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 "wine/port.h"
  34
  35 #include <limits.h>
  36 #include <stdio.h>
  37
  38 #include "wined3d_private.h"
  39
  40 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  41 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  42 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  43 WINE_DECLARE_DEBUG_CHANNEL(winediag);
  44
  45 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  46 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  47 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  48 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  49
  50 struct glsl_dst_param
  51 {
  52     char reg_name[150];
  53     char mask_str[6];
  54 };
  55
  56 struct glsl_src_param
  57 {
  58     char reg_name[150];
  59     char param_str[200];
  60 };
  61
  62 struct glsl_sample_function
  63 {
  64     const char *name;
  65     DWORD coord_mask;
  66 };
  67
  68 enum heap_node_op
  69 {
  70     HEAP_NODE_TRAVERSE_LEFT,
  71     HEAP_NODE_TRAVERSE_RIGHT,
  72     HEAP_NODE_POP,
  73 };
  74
  75 struct constant_entry
  76 {
  77     unsigned int idx;
  78     unsigned int version;
  79 };
  80
  81 struct constant_heap
  82 {
  83     struct constant_entry *entries;
  84     unsigned int *positions;
  85     unsigned int size;
  86 };
  87
  88 /* GLSL shader private data */
  89 struct shader_glsl_priv {
  90     struct wined3d_shader_buffer shader_buffer;
  91     struct wine_rb_tree program_lookup;
  92     struct glsl_shader_prog_link *glsl_program;
  93     struct constant_heap vconst_heap;
  94     struct constant_heap pconst_heap;
  95     unsigned char *stack;
  96     GLhandleARB depth_blt_program_full[tex_type_count];
  97     GLhandleARB depth_blt_program_masked[tex_type_count];
  98     UINT next_constant_version;
  99 };
 100
 101 /* Struct to maintain data about a linked GLSL program */
 102 struct glsl_shader_prog_link {
 103     struct wine_rb_entry        program_lookup_entry;
 104     struct list                 vshader_entry;
 105     struct list                 pshader_entry;
 106     GLhandleARB                 programId;
 107     GLint                       *vuniformF_locations;
 108     GLint                       *puniformF_locations;
 109     GLint                       vuniformI_locations[MAX_CONST_I];
 110     GLint                       puniformI_locations[MAX_CONST_I];
 111     GLint                       posFixup_location;
 112     GLint                       np2Fixup_location;
 113     GLint                       bumpenvmat_location[MAX_TEXTURES];
 114     GLint                       luminancescale_location[MAX_TEXTURES];
 115     GLint                       luminanceoffset_location[MAX_TEXTURES];
 116     GLint                       ycorrection_location;
 117     GLenum                      vertex_color_clamp;
 118     const struct wined3d_shader *vshader;
 119     const struct wined3d_shader *pshader;
 120     struct vs_compile_args      vs_args;
 121     struct ps_compile_args      ps_args;
 122     UINT                        constant_version;
 123     const struct ps_np2fixup_info *np2Fixup_info;
 124 };
 125
 126 struct glsl_program_key
 127 {
 128     const struct wined3d_shader *vshader;
 129     const struct wined3d_shader *pshader;
 130     struct ps_compile_args      ps_args;
 131     struct vs_compile_args      vs_args;
 132 };
 133
 134 struct shader_glsl_ctx_priv {
 135     const struct vs_compile_args    *cur_vs_args;
 136     const struct ps_compile_args    *cur_ps_args;
 137     struct ps_np2fixup_info         *cur_np2fixup_info;
 138 };
 139
 140 struct glsl_ps_compiled_shader
 141 {
 142     struct ps_compile_args          args;
 143     struct ps_np2fixup_info         np2fixup;
 144     GLhandleARB                     prgId;
 145 };
 146
 147 struct glsl_pshader_private
 148 {
 149     struct glsl_ps_compiled_shader  *gl_shaders;
 150     UINT                            num_gl_shaders, shader_array_size;
 151 };
 152
 153 struct glsl_vs_compiled_shader
 154 {
 155     struct vs_compile_args          args;
 156     GLhandleARB                     prgId;
 157 };
 158
 159 struct glsl_vshader_private
 160 {
 161     struct glsl_vs_compiled_shader  *gl_shaders;
 162     UINT                            num_gl_shaders, shader_array_size;
 163 };
 164
 165 static const char *debug_gl_shader_type(GLenum type)
 166 {
 167     switch (type)
 168     {
 169 #define WINED3D_TO_STR(u) case u: return #u
 170         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
 171         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
 172         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
 173 #undef WINED3D_TO_STR
 174         default:
 175             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
 176     }
 177 }
 178
 179 /* Extract a line from the info log.
 180  * Note that this modifies the source string. */
 181 static char *get_info_log_line(char **ptr)
 182 {
 183     char *p, *q;
 184
 185     p = *ptr;
 186     if (!(q = strstr(p, "\n")))
 187     {
 188         if (!*p) return NULL;
 189         *ptr += strlen(p);
 190         return p;
 191     }
 192     *q = '\0';
 193     *ptr = q + 1;
 194
 195     return p;
 196 }
 197
 198 /** Prints the GLSL info log which will contain error messages if they exist */
 199 /* GL locking is done by the caller */
 200 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 201 {
 202     int infologLength = 0;
 203     char *infoLog;
 204
 205     if (!WARN_ON(d3d_shader) && !FIXME_ON(d3d_shader))
 206         return;
 207
 208     GL_EXTCALL(glGetObjectParameterivARB(obj,
 209                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 210                &infologLength));
 211
 212     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 213      * that if there are errors. */
 214     if (infologLength > 1)
 215     {
 216         char *ptr, *line;
 217
 218         infoLog = HeapAlloc(GetProcessHeap(), 0, infologLength);
 219         /* The info log is supposed to be zero-terminated, but at least some
 220          * versions of fglrx don't terminate the string properly. The reported
 221          * length does include the terminator, so explicitly set it to zero
 222          * here. */
 223         infoLog[infologLength - 1] = 0;
 224         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 225
 226         ptr = infoLog;
 227         if (gl_info->quirks & WINED3D_QUIRK_INFO_LOG_SPAM)
 228         {
 229             WARN("Info log received from GLSL shader #%u:\n", obj);
 230             while ((line = get_info_log_line(&ptr))) WARN("    %s\n", line);
 231         }
 232         else
 233         {
 234             FIXME("Info log received from GLSL shader #%u:\n", obj);
 235             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 236         }
 237         HeapFree(GetProcessHeap(), 0, infoLog);
 238     }
 239 }
 240
 241 /* GL locking is done by the caller. */
 242 static void shader_glsl_compile(const struct wined3d_gl_info *gl_info, GLhandleARB shader, const char *src)
 243 {
 244     TRACE("Compiling shader object %u.\n", shader);
 245     GL_EXTCALL(glShaderSourceARB(shader, 1, &src, NULL));
 246     checkGLcall("glShaderSourceARB");
 247     GL_EXTCALL(glCompileShaderARB(shader));
 248     checkGLcall("glCompileShaderARB");
 249     print_glsl_info_log(gl_info, shader);
 250 }
 251
 252 /* GL locking is done by the caller. */
 253 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 254 {
 255     GLint i, object_count, source_size = -1;
 256     GLhandleARB *objects;
 257     char *source = NULL;
 258
 259     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
 260     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
 261     if (!objects)
 262     {
 263         ERR("Failed to allocate object array memory.\n");
 264         return;
 265     }
 266
 267     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
 268     for (i = 0; i < object_count; ++i)
 269     {
 270         char *ptr, *line;
 271         GLint tmp;
 272
 273         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
 274
 275         if (source_size < tmp)
 276         {
 277             HeapFree(GetProcessHeap(), 0, source);
 278
 279             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
 280             if (!source)
 281             {
 282                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
 283                 HeapFree(GetProcessHeap(), 0, objects);
 284                 return;
 285             }
 286             source_size = tmp;
 287         }
 288
 289         FIXME("Object %u:\n", objects[i]);
 290         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
 291         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
 292         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
 293         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
 294         FIXME("\n");
 295
 296         ptr = source;
 297         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
 298         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 299         FIXME("\n");
 300     }
 301
 302     HeapFree(GetProcessHeap(), 0, source);
 303     HeapFree(GetProcessHeap(), 0, objects);
 304 }
 305
 306 /* GL locking is done by the caller. */
 307 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 308 {
 309     GLint tmp;
 310
 311     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 312
 313     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
 314     if (tmp == GL_PROGRAM_OBJECT_ARB)
 315     {
 316         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
 317         if (!tmp)
 318         {
 319             FIXME("Program %u link status invalid.\n", program);
 320             shader_glsl_dump_program_source(gl_info, program);
 321         }
 322     }
 323
 324     print_glsl_info_log(gl_info, program);
 325 }
 326
 327 /**
 328  * Loads (pixel shader) samplers
 329  */
 330 /* GL locking is done by the caller */
 331 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 332         const DWORD *tex_unit_map, GLhandleARB programId)
 333 {
 334     GLint name_loc;
 335     int i;
 336     char sampler_name[20];
 337
 338     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 339         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 340         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 341         if (name_loc != -1) {
 342             DWORD mapped_unit = tex_unit_map[i];
 343             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 344             {
 345                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 346                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 347                 checkGLcall("glUniform1iARB");
 348             } else {
 349                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 350             }
 351         }
 352     }
 353 }
 354
 355 /* GL locking is done by the caller */
 356 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 357         const DWORD *tex_unit_map, GLhandleARB programId)
 358 {
 359     GLint name_loc;
 360     char sampler_name[20];
 361     int i;
 362
 363     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 364         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 365         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 366         if (name_loc != -1) {
 367             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 368             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 369             {
 370                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 371                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 372                 checkGLcall("glUniform1iARB");
 373             } else {
 374                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 375             }
 376         }
 377     }
 378 }
 379
 380 /* GL locking is done by the caller */
 381 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 382         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 383 {
 384     int stack_idx = 0;
 385     unsigned int heap_idx = 1;
 386     unsigned int idx;
 387
 388     if (heap->entries[heap_idx].version <= version) return;
 389
 390     idx = heap->entries[heap_idx].idx;
 391     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 392     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 393
 394     while (stack_idx >= 0)
 395     {
 396         /* Note that we fall through to the next case statement. */
 397         switch(stack[stack_idx])
 398         {
 399             case HEAP_NODE_TRAVERSE_LEFT:
 400             {
 401                 unsigned int left_idx = heap_idx << 1;
 402                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 403                 {
 404                     heap_idx = left_idx;
 405                     idx = heap->entries[heap_idx].idx;
 406                     if (constant_locations[idx] != -1)
 407                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 408
 409                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 410                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 411                     break;
 412                 }
 413             }
 414
 415             case HEAP_NODE_TRAVERSE_RIGHT:
 416             {
 417                 unsigned int right_idx = (heap_idx << 1) + 1;
 418                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 419                 {
 420                     heap_idx = right_idx;
 421                     idx = heap->entries[heap_idx].idx;
 422                     if (constant_locations[idx] != -1)
 423                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 424
 425                     stack[stack_idx++] = HEAP_NODE_POP;
 426                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 427                     break;
 428                 }
 429             }
 430
 431             case HEAP_NODE_POP:
 432                 heap_idx >>= 1;
 433                 --stack_idx;
 434                 break;
 435         }
 436     }
 437     checkGLcall("walk_constant_heap()");
 438 }
 439
 440 /* GL locking is done by the caller */
 441 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 442 {
 443     GLfloat clamped_constant[4];
 444
 445     if (location == -1) return;
 446
 447     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 448     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 449     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 450     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 451
 452     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 453 }
 454
 455 /* GL locking is done by the caller */
 456 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 457         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 458 {
 459     int stack_idx = 0;
 460     unsigned int heap_idx = 1;
 461     unsigned int idx;
 462
 463     if (heap->entries[heap_idx].version <= version) return;
 464
 465     idx = heap->entries[heap_idx].idx;
 466     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 467     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 468
 469     while (stack_idx >= 0)
 470     {
 471         /* Note that we fall through to the next case statement. */
 472         switch(stack[stack_idx])
 473         {
 474             case HEAP_NODE_TRAVERSE_LEFT:
 475             {
 476                 unsigned int left_idx = heap_idx << 1;
 477                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 478                 {
 479                     heap_idx = left_idx;
 480                     idx = heap->entries[heap_idx].idx;
 481                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 482
 483                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 484                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 485                     break;
 486                 }
 487             }
 488
 489             case HEAP_NODE_TRAVERSE_RIGHT:
 490             {
 491                 unsigned int right_idx = (heap_idx << 1) + 1;
 492                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 493                 {
 494                     heap_idx = right_idx;
 495                     idx = heap->entries[heap_idx].idx;
 496                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 497
 498                     stack[stack_idx++] = HEAP_NODE_POP;
 499                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 500                     break;
 501                 }
 502             }
 503
 504             case HEAP_NODE_POP:
 505                 heap_idx >>= 1;
 506                 --stack_idx;
 507                 break;
 508         }
 509     }
 510     checkGLcall("walk_constant_heap_clamped()");
 511 }
 512
 513 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 514 /* GL locking is done by the caller */
 515 static void shader_glsl_load_constantsF(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 516         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 517         unsigned char *stack, UINT version)
 518 {
 519     const struct wined3d_shader_lconst *lconst;
 520
 521     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 522     if (shader->reg_maps.shader_version.major == 1
 523             && shader_is_pshader_version(shader->reg_maps.shader_version.type))
 524         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 525     else
 526         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 527
 528     if (!shader->load_local_constsF)
 529     {
 530         TRACE("No need to load local float constants for this shader\n");
 531         return;
 532     }
 533
 534     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 535     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 536     {
 537         GLint location = constant_locations[lconst->idx];
 538         /* We found this uniform name in the program - go ahead and send the data */
 539         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 540     }
 541     checkGLcall("glUniform4fvARB()");
 542 }
 543
 544 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 545 /* GL locking is done by the caller */
 546 static void shader_glsl_load_constantsI(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 547         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 548 {
 549     unsigned int i;
 550     struct list* ptr;
 551
 552     for (i = 0; constants_set; constants_set >>= 1, ++i)
 553     {
 554         if (!(constants_set & 1)) continue;
 555
 556         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 557                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 558
 559         /* We found this uniform name in the program - go ahead and send the data */
 560         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 561         checkGLcall("glUniform4ivARB");
 562     }
 563
 564     /* Load immediate constants */
 565     ptr = list_head(&shader->constantsI);
 566     while (ptr)
 567     {
 568         const struct wined3d_shader_lconst *lconst = LIST_ENTRY(ptr, const struct wined3d_shader_lconst, entry);
 569         unsigned int idx = lconst->idx;
 570         const GLint *values = (const GLint *)lconst->value;
 571
 572         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 573             values[0], values[1], values[2], values[3]);
 574
 575         /* We found this uniform name in the program - go ahead and send the data */
 576         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 577         checkGLcall("glUniform4ivARB");
 578         ptr = list_next(&shader->constantsI, ptr);
 579     }
 580 }
 581
 582 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 583 /* GL locking is done by the caller */
 584 static void shader_glsl_load_constantsB(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
 585         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 586 {
 587     GLint tmp_loc;
 588     unsigned int i;
 589     char tmp_name[8];
 590     const char *prefix;
 591     struct list* ptr;
 592
 593     switch (shader->reg_maps.shader_version.type)
 594     {
 595         case WINED3D_SHADER_TYPE_VERTEX:
 596             prefix = "VB";
 597             break;
 598
 599         case WINED3D_SHADER_TYPE_GEOMETRY:
 600             prefix = "GB";
 601             break;
 602
 603         case WINED3D_SHADER_TYPE_PIXEL:
 604             prefix = "PB";
 605             break;
 606
 607         default:
 608             FIXME("Unknown shader type %#x.\n",
 609                     shader->reg_maps.shader_version.type);
 610             prefix = "UB";
 611             break;
 612     }
 613
 614     /* TODO: Benchmark and see if it would be beneficial to store the
 615      * locations of the constants to avoid looking up each time */
 616     for (i = 0; constants_set; constants_set >>= 1, ++i)
 617     {
 618         if (!(constants_set & 1)) continue;
 619
 620         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 621
 622         /* TODO: Benchmark and see if it would be beneficial to store the
 623          * locations of the constants to avoid looking up each time */
 624         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 625         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 626         if (tmp_loc != -1)
 627         {
 628             /* We found this uniform name in the program - go ahead and send the data */
 629             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 630             checkGLcall("glUniform1ivARB");
 631         }
 632     }
 633
 634     /* Load immediate constants */
 635     ptr = list_head(&shader->constantsB);
 636     while (ptr)
 637     {
 638         const struct wined3d_shader_lconst *lconst = LIST_ENTRY(ptr, const struct wined3d_shader_lconst, entry);
 639         unsigned int idx = lconst->idx;
 640         const GLint *values = (const GLint *)lconst->value;
 641
 642         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 643
 644         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 645         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 646         if (tmp_loc != -1) {
 647             /* We found this uniform name in the program - go ahead and send the data */
 648             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 649             checkGLcall("glUniform1ivARB");
 650         }
 651         ptr = list_next(&shader->constantsB, ptr);
 652     }
 653 }
 654
 655 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 656 {
 657     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 658 }
 659
 660 /**
 661  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 662  */
 663 /* GL locking is done by the caller (state handler) */
 664 static void shader_glsl_load_np2fixup_constants(void *shader_priv,
 665         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 666 {
 667     struct shader_glsl_priv *glsl_priv = shader_priv;
 668     const struct glsl_shader_prog_link *prog = glsl_priv->glsl_program;
 669
 670     /* No GLSL program set - nothing to do. */
 671     if (!prog) return;
 672
 673     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 674     if (!use_ps(state)) return;
 675
 676     if (prog->ps_args.np2_fixup && prog->np2Fixup_location != -1)
 677     {
 678         UINT i;
 679         UINT fixup = prog->ps_args.np2_fixup;
 680         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 681
 682         for (i = 0; fixup; fixup >>= 1, ++i)
 683         {
 684             const struct wined3d_texture *tex = state->textures[i];
 685             const unsigned char idx = prog->np2Fixup_info->idx[i];
 686             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 687
 688             if (!tex)
 689             {
 690                 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
 691                 continue;
 692             }
 693
 694             if (idx % 2)
 695             {
 696                 tex_dim[2] = tex->pow2_matrix[0];
 697                 tex_dim[3] = tex->pow2_matrix[5];
 698             }
 699             else
 700             {
 701                 tex_dim[0] = tex->pow2_matrix[0];
 702                 tex_dim[1] = tex->pow2_matrix[5];
 703             }
 704         }
 705
 706         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 707     }
 708 }
 709
 710 /**
 711  * Loads the app-supplied constants into the currently set GLSL program.
