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 2009-2011 Henri Verbeet for CodeWeavers
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  22  */
  23
  24 /*
  25  * D3D shader asm has swizzles on source parameters, and write masks for
  26  * destination parameters. GLSL uses swizzles for both. The result of this is
  27  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  28  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  29  * mask for the destination parameter into account.
  30  */
  31
  32 #include "config.h"
  33 #include <limits.h>
  34 #include <stdio.h>
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  43 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  44 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  45 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  46
  47 struct glsl_dst_param
  48 {
  49     char reg_name[150];
  50     char mask_str[6];
  51 };
  52
  53 struct glsl_src_param
  54 {
  55     char reg_name[150];
  56     char param_str[200];
  57 };
  58
  59 struct glsl_sample_function
  60 {
  61     const char *name;
  62     DWORD coord_mask;
  63 };
  64
  65 enum heap_node_op
  66 {
  67     HEAP_NODE_TRAVERSE_LEFT,
  68     HEAP_NODE_TRAVERSE_RIGHT,
  69     HEAP_NODE_POP,
  70 };
  71
  72 struct constant_entry
  73 {
  74     unsigned int idx;
  75     unsigned int version;
  76 };
  77
  78 struct constant_heap
  79 {
  80     struct constant_entry *entries;
  81     unsigned int *positions;
  82     unsigned int size;
  83 };
  84
  85 /* GLSL shader private data */
  86 struct shader_glsl_priv {
  87     struct wined3d_shader_buffer shader_buffer;
  88     struct wine_rb_tree program_lookup;
  89     struct glsl_shader_prog_link *glsl_program;
  90     struct constant_heap vconst_heap;
  91     struct constant_heap pconst_heap;
  92     unsigned char *stack;
  93     GLhandleARB depth_blt_program_full[tex_type_count];
  94     GLhandleARB depth_blt_program_masked[tex_type_count];
  95     UINT next_constant_version;
  96 };
  97
  98 /* Struct to maintain data about a linked GLSL program */
  99 struct glsl_shader_prog_link {
 100     struct wine_rb_entry        program_lookup_entry;
 101     struct list                 vshader_entry;
 102     struct list                 pshader_entry;
 103     GLhandleARB                 programId;
 104     GLint                       *vuniformF_locations;
 105     GLint                       *puniformF_locations;
 106     GLint                       vuniformI_locations[MAX_CONST_I];
 107     GLint                       puniformI_locations[MAX_CONST_I];
 108     GLint                       posFixup_location;
 109     GLint                       np2Fixup_location;
 110     GLint                       bumpenvmat_location[MAX_TEXTURES];
 111     GLint                       luminancescale_location[MAX_TEXTURES];
 112     GLint                       luminanceoffset_location[MAX_TEXTURES];
 113     GLint                       ycorrection_location;
 114     GLenum                      vertex_color_clamp;
 115     const struct wined3d_shader *vshader;
 116     const struct wined3d_shader *pshader;
 117     struct vs_compile_args      vs_args;
 118     struct ps_compile_args      ps_args;
 119     UINT                        constant_version;
 120     const struct ps_np2fixup_info *np2Fixup_info;
 121 };
 122
 123 struct glsl_program_key
 124 {
 125     const struct wined3d_shader *vshader;
 126     const struct wined3d_shader *pshader;
 127     struct ps_compile_args      ps_args;
 128     struct vs_compile_args      vs_args;
 129 };
 130
 131 struct shader_glsl_ctx_priv {
 132     const struct vs_compile_args    *cur_vs_args;
 133     const struct ps_compile_args    *cur_ps_args;
 134     struct ps_np2fixup_info         *cur_np2fixup_info;
 135 };
 136
 137 struct glsl_ps_compiled_shader
 138 {
 139     struct ps_compile_args          args;
 140     struct ps_np2fixup_info         np2fixup;
 141     GLhandleARB                     prgId;
 142 };
 143
 144 struct glsl_pshader_private
 145 {
 146     struct glsl_ps_compiled_shader  *gl_shaders;
 147     UINT                            num_gl_shaders, shader_array_size;
 148 };
 149
 150 struct glsl_vs_compiled_shader
 151 {
 152     struct vs_compile_args          args;
 153     GLhandleARB                     prgId;
 154 };
 155
 156 struct glsl_vshader_private
 157 {
 158     struct glsl_vs_compiled_shader  *gl_shaders;
 159     UINT                            num_gl_shaders, shader_array_size;
 160 };
 161
 162 static const char *debug_gl_shader_type(GLenum type)
 163 {
 164     switch (type)
 165     {
 166 #define WINED3D_TO_STR(u) case u: return #u
 167         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
 168         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
 169         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
 170 #undef WINED3D_TO_STR
 171         default:
 172             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
 173     }
 174 }
 175
 176 /* Extract a line from the info log.
 177  * Note that this modifies the source string. */
 178 static char *get_info_log_line(char **ptr)
 179 {
 180     char *p, *q;
 181
 182     p = *ptr;
 183     if (!(q = strstr(p, "\n")))
 184     {
 185         if (!*p) return NULL;
 186         *ptr += strlen(p);
 187         return p;
 188     }
 189     *q = '\0';
 190     *ptr = q + 1;
 191
 192     return p;
 193 }
 194
 195 /** Prints the GLSL info log which will contain error messages if they exist */
 196 /* GL locking is done by the caller */
 197 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 198 {
 199     int infologLength = 0;
 200     char *infoLog;
 201
 202     if (!WARN_ON(d3d_shader) && !FIXME_ON(d3d_shader))
 203         return;
 204
 205     GL_EXTCALL(glGetObjectParameterivARB(obj,
 206                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 207                &infologLength));
 208
 209     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 210      * that if there are errors. */
 211     if (infologLength > 1)
 212     {
 213         char *ptr, *line;
 214
 215         infoLog = HeapAlloc(GetProcessHeap(), 0, infologLength);
 216         /* The info log is supposed to be zero-terminated, but at least some
 217          * versions of fglrx don't terminate the string properly. The reported
 218          * length does include the terminator, so explicitly set it to zero
 219          * here. */
 220         infoLog[infologLength - 1] = 0;
 221         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 222
 223         ptr = infoLog;
 224         if (gl_info->quirks & WINED3D_QUIRK_INFO_LOG_SPAM)
 225         {
 226             WARN("Info log received from GLSL shader #%u:\n", obj);
 227             while ((line = get_info_log_line(&ptr))) WARN("    %s\n", line);
 228         }
 229         else
 230         {
 231             FIXME("Info log received from GLSL shader #%u:\n", obj);
 232             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 233         }
 234         HeapFree(GetProcessHeap(), 0, infoLog);
 235     }
 236 }
 237
 238 /* GL locking is done by the caller. */
 239 static void shader_glsl_compile(const struct wined3d_gl_info *gl_info, GLhandleARB shader, const char *src)
 240 {
 241     TRACE("Compiling shader object %u.\n", shader);
 242     GL_EXTCALL(glShaderSourceARB(shader, 1, &src, NULL));
 243     checkGLcall("glShaderSourceARB");
 244     GL_EXTCALL(glCompileShaderARB(shader));
 245     checkGLcall("glCompileShaderARB");
 246     print_glsl_info_log(gl_info, shader);
 247 }
 248
 249 /* GL locking is done by the caller. */
 250 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 251 {
 252     GLint i, object_count, source_size = -1;
 253     GLhandleARB *objects;
 254     char *source = NULL;
 255
 256     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
 257     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
 258     if (!objects)
 259     {
 260         ERR("Failed to allocate object array memory.\n");
 261         return;
 262     }
 263
 264     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
 265     for (i = 0; i < object_count; ++i)
 266     {
 267         char *ptr, *line;
 268         GLint tmp;
 269
 270         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
 271
 272         if (source_size < tmp)
 273         {
 274             HeapFree(GetProcessHeap(), 0, source);
 275
 276             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
 277             if (!source)
 278             {
 279                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
 280                 HeapFree(GetProcessHeap(), 0, objects);
 281                 return;
 282             }
 283             source_size = tmp;
 284         }
 285
 286         FIXME("Object %u:\n", objects[i]);
 287         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
 288         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
 289         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
 290         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
 291         FIXME("\n");
 292
 293         ptr = source;
 294         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
 295         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 296         FIXME("\n");
 297     }
 298
 299     HeapFree(GetProcessHeap(), 0, source);
 300     HeapFree(GetProcessHeap(), 0, objects);
 301 }
 302
 303 /* GL locking is done by the caller. */
 304 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 305 {
 306     GLint tmp;
 307
 308     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 309
 310     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
 311     if (tmp == GL_PROGRAM_OBJECT_ARB)
 312     {
 313         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
 314         if (!tmp)
 315         {
 316             FIXME("Program %u link status invalid.\n", program);
 317             shader_glsl_dump_program_source(gl_info, program);
 318         }
 319     }
 320
 321     print_glsl_info_log(gl_info, program);
 322 }
 323
 324 /**
 325  * Loads (pixel shader) samplers
 326  */
 327 /* GL locking is done by the caller */
 328 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 329         const DWORD *tex_unit_map, GLhandleARB programId)
 330 {
 331     GLint name_loc;
 332     int i;
 333     char sampler_name[20];
 334
 335     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 336         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 337         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 338         if (name_loc != -1) {
 339             DWORD mapped_unit = tex_unit_map[i];
 340             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 341             {
 342                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 343                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 344                 checkGLcall("glUniform1iARB");
 345             } else {
 346                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 347             }
 348         }
 349     }
 350 }
 351
 352 /* GL locking is done by the caller */
 353 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 354         const DWORD *tex_unit_map, GLhandleARB programId)
 355 {
 356     GLint name_loc;
 357     char sampler_name[20];
 358     int i;
 359
 360     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 361         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 362         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 363         if (name_loc != -1) {
 364             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 365             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 366             {
 367                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 368                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 369                 checkGLcall("glUniform1iARB");
 370             } else {
 371                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 372             }
 373         }
 374     }
 375 }
 376
 377 /* GL locking is done by the caller */
 378 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 379         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 380 {
 381     int stack_idx = 0;
 382     unsigned int heap_idx = 1;
 383     unsigned int idx;
 384
 385     if (heap->entries[heap_idx].version <= version) return;
 386
 387     idx = heap->entries[heap_idx].idx;
 388     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 389     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 390
 391     while (stack_idx >= 0)
 392     {
 393         /* Note that we fall through to the next case statement. */
 394         switch(stack[stack_idx])
 395         {
 396             case HEAP_NODE_TRAVERSE_LEFT:
 397             {
 398                 unsigned int left_idx = heap_idx << 1;
 399                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 400                 {
 401                     heap_idx = left_idx;
 402                     idx = heap->entries[heap_idx].idx;
 403                     if (constant_locations[idx] != -1)
 404                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 405
 406                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 407                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 408                     break;
 409                 }
 410             }
 411
 412             case HEAP_NODE_TRAVERSE_RIGHT:
 413             {
 414                 unsigned int right_idx = (heap_idx << 1) + 1;
 415                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 416                 {
 417                     heap_idx = right_idx;
 418                     idx = heap->entries[heap_idx].idx;
 419                     if (constant_locations[idx] != -1)
 420                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 421
 422                     stack[stack_idx++] = HEAP_NODE_POP;
 423                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 424                     break;
 425                 }
 426             }
 427
 428             case HEAP_NODE_POP:
 429                 heap_idx >>= 1;
 430                 --stack_idx;
 431                 break;
 432         }
 433     }
 434     checkGLcall("walk_constant_heap()");
 435 }
 436
 437 /* GL locking is done by the caller */
 438 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 439 {
 440     GLfloat clamped_constant[4];
 441
 442     if (location == -1) return;
 443
 444     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 445     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 446     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 447     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 448
 449     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 450 }
 451
 452 /* GL locking is done by the caller */
 453 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 454         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 455 {
 456     int stack_idx = 0;
 457     unsigned int heap_idx = 1;
 458     unsigned int idx;
 459
 460     if (heap->entries[heap_idx].version <= version) return;
 461
 462     idx = heap->entries[heap_idx].idx;
 463     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 464     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 465
 466     while (stack_idx >= 0)
 467     {
 468         /* Note that we fall through to the next case statement. */
 469         switch(stack[stack_idx])
 470         {
 471             case HEAP_NODE_TRAVERSE_LEFT:
 472             {
 473                 unsigned int left_idx = heap_idx << 1;
 474                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 475                 {
 476                     heap_idx = left_idx;
 477                     idx = heap->entries[heap_idx].idx;
 478                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 479
 480                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 481                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 482                     break;
 483                 }
 484             }
 485
 486             case HEAP_NODE_TRAVERSE_RIGHT:
 487             {
 488                 unsigned int right_idx = (heap_idx << 1) + 1;
 489                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 490                 {
 491                     heap_idx = right_idx;
 492                     idx = heap->entries[heap_idx].idx;
 493                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 494
 495                     stack[stack_idx++] = HEAP_NODE_POP;
 496                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 497                     break;
 498                 }
 499             }
 500
 501             case HEAP_NODE_POP:
 502                 heap_idx >>= 1;
 503                 --stack_idx;
 504                 break;
 505         }
 506     }
 507     checkGLcall("walk_constant_heap_clamped()");
 508 }
 509
 510 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 511 /* GL locking is done by the caller */
 512 static void shader_glsl_load_constantsF(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 513         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 514         unsigned char *stack, UINT version)
 515 {
 516     const struct wined3d_shader_lconst *lconst;
 517
 518     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 519     if (shader->reg_maps.shader_version.major == 1
 520             && shader_is_pshader_version(shader->reg_maps.shader_version.type))
 521         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 522     else
 523         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 524
 525     if (!shader->load_local_constsF)
 526     {
 527         TRACE("No need to load local float constants for this shader\n");
 528         return;
 529     }
 530
 531     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 532     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 533     {
 534         GLint location = constant_locations[lconst->idx];
 535         /* We found this uniform name in the program - go ahead and send the data */
 536         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 537     }
 538     checkGLcall("glUniform4fvARB()");
 539 }
 540
 541 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 542 /* GL locking is done by the caller */
 543 static void shader_glsl_load_constantsI(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 544         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 545 {
 546     unsigned int i;
 547     struct list* ptr;
 548
 549     for (i = 0; constants_set; constants_set >>= 1, ++i)
 550     {
 551         if (!(constants_set & 1)) continue;
 552
 553         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 554                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 555
 556         /* We found this uniform name in the program - go ahead and send the data */
 557         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 558         checkGLcall("glUniform4ivARB");
 559     }
 560
 561     /* Load immediate constants */
 562     ptr = list_head(&shader->constantsI);
 563     while (ptr)
 564     {
 565         const struct wined3d_shader_lconst *lconst = LIST_ENTRY(ptr, const struct wined3d_shader_lconst, entry);
 566         unsigned int idx = lconst->idx;
 567         const GLint *values = (const GLint *)lconst->value;
 568
 569         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 570             values[0], values[1], values[2], values[3]);
 571
 572         /* We found this uniform name in the program - go ahead and send the data */
 573         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 574         checkGLcall("glUniform4ivARB");
 575         ptr = list_next(&shader->constantsI, ptr);
 576     }
 577 }
 578
 579 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 580 /* GL locking is done by the caller */
 581 static void shader_glsl_load_constantsB(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 582         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 583 {
 584     GLint tmp_loc;
 585     unsigned int i;
 586     char tmp_name[8];
 587     const char *prefix;
 588     struct list* ptr;
 589
 590     switch (shader->reg_maps.shader_version.type)
 591     {
 592         case WINED3D_SHADER_TYPE_VERTEX:
 593             prefix = "VB";
 594             break;
 595
 596         case WINED3D_SHADER_TYPE_GEOMETRY:
 597             prefix = "GB";
 598             break;
 599
 600         case WINED3D_SHADER_TYPE_PIXEL:
 601             prefix = "PB";
 602             break;
 603
 604         default:
 605             FIXME("Unknown shader type %#x.\n",
 606                     shader->reg_maps.shader_version.type);
 607             prefix = "UB";
 608             break;
 609     }
 610
 611     /* TODO: Benchmark and see if it would be beneficial to store the
 612      * locations of the constants to avoid looking up each time */
 613     for (i = 0; constants_set; constants_set >>= 1, ++i)
 614     {
 615         if (!(constants_set & 1)) continue;
 616
 617         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 618
 619         /* TODO: Benchmark and see if it would be beneficial to store the
 620          * locations of the constants to avoid looking up each time */
 621         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 622         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 623         if (tmp_loc != -1)
 624         {
 625             /* We found this uniform name in the program - go ahead and send the data */
 626             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 627             checkGLcall("glUniform1ivARB");
 628         }
 629     }
 630
 631     /* Load immediate constants */
 632     ptr = list_head(&shader->constantsB);
 633     while (ptr)
 634     {
 635         const struct wined3d_shader_lconst *lconst = LIST_ENTRY(ptr, const struct wined3d_shader_lconst, entry);
 636         unsigned int idx = lconst->idx;
 637         const GLint *values = (const GLint *)lconst->value;
 638
 639         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 640
 641         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 642         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 643         if (tmp_loc != -1) {
 644             /* We found this uniform name in the program - go ahead and send the data */
 645             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 646             checkGLcall("glUniform1ivARB");
 647         }
 648         ptr = list_next(&shader->constantsB, ptr);
 649     }
 650 }
 651
 652 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 653 {
 654     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 655 }
 656
 657 /**
 658  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 659  */
 660 /* GL locking is done by the caller (state handler) */
 661 static void shader_glsl_load_np2fixup_constants(void *shader_priv,
 662         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 663 {
 664     struct shader_glsl_priv *glsl_priv = shader_priv;
 665     const struct glsl_shader_prog_link *prog = glsl_priv->glsl_program;
 666
 667     /* No GLSL program set - nothing to do. */
 668     if (!prog) return;
 669
 670     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 671     if (!use_ps(state)) return;
 672
 673     if (prog->ps_args.np2_fixup && prog->np2Fixup_location != -1)
 674     {
 675         UINT i;
 676         UINT fixup = prog->ps_args.np2_fixup;
 677         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 678
 679         for (i = 0; fixup; fixup >>= 1, ++i)
 680         {
 681             const struct wined3d_texture *tex = state->textures[i];
 682             const unsigned char idx = prog->np2Fixup_info->idx[i];
 683             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 684
 685             if (!tex)
 686             {
 687                 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
 688                 continue;
 689             }
 690
 691             if (idx % 2)
 692             {
 693                 tex_dim[2] = tex->pow2_matrix[0];
 694                 tex_dim[3] = tex->pow2_matrix[5];
 695             }
 696             else
 697             {
 698                 tex_dim[0] = tex->pow2_matrix[0];
 699                 tex_dim[1] = tex->pow2_matrix[5];
 700             }
 701         }
 702
 703         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 704     }
 705 }
 706
 707 /**
 708  * Loads the app-supplied constants into the currently set GLSL program.
