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     const struct wined3d_state *state = &stateBlock->state;
 718     struct shader_glsl_priv *priv = device->shader_priv;
 719     float position_fixup[4];
 720
 721     GLhandleARB programId;
 722     struct glsl_shader_prog_link *prog = priv->glsl_program;
 723     UINT constant_version;
 724     int i;
 725
 726     if (!prog) {
 727         /* No GLSL program set - nothing to do. */
 728         return;
 729     }
 730     programId = prog->programId;
 731     constant_version = prog->constant_version;
 732
 733     if (useVertexShader)
 734     {
 735         const struct wined3d_shader *vshader = state->vertex_shader;
 736
 737         /* Load DirectX 9 float constants/uniforms for vertex shader */
 738         shader_glsl_load_constantsF(vshader, gl_info, state->vs_consts_f,
 739                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 740
 741         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 742         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, state->vs_consts_i,
 743                 stateBlock->changed.vertexShaderConstantsI & vshader->reg_maps.integer_constants);
 744
 745         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 746         shader_glsl_load_constantsB(vshader, gl_info, programId, state->vs_consts_b,
 747                 stateBlock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants);
 748
 749         /* Upload the position fixup params */
 750         shader_get_position_fixup(context, state, position_fixup);
 751         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, position_fixup));
 752         checkGLcall("glUniform4fvARB");
 753     }
 754
 755     if (usePixelShader)
 756     {
 757         const struct wined3d_shader *pshader = state->pixel_shader;
 758
 759         /* Load DirectX 9 float constants/uniforms for pixel shader */
 760         shader_glsl_load_constantsF(pshader, gl_info, state->ps_consts_f,
 761                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 762
 763         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 764         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, state->ps_consts_i,
 765                 stateBlock->changed.pixelShaderConstantsI & pshader->reg_maps.integer_constants);
 766
 767         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 768         shader_glsl_load_constantsB(pshader, gl_info, programId, state->ps_consts_b,
 769                 stateBlock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants);
 770
 771         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 772          * It can't be 0 for a valid texbem instruction.
 773          */
 774         for(i = 0; i < MAX_TEXTURES; i++) {
 775             const float *data;
 776
 777             if(prog->bumpenvmat_location[i] == -1) continue;
 778
 779             data = (const float *)&state->texture_states[i][WINED3DTSS_BUMPENVMAT00];
 780             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 781             checkGLcall("glUniformMatrix2fvARB");
 782
 783             /* texbeml needs the luminance scale and offset too. If texbeml
 784              * is used, needsbumpmat is set too, so we can check that in the
 785              * needsbumpmat check. */
 786             if (prog->luminancescale_location[i] != -1)
 787             {
 788                 const GLfloat *scale = (const GLfloat *)&state->texture_states[i][WINED3DTSS_BUMPENVLSCALE];
 789                 const GLfloat *offset = (const GLfloat *)&state->texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
 790
 791                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 792                 checkGLcall("glUniform1fvARB");
 793                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 794                 checkGLcall("glUniform1fvARB");
 795             }
 796         }
 797
 798         if (prog->ycorrection_location != -1)
 799         {
 800             float correction_params[4];
 801
 802             if (context->render_offscreen)
 803             {
 804                 correction_params[0] = 0.0f;
 805                 correction_params[1] = 1.0f;
 806             } else {
 807                 /* position is window relative, not viewport relative */
 808                 correction_params[0] = (float) context->current_rt->resource.height;
 809                 correction_params[1] = -1.0f;
 810             }
 811             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 812         }
 813     }
 814
 815     if (priv->next_constant_version == UINT_MAX)
 816     {
 817         TRACE("Max constant version reached, resetting to 0.\n");
 818         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 819         priv->next_constant_version = 1;
 820     }
 821     else
 822     {
 823         prog->constant_version = priv->next_constant_version++;
 824     }
 825 }
 826
 827 static void update_heap_entry(const struct constant_heap *heap, unsigned int idx,
 828         unsigned int heap_idx, DWORD new_version)
 829 {
 830     struct constant_entry *entries = heap->entries;
 831     unsigned int *positions = heap->positions;
 832     unsigned int parent_idx;
 833
 834     while (heap_idx > 1)
 835     {
 836         parent_idx = heap_idx >> 1;
 837
 838         if (new_version <= entries[parent_idx].version) break;
 839
 840         entries[heap_idx] = entries[parent_idx];
 841         positions[entries[parent_idx].idx] = heap_idx;
 842         heap_idx = parent_idx;
 843     }
 844
 845     entries[heap_idx].version = new_version;
 846     entries[heap_idx].idx = idx;
 847     positions[idx] = heap_idx;
 848 }
 849
 850 static void shader_glsl_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 851 {
 852     struct shader_glsl_priv *priv = device->shader_priv;
 853     struct constant_heap *heap = &priv->vconst_heap;
 854     UINT i;
 855
 856     for (i = start; i < count + start; ++i)
 857     {
 858         if (!device->stateBlock->changed.vertexShaderConstantsF[i])
 859             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 860         else
 861             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 862     }
 863 }
 864
 865 static void shader_glsl_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 866 {
 867     struct shader_glsl_priv *priv = device->shader_priv;
 868     struct constant_heap *heap = &priv->pconst_heap;
 869     UINT i;
 870
 871     for (i = start; i < count + start; ++i)
 872     {
 873         if (!device->stateBlock->changed.pixelShaderConstantsF[i])
 874             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 875         else
 876             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 877     }
 878 }
 879
 880 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 881 {
 882     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 883     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 884     if(shader_major > 3) return ret;
 885
 886     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 887     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 888     return ret;
 889 }
 890
 891 /** Generate the variable & register declarations for the GLSL output target */
 892 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 893         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
 894         const struct wined3d_shader_reg_maps *reg_maps, const struct shader_glsl_ctx_priv *ctx_priv)
 895 {
 896     const struct wined3d_state *state = &shader->device->stateBlock->state;
 897     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 898     const struct wined3d_gl_info *gl_info = context->gl_info;
 899     const struct wined3d_fb_state *fb = &shader->device->fb;
 900     unsigned int i, extra_constants_needed = 0;
 901     const struct wined3d_shader_lconst *lconst;
 902     DWORD map;
 903
 904     /* There are some minor differences between pixel and vertex shaders */
 905     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 906     char prefix = pshader ? 'P' : 'V';
 907
 908     /* Prototype the subroutines */
 909     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 910     {
 911         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 912     }
 913
 914     /* Declare the constants (aka uniforms) */
 915     if (shader->limits.constant_float > 0)
 916     {
 917         unsigned max_constantsF;
 918         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 919          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 920          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 921          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 922          * a dx9 card, as long as it doesn't also use all the other constants.
 923          *
 924          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 925          * declare only the amount that we're assured to have.
 926          *
 927          * Thus we run into problems in these two cases:
 928          * 1) The shader really uses more uniforms than supported
 929          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 930          */
 931         if (pshader)
 932         {
 933             /* No indirect addressing here. */
 934             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 935         }
 936         else
 937         {
 938             if (reg_maps->usesrelconstF)
 939             {
 940                 /* Subtract the other potential uniforms from the max
 941                  * available (bools, ints, and 1 row of projection matrix).
 942                  * Subtract another uniform for immediate values, which have
 943                  * to be loaded via uniform by the driver as well. The shader
 944                  * code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex
 945                  * shader code, so one vec4 should be enough. (Unfortunately
 946                  * the Nvidia driver doesn't store 128 and -128 in one float).
 947                  *
 948                  * Writing gl_ClipVertex requires one uniform for each
 949                  * clipplane as well. */
 950                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 951                 if(ctx_priv->cur_vs_args->clip_enabled)
 952                 {
 953                     max_constantsF -= gl_info->limits.clipplanes;
 954                 }
 955                 max_constantsF -= count_bits(reg_maps->integer_constants);
 956                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 957                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 958                  * for now take this into account when calculating the number of available constants
 959                  */
 960                 max_constantsF -= count_bits(reg_maps->boolean_constants);
 961                 /* Set by driver quirks in directx.c */
 962                 max_constantsF -= gl_info->reserved_glsl_constants;
 963             }
 964             else
 965             {
 966                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 967             }
 968         }
 969         max_constantsF = min(shader->limits.constant_float, max_constantsF);
 970         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 971     }
 972
 973     /* Always declare the full set of constants, the compiler can remove the
 974      * unused ones because d3d doesn't (yet) support indirect int and bool
 975      * constant addressing. This avoids problems if the app uses e.g. i0 and i9. */
 976     if (shader->limits.constant_int > 0 && reg_maps->integer_constants)
 977         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, shader->limits.constant_int);
 978
 979     if (shader->limits.constant_bool > 0 && reg_maps->boolean_constants)
 980         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, shader->limits.constant_bool);
 981
 982     if (!pshader)
 983     {
 984         shader_addline(buffer, "uniform vec4 posFixup;\n");
 985         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 986     }
 987     else
 988     {
 989         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
 990         {
 991             if (!(map & 1)) continue;
 992
 993             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
 994
 995             if (reg_maps->luminanceparams & (1 << i))
 996             {
 997                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
 998                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
 999                 extra_constants_needed++;
1000             }
1001
1002             extra_constants_needed++;
1003         }
1004
1005         if (ps_args->srgb_correction)
1006         {
1007             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1008                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1009             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1010                     srgb_cmp);
1011         }
1012         if (reg_maps->vpos || reg_maps->usesdsy)
1013         {
1014             if (shader->limits.constant_float + extra_constants_needed
1015                     + 1 < gl_info->limits.glsl_ps_float_constants)
1016             {
1017                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1018                 extra_constants_needed++;
1019             }
1020             else
1021             {
1022                 /* This happens because we do not have proper tracking of the constant registers that are
1023                  * actually used, only the max limit of the shader version
1024                  */
1025                 FIXME("Cannot find a free uniform for vpos correction params\n");
1026                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1027                         context->render_offscreen ? 0.0f : fb->render_targets[0]->resource.height,
1028                         context->render_offscreen ? 1.0f : -1.0f);
1029             }
1030             shader_addline(buffer, "vec4 vpos;\n");
1031         }
1032     }
1033
1034     /* Declare texture samplers */
1035     for (i = 0; i < shader->limits.sampler; ++i)
1036     {
1037         if (reg_maps->sampler_type[i])
1038         {
1039             const struct wined3d_texture *texture;
1040
1041             switch (reg_maps->sampler_type[i])
1042             {
1043                 case WINED3DSTT_1D:
1044                     if (pshader && ps_args->shadow & (1 << i))
1045                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1046                     else
1047                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1048                     break;
1049                 case WINED3DSTT_2D:
1050                     texture = state->textures[i];
1051                     if (pshader && ps_args->shadow & (1 << i))
1052                     {
1053                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1054                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1055                         else
1056                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1057                     }
1058                     else
1059                     {
1060                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1061                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1062                         else
1063                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1064                     }
1065                     break;
1066                 case WINED3DSTT_CUBE:
1067                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1068                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1069                     break;
1070                 case WINED3DSTT_VOLUME:
1071                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1072                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1073                     break;
1074                 default:
1075                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1076                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1077                     break;
1078             }
1079         }
1080     }
1081
1082     /* Declare uniforms for NP2 texcoord fixup:
1083      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1084      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1085      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1086     if (pshader && ps_args->np2_fixup) {
1087
1088         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1089         UINT cur = 0;
1090
1091         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1092          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1093          * samplerNP2Fixup stores texture dimensions and is updated through
1094          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1095
1096         for (i = 0; i < shader->limits.sampler; ++i)
1097         {
1098             if (reg_maps->sampler_type[i])
1099             {
1100                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1101
1102                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1103                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1104                     continue;
1105                 }
1106
1107                 fixup->idx[i] = cur++;
1108             }
1109         }
1110
1111         fixup->num_consts = (cur + 1) >> 1;
1112         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1113     }
1114
1115     /* Declare address variables */
1116     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1117     {
1118         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1119     }
1120
1121     /* Declare texture coordinate temporaries and initialize them */
1122     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1123     {
1124         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1125     }
1126
1127     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1128      * helper function shader that is linked in at link time
1129      */
1130     if (pshader && reg_maps->shader_version.major >= 3)
1131     {
1132         UINT in_count = min(vec4_varyings(reg_maps->shader_version.major, gl_info), shader->limits.packed_input);
1133
1134         if (use_vs(state))
1135             shader_addline(buffer, "varying vec4 IN[%u];\n", in_count);
1136         else
1137             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1138              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1139              * pixel shader that reads the fixed function color into the packed input registers. */
1140             shader_addline(buffer, "vec4 IN[%u];\n", in_count);
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(enum wined3d_shader_src_modifier src_modifier,
1237         const char *in_reg, const char *in_regswizzle, char *out_str)
1238 {
1239     out_str[0] = 0;
1240
1241     switch (src_modifier)
1242     {
1243     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1244     case WINED3DSPSM_DW:
1245     case WINED3DSPSM_NONE:
1246         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1247         break;
1248     case WINED3DSPSM_NEG:
1249         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1250         break;
1251     case WINED3DSPSM_NOT:
1252         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1253         break;
1254     case WINED3DSPSM_BIAS:
1255         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1256         break;
1257     case WINED3DSPSM_BIASNEG:
1258         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1259         break;
1260     case WINED3DSPSM_SIGN:
1261         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1262         break;
1263     case WINED3DSPSM_SIGNNEG:
1264         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1265         break;
1266     case WINED3DSPSM_COMP:
1267         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1268         break;
1269     case WINED3DSPSM_X2:
1270         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1271         break;
1272     case WINED3DSPSM_X2NEG:
1273         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1274         break;
1275     case WINED3DSPSM_ABS:
1276         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1277         break;
1278     case WINED3DSPSM_ABSNEG:
1279         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1280         break;
1281     default:
1282         FIXME("Unhandled modifier %u\n", src_modifier);
1283         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1284     }
1285 }
1286
1287 /** Writes the GLSL variable name that corresponds to the register that the
1288  * DX opcode parameter is trying to access */
1289 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1290         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1291 {
1292     /* oPos, oFog and oPts in D3D */
1293     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1294
1295     const struct wined3d_shader *shader = ins->ctx->shader;
1296     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1297     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1298     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1299
1300     *is_color = FALSE;
1301
1302     switch (reg->type)
1303     {
1304         case WINED3DSPR_TEMP:
1305             sprintf(register_name, "R%u", reg->idx);
1306             break;
1307
1308         case WINED3DSPR_INPUT:
1309             /* vertex shaders */
1310             if (!pshader)
1311             {
1312                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1313                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1314                 sprintf(register_name, "attrib%u", reg->idx);
1315                 break;
1316             }
1317
1318             /* pixel shaders >= 3.0 */
1319             if (reg_maps->shader_version.major >= 3)
1320             {
1321                 DWORD idx = shader->u.ps.input_reg_map[reg->idx];
1322                 unsigned int in_count = vec4_varyings(reg_maps->shader_version.major, gl_info);
1323
1324                 if (reg->rel_addr)
1325                 {
1326                     struct glsl_src_param rel_param;
1327
1328                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1329
1330                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1331                      * operation there */
1332                     if (idx)
1333                     {
1334                         if (shader->u.ps.declared_in_count > in_count)
1335                         {
1336                             sprintf(register_name,
1337                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1338                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1339                                     rel_param.param_str, idx);
1340                         }
1341                         else
1342                         {
1343                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1344                         }
1345                     }
1346                     else
1347                     {
1348                         if (shader->u.ps.declared_in_count > in_count)
1349                         {
1350                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1351                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1352                                     rel_param.param_str);
1353                         }
1354                         else
1355                         {
1356                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1357                         }
1358                     }
1359                 }
1360                 else
1361                 {
1362                     if (idx == in_count) sprintf(register_name, "gl_Color");
1363                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1364                     else sprintf(register_name, "IN[%u]", idx);
1365                 }
1366             }
1367             else
1368             {
1369                 if (!reg->idx) strcpy(register_name, "gl_Color");
1370                 else strcpy(register_name, "gl_SecondaryColor");