 712  */
 713 /* GL locking is done by the caller (state handler) */
 714 static void shader_glsl_load_constants(const struct wined3d_context *context,
 715         char usePixelShader, char useVertexShader)
 716 {
 717     const struct wined3d_gl_info *gl_info = context->gl_info;
 718     struct wined3d_device *device = context->swapchain->device;
 719     struct wined3d_stateblock *stateBlock = device->stateBlock;
 720     const struct wined3d_state *state = &stateBlock->state;
 721     struct shader_glsl_priv *priv = device->shader_priv;
 722     float position_fixup[4];
 723
 724     GLhandleARB programId;
 725     struct glsl_shader_prog_link *prog = priv->glsl_program;
 726     UINT constant_version;
 727     int i;
 728
 729     if (!prog) {
 730         /* No GLSL program set - nothing to do. */
 731         return;
 732     }
 733     programId = prog->programId;
 734     constant_version = prog->constant_version;
 735
 736     if (useVertexShader)
 737     {
 738         const struct wined3d_shader *vshader = state->vertex_shader;
 739
 740         /* Load DirectX 9 float constants/uniforms for vertex shader */
 741         shader_glsl_load_constantsF(vshader, gl_info, state->vs_consts_f,
 742                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 743
 744         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 745         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, state->vs_consts_i,
 746                 stateBlock->changed.vertexShaderConstantsI & vshader->reg_maps.integer_constants);
 747
 748         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 749         shader_glsl_load_constantsB(vshader, gl_info, programId, state->vs_consts_b,
 750                 stateBlock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants);
 751
 752         /* Upload the position fixup params */
 753         shader_get_position_fixup(context, state, position_fixup);
 754         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, position_fixup));
 755         checkGLcall("glUniform4fvARB");
 756     }
 757
 758     if (usePixelShader)
 759     {
 760         const struct wined3d_shader *pshader = state->pixel_shader;
 761
 762         /* Load DirectX 9 float constants/uniforms for pixel shader */
 763         shader_glsl_load_constantsF(pshader, gl_info, state->ps_consts_f,
 764                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 765
 766         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 767         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, state->ps_consts_i,
 768                 stateBlock->changed.pixelShaderConstantsI & pshader->reg_maps.integer_constants);
 769
 770         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 771         shader_glsl_load_constantsB(pshader, gl_info, programId, state->ps_consts_b,
 772                 stateBlock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants);
 773
 774         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 775          * It can't be 0 for a valid texbem instruction.
 776          */
 777         for(i = 0; i < MAX_TEXTURES; i++) {
 778             const float *data;
 779
 780             if(prog->bumpenvmat_location[i] == -1) continue;
 781
 782             data = (const float *)&state->texture_states[i][WINED3D_TSS_BUMPENV_MAT00];
 783             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 784             checkGLcall("glUniformMatrix2fvARB");
 785
 786             /* texbeml needs the luminance scale and offset too. If texbeml
 787              * is used, needsbumpmat is set too, so we can check that in the
 788              * needsbumpmat check. */
 789             if (prog->luminancescale_location[i] != -1)
 790             {
 791                 const GLfloat *scale = (const GLfloat *)&state->texture_states[i][WINED3D_TSS_BUMPENV_LSCALE];
 792                 const GLfloat *offset = (const GLfloat *)&state->texture_states[i][WINED3D_TSS_BUMPENV_LOFFSET];
 793
 794                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 795                 checkGLcall("glUniform1fvARB");
 796                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 797                 checkGLcall("glUniform1fvARB");
 798             }
 799         }
 800
 801         if (prog->ycorrection_location != -1)
 802         {
 803             float correction_params[4];
 804
 805             if (context->render_offscreen)
 806             {
 807                 correction_params[0] = 0.0f;
 808                 correction_params[1] = 1.0f;
 809             } else {
 810                 /* position is window relative, not viewport relative */
 811                 correction_params[0] = (float) context->current_rt->resource.height;
 812                 correction_params[1] = -1.0f;
 813             }
 814             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 815         }
 816     }
 817
 818     if (priv->next_constant_version == UINT_MAX)
 819     {
 820         TRACE("Max constant version reached, resetting to 0.\n");
 821         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 822         priv->next_constant_version = 1;
 823     }
 824     else
 825     {
 826         prog->constant_version = priv->next_constant_version++;
 827     }
 828 }
 829
 830 static void update_heap_entry(const struct constant_heap *heap, unsigned int idx,
 831         unsigned int heap_idx, DWORD new_version)
 832 {
 833     struct constant_entry *entries = heap->entries;
 834     unsigned int *positions = heap->positions;
 835     unsigned int parent_idx;
 836
 837     while (heap_idx > 1)
 838     {
 839         parent_idx = heap_idx >> 1;
 840
 841         if (new_version <= entries[parent_idx].version) break;
 842
 843         entries[heap_idx] = entries[parent_idx];
 844         positions[entries[parent_idx].idx] = heap_idx;
 845         heap_idx = parent_idx;
 846     }
 847
 848     entries[heap_idx].version = new_version;
 849     entries[heap_idx].idx = idx;
 850     positions[idx] = heap_idx;
 851 }
 852
 853 static void shader_glsl_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 854 {
 855     struct shader_glsl_priv *priv = device->shader_priv;
 856     struct constant_heap *heap = &priv->vconst_heap;
 857     UINT i;
 858
 859     for (i = start; i < count + start; ++i)
 860     {
 861         if (!device->stateBlock->changed.vertexShaderConstantsF[i])
 862             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 863         else
 864             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 865     }
 866 }
 867
 868 static void shader_glsl_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 869 {
 870     struct shader_glsl_priv *priv = device->shader_priv;
 871     struct constant_heap *heap = &priv->pconst_heap;
 872     UINT i;
 873
 874     for (i = start; i < count + start; ++i)
 875     {
 876         if (!device->stateBlock->changed.pixelShaderConstantsF[i])
 877             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 878         else
 879             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 880     }
 881 }
 882
 883 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 884 {
 885     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 886     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 887     if(shader_major > 3) return ret;
 888
 889     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 890     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 891     return ret;
 892 }
 893
 894 /** Generate the variable & register declarations for the GLSL output target */
 895 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 896         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
 897         const struct wined3d_shader_reg_maps *reg_maps, const struct shader_glsl_ctx_priv *ctx_priv)
 898 {
 899     const struct wined3d_state *state = &shader->device->stateBlock->state;
 900     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 901     const struct wined3d_gl_info *gl_info = context->gl_info;
 902     const struct wined3d_fb_state *fb = &shader->device->fb;
 903     unsigned int i, extra_constants_needed = 0;
 904     const struct wined3d_shader_lconst *lconst;
 905     DWORD map;
 906
 907     /* There are some minor differences between pixel and vertex shaders */
 908     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 909     char prefix = pshader ? 'P' : 'V';
 910
 911     /* Prototype the subroutines */
 912     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 913     {
 914         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 915     }
 916
 917     /* Declare the constants (aka uniforms) */
 918     if (shader->limits.constant_float > 0)
 919     {
 920         unsigned max_constantsF;
 921         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 922          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 923          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 924          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 925          * a dx9 card, as long as it doesn't also use all the other constants.
 926          *
 927          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 928          * declare only the amount that we're assured to have.
 929          *
 930          * Thus we run into problems in these two cases:
 931          * 1) The shader really uses more uniforms than supported
 932          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 933          */
 934         if (pshader)
 935         {
 936             /* No indirect addressing here. */
 937             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 938         }
 939         else
 940         {
 941             if (reg_maps->usesrelconstF)
 942             {
 943                 /* Subtract the other potential uniforms from the max
 944                  * available (bools, ints, and 1 row of projection matrix).
 945                  * Subtract another uniform for immediate values, which have
 946                  * to be loaded via uniform by the driver as well. The shader
 947                  * code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex
 948                  * shader code, so one vec4 should be enough. (Unfortunately
 949                  * the Nvidia driver doesn't store 128 and -128 in one float).
 950                  *
 951                  * Writing gl_ClipVertex requires one uniform for each
 952                  * clipplane as well. */
 953                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 954                 if(ctx_priv->cur_vs_args->clip_enabled)
 955                 {
 956                     max_constantsF -= gl_info->limits.clipplanes;
 957                 }
 958                 max_constantsF -= count_bits(reg_maps->integer_constants);
 959                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 960                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 961                  * for now take this into account when calculating the number of available constants
 962                  */
 963                 max_constantsF -= count_bits(reg_maps->boolean_constants);
 964                 /* Set by driver quirks in directx.c */
 965                 max_constantsF -= gl_info->reserved_glsl_constants;
 966
 967                 if (max_constantsF < shader->limits.constant_float)
 968                 {
 969                     static unsigned int once;
 970
 971                     if (!once++)
 972                         ERR_(winediag)("The hardware does not support enough uniform components to run this shader,"
 973                                 " it may not render correctly.\n");
 974                     else
 975                         WARN("The hardware does not support enough uniform components to run this shader.\n");
 976                 }
 977             }
 978             else
 979             {
 980                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 981             }
 982         }
 983         max_constantsF = min(shader->limits.constant_float, max_constantsF);
 984         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 985     }
 986
 987     /* Always declare the full set of constants, the compiler can remove the
 988      * unused ones because d3d doesn't (yet) support indirect int and bool
 989      * constant addressing. This avoids problems if the app uses e.g. i0 and i9. */
 990     if (shader->limits.constant_int > 0 && reg_maps->integer_constants)
 991         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, shader->limits.constant_int);
 992
 993     if (shader->limits.constant_bool > 0 && reg_maps->boolean_constants)
 994         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, shader->limits.constant_bool);
 995
 996     if (!pshader)
 997     {
 998         shader_addline(buffer, "uniform vec4 posFixup;\n");
 999         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
1000     }
1001     else
1002     {
1003         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1004         {
1005             if (!(map & 1)) continue;
1006
1007             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1008
1009             if (reg_maps->luminanceparams & (1 << i))
1010             {
1011                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1012                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1013                 extra_constants_needed++;
1014             }
1015
1016             extra_constants_needed++;
1017         }
1018
1019         if (ps_args->srgb_correction)
1020         {
1021             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1022                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1023             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1024                     srgb_cmp);
1025         }
1026         if (reg_maps->vpos || reg_maps->usesdsy)
1027         {
1028             if (shader->limits.constant_float + extra_constants_needed
1029                     + 1 < gl_info->limits.glsl_ps_float_constants)
1030             {
1031                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1032                 extra_constants_needed++;
1033             }
1034             else
1035             {
1036                 /* This happens because we do not have proper tracking of the constant registers that are
1037                  * actually used, only the max limit of the shader version
1038                  */
1039                 FIXME("Cannot find a free uniform for vpos correction params\n");
1040                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1041                         context->render_offscreen ? 0.0f : fb->render_targets[0]->resource.height,
1042                         context->render_offscreen ? 1.0f : -1.0f);
1043             }
1044             shader_addline(buffer, "vec4 vpos;\n");
1045         }
1046     }
1047
1048     /* Declare texture samplers */
1049     for (i = 0; i < shader->limits.sampler; ++i)
1050     {
1051         if (reg_maps->sampler_type[i])
1052         {
1053             const struct wined3d_texture *texture;
1054
1055             switch (reg_maps->sampler_type[i])
1056             {
1057                 case WINED3DSTT_1D:
1058                     if (pshader && ps_args->shadow & (1 << i))
1059                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1060                     else
1061                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1062                     break;
1063                 case WINED3DSTT_2D:
1064                     texture = state->textures[i];
1065                     if (pshader && ps_args->shadow & (1 << i))
1066                     {
1067                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1068                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1069                         else
1070                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1071                     }
1072                     else
1073                     {
1074                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1075                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1076                         else
1077                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1078                     }
1079                     break;
1080                 case WINED3DSTT_CUBE:
1081                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1082                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1083                     break;
1084                 case WINED3DSTT_VOLUME:
1085                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1086                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1087                     break;
1088                 default:
1089                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1090                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1091                     break;
1092             }
1093         }
1094     }
1095
1096     /* Declare uniforms for NP2 texcoord fixup:
1097      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1098      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1099      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1100     if (pshader && ps_args->np2_fixup) {
1101
1102         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1103         UINT cur = 0;
1104
1105         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1106          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1107          * samplerNP2Fixup stores texture dimensions and is updated through
1108          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1109
1110         for (i = 0; i < shader->limits.sampler; ++i)
1111         {
1112             if (reg_maps->sampler_type[i])
1113             {
1114                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1115
1116                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1117                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1118                     continue;
1119                 }
1120
1121                 fixup->idx[i] = cur++;
1122             }
1123         }
1124
1125         fixup->num_consts = (cur + 1) >> 1;
1126         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1127     }
1128
1129     /* Declare address variables */
1130     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1131     {
1132         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1133     }
1134
1135     /* Declare texture coordinate temporaries and initialize them */
1136     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1137     {
1138         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1139     }
1140
1141     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1142      * helper function shader that is linked in at link time
1143      */
1144     if (pshader && reg_maps->shader_version.major >= 3)
1145     {
1146         UINT in_count = min(vec4_varyings(reg_maps->shader_version.major, gl_info), shader->limits.packed_input);
1147
1148         if (use_vs(state))
1149             shader_addline(buffer, "varying vec4 IN[%u];\n", in_count);
1150         else
1151             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1152              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1153              * pixel shader that reads the fixed function color into the packed input registers. */
1154             shader_addline(buffer, "vec4 IN[%u];\n", in_count);
1155     }
1156
1157     /* Declare output register temporaries */
1158     if (shader->limits.packed_output)
1159         shader_addline(buffer, "vec4 OUT[%u];\n", shader->limits.packed_output);
1160
1161     /* Declare temporary variables */
1162     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1163     {
1164         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1165     }
1166
1167     /* Declare attributes */
1168     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1169     {
1170         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1171         {
1172             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1173         }
1174     }
1175
1176     /* Declare loop registers aLx */
1177     for (i = 0; i < reg_maps->loop_depth; i++) {
1178         shader_addline(buffer, "int aL%u;\n", i);
1179         shader_addline(buffer, "int tmpInt%u;\n", i);
1180     }
1181
1182     /* Temporary variables for matrix operations */
1183     shader_addline(buffer, "vec4 tmp0;\n");
1184     shader_addline(buffer, "vec4 tmp1;\n");
1185
1186     /* Local constants use a different name so they can be loaded once at shader link time
1187      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1188      * float -> string conversion can cause precision loss.
1189      */
1190     if (!shader->load_local_constsF)
1191     {
1192         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
1193         {
1194             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1195         }
1196     }
1197
1198     /* Start the main program */
1199     shader_addline(buffer, "void main() {\n");
1200     if(pshader && reg_maps->vpos) {
1201         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1202          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1203          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1204          * precision troubles when we just subtract 0.5.
1205          *
1206          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1207          *
1208          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1209          *
1210          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1211          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1212          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1213          * correctly on drivers that returns integer values.