 709  */
 710 /* GL locking is done by the caller (state handler) */
 711 static void shader_glsl_load_constants(const struct wined3d_context *context,
 712         char usePixelShader, char useVertexShader)
 713 {
 714     const struct wined3d_gl_info *gl_info = context->gl_info;
 715     struct wined3d_device *device = context->swapchain->device;
 716     struct wined3d_stateblock *stateBlock = device->stateBlock;
 717     struct shader_glsl_priv *priv = device->shader_priv;
 718     float position_fixup[4];
 719
 720     GLhandleARB programId;
 721     struct glsl_shader_prog_link *prog = priv->glsl_program;
 722     UINT constant_version;
 723     int i;
 724
 725     if (!prog) {
 726         /* No GLSL program set - nothing to do. */
 727         return;
 728     }
 729     programId = prog->programId;
 730     constant_version = prog->constant_version;
 731
 732     if (useVertexShader)
 733     {
 734         const struct wined3d_shader *vshader = stateBlock->state.vertex_shader;
 735
 736         /* Load DirectX 9 float constants/uniforms for vertex shader */
 737         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->state.vs_consts_f,
 738                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 739
 740         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 741         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->state.vs_consts_i,
 742                 stateBlock->changed.vertexShaderConstantsI & vshader->reg_maps.integer_constants);
 743
 744         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 745         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->state.vs_consts_b,
 746                 stateBlock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants);
 747
 748         /* Upload the position fixup params */
 749         shader_get_position_fixup(context, &stateBlock->state, position_fixup);
 750         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, position_fixup));
 751         checkGLcall("glUniform4fvARB");
 752     }
 753
 754     if (usePixelShader)
 755     {
 756         const struct wined3d_shader *pshader = stateBlock->state.pixel_shader;
 757
 758         /* Load DirectX 9 float constants/uniforms for pixel shader */
 759         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->state.ps_consts_f,
 760                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 761
 762         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 763         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->state.ps_consts_i,
 764                 stateBlock->changed.pixelShaderConstantsI & pshader->reg_maps.integer_constants);
 765
 766         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 767         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->state.ps_consts_b,
 768                 stateBlock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants);
 769
 770         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 771          * It can't be 0 for a valid texbem instruction.
 772          */
 773         for(i = 0; i < MAX_TEXTURES; i++) {
 774             const float *data;
 775
 776             if(prog->bumpenvmat_location[i] == -1) continue;
 777
 778             data = (const float *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVMAT00];
 779             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 780             checkGLcall("glUniformMatrix2fvARB");
 781
 782             /* texbeml needs the luminance scale and offset too. If texbeml
 783              * is used, needsbumpmat is set too, so we can check that in the
 784              * needsbumpmat check. */
 785             if (prog->luminancescale_location[i] != -1)
 786             {
 787                 const GLfloat *scale = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLSCALE];
 788                 const GLfloat *offset = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
 789
 790                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 791                 checkGLcall("glUniform1fvARB");
 792                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 793                 checkGLcall("glUniform1fvARB");
 794             }
 795         }
 796
 797         if (prog->ycorrection_location != -1)
 798         {
 799             float correction_params[4];
 800
 801             if (context->render_offscreen)
 802             {
 803                 correction_params[0] = 0.0f;
 804                 correction_params[1] = 1.0f;
 805             } else {
 806                 /* position is window relative, not viewport relative */
 807                 correction_params[0] = (float) context->current_rt->resource.height;
 808                 correction_params[1] = -1.0f;
 809             }
 810             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 811         }
 812     }
 813
 814     if (priv->next_constant_version == UINT_MAX)
 815     {
 816         TRACE("Max constant version reached, resetting to 0.\n");
 817         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 818         priv->next_constant_version = 1;
 819     }
 820     else
 821     {
 822         prog->constant_version = priv->next_constant_version++;
 823     }
 824 }
 825
 826 static void update_heap_entry(const struct constant_heap *heap, unsigned int idx,
 827         unsigned int heap_idx, DWORD new_version)
 828 {
 829     struct constant_entry *entries = heap->entries;
 830     unsigned int *positions = heap->positions;
 831     unsigned int parent_idx;
 832
 833     while (heap_idx > 1)
 834     {
 835         parent_idx = heap_idx >> 1;
 836
 837         if (new_version <= entries[parent_idx].version) break;
 838
 839         entries[heap_idx] = entries[parent_idx];
 840         positions[entries[parent_idx].idx] = heap_idx;
 841         heap_idx = parent_idx;
 842     }
 843
 844     entries[heap_idx].version = new_version;
 845     entries[heap_idx].idx = idx;
 846     positions[idx] = heap_idx;
 847 }
 848
 849 static void shader_glsl_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 850 {
 851     struct shader_glsl_priv *priv = device->shader_priv;
 852     struct constant_heap *heap = &priv->vconst_heap;
 853     UINT i;
 854
 855     for (i = start; i < count + start; ++i)
 856     {
 857         if (!device->stateBlock->changed.vertexShaderConstantsF[i])
 858             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 859         else
 860             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 861     }
 862 }
 863
 864 static void shader_glsl_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 865 {
 866     struct shader_glsl_priv *priv = device->shader_priv;
 867     struct constant_heap *heap = &priv->pconst_heap;
 868     UINT i;
 869
 870     for (i = start; i < count + start; ++i)
 871     {
 872         if (!device->stateBlock->changed.pixelShaderConstantsF[i])
 873             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 874         else
 875             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 876     }
 877 }
 878
 879 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 880 {
 881     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 882     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 883     if(shader_major > 3) return ret;
 884
 885     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 886     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 887     return ret;
 888 }
 889
 890 /** Generate the variable & register declarations for the GLSL output target */
 891 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 892         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
 893         const struct wined3d_shader_reg_maps *reg_maps, const struct shader_glsl_ctx_priv *ctx_priv)
 894 {
 895     const struct wined3d_state *state = &shader->device->stateBlock->state;
 896     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 897     const struct wined3d_gl_info *gl_info = context->gl_info;
 898     const struct wined3d_fb_state *fb = &shader->device->fb;
 899     unsigned int i, extra_constants_needed = 0;
 900     const struct wined3d_shader_lconst *lconst;
 901     DWORD map;
 902
 903     /* There are some minor differences between pixel and vertex shaders */
 904     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 905     char prefix = pshader ? 'P' : 'V';
 906
 907     /* Prototype the subroutines */
 908     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 909     {
 910         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 911     }
 912
 913     /* Declare the constants (aka uniforms) */
 914     if (shader->limits.constant_float > 0)
 915     {
 916         unsigned max_constantsF;
 917         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 918          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 919          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 920          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 921          * a dx9 card, as long as it doesn't also use all the other constants.
 922          *
 923          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 924          * declare only the amount that we're assured to have.
 925          *
 926          * Thus we run into problems in these two cases:
 927          * 1) The shader really uses more uniforms than supported
 928          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 929          */
 930         if (pshader)
 931         {
 932             /* No indirect addressing here. */
 933             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 934         }
 935         else
 936         {
 937             if (reg_maps->usesrelconstF)
 938             {
 939                 /* Subtract the other potential uniforms from the max
 940                  * available (bools, ints, and 1 row of projection matrix).
 941                  * Subtract another uniform for immediate values, which have
 942                  * to be loaded via uniform by the driver as well. The shader
 943                  * code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex
 944                  * shader code, so one vec4 should be enough. (Unfortunately
 945                  * the Nvidia driver doesn't store 128 and -128 in one float).
 946                  *
 947                  * Writing gl_ClipVertex requires one uniform for each
 948                  * clipplane as well. */
 949                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 950                 if(ctx_priv->cur_vs_args->clip_enabled)
 951                 {
 952                     max_constantsF -= gl_info->limits.clipplanes;
 953                 }
 954                 max_constantsF -= count_bits(reg_maps->integer_constants);
 955                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 956                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 957                  * for now take this into account when calculating the number of available constants
 958                  */
 959                 max_constantsF -= count_bits(reg_maps->boolean_constants);
 960                 /* Set by driver quirks in directx.c */
 961                 max_constantsF -= gl_info->reserved_glsl_constants;
 962             }
 963             else
 964             {
 965                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 966             }
 967         }
 968         max_constantsF = min(shader->limits.constant_float, max_constantsF);
 969         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 970     }
 971
 972     /* Always declare the full set of constants, the compiler can remove the
 973      * unused ones because d3d doesn't (yet) support indirect int and bool
 974      * constant addressing. This avoids problems if the app uses e.g. i0 and i9. */
 975     if (shader->limits.constant_int > 0 && reg_maps->integer_constants)
 976         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, shader->limits.constant_int);
 977
 978     if (shader->limits.constant_bool > 0 && reg_maps->boolean_constants)
 979         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, shader->limits.constant_bool);
 980
 981     if (!pshader)
 982     {
 983         shader_addline(buffer, "uniform vec4 posFixup;\n");
 984         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 985     }
 986     else
 987     {
 988         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
 989         {
 990             if (!(map & 1)) continue;
 991
 992             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 993
 994             if (reg_maps->luminanceparams & (1 << i))
 995             {
 996                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
 997                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
 998                 extra_constants_needed++;
 999             }
1000
1001             extra_constants_needed++;
1002         }
1003
1004         if (ps_args->srgb_correction)
1005         {
1006             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1007                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1008             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1009                     srgb_cmp);
1010         }
1011         if (reg_maps->vpos || reg_maps->usesdsy)
1012         {
1013             if (shader->limits.constant_float + extra_constants_needed
1014                     + 1 < gl_info->limits.glsl_ps_float_constants)
1015             {
1016                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1017                 extra_constants_needed++;
1018             }
1019             else
1020             {
1021                 /* This happens because we do not have proper tracking of the constant registers that are
1022                  * actually used, only the max limit of the shader version
1023                  */
1024                 FIXME("Cannot find a free uniform for vpos correction params\n");
1025                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1026                         context->render_offscreen ? 0.0f : fb->render_targets[0]->resource.height,
1027                         context->render_offscreen ? 1.0f : -1.0f);
1028             }
1029             shader_addline(buffer, "vec4 vpos;\n");
1030         }
1031     }
1032
1033     /* Declare texture samplers */
1034     for (i = 0; i < shader->limits.sampler; ++i)
1035     {
1036         if (reg_maps->sampler_type[i])
1037         {
1038             const struct wined3d_texture *texture;
1039
1040             switch (reg_maps->sampler_type[i])
1041             {
1042                 case WINED3DSTT_1D:
1043                     if (pshader && ps_args->shadow & (1 << i))
1044                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1045                     else
1046                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1047                     break;
1048                 case WINED3DSTT_2D:
1049                     texture = state->textures[i];
1050                     if (pshader && ps_args->shadow & (1 << i))
1051                     {
1052                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1053                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1054                         else
1055                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1056                     }
1057                     else
1058                     {
1059                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1060                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1061                         else
1062                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1063                     }
1064                     break;
1065                 case WINED3DSTT_CUBE:
1066                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1067                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1068                     break;
1069                 case WINED3DSTT_VOLUME:
1070                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1071                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1072                     break;
1073                 default:
1074                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1075                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1076                     break;
1077             }
1078         }
1079     }
1080
1081     /* Declare uniforms for NP2 texcoord fixup:
1082      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1083      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1084      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1085     if (pshader && ps_args->np2_fixup) {
1086
1087         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1088         UINT cur = 0;
1089
1090         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1091          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1092          * samplerNP2Fixup stores texture dimensions and is updated through
1093          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1094
1095         for (i = 0; i < shader->limits.sampler; ++i)
1096         {
1097             if (reg_maps->sampler_type[i])
1098             {
1099                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1100
1101                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1102                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1103                     continue;
1104                 }
1105
1106                 fixup->idx[i] = cur++;
1107             }
1108         }
1109
1110         fixup->num_consts = (cur + 1) >> 1;
1111         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1112     }
1113
1114     /* Declare address variables */
1115     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1116     {
1117         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1118     }
1119
1120     /* Declare texture coordinate temporaries and initialize them */
1121     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1122     {
1123         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1124     }
1125
1126     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1127      * helper function shader that is linked in at link time
1128      */
1129     if (pshader && reg_maps->shader_version.major >= 3)
1130     {
1131         if (use_vs(state))
1132         {
1133             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1134         } else {
1135             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1136              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1137              * pixel shader that reads the fixed function color into the packed input registers.
1138              */
1139             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1140         }
1141     }
1142
1143     /* Declare output register temporaries */
1144     if (shader->limits.packed_output)
1145         shader_addline(buffer, "vec4 OUT[%u];\n", shader->limits.packed_output);
1146
1147     /* Declare temporary variables */
1148     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1149     {
1150         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1151     }
1152
1153     /* Declare attributes */
1154     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1155     {
1156         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1157         {
1158             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1159         }
1160     }
1161
1162     /* Declare loop registers aLx */
1163     for (i = 0; i < reg_maps->loop_depth; i++) {
1164         shader_addline(buffer, "int aL%u;\n", i);
1165         shader_addline(buffer, "int tmpInt%u;\n", i);
1166     }
1167
1168     /* Temporary variables for matrix operations */
1169     shader_addline(buffer, "vec4 tmp0;\n");
1170     shader_addline(buffer, "vec4 tmp1;\n");
1171
1172     /* Local constants use a different name so they can be loaded once at shader link time
1173      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1174      * float -> string conversion can cause precision loss.
1175      */
1176     if (!shader->load_local_constsF)
1177     {
1178         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
1179         {
1180             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1181         }
1182     }
1183
1184     /* Start the main program */
1185     shader_addline(buffer, "void main() {\n");
1186     if(pshader && reg_maps->vpos) {
1187         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1188          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1189          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1190          * precision troubles when we just subtract 0.5.
1191          *
1192          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1193          *
1194          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1195          *
1196          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1197          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1198          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1199          * correctly on drivers that returns integer values.