1371                 break;
1372             }
1373             break;
1374
1375         case WINED3DSPR_CONST:
1376             {
1377                 const char prefix = pshader ? 'P' : 'V';
1378
1379                 /* Relative addressing */
1380                 if (reg->rel_addr)
1381                 {
1382                     struct glsl_src_param rel_param;
1383                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1384                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1385                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1386                 }
1387                 else
1388                 {
1389                     if (shader_constant_is_local(shader, reg->idx))
1390                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1391                     else
1392                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1393                 }
1394             }
1395             break;
1396
1397         case WINED3DSPR_CONSTINT:
1398             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1399             else sprintf(register_name, "VI[%u]", reg->idx);
1400             break;
1401
1402         case WINED3DSPR_CONSTBOOL:
1403             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1404             else sprintf(register_name, "VB[%u]", reg->idx);
1405             break;
1406
1407         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1408             if (pshader) sprintf(register_name, "T%u", reg->idx);
1409             else sprintf(register_name, "A%u", reg->idx);
1410             break;
1411
1412         case WINED3DSPR_LOOP:
1413             sprintf(register_name, "aL%u", ins->ctx->loop_state->current_reg - 1);
1414             break;
1415
1416         case WINED3DSPR_SAMPLER:
1417             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1418             else sprintf(register_name, "Vsampler%u", reg->idx);
1419             break;
1420
1421         case WINED3DSPR_COLOROUT:
1422             if (reg->idx >= gl_info->limits.buffers)
1423                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1424
1425             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1426             break;
1427
1428         case WINED3DSPR_RASTOUT:
1429             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1430             break;
1431
1432         case WINED3DSPR_DEPTHOUT:
1433             sprintf(register_name, "gl_FragDepth");
1434             break;
1435
1436         case WINED3DSPR_ATTROUT:
1437             if (!reg->idx) sprintf(register_name, "OUT[8]");
1438             else sprintf(register_name, "OUT[9]");
1439             break;
1440
1441         case WINED3DSPR_TEXCRDOUT:
1442             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1443             sprintf(register_name, "OUT[%u]", reg->idx);
1444             break;
1445
1446         case WINED3DSPR_MISCTYPE:
1447             if (!reg->idx)
1448             {
1449                 /* vPos */
1450                 sprintf(register_name, "vpos");
1451             }
1452             else if (reg->idx == 1)
1453             {
1454                 /* Note that gl_FrontFacing is a bool, while vFace is
1455                  * a float for which the sign determines front/back */
1456                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1457             }
1458             else
1459             {
1460                 FIXME("Unhandled misctype register %d\n", reg->idx);
1461                 sprintf(register_name, "unrecognized_register");
1462             }
1463             break;
1464
1465         case WINED3DSPR_IMMCONST:
1466             switch (reg->immconst_type)
1467             {
1468                 case WINED3D_IMMCONST_SCALAR:
1469                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1470                     break;
1471
1472                 case WINED3D_IMMCONST_VEC4:
1473                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1474                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1475                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1476                     break;
1477
1478                 default:
1479                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1480                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1481             }
1482             break;
1483
1484         default:
1485             FIXME("Unhandled register name Type(%d)\n", reg->type);
1486             sprintf(register_name, "unrecognized_register");
1487             break;
1488     }
1489 }
1490
1491 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1492 {
1493     *str++ = '.';
1494     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1495     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1496     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1497     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1498     *str = '\0';
1499 }
1500
1501 /* Get the GLSL write mask for the destination register */
1502 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1503 {
1504     DWORD mask = param->write_mask;
1505
1506     if (shader_is_scalar(&param->reg))
1507     {
1508         mask = WINED3DSP_WRITEMASK_0;
1509         *write_mask = '\0';
1510     }
1511     else
1512     {
1513         shader_glsl_write_mask_to_str(mask, write_mask);
1514     }
1515
1516     return mask;
1517 }
1518
1519 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1520     unsigned int size = 0;
1521
1522     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1523     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1524     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1525     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1526
1527     return size;
1528 }
1529
1530 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1531 {
1532     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1533      * but addressed as "rgba". To fix this we need to swap the register's x
1534      * and z components. */
1535     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1536
1537     *str++ = '.';
1538     /* swizzle bits fields: wwzzyyxx */
1539     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1540     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1541     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1542     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1543     *str = '\0';
1544 }
1545
1546 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1547         BOOL fixup, DWORD mask, char *swizzle_str)
1548 {
1549     if (shader_is_scalar(&param->reg))
1550         *swizzle_str = '\0';
1551     else
1552         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1553 }
1554
1555 /* From a given parameter token, generate the corresponding GLSL string.
1556  * Also, return the actual register name and swizzle in case the
1557  * caller needs this information as well. */
1558 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1559         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src)
1560 {
1561     BOOL is_color = FALSE;
1562     char swizzle_str[6];
1563
1564     glsl_src->reg_name[0] = '\0';
1565     glsl_src->param_str[0] = '\0';
1566     swizzle_str[0] = '\0';
1567
1568     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1569     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1570     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1571 }
1572
1573 /* From a given parameter token, generate the corresponding GLSL string.
1574  * Also, return the actual register name and swizzle in case the
1575  * caller needs this information as well. */
1576 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1577         const struct wined3d_shader_dst_param *wined3d_dst, struct glsl_dst_param *glsl_dst)
1578 {
1579     BOOL is_color = FALSE;
1580
1581     glsl_dst->mask_str[0] = '\0';
1582     glsl_dst->reg_name[0] = '\0';
1583
1584     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1585     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1586 }
1587
1588 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1589 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1590         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1591 {
1592     struct glsl_dst_param glsl_dst;
1593     DWORD mask;
1594
1595     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1596     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1597
1598     return mask;
1599 }
1600
1601 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1602 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1603 {
1604     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1605 }
1606
1607 /** Process GLSL instruction modifiers */
1608 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1609 {
1610     struct glsl_dst_param dst_param;
1611     DWORD modifiers;
1612
1613     if (!ins->dst_count) return;
1614
1615     modifiers = ins->dst[0].modifiers;
1616     if (!modifiers) return;
1617
1618     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1619
1620     if (modifiers & WINED3DSPDM_SATURATE)
1621     {
1622         /* _SAT means to clamp the value of the register to between 0 and 1 */
1623         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1624                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1625     }
1626
1627     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1628     {
1629         FIXME("_centroid modifier not handled\n");
1630     }
1631
1632     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1633     {
1634         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1635     }
1636 }
1637
1638 static const char *shader_glsl_get_rel_op(enum wined3d_shader_rel_op op)
1639 {
1640     switch (op)
1641     {
1642         case WINED3D_SHADER_REL_OP_GT: return ">";
1643         case WINED3D_SHADER_REL_OP_EQ: return "==";
1644         case WINED3D_SHADER_REL_OP_GE: return ">=";
1645         case WINED3D_SHADER_REL_OP_LT: return "<";
1646         case WINED3D_SHADER_REL_OP_NE: return "!=";
1647         case WINED3D_SHADER_REL_OP_LE: return "<=";
1648         default:
1649             FIXME("Unrecognized operator %#x.\n", op);
1650             return "(\?\?)";
1651     }
1652 }
1653
1654 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1655         DWORD sampler_idx, DWORD flags, struct glsl_sample_function *sample_function)
1656 {
1657     enum wined3d_sampler_texture_type sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1658     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1659     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1660             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1661     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1662     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1663     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1664     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1665
1666     /* Note that there's no such thing as a projected cube texture. */
1667     switch(sampler_type) {
1668         case WINED3DSTT_1D:
1669             if (shadow)
1670             {
1671                 if (lod)
1672                 {
1673                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1674                 }
1675                 else if (grad)
1676                 {
1677                     if (gl_info->supported[EXT_GPU_SHADER4])
1678                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1679                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1680                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1681                     else
1682                     {
1683                         FIXME("Unsupported 1D shadow grad function.\n");
1684                         sample_function->name = "unsupported1DGrad";
1685                     }
1686                 }
1687                 else
1688                 {
1689                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1690                 }
1691                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1692             }
1693             else
1694             {
1695                 if (lod)
1696                 {
1697                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1698                 }
1699                 else if (grad)
1700                 {
1701                     if (gl_info->supported[EXT_GPU_SHADER4])
1702                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1703                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1704                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1705                     else
1706                     {
1707                         FIXME("Unsupported 1D grad function.\n");
1708                         sample_function->name = "unsupported1DGrad";
1709                     }
1710                 }
1711                 else
1712                 {
1713                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1714                 }
1715                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1716             }
1717             break;
1718
1719         case WINED3DSTT_2D:
1720             if (shadow)
1721             {
1722                 if (texrect)
1723                 {
1724                     if (lod)
1725                     {
1726                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1727                     }
1728                     else if (grad)
1729                     {
1730                         if (gl_info->supported[EXT_GPU_SHADER4])
1731                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1732                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1733                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1734                         else
1735                         {
1736                             FIXME("Unsupported RECT shadow grad function.\n");
1737                             sample_function->name = "unsupported2DRectGrad";
1738                         }
1739                     }
1740                     else
1741                     {
1742                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1743                     }
1744                 }
1745                 else
1746                 {
1747                     if (lod)
1748                     {
1749                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1750                     }
1751                     else if (grad)
1752                     {
1753                         if (gl_info->supported[EXT_GPU_SHADER4])
1754                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1755                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1756                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1757                         else
1758                         {
1759                             FIXME("Unsupported 2D shadow grad function.\n");
1760                             sample_function->name = "unsupported2DGrad";
1761                         }
1762                     }
1763                     else
1764                     {
1765                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1766                     }
1767                 }
1768                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1769             }
1770             else
1771             {
1772                 if (texrect)
1773                 {
1774                     if (lod)
1775                     {
1776                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1777                     }
1778                     else if (grad)
1779                     {
1780                         if (gl_info->supported[EXT_GPU_SHADER4])
1781                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1782                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1783                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1784                         else
1785                         {
1786                             FIXME("Unsupported RECT grad function.\n");
1787                             sample_function->name = "unsupported2DRectGrad";
1788                         }
1789                     }
1790                     else
1791                     {
1792                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1793                     }
1794                 }
1795                 else
1796                 {
1797                     if (lod)
1798                     {
1799                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1800                     }
1801                     else if (grad)
1802                     {
1803                         if (gl_info->supported[EXT_GPU_SHADER4])
1804                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1805                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1806                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1807                         else
1808                         {
1809                             FIXME("Unsupported 2D grad function.\n");
1810                             sample_function->name = "unsupported2DGrad";
1811                         }
1812                     }
1813                     else
1814                     {
1815                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1816                     }
1817                 }
1818                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1819             }
1820             break;
1821
1822         case WINED3DSTT_CUBE:
1823             if (shadow)
1824             {
1825                 FIXME("Unsupported Cube shadow function.\n");
1826                 sample_function->name = "unsupportedCubeShadow";
1827                 sample_function->coord_mask = 0;
1828             }
1829             else
1830             {
1831                 if (lod)
1832                 {
1833                     sample_function->name = "textureCubeLod";
1834                 }
1835                 else if (grad)
1836                 {
1837                     if (gl_info->supported[EXT_GPU_SHADER4])
1838                         sample_function->name = "textureCubeGrad";
1839                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1840                         sample_function->name = "textureCubeGradARB";
1841                     else
1842                     {
1843                         FIXME("Unsupported Cube grad function.\n");
1844                         sample_function->name = "unsupportedCubeGrad";
1845                     }
1846                 }
1847                 else
1848                 {
1849                     sample_function->name = "textureCube";
1850                 }
1851                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1852             }
1853             break;
1854
1855         case WINED3DSTT_VOLUME:
1856             if (shadow)
1857             {
1858                 FIXME("Unsupported 3D shadow function.\n");
1859                 sample_function->name = "unsupported3DShadow";
1860                 sample_function->coord_mask = 0;
1861             }
1862             else
1863             {
1864                 if (lod)
1865                 {
1866                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1867                 }
1868                 else  if (grad)
1869                 {
1870                     if (gl_info->supported[EXT_GPU_SHADER4])
1871                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1872                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1873                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1874                     else
1875                     {
1876                         FIXME("Unsupported 3D grad function.\n");
1877                         sample_function->name = "unsupported3DGrad";
1878                     }
1879                 }
1880                 else
1881                 {
1882                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1883                 }
1884                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1885             }
1886             break;
1887
1888         default:
1889             sample_function->name = "";
1890             sample_function->coord_mask = 0;
1891             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1892             break;
1893     }
1894 }
1895
1896 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1897         BOOL sign_fixup, enum fixup_channel_source channel_source)
1898 {
1899     switch(channel_source)
1900     {
1901         case CHANNEL_SOURCE_ZERO:
1902             strcat(arguments, "0.0");
1903             break;
1904
1905         case CHANNEL_SOURCE_ONE:
1906             strcat(arguments, "1.0");
1907             break;
1908
1909         case CHANNEL_SOURCE_X:
1910             strcat(arguments, reg_name);
1911             strcat(arguments, ".x");
1912             break;
1913
1914         case CHANNEL_SOURCE_Y:
1915             strcat(arguments, reg_name);
1916             strcat(arguments, ".y");
1917             break;
1918
1919         case CHANNEL_SOURCE_Z:
1920             strcat(arguments, reg_name);
1921             strcat(arguments, ".z");
1922             break;
1923
1924         case CHANNEL_SOURCE_W:
1925             strcat(arguments, reg_name);
1926             strcat(arguments, ".w");
1927             break;
1928
1929         default:
1930             FIXME("Unhandled channel source %#x\n", channel_source);
1931             strcat(arguments, "undefined");
1932             break;
1933     }
1934
1935     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1936 }
1937
1938 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1939 {
1940     struct wined3d_shader_dst_param dst;
1941     unsigned int mask_size, remaining;
1942     struct glsl_dst_param dst_param;
1943     char arguments[256];
1944     DWORD mask;
1945
1946     mask = 0;
1947     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1948     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1949     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1950     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1951     mask &= ins->dst[0].write_mask;
1952
1953     if (!mask) return; /* Nothing to do */
1954
1955     if (is_complex_fixup(fixup))
1956     {
1957         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1958         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1959         return;
1960     }
1961
1962     mask_size = shader_glsl_get_write_mask_size(mask);
1963
1964     dst = ins->dst[0];
1965     dst.write_mask = mask;
1966     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1967
1968     arguments[0] = '\0';
1969     remaining = mask_size;
1970     if (mask & WINED3DSP_WRITEMASK_0)
1971     {
1972         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1973         if (--remaining) strcat(arguments, ", ");
1974     }
1975     if (mask & WINED3DSP_WRITEMASK_1)
1976     {
1977         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1978         if (--remaining) strcat(arguments, ", ");
1979     }
1980     if (mask & WINED3DSP_WRITEMASK_2)
1981     {
1982         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1983         if (--remaining) strcat(arguments, ", ");
1984     }
1985     if (mask & WINED3DSP_WRITEMASK_3)
1986     {
1987         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1988         if (--remaining) strcat(arguments, ", ");
1989     }
1990
1991     if (mask_size > 1)
1992     {
1993         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1994                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1995     }
1996     else
1997     {
1998         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1999     }
2000 }
2001
2002 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2003         DWORD sampler, const struct glsl_sample_function *sample_function, DWORD swizzle,
2004         const char *dx, const char *dy, const char *bias, const char *coord_reg_fmt, ...)
