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

   1 /*
   2  * GLSL pixel and vertex shader implementation
   3  *
   4  * Copyright 2006 Jason Green
   5  * Copyright 2006-2007 Henri Verbeet
   6  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
   7  * Copyright 2009 Henri Verbeet for CodeWeavers
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  22  */
  23
  24 /*
  25  * D3D shader asm has swizzles on source parameters, and write masks for
  26  * destination parameters. GLSL uses swizzles for both. The result of this is
  27  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
  28  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
  29  * mask for the destination parameter into account.
  30  */
  31
  32 #include "config.h"
  33 #include <limits.h>
  34 #include <stdio.h>
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 #define GLINFO_LOCATION      (*gl_info)
  43
  44 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
  45 #define WINED3D_GLSL_SAMPLE_RECT        0x2
  46 #define WINED3D_GLSL_SAMPLE_LOD         0x4
  47 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
  48
  49 typedef struct {
  50     char reg_name[150];
  51     char mask_str[6];
  52 } glsl_dst_param_t;
  53
  54 typedef struct {
  55     char reg_name[150];
  56     char param_str[200];
  57 } glsl_src_param_t;
  58
  59 typedef struct {
  60     const char *name;
  61     DWORD coord_mask;
  62 } glsl_sample_function_t;
  63
  64 enum heap_node_op
  65 {
  66     HEAP_NODE_TRAVERSE_LEFT,
  67     HEAP_NODE_TRAVERSE_RIGHT,
  68     HEAP_NODE_POP,
  69 };
  70
  71 struct constant_entry
  72 {
  73     unsigned int idx;
  74     unsigned int version;
  75 };
  76
  77 struct constant_heap
  78 {
  79     struct constant_entry *entries;
  80     unsigned int *positions;
  81     unsigned int size;
  82 };
  83
  84 /* GLSL shader private data */
  85 struct shader_glsl_priv {
  86     struct wined3d_shader_buffer shader_buffer;
  87     struct wine_rb_tree program_lookup;
  88     struct glsl_shader_prog_link *glsl_program;
  89     struct constant_heap vconst_heap;
  90     struct constant_heap pconst_heap;
  91     unsigned char *stack;
  92     GLhandleARB depth_blt_program[tex_type_count];
  93     UINT next_constant_version;
  94 };
  95
  96 /* Struct to maintain data about a linked GLSL program */
  97 struct glsl_shader_prog_link {
  98     struct wine_rb_entry        program_lookup_entry;
  99     struct list                 vshader_entry;
 100     struct list                 pshader_entry;
 101     GLhandleARB                 programId;
 102     GLint                       *vuniformF_locations;
 103     GLint                       *puniformF_locations;
 104     GLint                       vuniformI_locations[MAX_CONST_I];
 105     GLint                       puniformI_locations[MAX_CONST_I];
 106     GLint                       posFixup_location;
 107     GLint                       np2Fixup_location;
 108     GLint                       bumpenvmat_location[MAX_TEXTURES];
 109     GLint                       luminancescale_location[MAX_TEXTURES];
 110     GLint                       luminanceoffset_location[MAX_TEXTURES];
 111     GLint                       ycorrection_location;
 112     GLenum                      vertex_color_clamp;
 113     IWineD3DVertexShader        *vshader;
 114     IWineD3DPixelShader         *pshader;
 115     struct vs_compile_args      vs_args;
 116     struct ps_compile_args      ps_args;
 117     UINT                        constant_version;
 118     const struct ps_np2fixup_info *np2Fixup_info;
 119 };
 120
 121 typedef struct {
 122     IWineD3DVertexShader        *vshader;
 123     IWineD3DPixelShader         *pshader;
 124     struct ps_compile_args      ps_args;
 125     struct vs_compile_args      vs_args;
 126 } glsl_program_key_t;
 127
 128 struct shader_glsl_ctx_priv {
 129     const struct vs_compile_args    *cur_vs_args;
 130     const struct ps_compile_args    *cur_ps_args;
 131     struct ps_np2fixup_info         *cur_np2fixup_info;
 132 };
 133
 134 struct glsl_ps_compiled_shader
 135 {
 136     struct ps_compile_args          args;
 137     struct ps_np2fixup_info         np2fixup;
 138     GLhandleARB                     prgId;
 139 };
 140
 141 struct glsl_pshader_private
 142 {
 143     struct glsl_ps_compiled_shader  *gl_shaders;
 144     UINT                            num_gl_shaders, shader_array_size;
 145 };
 146
 147 struct glsl_vs_compiled_shader
 148 {
 149     struct vs_compile_args          args;
 150     GLhandleARB                     prgId;
 151 };
 152
 153 struct glsl_vshader_private
 154 {
 155     struct glsl_vs_compiled_shader  *gl_shaders;
 156     UINT                            num_gl_shaders, shader_array_size;
 157 };
 158
 159 static const char *debug_gl_shader_type(GLenum type)
 160 {
 161     switch (type)
 162     {
 163 #define WINED3D_TO_STR(u) case u: return #u
 164         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
 165         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
 166         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
 167 #undef WINED3D_TO_STR
 168         default:
 169             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
 170     }
 171 }
 172
 173 /* Extract a line from the info log.