1200          */
1201         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1202     }
1203 }
1204
1205 /*****************************************************************************
1206  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1207  *
1208  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1209  ****************************************************************************/
1210
1211 /* Prototypes */
1212 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1213         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src);
1214
1215 /** Used for opcode modifiers - They multiply the result by the specified amount */
1216 static const char * const shift_glsl_tab[] = {
1217     "",           /*  0 (none) */
1218     "2.0 * ",     /*  1 (x2)   */
1219     "4.0 * ",     /*  2 (x4)   */
1220     "8.0 * ",     /*  3 (x8)   */
1221     "16.0 * ",    /*  4 (x16)  */
1222     "32.0 * ",    /*  5 (x32)  */
1223     "",           /*  6 (x64)  */
1224     "",           /*  7 (x128) */
1225     "",           /*  8 (d256) */
1226     "",           /*  9 (d128) */
1227     "",           /* 10 (d64)  */
1228     "",           /* 11 (d32)  */
1229     "0.0625 * ",  /* 12 (d16)  */
1230     "0.125 * ",   /* 13 (d8)   */
1231     "0.25 * ",    /* 14 (d4)   */
1232     "0.5 * "      /* 15 (d2)   */
1233 };
1234
1235 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1236 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1237 {
1238     out_str[0] = 0;
1239
1240     switch (src_modifier)
1241     {
1242     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1243     case WINED3DSPSM_DW:
1244     case WINED3DSPSM_NONE:
1245         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1246         break;
1247     case WINED3DSPSM_NEG:
1248         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1249         break;
1250     case WINED3DSPSM_NOT:
1251         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1252         break;
1253     case WINED3DSPSM_BIAS:
1254         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1255         break;
1256     case WINED3DSPSM_BIASNEG:
1257         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1258         break;
1259     case WINED3DSPSM_SIGN:
1260         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1261         break;
1262     case WINED3DSPSM_SIGNNEG:
1263         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1264         break;
1265     case WINED3DSPSM_COMP:
1266         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1267         break;
1268     case WINED3DSPSM_X2:
1269         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1270         break;
1271     case WINED3DSPSM_X2NEG:
1272         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1273         break;
1274     case WINED3DSPSM_ABS:
1275         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1276         break;
1277     case WINED3DSPSM_ABSNEG:
1278         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1279         break;
1280     default:
1281         FIXME("Unhandled modifier %u\n", src_modifier);
1282         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1283     }
1284 }
1285
1286 /** Writes the GLSL variable name that corresponds to the register that the
1287  * DX opcode parameter is trying to access */
1288 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1289         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1290 {
1291     /* oPos, oFog and oPts in D3D */
1292     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1293
1294     const struct wined3d_shader *shader = ins->ctx->shader;
1295     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1296     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1297     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1298
1299     *is_color = FALSE;
1300
1301     switch (reg->type)
1302     {
1303         case WINED3DSPR_TEMP:
1304             sprintf(register_name, "R%u", reg->idx);
1305             break;
1306
1307         case WINED3DSPR_INPUT:
1308             /* vertex shaders */
1309             if (!pshader)
1310             {
1311                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1312                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1313                 sprintf(register_name, "attrib%u", reg->idx);
1314                 break;
1315             }
1316
1317             /* pixel shaders >= 3.0 */
1318             if (reg_maps->shader_version.major >= 3)
1319             {
1320                 DWORD idx = shader->u.ps.input_reg_map[reg->idx];
1321                 unsigned int in_count = vec4_varyings(reg_maps->shader_version.major, gl_info);
1322
1323                 if (reg->rel_addr)
1324                 {
1325                     struct glsl_src_param rel_param;
1326
1327                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1328
1329                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1330                      * operation there */
1331                     if (idx)
1332                     {
1333                         if (shader->u.ps.declared_in_count > in_count)
1334                         {
1335                             sprintf(register_name,
1336                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1337                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1338                                     rel_param.param_str, idx);
1339                         }
1340                         else
1341                         {
1342                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1343                         }
1344                     }
1345                     else
1346                     {
1347                         if (shader->u.ps.declared_in_count > in_count)
1348                         {
1349                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1350                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1351                                     rel_param.param_str);
1352                         }
1353                         else
1354                         {
1355                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1356                         }
1357                     }
1358                 }
1359                 else
1360                 {
1361                     if (idx == in_count) sprintf(register_name, "gl_Color");
1362                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1363                     else sprintf(register_name, "IN[%u]", idx);
1364                 }
1365             }
1366             else
1367             {
1368                 if (!reg->idx) strcpy(register_name, "gl_Color");
1369                 else strcpy(register_name, "gl_SecondaryColor");
1370                 break;
1371             }
1372             break;
1373
1374         case WINED3DSPR_CONST:
1375             {
1376                 const char prefix = pshader ? 'P' : 'V';
1377
1378                 /* Relative addressing */
1379                 if (reg->rel_addr)
1380                 {
1381                     struct glsl_src_param rel_param;
1382                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1383                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1384                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1385                 }
1386                 else
1387                 {
1388                     if (shader_constant_is_local(shader, reg->idx))
1389                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1390                     else
1391                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1392                 }
1393             }
1394             break;
1395
1396         case WINED3DSPR_CONSTINT:
1397             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1398             else sprintf(register_name, "VI[%u]", reg->idx);
1399             break;
1400
1401         case WINED3DSPR_CONSTBOOL:
1402             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1403             else sprintf(register_name, "VB[%u]", reg->idx);
1404             break;
1405
1406         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1407             if (pshader) sprintf(register_name, "T%u", reg->idx);
1408             else sprintf(register_name, "A%u", reg->idx);
1409             break;
1410
1411         case WINED3DSPR_LOOP:
1412             sprintf(register_name, "aL%u", ins->ctx->loop_state->current_reg - 1);
1413             break;
1414
1415         case WINED3DSPR_SAMPLER:
1416             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1417             else sprintf(register_name, "Vsampler%u", reg->idx);
1418             break;
1419
1420         case WINED3DSPR_COLOROUT:
1421             if (reg->idx >= gl_info->limits.buffers)
1422                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1423
1424             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1425             break;
1426
1427         case WINED3DSPR_RASTOUT:
1428             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1429             break;
1430
1431         case WINED3DSPR_DEPTHOUT:
1432             sprintf(register_name, "gl_FragDepth");
1433             break;
1434
1435         case WINED3DSPR_ATTROUT:
1436             if (!reg->idx) sprintf(register_name, "OUT[8]");
1437             else sprintf(register_name, "OUT[9]");
1438             break;
1439
1440         case WINED3DSPR_TEXCRDOUT:
1441             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1442             sprintf(register_name, "OUT[%u]", reg->idx);
1443             break;
1444
1445         case WINED3DSPR_MISCTYPE:
1446             if (!reg->idx)
1447             {
1448                 /* vPos */
1449                 sprintf(register_name, "vpos");
1450             }
1451             else if (reg->idx == 1)
1452             {
1453                 /* Note that gl_FrontFacing is a bool, while vFace is
1454                  * a float for which the sign determines front/back */
1455                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1456             }
1457             else
1458             {
1459                 FIXME("Unhandled misctype register %d\n", reg->idx);
1460                 sprintf(register_name, "unrecognized_register");
1461             }
1462             break;
1463
1464         case WINED3DSPR_IMMCONST:
1465             switch (reg->immconst_type)
1466             {
1467                 case WINED3D_IMMCONST_SCALAR:
1468                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1469                     break;
1470
1471                 case WINED3D_IMMCONST_VEC4:
1472                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1473                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1474                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1475                     break;
1476
1477                 default:
1478                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1479                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1480             }
1481             break;
1482
1483         default:
1484             FIXME("Unhandled register name Type(%d)\n", reg->type);
1485             sprintf(register_name, "unrecognized_register");
1486             break;
1487     }
1488 }
1489
1490 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1491 {
1492     *str++ = '.';
1493     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1494     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1495     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1496     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1497     *str = '\0';
1498 }
1499
1500 /* Get the GLSL write mask for the destination register */
1501 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1502 {
1503     DWORD mask = param->write_mask;
1504
1505     if (shader_is_scalar(&param->reg))
1506     {
1507         mask = WINED3DSP_WRITEMASK_0;
1508         *write_mask = '\0';
1509     }
1510     else
1511     {
1512         shader_glsl_write_mask_to_str(mask, write_mask);
1513     }
1514
1515     return mask;
1516 }
1517
1518 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1519     unsigned int size = 0;
1520
1521     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1522     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1523     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1524     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1525
1526     return size;
1527 }
1528
1529 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1530 {
1531     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1532      * but addressed as "rgba". To fix this we need to swap the register's x
1533      * and z components. */
1534     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1535
1536     *str++ = '.';
1537     /* swizzle bits fields: wwzzyyxx */
1538     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1539     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1540     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1541     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1542     *str = '\0';
1543 }
1544
1545 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1546         BOOL fixup, DWORD mask, char *swizzle_str)
1547 {
1548     if (shader_is_scalar(&param->reg))
1549         *swizzle_str = '\0';
1550     else
1551         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1552 }
1553
1554 /* From a given parameter token, generate the corresponding GLSL string.
1555  * Also, return the actual register name and swizzle in case the
1556  * caller needs this information as well. */
1557 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1558         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src)
1559 {
1560     BOOL is_color = FALSE;
1561     char swizzle_str[6];
1562
1563     glsl_src->reg_name[0] = '\0';
1564     glsl_src->param_str[0] = '\0';
1565     swizzle_str[0] = '\0';
1566
1567     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1568     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1569     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1570 }
1571
1572 /* From a given parameter token, generate the corresponding GLSL string.
1573  * Also, return the actual register name and swizzle in case the
1574  * caller needs this information as well. */
1575 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1576         const struct wined3d_shader_dst_param *wined3d_dst, struct glsl_dst_param *glsl_dst)
1577 {
1578     BOOL is_color = FALSE;
1579
1580     glsl_dst->mask_str[0] = '\0';
1581     glsl_dst->reg_name[0] = '\0';
1582
1583     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1584     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1585 }
1586
1587 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1588 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1589         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1590 {
1591     struct glsl_dst_param glsl_dst;
1592     DWORD mask;
1593
1594     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1595     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1596
1597     return mask;
1598 }
1599
1600 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1601 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1602 {
1603     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1604 }
1605
1606 /** Process GLSL instruction modifiers */
1607 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1608 {
1609     struct glsl_dst_param dst_param;
1610     DWORD modifiers;
1611
1612     if (!ins->dst_count) return;
1613
1614     modifiers = ins->dst[0].modifiers;
1615     if (!modifiers) return;
1616
1617     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1618
1619     if (modifiers & WINED3DSPDM_SATURATE)
1620     {
1621         /* _SAT means to clamp the value of the register to between 0 and 1 */
1622         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1623                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1624     }
1625
1626     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1627     {
1628         FIXME("_centroid modifier not handled\n");
1629     }
1630
1631     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1632     {
1633         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1634     }
1635 }
1636
1637 static const char *shader_glsl_get_rel_op(enum wined3d_shader_rel_op op)
1638 {
1639     switch (op)
1640     {
1641         case WINED3D_SHADER_REL_OP_GT: return ">";
1642         case WINED3D_SHADER_REL_OP_EQ: return "==";
1643         case WINED3D_SHADER_REL_OP_GE: return ">=";
1644         case WINED3D_SHADER_REL_OP_LT: return "<";
1645         case WINED3D_SHADER_REL_OP_NE: return "!=";
1646         case WINED3D_SHADER_REL_OP_LE: return "<=";
1647         default:
1648             FIXME("Unrecognized operator %#x.\n", op);
1649             return "(\?\?)";
1650     }
1651 }
1652
1653 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1654         DWORD sampler_idx, DWORD flags, struct glsl_sample_function *sample_function)
1655 {
1656     enum wined3d_sampler_texture_type sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1657     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1658     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1659             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1660     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1661     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1662     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1663     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1664
1665     /* Note that there's no such thing as a projected cube texture. */
1666     switch(sampler_type) {
1667         case WINED3DSTT_1D:
1668             if (shadow)
1669             {
1670                 if (lod)
1671                 {
1672                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1673                 }
1674                 else if (grad)
1675                 {
1676                     if (gl_info->supported[EXT_GPU_SHADER4])
1677                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1678                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1679                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1680                     else
1681                     {
1682                         FIXME("Unsupported 1D shadow grad function.\n");
1683                         sample_function->name = "unsupported1DGrad";
1684                     }
1685                 }
1686                 else
1687                 {
1688                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1689                 }
1690                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1691             }
1692             else
1693             {
1694                 if (lod)
1695                 {
1696                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1697                 }
1698                 else if (grad)
1699                 {
1700                     if (gl_info->supported[EXT_GPU_SHADER4])
1701                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1702                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1703                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1704                     else
1705                     {
1706                         FIXME("Unsupported 1D grad function.\n");
1707                         sample_function->name = "unsupported1DGrad";
1708                     }
1709                 }
1710                 else
1711                 {
1712                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1713                 }
1714                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1715             }
1716             break;
1717
1718         case WINED3DSTT_2D:
1719             if (shadow)
1720             {
1721                 if (texrect)
1722                 {
1723                     if (lod)
1724                     {
1725                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1726                     }
1727                     else if (grad)
1728                     {
1729                         if (gl_info->supported[EXT_GPU_SHADER4])
1730                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1731                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1732                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1733                         else
1734                         {
1735                             FIXME("Unsupported RECT shadow grad function.\n");
1736                             sample_function->name = "unsupported2DRectGrad";
1737                         }
1738                     }
1739                     else
1740                     {
1741                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1742                     }
1743                 }
1744                 else
1745                 {
1746                     if (lod)
1747                     {
1748                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1749                     }
1750                     else if (grad)
1751                     {
1752                         if (gl_info->supported[EXT_GPU_SHADER4])
1753                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1754                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1755                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1756                         else
1757                         {
1758                             FIXME("Unsupported 2D shadow grad function.\n");
1759                             sample_function->name = "unsupported2DGrad";
1760                         }
1761                     }
1762                     else
1763                     {
1764                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1765                     }
1766                 }
1767                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1768             }
1769             else
1770             {
1771                 if (texrect)
1772                 {
1773                     if (lod)
1774                     {
1775                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1776                     }
1777                     else if (grad)
1778                     {
1779                         if (gl_info->supported[EXT_GPU_SHADER4])
1780                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1781                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1782                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1783                         else
1784                         {
1785                             FIXME("Unsupported RECT grad function.\n");
1786                             sample_function->name = "unsupported2DRectGrad";
1787                         }
1788                     }
1789                     else
1790                     {
1791                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1792                     }
1793                 }
1794                 else
1795                 {
1796                     if (lod)
1797                     {
1798                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1799                     }
1800                     else if (grad)
1801                     {
1802                         if (gl_info->supported[EXT_GPU_SHADER4])
1803                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1804                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1805                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1806                         else
1807                         {
1808                             FIXME("Unsupported 2D grad function.\n");
1809                             sample_function->name = "unsupported2DGrad";
1810                         }
1811                     }
1812                     else
1813                     {
1814                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1815                     }
1816                 }
1817                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1818             }
1819             break;
1820
1821         case WINED3DSTT_CUBE:
1822             if (shadow)
1823             {
1824                 FIXME("Unsupported Cube shadow function.\n");
1825                 sample_function->name = "unsupportedCubeShadow";
1826                 sample_function->coord_mask = 0;
1827             }
1828             else
1829             {
1830                 if (lod)
1831                 {
1832                     sample_function->name = "textureCubeLod";
1833                 }
1834                 else if (grad)
1835                 {
1836                     if (gl_info->supported[EXT_GPU_SHADER4])
1837                         sample_function->name = "textureCubeGrad";
1838                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1839                         sample_function->name = "textureCubeGradARB";
1840                     else
1841                     {
1842                         FIXME("Unsupported Cube grad function.\n");
1843                         sample_function->name = "unsupportedCubeGrad";
1844                     }
1845                 }
1846                 else
1847                 {
1848                     sample_function->name = "textureCube";
1849                 }
1850                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1851             }
1852             break;
1853
1854         case WINED3DSTT_VOLUME:
1855             if (shadow)
1856             {
1857                 FIXME("Unsupported 3D shadow function.\n");
1858                 sample_function->name = "unsupported3DShadow";
1859                 sample_function->coord_mask = 0;
1860             }
1861             else
1862             {
1863                 if (lod)
1864                 {
1865                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1866                 }
1867                 else  if (grad)
1868                 {
1869                     if (gl_info->supported[EXT_GPU_SHADER4])
1870                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1871                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1872                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1873                     else
1874                     {
1875                         FIXME("Unsupported 3D grad function.\n");
1876                         sample_function->name = "unsupported3DGrad";
1877                     }
1878                 }
1879                 else
1880                 {
1881                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1882                 }
1883                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1884             }
1885             break;
1886
1887         default:
1888             sample_function->name = "";
1889             sample_function->coord_mask = 0;
1890             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1891             break;
1892     }
1893 }
1894
1895 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1896         BOOL sign_fixup, enum fixup_channel_source channel_source)
1897 {
1898     switch(channel_source)
1899     {
1900         case CHANNEL_SOURCE_ZERO:
1901             strcat(arguments, "0.0");
1902             break;
1903
1904         case CHANNEL_SOURCE_ONE:
1905             strcat(arguments, "1.0");
1906             break;
1907
1908         case CHANNEL_SOURCE_X:
1909             strcat(arguments, reg_name);
1910             strcat(arguments, ".x");
1911             break;
1912
1913         case CHANNEL_SOURCE_Y:
1914             strcat(arguments, reg_name);
1915             strcat(arguments, ".y");
1916             break;
1917
1918         case CHANNEL_SOURCE_Z:
1919             strcat(arguments, reg_name);
1920             strcat(arguments, ".z");
1921             break;
1922
1923         case CHANNEL_SOURCE_W:
1924             strcat(arguments, reg_name);
1925             strcat(arguments, ".w");
1926             break;
1927
1928         default:
1929             FIXME("Unhandled channel source %#x\n", channel_source);
1930             strcat(arguments, "undefined");
1931             break;
1932     }
1933
1934     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1935 }
1936
1937 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1938 {
1939     struct wined3d_shader_dst_param dst;
1940     unsigned int mask_size, remaining;
1941     struct glsl_dst_param dst_param;
1942     char arguments[256];
1943     DWORD mask;
1944
1945     mask = 0;
1946     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1947     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1948     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1949     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1950     mask &= ins->dst[0].write_mask;
1951
1952     if (!mask) return; /* Nothing to do */
1953
1954     if (is_complex_fixup(fixup))
1955     {
1956         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1957         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1958         return;
1959     }
1960
1961     mask_size = shader_glsl_get_write_mask_size(mask);
1962
1963     dst = ins->dst[0];
1964     dst.write_mask = mask;
1965     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1966
1967     arguments[0] = '\0';
1968     remaining = mask_size;
1969     if (mask & WINED3DSP_WRITEMASK_0)
1970     {
1971         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1972         if (--remaining) strcat(arguments, ", ");
1973     }
1974     if (mask & WINED3DSP_WRITEMASK_1)
1975     {
1976         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1977         if (--remaining) strcat(arguments, ", ");
1978     }
1979     if (mask & WINED3DSP_WRITEMASK_2)
1980     {
1981         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1982         if (--remaining) strcat(arguments, ", ");
1983     }
1984     if (mask & WINED3DSP_WRITEMASK_3)
1985     {
1986         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1987         if (--remaining) strcat(arguments, ", ");
1988     }
1989
1990     if (mask_size > 1)
1991     {
1992         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1993                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1994     }
1995     else
1996     {
1997         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1998     }
1999 }
2000
2001 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2002         DWORD sampler, const struct glsl_sample_function *sample_function, DWORD swizzle,
2003         const char *dx, const char *dy, const char *bias, const char *coord_reg_fmt, ...)