2005 {
2006     const char *sampler_base;
2007     char dst_swizzle[6];
2008     struct color_fixup_desc fixup;
2009     BOOL np2_fixup = FALSE;
2010     va_list args;
2011
2012     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2013
2014     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2015     {
2016         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2017         fixup = priv->cur_ps_args->color_fixup[sampler];
2018         sampler_base = "Psampler";
2019
2020         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2021             if(bias) {
2022                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2023             } else {
2024                 np2_fixup = TRUE;
2025             }
2026         }
2027     } else {
2028         sampler_base = "Vsampler";
2029         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2030     }
2031
2032     shader_glsl_append_dst(ins->ctx->buffer, ins);
2033
2034     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2035
2036     va_start(args, coord_reg_fmt);
2037     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2038     va_end(args);
2039
2040     if(bias) {
2041         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2042     } else {
2043         if (np2_fixup) {
2044             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2045             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2046
2047             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2048                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2049         } else if(dx && dy) {
2050             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2051         } else {
2052             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2053         }
2054     }
2055
2056     if(!is_identity_fixup(fixup)) {
2057         shader_glsl_color_correction(ins, fixup);
2058     }
2059 }
2060
2061 /*****************************************************************************
2062  * Begin processing individual instruction opcodes
2063  ****************************************************************************/
2064
2065 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2066 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2067 {
2068     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2069     struct glsl_src_param src0_param;
2070     struct glsl_src_param src1_param;
2071     DWORD write_mask;
2072     char op;
2073
2074     /* Determine the GLSL operator to use based on the opcode */
2075     switch (ins->handler_idx)
2076     {
2077         case WINED3DSIH_MUL: op = '*'; break;
2078         case WINED3DSIH_ADD: op = '+'; break;
2079         case WINED3DSIH_SUB: op = '-'; break;
2080         default:
2081             op = ' ';
2082             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2083             break;
2084     }
2085
2086     write_mask = shader_glsl_append_dst(buffer, ins);
2087     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2088     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2089     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2090 }
2091
2092 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2093 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2094 {
2095     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2096     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2097     struct glsl_src_param src0_param;
2098     DWORD write_mask;
2099
2100     write_mask = shader_glsl_append_dst(buffer, ins);
2101     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2102
2103     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2104      * shader versions WINED3DSIO_MOVA is used for this. */
2105     if (ins->ctx->reg_maps->shader_version.major == 1
2106             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2107             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2108     {
2109         /* This is a simple floor() */
2110         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2111         if (mask_size > 1) {
2112             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2113         } else {
2114             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2115         }
2116     }
2117     else if(ins->handler_idx == WINED3DSIH_MOVA)
2118     {
2119         /* We need to *round* to the nearest int here. */
2120         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2121
2122         if (gl_info->supported[EXT_GPU_SHADER4])
2123         {
2124             if (mask_size > 1)
2125                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2126             else
2127                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2128         }
2129         else
2130         {
2131             if (mask_size > 1)
2132                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2133                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2134             else
2135                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2136                         src0_param.param_str, src0_param.param_str);
2137         }
2138     }
2139     else
2140     {
2141         shader_addline(buffer, "%s);\n", src0_param.param_str);
2142     }
2143 }
2144
2145 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2146 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2147 {
2148     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2149     struct glsl_src_param src0_param;
2150     struct glsl_src_param src1_param;
2151     DWORD dst_write_mask, src_write_mask;
2152     unsigned int dst_size = 0;
2153
2154     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2155     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2156
2157     /* dp3 works on vec3, dp4 on vec4 */
2158     if (ins->handler_idx == WINED3DSIH_DP4)
2159     {
2160         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2161     } else {
2162         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2163     }
2164
2165     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2166     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2167
2168     if (dst_size > 1) {
2169         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2170     } else {
2171         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2172     }
2173 }
2174
2175 /* Note that this instruction has some restrictions. The destination write mask
2176  * can't contain the w component, and the source swizzles have to be .xyzw */
2177 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2178 {
2179     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2180     struct glsl_src_param src0_param;
2181     struct glsl_src_param src1_param;
2182     char dst_mask[6];
2183
2184     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2185     shader_glsl_append_dst(ins->ctx->buffer, ins);
2186     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2187     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2188     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2189 }
2190
2191 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2192  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2193  * GLSL uses the value as-is. */
2194 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2195 {
2196     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2197     struct glsl_src_param src0_param;
2198     struct glsl_src_param src1_param;
2199     DWORD dst_write_mask;
2200     unsigned int dst_size;
2201
2202     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2203     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2204
2205     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2206     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2207
2208     if (dst_size > 1)
2209     {
2210         shader_addline(buffer, "vec%u(%s == 0.0 ? 1.0 : pow(abs(%s), %s)));\n",
2211                 dst_size, src1_param.param_str, src0_param.param_str, src1_param.param_str);
2212     }
2213     else
2214     {
2215         shader_addline(buffer, "%s == 0.0 ? 1.0 : pow(abs(%s), %s));\n",
2216                 src1_param.param_str, src0_param.param_str, src1_param.param_str);
2217     }
2218 }
2219
2220 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2221  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2222  * GLSL uses the value as-is. */
2223 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2224 {
2225     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2226     struct glsl_src_param src0_param;
2227     DWORD dst_write_mask;
2228     unsigned int dst_size;
2229
2230     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2231     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2232
2233     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2234
2235     if (dst_size > 1)
2236     {
2237         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2238                 dst_size, src0_param.param_str);
2239     }
2240     else
2241     {
2242         shader_addline(buffer, "log2(abs(%s)));\n",
2243                 src0_param.param_str);
2244     }
2245 }
2246
2247 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2248 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2249 {
2250     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2251     struct glsl_src_param src_param;
2252     const char *instruction;
2253     DWORD write_mask;
2254     unsigned i;
2255
2256     /* Determine the GLSL function to use based on the opcode */
2257     /* TODO: Possibly make this a table for faster lookups */
2258     switch (ins->handler_idx)
2259     {
2260         case WINED3DSIH_MIN: instruction = "min"; break;
2261         case WINED3DSIH_MAX: instruction = "max"; break;
2262         case WINED3DSIH_ABS: instruction = "abs"; break;
2263         case WINED3DSIH_FRC: instruction = "fract"; break;
2264         case WINED3DSIH_EXP: instruction = "exp2"; break;
2265         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2266         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2267         default: instruction = "";
2268             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2269             break;
2270     }
2271
2272     write_mask = shader_glsl_append_dst(buffer, ins);
2273
2274     shader_addline(buffer, "%s(", instruction);
2275
2276     if (ins->src_count)
2277     {
2278         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2279         shader_addline(buffer, "%s", src_param.param_str);
2280         for (i = 1; i < ins->src_count; ++i)
2281         {
2282             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2283             shader_addline(buffer, ", %s", src_param.param_str);
2284         }
2285     }
2286
2287     shader_addline(buffer, "));\n");
2288 }
2289
2290 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2291 {
2292     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2293     struct glsl_src_param src_param;
2294     unsigned int mask_size;
2295     DWORD write_mask;
2296     char dst_mask[6];
2297
2298     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2299     mask_size = shader_glsl_get_write_mask_size(write_mask);
2300     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2301
2302     shader_addline(buffer, "tmp0.x = dot(%s, %s);\n",
2303             src_param.param_str, src_param.param_str);
2304     shader_glsl_append_dst(buffer, ins);
2305
2306     if (mask_size > 1)
2307     {
2308         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s * inversesqrt(tmp0.x)));\n",
2309                 mask_size, src_param.param_str);
2310     }
2311     else
2312     {
2313         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s * inversesqrt(tmp0.x)));\n",
2314                 src_param.param_str);
2315     }
2316 }
2317
2318 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2319  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2320  *   dst.x = 2^(floor(src))
2321  *   dst.y = src - floor(src)
2322  *   dst.z = 2^src   (partial precision is allowed, but optional)
2323  *   dst.w = 1.0;
2324  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2325  *   dst = 2^src;    (partial precision is allowed, but optional)
2326  */
2327 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2328 {
2329     struct glsl_src_param src_param;
2330
2331     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2332
2333     if (ins->ctx->reg_maps->shader_version.major < 2)
2334     {
2335         char dst_mask[6];
2336
2337         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2338         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2339         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2340         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2341
2342         shader_glsl_append_dst(ins->ctx->buffer, ins);
2343         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2344         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2345     } else {
2346         DWORD write_mask;
2347         unsigned int mask_size;
2348
2349         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2350         mask_size = shader_glsl_get_write_mask_size(write_mask);
2351
2352         if (mask_size > 1) {
2353             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2354         } else {
2355             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2356         }
2357     }
2358 }
2359
2360 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2361 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2362 {
2363     struct glsl_src_param src_param;
2364     DWORD write_mask;
2365     unsigned int mask_size;
2366
2367     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2368     mask_size = shader_glsl_get_write_mask_size(write_mask);
2369     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2370
2371     if (mask_size > 1)
2372     {
2373         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2374                 mask_size, src_param.param_str);
2375     }
2376     else
2377     {
2378         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2379                 src_param.param_str);
2380     }
2381 }
2382
2383 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2384 {
2385     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2386     struct glsl_src_param src_param;
2387     DWORD write_mask;
2388     unsigned int mask_size;
2389
2390     write_mask = shader_glsl_append_dst(buffer, ins);
2391     mask_size = shader_glsl_get_write_mask_size(write_mask);
2392
2393     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2394
2395     if (mask_size > 1)
2396     {
2397         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2398                 mask_size, src_param.param_str);
2399     }
2400     else
2401     {
2402         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2403                 src_param.param_str);
2404     }
2405 }
2406
2407 /** Process signed comparison opcodes in GLSL. */
2408 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2409 {
2410     struct glsl_src_param src0_param;
2411     struct glsl_src_param src1_param;
2412     DWORD write_mask;
2413     unsigned int mask_size;
2414
2415     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2416     mask_size = shader_glsl_get_write_mask_size(write_mask);
2417     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2418     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2419
2420     if (mask_size > 1) {
2421         const char *compare;
2422
2423         switch(ins->handler_idx)
2424         {
2425             case WINED3DSIH_SLT: compare = "lessThan"; break;
2426             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2427             default: compare = "";
2428                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2429         }
2430
2431         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2432                 src0_param.param_str, src1_param.param_str);
2433     } else {
2434         switch(ins->handler_idx)
2435         {
2436             case WINED3DSIH_SLT:
2437                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2438                  * to return 0.0 but step returns 1.0 because step is not < x
2439                  * An alternative is a bvec compare padded with an unused second component.
