->input_reg_map[i], reg_mask, reg_mask);
3351             else
3352                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3353                         This->input_reg_map[i], reg_mask, reg_mask);
3354         }
3355         else
3356         {
3357             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3358                     This->input_reg_map[i], reg_mask, reg_mask);
3359         }
3360     }
3361 }
3362
3363 /*********************************************
3364  * Vertex Shader Specific Code begins here
3365  ********************************************/
3366
3367 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3368     glsl_program_key_t key;
3369
3370     key.vshader = entry->vshader;
3371     key.pshader = entry->pshader;
3372     key.vs_args = entry->vs_args;
3373     key.ps_args = entry->ps_args;
3374
3375     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3376     {
3377         ERR("Failed to insert program entry.\n");
3378     }
3379 }
3380
3381 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3382         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3383         struct ps_compile_args *ps_args) {
3384     struct wine_rb_entry *entry;
3385     glsl_program_key_t key;
3386
3387     key.vshader = vshader;
3388     key.pshader = pshader;
3389     key.vs_args = *vs_args;
3390     key.ps_args = *ps_args;
3391
3392     entry = wine_rb_get(&priv->program_lookup, &key);
3393     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3394 }
3395
3396 /* GL locking is done by the caller */
3397 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3398         struct glsl_shader_prog_link *entry)
3399 {
3400     glsl_program_key_t key;
3401
3402     key.vshader = entry->vshader;
3403     key.pshader = entry->pshader;
3404     key.vs_args = entry->vs_args;
3405     key.ps_args = entry->ps_args;
3406     wine_rb_remove(&priv->program_lookup, &key);
3407
3408     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3409     if (entry->vshader) list_remove(&entry->vshader_entry);
3410     if (entry->pshader) list_remove(&entry->pshader_entry);
3411     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3412     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3413     HeapFree(GetProcessHeap(), 0, entry);
3414 }
3415
3416 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3417         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3418         const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3419 {
3420     unsigned int i, j;
3421     const char *semantic_name_in, *semantic_name_out;
3422     UINT semantic_idx_in, semantic_idx_out;
3423     DWORD *set;
3424     DWORD in_idx;
3425     unsigned int in_count = vec4_varyings(3, gl_info);
3426     char reg_mask[6], reg_mask_out[6];
3427     char destination[50];
3428     WORD input_map, output_map;
3429
3430     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3431
3432     if (!output_signature)
3433     {
3434         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3435         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3436         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3437     }
3438
3439     input_map = reg_maps_in->input_registers;
3440     for (i = 0; input_map; input_map >>= 1, ++i)
3441     {
3442         if (!(input_map & 1)) continue;
3443
3444         in_idx = map[i];
3445         if (in_idx >= (in_count + 2)) {
3446             FIXME("More input varyings declared than supported, expect issues\n");
3447             continue;
3448         }
3449         else if (map[i] == ~0U)
3450         {
3451             /* Declared, but not read register */
3452             continue;
3453         }
3454
3455         if (in_idx == in_count) {
3456             sprintf(destination, "gl_FrontColor");
3457         } else if (in_idx == in_count + 1) {
3458             sprintf(destination, "gl_FrontSecondaryColor");
3459         } else {
3460             sprintf(destination, "IN[%u]", in_idx);
3461         }
3462
3463         semantic_name_in = input_signature[i].semantic_name;
3464         semantic_idx_in = input_signature[i].semantic_idx;
3465         set[map[i]] = input_signature[i].mask;
3466         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3467
3468         if (!output_signature)
3469         {
3470             if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_COLOR))
3471             {
3472                 if (semantic_idx_in == 0)
3473                     shader_addline(buffer, "%s%s = front_color%s;\n",
3474                             destination, reg_mask, reg_mask);
3475                 else if (semantic_idx_in == 1)
3476                     shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3477                             destination, reg_mask, reg_mask);
3478                 else
3479                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3480                             destination, reg_mask, reg_mask);
3481             }
3482             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_TEXCOORD))
3483             {
3484                 if (semantic_idx_in < 8)
3485                 {
3486                     shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3487                             destination, reg_mask, semantic_idx_in, reg_mask);
3488                 }
3489                 else
3490                 {
3491                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3492                             destination, reg_mask, reg_mask);
3493                 }
3494             }
3495             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_FOG))
3496             {
3497                 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3498                         destination, reg_mask, reg_mask);
3499             }
3500             else
3501             {
3502                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3503                         destination, reg_mask, reg_mask);
3504             }
3505         } else {
3506             BOOL found = FALSE;
3507
3508             output_map = reg_maps_out->output_registers;
3509             for (j = 0; output_map; output_map >>= 1, ++j)
3510             {
3511                 if (!(output_map & 1)) continue;
3512
3513                 semantic_name_out = output_signature[j].semantic_name;
3514                 semantic_idx_out = output_signature[j].semantic_idx;
3515                 shader_glsl_write_mask_to_str(output_signature[j].mask, reg_mask_out);
3516
3517                 if (semantic_idx_in == semantic_idx_out
3518                         && !strcmp(semantic_name_in, semantic_name_out))
3519                 {
3520                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3521                             destination, reg_mask, j, reg_mask);
3522                     found = TRUE;
3523                 }
3524             }
3525             if(!found) {
3526                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3527                                destination, reg_mask, reg_mask);
3528             }
3529         }
3530     }
3531
3532     /* This is solely to make the compiler / linker happy and avoid warning about undefined
3533      * varyings. It shouldn't result in any real code executed on the GPU, since all read
3534      * input varyings are assigned above, if the optimizer works properly.