2004 {
2005     const char *sampler_base;
2006     char dst_swizzle[6];
2007     struct color_fixup_desc fixup;
2008     BOOL np2_fixup = FALSE;
2009     va_list args;
2010
2011     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2012
2013     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2014     {
2015         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2016         fixup = priv->cur_ps_args->color_fixup[sampler];
2017         sampler_base = "Psampler";
2018
2019         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2020             if(bias) {
2021                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2022             } else {
2023                 np2_fixup = TRUE;
2024             }
2025         }
2026     } else {
2027         sampler_base = "Vsampler";
2028         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2029     }
2030
2031     shader_glsl_append_dst(ins->ctx->buffer, ins);
2032
2033     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2034
2035     va_start(args, coord_reg_fmt);
2036     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2037     va_end(args);
2038
2039     if(bias) {
2040         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2041     } else {
2042         if (np2_fixup) {
2043             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2044             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2045
2046             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2047                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2048         } else if(dx && dy) {
2049             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2050         } else {
2051             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2052         }
2053     }
2054
2055     if(!is_identity_fixup(fixup)) {
2056         shader_glsl_color_correction(ins, fixup);
2057     }
2058 }
2059
2060 /*****************************************************************************
2061  * Begin processing individual instruction opcodes
2062  ****************************************************************************/
2063
2064 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2065 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2066 {
2067     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2068     struct glsl_src_param src0_param;
2069     struct glsl_src_param src1_param;
2070     DWORD write_mask;
2071     char op;
2072
2073     /* Determine the GLSL operator to use based on the opcode */
2074     switch (ins->handler_idx)
2075     {
2076         case WINED3DSIH_MUL: op = '*'; break;
2077         case WINED3DSIH_ADD: op = '+'; break;
2078         case WINED3DSIH_SUB: op = '-'; break;
2079         default:
2080             op = ' ';
2081             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2082             break;
2083     }
2084
2085     write_mask = shader_glsl_append_dst(buffer, ins);
2086     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2087     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2088     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2089 }
2090
2091 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2092 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2093 {
2094     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2095     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2096     struct glsl_src_param src0_param;
2097     DWORD write_mask;
2098
2099     write_mask = shader_glsl_append_dst(buffer, ins);
2100     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2101
2102     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2103      * shader versions WINED3DSIO_MOVA is used for this. */
2104     if (ins->ctx->reg_maps->shader_version.major == 1
2105             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2106             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2107     {
2108         /* This is a simple floor() */
2109         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2110         if (mask_size > 1) {
2111             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2112         } else {
2113             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2114         }
2115     }
2116     else if(ins->handler_idx == WINED3DSIH_MOVA)
2117     {
2118         /* We need to *round* to the nearest int here. */
2119         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2120
2121         if (gl_info->supported[EXT_GPU_SHADER4])
2122         {
2123             if (mask_size > 1)
2124                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2125             else
2126                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2127         }
2128         else
2129         {
2130             if (mask_size > 1)
2131                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2132                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2133             else
2134                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2135                         src0_param.param_str, src0_param.param_str);
2136         }
2137     }
2138     else
2139     {
2140         shader_addline(buffer, "%s);\n", src0_param.param_str);
2141     }
2142 }
2143
2144 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2145 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2146 {
2147     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2148     struct glsl_src_param src0_param;
2149     struct glsl_src_param src1_param;
2150     DWORD dst_write_mask, src_write_mask;
2151     unsigned int dst_size = 0;
2152
2153     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2154     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2155
2156     /* dp3 works on vec3, dp4 on vec4 */
2157     if (ins->handler_idx == WINED3DSIH_DP4)
2158     {
2159         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2160     } else {
2161         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2162     }
2163
2164     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2165     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2166
2167     if (dst_size > 1) {
2168         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2169     } else {
2170         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2171     }
2172 }
2173
2174 /* Note that this instruction has some restrictions. The destination write mask
2175  * can't contain the w component, and the source swizzles have to be .xyzw */
2176 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2177 {
2178     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2179     struct glsl_src_param src0_param;
2180     struct glsl_src_param src1_param;
2181     char dst_mask[6];
2182
2183     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2184     shader_glsl_append_dst(ins->ctx->buffer, ins);
2185     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2186     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2187     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2188 }
2189
2190 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2191  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2192  * GLSL uses the value as-is. */
2193 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2194 {
2195     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2196     struct glsl_src_param src0_param;
2197     struct glsl_src_param src1_param;
2198     DWORD dst_write_mask;
2199     unsigned int dst_size;
2200
2201     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2202     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2203
2204     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2205     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2206
2207     if (dst_size > 1)
2208     {
2209         shader_addline(buffer, "vec%u(%s == 0.0 ? 1.0 : pow(abs(%s), %s)));\n",
2210                 dst_size, src1_param.param_str, src0_param.param_str, src1_param.param_str);
2211     }
2212     else
2213     {
2214         shader_addline(buffer, "%s == 0.0 ? 1.0 : pow(abs(%s), %s));\n",
2215                 src1_param.param_str, src0_param.param_str, src1_param.param_str);
2216     }
2217 }
2218
2219 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2220  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2221  * GLSL uses the value as-is. */
2222 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2223 {
2224     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2225     struct glsl_src_param src0_param;
2226     DWORD dst_write_mask;
2227     unsigned int dst_size;
2228
2229     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2230     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2231
2232     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2233
2234     if (dst_size > 1)
2235     {
2236         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2237                 dst_size, src0_param.param_str);
2238     }
2239     else
2240     {
2241         shader_addline(buffer, "log2(abs(%s)));\n",
2242                 src0_param.param_str);
2243     }
2244 }
2245
2246 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2247 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2248 {
2249     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2250     struct glsl_src_param src_param;
2251     const char *instruction;
2252     DWORD write_mask;
2253     unsigned i;
2254
2255     /* Determine the GLSL function to use based on the opcode */
2256     /* TODO: Possibly make this a table for faster lookups */
2257     switch (ins->handler_idx)
2258     {
2259         case WINED3DSIH_MIN: instruction = "min"; break;
2260         case WINED3DSIH_MAX: instruction = "max"; break;
2261         case WINED3DSIH_ABS: instruction = "abs"; break;
2262         case WINED3DSIH_FRC: instruction = "fract"; break;
2263         case WINED3DSIH_EXP: instruction = "exp2"; break;
2264         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2265         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2266         default: instruction = "";
2267             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2268             break;
2269     }
2270
2271     write_mask = shader_glsl_append_dst(buffer, ins);
2272
2273     shader_addline(buffer, "%s(", instruction);
2274
2275     if (ins->src_count)
2276     {
2277         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2278         shader_addline(buffer, "%s", src_param.param_str);
2279         for (i = 1; i < ins->src_count; ++i)
2280         {
2281             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2282             shader_addline(buffer, ", %s", src_param.param_str);
2283         }
2284     }
2285
2286     shader_addline(buffer, "));\n");
2287 }
2288
2289 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2290 {
2291     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2292     struct glsl_src_param src_param;
2293     unsigned int mask_size;
2294     DWORD write_mask;
2295     char dst_mask[6];
2296
2297     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2298     mask_size = shader_glsl_get_write_mask_size(write_mask);
2299     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2300
2301     shader_addline(buffer, "tmp0.x = dot(%s, %s);\n",
2302             src_param.param_str, src_param.param_str);
2303     shader_glsl_append_dst(buffer, ins);
2304
2305     if (mask_size > 1)
2306     {
2307         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s * inversesqrt(tmp0.x)));\n",
2308                 mask_size, src_param.param_str);
2309     }
2310     else
2311     {
2312         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s * inversesqrt(tmp0.x)));\n",
2313                 src_param.param_str);
2314     }
2315 }
2316
2317 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2318  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2319  *   dst.x = 2^(floor(src))
2320  *   dst.y = src - floor(src)
2321  *   dst.z = 2^src   (partial precision is allowed, but optional)
2322  *   dst.w = 1.0;
2323  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2324  *   dst = 2^src;    (partial precision is allowed, but optional)
2325  */
2326 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2327 {
2328     struct glsl_src_param src_param;
2329
2330     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2331
2332     if (ins->ctx->reg_maps->shader_version.major < 2)
2333     {
2334         char dst_mask[6];
2335
2336         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2337         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2338         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2339         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2340
2341         shader_glsl_append_dst(ins->ctx->buffer, ins);
2342         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2343         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2344     } else {
2345         DWORD write_mask;
2346         unsigned int mask_size;
2347
2348         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2349         mask_size = shader_glsl_get_write_mask_size(write_mask);
2350
2351         if (mask_size > 1) {
2352             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2353         } else {
2354             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2355         }
2356     }
2357 }
2358
2359 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2360 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2361 {
2362     struct glsl_src_param src_param;
2363     DWORD write_mask;
2364     unsigned int mask_size;
2365
2366     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2367     mask_size = shader_glsl_get_write_mask_size(write_mask);
2368     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2369
2370     if (mask_size > 1)
2371     {
2372         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2373                 mask_size, src_param.param_str);
2374     }
2375     else
2376     {
2377         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2378                 src_param.param_str);
2379     }
2380 }
2381
2382 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2383 {
2384     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2385     struct glsl_src_param src_param;
2386     DWORD write_mask;
2387     unsigned int mask_size;
2388
2389     write_mask = shader_glsl_append_dst(buffer, ins);
2390     mask_size = shader_glsl_get_write_mask_size(write_mask);
2391
2392     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2393
2394     if (mask_size > 1)
2395     {
2396         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2397                 mask_size, src_param.param_str);
2398     }
2399     else
2400     {
2401         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2402                 src_param.param_str);
2403     }
2404 }
2405
2406 /** Process signed comparison opcodes in GLSL. */
2407 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2408 {
2409     struct glsl_src_param src0_param;
2410     struct glsl_src_param src1_param;
2411     DWORD write_mask;
2412     unsigned int mask_size;
2413
2414     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2415     mask_size = shader_glsl_get_write_mask_size(write_mask);
2416     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2417     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2418
2419     if (mask_size > 1) {
2420         const char *compare;
2421
2422         switch(ins->handler_idx)
2423         {
2424             case WINED3DSIH_SLT: compare = "lessThan"; break;
2425             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2426             default: compare = "";
2427                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2428         }
2429
2430         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2431                 src0_param.param_str, src1_param.param_str);
2432     } else {
2433         switch(ins->handler_idx)
2434         {
2435             case WINED3DSIH_SLT:
2436                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2437                  * to return 0.0 but step returns 1.0 because step is not < x
2438                  * An alternative is a bvec compare padded with an unused second component.
2439                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2440                  * issue. Playing with not() is not possible either because not() does not accept
2441                  * a scalar.
2442                  */
2443                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2444                         src0_param.param_str, src1_param.param_str);
2445                 break;
2446             case WINED3DSIH_SGE:
2447                 /* Here we can use the step() function and safe a conditional */
2448                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2449                 break;
2450             default:
2451                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2452         }
2453
2454     }
2455 }
2456
2457 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2458 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2459 {
2460     struct glsl_src_param src0_param;
2461     struct glsl_src_param src1_param;
2462     struct glsl_src_param src2_param;
2463     DWORD write_mask, cmp_channel = 0;
2464     unsigned int i, j;
2465     char mask_char[6];
2466     BOOL temp_destination = FALSE;
2467
2468     if (shader_is_scalar(&ins->src[0].reg))
2469     {
2470         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2471
2472         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2473         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2474         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2475
2476         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2477                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2478     } else {
2479         DWORD dst_mask = ins->dst[0].write_mask;
2480         struct wined3d_shader_dst_param dst = ins->dst[0];
2481
2482         /* Cycle through all source0 channels */
2483         for (i=0; i<4; i++) {
2484             write_mask = 0;
2485             /* Find the destination channels which use the current source0 channel */
2486             for (j=0; j<4; j++) {
2487                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2488                 {
2489                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2490                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2491                 }
2492             }
2493             dst.write_mask = dst_mask & write_mask;
2494
2495             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2496             * The first lines may overwrite source parameters of the following lines.