2440                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2441                  * issue. Playing with not() is not possible either because not() does not accept
2442                  * a scalar.
2443                  */
2444                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2445                         src0_param.param_str, src1_param.param_str);
2446                 break;
2447             case WINED3DSIH_SGE:
2448                 /* Here we can use the step() function and safe a conditional */
2449                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2450                 break;
2451             default:
2452                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2453         }
2454
2455     }
2456 }
2457
2458 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2459 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2460 {
2461     struct glsl_src_param src0_param;
2462     struct glsl_src_param src1_param;
2463     struct glsl_src_param src2_param;
2464     DWORD write_mask, cmp_channel = 0;
2465     unsigned int i, j;
2466     char mask_char[6];
2467     BOOL temp_destination = FALSE;
2468
2469     if (shader_is_scalar(&ins->src[0].reg))
2470     {
2471         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2472
2473         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2474         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2475         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2476
2477         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2478                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2479     } else {
2480         DWORD dst_mask = ins->dst[0].write_mask;
2481         struct wined3d_shader_dst_param dst = ins->dst[0];
2482
2483         /* Cycle through all source0 channels */
2484         for (i=0; i<4; i++) {
2485             write_mask = 0;
2486             /* Find the destination channels which use the current source0 channel */
2487             for (j=0; j<4; j++) {
2488                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2489                 {
2490                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2491                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2492                 }
2493             }
2494             dst.write_mask = dst_mask & write_mask;
2495
2496             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2497             * The first lines may overwrite source parameters of the following lines.
2498             * Deal with that by using a temporary destination register if needed
2499             */
2500             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2501                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2502                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2503                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2504                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2505                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2506             {
2507                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2508                 if (!write_mask) continue;
2509                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2510                 temp_destination = TRUE;
2511             } else {
2512                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2513                 if (!write_mask) continue;
2514             }
2515
2516             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2517             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2518             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2519
2520             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2521                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2522         }
2523
2524         if(temp_destination) {
2525             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2526             shader_glsl_append_dst(ins->ctx->buffer, ins);
2527             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2528         }
2529     }
2530
2531 }
2532
2533 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2534 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2535  * the compare is done per component of src0. */
2536 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2537 {
2538     struct wined3d_shader_dst_param dst;
2539     struct glsl_src_param src0_param;
2540     struct glsl_src_param src1_param;
2541     struct glsl_src_param src2_param;
2542     DWORD write_mask, cmp_channel = 0;
2543     unsigned int i, j;
2544     DWORD dst_mask;
2545     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2546             ins->ctx->reg_maps->shader_version.minor);
2547
2548     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2549     {
2550         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2551         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2552         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2553         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2554
2555         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2556         if (ins->coissue)
2557         {
2558             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2559         } else {
2560             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2561                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2562         }
2563         return;
2564     }
2565     /* Cycle through all source0 channels */
2566     dst_mask = ins->dst[0].write_mask;
2567     dst = ins->dst[0];
2568     for (i=0; i<4; i++) {
2569         write_mask = 0;
2570         /* Find the destination channels which use the current source0 channel */
2571         for (j=0; j<4; j++) {
2572             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2573             {
2574                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2575                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2576             }
2577         }
2578
2579         dst.write_mask = dst_mask & write_mask;
2580         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2581         if (!write_mask) continue;
2582
2583         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2584         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2585         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2586
2587         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2588                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2589     }
2590 }
2591
2592 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2593 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2594 {
2595     struct glsl_src_param src0_param;
2596     struct glsl_src_param src1_param;
2597     struct glsl_src_param src2_param;
2598     DWORD write_mask;
2599
2600     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2601     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2602     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2603     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2604     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2605             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2606 }
2607
2608 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2609    Vertex shaders to GLSL codes */
2610 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2611 {
2612     int i;
2613     int nComponents = 0;
2614     struct wined3d_shader_dst_param tmp_dst = {{0}};
2615     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2616     struct wined3d_shader_instruction tmp_ins;
2617
2618     memset(&tmp_ins, 0, sizeof(tmp_ins));
2619
2620     /* Set constants for the temporary argument */
2621     tmp_ins.ctx = ins->ctx;
2622     tmp_ins.dst_count = 1;
2623     tmp_ins.dst = &tmp_dst;
2624     tmp_ins.src_count = 2;
2625     tmp_ins.src = tmp_src;
2626
2627     switch(ins->handler_idx)
2628     {
2629         case WINED3DSIH_M4x4:
2630             nComponents = 4;
2631             tmp_ins.handler_idx = WINED3DSIH_DP4;
2632             break;
2633         case WINED3DSIH_M4x3:
2634             nComponents = 3;
2635             tmp_ins.handler_idx = WINED3DSIH_DP4;
2636             break;
2637         case WINED3DSIH_M3x4:
2638             nComponents = 4;
2639             tmp_ins.handler_idx = WINED3DSIH_DP3;
2640             break;
2641         case WINED3DSIH_M3x3:
2642             nComponents = 3;
2643             tmp_ins.handler_idx = WINED3DSIH_DP3;
2644             break;
2645         case WINED3DSIH_M3x2:
2646             nComponents = 2;
2647             tmp_ins.handler_idx = WINED3DSIH_DP3;
2648             break;
2649         default:
2650             break;
2651     }
2652
2653     tmp_dst = ins->dst[0];
2654     tmp_src[0] = ins->src[0];
2655     tmp_src[1] = ins->src[1];
2656     for (i = 0; i < nComponents; ++i)
2657     {
2658         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2659         shader_glsl_dot(&tmp_ins);
2660         ++tmp_src[1].reg.idx;
2661     }
2662 }
2663
2664 /**
2665     The LRP instruction performs a component-wise linear interpolation
2666     between the second and third operands using the first operand as the
2667     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2668     This is equivalent to mix(src2, src1, src0);
2669 */
2670 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2671 {
2672     struct glsl_src_param src0_param;
2673     struct glsl_src_param src1_param;
2674     struct glsl_src_param src2_param;
2675     DWORD write_mask;
2676
2677     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2678
2679     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2680     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2681     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2682
2683     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2684             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2685 }
2686
2687 /** Process the WINED3DSIO_LIT instruction in GLSL:
2688  * dst.x = dst.w = 1.0
2689  * dst.y = (src0.x > 0) ? src0.x
2690  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2691  *                                        where src.w is clamped at +- 128
2692  */
2693 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2694 {
2695     struct glsl_src_param src0_param;
2696     struct glsl_src_param src1_param;
2697     struct glsl_src_param src3_param;
2698     char dst_mask[6];
2699
2700     shader_glsl_append_dst(ins->ctx->buffer, ins);
2701     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2702
2703     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2704     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2705     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2706
2707     /* The sdk specifies the instruction like this
2708      * dst.x = 1.0;
2709      * if(src.x > 0.0) dst.y = src.x
2710      * else dst.y = 0.0.
2711      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2712      * else dst.z = 0.0;
2713      * dst.w = 1.0;
2714      * (where power = src.w clamped between -128 and 128)
2715      *
2716      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2717      * dst.x = 1.0                                  ... No further explanation needed
2718      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2719      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2720      * dst.w = 1.0.                                 ... Nothing fancy.
2721      *
2722      * So we still have one conditional in there. So do this:
2723      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2724      *
2725      * 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),
2726      * 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.
2727      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to.
2728      *
2729      * Unfortunately pow(0.0 ^ 0.0) returns NaN on most GPUs, but lit with src.y = 0 and src.w = 0 returns
2730      * a non-NaN value in dst.z. What we return doesn't matter, as long as it is not NaN. Return 0, which is
2731      * what all Windows HW drivers and GL_ARB_vertex_program's LIT do.
2732      */
2733     shader_addline(ins->ctx->buffer,
2734             "vec4(1.0, max(%s, 0.0), %s == 0.0 ? 0.0 : "
2735             "pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
2736             src0_param.param_str, src3_param.param_str, src1_param.param_str,
2737             src0_param.param_str, src3_param.param_str, dst_mask);
2738 }
2739
2740 /** Process the WINED3DSIO_DST instruction in GLSL:
2741  * dst.x = 1.0
2742  * dst.y = src0.x * src0.y
2743  * dst.z = src0.z
2744  * dst.w = src1.w
2745  */
2746 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2747 {
2748     struct glsl_src_param src0y_param;
2749     struct glsl_src_param src0z_param;
2750     struct glsl_src_param src1y_param;
2751     struct glsl_src_param src1w_param;
2752     char dst_mask[6];
2753
2754     shader_glsl_append_dst(ins->ctx->buffer, ins);
2755     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2756
2757     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2758     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2759     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2760     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2761
2762     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2763             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2764 }
2765
2766 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2767  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2768  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2769  *
2770  * dst.x = cos(src0.?)
2771  * dst.y = sin(src0.?)
2772  * dst.z = dst.z
2773  * dst.w = dst.w
2774  */
2775 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2776 {
2777     struct glsl_src_param src0_param;
2778     DWORD write_mask;
2779
2780     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2781     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2782
2783     switch (write_mask) {
2784         case WINED3DSP_WRITEMASK_0:
2785             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2786             break;
2787
2788         case WINED3DSP_WRITEMASK_1:
2789             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2790             break;
2791
2792         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2793             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2794             break;
2795
2796         default:
2797             ERR("Write mask should be .x, .y or .xy\n");
2798             break;
2799     }
2800 }
2801
2802 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2803  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2804  * generate invalid code
2805  */
2806 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2807 {
2808     struct glsl_src_param src0_param;
2809     DWORD write_mask;
2810
2811     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2812     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2813
2814     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2815 }
2816
2817 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2818  * Start a for() loop where src1.y is the initial value of aL,
2819  *  increment aL by src1.z for a total of src1.x iterations.
2820  *  Need to use a temporary variable for this operation.
2821  */
2822 /* FIXME: I don't think nested loops will work correctly this way. */
2823 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2824 {
2825     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2826     const struct wined3d_shader *shader = ins->ctx->shader;
2827     const struct wined3d_shader_lconst *constant;
2828     struct glsl_src_param src1_param;
2829     const DWORD *control_values = NULL;
2830
2831     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2832
2833     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2834      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2835      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2836      * addressing.
2837      */
2838     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2839     {
2840         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2841         {
2842             if (constant->idx == ins->src[1].reg.idx)
2843             {
2844                 control_values = constant->value;
2845                 break;
2846             }
2847         }
2848     }
2849
2850     if (control_values)
2851     {
2852         struct wined3d_shader_loop_control loop_control;
2853         loop_control.count = control_values[0];
2854         loop_control.start = control_values[1];
2855         loop_control.step = (int)control_values[2];
2856
2857         if (loop_control.step > 0)
2858         {
2859             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2860                     loop_state->current_depth, loop_control.start,
2861                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2862                     loop_state->current_depth, loop_control.step);
2863         }
2864         else if (loop_control.step < 0)
2865         {
2866             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2867                     loop_state->current_depth, loop_control.start,
2868                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2869                     loop_state->current_depth, loop_control.step);
2870         }
2871         else
2872         {
2873             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2874                     loop_state->current_depth, loop_control.start, loop_state->current_depth,
2875                     loop_state->current_depth, loop_control.count,
2876                     loop_state->current_depth);
2877         }
2878     } else {
2879         shader_addline(ins->ctx->buffer,
2880                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2881                 loop_state->current_depth, loop_state->current_reg,
2882                 src1_param.reg_name, loop_state->current_depth, src1_param.reg_name,
2883                 loop_state->current_depth, loop_state->current_reg, src1_param.reg_name);
2884     }
2885
2886     ++loop_state->current_depth;
2887     ++loop_state->current_reg;
2888 }
2889
2890 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2891 {
2892     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2893
2894     shader_addline(ins->ctx->buffer, "}\n");
2895
2896     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2897     {
2898         --loop_state->current_depth;
2899         --loop_state->current_reg;
2900     }
2901
2902     if (ins->handler_idx == WINED3DSIH_ENDREP)
2903     {
2904         --loop_state->current_depth;
2905     }
2906 }
2907
2908 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2909 {
2910     const struct wined3d_shader *shader = ins->ctx->shader;
2911     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2912     const struct wined3d_shader_lconst *constant;
2913     struct glsl_src_param src0_param;
2914     const DWORD *control_values = NULL;
2915
2916     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2917     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2918     {
2919         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2920         {
2921             if (constant->idx == ins->src[0].reg.idx)
2922             {
2923                 control_values = constant->value;
2924                 break;
2925             }
2926         }
2927     }
2928
2929     if (control_values)
2930     {
2931         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2932                 loop_state->current_depth, loop_state->current_depth,
2933                 control_values[0], loop_state->current_depth);
2934     }
2935     else
2936     {
2937         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2938         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2939                 loop_state->current_depth, loop_state->current_depth,
2940                 src0_param.param_str, loop_state->current_depth);
2941     }
2942
2943     ++loop_state->current_depth;
2944 }
2945
2946 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2947 {
2948     struct glsl_src_param src0_param;
2949
2950     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2951     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2952 }
2953
2954 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2955 {
2956     struct glsl_src_param src0_param;
2957     struct glsl_src_param src1_param;
2958
2959     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2960     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2961
2962     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2963             src0_param.param_str, shader_glsl_get_rel_op(ins->flags), src1_param.param_str);
2964 }
2965
2966 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2967 {
2968     shader_addline(ins->ctx->buffer, "} else {\n");
2969 }
2970
2971 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2972 {
2973     shader_addline(ins->ctx->buffer, "break;\n");
2974 }
2975
2976 /* FIXME: According to MSDN the compare is done per component. */
2977 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2978 {
2979     struct glsl_src_param src0_param;
2980     struct glsl_src_param src1_param;
2981
2982     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2983     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2984
2985     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2986             src0_param.param_str, shader_glsl_get_rel_op(ins->flags), src1_param.param_str);
2987 }
2988
2989 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2990 {
2991     shader_addline(ins->ctx->buffer, "}\n");
2992     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2993 }
2994
2995 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2996 {
2997     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2998 }
2999
3000 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3001 {
3002     struct glsl_src_param src1_param;
3003
3004     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3005     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3006 }
3007
3008 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3009 {
3010     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3011      * function only suppresses the unhandled instruction warning
3012      */
3013 }
3014
3015 /*********************************************
3016  * Pixel Shader Specific Code begins here
3017  ********************************************/
3018 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3019 {
3020     const struct wined3d_shader *shader = ins->ctx->shader;
3021     struct wined3d_device *device = shader->device;
3022     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3023             ins->ctx->reg_maps->shader_version.minor);
3024     struct glsl_sample_function sample_function;
3025     const struct wined3d_texture *texture;
3026     DWORD sample_flags = 0;
3027     DWORD sampler_idx;
3028     DWORD mask = 0, swizzle;
3029
3030     /* 1.0-1.4: Use destination register as sampler source.