1214          */
1215         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1216     }
1217 }
1218
1219 /*****************************************************************************
1220  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1221  *
1222  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1223  ****************************************************************************/
1224
1225 /* Prototypes */
1226 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1227         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src);
1228
1229 /** Used for opcode modifiers - They multiply the result by the specified amount */
1230 static const char * const shift_glsl_tab[] = {
1231     "",           /*  0 (none) */
1232     "2.0 * ",     /*  1 (x2)   */
1233     "4.0 * ",     /*  2 (x4)   */
1234     "8.0 * ",     /*  3 (x8)   */
1235     "16.0 * ",    /*  4 (x16)  */
1236     "32.0 * ",    /*  5 (x32)  */
1237     "",           /*  6 (x64)  */
1238     "",           /*  7 (x128) */
1239     "",           /*  8 (d256) */
1240     "",           /*  9 (d128) */
1241     "",           /* 10 (d64)  */
1242     "",           /* 11 (d32)  */
1243     "0.0625 * ",  /* 12 (d16)  */
1244     "0.125 * ",   /* 13 (d8)   */
1245     "0.25 * ",    /* 14 (d4)   */
1246     "0.5 * "      /* 15 (d2)   */
1247 };
1248
1249 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1250 static void shader_glsl_gen_modifier(enum wined3d_shader_src_modifier src_modifier,
1251         const char *in_reg, const char *in_regswizzle, char *out_str)
1252 {
1253     out_str[0] = 0;
1254
1255     switch (src_modifier)
1256     {
1257     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1258     case WINED3DSPSM_DW:
1259     case WINED3DSPSM_NONE:
1260         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1261         break;
1262     case WINED3DSPSM_NEG:
1263         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1264         break;
1265     case WINED3DSPSM_NOT:
1266         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1267         break;
1268     case WINED3DSPSM_BIAS:
1269         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1270         break;
1271     case WINED3DSPSM_BIASNEG:
1272         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1273         break;
1274     case WINED3DSPSM_SIGN:
1275         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1276         break;
1277     case WINED3DSPSM_SIGNNEG:
1278         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1279         break;
1280     case WINED3DSPSM_COMP:
1281         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1282         break;
1283     case WINED3DSPSM_X2:
1284         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1285         break;
1286     case WINED3DSPSM_X2NEG:
1287         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1288         break;
1289     case WINED3DSPSM_ABS:
1290         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1291         break;
1292     case WINED3DSPSM_ABSNEG:
1293         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1294         break;
1295     default:
1296         FIXME("Unhandled modifier %u\n", src_modifier);
1297         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1298     }
1299 }
1300
1301 /** Writes the GLSL variable name that corresponds to the register that the
1302  * DX opcode parameter is trying to access */
1303 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1304         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1305 {
1306     /* oPos, oFog and oPts in D3D */
1307     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1308
1309     const struct wined3d_shader *shader = ins->ctx->shader;
1310     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1311     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1312     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1313
1314     *is_color = FALSE;
1315
1316     switch (reg->type)
1317     {
1318         case WINED3DSPR_TEMP:
1319             sprintf(register_name, "R%u", reg->idx);
1320             break;
1321
1322         case WINED3DSPR_INPUT:
1323             /* vertex shaders */
1324             if (!pshader)
1325             {
1326                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1327                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1328                 sprintf(register_name, "attrib%u", reg->idx);
1329                 break;
1330             }
1331
1332             /* pixel shaders >= 3.0 */
1333             if (reg_maps->shader_version.major >= 3)
1334             {
1335                 DWORD idx = shader->u.ps.input_reg_map[reg->idx];
1336                 unsigned int in_count = vec4_varyings(reg_maps->shader_version.major, gl_info);
1337
1338                 if (reg->rel_addr)
1339                 {
1340                     struct glsl_src_param rel_param;
1341
1342                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1343
1344                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1345                      * operation there */
1346                     if (idx)
1347                     {
1348                         if (shader->u.ps.declared_in_count > in_count)
1349                         {
1350                             sprintf(register_name,
1351                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1352                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1353                                     rel_param.param_str, idx);
1354                         }
1355                         else
1356                         {
1357                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1358                         }
1359                     }
1360                     else
1361                     {
1362                         if (shader->u.ps.declared_in_count > in_count)
1363                         {
1364                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1365                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1366                                     rel_param.param_str);
1367                         }
1368                         else
1369                         {
1370                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1371                         }
1372                     }
1373                 }
1374                 else
1375                 {
1376                     if (idx == in_count) sprintf(register_name, "gl_Color");
1377                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1378                     else sprintf(register_name, "IN[%u]", idx);
1379                 }
1380             }
1381             else
1382             {
1383                 if (!reg->idx) strcpy(register_name, "gl_Color");
1384                 else strcpy(register_name, "gl_SecondaryColor");
1385                 break;
1386             }
1387             break;
1388
1389         case WINED3DSPR_CONST:
1390             {
1391                 const char prefix = pshader ? 'P' : 'V';
1392
1393                 /* Relative addressing */
1394                 if (reg->rel_addr)
1395                 {
1396                     struct glsl_src_param rel_param;
1397                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1398                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1399                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1400                 }
1401                 else
1402                 {
1403                     if (shader_constant_is_local(shader, reg->idx))
1404                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1405                     else
1406                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1407                 }
1408             }
1409             break;
1410
1411         case WINED3DSPR_CONSTINT:
1412             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1413             else sprintf(register_name, "VI[%u]", reg->idx);
1414             break;
1415
1416         case WINED3DSPR_CONSTBOOL:
1417             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1418             else sprintf(register_name, "VB[%u]", reg->idx);
1419             break;
1420
1421         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1422             if (pshader) sprintf(register_name, "T%u", reg->idx);
1423             else sprintf(register_name, "A%u", reg->idx);
1424             break;
1425
1426         case WINED3DSPR_LOOP:
1427             sprintf(register_name, "aL%u", ins->ctx->loop_state->current_reg - 1);
1428             break;
1429
1430         case WINED3DSPR_SAMPLER:
1431             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1432             else sprintf(register_name, "Vsampler%u", reg->idx);
1433             break;
1434
1435         case WINED3DSPR_COLOROUT:
1436             if (reg->idx >= gl_info->limits.buffers)
1437                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1438
1439             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1440             break;
1441
1442         case WINED3DSPR_RASTOUT:
1443             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1444             break;
1445
1446         case WINED3DSPR_DEPTHOUT:
1447             sprintf(register_name, "gl_FragDepth");
1448             break;
1449
1450         case WINED3DSPR_ATTROUT:
1451             if (!reg->idx) sprintf(register_name, "OUT[8]");
1452             else sprintf(register_name, "OUT[9]");
1453             break;
1454
1455         case WINED3DSPR_TEXCRDOUT:
1456             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1457             sprintf(register_name, "OUT[%u]", reg->idx);
1458             break;
1459
1460         case WINED3DSPR_MISCTYPE:
1461             if (!reg->idx)
1462             {
1463                 /* vPos */
1464                 sprintf(register_name, "vpos");
1465             }
1466             else if (reg->idx == 1)
1467             {
1468                 /* Note that gl_FrontFacing is a bool, while vFace is
1469                  * a float for which the sign determines front/back */
1470                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1471             }
1472             else
1473             {
1474                 FIXME("Unhandled misctype register %d\n", reg->idx);
1475                 sprintf(register_name, "unrecognized_register");
1476             }
1477             break;
1478
1479         case WINED3DSPR_IMMCONST:
1480             switch (reg->immconst_type)
1481             {
1482                 case WINED3D_IMMCONST_SCALAR:
1483                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1484                     break;
1485
1486                 case WINED3D_IMMCONST_VEC4:
1487                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1488                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1489                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1490                     break;
1491
1492                 default:
1493                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1494                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1495             }
1496             break;
1497
1498         default:
1499             FIXME("Unhandled register name Type(%d)\n", reg->type);
1500             sprintf(register_name, "unrecognized_register");
1501             break;
1502     }
1503 }
1504
1505 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1506 {
1507     *str++ = '.';
1508     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1509     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1510     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1511     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1512     *str = '\0';
1513 }
1514
1515 /* Get the GLSL write mask for the destination register */
1516 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1517 {
1518     DWORD mask = param->write_mask;
1519
1520     if (shader_is_scalar(&param->reg))
1521     {
1522         mask = WINED3DSP_WRITEMASK_0;
1523         *write_mask = '\0';
1524     }
1525     else
1526     {
1527         shader_glsl_write_mask_to_str(mask, write_mask);
1528     }
1529
1530     return mask;
1531 }
1532
1533 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1534     unsigned int size = 0;
1535
1536     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1537     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1538     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1539     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1540
1541     return size;
1542 }
1543
1544 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1545 {
1546     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1547      * but addressed as "rgba". To fix this we need to swap the register's x
1548      * and z components. */
1549     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1550
1551     *str++ = '.';
1552     /* swizzle bits fields: wwzzyyxx */
1553     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1554     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1555     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1556     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1557     *str = '\0';
1558 }
1559
1560 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1561         BOOL fixup, DWORD mask, char *swizzle_str)
1562 {
1563     if (shader_is_scalar(&param->reg))
1564         *swizzle_str = '\0';
1565     else
1566         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1567 }
1568
1569 /* From a given parameter token, generate the corresponding GLSL string.
1570  * Also, return the actual register name and swizzle in case the
1571  * caller needs this information as well. */
1572 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1573         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src)
1574 {
1575     BOOL is_color = FALSE;
1576     char swizzle_str[6];
1577
1578     glsl_src->reg_name[0] = '\0';
1579     glsl_src->param_str[0] = '\0';
1580     swizzle_str[0] = '\0';
1581
1582     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1583     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1584     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1585 }
1586
1587 /* From a given parameter token, generate the corresponding GLSL string.
1588  * Also, return the actual register name and swizzle in case the
1589  * caller needs this information as well. */
1590 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1591         const struct wined3d_shader_dst_param *wined3d_dst, struct glsl_dst_param *glsl_dst)
1592 {
1593     BOOL is_color = FALSE;
1594
1595     glsl_dst->mask_str[0] = '\0';
1596     glsl_dst->reg_name[0] = '\0';
1597
1598     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1599     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1600 }
1601
1602 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1603 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1604         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1605 {
1606     struct glsl_dst_param glsl_dst;
1607     DWORD mask;
1608
1609     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1610     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1611
1612     return mask;
1613 }
1614
1615 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1616 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1617 {
1618     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1619 }
1620
1621 /** Process GLSL instruction modifiers */
1622 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1623 {
1624     struct glsl_dst_param dst_param;
1625     DWORD modifiers;
1626
1627     if (!ins->dst_count) return;
1628
1629     modifiers = ins->dst[0].modifiers;
1630     if (!modifiers) return;
1631
1632     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1633
1634     if (modifiers & WINED3DSPDM_SATURATE)
1635     {
1636         /* _SAT means to clamp the value of the register to between 0 and 1 */
1637         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1638                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1639     }
1640
1641     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1642     {
1643         FIXME("_centroid modifier not handled\n");
1644     }
1645
1646     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1647     {
1648         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1649     }
1650 }
1651
1652 static const char *shader_glsl_get_rel_op(enum wined3d_shader_rel_op op)
1653 {
1654     switch (op)
1655     {
1656         case WINED3D_SHADER_REL_OP_GT: return ">";
1657         case WINED3D_SHADER_REL_OP_EQ: return "==";
1658         case WINED3D_SHADER_REL_OP_GE: return ">=";
1659         case WINED3D_SHADER_REL_OP_LT: return "<";
1660         case WINED3D_SHADER_REL_OP_NE: return "!=";
1661         case WINED3D_SHADER_REL_OP_LE: return "<=";
1662         default:
1663             FIXME("Unrecognized operator %#x.\n", op);
1664             return "(\?\?)";
1665     }
1666 }
1667
1668 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1669         DWORD sampler_idx, DWORD flags, struct glsl_sample_function *sample_function)
1670 {
1671     enum wined3d_sampler_texture_type sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1672     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1673     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1674             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1675     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1676     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1677     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1678     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1679
1680     /* Note that there's no such thing as a projected cube texture. */
1681     switch(sampler_type) {
1682         case WINED3DSTT_1D:
1683             if (shadow)
1684             {
1685                 if (lod)
1686                 {
1687                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1688                 }
1689                 else if (grad)
1690                 {
1691                     if (gl_info->supported[EXT_GPU_SHADER4])
1692                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1693                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1694                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1695                     else
1696                     {
1697                         FIXME("Unsupported 1D shadow grad function.\n");
1698                         sample_function->name = "unsupported1DGrad";
1699                     }
1700                 }
1701                 else
1702                 {
1703                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1704                 }
1705                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1706             }
1707             else
1708             {
1709                 if (lod)
1710                 {
1711                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1712                 }
1713                 else if (grad)
1714                 {
1715                     if (gl_info->supported[EXT_GPU_SHADER4])
1716                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1717                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1718                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1719                     else
1720                     {
1721                         FIXME("Unsupported 1D grad function.\n");
1722                         sample_function->name = "unsupported1DGrad";
1723                     }
1724                 }
1725                 else
1726                 {
1727                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1728                 }
1729                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1730             }
1731             break;
1732
1733         case WINED3DSTT_2D:
1734             if (shadow)
1735             {
1736                 if (texrect)
1737                 {
1738                     if (lod)
1739                     {
1740                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1741                     }
1742                     else if (grad)
1743                     {
1744                         if (gl_info->supported[EXT_GPU_SHADER4])
1745                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1746                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1747                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1748                         else
1749                         {
1750                             FIXME("Unsupported RECT shadow grad function.\n");
1751                             sample_function->name = "unsupported2DRectGrad";
1752                         }
1753                     }
1754                     else
1755                     {
1756                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1757                     }
1758                 }
1759                 else
1760                 {
1761                     if (lod)
1762                     {
1763                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1764                     }
1765                     else if (grad)
1766                     {
1767                         if (gl_info->supported[EXT_GPU_SHADER4])
1768                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1769                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1770                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1771                         else
1772                         {
1773                             FIXME("Unsupported 2D shadow grad function.\n");
1774                             sample_function->name = "unsupported2DGrad";
1775                         }
1776                     }
1777                     else
1778                     {
1779                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1780                     }
1781                 }
1782                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1783             }
1784             else
1785             {
1786                 if (texrect)
1787                 {
1788                     if (lod)
1789                     {
1790                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1791                     }
1792                     else if (grad)
1793                     {
1794                         if (gl_info->supported[EXT_GPU_SHADER4])
1795                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1796                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1797                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1798                         else
1799                         {
1800                             FIXME("Unsupported RECT grad function.\n");
1801                             sample_function->name = "unsupported2DRectGrad";
1802                         }
1803                     }
1804                     else
1805                     {
1806                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1807                     }
1808                 }
1809                 else
1810                 {
1811                     if (lod)
1812                     {
1813                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1814                     }
1815                     else if (grad)
1816                     {
1817                         if (gl_info->supported[EXT_GPU_SHADER4])
1818                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1819                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1820                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1821                         else
1822                         {
1823                             FIXME("Unsupported 2D grad function.\n");
1824                             sample_function->name = "unsupported2DGrad";
1825                         }
1826                     }
1827                     else
1828                     {
1829                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1830                     }
1831                 }
1832                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1833             }
1834             break;
1835
1836         case WINED3DSTT_CUBE:
1837             if (shadow)
1838             {
1839                 FIXME("Unsupported Cube shadow function.\n");
1840                 sample_function->name = "unsupportedCubeShadow";
1841                 sample_function->coord_mask = 0;
1842             }
1843             else
1844             {
1845                 if (lod)
1846                 {
1847                     sample_function->name = "textureCubeLod";
1848                 }
1849                 else if (grad)
1850                 {
1851                     if (gl_info->supported[EXT_GPU_SHADER4])
1852                         sample_function->name = "textureCubeGrad";
1853                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1854                         sample_function->name = "textureCubeGradARB";
1855                     else
1856                     {
1857                         FIXME("Unsupported Cube grad function.\n");
1858                         sample_function->name = "unsupportedCubeGrad";
1859                     }
1860                 }
1861                 else
1862                 {
1863                     sample_function->name = "textureCube";
1864                 }
1865                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1866             }
1867             break;
1868
1869         case WINED3DSTT_VOLUME:
1870             if (shadow)
1871             {
1872                 FIXME("Unsupported 3D shadow function.\n");
1873                 sample_function->name = "unsupported3DShadow";
1874                 sample_function->coord_mask = 0;
1875             }
1876             else
1877             {
1878                 if (lod)
1879                 {
1880                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1881                 }
1882                 else  if (grad)
1883                 {
1884                     if (gl_info->supported[EXT_GPU_SHADER4])
1885                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1886                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1887                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1888                     else
1889                     {
1890                         FIXME("Unsupported 3D grad function.\n");
1891                         sample_function->name = "unsupported3DGrad";
1892                     }
1893                 }
1894                 else
1895                 {
1896                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1897                 }
1898                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1899             }
1900             break;
1901
1902         default:
1903             sample_function->name = "";
1904             sample_function->coord_mask = 0;
1905             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1906             break;
1907     }
1908 }
1909
1910 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1911         BOOL sign_fixup, enum fixup_channel_source channel_source)
1912 {
1913     switch(channel_source)
1914     {
1915         case CHANNEL_SOURCE_ZERO:
1916             strcat(arguments, "0.0");
1917             break;
1918
1919         case CHANNEL_SOURCE_ONE:
1920             strcat(arguments, "1.0");
1921             break;
1922
1923         case CHANNEL_SOURCE_X:
1924             strcat(arguments, reg_name);
1925             strcat(arguments, ".x");
1926             break;
1927
1928         case CHANNEL_SOURCE_Y:
1929             strcat(arguments, reg_name);
1930             strcat(arguments, ".y");
1931             break;
1932
1933         case CHANNEL_SOURCE_Z:
1934             strcat(arguments, reg_name);
1935             strcat(arguments, ".z");
1936             break;
1937
1938         case CHANNEL_SOURCE_W:
1939             strcat(arguments, reg_name);
1940             strcat(arguments, ".w");
1941             break;
1942
1943         default:
1944             FIXME("Unhandled channel source %#x\n", channel_source);
1945             strcat(arguments, "undefined");
1946             break;
1947     }
1948
1949     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1950 }
1951
1952 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1953 {
1954     struct wined3d_shader_dst_param dst;
1955     unsigned int mask_size, remaining;
1956     struct glsl_dst_param dst_param;
1957     char arguments[256];
1958     DWORD mask;
1959
1960     mask = 0;
1961     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1962     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1963     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1964     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1965     mask &= ins->dst[0].write_mask;
1966
1967     if (!mask) return; /* Nothing to do */
1968
1969     if (is_complex_fixup(fixup))
1970     {
1971         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1972         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1973         return;
1974     }
1975
1976     mask_size = shader_glsl_get_write_mask_size(mask);
1977
1978     dst = ins->dst[0];
1979     dst.write_mask = mask;
1980     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1981
1982     arguments[0] = '\0';
1983     remaining = mask_size;
1984     if (mask & WINED3DSP_WRITEMASK_0)
1985     {
1986         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1987         if (--remaining) strcat(arguments, ", ");
1988     }
1989     if (mask & WINED3DSP_WRITEMASK_1)
1990     {
1991         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1992         if (--remaining) strcat(arguments, ", ");
1993     }
1994     if (mask & WINED3DSP_WRITEMASK_2)
1995     {
1996         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1997         if (--remaining) strcat(arguments, ", ");
1998     }
1999     if (mask & WINED3DSP_WRITEMASK_3)
2000     {
2001         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
2002         if (--remaining) strcat(arguments, ", ");
2003     }
2004
2005     if (mask_size > 1)
2006     {
2007         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
2008                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
2009     }
2010     else
2011     {
2012         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2013     }
2014 }
2015
2016 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2017         DWORD sampler, const struct glsl_sample_function *sample_function, DWORD swizzle,
2018         const char *dx, const char *dy, const char *bias, const char *coord_reg_fmt, ...)