 174  * Note that this modifies the source string. */
 175 static char *get_info_log_line(char **ptr)
 176 {
 177     char *p, *q;
 178
 179     p = *ptr;
 180     if (!(q = strstr(p, "\n")))
 181     {
 182         if (!*p) return NULL;
 183         *ptr += strlen(p);
 184         return p;
 185     }
 186     *q = '\0';
 187     *ptr = q + 1;
 188
 189     return p;
 190 }
 191
 192 /** Prints the GLSL info log which will contain error messages if they exist */
 193 /* GL locking is done by the caller */
 194 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 195 {
 196     int infologLength = 0;
 197     char *infoLog;
 198     unsigned int i;
 199     BOOL is_spam;
 200
 201     static const char * const spam[] =
 202     {
 203         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 204         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx, with \n */
 205         "Fragment shader was successfully compiled to run on hardware.",    /* fglrx, no \n   */
 206         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 207         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 208         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 209         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 210         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 211         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 212     };
 213
 214     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 215
 216     GL_EXTCALL(glGetObjectParameterivARB(obj,
 217                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 218                &infologLength));
 219
 220     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 221      * that if there are errors. */
 222     if (infologLength > 1)
 223     {
 224         char *ptr, *line;
 225
 226         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 227          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 228          */
 229         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 230         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 231         is_spam = FALSE;
 232
 233         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 234             if(strcmp(infoLog, spam[i]) == 0) {
 235                 is_spam = TRUE;
 236                 break;
 237             }
 238         }
 239
 240         ptr = infoLog;
 241         if (is_spam)
 242         {
 243             TRACE("Spam received from GLSL shader #%u:\n", obj);
 244             while ((line = get_info_log_line(&ptr))) TRACE("    %s\n", line);
 245         }
 246         else
 247         {
 248             FIXME("Error received from GLSL shader #%u:\n", obj);
 249             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 250         }
 251         HeapFree(GetProcessHeap(), 0, infoLog);
 252     }
 253 }
 254
 255 /* GL locking is done by the caller. */
 256 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 257 {
 258     GLint i, object_count, source_size;
 259     GLhandleARB *objects;
 260     char *source = NULL;
 261
 262     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
 263     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
 264     if (!objects)
 265     {
 266         ERR("Failed to allocate object array memory.\n");
 267         return;
 268     }
 269
 270     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
 271     for (i = 0; i < object_count; ++i)
 272     {
 273         char *ptr, *line;
 274         GLint tmp;
 275
 276         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
 277
 278         if (!source || source_size < tmp)
 279         {
 280             HeapFree(GetProcessHeap(), 0, source);
 281
 282             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
 283             if (!source)
 284             {
 285                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
 286                 HeapFree(GetProcessHeap(), 0, objects);
 287                 return;
 288             }
 289             source_size = tmp;
 290         }
 291
 292         FIXME("Object %u:\n", objects[i]);
 293         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
 294         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
 295         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
 296         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
 297         FIXME("\n");
 298
 299         ptr = source;
 300         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
 301         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 302         FIXME("\n");
 303     }
 304
 305     HeapFree(GetProcessHeap(), 0, source);
 306     HeapFree(GetProcessHeap(), 0, objects);
 307 }
 308
 309 /* GL locking is done by the caller. */
 310 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 311 {
 312     GLint tmp;
 313
 314     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 315
 316     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
 317     if (tmp == GL_PROGRAM_OBJECT_ARB)
 318     {
 319         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
 320         if (!tmp)
 321         {
 322             FIXME("Program %u link status invalid.\n", program);
 323             shader_glsl_dump_program_source(gl_info, program);
 324         }
 325     }
 326
 327     print_glsl_info_log(gl_info, program);
 328 }
 329
 330 /**
 331  * Loads (pixel shader) samplers
 332  */
 333 /* GL locking is done by the caller */
 334 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 335         DWORD *tex_unit_map, GLhandleARB programId)
 336 {
 337     GLint name_loc;
 338     int i;
 339     char sampler_name[20];
 340
 341     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 342         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 343         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 344         if (name_loc != -1) {
 345             DWORD mapped_unit = tex_unit_map[i];
 346             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 347             {
 348                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 349                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 350                 checkGLcall("glUniform1iARB");
 351             } else {
 352                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 353             }
 354         }
 355     }
 356 }
 357
 358 /* GL locking is done by the caller */
 359 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 360         DWORD *tex_unit_map, GLhandleARB programId)
 361 {
 362     GLint name_loc;
 363     char sampler_name[20];
 364     int i;
 365
 366     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 367         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 368         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 369         if (name_loc != -1) {
 370             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 371             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 372             {
 373                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 374                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 375                 checkGLcall("glUniform1iARB");
 376             } else {
 377                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 378             }
 379         }
 380     }
 381 }
 382
 383 /* GL locking is done by the caller */
 384 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 385         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 386 {