3031      * 2.0+: Use provided sampler source. */
3032     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3033     else sampler_idx = ins->src[1].reg.idx;
3034     texture = device->stateBlock->state.textures[sampler_idx];
3035
3036     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3037     {
3038         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3039         DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3040                 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3041         enum wined3d_sampler_texture_type sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3042
3043         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3044         if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3045             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3046             switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3047                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3048                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3049                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3050                 case WINED3DTTFF_COUNT4:
3051                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3052             }
3053         }
3054     }
3055     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3056     {
3057         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
3058
3059         if (src_mod == WINED3DSPSM_DZ) {
3060             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3061             mask = WINED3DSP_WRITEMASK_2;
3062         } else if (src_mod == WINED3DSPSM_DW) {
3063             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3064             mask = WINED3DSP_WRITEMASK_3;
3065         }
3066     } else {
3067         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3068         {
3069             /* ps 2.0 texldp instruction always divides by the fourth component. */
3070             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3071             mask = WINED3DSP_WRITEMASK_3;
3072         }
3073     }
3074
3075     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3076         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3077
3078     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3079     mask |= sample_function.coord_mask;
3080
3081     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3082     else swizzle = ins->src[1].swizzle;
3083
3084     /* 1.0-1.3: Use destination register as coordinate source.
3085        1.4+: Use provided coordinate source register. */
3086     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3087     {
3088         char coord_mask[6];
3089         shader_glsl_write_mask_to_str(mask, coord_mask);
3090         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3091                 "T%u%s", sampler_idx, coord_mask);
3092     }
3093     else
3094     {
3095         struct glsl_src_param coord_param;
3096         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3097         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3098         {
3099             struct glsl_src_param bias;
3100             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3101             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3102                     "%s", coord_param.param_str);
3103         } else {
3104             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3105                     "%s", coord_param.param_str);
3106         }
3107     }
3108 }
3109
3110 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3111 {
3112     const struct wined3d_shader *shader = ins->ctx->shader;
3113     struct wined3d_device *device = shader->device;
3114     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3115     struct glsl_src_param coord_param, dx_param, dy_param;
3116     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3117     struct glsl_sample_function sample_function;
3118     DWORD sampler_idx;
3119     DWORD swizzle = ins->src[1].swizzle;
3120     const struct wined3d_texture *texture;
3121
3122     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3123     {
3124         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3125         shader_glsl_tex(ins);
3126         return;
3127     }
3128
3129     sampler_idx = ins->src[1].reg.idx;
3130     texture = device->stateBlock->state.textures[sampler_idx];
3131     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3132         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3133
3134     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3135     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3136     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3137     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3138
3139     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3140                                 "%s", coord_param.param_str);
3141 }
3142
3143 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3144 {
3145     const struct wined3d_shader *shader = ins->ctx->shader;
3146     struct wined3d_device *device = shader->device;
3147     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3148     struct glsl_src_param coord_param, lod_param;
3149     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3150     struct glsl_sample_function sample_function;
3151     DWORD sampler_idx;
3152     DWORD swizzle = ins->src[1].swizzle;
3153     const struct wined3d_texture *texture;
3154
3155     sampler_idx = ins->src[1].reg.idx;
3156     texture = device->stateBlock->state.textures[sampler_idx];
3157     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3158         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3159
3160     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3161     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3162
3163     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3164
3165     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3166             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3167     {
3168         /* Plain GLSL only supports Lod sampling functions in vertex shaders.
3169          * However, the NVIDIA drivers allow them in fragment shaders as well,
3170          * even without the appropriate extension. */
3171         WARN("Using %s in fragment shader.\n", sample_function.name);
3172     }
3173     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3174             "%s", coord_param.param_str);
3175 }
3176
3177 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3178 {
3179     /* FIXME: Make this work for more than just 2D textures */
3180     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3181     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3182
3183     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3184     {
3185         char dst_mask[6];
3186
3187         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3188         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3189                 ins->dst[0].reg.idx, dst_mask);
3190     }
3191     else
3192     {
3193         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
3194         DWORD reg = ins->src[0].reg.idx;
3195         char dst_swizzle[6];
3196
3197         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3198
3199         if (src_mod == WINED3DSPSM_DZ)
3200         {
3201             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3202             struct glsl_src_param div_param;
3203
3204             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3205
3206             if (mask_size > 1) {
3207                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3208             } else {
3209                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3210             }
3211         }
3212         else if (src_mod == WINED3DSPSM_DW)
3213         {
3214             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3215             struct glsl_src_param div_param;
3216
3217             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3218
3219             if (mask_size > 1) {
3220                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3221             } else {
3222                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3223             }
3224         } else {
3225             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3226         }
3227     }
3228 }
3229
3230 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3231  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3232  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3233 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3234 {
3235     DWORD sampler_idx = ins->dst[0].reg.idx;
3236     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3237     struct glsl_sample_function sample_function;
3238     struct glsl_src_param src0_param;
3239     UINT mask_size;
3240
3241     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3242
3243     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3244      * scalar, and projected sampling would require 4.
3245      *
3246      * It is a dependent read - not valid with conditional NP2 textures
3247      */
3248     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3249     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3250
3251     switch(mask_size)
3252     {
3253         case 1:
3254             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3255                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3256             break;
3257
3258         case 2:
3259             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3260                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3261             break;
3262
3263         case 3:
3264             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3265                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3266             break;
3267
3268         default:
3269             FIXME("Unexpected mask size %u\n", mask_size);
3270             break;
3271     }
3272 }
3273
3274 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3275  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3276 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3277 {
3278     DWORD dstreg = ins->dst[0].reg.idx;
3279     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3280     struct glsl_src_param src0_param;
3281     DWORD dst_mask;
3282     unsigned int mask_size;
3283
3284     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3285     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3286     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3287
3288     if (mask_size > 1) {
3289         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3290     } else {
3291         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3292     }
3293 }
3294
3295 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3296  * Calculate the depth as dst.x / dst.y   */
3297 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3298 {
3299     struct glsl_dst_param dst_param;
3300
3301     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3302
3303     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3304      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3305      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3306      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3307      * >= 1.0 or < 0.0
3308      */
3309     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3310             dst_param.reg_name, dst_param.reg_name);
3311 }
3312
3313 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3314  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3315  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3316  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3317  */
3318 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3319 {
3320     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3321     DWORD dstreg = ins->dst[0].reg.idx;
3322     struct glsl_src_param src0_param;
3323
3324     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3325
3326     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3327     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3328 }
3329
3330 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3331  * Calculate the 1st of a 2-row matrix multiplication. */
3332 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3333 {
3334     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3335     DWORD reg = ins->dst[0].reg.idx;
3336     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3337     struct glsl_src_param src0_param;
3338
3339     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3340     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3341 }
3342
3343 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3344  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3345 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3346 {
3347     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3348     DWORD reg = ins->dst[0].reg.idx;
3349     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3350     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3351     struct glsl_src_param src0_param;
3352
3353     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3354     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + tex_mx->current_row, reg, src0_param.param_str);
3355     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
3356 }
3357
3358 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3359 {
3360     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3361     DWORD reg = ins->dst[0].reg.idx;
3362     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3363     struct glsl_sample_function sample_function;
3364     struct glsl_src_param src0_param;
3365
3366     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3367     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3368
3369     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3370
3371     /* Sample the texture using the calculated coordinates */
3372     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3373 }
3374
3375 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3376  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3377 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3378 {
3379     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3380     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3381     struct glsl_sample_function sample_function;
3382     struct glsl_src_param src0_param;
3383     DWORD reg = ins->dst[0].reg.idx;
3384
3385     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3386     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3387
3388     /* Dependent read, not valid with conditional NP2 */
3389     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3390
3391     /* Sample the texture using the calculated coordinates */
3392     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3393
3394     tex_mx->current_row = 0;
3395 }
3396
3397 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3398  * Perform the 3rd row of a 3x3 matrix multiply */
3399 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3400 {
3401     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3402     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3403     struct glsl_src_param src0_param;
3404     char dst_mask[6];
3405     DWORD reg = ins->dst[0].reg.idx;
3406
3407     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3408
3409     shader_glsl_append_dst(ins->ctx->buffer, ins);
3410     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3411     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3412
3413     tex_mx->current_row = 0;
3414 }
3415
3416 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3417  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3418 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3419 {
3420     struct glsl_src_param src0_param;
3421     struct glsl_src_param src1_param;
3422     DWORD reg = ins->dst[0].reg.idx;
3423     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3424     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3425     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3426     struct glsl_sample_function sample_function;
3427     char coord_mask[6];
3428
3429     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3430     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3431
3432     /* Perform the last matrix multiply operation */
3433     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3434     /* Reflection calculation */
3435     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3436
3437     /* Dependent read, not valid with conditional NP2 */
3438     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3439     shader_glsl_write_mask_to_str(sample_function.coord_mask, coord_mask);
3440
3441     /* Sample the texture */
3442     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE,
3443             NULL, NULL, NULL, "tmp0%s", coord_mask);
3444
3445     tex_mx->current_row = 0;
3446 }
3447
3448 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3449  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3450 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3451 {
3452     DWORD reg = ins->dst[0].reg.idx;
3453     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3454     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3455     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3456     struct glsl_sample_function sample_function;
3457     struct glsl_src_param src0_param;
3458     char coord_mask[6];
3459
3460     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3461
3462     /* Perform the last matrix multiply operation */
3463     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3464
3465     /* Construct the eye-ray vector from w coordinates */
3466     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3467             tex_mx->texcoord_w[0], tex_mx->texcoord_w[1], reg);
3468     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3469
3470     /* Dependent read, not valid with conditional NP2 */
3471     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3472     shader_glsl_write_mask_to_str(sample_function.coord_mask, coord_mask);
3473
3474     /* Sample the texture using the calculated coordinates */
3475     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE,
3476             NULL, NULL, NULL, "tmp0%s", coord_mask);
3477
3478     tex_mx->current_row = 0;
3479 }
3480
3481 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3482  * Apply a fake bump map transform.
3483  * texbem is pshader <= 1.3 only, this saves a few version checks
3484  */
3485 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3486 {
3487     const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3488     struct glsl_sample_function sample_function;
3489     struct glsl_src_param coord_param;
3490     DWORD sampler_idx;
3491     DWORD mask;
3492     DWORD flags;
3493     char coord_mask[6];
3494
3495     sampler_idx = ins->dst[0].reg.idx;
3496     flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3497             & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3498
3499     /* Dependent read, not valid with conditional NP2 */
3500     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3501     mask = sample_function.coord_mask;
3502
3503     shader_glsl_write_mask_to_str(mask, coord_mask);
3504
3505     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3506          * so we can't let the GL handle this.