2019 {
2020     const char *sampler_base;
2021     char dst_swizzle[6];
2022     struct color_fixup_desc fixup;
2023     BOOL np2_fixup = FALSE;
2024     va_list args;
2025
2026     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2027
2028     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2029     {
2030         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2031         fixup = priv->cur_ps_args->color_fixup[sampler];
2032         sampler_base = "Psampler";
2033
2034         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2035             if(bias) {
2036                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2037             } else {
2038                 np2_fixup = TRUE;
2039             }
2040         }
2041     } else {
2042         sampler_base = "Vsampler";
2043         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2044     }
2045
2046     shader_glsl_append_dst(ins->ctx->buffer, ins);
2047
2048     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2049
2050     va_start(args, coord_reg_fmt);
2051     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2052     va_end(args);
2053
2054     if(bias) {
2055         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2056     } else {
2057         if (np2_fixup) {
2058             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2059             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2060
2061             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2062                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2063         } else if(dx && dy) {
2064             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2065         } else {
2066             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2067         }
2068     }
2069
2070     if(!is_identity_fixup(fixup)) {
2071         shader_glsl_color_correction(ins, fixup);
2072     }
2073 }
2074
2075 /*****************************************************************************
2076  * Begin processing individual instruction opcodes
2077  ****************************************************************************/
2078
2079 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2080 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2081 {
2082     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2083     struct glsl_src_param src0_param;
2084     struct glsl_src_param src1_param;
2085     DWORD write_mask;
2086     char op;
2087
2088     /* Determine the GLSL operator to use based on the opcode */
2089     switch (ins->handler_idx)
2090     {
2091         case WINED3DSIH_MUL: op = '*'; break;
2092         case WINED3DSIH_ADD: op = '+'; break;
2093         case WINED3DSIH_SUB: op = '-'; break;
2094         default:
2095             op = ' ';
2096             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2097             break;
2098     }
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     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2103     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2104 }
2105
2106 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2107 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2108 {
2109     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2110     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2111     struct glsl_src_param src0_param;
2112     DWORD write_mask;
2113
2114     write_mask = shader_glsl_append_dst(buffer, ins);
2115     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2116
2117     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2118      * shader versions WINED3DSIO_MOVA is used for this. */
2119     if (ins->ctx->reg_maps->shader_version.major == 1
2120             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2121             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2122     {
2123         /* This is a simple floor() */
2124         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2125         if (mask_size > 1) {
2126             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2127         } else {
2128             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2129         }
2130     }
2131     else if(ins->handler_idx == WINED3DSIH_MOVA)
2132     {
2133         /* We need to *round* to the nearest int here. */
2134         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2135
2136         if (gl_info->supported[EXT_GPU_SHADER4])
2137         {
2138             if (mask_size > 1)
2139                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2140             else
2141                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2142         }
2143         else
2144         {
2145             if (mask_size > 1)
2146                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2147                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2148             else
2149                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2150                         src0_param.param_str, src0_param.param_str);
2151         }
2152     }
2153     else
2154     {
2155         shader_addline(buffer, "%s);\n", src0_param.param_str);
2156     }
2157 }
2158
2159 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2160 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2161 {
2162     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2163     struct glsl_src_param src0_param;
2164     struct glsl_src_param src1_param;
2165     DWORD dst_write_mask, src_write_mask;
2166     unsigned int dst_size = 0;
2167
2168     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2169     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2170
2171     /* dp3 works on vec3, dp4 on vec4 */
2172     if (ins->handler_idx == WINED3DSIH_DP4)
2173     {
2174         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2175     } else {
2176         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2177     }
2178
2179     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2180     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2181
2182     if (dst_size > 1) {
2183         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2184     } else {
2185         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2186     }
2187 }
2188
2189 /* Note that this instruction has some restrictions. The destination write mask
2190  * can't contain the w component, and the source swizzles have to be .xyzw */
2191 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2192 {
2193     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2194     struct glsl_src_param src0_param;
2195     struct glsl_src_param src1_param;
2196     char dst_mask[6];
2197
2198     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2199     shader_glsl_append_dst(ins->ctx->buffer, ins);
2200     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2201     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2202     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2203 }
2204
2205 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2206  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2207  * GLSL uses the value as-is. */
2208 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2209 {
2210     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2211     struct glsl_src_param src0_param;
2212     struct glsl_src_param src1_param;
2213     DWORD dst_write_mask;
2214     unsigned int dst_size;
2215
2216     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2217     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2218
2219     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2220     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2221
2222     if (dst_size > 1)
2223     {
2224         shader_addline(buffer, "vec%u(%s == 0.0 ? 1.0 : pow(abs(%s), %s)));\n",
2225                 dst_size, src1_param.param_str, src0_param.param_str, src1_param.param_str);
2226     }
2227     else
2228     {
2229         shader_addline(buffer, "%s == 0.0 ? 1.0 : pow(abs(%s), %s));\n",
2230                 src1_param.param_str, src0_param.param_str, src1_param.param_str);
2231     }
2232 }
2233
2234 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2235  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2236  * GLSL uses the value as-is. */
2237 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2238 {
2239     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2240     struct glsl_src_param src0_param;
2241     DWORD dst_write_mask;
2242     unsigned int dst_size;
2243
2244     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2245     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2246
2247     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2248
2249     if (dst_size > 1)
2250     {
2251         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2252                 dst_size, src0_param.param_str);
2253     }
2254     else
2255     {
2256         shader_addline(buffer, "log2(abs(%s)));\n",
2257                 src0_param.param_str);
2258     }
2259 }
2260
2261 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2262 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2263 {
2264     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2265     struct glsl_src_param src_param;
2266     const char *instruction;
2267     DWORD write_mask;
2268     unsigned i;
2269
2270     /* Determine the GLSL function to use based on the opcode */
2271     /* TODO: Possibly make this a table for faster lookups */
2272     switch (ins->handler_idx)
2273     {
2274         case WINED3DSIH_MIN: instruction = "min"; break;
2275         case WINED3DSIH_MAX: instruction = "max"; break;
2276         case WINED3DSIH_ABS: instruction = "abs"; break;
2277         case WINED3DSIH_FRC: instruction = "fract"; break;
2278         case WINED3DSIH_EXP: instruction = "exp2"; break;
2279         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2280         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2281         default: instruction = "";
2282             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2283             break;
2284     }
2285
2286     write_mask = shader_glsl_append_dst(buffer, ins);
2287
2288     shader_addline(buffer, "%s(", instruction);
2289
2290     if (ins->src_count)
2291     {
2292         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2293         shader_addline(buffer, "%s", src_param.param_str);
2294         for (i = 1; i < ins->src_count; ++i)
2295         {
2296             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2297             shader_addline(buffer, ", %s", src_param.param_str);
2298         }
2299     }
2300
2301     shader_addline(buffer, "));\n");
2302 }
2303
2304 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2305 {
2306     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2307     struct glsl_src_param src_param;
2308     unsigned int mask_size;
2309     DWORD write_mask;
2310     char dst_mask[6];
2311
2312     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2313     mask_size = shader_glsl_get_write_mask_size(write_mask);
2314     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2315
2316     shader_addline(buffer, "tmp0.x = dot(%s, %s);\n",
2317             src_param.param_str, src_param.param_str);
2318     shader_glsl_append_dst(buffer, ins);
2319
2320     if (mask_size > 1)
2321     {
2322         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s * inversesqrt(tmp0.x)));\n",
2323                 mask_size, src_param.param_str);
2324     }
2325     else
2326     {
2327         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s * inversesqrt(tmp0.x)));\n",
2328                 src_param.param_str);
2329     }
2330 }
2331
2332 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2333  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2334  *   dst.x = 2^(floor(src))
2335  *   dst.y = src - floor(src)
2336  *   dst.z = 2^src   (partial precision is allowed, but optional)
2337  *   dst.w = 1.0;
2338  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2339  *   dst = 2^src;    (partial precision is allowed, but optional)
2340  */
2341 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2342 {
2343     struct glsl_src_param src_param;
2344
2345     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2346
2347     if (ins->ctx->reg_maps->shader_version.major < 2)
2348     {
2349         char dst_mask[6];
2350
2351         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2352         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2353         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2354         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2355
2356         shader_glsl_append_dst(ins->ctx->buffer, ins);
2357         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2358         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2359     } else {
2360         DWORD write_mask;
2361         unsigned int mask_size;
2362
2363         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2364         mask_size = shader_glsl_get_write_mask_size(write_mask);
2365
2366         if (mask_size > 1) {
2367             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2368         } else {
2369             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2370         }
2371     }
2372 }
2373
2374 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2375 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2376 {
2377     struct glsl_src_param src_param;
2378     DWORD write_mask;
2379     unsigned int mask_size;
2380
2381     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2382     mask_size = shader_glsl_get_write_mask_size(write_mask);
2383     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2384
2385     if (mask_size > 1)
2386     {
2387         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2388                 mask_size, src_param.param_str);
2389     }
2390     else
2391     {
2392         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2393                 src_param.param_str);
2394     }
2395 }
2396
2397 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2398 {
2399     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2400     struct glsl_src_param src_param;
2401     DWORD write_mask;
2402     unsigned int mask_size;
2403
2404     write_mask = shader_glsl_append_dst(buffer, ins);
2405     mask_size = shader_glsl_get_write_mask_size(write_mask);
2406
2407     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2408
2409     if (mask_size > 1)
2410     {
2411         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2412                 mask_size, src_param.param_str);
2413     }
2414     else
2415     {
2416         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2417                 src_param.param_str);
2418     }
2419 }
2420
2421 /** Process signed comparison opcodes in GLSL. */
2422 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2423 {
2424     struct glsl_src_param src0_param;
2425     struct glsl_src_param src1_param;
2426     DWORD write_mask;
2427     unsigned int mask_size;
2428
2429     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2430     mask_size = shader_glsl_get_write_mask_size(write_mask);
2431     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2432     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2433
2434     if (mask_size > 1) {
2435         const char *compare;
2436
2437         switch(ins->handler_idx)
2438         {
2439             case WINED3DSIH_SLT: compare = "lessThan"; break;
2440             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2441             default: compare = "";
2442                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2443         }
2444
2445         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2446                 src0_param.param_str, src1_param.param_str);
2447     } else {
2448         switch(ins->handler_idx)
2449         {
2450             case WINED3DSIH_SLT:
2451                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2452                  * to return 0.0 but step returns 1.0 because step is not < x
2453                  * An alternative is a bvec compare padded with an unused second component.
2454                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2455                  * issue. Playing with not() is not possible either because not() does not accept
2456                  * a scalar.
2457                  */
2458                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2459                         src0_param.param_str, src1_param.param_str);
2460                 break;
2461             case WINED3DSIH_SGE:
2462                 /* Here we can use the step() function and safe a conditional */
2463                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2464                 break;
2465             default:
2466                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2467         }
2468
2469     }
2470 }
2471
2472 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2473 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2474 {
2475     struct glsl_src_param src0_param;
2476     struct glsl_src_param src1_param;
2477     struct glsl_src_param src2_param;
2478     DWORD write_mask, cmp_channel = 0;
2479     unsigned int i, j;
2480     char mask_char[6];
2481     BOOL temp_destination = FALSE;
2482
2483     if (shader_is_scalar(&ins->src[0].reg))
2484     {
2485         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2486
2487         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2488         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2489         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2490
2491         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2492                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2493     } else {
2494         DWORD dst_mask = ins->dst[0].write_mask;
2495         struct wined3d_shader_dst_param dst = ins->dst[0];
2496
2497         /* Cycle through all source0 channels */
2498         for (i=0; i<4; i++) {
2499             write_mask = 0;
2500             /* Find the destination channels which use the current source0 channel */
2501             for (j=0; j<4; j++) {
2502                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2503                 {
2504                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2505                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2506                 }
2507             }
2508             dst.write_mask = dst_mask & write_mask;
2509
2510             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2511             * The first lines may overwrite source parameters of the following lines.