2497             * Deal with that by using a temporary destination register if needed
2498             */
2499             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2500                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2501                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2502                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2503                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2504                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2505             {
2506                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2507                 if (!write_mask) continue;
2508                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2509                 temp_destination = TRUE;
2510             } else {
2511                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2512                 if (!write_mask) continue;
2513             }
2514
2515             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2516             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2517             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2518
2519             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2520                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2521         }
2522
2523         if(temp_destination) {
2524             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2525             shader_glsl_append_dst(ins->ctx->buffer, ins);
2526             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2527         }
2528     }
2529
2530 }
2531
2532 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2533 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2534  * the compare is done per component of src0. */
2535 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2536 {
2537     struct wined3d_shader_dst_param dst;
2538     struct glsl_src_param src0_param;
2539     struct glsl_src_param src1_param;
2540     struct glsl_src_param src2_param;
2541     DWORD write_mask, cmp_channel = 0;
2542     unsigned int i, j;
2543     DWORD dst_mask;
2544     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2545             ins->ctx->reg_maps->shader_version.minor);
2546
2547     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2548     {
2549         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2550         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2551         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2552         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2553
2554         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2555         if (ins->coissue)
2556         {
2557             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2558         } else {
2559             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2560                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2561         }
2562         return;
2563     }
2564     /* Cycle through all source0 channels */
2565     dst_mask = ins->dst[0].write_mask;
2566     dst = ins->dst[0];
2567     for (i=0; i<4; i++) {
2568         write_mask = 0;
2569         /* Find the destination channels which use the current source0 channel */
2570         for (j=0; j<4; j++) {
2571             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2572             {
2573                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2574                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2575             }
2576         }
2577
2578         dst.write_mask = dst_mask & write_mask;
2579         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2580         if (!write_mask) continue;
2581
2582         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2583         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2584         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2585
2586         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2587                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2588     }
2589 }
2590
2591 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2592 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2593 {
2594     struct glsl_src_param src0_param;
2595     struct glsl_src_param src1_param;
2596     struct glsl_src_param src2_param;
2597     DWORD write_mask;
2598
2599     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2600     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2601     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2602     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2603     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2604             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2605 }
2606
2607 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2608    Vertex shaders to GLSL codes */
2609 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2610 {
2611     int i;
2612     int nComponents = 0;
2613     struct wined3d_shader_dst_param tmp_dst = {{0}};
2614     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2615     struct wined3d_shader_instruction tmp_ins;
2616
2617     memset(&tmp_ins, 0, sizeof(tmp_ins));
2618
2619     /* Set constants for the temporary argument */
2620     tmp_ins.ctx = ins->ctx;
2621     tmp_ins.dst_count = 1;
2622     tmp_ins.dst = &tmp_dst;
2623     tmp_ins.src_count = 2;
2624     tmp_ins.src = tmp_src;
2625
2626     switch(ins->handler_idx)
2627     {
2628         case WINED3DSIH_M4x4:
2629             nComponents = 4;
2630             tmp_ins.handler_idx = WINED3DSIH_DP4;
2631             break;
2632         case WINED3DSIH_M4x3:
2633             nComponents = 3;
2634             tmp_ins.handler_idx = WINED3DSIH_DP4;
2635             break;
2636         case WINED3DSIH_M3x4:
2637             nComponents = 4;
2638             tmp_ins.handler_idx = WINED3DSIH_DP3;
2639             break;
2640         case WINED3DSIH_M3x3:
2641             nComponents = 3;
2642             tmp_ins.handler_idx = WINED3DSIH_DP3;
2643             break;
2644         case WINED3DSIH_M3x2:
2645             nComponents = 2;
2646             tmp_ins.handler_idx = WINED3DSIH_DP3;
2647             break;
2648         default:
2649             break;
2650     }
2651
2652     tmp_dst = ins->dst[0];
2653     tmp_src[0] = ins->src[0];
2654     tmp_src[1] = ins->src[1];
2655     for (i = 0; i < nComponents; ++i)
2656     {
2657         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2658         shader_glsl_dot(&tmp_ins);
2659         ++tmp_src[1].reg.idx;
2660     }
2661 }
2662
2663 /**
2664     The LRP instruction performs a component-wise linear interpolation
2665     between the second and third operands using the first operand as the
2666     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2667     This is equivalent to mix(src2, src1, src0);
2668 */
2669 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2670 {
2671     struct glsl_src_param src0_param;
2672     struct glsl_src_param src1_param;
2673     struct glsl_src_param src2_param;
2674     DWORD write_mask;
2675
2676     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2677
2678     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2679     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2680     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2681
2682     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2683             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2684 }
2685
2686 /** Process the WINED3DSIO_LIT instruction in GLSL:
2687  * dst.x = dst.w = 1.0
2688  * dst.y = (src0.x > 0) ? src0.x
2689  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2690  *                                        where src.w is clamped at +- 128
2691  */
2692 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2693 {
2694     struct glsl_src_param src0_param;
2695     struct glsl_src_param src1_param;
2696     struct glsl_src_param src3_param;
2697     char dst_mask[6];
2698
2699     shader_glsl_append_dst(ins->ctx->buffer, ins);
2700     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2701
2702     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2703     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2704     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2705
2706     /* The sdk specifies the instruction like this
2707      * dst.x = 1.0;
2708      * if(src.x > 0.0) dst.y = src.x
2709      * else dst.y = 0.0.
2710      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2711      * else dst.z = 0.0;
2712      * dst.w = 1.0;
2713      * (where power = src.w clamped between -128 and 128)
2714      *
2715      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2716      * dst.x = 1.0                                  ... No further explanation needed
2717      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2718      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2719      * dst.w = 1.0.                                 ... Nothing fancy.
2720      *
2721      * So we still have one conditional in there. So do this:
2722      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2723      *
2724      * 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),
2725      * 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.
2726      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to.
2727      *
2728      * Unfortunately pow(0.0 ^ 0.0) returns NaN on most GPUs, but lit with src.y = 0 and src.w = 0 returns
2729      * a non-NaN value in dst.z. What we return doesn't matter, as long as it is not NaN. Return 0, which is
2730      * what all Windows HW drivers and GL_ARB_vertex_program's LIT do.
2731      */
2732     shader_addline(ins->ctx->buffer,
2733             "vec4(1.0, max(%s, 0.0), %s == 0.0 ? 0.0 : "
2734             "pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
2735             src0_param.param_str, src3_param.param_str, src1_param.param_str,
2736             src0_param.param_str, src3_param.param_str, dst_mask);
2737 }
2738
2739 /** Process the WINED3DSIO_DST instruction in GLSL:
2740  * dst.x = 1.0
2741  * dst.y = src0.x * src0.y
2742  * dst.z = src0.z
2743  * dst.w = src1.w
2744  */
2745 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2746 {
2747     struct glsl_src_param src0y_param;
2748     struct glsl_src_param src0z_param;
2749     struct glsl_src_param src1y_param;
2750     struct glsl_src_param src1w_param;
2751     char dst_mask[6];
2752
2753     shader_glsl_append_dst(ins->ctx->buffer, ins);
2754     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2755
2756     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2757     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2758     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2759     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2760
2761     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2762             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2763 }
2764
2765 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2766  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2767  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2768  *
2769  * dst.x = cos(src0.?)
2770  * dst.y = sin(src0.?)
2771  * dst.z = dst.z
2772  * dst.w = dst.w
2773  */
2774 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2775 {
2776     struct glsl_src_param src0_param;
2777     DWORD write_mask;
2778
2779     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2780     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2781
2782     switch (write_mask) {
2783         case WINED3DSP_WRITEMASK_0:
2784             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2785             break;
2786
2787         case WINED3DSP_WRITEMASK_1:
2788             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2789             break;
2790
2791         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2792             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2793             break;
2794
2795         default:
2796             ERR("Write mask should be .x, .y or .xy\n");
2797             break;
2798     }
2799 }
2800
2801 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2802  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2803  * generate invalid code
2804  */
2805 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2806 {
2807     struct glsl_src_param src0_param;
2808     DWORD write_mask;
2809
2810     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2811     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2812
2813     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2814 }
2815
2816 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2817  * Start a for() loop where src1.y is the initial value of aL,
2818  *  increment aL by src1.z for a total of src1.x iterations.
2819  *  Need to use a temporary variable for this operation.
2820  */
2821 /* FIXME: I don't think nested loops will work correctly this way. */
2822 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2823 {
2824     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2825     const struct wined3d_shader *shader = ins->ctx->shader;
2826     const struct wined3d_shader_lconst *constant;
2827     struct glsl_src_param src1_param;
2828     const DWORD *control_values = NULL;
2829
2830     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2831
2832     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2833      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2834      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2835      * addressing.
2836      */
2837     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2838     {
2839         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2840         {
2841             if (constant->idx == ins->src[1].reg.idx)
2842             {
2843                 control_values = constant->value;
2844                 break;
2845             }
2846         }
2847     }
2848
2849     if (control_values)
2850     {
2851         struct wined3d_shader_loop_control loop_control;
2852         loop_control.count = control_values[0];
2853         loop_control.start = control_values[1];
2854         loop_control.step = (int)control_values[2];
2855
2856         if (loop_control.step > 0)
2857         {
2858             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2859                     loop_state->current_depth, loop_control.start,
2860                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2861                     loop_state->current_depth, loop_control.step);
2862         }
2863         else if (loop_control.step < 0)
2864         {
2865             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2866                     loop_state->current_depth, loop_control.start,
2867                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2868                     loop_state->current_depth, loop_control.step);
2869         }
2870         else
2871         {
2872             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2873                     loop_state->current_depth, loop_control.start, loop_state->current_depth,
2874                     loop_state->current_depth, loop_control.count,
2875                     loop_state->current_depth);
2876         }
2877     } else {
2878         shader_addline(ins->ctx->buffer,
2879                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2880                 loop_state->current_depth, loop_state->current_reg,
2881                 src1_param.reg_name, loop_state->current_depth, src1_param.reg_name,
2882                 loop_state->current_depth, loop_state->current_reg, src1_param.reg_name);
2883     }
2884
2885     ++loop_state->current_depth;
2886     ++loop_state->current_reg;
2887 }
2888
2889 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2890 {
2891     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2892
2893     shader_addline(ins->ctx->buffer, "}\n");
2894
2895     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2896     {
2897         --loop_state->current_depth;
2898         --loop_state->current_reg;
2899     }
2900
2901     if (ins->handler_idx == WINED3DSIH_ENDREP)
2902     {
2903         --loop_state->current_depth;
2904     }
2905 }
2906
2907 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2908 {
2909     const struct wined3d_shader *shader = ins->ctx->shader;
2910     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2911     const struct wined3d_shader_lconst *constant;
2912     struct glsl_src_param src0_param;
2913     const DWORD *control_values = NULL;
2914
2915     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2916     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2917     {
2918         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2919         {
2920             if (constant->idx == ins->src[0].reg.idx)
2921             {
2922                 control_values = constant->value;
2923                 break;
2924             }
2925         }
2926     }
2927
2928     if (control_values)
2929     {
2930         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2931                 loop_state->current_depth, loop_state->current_depth,
2932                 control_values[0], loop_state->current_depth);
2933     }
2934     else
2935     {
2936         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2937         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2938                 loop_state->current_depth, loop_state->current_depth,
2939                 src0_param.param_str, loop_state->current_depth);
2940     }
2941
2942     ++loop_state->current_depth;
2943 }
2944
2945 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2946 {
2947     struct glsl_src_param src0_param;
2948
2949     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2950     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2951 }
2952
2953 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2954 {
2955     struct glsl_src_param src0_param;
2956     struct glsl_src_param src1_param;
2957
2958     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2959     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2960
2961     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2962             src0_param.param_str, shader_glsl_get_rel_op(ins->flags), src1_param.param_str);
2963 }
2964
2965 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2966 {
2967     shader_addline(ins->ctx->buffer, "} else {\n");
2968 }
2969
2970 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2971 {
2972     shader_addline(ins->ctx->buffer, "break;\n");
2973 }
2974
2975 /* FIXME: According to MSDN the compare is done per component. */
2976 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2977 {
2978     struct glsl_src_param src0_param;
2979     struct glsl_src_param src1_param;
2980
2981     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2982     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2983
2984     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2985             src0_param.param_str, shader_glsl_get_rel_op(ins->flags), src1_param.param_str);
2986 }
2987
2988 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2989 {
2990     shader_addline(ins->ctx->buffer, "}\n");
2991     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2992 }
2993
2994 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2995 {
2996     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2997 }
2998
2999 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3000 {
3001     struct glsl_src_param src1_param;
3002
3003     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3004     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3005 }
3006
3007 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3008 {
3009     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3010      * function only suppresses the unhandled instruction warning
3011      */
3012 }
3013
3014 /*********************************************
3015  * Pixel Shader Specific Code begins here
3016  ********************************************/
3017 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3018 {
3019     const struct wined3d_shader *shader = ins->ctx->shader;
3020     struct wined3d_device *device = shader->device;
3021     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3022             ins->ctx->reg_maps->shader_version.minor);
3023     struct glsl_sample_function sample_function;
3024     const struct wined3d_texture *texture;
3025     DWORD sample_flags = 0;
3026     DWORD sampler_idx;
3027     DWORD mask = 0, swizzle;
3028
3029     /* 1.0-1.4: Use destination register as sampler source.
3030      * 2.0+: Use provided sampler source. */
3031     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3032     else sampler_idx = ins->src[1].reg.idx;
3033     texture = device->stateBlock->state.textures[sampler_idx];
3034
3035     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3036     {
3037         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3038         DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3039                 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3040         enum wined3d_sampler_texture_type sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3041
3042         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3043         if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3044             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3045             switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3046                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3047                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3048                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3049                 case WINED3DTTFF_COUNT4:
3050                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3051             }
3052         }
3053     }
3054     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3055     {
3056         DWORD src_mod = ins->src[0].modifiers;
3057
3058         if (src_mod == WINED3DSPSM_DZ) {
3059             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3060             mask = WINED3DSP_WRITEMASK_2;
3061         } else if (src_mod == WINED3DSPSM_DW) {
3062             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3063             mask = WINED3DSP_WRITEMASK_3;
3064         }
3065     } else {
3066         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3067         {
3068             /* ps 2.0 texldp instruction always divides by the fourth component. */
3069             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3070             mask = WINED3DSP_WRITEMASK_3;
3071         }
3072     }
3073
3074     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3075         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3076
3077     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3078     mask |= sample_function.coord_mask;
3079
3080     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3081     else swizzle = ins->src[1].swizzle;
3082
3083     /* 1.0-1.3: Use destination register as coordinate source.
3084        1.4+: Use provided coordinate source register. */
3085     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3086     {
3087         char coord_mask[6];
3088         shader_glsl_write_mask_to_str(mask, coord_mask);
3089         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3090                 "T%u%s", sampler_idx, coord_mask);
3091     }
3092     else
3093     {
3094         struct glsl_src_param coord_param;
3095         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3096         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3097         {
3098             struct glsl_src_param bias;
3099             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3100             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3101                     "%s", coord_param.param_str);
3102         } else {
3103             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3104                     "%s", coord_param.param_str);
3105         }
3106     }
3107 }
3108
3109 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3110 {
3111     const struct wined3d_shader *shader = ins->ctx->shader;
3112     struct wined3d_device *device = shader->device;
3113     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3114     struct glsl_src_param coord_param, dx_param, dy_param;
3115     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3116     struct glsl_sample_function sample_function;
3117     DWORD sampler_idx;
3118     DWORD swizzle = ins->src[1].swizzle;
3119     const struct wined3d_texture *texture;
3120
3121     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3122     {
3123         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3124         shader_glsl_tex(ins);
3125         return;
3126     }
3127
3128     sampler_idx = ins->src[1].reg.idx;
3129     texture = device->stateBlock->state.textures[sampler_idx];
3130     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3131         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3132
3133     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3134     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3135     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3136     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3137
3138     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3139                                 "%s", coord_param.param_str);
3140 }
3141
3142 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3143 {
3144     const struct wined3d_shader *shader = ins->ctx->shader;
3145     struct wined3d_device *device = shader->device;
3146     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3147     struct glsl_src_param coord_param, lod_param;
3148     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3149     struct glsl_sample_function sample_function;
3150     DWORD sampler_idx;
3151     DWORD swizzle = ins->src[1].swizzle;
3152     const struct wined3d_texture *texture;
3153
3154     sampler_idx = ins->src[1].reg.idx;
3155     texture = device->stateBlock->state.textures[sampler_idx];
3156     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3157         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3158
3159     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3160     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3161
3162     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3163
3164     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3165             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3166     {
3167         /* Plain GLSL only supports Lod sampling functions in vertex shaders.