3507          */
3508     if (flags & WINED3D_PSARGS_PROJECTED) {
3509         DWORD div_mask=0;
3510         char coord_div_mask[3];
3511         switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3512             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3513             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3514             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3515             case WINED3DTTFF_COUNT4:
3516             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3517         }
3518         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3519         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3520     }
3521
3522     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3523
3524     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3525             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3526             coord_param.param_str, coord_mask);
3527
3528     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3529     {
3530         struct glsl_src_param luminance_param;
3531         struct glsl_dst_param dst_param;
3532
3533         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3534         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3535
3536         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3537                 dst_param.reg_name, dst_param.mask_str,
3538                 luminance_param.param_str, sampler_idx, sampler_idx);
3539     }
3540 }
3541
3542 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3543 {
3544     struct glsl_src_param src0_param, src1_param;
3545     DWORD sampler_idx = ins->dst[0].reg.idx;
3546
3547     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3548     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3549
3550     shader_glsl_append_dst(ins->ctx->buffer, ins);
3551     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3552             src0_param.param_str, sampler_idx, src1_param.param_str);
3553 }
3554
3555 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3556  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3557 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3558 {
3559     struct glsl_sample_function sample_function;
3560     struct glsl_src_param src0_param;
3561     DWORD sampler_idx = ins->dst[0].reg.idx;
3562
3563     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3564
3565     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3566     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3567             "%s.wx", src0_param.reg_name);
3568 }
3569
3570 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3571  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3572 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3573 {
3574     struct glsl_sample_function sample_function;
3575     struct glsl_src_param src0_param;
3576     DWORD sampler_idx = ins->dst[0].reg.idx;
3577
3578     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3579
3580     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3581     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3582             "%s.yz", src0_param.reg_name);
3583 }
3584
3585 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3586  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3587 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3588 {
3589     struct glsl_sample_function sample_function;
3590     struct glsl_src_param src0_param;
3591     DWORD sampler_idx = ins->dst[0].reg.idx;
3592
3593     /* Dependent read, not valid with conditional NP2 */
3594     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3595     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3596
3597     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3598             "%s", src0_param.param_str);
3599 }
3600
3601 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3602  * If any of the first 3 components are < 0, discard this pixel */
3603 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3604 {
3605     struct glsl_dst_param dst_param;
3606
3607     /* The argument is a destination parameter, and no writemasks are allowed */
3608     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3609     if (ins->ctx->reg_maps->shader_version.major >= 2)
3610     {
3611         /* 2.0 shaders compare all 4 components in texkill */
3612         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3613     } else {
3614         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3615          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3616          * 4 components are defined, only the first 3 are used
3617          */
3618         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3619     }
3620 }
3621
3622 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3623  * dst = dot2(src0, src1) + src2 */
3624 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3625 {
3626     struct glsl_src_param src0_param;
3627     struct glsl_src_param src1_param;
3628     struct glsl_src_param src2_param;
3629     DWORD write_mask;
3630     unsigned int mask_size;
3631
3632     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3633     mask_size = shader_glsl_get_write_mask_size(write_mask);
3634
3635     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3636     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3637     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3638
3639     if (mask_size > 1) {
3640         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3641                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3642     } else {
3643         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3644                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3645     }
3646 }
3647
3648 static void shader_glsl_input_pack(const struct wined3d_shader *shader, struct wined3d_shader_buffer *buffer,
3649         const struct wined3d_shader_signature_element *input_signature,
3650         const struct wined3d_shader_reg_maps *reg_maps,
3651         enum vertexprocessing_mode vertexprocessing)
3652 {
3653     WORD map = reg_maps->input_registers;
3654     unsigned int i;
3655
3656     for (i = 0; map; map >>= 1, ++i)
3657     {
3658         const char *semantic_name;
3659         UINT semantic_idx;
3660         char reg_mask[6];
3661
3662         /* Unused */
3663         if (!(map & 1)) continue;
3664
3665         semantic_name = input_signature[i].semantic_name;
3666         semantic_idx = input_signature[i].semantic_idx;
3667         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3668
3669         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3670         {
3671             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3672                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3673                         shader->u.ps.input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3674             else
3675                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3676                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3677         }
3678         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3679         {
3680             if (!semantic_idx)
3681                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3682                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3683             else if (semantic_idx == 1)
3684                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3685                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3686             else
3687                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3688                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3689         }
3690         else
3691         {
3692             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3693                     shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3694         }
3695     }
3696 }
3697
3698 /*********************************************
3699  * Vertex Shader Specific Code begins here
3700  ********************************************/
3701
3702 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry)
3703 {
3704     struct glsl_program_key key;
3705
3706     key.vshader = entry->vshader;
3707     key.pshader = entry->pshader;
3708     key.vs_args = entry->vs_args;
3709     key.ps_args = entry->ps_args;
3710
3711     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3712     {
3713         ERR("Failed to insert program entry.\n");
3714     }
3715 }
3716
3717 static struct glsl_shader_prog_link *get_glsl_program_entry(const struct shader_glsl_priv *priv,
3718         const struct wined3d_shader *vshader, const struct wined3d_shader *pshader,
3719         const struct vs_compile_args *vs_args, const struct ps_compile_args *ps_args)
3720 {
3721     struct wine_rb_entry *entry;
3722     struct glsl_program_key key;
3723
3724     key.vshader = vshader;
3725     key.pshader = pshader;
3726     key.vs_args = *vs_args;
3727     key.ps_args = *ps_args;
3728
3729     entry = wine_rb_get(&priv->program_lookup, &key);
3730     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3731 }
3732
3733 /* GL locking is done by the caller */
3734 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3735         struct glsl_shader_prog_link *entry)
3736 {
3737     struct glsl_program_key key;
3738
3739     key.vshader = entry->vshader;
3740     key.pshader = entry->pshader;
3741     key.vs_args = entry->vs_args;
3742     key.ps_args = entry->ps_args;
3743     wine_rb_remove(&priv->program_lookup, &key);
3744
3745     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3746     if (entry->vshader) list_remove(&entry->vshader_entry);
3747     if (entry->pshader) list_remove(&entry->pshader_entry);
3748     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3749     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3750     HeapFree(GetProcessHeap(), 0, entry);
3751 }
3752
3753 static void handle_ps3_input(struct wined3d_shader_buffer *buffer,
3754         const struct wined3d_gl_info *gl_info, const DWORD *map,
3755         const struct wined3d_shader_signature_element *input_signature,
3756         const struct wined3d_shader_reg_maps *reg_maps_in,
3757         const struct wined3d_shader_signature_element *output_signature,
3758         const struct wined3d_shader_reg_maps *reg_maps_out)
3759 {
3760     unsigned int i, j;
3761     const char *semantic_name_in;
3762     UINT semantic_idx_in;
3763     DWORD *set;
3764     DWORD in_idx;
3765     unsigned int in_count = vec4_varyings(3, gl_info);
3766     char reg_mask[6];
3767     char destination[50];
3768     WORD input_map, output_map;
3769
3770     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3771
3772     input_map = reg_maps_in->input_registers;
3773     for (i = 0; input_map; input_map >>= 1, ++i)
3774     {
3775         if (!(input_map & 1)) continue;
3776
3777         in_idx = map[i];
3778         /* Declared, but not read register */
3779         if (in_idx == ~0U) continue;
3780         if (in_idx >= (in_count + 2))
3781         {
3782             FIXME("More input varyings declared than supported, expect issues.\n");
3783             continue;
3784         }
3785
3786         if (in_idx == in_count) {
3787             sprintf(destination, "gl_FrontColor");
3788         } else if (in_idx == in_count + 1) {
3789             sprintf(destination, "gl_FrontSecondaryColor");
3790         } else {
3791             sprintf(destination, "IN[%u]", in_idx);
3792         }
3793
3794         semantic_name_in = input_signature[i].semantic_name;
3795         semantic_idx_in = input_signature[i].semantic_idx;
3796         set[in_idx] = ~0U;
3797
3798         output_map = reg_maps_out->output_registers;
3799         for (j = 0; output_map; output_map >>= 1, ++j)
3800         {
3801             DWORD mask;
3802
3803             if (!(output_map & 1)
3804                     || semantic_idx_in != output_signature[j].semantic_idx
3805                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3806                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3807                 continue;
3808
3809             set[in_idx] = mask;
3810             shader_glsl_write_mask_to_str(mask, reg_mask);
3811
3812             shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3813                     destination, reg_mask, j, reg_mask);
3814         }
3815     }
3816
3817     for (i = 0; i < in_count + 2; ++i)
3818     {
3819         unsigned int size;
3820
3821         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3822             continue;
3823
3824         if (set[i] == ~0U) set[i] = 0;
3825
3826         size = 0;
3827         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3828         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3829         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3830         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3831         reg_mask[size] = '\0';
3832
3833         if (i == in_count) sprintf(destination, "gl_FrontColor");
3834         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3835         else sprintf(destination, "IN[%u]", i);
3836
3837         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3838         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3839     }
3840
3841     HeapFree(GetProcessHeap(), 0, set);
3842 }
3843
3844 /* GL locking is done by the caller */
3845 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3846         const struct wined3d_shader *vs, const struct wined3d_shader *ps,
3847         const struct wined3d_gl_info *gl_info)
3848 {
3849     GLhandleARB ret = 0;
3850     DWORD ps_major = ps ? ps->reg_maps.shader_version.major : 0;
3851     unsigned int i;
3852     const char *semantic_name;
3853     UINT semantic_idx;
3854     char reg_mask[6];
3855     const struct wined3d_shader_signature_element *output_signature = vs->output_signature;
3856     WORD map = vs->reg_maps.output_registers;
3857
3858     shader_buffer_clear(buffer);
3859
3860     shader_addline(buffer, "#version 120\n");
3861
3862     if (ps_major < 3)
3863     {
3864         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3865
3866         for (i = 0; map; map >>= 1, ++i)
3867         {
3868             DWORD write_mask;
3869
3870             if (!(map & 1)) continue;
3871
3872             semantic_name = output_signature[i].semantic_name;
3873             semantic_idx = output_signature[i].semantic_idx;
3874             write_mask = output_signature[i].mask;
3875             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3876
3877             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3878             {
3879                 if (!semantic_idx)
3880                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n",
3881                             reg_mask, i, reg_mask);
3882                 else if (semantic_idx == 1)
3883                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n",
3884                             reg_mask, i, reg_mask);
3885             }
3886             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3887             {
3888                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3889                         reg_mask, i, reg_mask);
3890             }
3891             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3892             {
3893                 if (semantic_idx < 8)
3894                 {
3895                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3896                         write_mask |= WINED3DSP_WRITEMASK_3;
3897
3898                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3899                             semantic_idx, reg_mask, i, reg_mask);
3900                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3901                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3902                 }
3903             }
3904             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3905             {
3906                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3907             }
3908             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3909             {
3910                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3911             }
3912         }
3913         shader_addline(buffer, "}\n");
3914
3915     }
3916     else
3917     {
3918         UINT in_count = min(vec4_varyings(ps_major, gl_info), ps->limits.packed_input);
3919         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3920         shader_addline(buffer, "varying vec4 IN[%u];\n", in_count);
3921         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3922
3923         /* First, sort out position and point size. Those are not passed to the pixel shader */
3924         for (i = 0; map; map >>= 1, ++i)
3925         {
3926             if (!(map & 1)) continue;
3927
3928             semantic_name = output_signature[i].semantic_name;
3929             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3930
3931             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3932             {
3933                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3934                         reg_mask, i, reg_mask);
3935             }
3936             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3937             {
3938                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3939             }
3940         }
3941
3942         /* Then, fix the pixel shader input */
3943         handle_ps3_input(buffer, gl_info, ps->u.ps.input_reg_map, ps->input_signature,
3944                 &ps->reg_maps, output_signature, &vs->reg_maps);
3945
3946         shader_addline(buffer, "}\n");
3947     }
3948
3949     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3950     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3951     shader_glsl_compile(gl_info, ret, buffer->buffer);
3952
3953     return ret;
3954 }
3955
3956 /* GL locking is done by the caller */
3957 static void hardcode_local_constants(const struct wined3d_shader *shader,
3958         const struct wined3d_gl_info *gl_info, GLhandleARB programId, char prefix)
3959 {
3960     const struct wined3d_shader_lconst *lconst;
3961     GLint tmp_loc;
3962     const float *value;
3963     char glsl_name[8];
3964
3965     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3966     {
3967         value = (const float *)lconst->value;
3968         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3969         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3970         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3971     }
3972     checkGLcall("Hardcoding local constants");
3973 }
3974
3975 /* GL locking is done by the caller */
3976 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3977         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
3978         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3979 {
3980     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3981     const struct wined3d_gl_info *gl_info = context->gl_info;
3982     const DWORD *function = shader->function;
3983     struct shader_glsl_ctx_priv priv_ctx;
3984
3985     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3986     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3987
3988     memset(&priv_ctx, 0, sizeof(priv_ctx));
3989     priv_ctx.cur_ps_args = args;
3990     priv_ctx.cur_np2fixup_info = np2fixup_info;
3991
3992     shader_addline(buffer, "#version 120\n");
3993
3994     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
3995     {
3996         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3997     }
3998     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3999     {
4000         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
4001          * drivers write a warning if we don't do so
4002          */
4003         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
4004     }
4005     if (gl_info->supported[EXT_GPU_SHADER4])
4006     {
4007         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4008     }
4009
4010     /* Base Declarations */
4011     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4012
4013     /* Pack 3.0 inputs */
4014     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4015         shader_glsl_input_pack(shader, buffer, shader->input_signature, reg_maps, args->vp_mode);
4016
4017     /* Base Shader Body */
4018     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4019
4020     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4021     if (reg_maps->shader_version.major < 2)
4022     {
4023         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4024         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4025     }
4026
4027     if (args->srgb_correction)
4028     {
4029         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4030         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4031         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4032         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4033         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4034         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4035     }
4036     /* Pixel shader < 3.0 do not replace the fog stage.
4037      * This implements linear fog computation and blending.
4038      * TODO: non linear fog
4039      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4040      * -1/(e-s) and e/(e-s) respectively.
4041      */
4042     if (reg_maps->shader_version.major < 3)
4043     {
4044         switch(args->fog) {
4045             case FOG_OFF: break;
4046             case FOG_LINEAR:
4047                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4048                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4049                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4050                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4051                 break;
4052             case FOG_EXP:
4053                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4054                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4055                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4056                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4057                 break;
4058             case FOG_EXP2:
4059                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4060                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4061                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4062                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4063                 break;
4064         }
4065     }
4066
4067     shader_addline(buffer, "}\n");
4068
4069     TRACE("Compiling shader object %u\n", shader_obj);
4070     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4071
4072     /* Store the shader object */
4073     return shader_obj;
4074 }
4075
4076 /* GL locking is done by the caller */
4077 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4078         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
4079         const struct vs_compile_args *args)
4080 {
4081     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4082     const struct wined3d_gl_info *gl_info = context->gl_info;
4083     const DWORD *function = shader->function;
4084     struct shader_glsl_ctx_priv priv_ctx;
4085
4086     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4087     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4088
4089     shader_addline(buffer, "#version 120\n");
4090
4091     if (gl_info->supported[EXT_GPU_SHADER4])
4092         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4093
4094     memset(&priv_ctx, 0, sizeof(priv_ctx));
4095     priv_ctx.cur_vs_args = args;
4096
4097     /* Base Declarations */
4098     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4099
4100     /* Base Shader Body */
4101     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4102
4103     /* Unpack outputs */
4104     shader_addline(buffer, "order_ps_input(OUT);\n");
4105
4106     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4107      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4108      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4109      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4110      */
4111     if (args->fog_src == VS_FOG_Z)
4112         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4113     else if (!reg_maps->fog)
4114         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4115
4116     /* We always store the clipplanes without y inversion */
4117     if (args->clip_enabled)
4118         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4119
4120     /* Write the final position.