2512             * Deal with that by using a temporary destination register if needed
2513             */
2514             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2515                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2516                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2517                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2518                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2519                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2520             {
2521                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2522                 if (!write_mask) continue;
2523                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2524                 temp_destination = TRUE;
2525             } else {
2526                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2527                 if (!write_mask) continue;
2528             }
2529
2530             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2531             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2532             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2533
2534             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2535                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2536         }
2537
2538         if(temp_destination) {
2539             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2540             shader_glsl_append_dst(ins->ctx->buffer, ins);
2541             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2542         }
2543     }
2544
2545 }
2546
2547 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2548 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2549  * the compare is done per component of src0. */
2550 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2551 {
2552     struct wined3d_shader_dst_param dst;
2553     struct glsl_src_param src0_param;
2554     struct glsl_src_param src1_param;
2555     struct glsl_src_param src2_param;
2556     DWORD write_mask, cmp_channel = 0;
2557     unsigned int i, j;
2558     DWORD dst_mask;
2559     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2560             ins->ctx->reg_maps->shader_version.minor);
2561
2562     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2563     {
2564         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2565         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2566         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2567         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2568
2569         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2570         if (ins->coissue)
2571         {
2572             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2573         } else {
2574             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2575                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2576         }
2577         return;
2578     }
2579     /* Cycle through all source0 channels */
2580     dst_mask = ins->dst[0].write_mask;
2581     dst = ins->dst[0];
2582     for (i=0; i<4; i++) {
2583         write_mask = 0;
2584         /* Find the destination channels which use the current source0 channel */
2585         for (j=0; j<4; j++) {
2586             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2587             {
2588                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2589                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2590             }
2591         }
2592
2593         dst.write_mask = dst_mask & write_mask;
2594         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2595         if (!write_mask) continue;
2596
2597         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2598         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2599         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2600
2601         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2602                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2603     }
2604 }
2605
2606 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2607 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2608 {
2609     struct glsl_src_param src0_param;
2610     struct glsl_src_param src1_param;
2611     struct glsl_src_param src2_param;
2612     DWORD write_mask;
2613
2614     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2615     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2616     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2617     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2618     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2619             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2620 }
2621
2622 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2623    Vertex shaders to GLSL codes */
2624 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2625 {
2626     int i;
2627     int nComponents = 0;
2628     struct wined3d_shader_dst_param tmp_dst = {{0}};
2629     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2630     struct wined3d_shader_instruction tmp_ins;
2631
2632     memset(&tmp_ins, 0, sizeof(tmp_ins));
2633
2634     /* Set constants for the temporary argument */
2635     tmp_ins.ctx = ins->ctx;
2636     tmp_ins.dst_count = 1;
2637     tmp_ins.dst = &tmp_dst;
2638     tmp_ins.src_count = 2;
2639     tmp_ins.src = tmp_src;
2640
2641     switch(ins->handler_idx)
2642     {
2643         case WINED3DSIH_M4x4:
2644             nComponents = 4;
2645             tmp_ins.handler_idx = WINED3DSIH_DP4;
2646             break;
2647         case WINED3DSIH_M4x3:
2648             nComponents = 3;
2649             tmp_ins.handler_idx = WINED3DSIH_DP4;
2650             break;
2651         case WINED3DSIH_M3x4:
2652             nComponents = 4;
2653             tmp_ins.handler_idx = WINED3DSIH_DP3;
2654             break;
2655         case WINED3DSIH_M3x3:
2656             nComponents = 3;
2657             tmp_ins.handler_idx = WINED3DSIH_DP3;
2658             break;
2659         case WINED3DSIH_M3x2:
2660             nComponents = 2;
2661             tmp_ins.handler_idx = WINED3DSIH_DP3;
2662             break;
2663         default:
2664             break;
2665     }
2666
2667     tmp_dst = ins->dst[0];
2668     tmp_src[0] = ins->src[0];
2669     tmp_src[1] = ins->src[1];
2670     for (i = 0; i < nComponents; ++i)
2671     {
2672         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2673         shader_glsl_dot(&tmp_ins);
2674         ++tmp_src[1].reg.idx;
2675     }
2676 }
2677
2678 /**
2679     The LRP instruction performs a component-wise linear interpolation
2680     between the second and third operands using the first operand as the
2681     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2682     This is equivalent to mix(src2, src1, src0);
2683 */
2684 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2685 {
2686     struct glsl_src_param src0_param;
2687     struct glsl_src_param src1_param;
2688     struct glsl_src_param src2_param;
2689     DWORD write_mask;
2690
2691     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2692
2693     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2694     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2695     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2696
2697     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2698             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2699 }
2700
2701 /** Process the WINED3DSIO_LIT instruction in GLSL:
2702  * dst.x = dst.w = 1.0
2703  * dst.y = (src0.x > 0) ? src0.x
2704  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2705  *                                        where src.w is clamped at +- 128
2706  */
2707 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2708 {
2709     struct glsl_src_param src0_param;
2710     struct glsl_src_param src1_param;
2711     struct glsl_src_param src3_param;
2712     char dst_mask[6];
2713
2714     shader_glsl_append_dst(ins->ctx->buffer, ins);
2715     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2716
2717     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2718     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2719     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2720
2721     /* The sdk specifies the instruction like this
2722      * dst.x = 1.0;
2723      * if(src.x > 0.0) dst.y = src.x
2724      * else dst.y = 0.0.
2725      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2726      * else dst.z = 0.0;
2727      * dst.w = 1.0;
2728      * (where power = src.w clamped between -128 and 128)
2729      *
2730      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2731      * dst.x = 1.0                                  ... No further explanation needed
2732      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2733      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2734      * dst.w = 1.0.                                 ... Nothing fancy.
2735      *
2736      * So we still have one conditional in there. So do this:
2737      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2738      *
2739      * step(0.0, x) will return 1 if src.x > 0.0, and 0 otherwise. So if y is 0 we get pow(0.0 * 1.0, power),
2740      * which sets dst.z to 0. If y > 0, but x = 0.0, we get pow(y * 0.0, power), which results in 0 too.
2741      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to.
2742      *
2743      * Unfortunately pow(0.0 ^ 0.0) returns NaN on most GPUs, but lit with src.y = 0 and src.w = 0 returns
2744      * a non-NaN value in dst.z. What we return doesn't matter, as long as it is not NaN. Return 0, which is
2745      * what all Windows HW drivers and GL_ARB_vertex_program's LIT do.
2746      */
2747     shader_addline(ins->ctx->buffer,
2748             "vec4(1.0, max(%s, 0.0), %s == 0.0 ? 0.0 : "
2749             "pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
2750             src0_param.param_str, src3_param.param_str, src1_param.param_str,
2751             src0_param.param_str, src3_param.param_str, dst_mask);
2752 }
2753
2754 /** Process the WINED3DSIO_DST instruction in GLSL:
2755  * dst.x = 1.0
2756  * dst.y = src0.x * src0.y
2757  * dst.z = src0.z
2758  * dst.w = src1.w
2759  */
2760 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2761 {
2762     struct glsl_src_param src0y_param;
2763     struct glsl_src_param src0z_param;
2764     struct glsl_src_param src1y_param;
2765     struct glsl_src_param src1w_param;
2766     char dst_mask[6];
2767
2768     shader_glsl_append_dst(ins->ctx->buffer, ins);
2769     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2770
2771     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2772     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2773     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2774     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2775
2776     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2777             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2778 }
2779
2780 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2781  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2782  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2783  *
2784  * dst.x = cos(src0.?)
2785  * dst.y = sin(src0.?)
2786  * dst.z = dst.z
2787  * dst.w = dst.w
2788  */
2789 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2790 {
2791     struct glsl_src_param src0_param;
2792     DWORD write_mask;
2793
2794     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2795     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2796
2797     switch (write_mask) {
2798         case WINED3DSP_WRITEMASK_0:
2799             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2800             break;
2801
2802         case WINED3DSP_WRITEMASK_1:
2803             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2804             break;
2805
2806         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2807             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2808             break;
2809
2810         default:
2811             ERR("Write mask should be .x, .y or .xy\n");
2812             break;
2813     }
2814 }
2815
2816 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2817  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2818  * generate invalid code
2819  */
2820 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2821 {
2822     struct glsl_src_param src0_param;
2823     DWORD write_mask;
2824
2825     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2826     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2827
2828     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2829 }
2830
2831 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2832  * Start a for() loop where src1.y is the initial value of aL,
2833  *  increment aL by src1.z for a total of src1.x iterations.
2834  *  Need to use a temporary variable for this operation.
2835  */
2836 /* FIXME: I don't think nested loops will work correctly this way. */
2837 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2838 {
2839     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2840     const struct wined3d_shader *shader = ins->ctx->shader;
2841     const struct wined3d_shader_lconst *constant;
2842     struct glsl_src_param src1_param;
2843     const DWORD *control_values = NULL;
2844
2845     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2846
2847     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2848      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2849      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2850      * addressing.
2851      */
2852     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2853     {
2854         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2855         {
2856             if (constant->idx == ins->src[1].reg.idx)
2857             {
2858                 control_values = constant->value;
2859                 break;
2860             }
2861         }
2862     }
2863
2864     if (control_values)
2865     {
2866         struct wined3d_shader_loop_control loop_control;
2867         loop_control.count = control_values[0];
2868         loop_control.start = control_values[1];
2869         loop_control.step = (int)control_values[2];
2870
2871         if (loop_control.step > 0)
2872         {
2873             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2874                     loop_state->current_depth, loop_control.start,
2875                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2876                     loop_state->current_depth, loop_control.step);
2877         }
2878         else if (loop_control.step < 0)
2879         {
2880             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2881                     loop_state->current_depth, loop_control.start,
2882                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2883                     loop_state->current_depth, loop_control.step);
2884         }
2885         else
2886         {
2887             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2888                     loop_state->current_depth, loop_control.start, loop_state->current_depth,
2889                     loop_state->current_depth, loop_control.count,
2890                     loop_state->current_depth);
2891         }
2892     } else {
2893         shader_addline(ins->ctx->buffer,
2894                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2895                 loop_state->current_depth, loop_state->current_reg,
2896                 src1_param.reg_name, loop_state->current_depth, src1_param.reg_name,
2897                 loop_state->current_depth, loop_state->current_reg, src1_param.reg_name);
2898     }
2899
2900     ++loop_state->current_depth;
2901     ++loop_state->current_reg;
2902 }
2903
2904 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2905 {
2906     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2907
2908     shader_addline(ins->ctx->buffer, "}\n");
2909
2910     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2911     {
2912         --loop_state->current_depth;
2913         --loop_state->current_reg;
2914     }
2915
2916     if (ins->handler_idx == WINED3DSIH_ENDREP)
2917     {
2918         --loop_state->current_depth;
2919     }
2920 }
2921
2922 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2923 {
2924     const struct wined3d_shader *shader = ins->ctx->shader;
2925     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2926     const struct wined3d_shader_lconst *constant;
2927     struct glsl_src_param src0_param;
2928     const DWORD *control_values = NULL;
2929
2930     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2931     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2932     {
2933         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2934         {
2935             if (constant->idx == ins->src[0].reg.idx)
2936             {
2937                 control_values = constant->value;
2938                 break;
2939             }
2940         }
2941     }
2942
2943     if (control_values)
2944     {
2945         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2946                 loop_state->current_depth, loop_state->current_depth,
2947                 control_values[0], loop_state->current_depth);
2948     }
2949     else
2950     {
2951         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2952         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2953                 loop_state->current_depth, loop_state->current_depth,
2954                 src0_param.param_str, loop_state->current_depth);
2955     }
2956
2957     ++loop_state->current_depth;
2958 }
2959
2960 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2961 {
2962     struct glsl_src_param src0_param;
2963
2964     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2965     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2966 }
2967
2968 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2969 {
2970     struct glsl_src_param src0_param;
2971     struct glsl_src_param src1_param;
2972
2973     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2974     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2975
2976     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2977             src0_param.param_str, shader_glsl_get_rel_op(ins->flags), src1_param.param_str);
2978 }
2979
2980 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2981 {
2982     shader_addline(ins->ctx->buffer, "} else {\n");
2983 }
2984
2985 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2986 {
2987     shader_addline(ins->ctx->buffer, "break;\n");
2988 }
2989
2990 /* FIXME: According to MSDN the compare is done per component. */
2991 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2992 {
2993     struct glsl_src_param src0_param;
2994     struct glsl_src_param src1_param;
2995
2996     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2997     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2998
2999     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
3000             src0_param.param_str, shader_glsl_get_rel_op(ins->flags), src1_param.param_str);
3001 }
3002
3003 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
3004 {
3005     shader_addline(ins->ctx->buffer, "}\n");
3006     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
3007 }
3008
3009 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
3010 {
3011     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
3012 }
3013
3014 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3015 {
3016     struct glsl_src_param src1_param;
3017
3018     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3019     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3020 }
3021
3022 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3023 {
3024     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3025      * function only suppresses the unhandled instruction warning
3026      */
3027 }
3028
3029 /*********************************************
3030  * Pixel Shader Specific Code begins here
3031  ********************************************/
3032 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3033 {
3034     const struct wined3d_shader *shader = ins->ctx->shader;
3035     struct wined3d_device *device = shader->device;
3036     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3037             ins->ctx->reg_maps->shader_version.minor);
3038     struct glsl_sample_function sample_function;
3039     const struct wined3d_texture *texture;
3040     DWORD sample_flags = 0;
3041     DWORD sampler_idx;
3042     DWORD mask = 0, swizzle;
3043
3044     /* 1.0-1.4: Use destination register as sampler source.
3045      * 2.0+: Use provided sampler source. */
3046     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3047     else sampler_idx = ins->src[1].reg.idx;
3048     texture = device->stateBlock->state.textures[sampler_idx];
3049
3050     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3051     {
3052         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3053         DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3054                 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3055         enum wined3d_sampler_texture_type sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3056
3057         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3058         if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE)
3059         {
3060             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3061             switch (flags & ~WINED3D_PSARGS_PROJECTED)
3062             {
3063                 case WINED3D_TTFF_COUNT1:
3064                     FIXME("WINED3D_TTFF_PROJECTED with WINED3D_TTFF_COUNT1?\n");
3065                     break;
3066                 case WINED3D_TTFF_COUNT2:
3067                     mask = WINED3DSP_WRITEMASK_1;
3068                     break;
3069                 case WINED3D_TTFF_COUNT3:
3070                     mask = WINED3DSP_WRITEMASK_2;
3071                     break;
3072                 case WINED3D_TTFF_COUNT4:
3073                 case WINED3D_TTFF_DISABLE:
3074                     mask = WINED3DSP_WRITEMASK_3;
3075                     break;
3076             }
3077         }
3078     }
3079     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3080     {
3081         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
3082
3083         if (src_mod == WINED3DSPSM_DZ) {
3084             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3085             mask = WINED3DSP_WRITEMASK_2;
3086         } else if (src_mod == WINED3DSPSM_DW) {
3087             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3088             mask = WINED3DSP_WRITEMASK_3;
3089         }
3090     } else {
3091         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3092         {
3093             /* ps 2.0 texldp instruction always divides by the fourth component. */
3094             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3095             mask = WINED3DSP_WRITEMASK_3;
3096         }
3097     }
3098
3099     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3100         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3101
3102     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3103     mask |= sample_function.coord_mask;
3104
3105     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3106     else swizzle = ins->src[1].swizzle;
3107
3108     /* 1.0-1.3: Use destination register as coordinate source.
3109        1.4+: Use provided coordinate source register. */
3110     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3111     {
3112         char coord_mask[6];
3113         shader_glsl_write_mask_to_str(mask, coord_mask);
3114         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3115                 "T%u%s", sampler_idx, coord_mask);
3116     }
3117     else
3118     {
3119         struct glsl_src_param coord_param;
3120         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3121         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3122         {
3123             struct glsl_src_param bias;
3124             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3125             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3126                     "%s", coord_param.param_str);
3127         } else {
3128             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3129                     "%s", coord_param.param_str);
3130         }
3131     }
3132 }
3133
3134 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3135 {
3136     const struct wined3d_shader *shader = ins->ctx->shader;
3137     struct wined3d_device *device = shader->device;
3138     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3139     struct glsl_src_param coord_param, dx_param, dy_param;
3140     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3141     struct glsl_sample_function sample_function;
3142     DWORD sampler_idx;
3143     DWORD swizzle = ins->src[1].swizzle;
3144     const struct wined3d_texture *texture;
3145
3146     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3147     {
3148         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3149         shader_glsl_tex(ins);
3150         return;
3151     }
3152
3153     sampler_idx = ins->src[1].reg.idx;
3154     texture = device->stateBlock->state.textures[sampler_idx];
3155     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3156         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3157
3158     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3159     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3160     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3161     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3162
3163     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3164                                 "%s", coord_param.param_str);
3165 }
3166
3167 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3168 {
3169     const struct wined3d_shader *shader = ins->ctx->shader;
3170     struct wined3d_device *device = shader->device;
3171     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3172     struct glsl_src_param coord_param, lod_param;
3173     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3174     struct glsl_sample_function sample_function;
3175     DWORD sampler_idx;
3176     DWORD swizzle = ins->src[1].swizzle;
3177     const struct wined3d_texture *texture;
3178
3179     sampler_idx = ins->src[1].reg.idx;
3180     texture = device->stateBlock->state.textures[sampler_idx];
3181     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3182         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3183
3184     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3185     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3186
3187     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3188
3189     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3190             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3191     {
3192         /* Plain GLSL only supports Lod sampling functions in vertex shaders.
3193          * However, the NVIDIA drivers allow them in fragment shaders as well,
3194          * even without the appropriate extension. */
3195         WARN("Using %s in fragment shader.\n", sample_function.name);
3196     }
3197     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3198             "%s", coord_param.param_str);
3199 }
3200
3201 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3202 {
3203     /* FIXME: Make this work for more than just 2D textures */
3204     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3205     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3206
3207     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3208     {
3209         char dst_mask[6];
3210
3211         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3212         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3213                 ins->dst[0].reg.idx, dst_mask);
3214     }
3215     else
3216     {
3217         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
3218         DWORD reg = ins->src[0].reg.idx;
3219         char dst_swizzle[6];
3220
3221         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3222
3223         if (src_mod == WINED3DSPSM_DZ)
3224         {
3225             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3226             struct glsl_src_param div_param;
3227
3228             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3229
3230             if (mask_size > 1) {
3231                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3232             } else {
3233                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3234             }
3235         }
3236         else if (src_mod == WINED3DSPSM_DW)
3237         {
3238             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3239             struct glsl_src_param div_param;
3240
3241             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3242
3243             if (mask_size > 1) {
3244                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3245             } else {
3246                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3247             }
3248         } else {
3249             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3250         }
3251     }
3252 }
3253
3254 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3255  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3256  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3257 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3258 {
3259     DWORD sampler_idx = ins->dst[0].reg.idx;
3260     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3261     struct glsl_sample_function sample_function;
3262     struct glsl_src_param src0_param;
3263     UINT mask_size;
3264
3265     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3266
3267     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3268      * scalar, and projected sampling would require 4.