3168          * However, the NVIDIA drivers allow them in fragment shaders as well,
3169          * even without the appropriate extension. */
3170         WARN("Using %s in fragment shader.\n", sample_function.name);
3171     }
3172     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3173             "%s", coord_param.param_str);
3174 }
3175
3176 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3177 {
3178     /* FIXME: Make this work for more than just 2D textures */
3179     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3180     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3181
3182     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3183     {
3184         char dst_mask[6];
3185
3186         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3187         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3188                 ins->dst[0].reg.idx, dst_mask);
3189     } else {
3190         DWORD reg = ins->src[0].reg.idx;
3191         DWORD src_mod = ins->src[0].modifiers;
3192         char dst_swizzle[6];
3193
3194         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3195
3196         if (src_mod == WINED3DSPSM_DZ)
3197         {
3198             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3199             struct glsl_src_param div_param;
3200
3201             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3202
3203             if (mask_size > 1) {
3204                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3205             } else {
3206                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3207             }
3208         }
3209         else if (src_mod == WINED3DSPSM_DW)
3210         {
3211             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3212             struct glsl_src_param div_param;
3213
3214             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3215
3216             if (mask_size > 1) {
3217                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3218             } else {
3219                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3220             }
3221         } else {
3222             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3223         }
3224     }
3225 }
3226
3227 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3228  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3229  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3230 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3231 {
3232     DWORD sampler_idx = ins->dst[0].reg.idx;
3233     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3234     struct glsl_sample_function sample_function;
3235     struct glsl_src_param src0_param;
3236     UINT mask_size;
3237
3238     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3239
3240     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3241      * scalar, and projected sampling would require 4.
3242      *
3243      * It is a dependent read - not valid with conditional NP2 textures
3244      */
3245     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3246     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3247
3248     switch(mask_size)
3249     {
3250         case 1:
3251             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3252                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3253             break;
3254
3255         case 2:
3256             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3257                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3258             break;
3259
3260         case 3:
3261             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3262                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3263             break;
3264
3265         default:
3266             FIXME("Unexpected mask size %u\n", mask_size);
3267             break;
3268     }
3269 }
3270
3271 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3272  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3273 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3274 {
3275     DWORD dstreg = ins->dst[0].reg.idx;
3276     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3277     struct glsl_src_param src0_param;
3278     DWORD dst_mask;
3279     unsigned int mask_size;
3280
3281     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3282     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3283     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3284
3285     if (mask_size > 1) {
3286         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3287     } else {
3288         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3289     }
3290 }
3291
3292 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3293  * Calculate the depth as dst.x / dst.y   */
3294 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3295 {
3296     struct glsl_dst_param dst_param;
3297
3298     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3299
3300     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3301      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3302      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3303      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3304      * >= 1.0 or < 0.0
3305      */
3306     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3307             dst_param.reg_name, dst_param.reg_name);
3308 }
3309
3310 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3311  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3312  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3313  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3314  */
3315 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3316 {
3317     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3318     DWORD dstreg = ins->dst[0].reg.idx;
3319     struct glsl_src_param src0_param;
3320
3321     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3322
3323     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3324     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3325 }
3326
3327 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3328  * Calculate the 1st of a 2-row matrix multiplication. */
3329 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3330 {
3331     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3332     DWORD reg = ins->dst[0].reg.idx;
3333     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3334     struct glsl_src_param src0_param;
3335
3336     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3337     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3338 }
3339
3340 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3341  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3342 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3343 {
3344     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3345     DWORD reg = ins->dst[0].reg.idx;
3346     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3347     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3348     struct glsl_src_param src0_param;
3349
3350     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3351     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + tex_mx->current_row, reg, src0_param.param_str);
3352     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
3353 }
3354
3355 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3356 {
3357     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3358     DWORD reg = ins->dst[0].reg.idx;
3359     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3360     struct glsl_sample_function sample_function;
3361     struct glsl_src_param src0_param;
3362
3363     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3364     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3365
3366     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3367
3368     /* Sample the texture using the calculated coordinates */
3369     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3370 }
3371
3372 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3373  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3374 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3375 {
3376     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3377     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3378     struct glsl_sample_function sample_function;
3379     struct glsl_src_param src0_param;
3380     DWORD reg = ins->dst[0].reg.idx;
3381
3382     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3383     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3384
3385     /* Dependent read, not valid with conditional NP2 */
3386     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3387
3388     /* Sample the texture using the calculated coordinates */
3389     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3390
3391     tex_mx->current_row = 0;
3392 }
3393
3394 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3395  * Perform the 3rd row of a 3x3 matrix multiply */
3396 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3397 {
3398     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3399     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3400     struct glsl_src_param src0_param;
3401     char dst_mask[6];
3402     DWORD reg = ins->dst[0].reg.idx;
3403
3404     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3405
3406     shader_glsl_append_dst(ins->ctx->buffer, ins);
3407     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3408     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3409
3410     tex_mx->current_row = 0;
3411 }
3412
3413 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3414  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3415 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3416 {
3417     struct glsl_src_param src0_param;
3418     struct glsl_src_param src1_param;
3419     DWORD reg = ins->dst[0].reg.idx;
3420     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3421     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3422     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3423     struct glsl_sample_function sample_function;
3424
3425     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3426     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3427
3428     /* Perform the last matrix multiply operation */
3429     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3430     /* Reflection calculation */
3431     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3432
3433     /* Dependent read, not valid with conditional NP2 */
3434     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3435
3436     /* Sample the texture */
3437     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3438
3439     tex_mx->current_row = 0;
3440 }
3441
3442 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3443  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3444 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3445 {
3446     DWORD reg = ins->dst[0].reg.idx;
3447     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3448     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3449     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3450     struct glsl_sample_function sample_function;
3451     struct glsl_src_param src0_param;
3452
3453     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3454
3455     /* Perform the last matrix multiply operation */
3456     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3457
3458     /* Construct the eye-ray vector from w coordinates */
3459     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3460             tex_mx->texcoord_w[0], tex_mx->texcoord_w[1], reg);
3461     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3462
3463     /* Dependent read, not valid with conditional NP2 */
3464     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3465
3466     /* Sample the texture using the calculated coordinates */
3467     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3468
3469     tex_mx->current_row = 0;
3470 }
3471
3472 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3473  * Apply a fake bump map transform.
3474  * texbem is pshader <= 1.3 only, this saves a few version checks
3475  */
3476 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3477 {
3478     const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3479     struct glsl_sample_function sample_function;
3480     struct glsl_src_param coord_param;
3481     DWORD sampler_idx;
3482     DWORD mask;
3483     DWORD flags;
3484     char coord_mask[6];
3485
3486     sampler_idx = ins->dst[0].reg.idx;
3487     flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3488             & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3489
3490     /* Dependent read, not valid with conditional NP2 */
3491     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3492     mask = sample_function.coord_mask;
3493
3494     shader_glsl_write_mask_to_str(mask, coord_mask);
3495
3496     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3497          * so we can't let the GL handle this.
3498          */
3499     if (flags & WINED3D_PSARGS_PROJECTED) {
3500         DWORD div_mask=0;
3501         char coord_div_mask[3];
3502         switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3503             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3504             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3505             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3506             case WINED3DTTFF_COUNT4:
3507             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3508         }
3509         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3510         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3511     }
3512
3513     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3514
3515     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3516             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3517             coord_param.param_str, coord_mask);
3518
3519     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3520     {
3521         struct glsl_src_param luminance_param;
3522         struct glsl_dst_param dst_param;
3523
3524         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3525         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3526
3527         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3528                 dst_param.reg_name, dst_param.mask_str,
3529                 luminance_param.param_str, sampler_idx, sampler_idx);
3530     }
3531 }
3532
3533 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3534 {
3535     struct glsl_src_param src0_param, src1_param;
3536     DWORD sampler_idx = ins->dst[0].reg.idx;
3537
3538     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3539     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3540
3541     shader_glsl_append_dst(ins->ctx->buffer, ins);
3542     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3543             src0_param.param_str, sampler_idx, src1_param.param_str);
3544 }
3545
3546 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3547  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3548 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3549 {
3550     struct glsl_sample_function sample_function;
3551     struct glsl_src_param src0_param;
3552     DWORD sampler_idx = ins->dst[0].reg.idx;
3553
3554     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3555
3556     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3557     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3558             "%s.wx", src0_param.reg_name);
3559 }
3560
3561 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3562  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3563 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3564 {
3565     struct glsl_sample_function sample_function;
3566     struct glsl_src_param src0_param;
3567     DWORD sampler_idx = ins->dst[0].reg.idx;
3568
3569     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3570
3571     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3572     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3573             "%s.yz", src0_param.reg_name);
3574 }
3575
3576 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3577  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3578 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3579 {
3580     struct glsl_sample_function sample_function;
3581     struct glsl_src_param src0_param;
3582     DWORD sampler_idx = ins->dst[0].reg.idx;
3583
3584     /* Dependent read, not valid with conditional NP2 */
3585     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3586     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3587
3588     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3589             "%s", src0_param.param_str);
3590 }
3591
3592 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3593  * If any of the first 3 components are < 0, discard this pixel */
3594 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3595 {
3596     struct glsl_dst_param dst_param;
3597
3598     /* The argument is a destination parameter, and no writemasks are allowed */
3599     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3600     if (ins->ctx->reg_maps->shader_version.major >= 2)
3601     {
3602         /* 2.0 shaders compare all 4 components in texkill */
3603         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3604     } else {
3605         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3606          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3607          * 4 components are defined, only the first 3 are used
3608          */
3609         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3610     }
3611 }
3612
3613 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3614  * dst = dot2(src0, src1) + src2 */
3615 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3616 {
3617     struct glsl_src_param src0_param;
3618     struct glsl_src_param src1_param;
3619     struct glsl_src_param src2_param;
3620     DWORD write_mask;
3621     unsigned int mask_size;
3622
3623     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3624     mask_size = shader_glsl_get_write_mask_size(write_mask);
3625
3626     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3627     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3628     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3629
3630     if (mask_size > 1) {
3631         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3632                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3633     } else {
3634         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3635                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3636     }
3637 }
3638
3639 static void shader_glsl_input_pack(const struct wined3d_shader *shader, struct wined3d_shader_buffer *buffer,
3640         const struct wined3d_shader_signature_element *input_signature,
3641         const struct wined3d_shader_reg_maps *reg_maps,
3642         enum vertexprocessing_mode vertexprocessing)
3643 {
3644     WORD map = reg_maps->input_registers;
3645     unsigned int i;
3646
3647     for (i = 0; map; map >>= 1, ++i)
3648     {
3649         const char *semantic_name;
3650         UINT semantic_idx;
3651         char reg_mask[6];
3652
3653         /* Unused */
3654         if (!(map & 1)) continue;
3655
3656         semantic_name = input_signature[i].semantic_name;
3657         semantic_idx = input_signature[i].semantic_idx;
3658         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3659
3660         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3661         {
3662             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3663                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3664                         shader->u.ps.input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3665             else
3666                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3667                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3668         }
3669         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3670         {
3671             if (!semantic_idx)
3672                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3673                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3674             else if (semantic_idx == 1)
3675                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3676                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3677             else
3678                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3679                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3680         }
3681         else
3682         {
3683             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3684                     shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3685         }
3686     }
3687 }
3688
3689 /*********************************************
3690  * Vertex Shader Specific Code begins here
3691  ********************************************/
3692
3693 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry)
3694 {
3695     struct glsl_program_key key;
3696
3697     key.vshader = entry->vshader;
3698     key.pshader = entry->pshader;
3699     key.vs_args = entry->vs_args;
3700     key.ps_args = entry->ps_args;
3701
3702     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3703     {
3704         ERR("Failed to insert program entry.\n");
3705     }
3706 }
3707
3708 static struct glsl_shader_prog_link *get_glsl_program_entry(const struct shader_glsl_priv *priv,
3709         const struct wined3d_shader *vshader, const struct wined3d_shader *pshader,
3710         const struct vs_compile_args *vs_args, const struct ps_compile_args *ps_args)
3711 {
3712     struct wine_rb_entry *entry;
3713     struct glsl_program_key key;
3714
3715     key.vshader = vshader;
3716     key.pshader = pshader;
3717     key.vs_args = *vs_args;
3718     key.ps_args = *ps_args;
3719
3720     entry = wine_rb_get(&priv->program_lookup, &key);
3721     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3722 }
3723
3724 /* GL locking is done by the caller */
3725 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3726         struct glsl_shader_prog_link *entry)
3727 {
3728     struct glsl_program_key key;
3729
3730     key.vshader = entry->vshader;
3731     key.pshader = entry->pshader;
3732     key.vs_args = entry->vs_args;
3733     key.ps_args = entry->ps_args;
3734     wine_rb_remove(&priv->program_lookup, &key);
3735
3736     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3737     if (entry->vshader) list_remove(&entry->vshader_entry);
3738     if (entry->pshader) list_remove(&entry->pshader_entry);
3739     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3740     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3741     HeapFree(GetProcessHeap(), 0, entry);
3742 }
3743
3744 static void handle_ps3_input(struct wined3d_shader_buffer *buffer,
3745         const struct wined3d_gl_info *gl_info, const DWORD *map,
3746         const struct wined3d_shader_signature_element *input_signature,
3747         const struct wined3d_shader_reg_maps *reg_maps_in,
3748         const struct wined3d_shader_signature_element *output_signature,
3749         const struct wined3d_shader_reg_maps *reg_maps_out)
3750 {
3751     unsigned int i, j;
3752     const char *semantic_name_in;
3753     UINT semantic_idx_in;
3754     DWORD *set;
3755     DWORD in_idx;
3756     unsigned int in_count = vec4_varyings(3, gl_info);
3757     char reg_mask[6];
3758     char destination[50];
3759     WORD input_map, output_map;
3760
3761     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3762
3763     input_map = reg_maps_in->input_registers;
3764     for (i = 0; input_map; input_map >>= 1, ++i)
3765     {
3766         if (!(input_map & 1)) continue;
3767
3768         in_idx = map[i];
3769         /* Declared, but not read register */
3770         if (in_idx == ~0U) continue;
3771         if (in_idx >= (in_count + 2))
3772         {
3773             FIXME("More input varyings declared than supported, expect issues.\n");
3774             continue;
3775         }
3776
3777         if (in_idx == in_count) {
3778             sprintf(destination, "gl_FrontColor");
3779         } else if (in_idx == in_count + 1) {
3780             sprintf(destination, "gl_FrontSecondaryColor");
3781         } else {
3782             sprintf(destination, "IN[%u]", in_idx);
3783         }
3784
3785         semantic_name_in = input_signature[i].semantic_name;
3786         semantic_idx_in = input_signature[i].semantic_idx;
3787         set[in_idx] = ~0U;
3788
3789         output_map = reg_maps_out->output_registers;
3790         for (j = 0; output_map; output_map >>= 1, ++j)
3791         {
3792             DWORD mask;
3793
3794             if (!(output_map & 1)
3795                     || semantic_idx_in != output_signature[j].semantic_idx
3796                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3797                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3798                 continue;
3799
3800             set[in_idx] = mask;
3801             shader_glsl_write_mask_to_str(mask, reg_mask);
3802
3803             shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3804                     destination, reg_mask, j, reg_mask);
3805         }
3806     }
3807
3808     for (i = 0; i < in_count + 2; ++i)
3809     {
3810         unsigned int size;
3811
3812         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3813             continue;
3814
3815         if (set[i] == ~0U) set[i] = 0;
3816
3817         size = 0;
3818         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3819         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3820         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3821         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3822         reg_mask[size] = '\0';
3823
3824         if (i == in_count) sprintf(destination, "gl_FrontColor");
3825         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3826         else sprintf(destination, "IN[%u]", i);
3827
3828         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3829         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3830     }
3831
3832     HeapFree(GetProcessHeap(), 0, set);
3833 }
3834
3835 /* GL locking is done by the caller */
3836 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3837         const struct wined3d_shader *vs, const struct wined3d_shader *ps,
3838         const struct wined3d_gl_info *gl_info)
3839 {
3840     GLhandleARB ret = 0;
3841     DWORD ps_major = ps ? ps->reg_maps.shader_version.major : 0;
3842     unsigned int i;
3843     const char *semantic_name;
3844     UINT semantic_idx;
3845     char reg_mask[6];
3846     const struct wined3d_shader_signature_element *output_signature = vs->output_signature;
3847     WORD map = vs->reg_maps.output_registers;
3848
3849     shader_buffer_clear(buffer);
3850
3851     shader_addline(buffer, "#version 120\n");
3852
3853     if (ps_major < 3)
3854     {
3855         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3856
3857         for (i = 0; map; map >>= 1, ++i)
3858         {
3859             DWORD write_mask;
3860
3861             if (!(map & 1)) continue;
3862
3863             semantic_name = output_signature[i].semantic_name;
3864             semantic_idx = output_signature[i].semantic_idx;
3865             write_mask = output_signature[i].mask;
3866             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3867
3868             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3869             {
3870                 if (!semantic_idx)
3871                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n",
3872                             reg_mask, i, reg_mask);
3873                 else if (semantic_idx == 1)
3874                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n",
3875                             reg_mask, i, reg_mask);
3876             }
3877             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3878             {
3879                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3880                         reg_mask, i, reg_mask);
3881             }
3882             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3883             {
3884                 if (semantic_idx < 8)
3885                 {
3886                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3887                         write_mask |= WINED3DSP_WRITEMASK_3;
3888
3889                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3890                             semantic_idx, reg_mask, i, reg_mask);
3891                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3892                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3893                 }
3894             }
3895             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3896             {
3897                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3898             }
3899             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3900             {
3901                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3902             }
3903         }
3904         shader_addline(buffer, "}\n");
3905
3906     }
3907     else
3908     {
3909         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3910         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3911         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3912
3913         /* First, sort out position and point size. Those are not passed to the pixel shader */
3914         for (i = 0; map; map >>= 1, ++i)
3915         {
3916             if (!(map & 1)) continue;
3917
3918             semantic_name = output_signature[i].semantic_name;
3919             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3920
3921             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3922             {
3923                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3924                         reg_mask, i, reg_mask);
3925             }
3926             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3927             {
3928                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3929             }
3930         }
3931
3932         /* Then, fix the pixel shader input */
3933         handle_ps3_input(buffer, gl_info, ps->u.ps.input_reg_map, ps->input_signature,
3934                 &ps->reg_maps, output_signature, &vs->reg_maps);
3935
3936         shader_addline(buffer, "}\n");
3937     }
3938
3939     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3940     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3941     shader_glsl_compile(gl_info, ret, buffer->buffer);
3942
3943     return ret;
3944 }
3945
3946 /* GL locking is done by the caller */
3947 static void hardcode_local_constants(const struct wined3d_shader *shader,
3948         const struct wined3d_gl_info *gl_info, GLhandleARB programId, char prefix)
3949 {
3950     const struct wined3d_shader_lconst *lconst;
3951     GLint tmp_loc;
3952     const float *value;
3953     char glsl_name[8];
3954
3955     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3956     {
3957         value = (const float *)lconst->value;
3958         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3959         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3960         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3961     }
3962     checkGLcall("Hardcoding local constants");
3963 }
3964
3965 /* GL locking is done by the caller */
3966 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3967         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
3968         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3969 {
3970     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3971     const struct wined3d_gl_info *gl_info = context->gl_info;
3972     const DWORD *function = shader->function;
3973     struct shader_glsl_ctx_priv priv_ctx;
3974
3975     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3976     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3977
3978     memset(&priv_ctx, 0, sizeof(priv_ctx));
3979     priv_ctx.cur_ps_args = args;
3980     priv_ctx.cur_np2fixup_info = np2fixup_info;
3981
3982     shader_addline(buffer, "#version 120\n");
3983
3984     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
3985     {
3986         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3987     }
3988     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3989     {
3990         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3991          * drivers write a warning if we don't do so
3992          */
3993         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3994     }
3995     if (gl_info->supported[EXT_GPU_SHADER4])
3996     {
3997         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3998     }
3999
4000     /* Base Declarations */
4001     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4002
4003     /* Pack 3.0 inputs */
4004     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4005         shader_glsl_input_pack(shader, buffer, shader->input_signature, reg_maps, args->vp_mode);
4006
4007     /* Base Shader Body */
4008     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4009
4010     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4011     if (reg_maps->shader_version.major < 2)
4012     {
4013         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4014         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4015     }
4016
4017     if (args->srgb_correction)
4018     {
4019         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4020         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4021         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4022         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4023         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4024         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4025     }
4026     /* Pixel shader < 3.0 do not replace the fog stage.