3535      */
3536     for(i = 0; i < in_count + 2; i++) {
3537         if (set[i] && set[i] != WINED3DSP_WRITEMASK_ALL)
3538         {
3539             unsigned int size = 0;
3540             memset(reg_mask, 0, sizeof(reg_mask));
3541             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3542                 reg_mask[size] = 'x';
3543                 size++;
3544             }
3545             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3546                 reg_mask[size] = 'y';
3547                 size++;
3548             }
3549             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3550                 reg_mask[size] = 'z';
3551                 size++;
3552             }
3553             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3554                 reg_mask[size] = 'w';
3555                 size++;
3556             }
3557
3558             if (i == in_count) {
3559                 sprintf(destination, "gl_FrontColor");
3560             } else if (i == in_count + 1) {
3561                 sprintf(destination, "gl_FrontSecondaryColor");
3562             } else {
3563                 sprintf(destination, "IN[%u]", i);
3564             }
3565
3566             if (size == 1) {
3567                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3568             } else {
3569                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3570             }
3571         }
3572     }
3573
3574     HeapFree(GetProcessHeap(), 0, set);
3575 }
3576
3577 /* GL locking is done by the caller */
3578 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3579         IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3580 {
3581     GLhandleARB ret = 0;
3582     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3583     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3584     IWineD3DDeviceImpl *device;
3585     DWORD vs_major = vs->baseShader.reg_maps.shader_version.major;
3586     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3587     unsigned int i;
3588     const char *semantic_name;
3589     UINT semantic_idx;
3590     char reg_mask[6];
3591     const struct wined3d_shader_signature_element *output_signature;
3592
3593     shader_buffer_clear(buffer);
3594
3595     shader_addline(buffer, "#version 120\n");
3596
3597     if(vs_major < 3 && ps_major < 3) {
3598         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3599          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3600          */
3601         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3602         if ((gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W)
3603                 && ps_major == 0 && vs_major > 0 && !device->frag_pipe->ffp_proj_control)
3604         {
3605             shader_addline(buffer, "void order_ps_input() {\n");
3606             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3607                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3608                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3609                     shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3610                 }
3611             }
3612             shader_addline(buffer, "}\n");
3613         } else {
3614             shader_addline(buffer, "void order_ps_input() { /* do nothing */ }\n");
3615         }
3616     } else if(ps_major < 3 && vs_major >= 3) {
3617         WORD map = vs->baseShader.reg_maps.output_registers;
3618
3619         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3620         output_signature = vs->baseShader.output_signature;
3621
3622         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3623         for (i = 0; map; map >>= 1, ++i)
3624         {
3625             DWORD write_mask;
3626
3627             if (!(map & 1)) continue;
3628
3629             semantic_name = output_signature[i].semantic_name;
3630             semantic_idx = output_signature[i].semantic_idx;
3631             write_mask = output_signature[i].mask;
3632             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3633
3634             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3635             {
3636                 if (semantic_idx == 0)
3637                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3638                 else if (semantic_idx == 1)
3639                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3640             }
3641             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3642             {
3643                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3644             }
3645             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3646             {
3647                 if (semantic_idx < 8)
3648                 {
3649                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3650                         write_mask |= WINED3DSP_WRITEMASK_3;
3651
3652                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3653                             semantic_idx, reg_mask, i, reg_mask);
3654                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3655                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3656                 }
3657             }
3658             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3659             {
3660                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3661             }
3662             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3663             {
3664                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3665             }
3666         }
3667         shader_addline(buffer, "}\n");
3668
3669     } else if(ps_major >= 3 && vs_major >= 3) {
3670         WORD map = vs->baseShader.reg_maps.output_registers;
3671
3672         output_signature = vs->baseShader.output_signature;
3673
3674         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3675         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3676         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3677
3678         /* First, sort out position and point size. Those are not passed to the pixel shader */
3679         for (i = 0; map; map >>= 1, ++i)
3680         {
3681             if (!(map & 1)) continue;
3682
3683             semantic_name = output_signature[i].semantic_name;
3684             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3685
3686             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3687             {
3688                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3689             }
3690             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3691             {
3692                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3693             }
3694         }
3695
3696         /* Then, fix the pixel shader input */
3697         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3698                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3699
3700         shader_addline(buffer, "}\n");
3701     } else if(ps_major >= 3 && vs_major < 3) {
3702         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3703         shader_addline(buffer, "void order_ps_input() {\n");
3704         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3705          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3706          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3707          */
3708         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3709                 &ps->baseShader.reg_maps, NULL, NULL);
3710         shader_addline(buffer, "}\n");
3711     } else {
3712         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3713     }
3714
3715     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3716     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3717     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer->buffer, NULL));