 387     int stack_idx = 0;
 388     unsigned int heap_idx = 1;
 389     unsigned int idx;
 390
 391     if (heap->entries[heap_idx].version <= version) return;
 392
 393     idx = heap->entries[heap_idx].idx;
 394     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 395     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 396
 397     while (stack_idx >= 0)
 398     {
 399         /* Note that we fall through to the next case statement. */
 400         switch(stack[stack_idx])
 401         {
 402             case HEAP_NODE_TRAVERSE_LEFT:
 403             {
 404                 unsigned int left_idx = heap_idx << 1;
 405                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 406                 {
 407                     heap_idx = left_idx;
 408                     idx = heap->entries[heap_idx].idx;
 409                     if (constant_locations[idx] != -1)
 410                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 411
 412                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 413                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 414                     break;
 415                 }
 416             }
 417
 418             case HEAP_NODE_TRAVERSE_RIGHT:
 419             {
 420                 unsigned int right_idx = (heap_idx << 1) + 1;
 421                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 422                 {
 423                     heap_idx = right_idx;
 424                     idx = heap->entries[heap_idx].idx;
 425                     if (constant_locations[idx] != -1)
 426                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 427
 428                     stack[stack_idx++] = HEAP_NODE_POP;
 429                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 430                     break;
 431                 }
 432             }
 433
 434             case HEAP_NODE_POP:
 435             {
 436                 heap_idx >>= 1;
 437                 --stack_idx;
 438                 break;
 439             }
 440         }
 441     }
 442     checkGLcall("walk_constant_heap()");
 443 }
 444
 445 /* GL locking is done by the caller */
 446 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 447 {
 448     GLfloat clamped_constant[4];
 449
 450     if (location == -1) return;
 451
 452     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 453     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 454     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 455     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 456
 457     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 458 }
 459
 460 /* GL locking is done by the caller */
 461 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 462         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 463 {
 464     int stack_idx = 0;
 465     unsigned int heap_idx = 1;
 466     unsigned int idx;
 467
 468     if (heap->entries[heap_idx].version <= version) return;
 469
 470     idx = heap->entries[heap_idx].idx;
 471     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 472     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 473
 474     while (stack_idx >= 0)
 475     {
 476         /* Note that we fall through to the next case statement. */
 477         switch(stack[stack_idx])
 478         {
 479             case HEAP_NODE_TRAVERSE_LEFT:
 480             {
 481                 unsigned int left_idx = heap_idx << 1;
 482                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 483                 {
 484                     heap_idx = left_idx;
 485                     idx = heap->entries[heap_idx].idx;
 486                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 487
 488                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 489                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 490                     break;
 491                 }
 492             }
 493
 494             case HEAP_NODE_TRAVERSE_RIGHT:
 495             {
 496                 unsigned int right_idx = (heap_idx << 1) + 1;
 497                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 498                 {
 499                     heap_idx = right_idx;
 500                     idx = heap->entries[heap_idx].idx;
 501                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 502
 503                     stack[stack_idx++] = HEAP_NODE_POP;
 504                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 505                     break;
 506                 }
 507             }
 508
 509             case HEAP_NODE_POP:
 510             {
 511                 heap_idx >>= 1;
 512                 --stack_idx;
 513                 break;
 514             }
 515         }
 516     }
 517     checkGLcall("walk_constant_heap_clamped()");
 518 }
 519
 520 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 521 /* GL locking is done by the caller */
 522 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 523         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 524         unsigned char *stack, UINT version)
 525 {
 526     const local_constant *lconst;
 527
 528     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 529     if (This->baseShader.reg_maps.shader_version.major == 1
 530             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
 531         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 532     else
 533         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 534
 535     if (!This->baseShader.load_local_constsF)
 536     {
 537         TRACE("No need to load local float constants for this shader\n");
 538         return;
 539     }
 540
 541     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 542     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 543     {
 544         GLint location = constant_locations[lconst->idx];
 545         /* We found this uniform name in the program - go ahead and send the data */
 546         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 547     }
 548     checkGLcall("glUniform4fvARB()");
 549 }
 550
 551 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 552 /* GL locking is done by the caller */
 553 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 554         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 555 {
 556     unsigned int i;
 557     struct list* ptr;
 558
 559     for (i = 0; constants_set; constants_set >>= 1, ++i)
 560     {
 561         if (!(constants_set & 1)) continue;
 562
 563         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 564                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 565
 566         /* We found this uniform name in the program - go ahead and send the data */
 567         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 568         checkGLcall("glUniform4ivARB");
 569     }
 570
 571     /* Load immediate constants */
 572     ptr = list_head(&This->baseShader.constantsI);
 573     while (ptr) {
 574         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 575         unsigned int idx = lconst->idx;
 576         const GLint *values = (const GLint *)lconst->value;
 577
 578         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 579             values[0], values[1], values[2], values[3]);
 580
 581         /* We found this uniform name in the program - go ahead and send the data */
 582         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 583         checkGLcall("glUniform4ivARB");