4027      * This implements linear fog computation and blending.
4028      * TODO: non linear fog
4029      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4030      * -1/(e-s) and e/(e-s) respectively.
4031      */
4032     if (reg_maps->shader_version.major < 3)
4033     {
4034         switch(args->fog) {
4035             case FOG_OFF: break;
4036             case FOG_LINEAR:
4037                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4038                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4039                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4040                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4041                 break;
4042             case FOG_EXP:
4043                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4044                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4045                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4046                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4047                 break;
4048             case FOG_EXP2:
4049                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4050                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4051                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4052                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4053                 break;
4054         }
4055     }
4056
4057     shader_addline(buffer, "}\n");
4058
4059     TRACE("Compiling shader object %u\n", shader_obj);
4060     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4061
4062     /* Store the shader object */
4063     return shader_obj;
4064 }
4065
4066 /* GL locking is done by the caller */
4067 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4068         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
4069         const struct vs_compile_args *args)
4070 {
4071     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4072     const struct wined3d_gl_info *gl_info = context->gl_info;
4073     const DWORD *function = shader->function;
4074     struct shader_glsl_ctx_priv priv_ctx;
4075
4076     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4077     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4078
4079     shader_addline(buffer, "#version 120\n");
4080
4081     if (gl_info->supported[EXT_GPU_SHADER4])
4082         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4083
4084     memset(&priv_ctx, 0, sizeof(priv_ctx));
4085     priv_ctx.cur_vs_args = args;
4086
4087     /* Base Declarations */
4088     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4089
4090     /* Base Shader Body */
4091     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4092
4093     /* Unpack outputs */
4094     shader_addline(buffer, "order_ps_input(OUT);\n");
4095
4096     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4097      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4098      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4099      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4100      */
4101     if (args->fog_src == VS_FOG_Z)
4102         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4103     else if (!reg_maps->fog)
4104         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4105
4106     /* We always store the clipplanes without y inversion */
4107     if (args->clip_enabled)
4108         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4109
4110     /* Write the final position.
4111      *
4112      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4113      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4114      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4115      * contains 1.0 to allow a mad.
4116      */
4117     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4118     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4119
4120     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4121      *
4122      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4123      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4124      * which is the same as z = z * 2 - w.
4125      */
4126     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4127
4128     shader_addline(buffer, "}\n");
4129
4130     TRACE("Compiling shader object %u\n", shader_obj);
4131     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4132
4133     return shader_obj;
4134 }
4135
4136 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4137         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4138         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4139 {
4140     struct wined3d_state *state = &shader->device->stateBlock->state;
4141     UINT i;
4142     DWORD new_size;
4143     struct glsl_ps_compiled_shader *new_array;
4144     struct glsl_pshader_private    *shader_data;
4145     struct ps_np2fixup_info        *np2fixup = NULL;
4146     GLhandleARB ret;
4147
4148     if (!shader->backend_data)
4149     {
4150         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4151         if (!shader->backend_data)
4152         {
4153             ERR("Failed to allocate backend data.\n");
4154             return 0;
4155         }
4156     }
4157     shader_data = shader->backend_data;
4158
4159     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4160      * so a linear search is more performant than a hashmap or a binary search
4161      * (cache coherency etc)
4162      */
4163     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4164     {
4165         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4166         {
4167             if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4168             return shader_data->gl_shaders[i].prgId;
4169         }
4170     }
4171
4172     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4173     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4174         if (shader_data->num_gl_shaders)
4175         {
4176             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4177             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4178                                     new_size * sizeof(*shader_data->gl_shaders));
4179         } else {
4180             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4181             new_size = 1;
4182         }
4183
4184         if(!new_array) {
4185             ERR("Out of memory\n");
4186             return 0;
4187         }
4188         shader_data->gl_shaders = new_array;
4189         shader_data->shader_array_size = new_size;
4190     }
4191
4192     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4193
4194     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4195     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4196
4197     pixelshader_update_samplers(&shader->reg_maps, state->textures);
4198
4199     shader_buffer_clear(buffer);
4200     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4201     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4202     *np2fixup_info = np2fixup;
4203
4204     return ret;
4205 }
4206
4207 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4208                                  const DWORD use_map) {
4209     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4210     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4211     return stored->fog_src == new->fog_src;
4212 }
4213
4214 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4215         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4216         const struct vs_compile_args *args)
4217 {
4218     UINT i;
4219     DWORD new_size;
4220     struct glsl_vs_compiled_shader *new_array;
4221     DWORD use_map = shader->device->strided_streams.use_map;
4222     struct glsl_vshader_private *shader_data;
4223     GLhandleARB ret;
4224
4225     if (!shader->backend_data)
4226     {
4227         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4228         if (!shader->backend_data)
4229         {
4230             ERR("Failed to allocate backend data.\n");
4231             return 0;
4232         }
4233     }
4234     shader_data = shader->backend_data;
4235
4236     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4237      * so a linear search is more performant than a hashmap or a binary search
4238      * (cache coherency etc)
4239      */
4240     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4241         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4242             return shader_data->gl_shaders[i].prgId;
4243         }
4244     }
4245
4246     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4247
4248     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4249         if (shader_data->num_gl_shaders)
4250         {
4251             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4252             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4253                                     new_size * sizeof(*shader_data->gl_shaders));
4254         } else {
4255             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4256             new_size = 1;
4257         }
4258
4259         if(!new_array) {
4260             ERR("Out of memory\n");
4261             return 0;
4262         }
4263         shader_data->gl_shaders = new_array;
4264         shader_data->shader_array_size = new_size;
4265     }
4266
4267     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4268
4269     shader_buffer_clear(buffer);
4270     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4271     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4272
4273     return ret;
4274 }
4275
4276 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4277  * It sets the programId on the current StateBlock (because it should be called
4278  * inside of the DrawPrimitive() part of the render loop).
4279  *
4280  * If a program for the given combination does not exist, create one, and store
4281  * the program in the hash table.  If it creates a program, it will link the
4282  * given objects, too.
4283  */
4284
4285 /* GL locking is done by the caller */
4286 static void set_glsl_shader_program(const struct wined3d_context *context,
4287         struct wined3d_device *device, BOOL use_ps, BOOL use_vs)
4288 {
4289     const struct wined3d_state *state = &device->stateBlock->state;
4290     struct wined3d_shader *vshader = use_vs ? state->vertex_shader : NULL;
4291     struct wined3d_shader *pshader = use_ps ? state->pixel_shader : NULL;
4292     const struct wined3d_gl_info *gl_info = context->gl_info;
4293     struct shader_glsl_priv *priv = device->shader_priv;
4294     struct glsl_shader_prog_link *entry    = NULL;
4295     GLhandleARB programId                  = 0;
4296     GLhandleARB reorder_shader_id          = 0;
4297     unsigned int i;
4298     char glsl_name[8];
4299     struct ps_compile_args ps_compile_args;
4300     struct vs_compile_args vs_compile_args;
4301
4302     if (vshader) find_vs_compile_args(state, vshader, &vs_compile_args);
4303     if (pshader) find_ps_compile_args(state, pshader, &ps_compile_args);
4304
4305     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4306     if (entry)
4307     {
4308         priv->glsl_program = entry;
4309         return;
4310     }
4311
4312     /* If we get to this point, then no matching program exists, so we create one */
4313     programId = GL_EXTCALL(glCreateProgramObjectARB());
4314     TRACE("Created new GLSL shader program %u\n", programId);
4315
4316     /* Create the entry */
4317     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4318     entry->programId = programId;
4319     entry->vshader = vshader;
4320     entry->pshader = pshader;
4321     entry->vs_args = vs_compile_args;
4322     entry->ps_args = ps_compile_args;
4323     entry->constant_version = 0;
4324     entry->np2Fixup_info = NULL;
4325     /* Add the hash table entry */
4326     add_glsl_program_entry(priv, entry);
4327
4328     /* Set the current program */
4329     priv->glsl_program = entry;
4330
4331     /* Attach GLSL vshader */
4332     if (vshader)
4333     {
4334         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer, vshader, &vs_compile_args);
4335         WORD map = vshader->reg_maps.input_registers;
4336         char tmp_name[10];
4337
4338         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4339         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4340         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4341         checkGLcall("glAttachObjectARB");
4342         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4343          * is destroyed
4344          */
4345         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4346
4347         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4348         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4349         checkGLcall("glAttachObjectARB");
4350
4351         /* Bind vertex attributes to a corresponding index number to match
4352          * the same index numbers as ARB_vertex_programs (makes loading
4353          * vertex attributes simpler).  With this method, we can use the
4354          * exact same code to load the attributes later for both ARB and
4355          * GLSL shaders.
4356          *
4357          * We have to do this here because we need to know the Program ID
4358          * in order to make the bindings work, and it has to be done prior
4359          * to linking the GLSL program. */
4360         for (i = 0; map; map >>= 1, ++i)
4361         {
4362             if (!(map & 1)) continue;
4363
4364             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4365             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4366         }
4367         checkGLcall("glBindAttribLocationARB");
4368
4369         list_add_head(&vshader->linked_programs, &entry->vshader_entry);
4370     }
4371
4372     /* Attach GLSL pshader */
4373     if (pshader)
4374     {
4375         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4376                 pshader, &ps_compile_args, &entry->np2Fixup_info);
4377         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4378         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4379         checkGLcall("glAttachObjectARB");
4380
4381         list_add_head(&pshader->linked_programs, &entry->pshader_entry);
4382     }
4383
4384     /* Link the program */
4385     TRACE("Linking GLSL shader program %u\n", programId);
4386     GL_EXTCALL(glLinkProgramARB(programId));
4387     shader_glsl_validate_link(gl_info, programId);
4388
4389     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4390             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4391     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4392     {
4393         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4394         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4395     }
4396     for (i = 0; i < MAX_CONST_I; ++i)
4397     {
4398         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4399         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4400     }
4401     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4402             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4403     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4404     {
4405         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4406         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4407     }
4408     for (i = 0; i < MAX_CONST_I; ++i)
4409     {
4410         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4411         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4412     }
4413
4414     if(pshader) {
4415         char name[32];
4416
4417         for(i = 0; i < MAX_TEXTURES; i++) {
4418             sprintf(name, "bumpenvmat%u", i);
4419             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4420             sprintf(name, "luminancescale%u", i);
4421             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4422             sprintf(name, "luminanceoffset%u", i);
4423             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4424         }
4425
4426         if (ps_compile_args.np2_fixup) {
4427             if (entry->np2Fixup_info) {
4428                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4429             } else {
4430                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4431             }
4432         }
4433     }
4434
4435     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4436     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4437     checkGLcall("Find glsl program uniform locations");
4438
4439     if (pshader && pshader->reg_maps.shader_version.major >= 3
4440             && pshader->u.ps.declared_in_count > vec4_varyings(3, gl_info))
4441     {
4442         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4443         entry->vertex_color_clamp = GL_FALSE;
4444     } else {
4445         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4446     }
4447
4448     /* Set the shader to allow uniform loading on it */
4449     GL_EXTCALL(glUseProgramObjectARB(programId));
4450     checkGLcall("glUseProgramObjectARB(programId)");
4451
4452     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4453      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4454      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4455      * vertex shader with fixed function pixel processing is used we make sure that the card
4456      * supports enough samplers to allow the max number of vertex samplers with all possible
4457      * fixed function fragment processing setups. So once the program is linked these samplers
4458      * won't change.