3718     checkGLcall("glShaderSourceARB(ret, 1, &buffer->buffer, NULL)");
3719     GL_EXTCALL(glCompileShaderARB(ret));
3720     checkGLcall("glCompileShaderARB(ret)");
3721
3722     return ret;
3723 }
3724
3725 /* GL locking is done by the caller */
3726 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3727         GLhandleARB programId, char prefix)
3728 {
3729     const local_constant *lconst;
3730     GLint tmp_loc;
3731     const float *value;
3732     char glsl_name[8];
3733
3734     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3735         value = (const float *)lconst->value;
3736         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3737         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3738         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3739     }
3740     checkGLcall("Hardcoding local constants");
3741 }
3742
3743 /* GL locking is done by the caller */
3744 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3745         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3746         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3747 {
3748     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3749     const struct wined3d_gl_info *gl_info = context->gl_info;
3750     CONST DWORD *function = This->baseShader.function;
3751     struct shader_glsl_ctx_priv priv_ctx;
3752
3753     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3754     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3755
3756     memset(&priv_ctx, 0, sizeof(priv_ctx));
3757     priv_ctx.cur_ps_args = args;
3758     priv_ctx.cur_np2fixup_info = np2fixup_info;
3759
3760     shader_addline(buffer, "#version 120\n");
3761
3762     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3763     {
3764         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3765     }
3766     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3767     {
3768         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3769          * drivers write a warning if we don't do so
3770          */
3771         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3772     }
3773     if (gl_info->supported[EXT_GPU_SHADER4])
3774     {
3775         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3776     }
3777
3778     /* Base Declarations */
3779     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3780
3781     /* Pack 3.0 inputs */
3782     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3783     {
3784         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
3785                 This->baseShader.input_signature, reg_maps, args->vp_mode);
3786     }
3787
3788     /* Base Shader Body */
3789     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3790
3791     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3792     if (reg_maps->shader_version.major < 2)
3793     {
3794         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3795         shader_addline(buffer, "gl_FragData[0] = R0;\n");
3796     }
3797
3798     if (args->srgb_correction)
3799     {
3800         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
3801         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
3802         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
3803         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
3804         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
3805         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
3806     }
3807     /* Pixel shader < 3.0 do not replace the fog stage.
3808      * This implements linear fog computation and blending.
3809      * TODO: non linear fog
3810      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3811      * -1/(e-s) and e/(e-s) respectively.
3812      */
3813     if (reg_maps->shader_version.major < 3)
3814     {
3815         switch(args->fog) {
3816             case FOG_OFF: break;
3817             case FOG_LINEAR:
3818                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
3819                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
3820                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
3821                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
3822                 break;
3823             case FOG_EXP:
3824                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
3825                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
3826                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
3827                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
3828                 break;
3829             case FOG_EXP2:
3830                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
3831                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
3832                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
3833                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
3834                 break;
3835         }
3836     }
3837
3838     shader_addline(buffer, "}\n");
3839
3840     TRACE("Compiling shader object %u\n", shader_obj);
3841     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3842     GL_EXTCALL(glCompileShaderARB(shader_obj));
3843     print_glsl_info_log(gl_info, shader_obj);
3844
3845     /* Store the shader object */
3846     return shader_obj;
3847 }
3848
3849 /* GL locking is done by the caller */
3850 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
3851         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
3852         const struct vs_compile_args *args)
3853 {
3854     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3855     const struct wined3d_gl_info *gl_info = context->gl_info;
3856     CONST DWORD *function = This->baseShader.function;
3857     struct shader_glsl_ctx_priv priv_ctx;
3858
3859     /* Create the hw GLSL shader program and assign it as the shader->prgId */
3860     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3861
3862     shader_addline(buffer, "#version 120\n");
3863
3864     if (gl_info->supported[EXT_GPU_SHADER4])
3865     {
3866         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3867     }
3868
3869     memset(&priv_ctx, 0, sizeof(priv_ctx));
3870     priv_ctx.cur_vs_args = args;
3871
3872     /* Base Declarations */
3873     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3874
3875     /* Base Shader Body */
3876     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
3877
3878     /* Unpack 3.0 outputs */
3879     if (reg_maps->shader_version.major >= 3) shader_addline(buffer, "order_ps_input(OUT);\n");
3880     else shader_addline(buffer, "order_ps_input();\n");
3881
3882     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3883      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3884      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3885      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3886      */
3887     if(args->fog_src == VS_FOG_Z) {
3888         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3889     } else if (!reg_maps->fog) {
3890         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
3891     }
3892
3893     /* Write the final position.
3894      *
3895      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3896      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3897      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3898      * contains 1.0 to allow a mad.
3899      */
3900     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3901     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3902     if(args->clip_enabled) {
3903         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
3904     }
3905
3906     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3907      *
3908      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3909      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3910      * which is the same as z = z * 2 - w.