4121      *
4122      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4123      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4124      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4125      * contains 1.0 to allow a mad.
4126      */
4127     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4128     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4129
4130     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4131      *
4132      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4133      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4134      * which is the same as z = z * 2 - w.
4135      */
4136     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4137
4138     shader_addline(buffer, "}\n");
4139
4140     TRACE("Compiling shader object %u\n", shader_obj);
4141     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4142
4143     return shader_obj;
4144 }
4145
4146 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4147         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4148         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4149 {
4150     struct wined3d_state *state = &shader->device->stateBlock->state;
4151     UINT i;
4152     DWORD new_size;
4153     struct glsl_ps_compiled_shader *new_array;
4154     struct glsl_pshader_private    *shader_data;
4155     struct ps_np2fixup_info        *np2fixup = NULL;
4156     GLhandleARB ret;
4157
4158     if (!shader->backend_data)
4159     {
4160         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4161         if (!shader->backend_data)
4162         {
4163             ERR("Failed to allocate backend data.\n");
4164             return 0;
4165         }
4166     }
4167     shader_data = shader->backend_data;
4168
4169     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4170      * so a linear search is more performant than a hashmap or a binary search
4171      * (cache coherency etc)
4172      */
4173     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4174     {
4175         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4176         {
4177             if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4178             return shader_data->gl_shaders[i].prgId;
4179         }
4180     }
4181
4182     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4183     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4184         if (shader_data->num_gl_shaders)
4185         {
4186             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4187             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4188                                     new_size * sizeof(*shader_data->gl_shaders));
4189         } else {
4190             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4191             new_size = 1;
4192         }
4193
4194         if(!new_array) {
4195             ERR("Out of memory\n");
4196             return 0;
4197         }
4198         shader_data->gl_shaders = new_array;
4199         shader_data->shader_array_size = new_size;
4200     }
4201
4202     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4203
4204     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4205     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4206
4207     pixelshader_update_samplers(&shader->reg_maps, state->textures);
4208
4209     shader_buffer_clear(buffer);
4210     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4211     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4212     *np2fixup_info = np2fixup;
4213
4214     return ret;
4215 }
4216
4217 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4218                                  const DWORD use_map) {
4219     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4220     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4221     return stored->fog_src == new->fog_src;
4222 }
4223
4224 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4225         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4226         const struct vs_compile_args *args)
4227 {
4228     UINT i;
4229     DWORD new_size;
4230     struct glsl_vs_compiled_shader *new_array;
4231     DWORD use_map = shader->device->strided_streams.use_map;
4232     struct glsl_vshader_private *shader_data;
4233     GLhandleARB ret;
4234
4235     if (!shader->backend_data)
4236     {
4237         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4238         if (!shader->backend_data)
4239         {
4240             ERR("Failed to allocate backend data.\n");
4241             return 0;
4242         }
4243     }
4244     shader_data = shader->backend_data;
4245
4246     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4247      * so a linear search is more performant than a hashmap or a binary search
4248      * (cache coherency etc)
4249      */
4250     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4251         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4252             return shader_data->gl_shaders[i].prgId;
4253         }
4254     }
4255
4256     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4257
4258     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4259         if (shader_data->num_gl_shaders)
4260         {
4261             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4262             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4263                                     new_size * sizeof(*shader_data->gl_shaders));
4264         } else {
4265             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4266             new_size = 1;
4267         }
4268
4269         if(!new_array) {
4270             ERR("Out of memory\n");
4271             return 0;
4272         }
4273         shader_data->gl_shaders = new_array;
4274         shader_data->shader_array_size = new_size;
4275     }
4276
4277     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4278
4279     shader_buffer_clear(buffer);
4280     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4281     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4282
4283     return ret;
4284 }
4285
4286 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4287  * It sets the programId on the current StateBlock (because it should be called
4288  * inside of the DrawPrimitive() part of the render loop).
4289  *
4290  * If a program for the given combination does not exist, create one, and store
4291  * the program in the hash table.  If it creates a program, it will link the
4292  * given objects, too.
4293  */
4294
4295 /* GL locking is done by the caller */
4296 static void set_glsl_shader_program(const struct wined3d_context *context,
4297         struct wined3d_device *device, BOOL use_ps, BOOL use_vs)
4298 {
4299     const struct wined3d_state *state = &device->stateBlock->state;
4300     struct wined3d_shader *vshader = use_vs ? state->vertex_shader : NULL;
4301     struct wined3d_shader *pshader = use_ps ? state->pixel_shader : NULL;
4302     const struct wined3d_gl_info *gl_info = context->gl_info;
4303     struct shader_glsl_priv *priv = device->shader_priv;
4304     struct glsl_shader_prog_link *entry    = NULL;
4305     GLhandleARB programId                  = 0;
4306     GLhandleARB reorder_shader_id          = 0;
4307     unsigned int i;
4308     char glsl_name[8];
4309     struct ps_compile_args ps_compile_args;
4310     struct vs_compile_args vs_compile_args;
4311
4312     if (vshader) find_vs_compile_args(state, vshader, &vs_compile_args);
4313     if (pshader) find_ps_compile_args(state, pshader, &ps_compile_args);
4314
4315     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4316     if (entry)
4317     {
4318         priv->glsl_program = entry;
4319         return;
4320     }
4321
4322     /* If we get to this point, then no matching program exists, so we create one */
4323     programId = GL_EXTCALL(glCreateProgramObjectARB());
4324     TRACE("Created new GLSL shader program %u\n", programId);
4325
4326     /* Create the entry */
4327     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4328     entry->programId = programId;
4329     entry->vshader = vshader;
4330     entry->pshader = pshader;
4331     entry->vs_args = vs_compile_args;
4332     entry->ps_args = ps_compile_args;
4333     entry->constant_version = 0;
4334     entry->np2Fixup_info = NULL;
4335     /* Add the hash table entry */
4336     add_glsl_program_entry(priv, entry);
4337
4338     /* Set the current program */
4339     priv->glsl_program = entry;
4340
4341     /* Attach GLSL vshader */
4342     if (vshader)
4343     {
4344         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer, vshader, &vs_compile_args);
4345         WORD map = vshader->reg_maps.input_registers;
4346         char tmp_name[10];
4347
4348         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4349         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4350         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4351         checkGLcall("glAttachObjectARB");
4352         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4353          * is destroyed
4354          */
4355         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4356
4357         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4358         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4359         checkGLcall("glAttachObjectARB");
4360
4361         /* Bind vertex attributes to a corresponding index number to match
4362          * the same index numbers as ARB_vertex_programs (makes loading
4363          * vertex attributes simpler).  With this method, we can use the
4364          * exact same code to load the attributes later for both ARB and
4365          * GLSL shaders.
4366          *
4367          * We have to do this here because we need to know the Program ID
4368          * in order to make the bindings work, and it has to be done prior
4369          * to linking the GLSL program. */
4370         for (i = 0; map; map >>= 1, ++i)
4371         {
4372             if (!(map & 1)) continue;
4373
4374             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4375             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4376         }
4377         checkGLcall("glBindAttribLocationARB");
4378
4379         list_add_head(&vshader->linked_programs, &entry->vshader_entry);
4380     }
4381
4382     /* Attach GLSL pshader */
4383     if (pshader)
4384     {
4385         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4386                 pshader, &ps_compile_args, &entry->np2Fixup_info);
4387         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4388         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4389         checkGLcall("glAttachObjectARB");
4390
4391         list_add_head(&pshader->linked_programs, &entry->pshader_entry);
4392     }
4393
4394     /* Link the program */
4395     TRACE("Linking GLSL shader program %u\n", programId);
4396     GL_EXTCALL(glLinkProgramARB(programId));
4397     shader_glsl_validate_link(gl_info, programId);
4398
4399     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4400             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4401     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4402     {
4403         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4404         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4405     }
4406     for (i = 0; i < MAX_CONST_I; ++i)
4407     {
4408         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4409         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4410     }
4411     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4412             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4413     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4414     {
4415         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4416         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4417     }
4418     for (i = 0; i < MAX_CONST_I; ++i)
4419     {
4420         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4421         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4422     }
4423
4424     if(pshader) {
4425         char name[32];
4426
4427         for(i = 0; i < MAX_TEXTURES; i++) {
4428             sprintf(name, "bumpenvmat%u", i);
4429             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4430             sprintf(name, "luminancescale%u", i);
4431             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4432             sprintf(name, "luminanceoffset%u", i);
4433             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4434         }
4435
4436         if (ps_compile_args.np2_fixup) {
4437             if (entry->np2Fixup_info) {
4438                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4439             } else {
4440                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4441             }
4442         }
4443     }
4444
4445     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4446     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4447     checkGLcall("Find glsl program uniform locations");
4448
4449     if (pshader && pshader->reg_maps.shader_version.major >= 3
4450             && pshader->u.ps.declared_in_count > vec4_varyings(3, gl_info))
4451     {
4452         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4453         entry->vertex_color_clamp = GL_FALSE;
4454     } else {
4455         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4456     }
4457
4458     /* Set the shader to allow uniform loading on it */
4459     GL_EXTCALL(glUseProgramObjectARB(programId));
4460     checkGLcall("glUseProgramObjectARB(programId)");
4461
4462     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4463      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4464      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4465      * vertex shader with fixed function pixel processing is used we make sure that the card
4466      * supports enough samplers to allow the max number of vertex samplers with all possible
4467      * fixed function fragment processing setups. So once the program is linked these samplers
4468      * won't change.
4469      */
4470     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4471     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4472
4473     /* If the local constants do not have to be loaded with the environment constants,
4474      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4475      * later
4476      */
4477     if (pshader && !pshader->load_local_constsF)
4478         hardcode_local_constants(pshader, gl_info, programId, 'P');
4479     if (vshader && !vshader->load_local_constsF)
4480         hardcode_local_constants(vshader, gl_info, programId, 'V');
4481 }
4482
4483 /* GL locking is done by the caller */
4484 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4485 {
4486     GLhandleARB program_id;
4487     GLhandleARB vshader_id, pshader_id;
4488     const char *blt_pshader;
4489
4490     static const char *blt_vshader =
4491         "#version 120\n"
4492         "void main(void)\n"
4493         "{\n"
4494         "    gl_Position = gl_Vertex;\n"
4495         "    gl_FrontColor = vec4(1.0);\n"
4496         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4497         "}\n";
4498
4499     static const char * const blt_pshaders_full[tex_type_count] =
4500     {
4501         /* tex_1d */
4502         NULL,
4503         /* tex_2d */
4504         "#version 120\n"
4505         "uniform sampler2D sampler;\n"
4506         "void main(void)\n"
4507         "{\n"
4508         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4509         "}\n",
4510         /* tex_3d */
4511         NULL,
4512         /* tex_cube */
4513         "#version 120\n"
4514         "uniform samplerCube sampler;\n"
4515         "void main(void)\n"
4516         "{\n"
4517         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4518         "}\n",
4519         /* tex_rect */
4520         "#version 120\n"
4521         "#extension GL_ARB_texture_rectangle : enable\n"
4522         "uniform sampler2DRect sampler;\n"
4523         "void main(void)\n"
4524         "{\n"
4525         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4526         "}\n",
4527     };
4528
4529     static const char * const blt_pshaders_masked[tex_type_count] =
4530     {
4531         /* tex_1d */
4532         NULL,
4533         /* tex_2d */
4534         "#version 120\n"
4535         "uniform sampler2D sampler;\n"
4536         "uniform vec4 mask;\n"
4537         "void main(void)\n"
4538         "{\n"
4539         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4540         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4541         "}\n",
4542         /* tex_3d */
4543         NULL,
4544         /* tex_cube */
4545         "#version 120\n"
4546         "uniform samplerCube 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 = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4552         "}\n",
4553         /* tex_rect */
4554         "#version 120\n"
4555         "#extension GL_ARB_texture_rectangle : enable\n"
4556         "uniform sampler2DRect sampler;\n"
4557         "uniform vec4 mask;\n"
4558         "void main(void)\n"
4559         "{\n"
4560         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4561         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4562         "}\n",
4563     };
4564
4565     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4566     if (!blt_pshader)
4567     {
4568         FIXME("tex_type %#x not supported\n", tex_type);
4569         return 0;
4570     }
4571
4572     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4573     shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4574
4575     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4576     shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4577
4578     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4579     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4580     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4581     GL_EXTCALL(glLinkProgramARB(program_id));
4582
4583     shader_glsl_validate_link(gl_info, program_id);
4584
4585     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4586      * is destroyed
4587      */
4588     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4589     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4590     return program_id;
4591 }
4592
4593 /* GL locking is done by the caller */
4594 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4595 {
4596     const struct wined3d_gl_info *gl_info = context->gl_info;
4597     struct wined3d_device *device = context->swapchain->device;
4598     struct shader_glsl_priv *priv = device->shader_priv;
4599     GLhandleARB program_id = 0;
4600     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4601
4602     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4603
4604     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4605     else priv->glsl_program = NULL;
4606
4607     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4608
4609     if (old_vertex_color_clamp != current_vertex_color_clamp)
4610     {
4611         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4612         {
4613             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4614             checkGLcall("glClampColorARB");
4615         }
4616         else
4617         {
4618             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4619         }
4620     }
4621