 584         ptr = list_next(&This->baseShader.constantsI, ptr);
 585     }
 586 }
 587
 588 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 589 /* GL locking is done by the caller */
 590 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 591         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 592 {
 593     GLint tmp_loc;
 594     unsigned int i;
 595     char tmp_name[8];
 596     const char *prefix;
 597     struct list* ptr;
 598
 599     switch (This->baseShader.reg_maps.shader_version.type)
 600     {
 601         case WINED3D_SHADER_TYPE_VERTEX:
 602             prefix = "VB";
 603             break;
 604
 605         case WINED3D_SHADER_TYPE_GEOMETRY:
 606             prefix = "GB";
 607             break;
 608
 609         case WINED3D_SHADER_TYPE_PIXEL:
 610             prefix = "PB";
 611             break;
 612
 613         default:
 614             FIXME("Unknown shader type %#x.\n",
 615                     This->baseShader.reg_maps.shader_version.type);
 616             prefix = "UB";
 617             break;
 618     }
 619
 620     /* TODO: Benchmark and see if it would be beneficial to store the
 621      * locations of the constants to avoid looking up each time */
 622     for (i = 0; constants_set; constants_set >>= 1, ++i)
 623     {
 624         if (!(constants_set & 1)) continue;
 625
 626         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 627
 628         /* TODO: Benchmark and see if it would be beneficial to store the
 629          * locations of the constants to avoid looking up each time */
 630         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 631         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 632         if (tmp_loc != -1)
 633         {
 634             /* We found this uniform name in the program - go ahead and send the data */
 635             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 636             checkGLcall("glUniform1ivARB");
 637         }
 638     }
 639
 640     /* Load immediate constants */
 641     ptr = list_head(&This->baseShader.constantsB);
 642     while (ptr) {
 643         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 644         unsigned int idx = lconst->idx;
 645         const GLint *values = (const GLint *)lconst->value;
 646
 647         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 648
 649         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 650         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 651         if (tmp_loc != -1) {
 652             /* We found this uniform name in the program - go ahead and send the data */
 653             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 654             checkGLcall("glUniform1ivARB");
 655         }
 656         ptr = list_next(&This->baseShader.constantsB, ptr);
 657     }
 658 }
 659
 660 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 661 {
 662     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 663 }
 664
 665 /**
 666  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 667  */
 668 /* GL locking is done by the caller (state handler) */
 669 static void shader_glsl_load_np2fixup_constants(
 670     IWineD3DDevice* device,
 671     char usePixelShader,
 672     char useVertexShader) {
 673
 674     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
 675     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
 676
 677     if (!prog) {
 678         /* No GLSL program set - nothing to do. */
 679         return;
 680     }
 681
 682     if (!usePixelShader) {
 683         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 684         return;
 685     }
 686
 687     if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
 688         const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 689         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
 690         UINT i;
 691         UINT fixup = prog->ps_args.np2_fixup;
 692         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 693
 694         for (i = 0; fixup; fixup >>= 1, ++i) {
 695             const unsigned char idx = prog->np2Fixup_info->idx[i];
 696             const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
 697             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 698
 699             if (!tex) {
 700                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 701                 continue;
 702             }
 703
 704             if (idx % 2) {
 705                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 706             } else {
 707                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 708             }
 709         }
 710
 711         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 712     }
 713 }
 714
 715 /**
 716  * Loads the app-supplied constants into the currently set GLSL program.
 717  */
 718 /* GL locking is done by the caller (state handler) */
 719 static void shader_glsl_load_constants(const struct wined3d_context *context,
 720         char usePixelShader, char useVertexShader)
 721 {
 722     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.device;
 723     const struct wined3d_gl_info *gl_info = context->gl_info;
 724     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 725     struct shader_glsl_priv *priv = device->shader_priv;
 726
 727     GLhandleARB programId;
 728     struct glsl_shader_prog_link *prog = priv->glsl_program;
 729     UINT constant_version;
 730     int i;
 731
 732     if (!prog) {
 733         /* No GLSL program set - nothing to do. */
 734         return;
 735     }
 736     programId = prog->programId;
 737     constant_version = prog->constant_version;
 738
 739     if (useVertexShader) {
 740         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 741
 742         /* Load DirectX 9 float constants/uniforms for vertex shader */
 743         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
 744                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 745
 746         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 747         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
 748                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 749
 750         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 751         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
 752                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 753
 754         /* Upload the position fixup params */
 755         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
 756         checkGLcall("glUniform4fvARB");
 757     }
 758
 759     if (usePixelShader) {
 760
 761         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 762
 763         /* Load DirectX 9 float constants/uniforms for pixel shader */
 764         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
 765                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 766
 767         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 768         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
 769                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 770
 771         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 772         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
 773                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 774
 775         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 776          * It can't be 0 for a valid texbem instruction.