3911      */
3912     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3913
3914     shader_addline(buffer, "}\n");
3915
3916     TRACE("Compiling shader object %u\n", shader_obj);
3917     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3918     GL_EXTCALL(glCompileShaderARB(shader_obj));
3919     print_glsl_info_log(gl_info, shader_obj);
3920
3921     return shader_obj;
3922 }
3923
3924 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
3925         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
3926         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
3927 {
3928     UINT i;
3929     DWORD new_size;
3930     struct glsl_ps_compiled_shader *new_array;
3931     struct glsl_pshader_private    *shader_data;
3932     struct ps_np2fixup_info        *np2fixup = NULL;
3933     GLhandleARB ret;
3934
3935     if (!shader->baseShader.backend_data)
3936     {
3937         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3938         if (!shader->baseShader.backend_data)
3939         {
3940             ERR("Failed to allocate backend data.\n");
3941             return 0;
3942         }
3943     }
3944     shader_data = shader->baseShader.backend_data;
3945
3946     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3947      * so a linear search is more performant than a hashmap or a binary search
3948      * (cache coherency etc)
3949      */
3950     for(i = 0; i < shader_data->num_gl_shaders; i++) {
3951         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
3952             if(args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
3953             return shader_data->gl_shaders[i].prgId;
3954         }
3955     }
3956
3957     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
3958     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
3959         if (shader_data->num_gl_shaders)
3960         {
3961             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
3962             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
3963                                     new_size * sizeof(*shader_data->gl_shaders));
3964         } else {
3965             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
3966             new_size = 1;
3967         }
3968
3969         if(!new_array) {
3970             ERR("Out of memory\n");
3971             return 0;
3972         }
3973         shader_data->gl_shaders = new_array;
3974         shader_data->shader_array_size = new_size;
3975     }
3976
3977     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
3978
3979     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
3980     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
3981
3982     pixelshader_update_samplers(&shader->baseShader.reg_maps,
3983             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
3984
3985     shader_buffer_clear(buffer);
3986     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
3987     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
3988     *np2fixup_info = np2fixup;
3989
3990     return ret;
3991 }
3992
3993 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
3994                                  const DWORD use_map) {
3995     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
3996     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
3997     return stored->fog_src == new->fog_src;
3998 }
3999
4000 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4001         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4002         const struct vs_compile_args *args)
4003 {
4004     UINT i;
4005     DWORD new_size;
4006     struct glsl_vs_compiled_shader *new_array;
4007     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4008     struct glsl_vshader_private *shader_data;
4009     GLhandleARB ret;
4010
4011     if (!shader->baseShader.backend_data)
4012     {
4013         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4014         if (!shader->baseShader.backend_data)
4015         {
4016             ERR("Failed to allocate backend data.\n");
4017             return 0;
4018         }
4019     }
4020     shader_data = shader->baseShader.backend_data;
4021
4022     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4023      * so a linear search is more performant than a hashmap or a binary search
4024      * (cache coherency etc)
4025      */
4026     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4027         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4028             return shader_data->gl_shaders[i].prgId;
4029         }
4030     }
4031
4032     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4033
4034     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4035         if (shader_data->num_gl_shaders)
4036         {
4037             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4038             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4039                                     new_size * sizeof(*shader_data->gl_shaders));
4040         } else {
4041             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4042             new_size = 1;
4043         }
4044
4045         if(!new_array) {
4046             ERR("Out of memory\n");
4047             return 0;
4048         }
4049         shader_data->gl_shaders = new_array;
4050         shader_data->shader_array_size = new_size;
4051     }
4052
4053     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4054
4055     shader_buffer_clear(buffer);
4056     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4057     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4058
4059     return ret;
4060 }
4061
4062 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4063  * It sets the programId on the current StateBlock (because it should be called
4064  * inside of the DrawPrimitive() part of the render loop).
4065  *
4066  * If a program for the given combination does not exist, create one, and store
4067  * the program in the hash table.  If it creates a program, it will link the
4068  * given objects, too.