4459      */
4460     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4461     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4462
4463     /* If the local constants do not have to be loaded with the environment constants,
4464      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4465      * later
4466      */
4467     if (pshader && !pshader->load_local_constsF)
4468         hardcode_local_constants(pshader, gl_info, programId, 'P');
4469     if (vshader && !vshader->load_local_constsF)
4470         hardcode_local_constants(vshader, gl_info, programId, 'V');
4471 }
4472
4473 /* GL locking is done by the caller */
4474 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4475 {
4476     GLhandleARB program_id;
4477     GLhandleARB vshader_id, pshader_id;
4478     const char *blt_pshader;
4479
4480     static const char *blt_vshader =
4481         "#version 120\n"
4482         "void main(void)\n"
4483         "{\n"
4484         "    gl_Position = gl_Vertex;\n"
4485         "    gl_FrontColor = vec4(1.0);\n"
4486         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4487         "}\n";
4488
4489     static const char * const blt_pshaders_full[tex_type_count] =
4490     {
4491         /* tex_1d */
4492         NULL,
4493         /* tex_2d */
4494         "#version 120\n"
4495         "uniform sampler2D sampler;\n"
4496         "void main(void)\n"
4497         "{\n"
4498         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4499         "}\n",
4500         /* tex_3d */
4501         NULL,
4502         /* tex_cube */
4503         "#version 120\n"
4504         "uniform samplerCube sampler;\n"
4505         "void main(void)\n"
4506         "{\n"
4507         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4508         "}\n",
4509         /* tex_rect */
4510         "#version 120\n"
4511         "#extension GL_ARB_texture_rectangle : enable\n"
4512         "uniform sampler2DRect sampler;\n"
4513         "void main(void)\n"
4514         "{\n"
4515         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4516         "}\n",
4517     };
4518
4519     static const char * const blt_pshaders_masked[tex_type_count] =
4520     {
4521         /* tex_1d */
4522         NULL,
4523         /* tex_2d */
4524         "#version 120\n"
4525         "uniform sampler2D sampler;\n"
4526         "uniform vec4 mask;\n"
4527         "void main(void)\n"
4528         "{\n"
4529         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4530         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4531         "}\n",
4532         /* tex_3d */
4533         NULL,
4534         /* tex_cube */
4535         "#version 120\n"
4536         "uniform samplerCube sampler;\n"
4537         "uniform vec4 mask;\n"
4538         "void main(void)\n"
4539         "{\n"
4540         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4541         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4542         "}\n",
4543         /* tex_rect */
4544         "#version 120\n"
4545         "#extension GL_ARB_texture_rectangle : enable\n"
4546         "uniform sampler2DRect sampler;\n"
4547         "uniform vec4 mask;\n"
4548         "void main(void)\n"
4549         "{\n"
4550         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4551         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4552         "}\n",
4553     };
4554
4555     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4556     if (!blt_pshader)
4557     {
4558         FIXME("tex_type %#x not supported\n", tex_type);
4559         return 0;
4560     }
4561
4562     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4563     shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4564
4565     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4566     shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4567
4568     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4569     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4570     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4571     GL_EXTCALL(glLinkProgramARB(program_id));
4572
4573     shader_glsl_validate_link(gl_info, program_id);
4574
4575     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4576      * is destroyed
4577      */
4578     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4579     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4580     return program_id;
4581 }
4582
4583 /* GL locking is done by the caller */
4584 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4585 {
4586     const struct wined3d_gl_info *gl_info = context->gl_info;
4587     struct wined3d_device *device = context->swapchain->device;
4588     struct shader_glsl_priv *priv = device->shader_priv;
4589     GLhandleARB program_id = 0;
4590     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4591
4592     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4593
4594     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4595     else priv->glsl_program = NULL;
4596
4597     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4598
4599     if (old_vertex_color_clamp != current_vertex_color_clamp)
4600     {
4601         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4602         {
4603             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4604             checkGLcall("glClampColorARB");
4605         }
4606         else
4607         {
4608             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4609         }
4610     }
4611
4612     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4613     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4614     GL_EXTCALL(glUseProgramObjectARB(program_id));
4615     checkGLcall("glUseProgramObjectARB");
4616
4617     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4618      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4619      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4620     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4621     {
4622         shader_glsl_load_np2fixup_constants(priv, gl_info, &device->stateBlock->state);
4623     }
4624 }
4625
4626 /* GL locking is done by the caller */
4627 static void shader_glsl_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4628         enum tex_types tex_type, const SIZE *ds_mask_size)
4629 {
4630     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4631     struct shader_glsl_priv *priv = shader_priv;
4632     GLhandleARB *blt_program;
4633     GLint loc;
4634
4635     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4636     if (!*blt_program)
4637     {
4638         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4639         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4640         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4641         GL_EXTCALL(glUniform1iARB(loc, 0));
4642     }
4643     else
4644     {
4645         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4646     }
4647
4648     if (masked)
4649     {
4650         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4651         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4652     }
4653 }
4654
4655 /* GL locking is done by the caller */
4656 static void shader_glsl_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4657 {
4658     struct shader_glsl_priv *priv = shader_priv;
4659     GLhandleARB program_id;
4660
4661     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4662     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4663
4664     GL_EXTCALL(glUseProgramObjectARB(program_id));
4665     checkGLcall("glUseProgramObjectARB");
4666 }
4667
4668 static void shader_glsl_destroy(struct wined3d_shader *shader)
4669 {
4670     struct wined3d_device *device = shader->device;
4671     struct shader_glsl_priv *priv = device->shader_priv;
4672     const struct wined3d_gl_info *gl_info;
4673     const struct list *linked_programs;
4674     struct wined3d_context *context;
4675
4676     char pshader = shader_is_pshader_version(shader->reg_maps.shader_version.type);
4677
4678     if (pshader)
4679     {
4680         struct glsl_pshader_private *shader_data = shader->backend_data;
4681
4682         if (!shader_data || !shader_data->num_gl_shaders)
4683         {
4684             HeapFree(GetProcessHeap(), 0, shader_data);
4685             shader->backend_data = NULL;
4686             return;
4687         }
4688
4689         context = context_acquire(device, NULL);
4690         gl_info = context->gl_info;
4691
4692         if (priv->glsl_program && priv->glsl_program->pshader == shader)
4693         {
4694             ENTER_GL();
4695             shader_glsl_select(context, FALSE, FALSE);
4696             LEAVE_GL();
4697         }
4698     }
4699     else
4700     {
4701         struct glsl_vshader_private *shader_data = shader->backend_data;
4702
4703         if (!shader_data || !shader_data->num_gl_shaders)
4704         {
4705             HeapFree(GetProcessHeap(), 0, shader_data);
4706             shader->backend_data = NULL;
4707             return;
4708         }
4709
4710         context = context_acquire(device, NULL);
4711         gl_info = context->gl_info;
4712
4713         if (priv->glsl_program && priv->glsl_program->vshader == shader)
4714         {
4715             ENTER_GL();
4716             shader_glsl_select(context, FALSE, FALSE);
4717             LEAVE_GL();
4718         }
4719     }
4720
4721     linked_programs = &shader->linked_programs;
4722
4723     TRACE("Deleting linked programs\n");
4724     if (linked_programs->next) {
4725         struct glsl_shader_prog_link *entry, *entry2;
4726
4727         ENTER_GL();
4728         if(pshader) {
4729             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4730                 delete_glsl_program_entry(priv, gl_info, entry);
4731             }
4732         } else {
4733             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4734                 delete_glsl_program_entry(priv, gl_info, entry);
4735             }
4736         }
4737         LEAVE_GL();
4738     }
4739
4740     if (pshader)
4741     {
4742         struct glsl_pshader_private *shader_data = shader->backend_data;
4743         UINT i;
4744
4745         ENTER_GL();
4746         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4747             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4748             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4749             checkGLcall("glDeleteObjectARB");
4750         }
4751         LEAVE_GL();
4752         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4753     }
4754     else
4755     {
4756         struct glsl_vshader_private *shader_data = shader->backend_data;
4757         UINT i;
4758
4759         ENTER_GL();
4760         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4761             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4762             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4763             checkGLcall("glDeleteObjectARB");
4764         }
4765         LEAVE_GL();
4766         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4767     }
4768
4769     HeapFree(GetProcessHeap(), 0, shader->backend_data);
4770     shader->backend_data = NULL;
4771
4772     context_release(context);
4773 }
4774
4775 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4776 {
4777     const struct glsl_program_key *k = key;
4778     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4779             const struct glsl_shader_prog_link, program_lookup_entry);
4780     int cmp;
4781
4782     if (k->vshader > prog->vshader) return 1;
4783     else if (k->vshader < prog->vshader) return -1;
4784
4785     if (k->pshader > prog->pshader) return 1;
4786     else if (k->pshader < prog->pshader) return -1;
4787
4788     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4789     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4790
4791     return 0;
4792 }
4793
4794 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4795 {
4796     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4797     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4798
4799     if (!mem)
4800     {
4801         ERR("Failed to allocate memory\n");
4802         return FALSE;
4803     }
4804
4805     heap->entries = mem;
4806     heap->entries[1].version = 0;
4807     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4808     heap->size = 1;
4809
4810     return TRUE;
4811 }
4812
4813 static void constant_heap_free(struct constant_heap *heap)
4814 {
4815     HeapFree(GetProcessHeap(), 0, heap->entries);
4816 }
4817
4818 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4819 {
4820     wined3d_rb_alloc,
4821     wined3d_rb_realloc,
4822     wined3d_rb_free,
4823     glsl_program_key_compare,
4824 };
4825
4826 static HRESULT shader_glsl_alloc(struct wined3d_device *device)
4827 {
4828     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4829     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4830     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4831             gl_info->limits.glsl_ps_float_constants)) + 1;
4832
4833     if (!shader_buffer_init(&priv->shader_buffer))
4834     {
4835         ERR("Failed to initialize shader buffer.\n");
4836         goto fail;
4837     }
4838
4839     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4840     if (!priv->stack)
4841     {
4842         ERR("Failed to allocate memory.\n");
4843         goto fail;
4844     }
4845
4846     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4847     {
4848         ERR("Failed to initialize vertex shader constant heap\n");
4849         goto fail;
4850     }
4851
4852     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4853     {
4854         ERR("Failed to initialize pixel shader constant heap\n");
4855         goto fail;
4856     }
4857
4858     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4859     {
4860         ERR("Failed to initialize rbtree.\n");
4861         goto fail;
4862     }
4863
4864     priv->next_constant_version = 1;
4865
4866     device->shader_priv = priv;
4867     return WINED3D_OK;
4868
4869 fail:
4870     constant_heap_free(&priv->pconst_heap);
4871     constant_heap_free(&priv->vconst_heap);
4872     HeapFree(GetProcessHeap(), 0, priv->stack);
4873     shader_buffer_free(&priv->shader_buffer);
4874     HeapFree(GetProcessHeap(), 0, priv);
4875     return E_OUTOFMEMORY;
4876 }
4877
4878 /* Context activation is done by the caller. */
4879 static void shader_glsl_free(struct wined3d_device *device)
4880 {
4881     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4882     struct shader_glsl_priv *priv = device->shader_priv;
4883     int i;
4884
4885     ENTER_GL();
4886     for (i = 0; i < tex_type_count; ++i)
4887     {
4888         if (priv->depth_blt_program_full[i])
4889         {
4890             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4891         }
4892         if (priv->depth_blt_program_masked[i])
4893         {
4894             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4895         }
4896     }
4897     LEAVE_GL();
4898
4899     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4900     constant_heap_free(&priv->pconst_heap);
4901     constant_heap_free(&priv->vconst_heap);
4902     HeapFree(GetProcessHeap(), 0, priv->stack);
4903     shader_buffer_free(&priv->shader_buffer);
4904
4905     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4906     device->shader_priv = NULL;
4907 }
4908
4909 static void shader_glsl_context_destroyed(void *shader_priv, const struct wined3d_context *context) {}
4910
4911 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4912 {
4913     if (gl_info->supported[EXT_GPU_SHADER4] && gl_info->supported[ARB_GEOMETRY_SHADER4]
4914             && gl_info->glsl_version >= MAKEDWORD_VERSION(1, 50))
4915     {
4916         caps->VertexShaderVersion = 4;
4917         caps->PixelShaderVersion = 4;
4918     }
4919     /* ARB_shader_texture_lod or EXT_gpu_shader4 is required for the SM3
4920      * texldd and texldl instructions. */
4921     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] || gl_info->supported[EXT_GPU_SHADER4])
4922     {
4923         caps->VertexShaderVersion = 3;
4924         caps->PixelShaderVersion = 3;
4925     }
4926     else
4927     {
4928         caps->VertexShaderVersion = 2;
4929         caps->PixelShaderVersion = 2;
4930     }
4931
4932     caps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4933     caps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4934
4935     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4936      * Direct3D minimum requirement.
4937      *
4938      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4939      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4940      *
4941      * The problem is that the refrast clamps temporary results in the shader to
4942      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4943      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4944      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4945      * offer a way to query this.
4946      */
4947     caps->PixelShader1xMaxValue = 8.0;
4948
4949     caps->VSClipping = TRUE;
4950
4951     TRACE_(d3d_caps)("Hardware vertex shader version %u enabled (GLSL).\n",
4952             caps->VertexShaderVersion);
4953     TRACE_(d3d_caps)("Hardware pixel shader version %u enabled (GLSL).\n",
4954             caps->PixelShaderVersion);
4955 }
4956
4957 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4958 {
4959     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4960     {
4961         TRACE("Checking support for fixup:\n");
4962         dump_color_fixup_desc(fixup);
4963     }
4964
4965     /* We support everything except YUV conversions. */
4966     if (!is_complex_fixup(fixup))
4967     {
4968         TRACE("[OK]\n");
4969         return TRUE;
4970     }
4971
4972     TRACE("[FAILED]\n");
4973     return FALSE;
4974 }
4975
4976 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4977 {
4978     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4979     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4980     /* WINED3DSIH_AND           */ NULL,
4981     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4982     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4983     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4984     /* WINED3DSIH_BREAKP        */ NULL,
4985     /* WINED3DSIH_CALL          */ shader_glsl_call,
4986     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4987     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4988     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4989     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4990     /* WINED3DSIH_CUT           */ NULL,
4991     /* WINED3DSIH_DCL           */ NULL,
4992     /* WINED3DSIH_DEF           */ NULL,
4993     /* WINED3DSIH_DEFB          */ NULL,
4994     /* WINED3DSIH_DEFI          */ NULL,
4995     /* WINED3DSIH_DIV           */ NULL,
4996     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
4997     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4998     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4999     /* WINED3DSIH_DST           */ shader_glsl_dst,
5000     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5001     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5002     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5003     /* WINED3DSIH_EMIT          */ NULL,
5004     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5005     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5006     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5007     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5008     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5009     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5010     /* WINED3DSIH_FTOI          */ NULL,
5011     /* WINED3DSIH_IADD          */ NULL,
5012     /* WINED3DSIH_IEQ           */ NULL,
5013     /* WINED3DSIH_IF            */ shader_glsl_if,
5014     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5015     /* WINED3DSIH_IGE           */ NULL,
5016     /* WINED3DSIH_IMUL          */ NULL,
5017     /* WINED3DSIH_ITOF          */ NULL,
5018     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5019     /* WINED3DSIH_LD            */ NULL,
5020     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5021     /* WINED3DSIH_LOG           */ shader_glsl_log,
5022     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5023     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5024     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5025     /* WINED3DSIH_LT            */ NULL,
5026     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5027     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5028     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5029     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5030     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5031     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5032     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5033     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5034     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5035     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5036     /* WINED3DSIH_MOVC          */ NULL,
5037     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5038     /* WINED3DSIH_NOP           */ NULL,
5039     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5040     /* WINED3DSIH_PHASE         */ NULL,
5041     /* WINED3DSIH_POW           */ shader_glsl_pow,
5042     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5043     /* WINED3DSIH_REP           */ shader_glsl_rep,
5044     /* WINED3DSIH_RET           */ shader_glsl_ret,
5045     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5046     /* WINED3DSIH_SAMPLE        */ NULL,
5047     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5048     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5049     /* WINED3DSIH_SETP          */ NULL,
5050     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5051     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5052     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5053     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5054     /* WINED3DSIH_SQRT          */ NULL,
5055     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5056     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5057     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5058     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5059     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5060     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5061     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5062     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5063     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5064     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5065     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5066     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5067     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5068     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5069     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5070     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5071     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5072     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5073     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5074     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5075     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5076     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5077     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5078     /* WINED3DSIH_UTOF          */ NULL,
5079 };
5080
5081 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5082     SHADER_HANDLER hw_fct;
5083
5084     /* Select handler */
5085     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5086
5087     /* Unhandled opcode */
5088     if (!hw_fct)
5089     {
5090         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5091         return;
5092     }
5093     hw_fct(ins);
5094
5095     shader_glsl_add_instruction_modifiers(ins);
5096 }
5097
5098 const struct wined3d_shader_backend_ops glsl_shader_backend =
5099 {
5100     shader_glsl_handle_instruction,
5101     shader_glsl_select,
5102     shader_glsl_select_depth_blt,
5103     shader_glsl_deselect_depth_blt,
5104     shader_glsl_update_float_vertex_constants,
5105     shader_glsl_update_float_pixel_constants,
5106     shader_glsl_load_constants,
5107     shader_glsl_load_np2fixup_constants,
5108     shader_glsl_destroy,
5109     shader_glsl_alloc,
5110     shader_glsl_free,
5111     shader_glsl_context_destroyed,
5112     shader_glsl_get_caps,
5113     shader_glsl_color_fixup_supported,
5114 };