 777          */
 778         for(i = 0; i < MAX_TEXTURES; i++) {
 779             const float *data;
 780
 781             if(prog->bumpenvmat_location[i] == -1) continue;
 782
 783             data = (const float *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVMAT00];
 784             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 785             checkGLcall("glUniformMatrix2fvARB");
 786
 787             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 788              * is set too, so we can check that in the needsbumpmat check
 789              */
 790             if(prog->luminancescale_location[i] != -1) {
 791                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLSCALE];
 792                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLOFFSET];
 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(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 802             float correction_params[4];
 803
 804             if (context->render_offscreen)
 805             {
 806                 correction_params[0] = 0.0f;
 807                 correction_params[1] = 1.0f;
 808             } else {
 809                 /* position is window relative, not viewport relative */
 810                 correction_params[0] = ((IWineD3DSurfaceImpl *)context->current_rt)->currentDesc.Height;
 811                 correction_params[1] = -1.0f;
 812             }
 813             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 814         }
 815     }
 816
 817     if (priv->next_constant_version == UINT_MAX)
 818     {
 819         TRACE("Max constant version reached, resetting to 0.\n");
 820         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 821         priv->next_constant_version = 1;
 822     }
 823     else
 824     {
 825         prog->constant_version = priv->next_constant_version++;
 826     }
 827 }
 828
 829 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 830         unsigned int heap_idx, DWORD new_version)
 831 {
 832     struct constant_entry *entries = heap->entries;
 833     unsigned int *positions = heap->positions;
 834     unsigned int parent_idx;
 835
 836     while (heap_idx > 1)
 837     {
 838         parent_idx = heap_idx >> 1;
 839
 840         if (new_version <= entries[parent_idx].version) break;
 841
 842         entries[heap_idx] = entries[parent_idx];
 843         positions[entries[parent_idx].idx] = heap_idx;
 844         heap_idx = parent_idx;
 845     }
 846
 847     entries[heap_idx].version = new_version;
 848     entries[heap_idx].idx = idx;
 849     positions[idx] = heap_idx;
 850 }
 851
 852 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 853 {
 854     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 855     struct shader_glsl_priv *priv = This->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 (!This->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(IWineD3DDevice *iface, UINT start, UINT count)
 869 {
 870     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 871     struct shader_glsl_priv *priv = This->shader_priv;
 872     struct constant_heap *heap = &priv->pconst_heap;
 873     UINT i;
 874
 875     for (i = start; i < count + start; ++i)
 876     {
 877         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 878             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 879         else
 880             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 881     }
 882 }
 883
 884 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 885 {
 886     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 887     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 888     if(shader_major > 3) return ret;
 889
 890     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 891     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 892     return ret;
 893 }
 894
 895 /** Generate the variable & register declarations for the GLSL output target */
 896 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 897         struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
 898         const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
 899 {
 900     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 901     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 902     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 903     const struct wined3d_gl_info *gl_info = context->gl_info;
 904     unsigned int i, extra_constants_needed = 0;
 905     const local_constant *lconst;
 906     DWORD map;
 907
 908     /* There are some minor differences between pixel and vertex shaders */
 909     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 910     char prefix = pshader ? 'P' : 'V';
 911
 912     /* Prototype the subroutines */
 913     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 914     {
 915         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 916     }
 917
 918     /* Declare the constants (aka uniforms) */
 919     if (This->baseShader.limits.constant_float > 0) {
 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(This->baseShader.reg_maps.usesrelconstF) {
 942                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
 943                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
 944                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
 945                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
 946                  *
 947                  * Writing gl_ClipVertex requires one uniform for each clipplane as well.
 948                  */
 949                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 950                 if(ctx_priv->cur_vs_args->clip_enabled)
 951                 {
 952                     max_constantsF -= gl_info->limits.clipplanes;
 953                 }
 954                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 955                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 956                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 957                  * for now take this into account when calculating the number of available constants
 958                  */
 959                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
 960                 /* Set by driver quirks in directx.c */
 961                 max_constantsF -= gl_info->reserved_glsl_constants;
 962             }
 963             else
 964             {
 965                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 966             }
 967         }
 968         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
 969         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 970     }
 971
 972     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
 973      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
 974      */
 975     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
 976         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 977
 978     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
 979         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 980
 981     if(!pshader) {
 982         shader_addline(buffer, "uniform vec4 posFixup;\n");
 983         /* Predeclaration; This function is added at link time based on the pixel shader.
 984          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 985          * that. We know the input to the reorder function at vertex shader compile time, so
 986          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 987          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 988          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 989          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 990          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 991          * inout.