3269      *
3270      * It is a dependent read - not valid with conditional NP2 textures
3271      */
3272     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3273     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3274
3275     switch(mask_size)
3276     {
3277         case 1:
3278             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3279                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3280             break;
3281
3282         case 2:
3283             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3284                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3285             break;
3286
3287         case 3:
3288             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3289                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3290             break;
3291
3292         default:
3293             FIXME("Unexpected mask size %u\n", mask_size);
3294             break;
3295     }
3296 }
3297
3298 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3299  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3300 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3301 {
3302     DWORD dstreg = ins->dst[0].reg.idx;
3303     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3304     struct glsl_src_param src0_param;
3305     DWORD dst_mask;
3306     unsigned int mask_size;
3307
3308     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3309     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3310     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3311
3312     if (mask_size > 1) {
3313         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3314     } else {
3315         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3316     }
3317 }
3318
3319 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3320  * Calculate the depth as dst.x / dst.y   */
3321 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3322 {
3323     struct glsl_dst_param dst_param;
3324
3325     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3326
3327     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3328      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3329      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3330      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3331      * >= 1.0 or < 0.0
3332      */
3333     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3334             dst_param.reg_name, dst_param.reg_name);
3335 }
3336
3337 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3338  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3339  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3340  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3341  */
3342 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3343 {
3344     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3345     DWORD dstreg = ins->dst[0].reg.idx;
3346     struct glsl_src_param src0_param;
3347
3348     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3349
3350     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3351     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3352 }
3353
3354 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3355  * Calculate the 1st of a 2-row matrix multiplication. */
3356 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3357 {
3358     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3359     DWORD reg = ins->dst[0].reg.idx;
3360     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3361     struct glsl_src_param src0_param;
3362
3363     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3364     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3365 }
3366
3367 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3368  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3369 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3370 {
3371     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3372     DWORD reg = ins->dst[0].reg.idx;
3373     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3374     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3375     struct glsl_src_param src0_param;
3376
3377     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3378     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + tex_mx->current_row, reg, src0_param.param_str);
3379     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
3380 }
3381
3382 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3383 {
3384     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3385     DWORD reg = ins->dst[0].reg.idx;
3386     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3387     struct glsl_sample_function sample_function;
3388     struct glsl_src_param src0_param;
3389
3390     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3391     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3392
3393     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3394
3395     /* Sample the texture using the calculated coordinates */
3396     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3397 }
3398
3399 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3400  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3401 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3402 {
3403     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3404     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3405     struct glsl_sample_function sample_function;
3406     struct glsl_src_param src0_param;
3407     DWORD reg = ins->dst[0].reg.idx;
3408
3409     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3410     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3411
3412     /* Dependent read, not valid with conditional NP2 */
3413     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3414
3415     /* Sample the texture using the calculated coordinates */
3416     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3417
3418     tex_mx->current_row = 0;
3419 }
3420
3421 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3422  * Perform the 3rd row of a 3x3 matrix multiply */
3423 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3424 {
3425     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3426     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3427     struct glsl_src_param src0_param;
3428     char dst_mask[6];
3429     DWORD reg = ins->dst[0].reg.idx;
3430
3431     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3432
3433     shader_glsl_append_dst(ins->ctx->buffer, ins);
3434     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3435     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3436
3437     tex_mx->current_row = 0;
3438 }
3439
3440 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3441  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3442 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3443 {
3444     struct glsl_src_param src0_param;
3445     struct glsl_src_param src1_param;
3446     DWORD reg = ins->dst[0].reg.idx;
3447     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3448     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3449     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3450     struct glsl_sample_function sample_function;
3451     char coord_mask[6];
3452
3453     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3454     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3455
3456     /* Perform the last matrix multiply operation */
3457     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3458     /* Reflection calculation */
3459     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3460
3461     /* Dependent read, not valid with conditional NP2 */
3462     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3463     shader_glsl_write_mask_to_str(sample_function.coord_mask, coord_mask);
3464
3465     /* Sample the texture */
3466     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE,
3467             NULL, NULL, NULL, "tmp0%s", coord_mask);
3468
3469     tex_mx->current_row = 0;
3470 }
3471
3472 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3473  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3474 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3475 {
3476     DWORD reg = ins->dst[0].reg.idx;
3477     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3478     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3479     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3480     struct glsl_sample_function sample_function;
3481     struct glsl_src_param src0_param;
3482     char coord_mask[6];
3483
3484     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3485
3486     /* Perform the last matrix multiply operation */
3487     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3488
3489     /* Construct the eye-ray vector from w coordinates */
3490     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3491             tex_mx->texcoord_w[0], tex_mx->texcoord_w[1], reg);
3492     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3493
3494     /* Dependent read, not valid with conditional NP2 */
3495     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3496     shader_glsl_write_mask_to_str(sample_function.coord_mask, coord_mask);
3497
3498     /* Sample the texture using the calculated coordinates */
3499     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE,
3500             NULL, NULL, NULL, "tmp0%s", coord_mask);
3501
3502     tex_mx->current_row = 0;
3503 }
3504
3505 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3506  * Apply a fake bump map transform.
3507  * texbem is pshader <= 1.3 only, this saves a few version checks
3508  */
3509 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3510 {
3511     const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3512     struct glsl_sample_function sample_function;
3513     struct glsl_src_param coord_param;
3514     DWORD sampler_idx;
3515     DWORD mask;
3516     DWORD flags;
3517     char coord_mask[6];
3518
3519     sampler_idx = ins->dst[0].reg.idx;
3520     flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3521             & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3522
3523     /* Dependent read, not valid with conditional NP2 */
3524     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3525     mask = sample_function.coord_mask;
3526
3527     shader_glsl_write_mask_to_str(mask, coord_mask);
3528
3529     /* With projected textures, texbem only divides the static texture coord,
3530      * not the displacement, so we can't let GL handle this. */
3531     if (flags & WINED3D_PSARGS_PROJECTED)
3532     {
3533         DWORD div_mask=0;
3534         char coord_div_mask[3];
3535         switch (flags & ~WINED3D_PSARGS_PROJECTED)
3536         {
3537             case WINED3D_TTFF_COUNT1:
3538                 FIXME("WINED3D_TTFF_PROJECTED with WINED3D_TTFF_COUNT1?\n");
3539                 break;
3540             case WINED3D_TTFF_COUNT2:
3541                 div_mask = WINED3DSP_WRITEMASK_1;
3542                 break;
3543             case WINED3D_TTFF_COUNT3:
3544                 div_mask = WINED3DSP_WRITEMASK_2;
3545                 break;
3546             case WINED3D_TTFF_COUNT4:
3547             case WINED3D_TTFF_DISABLE:
3548                 div_mask = WINED3DSP_WRITEMASK_3;
3549                 break;
3550         }
3551         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3552         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3553     }
3554
3555     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3556
3557     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3558             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3559             coord_param.param_str, coord_mask);
3560
3561     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3562     {
3563         struct glsl_src_param luminance_param;
3564         struct glsl_dst_param dst_param;
3565
3566         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3567         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3568
3569         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3570                 dst_param.reg_name, dst_param.mask_str,
3571                 luminance_param.param_str, sampler_idx, sampler_idx);
3572     }
3573 }
3574
3575 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3576 {
3577     struct glsl_src_param src0_param, src1_param;
3578     DWORD sampler_idx = ins->dst[0].reg.idx;
3579
3580     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3581     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3582
3583     shader_glsl_append_dst(ins->ctx->buffer, ins);
3584     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3585             src0_param.param_str, sampler_idx, src1_param.param_str);
3586 }
3587
3588 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3589  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3590 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3591 {
3592     struct glsl_sample_function sample_function;
3593     struct glsl_src_param src0_param;
3594     DWORD sampler_idx = ins->dst[0].reg.idx;
3595
3596     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3597
3598     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3599     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3600             "%s.wx", src0_param.reg_name);
3601 }
3602
3603 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3604  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3605 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3606 {
3607     struct glsl_sample_function sample_function;
3608     struct glsl_src_param src0_param;
3609     DWORD sampler_idx = ins->dst[0].reg.idx;
3610
3611     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3612
3613     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3614     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3615             "%s.yz", src0_param.reg_name);
3616 }
3617
3618 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3619  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3620 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3621 {
3622     struct glsl_sample_function sample_function;
3623     struct glsl_src_param src0_param;
3624     DWORD sampler_idx = ins->dst[0].reg.idx;
3625
3626     /* Dependent read, not valid with conditional NP2 */
3627     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3628     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3629
3630     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3631             "%s", src0_param.param_str);
3632 }
3633
3634 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3635  * If any of the first 3 components are < 0, discard this pixel */
3636 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3637 {
3638     struct glsl_dst_param dst_param;
3639
3640     /* The argument is a destination parameter, and no writemasks are allowed */
3641     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3642     if (ins->ctx->reg_maps->shader_version.major >= 2)
3643     {
3644         /* 2.0 shaders compare all 4 components in texkill */
3645         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3646     } else {
3647         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3648          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3649          * 4 components are defined, only the first 3 are used
3650          */
3651         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3652     }
3653 }
3654
3655 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3656  * dst = dot2(src0, src1) + src2 */
3657 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3658 {
3659     struct glsl_src_param src0_param;
3660     struct glsl_src_param src1_param;
3661     struct glsl_src_param src2_param;
3662     DWORD write_mask;
3663     unsigned int mask_size;
3664
3665     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3666     mask_size = shader_glsl_get_write_mask_size(write_mask);
3667
3668     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3669     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3670     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3671
3672     if (mask_size > 1) {
3673         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3674                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3675     } else {
3676         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3677                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3678     }
3679 }
3680
3681 static void shader_glsl_input_pack(const struct wined3d_shader *shader, struct wined3d_shader_buffer *buffer,
3682         const struct wined3d_shader_signature_element *input_signature,
3683         const struct wined3d_shader_reg_maps *reg_maps,
3684         enum vertexprocessing_mode vertexprocessing)
3685 {
3686     WORD map = reg_maps->input_registers;
3687     unsigned int i;
3688
3689     for (i = 0; map; map >>= 1, ++i)
3690     {
3691         const char *semantic_name;
3692         UINT semantic_idx;
3693         char reg_mask[6];
3694
3695         /* Unused */
3696         if (!(map & 1)) continue;
3697
3698         semantic_name = input_signature[i].semantic_name;
3699         semantic_idx = input_signature[i].semantic_idx;
3700         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3701
3702         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3703         {
3704             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3705                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3706                         shader->u.ps.input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3707             else
3708                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3709                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3710         }
3711         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3712         {
3713             if (!semantic_idx)
3714                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3715                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3716             else if (semantic_idx == 1)
3717                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3718                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3719             else
3720                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3721                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3722         }
3723         else
3724         {
3725             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3726                     shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3727         }
3728     }
3729 }
3730
3731 /*********************************************
3732  * Vertex Shader Specific Code begins here
3733  ********************************************/
3734
3735 static void add_glsl_program_entry(struct shader_glsl_priv *priv, 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
3744     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3745     {
3746         ERR("Failed to insert program entry.\n");
3747     }
3748 }
3749
3750 static struct glsl_shader_prog_link *get_glsl_program_entry(const struct shader_glsl_priv *priv,
3751         const struct wined3d_shader *vshader, const struct wined3d_shader *pshader,
3752         const struct vs_compile_args *vs_args, const struct ps_compile_args *ps_args)
3753 {
3754     struct wine_rb_entry *entry;
3755     struct glsl_program_key key;
3756
3757     key.vshader = vshader;
3758     key.pshader = pshader;
3759     key.vs_args = *vs_args;
3760     key.ps_args = *ps_args;
3761
3762     entry = wine_rb_get(&priv->program_lookup, &key);
3763     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3764 }
3765
3766 /* GL locking is done by the caller */
3767 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3768         struct glsl_shader_prog_link *entry)
3769 {
3770     struct glsl_program_key key;
3771
3772     key.vshader = entry->vshader;
3773     key.pshader = entry->pshader;
3774     key.vs_args = entry->vs_args;
3775     key.ps_args = entry->ps_args;
3776     wine_rb_remove(&priv->program_lookup, &key);
3777
3778     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3779     if (entry->vshader) list_remove(&entry->vshader_entry);
3780     if (entry->pshader) list_remove(&entry->pshader_entry);
3781     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3782     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3783     HeapFree(GetProcessHeap(), 0, entry);
3784 }
3785
3786 static void handle_ps3_input(struct wined3d_shader_buffer *buffer,
3787         const struct wined3d_gl_info *gl_info, const DWORD *map,
3788         const struct wined3d_shader_signature_element *input_signature,
3789         const struct wined3d_shader_reg_maps *reg_maps_in,
3790         const struct wined3d_shader_signature_element *output_signature,
3791         const struct wined3d_shader_reg_maps *reg_maps_out)
3792 {
3793     unsigned int i, j;
3794     const char *semantic_name_in;
3795     UINT semantic_idx_in;
3796     DWORD *set;
3797     DWORD in_idx;
3798     unsigned int in_count = vec4_varyings(3, gl_info);
3799     char reg_mask[6];
3800     char destination[50];
3801     WORD input_map, output_map;
3802
3803     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3804
3805     input_map = reg_maps_in->input_registers;
3806     for (i = 0; input_map; input_map >>= 1, ++i)
3807     {
3808         if (!(input_map & 1)) continue;
3809
3810         in_idx = map[i];
3811         /* Declared, but not read register */
3812         if (in_idx == ~0U) continue;
3813         if (in_idx >= (in_count + 2))
3814         {
3815             FIXME("More input varyings declared than supported, expect issues.\n");
3816             continue;
3817         }
3818
3819         if (in_idx == in_count) {
3820             sprintf(destination, "gl_FrontColor");
3821         } else if (in_idx == in_count + 1) {
3822             sprintf(destination, "gl_FrontSecondaryColor");
3823         } else {
3824             sprintf(destination, "IN[%u]", in_idx);
3825         }
3826
3827         semantic_name_in = input_signature[i].semantic_name;
3828         semantic_idx_in = input_signature[i].semantic_idx;
3829         set[in_idx] = ~0U;
3830
3831         output_map = reg_maps_out->output_registers;
3832         for (j = 0; output_map; output_map >>= 1, ++j)
3833         {
3834             DWORD mask;
3835
3836             if (!(output_map & 1)
3837                     || semantic_idx_in != output_signature[j].semantic_idx
3838                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3839                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3840                 continue;
3841
3842             set[in_idx] = mask;
3843             shader_glsl_write_mask_to_str(mask, reg_mask);
3844
3845             shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3846                     destination, reg_mask, j, reg_mask);
3847         }
3848     }
3849
3850     for (i = 0; i < in_count + 2; ++i)
3851     {
3852         unsigned int size;
3853
3854         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3855             continue;
3856
3857         if (set[i] == ~0U) set[i] = 0;
3858
3859         size = 0;
3860         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3861         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3862         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3863         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3864         reg_mask[size] = '\0';
3865
3866         if (i == in_count) sprintf(destination, "gl_FrontColor");
3867         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3868         else sprintf(destination, "IN[%u]", i);
3869
3870         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3871         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3872     }
3873
3874     HeapFree(GetProcessHeap(), 0, set);
3875 }
3876
3877 /* GL locking is done by the caller */
3878 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3879         const struct wined3d_shader *vs, const struct wined3d_shader *ps,
3880         const struct wined3d_gl_info *gl_info)
3881 {
3882     GLhandleARB ret = 0;
3883     DWORD ps_major = ps ? ps->reg_maps.shader_version.major : 0;
3884     unsigned int i;
3885     const char *semantic_name;
3886     UINT semantic_idx;
3887     char reg_mask[6];
3888     const struct wined3d_shader_signature_element *output_signature = vs->output_signature;
3889     WORD map = vs->reg_maps.output_registers;
3890
3891     shader_buffer_clear(buffer);
3892
3893     shader_addline(buffer, "#version 120\n");
3894
3895     if (ps_major < 3)
3896     {
3897         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3898
3899         for (i = 0; map; map >>= 1, ++i)
3900         {
3901             DWORD write_mask;
3902
3903             if (!(map & 1)) continue;
3904
3905             semantic_name = output_signature[i].semantic_name;
3906             semantic_idx = output_signature[i].semantic_idx;
3907             write_mask = output_signature[i].mask;
3908             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3909
3910             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3911             {
3912                 if (!semantic_idx)
3913                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n",
3914                             reg_mask, i, reg_mask);
3915                 else if (semantic_idx == 1)
3916                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n",
3917                             reg_mask, i, reg_mask);
3918             }
3919             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
3920             {
3921                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3922                         reg_mask, i, reg_mask);
3923             }
3924             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3925             {
3926                 if (semantic_idx < 8)
3927                 {
3928                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3929                         write_mask |= WINED3DSP_WRITEMASK_3;
3930
3931                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3932                             semantic_idx, reg_mask, i, reg_mask);
3933                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3934                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3935                 }
3936             }
3937             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
3938             {
3939                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3940             }
3941             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3942             {
3943                 shader_addline(buffer, "gl_FogFragCoord = clamp(OUT[%u].%c, 0.0, 1.0);\n", i, reg_mask[1]);
3944             }
3945         }
3946         shader_addline(buffer, "}\n");
3947     }
3948     else
3949     {
3950         UINT in_count = min(vec4_varyings(ps_major, gl_info), ps->limits.packed_input);
3951         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3952         shader_addline(buffer, "varying vec4 IN[%u];\n", in_count);
3953         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3954
3955         /* First, sort out position and point size. Those are not passed to the pixel shader */
3956         for (i = 0; map; map >>= 1, ++i)
3957         {
3958             if (!(map & 1)) continue;
3959
3960             semantic_name = output_signature[i].semantic_name;
3961             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3962
3963             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
3964             {
3965                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3966                         reg_mask, i, reg_mask);
3967             }
3968             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
3969             {
3970                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3971             }
3972         }
3973
3974         /* Then, fix the pixel shader input */
3975         handle_ps3_input(buffer, gl_info, ps->u.ps.input_reg_map, ps->input_signature,
3976                 &ps->reg_maps, output_signature, &vs->reg_maps);
3977
3978         shader_addline(buffer, "}\n");
3979     }
3980
3981     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3982     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3983     shader_glsl_compile(gl_info, ret, buffer->buffer);
3984
3985     return ret;
3986 }
3987
3988 /* GL locking is done by the caller */
3989 static void hardcode_local_constants(const struct wined3d_shader *shader,
3990         const struct wined3d_gl_info *gl_info, GLhandleARB programId, char prefix)
3991 {
3992     const struct wined3d_shader_lconst *lconst;
3993     GLint tmp_loc;
3994     const float *value;
3995     char glsl_name[8];
3996
3997     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3998     {
3999         value = (const float *)lconst->value;
4000         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
4001         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4002         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
4003     }
4004     checkGLcall("Hardcoding local constants");
4005 }
4006
4007 /* GL locking is done by the caller */
4008 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
4009         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
4010         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
4011 {
4012     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4013     const struct wined3d_gl_info *gl_info = context->gl_info;
4014     const DWORD *function = shader->function;
4015     struct shader_glsl_ctx_priv priv_ctx;
4016
4017     /* Create the hw GLSL shader object and assign it as the shader->prgId */
4018     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4019
4020     memset(&priv_ctx, 0, sizeof(priv_ctx));
4021     priv_ctx.cur_ps_args = args;
4022     priv_ctx.cur_np2fixup_info = np2fixup_info;
4023
4024     shader_addline(buffer, "#version 120\n");
4025
4026     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
4027     {
4028         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
4029     }
4030     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
4031     {
4032         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
4033          * drivers write a warning if we don't do so
4034          */
4035         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
4036     }
4037     if (gl_info->supported[EXT_GPU_SHADER4])
4038     {
4039         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4040     }
4041
4042     /* Base Declarations */
4043     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4044
4045     /* Pack 3.0 inputs */
4046     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4047         shader_glsl_input_pack(shader, buffer, shader->input_signature, reg_maps, args->vp_mode);
4048
4049     /* Base Shader Body */
4050     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4051
4052     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4053     if (reg_maps->shader_version.major < 2)
4054     {
4055         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4056         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4057     }
4058
4059     if (args->srgb_correction)
4060     {
4061         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4062         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4063         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4064         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4065         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4066         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4067     }
4068     /* Pixel shader < 3.0 do not replace the fog stage.