4069  */
4070
4071 /* GL locking is done by the caller */
4072 static void set_glsl_shader_program(const struct wined3d_context *context,
4073         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4074 {
4075     IWineD3DVertexShader *vshader = use_vs ? device->stateBlock->vertexShader : NULL;
4076     IWineD3DPixelShader *pshader = use_ps ? device->stateBlock->pixelShader : NULL;
4077     const struct wined3d_gl_info *gl_info = context->gl_info;
4078     struct shader_glsl_priv *priv = device->shader_priv;
4079     struct glsl_shader_prog_link *entry    = NULL;
4080     GLhandleARB programId                  = 0;
4081     GLhandleARB reorder_shader_id          = 0;
4082     unsigned int i;
4083     char glsl_name[8];
4084     struct ps_compile_args ps_compile_args;
4085     struct vs_compile_args vs_compile_args;
4086
4087     if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4088     if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4089
4090     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4091     if (entry) {
4092         priv->glsl_program = entry;
4093         return;
4094     }
4095
4096     /* If we get to this point, then no matching program exists, so we create one */
4097     programId = GL_EXTCALL(glCreateProgramObjectARB());
4098     TRACE("Created new GLSL shader program %u\n", programId);
4099
4100     /* Create the entry */
4101     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4102     entry->programId = programId;
4103     entry->vshader = vshader;
4104     entry->pshader = pshader;
4105     entry->vs_args = vs_compile_args;
4106     entry->ps_args = ps_compile_args;
4107     entry->constant_version = 0;
4108     entry->np2Fixup_info = NULL;
4109     /* Add the hash table entry */
4110     add_glsl_program_entry(priv, entry);
4111
4112     /* Set the current program */
4113     priv->glsl_program = entry;
4114
4115     /* Attach GLSL vshader */
4116     if (vshader)
4117     {
4118         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4119                 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4120         WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4121         char tmp_name[10];
4122
4123         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4124         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4125         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4126         checkGLcall("glAttachObjectARB");
4127         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4128          * is destroyed
4129          */
4130         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4131
4132         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4133         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4134         checkGLcall("glAttachObjectARB");
4135
4136         /* Bind vertex attributes to a corresponding index number to match
4137          * the same index numbers as ARB_vertex_programs (makes loading
4138          * vertex attributes simpler).  With this method, we can use the
4139          * exact same code to load the attributes later for both ARB and
4140          * GLSL shaders.
4141          *
4142          * We have to do this here because we need to know the Program ID
4143          * in order to make the bindings work, and it has to be done prior
4144          * to linking the GLSL program. */
4145         for (i = 0; map; map >>= 1, ++i)
4146         {
4147             if (!(map & 1)) continue;
4148
4149             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4150             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4151         }
4152         checkGLcall("glBindAttribLocationARB");
4153
4154         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4155     }
4156
4157     /* Attach GLSL pshader */
4158     if (pshader)
4159     {
4160         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4161                 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4162         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4163         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4164         checkGLcall("glAttachObjectARB");
4165
4166         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4167     }
4168
4169     /* Link the program */
4170     TRACE("Linking GLSL shader program %u\n", programId);
4171     GL_EXTCALL(glLinkProgramARB(programId));
4172     print_glsl_info_log(gl_info, programId);
4173
4174     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4175             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4176     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4177     {
4178         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4179         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4180     }
4181     for (i = 0; i < MAX_CONST_I; ++i)
4182     {
4183         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4184         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4185     }
4186     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4187             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4188     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4189     {
4190         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4191         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4192     }
4193     for (i = 0; i < MAX_CONST_I; ++i)
4194     {
4195         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4196         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4197     }
4198
4199     if(pshader) {
4200         char name[32];
4201
4202         for(i = 0; i < MAX_TEXTURES; i++) {
4203             sprintf(name, "bumpenvmat%u", i);
4204             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4205             sprintf(name, "luminancescale%u", i);
4206             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4207             sprintf(name, "luminanceoffset%u", i);
4208             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4209         }
4210
4211         if (ps_compile_args.np2_fixup) {
4212             if (entry->np2Fixup_info) {
4213                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4214             } else {
4215                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4216             }
4217         }
4218     }
4219
4220     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4221     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4222     checkGLcall("Find glsl program uniform locations");
4223
4224     if (pshader
4225             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4226             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4227     {
4228         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4229         entry->vertex_color_clamp = GL_FALSE;
4230     } else {
4231         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4232     }
4233
4234     /* Set the shader to allow uniform loading on it */
4235     GL_EXTCALL(glUseProgramObjectARB(programId));
4236     checkGLcall("glUseProgramObjectARB(programId)");
4237
4238     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4239      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4240      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4241      * vertex shader with fixed function pixel processing is used we make sure that the card
4242      * supports enough samplers to allow the max number of vertex samplers with all possible
4243      * fixed function fragment processing setups. So once the program is linked these samplers
4244      * won't change.