 992          */
 993         if (reg_maps->shader_version.major >= 3)
 994         {
 995             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 996         } else {
 997             shader_addline(buffer, "void order_ps_input();\n");
 998         }
 999     } else {
1000         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1001         {
1002             if (!(map & 1)) continue;
1003
1004             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1005
1006             if (reg_maps->luminanceparams & (1 << i))
1007             {
1008                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1009                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1010                 extra_constants_needed++;
1011             }
1012
1013             extra_constants_needed++;
1014         }
1015
1016         if (ps_args->srgb_correction)
1017         {
1018             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1019                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1020             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1021                     srgb_cmp);
1022         }
1023         if (reg_maps->vpos || reg_maps->usesdsy)
1024         {
1025             if (This->baseShader.limits.constant_float + extra_constants_needed
1026                     + 1 < gl_info->limits.glsl_ps_float_constants)
1027             {
1028                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1029                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1030                 extra_constants_needed++;
1031             } else {
1032                 /* This happens because we do not have proper tracking of the constant registers that are
1033                  * actually used, only the max limit of the shader version
1034                  */
1035                 FIXME("Cannot find a free uniform for vpos correction params\n");
1036                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1037                         context->render_offscreen ? 0.0f : ((IWineD3DSurfaceImpl *)device->render_targets[0])->currentDesc.Height,
1038                         context->render_offscreen ? 1.0f : -1.0f);
1039             }
1040             shader_addline(buffer, "vec4 vpos;\n");
1041         }
1042     }
1043
1044     /* Declare texture samplers */
1045     for (i = 0; i < This->baseShader.limits.sampler; i++) {
1046         if (reg_maps->sampler_type[i])
1047         {
1048             switch (reg_maps->sampler_type[i])
1049             {
1050                 case WINED3DSTT_1D:
1051                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1052                     break;
1053                 case WINED3DSTT_2D:
1054                     if(device->stateBlock->textures[i] &&
1055                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
1056                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1057                     } else {
1058                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1059                     }
1060                     break;
1061                 case WINED3DSTT_CUBE:
1062                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1063                     break;
1064                 case WINED3DSTT_VOLUME:
1065                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1066                     break;
1067                 default:
1068                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1069                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1070                     break;
1071             }
1072         }
1073     }
1074
1075     /* Declare uniforms for NP2 texcoord fixup:
1076      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1077      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1078      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1079     if (pshader && ps_args->np2_fixup) {
1080
1081         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1082         UINT cur = 0;
1083
1084         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1085          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1086          * samplerNP2Fixup stores texture dimensions and is updated through
1087          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1088
1089         for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1090             if (reg_maps->sampler_type[i]) {
1091                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1092
1093                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1094                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1095                     continue;
1096                 }
1097
1098                 fixup->idx[i] = cur++;
1099             }
1100         }
1101
1102         fixup->num_consts = (cur + 1) >> 1;
1103         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1104     }
1105
1106     /* Declare address variables */
1107     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1108     {
1109         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1110     }
1111
1112     /* Declare texture coordinate temporaries and initialize them */
1113     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1114     {
1115         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1116     }
1117
1118     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1119      * helper function shader that is linked in at link time
1120      */
1121     if (pshader && reg_maps->shader_version.major >= 3)
1122     {
1123         if (use_vs(device->stateBlock))
1124         {
1125             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1126         } else {
1127             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1128              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1129              * pixel shader that reads the fixed function color into the packed input registers.
1130              */
1131             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1132         }
1133     }
1134
1135     /* Declare output register temporaries */
1136     if(This->baseShader.limits.packed_output) {
1137         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1138     }
1139
1140     /* Declare temporary variables */
1141     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1142     {
1143         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1144     }
1145
1146     /* Declare attributes */
1147     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1148     {
1149         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1150         {
1151             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1152         }
1153     }
1154
1155     /* Declare loop registers aLx */
1156     for (i = 0; i < reg_maps->loop_depth; i++) {
1157         shader_addline(buffer, "int aL%u;\n", i);
1158         shader_addline(buffer, "int tmpInt%u;\n", i);
1159     }
1160
1161     /* Temporary variables for matrix operations */
1162     shader_addline(buffer, "vec4 tmp0;\n");
1163     shader_addline(buffer, "vec4 tmp1;\n");
1164
1165     /* Local constants use a different name so they can be loaded once at shader link time
1166      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1167      * float -> string conversion can cause precision loss.
1168      */
1169     if(!This->baseShader.load_local_constsF) {
1170         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1171             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1172         }
1173     }
1174
1175     shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
1176
1177     /* Start the main program */
1178     shader_addline(buffer, "void main() {\n");
1179     if(pshader && reg_maps->vpos) {
1180         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1181          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1182          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1183          * precision troubles when we just substract 0.5.
1184          *
1185          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1186          *
1187          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1188          *
1189          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1190          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1191          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1192          * correctly on drivers that returns integer values.