4069      * This implements linear fog computation and blending.
4070      * TODO: non linear fog
4071      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4072      * -1/(e-s) and e/(e-s) respectively.
4073      */
4074     if (reg_maps->shader_version.major < 3)
4075     {
4076         switch(args->fog) {
4077             case FOG_OFF: break;
4078             case FOG_LINEAR:
4079                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4080                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4081                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4082                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4083                 break;
4084             case FOG_EXP:
4085                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4086                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4087                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4088                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4089                 break;
4090             case FOG_EXP2:
4091                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4092                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4093                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4094                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4095                 break;
4096         }
4097     }
4098
4099     shader_addline(buffer, "}\n");
4100
4101     TRACE("Compiling shader object %u\n", shader_obj);
4102     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4103
4104     /* Store the shader object */
4105     return shader_obj;
4106 }
4107
4108 /* GL locking is done by the caller */
4109 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4110         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
4111         const struct vs_compile_args *args)
4112 {
4113     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4114     const struct wined3d_gl_info *gl_info = context->gl_info;
4115     const DWORD *function = shader->function;
4116     struct shader_glsl_ctx_priv priv_ctx;
4117
4118     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4119     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4120
4121     shader_addline(buffer, "#version 120\n");
4122
4123     if (gl_info->supported[EXT_GPU_SHADER4])
4124         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4125
4126     memset(&priv_ctx, 0, sizeof(priv_ctx));
4127     priv_ctx.cur_vs_args = args;
4128
4129     /* Base Declarations */
4130     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4131
4132     /* Base Shader Body */
4133     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4134
4135     /* Unpack outputs */
4136     shader_addline(buffer, "order_ps_input(OUT);\n");
4137
4138     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4139      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4140      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4141      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4142      */
4143     if (args->fog_src == VS_FOG_Z)
4144         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4145     else if (!reg_maps->fog)
4146         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4147
4148     /* We always store the clipplanes without y inversion */
4149     if (args->clip_enabled)
4150         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4151
4152     /* Write the final position.
4153      *
4154      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4155      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4156      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4157      * contains 1.0 to allow a mad.
4158      */
4159     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4160     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4161
4162     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4163      *
4164      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4165      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4166      * which is the same as z = z * 2 - w.
4167      */
4168     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4169
4170     shader_addline(buffer, "}\n");
4171
4172     TRACE("Compiling shader object %u\n", shader_obj);
4173     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4174
4175     return shader_obj;
4176 }
4177
4178 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4179         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4180         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4181 {
4182     struct wined3d_state *state = &shader->device->stateBlock->state;
4183     UINT i;
4184     DWORD new_size;
4185     struct glsl_ps_compiled_shader *new_array;
4186     struct glsl_pshader_private    *shader_data;
4187     struct ps_np2fixup_info        *np2fixup = NULL;
4188     GLhandleARB ret;
4189
4190     if (!shader->backend_data)
4191     {
4192         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4193         if (!shader->backend_data)
4194         {
4195             ERR("Failed to allocate backend data.\n");
4196             return 0;
4197         }
4198     }
4199     shader_data = shader->backend_data;
4200
4201     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4202      * so a linear search is more performant than a hashmap or a binary search
4203      * (cache coherency etc)
4204      */
4205     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4206     {
4207         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4208         {
4209             if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4210             return shader_data->gl_shaders[i].prgId;
4211         }
4212     }
4213
4214     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4215     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4216         if (shader_data->num_gl_shaders)
4217         {
4218             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4219             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4220                                     new_size * sizeof(*shader_data->gl_shaders));
4221         } else {
4222             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4223             new_size = 1;
4224         }
4225
4226         if(!new_array) {
4227             ERR("Out of memory\n");
4228             return 0;
4229         }
4230         shader_data->gl_shaders = new_array;
4231         shader_data->shader_array_size = new_size;
4232     }
4233
4234     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4235
4236     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4237     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4238
4239     pixelshader_update_samplers(&shader->reg_maps, state->textures);
4240
4241     shader_buffer_clear(buffer);
4242     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4243     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4244     *np2fixup_info = np2fixup;
4245
4246     return ret;
4247 }
4248
4249 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4250                                  const DWORD use_map) {
4251     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4252     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4253     return stored->fog_src == new->fog_src;
4254 }
4255
4256 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4257         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4258         const struct vs_compile_args *args)
4259 {
4260     UINT i;
4261     DWORD new_size;
4262     struct glsl_vs_compiled_shader *new_array;
4263     DWORD use_map = shader->device->strided_streams.use_map;
4264     struct glsl_vshader_private *shader_data;
4265     GLhandleARB ret;
4266
4267     if (!shader->backend_data)
4268     {
4269         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4270         if (!shader->backend_data)
4271         {
4272             ERR("Failed to allocate backend data.\n");
4273             return 0;
4274         }
4275     }
4276     shader_data = shader->backend_data;
4277
4278     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4279      * so a linear search is more performant than a hashmap or a binary search
4280      * (cache coherency etc)
4281      */
4282     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4283         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4284             return shader_data->gl_shaders[i].prgId;
4285         }
4286     }
4287
4288     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4289
4290     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4291         if (shader_data->num_gl_shaders)
4292         {
4293             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4294             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4295                                     new_size * sizeof(*shader_data->gl_shaders));
4296         } else {
4297             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4298             new_size = 1;
4299         }
4300
4301         if(!new_array) {
4302             ERR("Out of memory\n");
4303             return 0;
4304         }
4305         shader_data->gl_shaders = new_array;
4306         shader_data->shader_array_size = new_size;
4307     }
4308
4309     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4310
4311     shader_buffer_clear(buffer);
4312     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4313     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4314
4315     return ret;
4316 }
4317
4318 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4319  * It sets the programId on the current StateBlock (because it should be called
4320  * inside of the DrawPrimitive() part of the render loop).
4321  *
4322  * If a program for the given combination does not exist, create one, and store
4323  * the program in the hash table.  If it creates a program, it will link the
4324  * given objects, too.
4325  */
4326
4327 /* GL locking is done by the caller */
4328 static void set_glsl_shader_program(const struct wined3d_context *context,
4329         struct wined3d_device *device, BOOL use_ps, BOOL use_vs)
4330 {
4331     const struct wined3d_state *state = &device->stateBlock->state;
4332     struct wined3d_shader *vshader = use_vs ? state->vertex_shader : NULL;
4333     struct wined3d_shader *pshader = use_ps ? state->pixel_shader : NULL;
4334     const struct wined3d_gl_info *gl_info = context->gl_info;
4335     struct shader_glsl_priv *priv = device->shader_priv;
4336     struct glsl_shader_prog_link *entry    = NULL;
4337     GLhandleARB programId                  = 0;
4338     GLhandleARB reorder_shader_id          = 0;
4339     unsigned int i;
4340     char glsl_name[8];
4341     struct ps_compile_args ps_compile_args;
4342     struct vs_compile_args vs_compile_args;
4343
4344     if (vshader) find_vs_compile_args(state, vshader, &vs_compile_args);
4345     if (pshader) find_ps_compile_args(state, pshader, &ps_compile_args);
4346
4347     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4348     if (entry)
4349     {
4350         priv->glsl_program = entry;
4351         return;
4352     }
4353
4354     /* If we get to this point, then no matching program exists, so we create one */
4355     programId = GL_EXTCALL(glCreateProgramObjectARB());
4356     TRACE("Created new GLSL shader program %u\n", programId);
4357
4358     /* Create the entry */
4359     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4360     entry->programId = programId;
4361     entry->vshader = vshader;
4362     entry->pshader = pshader;
4363     entry->vs_args = vs_compile_args;
4364     entry->ps_args = ps_compile_args;
4365     entry->constant_version = 0;
4366     entry->np2Fixup_info = NULL;
4367     /* Add the hash table entry */
4368     add_glsl_program_entry(priv, entry);
4369
4370     /* Set the current program */
4371     priv->glsl_program = entry;
4372
4373     /* Attach GLSL vshader */
4374     if (vshader)
4375     {
4376         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer, vshader, &vs_compile_args);
4377         WORD map = vshader->reg_maps.input_registers;
4378         char tmp_name[10];
4379
4380         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4381         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4382         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4383         checkGLcall("glAttachObjectARB");
4384         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4385          * is destroyed
4386          */
4387         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4388
4389         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4390         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4391         checkGLcall("glAttachObjectARB");
4392
4393         /* Bind vertex attributes to a corresponding index number to match
4394          * the same index numbers as ARB_vertex_programs (makes loading
4395          * vertex attributes simpler).  With this method, we can use the
4396          * exact same code to load the attributes later for both ARB and
4397          * GLSL shaders.
4398          *
4399          * We have to do this here because we need to know the Program ID
4400          * in order to make the bindings work, and it has to be done prior
4401          * to linking the GLSL program. */
4402         for (i = 0; map; map >>= 1, ++i)
4403         {
4404             if (!(map & 1)) continue;
4405
4406             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4407             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4408         }
4409         checkGLcall("glBindAttribLocationARB");
4410
4411         list_add_head(&vshader->linked_programs, &entry->vshader_entry);
4412     }
4413
4414     /* Attach GLSL pshader */
4415     if (pshader)
4416     {
4417         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4418                 pshader, &ps_compile_args, &entry->np2Fixup_info);
4419         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4420         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4421         checkGLcall("glAttachObjectARB");
4422
4423         list_add_head(&pshader->linked_programs, &entry->pshader_entry);
4424     }
4425
4426     /* Link the program */
4427     TRACE("Linking GLSL shader program %u\n", programId);
4428     GL_EXTCALL(glLinkProgramARB(programId));
4429     shader_glsl_validate_link(gl_info, programId);
4430
4431     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4432             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4433     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4434     {
4435         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4436         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4437     }
4438     for (i = 0; i < MAX_CONST_I; ++i)
4439     {
4440         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4441         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4442     }
4443     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4444             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4445     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4446     {
4447         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4448         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4449     }
4450     for (i = 0; i < MAX_CONST_I; ++i)
4451     {
4452         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4453         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4454     }
4455
4456     if(pshader) {
4457         char name[32];
4458
4459         for(i = 0; i < MAX_TEXTURES; i++) {
4460             sprintf(name, "bumpenvmat%u", i);
4461             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4462             sprintf(name, "luminancescale%u", i);
4463             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4464             sprintf(name, "luminanceoffset%u", i);
4465             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4466         }
4467
4468         if (ps_compile_args.np2_fixup) {
4469             if (entry->np2Fixup_info) {
4470                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4471             } else {
4472                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4473             }
4474         }
4475     }
4476
4477     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4478     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4479     checkGLcall("Find glsl program uniform locations");
4480
4481     if (pshader && pshader->reg_maps.shader_version.major >= 3
4482             && pshader->u.ps.declared_in_count > vec4_varyings(3, gl_info))
4483     {
4484         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4485         entry->vertex_color_clamp = GL_FALSE;
4486     } else {
4487         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4488     }
4489
4490     /* Set the shader to allow uniform loading on it */
4491     GL_EXTCALL(glUseProgramObjectARB(programId));
4492     checkGLcall("glUseProgramObjectARB(programId)");
4493
4494     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4495      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4496      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4497      * vertex shader with fixed function pixel processing is used we make sure that the card
4498      * supports enough samplers to allow the max number of vertex samplers with all possible
4499      * fixed function fragment processing setups. So once the program is linked these samplers
4500      * won't change.