4622     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4623     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4624     GL_EXTCALL(glUseProgramObjectARB(program_id));
4625     checkGLcall("glUseProgramObjectARB");
4626
4627     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4628      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4629      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4630     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4631     {
4632         shader_glsl_load_np2fixup_constants(priv, gl_info, &device->stateBlock->state);
4633     }
4634 }
4635
4636 /* GL locking is done by the caller */
4637 static void shader_glsl_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4638         enum tex_types tex_type, const SIZE *ds_mask_size)
4639 {
4640     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4641     struct shader_glsl_priv *priv = shader_priv;
4642     GLhandleARB *blt_program;
4643     GLint loc;
4644
4645     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4646     if (!*blt_program)
4647     {
4648         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4649         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4650         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4651         GL_EXTCALL(glUniform1iARB(loc, 0));
4652     }
4653     else
4654     {
4655         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4656     }
4657
4658     if (masked)
4659     {
4660         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4661         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4662     }
4663 }
4664
4665 /* GL locking is done by the caller */
4666 static void shader_glsl_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4667 {
4668     struct shader_glsl_priv *priv = shader_priv;
4669     GLhandleARB program_id;
4670
4671     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4672     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4673
4674     GL_EXTCALL(glUseProgramObjectARB(program_id));
4675     checkGLcall("glUseProgramObjectARB");
4676 }
4677
4678 static void shader_glsl_destroy(struct wined3d_shader *shader)
4679 {
4680     struct wined3d_device *device = shader->device;
4681     struct shader_glsl_priv *priv = device->shader_priv;
4682     const struct wined3d_gl_info *gl_info;
4683     const struct list *linked_programs;
4684     struct wined3d_context *context;
4685
4686     char pshader = shader_is_pshader_version(shader->reg_maps.shader_version.type);
4687
4688     if (pshader)
4689     {
4690         struct glsl_pshader_private *shader_data = shader->backend_data;
4691
4692         if (!shader_data || !shader_data->num_gl_shaders)
4693         {
4694             HeapFree(GetProcessHeap(), 0, shader_data);
4695             shader->backend_data = NULL;
4696             return;
4697         }
4698
4699         context = context_acquire(device, NULL);
4700         gl_info = context->gl_info;
4701
4702         if (priv->glsl_program && priv->glsl_program->pshader == shader)
4703         {
4704             ENTER_GL();
4705             shader_glsl_select(context, FALSE, FALSE);
4706             LEAVE_GL();
4707         }
4708     }
4709     else
4710     {
4711         struct glsl_vshader_private *shader_data = shader->backend_data;
4712
4713         if (!shader_data || !shader_data->num_gl_shaders)
4714         {
4715             HeapFree(GetProcessHeap(), 0, shader_data);
4716             shader->backend_data = NULL;
4717             return;
4718         }
4719
4720         context = context_acquire(device, NULL);
4721         gl_info = context->gl_info;
4722
4723         if (priv->glsl_program && priv->glsl_program->vshader == shader)
4724         {
4725             ENTER_GL();
4726             shader_glsl_select(context, FALSE, FALSE);
4727             LEAVE_GL();
4728         }
4729     }
4730
4731     linked_programs = &shader->linked_programs;
4732
4733     TRACE("Deleting linked programs\n");
4734     if (linked_programs->next) {
4735         struct glsl_shader_prog_link *entry, *entry2;
4736
4737         ENTER_GL();
4738         if(pshader) {
4739             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4740                 delete_glsl_program_entry(priv, gl_info, entry);
4741             }
4742         } else {
4743             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4744                 delete_glsl_program_entry(priv, gl_info, entry);
4745             }
4746         }
4747         LEAVE_GL();
4748     }
4749
4750     if (pshader)
4751     {
4752         struct glsl_pshader_private *shader_data = shader->backend_data;
4753         UINT i;
4754
4755         ENTER_GL();
4756         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4757             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4758             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4759             checkGLcall("glDeleteObjectARB");
4760         }
4761         LEAVE_GL();
4762         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4763     }
4764     else
4765     {
4766         struct glsl_vshader_private *shader_data = shader->backend_data;
4767         UINT i;
4768
4769         ENTER_GL();
4770         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4771             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4772             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4773             checkGLcall("glDeleteObjectARB");
4774         }
4775         LEAVE_GL();
4776         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4777     }
4778
4779     HeapFree(GetProcessHeap(), 0, shader->backend_data);
4780     shader->backend_data = NULL;
4781
4782     context_release(context);
4783 }
4784
4785 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4786 {
4787     const struct glsl_program_key *k = key;
4788     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4789             const struct glsl_shader_prog_link, program_lookup_entry);
4790     int cmp;
4791
4792     if (k->vshader > prog->vshader) return 1;
4793     else if (k->vshader < prog->vshader) return -1;
4794
4795     if (k->pshader > prog->pshader) return 1;
4796     else if (k->pshader < prog->pshader) return -1;
4797
4798     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4799     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4800
4801     return 0;
4802 }
4803
4804 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4805 {
4806     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4807     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4808
4809     if (!mem)
4810     {
4811         ERR("Failed to allocate memory\n");
4812         return FALSE;
4813     }
4814
4815     heap->entries = mem;
4816     heap->entries[1].version = 0;
4817     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4818     heap->size = 1;
4819
4820     return TRUE;
4821 }
4822
4823 static void constant_heap_free(struct constant_heap *heap)
4824 {
4825     HeapFree(GetProcessHeap(), 0, heap->entries);
4826 }
4827
4828 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4829 {
4830     wined3d_rb_alloc,
4831     wined3d_rb_realloc,
4832     wined3d_rb_free,
4833     glsl_program_key_compare,
4834 };
4835
4836 static HRESULT shader_glsl_alloc(struct wined3d_device *device)
4837 {
4838     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4839     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4840     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4841             gl_info->limits.glsl_ps_float_constants)) + 1;
4842
4843     if (!shader_buffer_init(&priv->shader_buffer))
4844     {
4845         ERR("Failed to initialize shader buffer.\n");
4846         goto fail;
4847     }
4848
4849     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4850     if (!priv->stack)
4851     {
4852         ERR("Failed to allocate memory.\n");
4853         goto fail;
4854     }
4855
4856     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4857     {
4858         ERR("Failed to initialize vertex shader constant heap\n");
4859         goto fail;
4860     }
4861
4862     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4863     {
4864         ERR("Failed to initialize pixel shader constant heap\n");
4865         goto fail;
4866     }
4867
4868     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4869     {
4870         ERR("Failed to initialize rbtree.\n");
4871         goto fail;
4872     }
4873
4874     priv->next_constant_version = 1;
4875
4876     device->shader_priv = priv;
4877     return WINED3D_OK;
4878
4879 fail:
4880     constant_heap_free(&priv->pconst_heap);
4881     constant_heap_free(&priv->vconst_heap);
4882     HeapFree(GetProcessHeap(), 0, priv->stack);
4883     shader_buffer_free(&priv->shader_buffer);
4884     HeapFree(GetProcessHeap(), 0, priv);
4885     return E_OUTOFMEMORY;
4886 }
4887
4888 /* Context activation is done by the caller. */
4889 static void shader_glsl_free(struct wined3d_device *device)
4890 {
4891     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4892     struct shader_glsl_priv *priv = device->shader_priv;
4893     int i;
4894
4895     ENTER_GL();
4896     for (i = 0; i < tex_type_count; ++i)
4897     {
4898         if (priv->depth_blt_program_full[i])
4899         {
4900             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4901         }
4902         if (priv->depth_blt_program_masked[i])
4903         {
4904             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4905         }
4906     }
4907     LEAVE_GL();
4908
4909     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4910     constant_heap_free(&priv->pconst_heap);
4911     constant_heap_free(&priv->vconst_heap);
4912     HeapFree(GetProcessHeap(), 0, priv->stack);
4913     shader_buffer_free(&priv->shader_buffer);
4914
4915     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4916     device->shader_priv = NULL;
4917 }
4918
4919 static void shader_glsl_context_destroyed(void *shader_priv, const struct wined3d_context *context) {}
4920
4921 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4922 {
4923     if (gl_info->supported[EXT_GPU_SHADER4] && gl_info->supported[ARB_GEOMETRY_SHADER4]
4924             && gl_info->glsl_version >= MAKEDWORD_VERSION(1, 50))
4925     {
4926         caps->VertexShaderVersion = 4;
4927         caps->PixelShaderVersion = 4;
4928     }
4929     /* ARB_shader_texture_lod or EXT_gpu_shader4 is required for the SM3
4930      * texldd and texldl instructions. */
4931     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] || gl_info->supported[EXT_GPU_SHADER4])
4932     {
4933         caps->VertexShaderVersion = 3;
4934         caps->PixelShaderVersion = 3;
4935     }
4936     else
4937     {
4938         caps->VertexShaderVersion = 2;
4939         caps->PixelShaderVersion = 2;
4940     }
4941
4942     caps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4943     caps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4944
4945     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4946      * Direct3D minimum requirement.
4947      *
4948      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4949      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4950      *
4951      * The problem is that the refrast clamps temporary results in the shader to
4952      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4953      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4954      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4955      * offer a way to query this.
4956      */
4957     caps->PixelShader1xMaxValue = 8.0;
4958
4959     caps->VSClipping = TRUE;
4960
4961     TRACE_(d3d_caps)("Hardware vertex shader version %u enabled (GLSL).\n",
4962             caps->VertexShaderVersion);
4963     TRACE_(d3d_caps)("Hardware pixel shader version %u enabled (GLSL).\n",
4964             caps->PixelShaderVersion);
4965 }
4966
4967 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4968 {
4969     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4970     {
4971         TRACE("Checking support for fixup:\n");
4972         dump_color_fixup_desc(fixup);
4973     }
4974
4975     /* We support everything except YUV conversions. */
4976     if (!is_complex_fixup(fixup))
4977     {
4978         TRACE("[OK]\n");
4979         return TRUE;
4980     }
4981
4982     TRACE("[FAILED]\n");
4983     return FALSE;
4984 }
4985
4986 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4987 {
4988     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4989     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4990     /* WINED3DSIH_AND           */ NULL,
4991     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4992     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4993     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4994     /* WINED3DSIH_BREAKP        */ NULL,
4995     /* WINED3DSIH_CALL          */ shader_glsl_call,
4996     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4997     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4998     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4999     /* WINED3DSIH_CRS           */ shader_glsl_cross,
5000     /* WINED3DSIH_CUT           */ NULL,
5001     /* WINED3DSIH_DCL           */ NULL,
5002     /* WINED3DSIH_DEF           */ NULL,
5003     /* WINED3DSIH_DEFB          */ NULL,
5004     /* WINED3DSIH_DEFI          */ NULL,
5005     /* WINED3DSIH_DIV           */ NULL,
5006     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
5007     /* WINED3DSIH_DP3           */ shader_glsl_dot,
5008     /* WINED3DSIH_DP4           */ shader_glsl_dot,
5009     /* WINED3DSIH_DST           */ shader_glsl_dst,
5010     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5011     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5012     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5013     /* WINED3DSIH_EMIT          */ NULL,
5014     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5015     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5016     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5017     /* WINED3DSIH_EQ            */ NULL,
5018     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5019     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5020     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5021     /* WINED3DSIH_FTOI          */ NULL,
5022     /* WINED3DSIH_GE            */ NULL,
5023     /* WINED3DSIH_IADD          */ NULL,
5024     /* WINED3DSIH_IEQ           */ NULL,
5025     /* WINED3DSIH_IF            */ shader_glsl_if,
5026     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5027     /* WINED3DSIH_IGE           */ NULL,
5028     /* WINED3DSIH_IMUL          */ NULL,
5029     /* WINED3DSIH_ITOF          */ NULL,
5030     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5031     /* WINED3DSIH_LD            */ NULL,
5032     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5033     /* WINED3DSIH_LOG           */ shader_glsl_log,
5034     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5035     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5036     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5037     /* WINED3DSIH_LT            */ NULL,
5038     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5039     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5040     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5041     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5042     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5043     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5044     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5045     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5046     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5047     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5048     /* WINED3DSIH_MOVC          */ NULL,
5049     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5050     /* WINED3DSIH_NOP           */ NULL,
5051     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5052     /* WINED3DSIH_PHASE         */ NULL,
5053     /* WINED3DSIH_POW           */ shader_glsl_pow,
5054     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5055     /* WINED3DSIH_REP           */ shader_glsl_rep,
5056     /* WINED3DSIH_RET           */ shader_glsl_ret,
5057     /* WINED3DSIH_ROUND_NI      */ NULL,
5058     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5059     /* WINED3DSIH_SAMPLE        */ NULL,
5060     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5061     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5062     /* WINED3DSIH_SETP          */ NULL,
5063     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5064     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5065     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5066     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5067     /* WINED3DSIH_SQRT          */ NULL,
5068     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5069     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5070     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5071     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5072     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5073     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5074     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5075     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5076     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5077     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5078     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5079     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5080     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5081     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5082     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5083     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5084     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5085     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5086     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5087     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5088     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5089     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5090     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5091     /* WINED3DSIH_UDIV          */ NULL,
5092     /* WINED3DSIH_USHR          */ NULL,
5093     /* WINED3DSIH_UTOF          */ NULL,
5094     /* WINED3DSIH_XOR           */ NULL,
5095 };
5096
5097 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5098     SHADER_HANDLER hw_fct;
5099
5100     /* Select handler */
5101     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5102
5103     /* Unhandled opcode */
5104     if (!hw_fct)
5105     {
5106         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5107         return;
5108     }
5109     hw_fct(ins);
5110
5111     shader_glsl_add_instruction_modifiers(ins);
5112 }
5113
5114 const struct wined3d_shader_backend_ops glsl_shader_backend =
5115 {
5116     shader_glsl_handle_instruction,
5117     shader_glsl_select,
5118     shader_glsl_select_depth_blt,
5119     shader_glsl_deselect_depth_blt,
5120     shader_glsl_update_float_vertex_constants,
5121     shader_glsl_update_float_pixel_constants,
5122     shader_glsl_load_constants,
5123     shader_glsl_load_np2fixup_constants,
5124     shader_glsl_destroy,
5125     shader_glsl_alloc,
5126     shader_glsl_free,
5127     shader_glsl_context_destroyed,
5128     shader_glsl_get_caps,
5129     shader_glsl_color_fixup_supported,
5130 };