1193          */
1194         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1195     }
1196 }
1197
1198 /*****************************************************************************
1199  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1200  *
1201  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1202  ****************************************************************************/
1203
1204 /* Prototypes */
1205 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1206         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1207
1208 /** Used for opcode modifiers - They multiply the result by the specified amount */
1209 static const char * const shift_glsl_tab[] = {
1210     "",           /*  0 (none) */
1211     "2.0 * ",     /*  1 (x2)   */
1212     "4.0 * ",     /*  2 (x4)   */
1213     "8.0 * ",     /*  3 (x8)   */
1214     "16.0 * ",    /*  4 (x16)  */
1215     "32.0 * ",    /*  5 (x32)  */
1216     "",           /*  6 (x64)  */
1217     "",           /*  7 (x128) */
1218     "",           /*  8 (d256) */
1219     "",           /*  9 (d128) */
1220     "",           /* 10 (d64)  */
1221     "",           /* 11 (d32)  */
1222     "0.0625 * ",  /* 12 (d16)  */
1223     "0.125 * ",   /* 13 (d8)   */
1224     "0.25 * ",    /* 14 (d4)   */
1225     "0.5 * "      /* 15 (d2)   */
1226 };
1227
1228 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1229 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1230 {
1231     out_str[0] = 0;
1232
1233     switch (src_modifier)
1234     {
1235     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1236     case WINED3DSPSM_DW:
1237     case WINED3DSPSM_NONE:
1238         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1239         break;
1240     case WINED3DSPSM_NEG:
1241         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1242         break;
1243     case WINED3DSPSM_NOT:
1244         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1245         break;
1246     case WINED3DSPSM_BIAS:
1247         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1248         break;
1249     case WINED3DSPSM_BIASNEG:
1250         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1251         break;
1252     case WINED3DSPSM_SIGN:
1253         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1254         break;
1255     case WINED3DSPSM_SIGNNEG:
1256         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1257         break;
1258     case WINED3DSPSM_COMP:
1259         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1260         break;
1261     case WINED3DSPSM_X2:
1262         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1263         break;
1264     case WINED3DSPSM_X2NEG:
1265         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1266         break;
1267     case WINED3DSPSM_ABS:
1268         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1269         break;
1270     case WINED3DSPSM_ABSNEG:
1271         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1272         break;
1273     default:
1274         FIXME("Unhandled modifier %u\n", src_modifier);
1275         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1276     }
1277 }
1278
1279 /** Writes the GLSL variable name that corresponds to the register that the
1280  * DX opcode parameter is trying to access */
1281 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1282         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1283 {
1284     /* oPos, oFog and oPts in D3D */
1285     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
1286
1287     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1288     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1289     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1290
1291     *is_color = FALSE;
1292
1293     switch (reg->type)
1294     {
1295         case WINED3DSPR_TEMP:
1296             sprintf(register_name, "R%u", reg->idx);
1297             break;
1298
1299         case WINED3DSPR_INPUT:
1300             /* vertex shaders */
1301             if (!pshader)
1302             {
1303                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1304                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1305                 sprintf(register_name, "attrib%u", reg->idx);
1306                 break;
1307             }
1308
1309             /* pixel shaders >= 3.0 */
1310             if (This->baseShader.reg_maps.shader_version.major >= 3)
1311             {
1312                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1313                 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1314
1315                 if (reg->rel_addr)
1316                 {
1317                     glsl_src_param_t rel_param;
1318
1319                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1320
1321                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1322                      * operation there */
1323                     if (idx)
1324                     {
1325                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1326                         {
1327                             sprintf(register_name,
1328                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1329                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1330                                     rel_param.param_str, idx);
1331                         }
1332                         else
1333                         {
1334                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1335                         }
1336                     }
1337                     else
1338                     {
1339                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1340                         {
1341                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1342                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1343                                     rel_param.param_str);
1344                         }
1345                         else
1346                         {
1347                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1348                         }
1349                     }
1350                 }
1351                 else
1352                 {
1353                     if (idx == in_count) sprintf(register_name, "gl_Color");
1354                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1355                     else sprintf(register_name, "IN[%u]", idx);
1356                 }
1357             }
1358             else
1359             {
1360                 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1361                 else strcpy(register_name, "gl_SecondaryColor");
1362                 break;
1363             }
1364             break;
1365
1366         case WINED3DSPR_CONST:
1367             {
1368                 const char prefix = pshader ? 'P' : 'V';
1369
1370                 /* Relative addressing */
1371                 if (reg->rel_addr)
1372                 {
1373                     glsl_src_param_t rel_param;
1374                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1375                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1376                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1377                 }
1378                 else
1379                 {
1380                     if (shader_constant_is_local(This, reg->idx))
1381                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1382                     else