4501      */
4502     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4503     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4504
4505     /* If the local constants do not have to be loaded with the environment constants,
4506      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4507      * later
4508      */
4509     if (pshader && !pshader->load_local_constsF)
4510         hardcode_local_constants(pshader, gl_info, programId, 'P');
4511     if (vshader && !vshader->load_local_constsF)
4512         hardcode_local_constants(vshader, gl_info, programId, 'V');
4513 }
4514
4515 /* GL locking is done by the caller */
4516 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4517 {
4518     GLhandleARB program_id;
4519     GLhandleARB vshader_id, pshader_id;
4520     const char *blt_pshader;
4521
4522     static const char *blt_vshader =
4523         "#version 120\n"
4524         "void main(void)\n"
4525         "{\n"
4526         "    gl_Position = gl_Vertex;\n"
4527         "    gl_FrontColor = vec4(1.0);\n"
4528         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4529         "}\n";
4530
4531     static const char * const blt_pshaders_full[tex_type_count] =
4532     {
4533         /* tex_1d */
4534         NULL,
4535         /* tex_2d */
4536         "#version 120\n"
4537         "uniform sampler2D sampler;\n"
4538         "void main(void)\n"
4539         "{\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         "void main(void)\n"
4548         "{\n"
4549         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4550         "}\n",
4551         /* tex_rect */
4552         "#version 120\n"
4553         "#extension GL_ARB_texture_rectangle : enable\n"
4554         "uniform sampler2DRect sampler;\n"
4555         "void main(void)\n"
4556         "{\n"
4557         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4558         "}\n",
4559     };
4560
4561     static const char * const blt_pshaders_masked[tex_type_count] =
4562     {
4563         /* tex_1d */
4564         NULL,
4565         /* tex_2d */
4566         "#version 120\n"
4567         "uniform sampler2D sampler;\n"
4568         "uniform vec4 mask;\n"
4569         "void main(void)\n"
4570         "{\n"
4571         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4572         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4573         "}\n",
4574         /* tex_3d */
4575         NULL,
4576         /* tex_cube */
4577         "#version 120\n"
4578         "uniform samplerCube sampler;\n"
4579         "uniform vec4 mask;\n"
4580         "void main(void)\n"
4581         "{\n"
4582         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4583         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4584         "}\n",
4585         /* tex_rect */
4586         "#version 120\n"
4587         "#extension GL_ARB_texture_rectangle : enable\n"
4588         "uniform sampler2DRect sampler;\n"
4589         "uniform vec4 mask;\n"
4590         "void main(void)\n"
4591         "{\n"
4592         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4593         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4594         "}\n",
4595     };
4596
4597     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4598     if (!blt_pshader)
4599     {
4600         FIXME("tex_type %#x not supported\n", tex_type);
4601         return 0;
4602     }
4603
4604     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4605     shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4606
4607     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4608     shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4609
4610     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4611     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4612     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4613     GL_EXTCALL(glLinkProgramARB(program_id));
4614
4615     shader_glsl_validate_link(gl_info, program_id);
4616
4617     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4618      * is destroyed
4619      */
4620     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4621     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4622     return program_id;
4623 }
4624
4625 /* GL locking is done by the caller */
4626 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4627 {
4628     const struct wined3d_gl_info *gl_info = context->gl_info;
4629     struct wined3d_device *device = context->swapchain->device;
4630     struct shader_glsl_priv *priv = device->shader_priv;
4631     GLhandleARB program_id = 0;
4632     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4633
4634     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4635
4636     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4637     else priv->glsl_program = NULL;
4638
4639     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4640
4641     if (old_vertex_color_clamp != current_vertex_color_clamp)
4642     {
4643         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4644         {
4645             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4646             checkGLcall("glClampColorARB");
4647         }
4648         else
4649         {
4650             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4651         }
4652     }
4653
4654     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4655     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4656     GL_EXTCALL(glUseProgramObjectARB(program_id));
4657     checkGLcall("glUseProgramObjectARB");
4658
4659     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4660      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4661      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4662     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4663     {
4664         shader_glsl_load_np2fixup_constants(priv, gl_info, &device->stateBlock->state);
4665     }
4666 }
4667
4668 /* GL locking is done by the caller */
4669 static void shader_glsl_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4670         enum tex_types tex_type, const SIZE *ds_mask_size)
4671 {
4672     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4673     struct shader_glsl_priv *priv = shader_priv;
4674     GLhandleARB *blt_program;
4675     GLint loc;
4676
4677     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4678     if (!*blt_program)
4679     {
4680         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4681         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4682         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4683         GL_EXTCALL(glUniform1iARB(loc, 0));
4684     }
4685     else
4686     {
4687         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4688     }
4689
4690     if (masked)
4691     {
4692         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4693         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4694     }
4695 }
4696
4697 /* GL locking is done by the caller */
4698 static void shader_glsl_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4699 {
4700     struct shader_glsl_priv *priv = shader_priv;
4701     GLhandleARB program_id;
4702
4703     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4704     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4705
4706     GL_EXTCALL(glUseProgramObjectARB(program_id));
4707     checkGLcall("glUseProgramObjectARB");
4708 }
4709
4710 static void shader_glsl_destroy(struct wined3d_shader *shader)
4711 {
4712     struct wined3d_device *device = shader->device;
4713     struct shader_glsl_priv *priv = device->shader_priv;
4714     const struct wined3d_gl_info *gl_info;
4715     const struct list *linked_programs;
4716     struct wined3d_context *context;
4717
4718     char pshader = shader_is_pshader_version(shader->reg_maps.shader_version.type);
4719
4720     if (pshader)
4721     {
4722         struct glsl_pshader_private *shader_data = shader->backend_data;
4723
4724         if (!shader_data || !shader_data->num_gl_shaders)
4725         {
4726             HeapFree(GetProcessHeap(), 0, shader_data);
4727             shader->backend_data = NULL;
4728             return;
4729         }
4730
4731         context = context_acquire(device, NULL);
4732         gl_info = context->gl_info;
4733
4734         if (priv->glsl_program && priv->glsl_program->pshader == shader)
4735         {
4736             ENTER_GL();
4737             shader_glsl_select(context, FALSE, FALSE);
4738             LEAVE_GL();
4739         }
4740     }
4741     else
4742     {
4743         struct glsl_vshader_private *shader_data = shader->backend_data;
4744
4745         if (!shader_data || !shader_data->num_gl_shaders)
4746         {
4747             HeapFree(GetProcessHeap(), 0, shader_data);
4748             shader->backend_data = NULL;
4749             return;
4750         }
4751
4752         context = context_acquire(device, NULL);
4753         gl_info = context->gl_info;
4754
4755         if (priv->glsl_program && priv->glsl_program->vshader == shader)
4756         {
4757             ENTER_GL();
4758             shader_glsl_select(context, FALSE, FALSE);
4759             LEAVE_GL();
4760         }
4761     }
4762
4763     linked_programs = &shader->linked_programs;
4764
4765     TRACE("Deleting linked programs\n");
4766     if (linked_programs->next) {
4767         struct glsl_shader_prog_link *entry, *entry2;
4768
4769         ENTER_GL();
4770         if(pshader) {
4771             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4772                 delete_glsl_program_entry(priv, gl_info, entry);
4773             }
4774         } else {
4775             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4776                 delete_glsl_program_entry(priv, gl_info, entry);
4777             }
4778         }
4779         LEAVE_GL();
4780     }
4781
4782     if (pshader)
4783     {
4784         struct glsl_pshader_private *shader_data = shader->backend_data;
4785         UINT i;
4786
4787         ENTER_GL();
4788         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4789             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4790             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4791             checkGLcall("glDeleteObjectARB");
4792         }
4793         LEAVE_GL();
4794         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4795     }
4796     else
4797     {
4798         struct glsl_vshader_private *shader_data = shader->backend_data;
4799         UINT i;
4800
4801         ENTER_GL();
4802         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4803             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4804             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4805             checkGLcall("glDeleteObjectARB");
4806         }
4807         LEAVE_GL();
4808         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4809     }
4810
4811     HeapFree(GetProcessHeap(), 0, shader->backend_data);
4812     shader->backend_data = NULL;
4813
4814     context_release(context);
4815 }
4816
4817 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4818 {
4819     const struct glsl_program_key *k = key;
4820     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4821             const struct glsl_shader_prog_link, program_lookup_entry);
4822     int cmp;
4823
4824     if (k->vshader > prog->vshader) return 1;
4825     else if (k->vshader < prog->vshader) return -1;
4826
4827     if (k->pshader > prog->pshader) return 1;
4828     else if (k->pshader < prog->pshader) return -1;
4829
4830     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4831     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4832
4833     return 0;
4834 }
4835
4836 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4837 {
4838     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4839     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4840
4841     if (!mem)
4842     {
4843         ERR("Failed to allocate memory\n");
4844         return FALSE;
4845     }
4846
4847     heap->entries = mem;
4848     heap->entries[1].version = 0;
4849     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4850     heap->size = 1;
4851
4852     return TRUE;
4853 }
4854
4855 static void constant_heap_free(struct constant_heap *heap)
4856 {
4857     HeapFree(GetProcessHeap(), 0, heap->entries);
4858 }
4859
4860 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4861 {
4862     wined3d_rb_alloc,
4863     wined3d_rb_realloc,
4864     wined3d_rb_free,
4865     glsl_program_key_compare,
4866 };
4867
4868 static HRESULT shader_glsl_alloc(struct wined3d_device *device)
4869 {
4870     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4871     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4872     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4873             gl_info->limits.glsl_ps_float_constants)) + 1;
4874
4875     if (!shader_buffer_init(&priv->shader_buffer))
4876     {
4877         ERR("Failed to initialize shader buffer.\n");
4878         goto fail;
4879     }
4880
4881     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4882     if (!priv->stack)
4883     {
4884         ERR("Failed to allocate memory.\n");
4885         goto fail;
4886     }
4887
4888     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4889     {
4890         ERR("Failed to initialize vertex shader constant heap\n");
4891         goto fail;
4892     }
4893
4894     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4895     {
4896         ERR("Failed to initialize pixel shader constant heap\n");
4897         goto fail;
4898     }
4899
4900     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4901     {
4902         ERR("Failed to initialize rbtree.\n");
4903         goto fail;
4904     }
4905
4906     priv->next_constant_version = 1;
4907
4908     device->shader_priv = priv;
4909     return WINED3D_OK;
4910
4911 fail:
4912     constant_heap_free(&priv->pconst_heap);
4913     constant_heap_free(&priv->vconst_heap);
4914     HeapFree(GetProcessHeap(), 0, priv->stack);
4915     shader_buffer_free(&priv->shader_buffer);
4916     HeapFree(GetProcessHeap(), 0, priv);
4917     return E_OUTOFMEMORY;
4918 }
4919
4920 /* Context activation is done by the caller. */
4921 static void shader_glsl_free(struct wined3d_device *device)
4922 {
4923     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4924     struct shader_glsl_priv *priv = device->shader_priv;
4925     int i;
4926
4927     ENTER_GL();
4928     for (i = 0; i < tex_type_count; ++i)
4929     {
4930         if (priv->depth_blt_program_full[i])
4931         {
4932             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4933         }
4934         if (priv->depth_blt_program_masked[i])
4935         {
4936             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4937         }
4938     }
4939     LEAVE_GL();
4940
4941     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4942     constant_heap_free(&priv->pconst_heap);
4943     constant_heap_free(&priv->vconst_heap);
4944     HeapFree(GetProcessHeap(), 0, priv->stack);
4945     shader_buffer_free(&priv->shader_buffer);
4946
4947     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4948     device->shader_priv = NULL;
4949 }
4950
4951 static void shader_glsl_context_destroyed(void *shader_priv, const struct wined3d_context *context) {}
4952
4953 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4954 {
4955     if (gl_info->supported[EXT_GPU_SHADER4] && gl_info->supported[ARB_GEOMETRY_SHADER4]
4956             && gl_info->glsl_version >= MAKEDWORD_VERSION(1, 50))
4957     {
4958         caps->VertexShaderVersion = 4;
4959         caps->PixelShaderVersion = 4;
4960     }
4961     /* ARB_shader_texture_lod or EXT_gpu_shader4 is required for the SM3
4962      * texldd and texldl instructions. */
4963     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] || gl_info->supported[EXT_GPU_SHADER4])
4964     {
4965         caps->VertexShaderVersion = 3;
4966         caps->PixelShaderVersion = 3;
4967     }
4968     else
4969     {
4970         caps->VertexShaderVersion = 2;
4971         caps->PixelShaderVersion = 2;
4972     }
4973
4974     caps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4975     caps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4976
4977     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4978      * Direct3D minimum requirement.
4979      *
4980      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4981      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4982      *
4983      * The problem is that the refrast clamps temporary results in the shader to
4984      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4985      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4986      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4987      * offer a way to query this.
4988      */
4989     caps->PixelShader1xMaxValue = 8.0;
4990
4991     caps->VSClipping = TRUE;
4992
4993     TRACE("Hardware vertex shader version %u enabled (GLSL).\n", caps->VertexShaderVersion);
4994     TRACE("Hardware pixel shader version %u enabled (GLSL).\n", caps->PixelShaderVersion);
4995 }
4996
4997 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4998 {
4999     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5000     {
5001         TRACE("Checking support for fixup:\n");
5002         dump_color_fixup_desc(fixup);
5003     }
5004
5005     /* We support everything except YUV conversions. */
5006     if (!is_complex_fixup(fixup))
5007     {
5008         TRACE("[OK]\n");
5009         return TRUE;
5010     }
5011
5012     TRACE("[FAILED]\n");
5013     return FALSE;
5014 }
5015
5016 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5017 {
5018     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
5019     /* WINED3DSIH_ADD           */ shader_glsl_arith,
5020     /* WINED3DSIH_AND           */ NULL,
5021     /* WINED3DSIH_BEM           */ shader_glsl_bem,
5022     /* WINED3DSIH_BREAK         */ shader_glsl_break,
5023     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
5024     /* WINED3DSIH_BREAKP        */ NULL,
5025     /* WINED3DSIH_CALL          */ shader_glsl_call,
5026     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
5027     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
5028     /* WINED3DSIH_CND           */ shader_glsl_cnd,
5029     /* WINED3DSIH_CRS           */ shader_glsl_cross,
5030     /* WINED3DSIH_CUT           */ NULL,
5031     /* WINED3DSIH_DCL           */ NULL,
5032     /* WINED3DSIH_DEF           */ NULL,
5033     /* WINED3DSIH_DEFB          */ NULL,
5034     /* WINED3DSIH_DEFI          */ NULL,
5035     /* WINED3DSIH_DIV           */ NULL,
5036     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
5037     /* WINED3DSIH_DP3           */ shader_glsl_dot,
5038     /* WINED3DSIH_DP4           */ shader_glsl_dot,
5039     /* WINED3DSIH_DST           */ shader_glsl_dst,
5040     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5041     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5042     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5043     /* WINED3DSIH_EMIT          */ NULL,
5044     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5045     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5046     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5047     /* WINED3DSIH_EQ            */ NULL,
5048     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5049     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5050     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5051     /* WINED3DSIH_FTOI          */ NULL,
5052     /* WINED3DSIH_GE            */ NULL,
5053     /* WINED3DSIH_IADD          */ NULL,
5054     /* WINED3DSIH_IEQ           */ NULL,
5055     /* WINED3DSIH_IF            */ shader_glsl_if,
5056     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5057     /* WINED3DSIH_IGE           */ NULL,
5058     /* WINED3DSIH_IMUL          */ NULL,
5059     /* WINED3DSIH_ITOF          */ NULL,
5060     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5061     /* WINED3DSIH_LD            */ NULL,
5062     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5063     /* WINED3DSIH_LOG           */ shader_glsl_log,
5064     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5065     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5066     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5067     /* WINED3DSIH_LT            */ NULL,
5068     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5069     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5070     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5071     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5072     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5073     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5074     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5075     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5076     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5077     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5078     /* WINED3DSIH_MOVC          */ NULL,
5079     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5080     /* WINED3DSIH_NOP           */ NULL,
5081     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5082     /* WINED3DSIH_PHASE         */ NULL,
5083     /* WINED3DSIH_POW           */ shader_glsl_pow,
5084     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5085     /* WINED3DSIH_REP           */ shader_glsl_rep,
5086     /* WINED3DSIH_RET           */ shader_glsl_ret,
5087     /* WINED3DSIH_ROUND_NI      */ NULL,
5088     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5089     /* WINED3DSIH_SAMPLE        */ NULL,
5090     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5091     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5092     /* WINED3DSIH_SETP          */ NULL,
5093     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5094     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5095     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5096     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5097     /* WINED3DSIH_SQRT          */ NULL,
5098     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5099     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5100     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5101     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5102     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5103     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5104     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5105     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5106     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5107     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5108     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5109     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5110     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5111     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5112     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5113     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5114     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5115     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5116     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5117     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5118     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5119     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5120     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5121     /* WINED3DSIH_UDIV          */ NULL,
5122     /* WINED3DSIH_USHR          */ NULL,
5123     /* WINED3DSIH_UTOF          */ NULL,
5124     /* WINED3DSIH_XOR           */ NULL,
5125 };
5126
5127 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5128     SHADER_HANDLER hw_fct;
5129
5130     /* Select handler */
5131     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5132
5133     /* Unhandled opcode */
5134     if (!hw_fct)
5135     {
5136         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5137         return;
5138     }
5139     hw_fct(ins);
5140
5141     shader_glsl_add_instruction_modifiers(ins);
5142 }
5143
5144 const struct wined3d_shader_backend_ops glsl_shader_backend =
5145 {
5146     shader_glsl_handle_instruction,
5147     shader_glsl_select,
5148     shader_glsl_select_depth_blt,
5149     shader_glsl_deselect_depth_blt,
5150     shader_glsl_update_float_vertex_constants,
5151     shader_glsl_update_float_pixel_constants,
5152     shader_glsl_load_constants,
5153     shader_glsl_load_np2fixup_constants,
5154     shader_glsl_destroy,
5155     shader_glsl_alloc,
5156     shader_glsl_free,
5157     shader_glsl_context_destroyed,
5158     shader_glsl_get_caps,
5159     shader_glsl_color_fixup_supported,
5160 };