4245      */
4246     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4247     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4248
4249     /* If the local constants do not have to be loaded with the environment constants,
4250      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4251      * later
4252      */
4253     if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
4254         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4255     }
4256     if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
4257         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4258     }
4259 }
4260
4261 /* GL locking is done by the caller */
4262 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
4263 {
4264     GLhandleARB program_id;
4265     GLhandleARB vshader_id, pshader_id;
4266     static const char *blt_vshader[] =
4267     {
4268         "#version 120\n"
4269         "void main(void)\n"
4270         "{\n"
4271         "    gl_Position = gl_Vertex;\n"
4272         "    gl_FrontColor = vec4(1.0);\n"
4273         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4274         "}\n"
4275     };
4276
4277     static const char *blt_pshaders[tex_type_count] =
4278     {
4279         /* tex_1d */
4280         NULL,
4281         /* tex_2d */
4282         "#version 120\n"
4283         "uniform sampler2D sampler;\n"
4284         "void main(void)\n"
4285         "{\n"
4286         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4287         "}\n",
4288         /* tex_3d */
4289         NULL,
4290         /* tex_cube */
4291         "#version 120\n"
4292         "uniform samplerCube sampler;\n"
4293         "void main(void)\n"
4294         "{\n"
4295         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4296         "}\n",
4297         /* tex_rect */
4298         "#version 120\n"
4299         "#extension GL_ARB_texture_rectangle : enable\n"
4300         "uniform sampler2DRect sampler;\n"
4301         "void main(void)\n"
4302         "{\n"
4303         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4304         "}\n",
4305     };
4306
4307     if (!blt_pshaders[tex_type])
4308     {
4309         FIXME("tex_type %#x not supported\n", tex_type);
4310         tex_type = tex_2d;
4311     }
4312
4313     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4314     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
4315     GL_EXTCALL(glCompileShaderARB(vshader_id));
4316
4317     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4318     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
4319     GL_EXTCALL(glCompileShaderARB(pshader_id));
4320
4321     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4322     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4323     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4324     GL_EXTCALL(glLinkProgramARB(program_id));
4325
4326     print_glsl_info_log(gl_info, program_id);
4327
4328     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4329      * is destroyed
4330      */
4331     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4332     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4333     return program_id;
4334 }
4335
4336 /* GL locking is done by the caller */
4337 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4338 {
4339     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.device;
4340     const struct wined3d_gl_info *gl_info = context->gl_info;
4341     struct shader_glsl_priv *priv = device->shader_priv;
4342     GLhandleARB program_id = 0;
4343     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4344
4345     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4346
4347     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4348     else priv->glsl_program = NULL;
4349
4350     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4351
4352     if (old_vertex_color_clamp != current_vertex_color_clamp)
4353     {
4354         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4355         {
4356             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4357             checkGLcall("glClampColorARB");
4358         }
4359         else
4360         {
4361             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4362         }
4363     }
4364
4365     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4366     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4367     GL_EXTCALL(glUseProgramObjectARB(program_id));
4368     checkGLcall("glUseProgramObjectARB");
4369
4370     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4371      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4372      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4373     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4374     {
4375         shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4376     }
4377 }
4378
4379 /* GL locking is done by the caller */
4380 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4381     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4382     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4383     struct shader_glsl_priv *priv = This->shader_priv;
4384     GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
4385
4386     if (!*blt_program) {
4387         GLint loc;
4388         *blt_program = create_glsl_blt_shader(gl_info, tex_type);
4389         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4390         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4391         GL_EXTCALL(glUniform1iARB(loc, 0));
4392     } else {
4393         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4394     }
4395 }
4396
4397 /* GL locking is done by the caller */
4398 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4399     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4400     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4401     struct shader_glsl_priv *priv = This->shader_priv;
4402     GLhandleARB program_id;
4403
4404     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4405     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4406
4407     GL_EXTCALL(glUseProgramObjectARB(program_id));
4408     checkGLcall("glUseProgramObjectARB");
4409 }
4410
4411 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4412     const struct list *linked_programs;
4413     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4414     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4415     struct shader_glsl_priv *priv = device->shader_priv;
4416     const struct wined3d_gl_info *gl_info;
4417     IWineD3DPixelShaderImpl *ps = NULL;
4418     IWineD3DVertexShaderImpl *vs = NULL;
4419     struct wined3d_context *context;
4420
4421     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4422      * can be called from IWineD3DBaseShader::Release
4423      */
4424     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4425
4426     if(pshader) {
4427         struct glsl_pshader_private *shader_data;
4428         ps = (IWineD3DPixelShaderImpl *) This;
4429         shader_data = ps->baseShader.backend_data;
4430         if(!shader_data || shader_data->num_gl_shaders == 0)
4431         {
4432             HeapFree(GetProcessHeap(), 0, shader_data);
4433             ps->baseShader.backend_data = NULL;
4434             return;
4435         }
4436
4437         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4438         gl_info = context->gl_info;
4439
4440         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4441         {
4442             ENTER_GL();
4443             shader_glsl_select(context, FALSE, FALSE);
4444             LEAVE_GL();
4445         }
4446     } else {
4447         struct glsl_vshader_private *shader_data;
4448         vs = (IWineD3DVertexShaderImpl *) This;
4449         shader_data = vs->baseShader.backend_data;
4450         if(!shader_data || shader_data->num_gl_shaders == 0)
4451         {
4452             HeapFree(GetProcessHeap(), 0, shader_data);
4453             vs->baseShader.backend_data = NULL;
4454             return;
4455         }
4456
4457         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4458         gl_info = context->gl_info;
4459
4460         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4461         {
4462             ENTER_GL();
4463             shader_glsl_select(context, FALSE, FALSE);
4464             LEAVE_GL();
4465         }
4466     }
4467
4468     linked_programs = &This->baseShader.linked_programs;
4469
4470     TRACE("Deleting linked programs\n");
4471     if (linked_programs->next) {
4472         struct glsl_shader_prog_link *entry, *entry2;
4473
4474         ENTER_GL();
4475         if(pshader) {
4476             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4477                 delete_glsl_program_entry(priv, gl_info, entry);
4478             }
4479         } else {
4480             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4481                 delete_glsl_program_entry(priv, gl_info, entry);
4482             }
4483         }
4484         LEAVE_GL();
4485     }
4486
4487     if(pshader) {
4488         UINT i;
4489         struct glsl_pshader_private *shader_data = ps->baseShader.backend_data;
4490
4491         ENTER_GL();
4492         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4493             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4494             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4495             checkGLcall("glDeleteObjectARB");
4496         }
4497         LEAVE_GL();
4498         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4499         HeapFree(GetProcessHeap(), 0, shader_data);
4500         ps->baseShader.backend_data = NULL;
4501     } else {
4502         UINT i;
4503         struct glsl_vshader_private *shader_data = vs->baseShader.backend_data;
4504
4505         ENTER_GL();
4506         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4507             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4508             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4509             checkGLcall("glDeleteObjectARB");
4510         }
4511         LEAVE_GL();
4512         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4513         HeapFree(GetProcessHeap(), 0, shader_data);
4514         vs->baseShader.backend_data = NULL;
4515     }
4516
4517     context_release(context);
4518 }
4519
4520 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4521 {
4522     const glsl_program_key_t *k = key;
4523     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4524             const struct glsl_shader_prog_link, program_lookup_entry);
4525     int cmp;
4526
4527     if (k->vshader > prog->vshader) return 1;
4528     else if (k->vshader < prog->vshader) return -1;
4529
4530     if (k->pshader > prog->pshader) return 1;
4531     else if (k->pshader < prog->pshader) return -1;
4532
4533     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4534     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4535
4536     return 0;
4537 }
4538
4539 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4540 {
4541     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4542     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4543
4544     if (!mem)
4545     {
4546         ERR("Failed to allocate memory\n");
4547         return FALSE;
4548     }
4549
4550     heap->entries = mem;
4551     heap->entries[1].version = 0;
4552     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4553     heap->size = 1;
4554
4555     return TRUE;
4556 }
4557
4558 static void constant_heap_free(struct constant_heap *heap)
4559 {
4560     HeapFree(GetProcessHeap(), 0, heap->entries);
4561 }
4562
4563 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4564 {
4565     wined3d_rb_alloc,
4566     wined3d_rb_realloc,
4567     wined3d_rb_free,
4568     glsl_program_key_compare,
4569 };
4570
4571 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4572     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4573     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4574     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4575     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4576             gl_info->limits.glsl_ps_float_constants)) + 1;
4577
4578     if (!shader_buffer_init(&priv->shader_buffer))
4579     {
4580         ERR("Failed to initialize shader buffer.\n");
4581         goto fail;
4582     }
4583
4584     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4585     if (!priv->stack)
4586     {
4587         ERR("Failed to allocate memory.\n");
4588         goto fail;
4589     }
4590
4591     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4592     {
4593         ERR("Failed to initialize vertex shader constant heap\n");
4594         goto fail;
4595     }
4596
4597     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4598     {
4599         ERR("Failed to initialize pixel shader constant heap\n");
4600         goto fail;
4601     }
4602
4603     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4604     {
4605         ERR("Failed to initialize rbtree.\n");
4606         goto fail;
4607     }
4608
4609     priv->next_constant_version = 1;
4610
4611     This->shader_priv = priv;
4612     return WINED3D_OK;
4613
4614 fail:
4615     constant_heap_free(&priv->pconst_heap);
4616     constant_heap_free(&priv->vconst_heap);
4617     HeapFree(GetProcessHeap(), 0, priv->stack);
4618     shader_buffer_free(&priv->shader_buffer);
4619     HeapFree(GetProcessHeap(), 0, priv);
4620     return E_OUTOFMEMORY;
4621 }
4622
4623 /* Context activation is done by the caller. */
4624 static void shader_glsl_free(IWineD3DDevice *iface) {
4625     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4626     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4627     struct shader_glsl_priv *priv = This->shader_priv;
4628     int i;
4629
4630     ENTER_GL();
4631     for (i = 0; i < tex_type_count; ++i)
4632     {
4633         if (priv->depth_blt_program[i])
4634         {
4635             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
4636         }
4637     }
4638     LEAVE_GL();
4639
4640     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4641     constant_heap_free(&priv->pconst_heap);
4642     constant_heap_free(&priv->vconst_heap);
4643     HeapFree(GetProcessHeap(), 0, priv->stack);
4644     shader_buffer_free(&priv->shader_buffer);
4645
4646     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4647     This->shader_priv = NULL;
4648 }
4649
4650 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4651     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4652     return FALSE;
4653 }
4654
4655 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype,
4656         const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4657 {
4658     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4659      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4660      * on the version of NV_vertex_program.
4661      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4662      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4663      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4664      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4665      */
4666     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4667             || gl_info->limits.arb_ps_instructions <= 512)
4668         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4669     else
4670         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4671     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4672     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4673
4674     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4675      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4676      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4677      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4678      * in max native instructions. Intel and others also offer the info in this extension but they
4679      * don't support GLSL (at least on Windows).
4680      *
4681      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4682      * of instructions is 512 or less we have to do with ps2.0 hardware.
4683      * 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.
4684      */
4685     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4686             || gl_info->limits.arb_ps_instructions <= 512)
4687         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4688     else
4689         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4690
4691     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4692
4693     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4694      * Direct3D minimum requirement.
4695      *
4696      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4697      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4698      *
4699      * The problem is that the refrast clamps temporary results in the shader to
4700      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4701      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4702      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4703      * offer a way to query this.
4704      */
4705     pCaps->PixelShader1xMaxValue = 8.0;
4706     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4707
4708     pCaps->VSClipping = TRUE;
4709 }
4710
4711 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4712 {
4713     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4714     {
4715         TRACE("Checking support for fixup:\n");
4716         dump_color_fixup_desc(fixup);
4717     }
4718
4719     /* We support everything except YUV conversions. */
4720     if (!is_yuv_fixup(fixup))
4721     {
4722         TRACE("[OK]\n");
4723         return TRUE;
4724     }
4725
4726     TRACE("[FAILED]\n");
4727     return FALSE;
4728 }
4729
4730 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4731 {
4732     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4733     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4734     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4735     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4736     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4737     /* WINED3DSIH_BREAKP        */ NULL,
4738     /* WINED3DSIH_CALL          */ shader_glsl_call,
4739     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4740     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4741     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4742     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4743     /* WINED3DSIH_CUT           */ NULL,
4744     /* WINED3DSIH_DCL           */ NULL,
4745     /* WINED3DSIH_DEF           */ NULL,
4746     /* WINED3DSIH_DEFB          */ NULL,
4747     /* WINED3DSIH_DEFI          */ NULL,
4748     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
4749     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4750     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4751     /* WINED3DSIH_DST           */ shader_glsl_dst,
4752     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
4753     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
4754     /* WINED3DSIH_ELSE          */ shader_glsl_else,
4755     /* WINED3DSIH_EMIT          */ NULL,
4756     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
4757     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
4758     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
4759     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
4760     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
4761     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
4762     /* WINED3DSIH_IADD          */ NULL,
4763     /* WINED3DSIH_IF            */ shader_glsl_if,
4764     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
4765     /* WINED3DSIH_IGE           */ NULL,
4766     /* WINED3DSIH_LABEL         */ shader_glsl_label,
4767     /* WINED3DSIH_LIT           */ shader_glsl_lit,
4768     /* WINED3DSIH_LOG           */ shader_glsl_log,
4769     /* WINED3DSIH_LOGP          */ shader_glsl_log,
4770     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
4771     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
4772     /* WINED3DSIH_LT            */ NULL,
4773     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
4774     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
4775     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
4776     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
4777     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
4778     /* WINED3DSIH_MAD           */ shader_glsl_mad,
4779     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
4780     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
4781     /* WINED3DSIH_MOV           */ shader_glsl_mov,
4782     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
4783     /* WINED3DSIH_MUL           */ shader_glsl_arith,
4784     /* WINED3DSIH_NOP           */ NULL,
4785     /* WINED3DSIH_NRM           */ shader_glsl_map2gl,
4786     /* WINED3DSIH_PHASE         */ NULL,
4787     /* WINED3DSIH_POW           */ shader_glsl_pow,
4788     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
4789     /* WINED3DSIH_REP           */ shader_glsl_rep,
4790     /* WINED3DSIH_RET           */ shader_glsl_ret,
4791     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
4792     /* WINED3DSIH_SETP          */ NULL,
4793     /* WINED3DSIH_SGE           */ shader_glsl_compare,
4794     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
4795     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
4796     /* WINED3DSIH_SLT           */ shader_glsl_compare,
4797     /* WINED3DSIH_SUB           */ shader_glsl_arith,
4798     /* WINED3DSIH_TEX           */ shader_glsl_tex,
4799     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
4800     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
4801     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
4802     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
4803     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
4804     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
4805     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
4806     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
4807     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
4808     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
4809     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
4810     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
4811     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
4812     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4813     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
4814     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
4815     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
4816     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
4817     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
4818     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
4819     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
4820 };
4821
4822 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
4823     SHADER_HANDLER hw_fct;
4824
4825     /* Select handler */
4826     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
4827
4828     /* Unhandled opcode */
4829     if (!hw_fct)
4830     {
4831         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
4832         return;
4833     }
4834     hw_fct(ins);
4835
4836     shader_glsl_add_instruction_modifiers(ins);
4837 }
4838
4839 const shader_backend_t glsl_shader_backend = {
4840     shader_glsl_handle_instruction,
4841     shader_glsl_select,
4842     shader_glsl_select_depth_blt,
4843     shader_glsl_deselect_depth_blt,
4844     shader_glsl_update_float_vertex_constants,
4845     shader_glsl_update_float_pixel_constants,
4846     shader_glsl_load_constants,
4847     shader_glsl_load_np2fixup_constants,
4848     shader_glsl_destroy,
4849     shader_glsl_alloc,
4850     shader_glsl_free,
4851     shader_glsl_dirty_const,
4852     shader_glsl_get_caps,
4853     shader_glsl_color_fixup_supported,
4854 };