1383                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1384                 }
1385             }
1386             break;
1387
1388         case WINED3DSPR_CONSTINT:
1389             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1390             else sprintf(register_name, "VI[%u]", reg->idx);
1391             break;
1392
1393         case WINED3DSPR_CONSTBOOL:
1394             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1395             else sprintf(register_name, "VB[%u]", reg->idx);
1396             break;
1397
1398         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1399             if (pshader) sprintf(register_name, "T%u", reg->idx);
1400             else sprintf(register_name, "A%u", reg->idx);
1401             break;
1402
1403         case WINED3DSPR_LOOP:
1404             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1405             break;
1406
1407         case WINED3DSPR_SAMPLER:
1408             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1409             else sprintf(register_name, "Vsampler%u", reg->idx);
1410             break;
1411
1412         case WINED3DSPR_COLOROUT:
1413             if (reg->idx >= gl_info->limits.buffers)
1414                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1415
1416             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1417             break;
1418
1419         case WINED3DSPR_RASTOUT:
1420             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1421             break;
1422
1423         case WINED3DSPR_DEPTHOUT:
1424             sprintf(register_name, "gl_FragDepth");
1425             break;
1426
1427         case WINED3DSPR_ATTROUT:
1428             if (reg->idx == 0) sprintf(register_name, "gl_FrontColor");
1429             else sprintf(register_name, "gl_FrontSecondaryColor");
1430             break;
1431
1432         case WINED3DSPR_TEXCRDOUT:
1433             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1434             if (This->baseShader.reg_maps.shader_version.major >= 3) sprintf(register_name, "OUT[%u]", reg->idx);
1435             else sprintf(register_name, "gl_TexCoord[%u]", reg->idx);
1436             break;
1437
1438         case WINED3DSPR_MISCTYPE:
1439             if (reg->idx == 0)
1440             {
1441                 /* vPos */
1442                 sprintf(register_name, "vpos");
1443             }
1444             else if (reg->idx == 1)
1445             {
1446                 /* Note that gl_FrontFacing is a bool, while vFace is
1447                  * a float for which the sign determines front/back */
1448                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1449             }
1450             else
1451             {
1452                 FIXME("Unhandled misctype register %d\n", reg->idx);
1453                 sprintf(register_name, "unrecognized_register");
1454             }
1455             break;
1456
1457         case WINED3DSPR_IMMCONST:
1458             switch (reg->immconst_type)
1459             {
1460                 case WINED3D_IMMCONST_FLOAT:
1461                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1462                     break;
1463
1464                 case WINED3D_IMMCONST_FLOAT4:
1465                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1466                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1467                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1468                     break;
1469
1470                 default:
1471                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1472                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1473             }
1474             break;
1475
1476         default:
1477             FIXME("Unhandled register name Type(%d)\n", reg->type);
1478             sprintf(register_name, "unrecognized_register");
1479             break;
1480     }
1481 }
1482
1483 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1484 {
1485     *str++ = '.';
1486     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1487     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1488     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1489     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1490     *str = '\0';
1491 }
1492
1493 /* Get the GLSL write mask for the destination register */
1494 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1495 {
1496     DWORD mask = param->write_mask;
1497
1498     if (shader_is_scalar(&param->reg))
1499     {
1500         mask = WINED3DSP_WRITEMASK_0;
1501         *write_mask = '\0';
1502     }
1503     else
1504     {
1505         shader_glsl_write_mask_to_str(mask, write_mask);
1506     }
1507
1508     return mask;
1509 }
1510
1511 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1512     unsigned int size = 0;
1513
1514     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1515     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1516     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1517     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1518
1519     return size;
1520 }
1521
1522 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1523 {
1524     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1525      * but addressed as "rgba". To fix this we need to swap the register's x
1526      * and z components. */
1527     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1528
1529     *str++ = '.';
1530     /* swizzle bits fields: wwzzyyxx */
1531     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1532     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1533     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1534     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1535     *str = '\0';
1536 }
1537
1538 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1539         BOOL fixup, DWORD mask, char *swizzle_str)
1540 {
1541     if (shader_is_scalar(&param->reg))
1542         *swizzle_str = '\0';
1543     else
1544         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1545 }
1546
1547 /* From a given parameter token, generate the corresponding GLSL string.
1548  * Also, return the actual register name and swizzle in case the
1549  * caller needs this information as well. */
1550 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1551         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1552 {
1553     BOOL is_color = FALSE;
1554     char swizzle_str[6];
1555
1556     glsl_src->reg_name[0] = '\0';
1557     glsl_src->param_str[0] = '\0';
1558     swizzle_str[0] = '\0';
1559
1560     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1561     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1562     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1563 }
1564
1565 /* From a given parameter token, generate the corresponding GLSL string.
1566  * Also, return the actual register name and swizzle in case the
1567  * caller needs this information as well. */
1568 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1569         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1570 {
1571     BOOL is_color = FALSE;
1572
1573     glsl_dst->mask_str[0] = '\0';
1574     glsl_dst->reg_name[0] = '\0';
1575
1576     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1577     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1578 }
1579
1580 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1581 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1582         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1583 {
1584     glsl_dst_param_t glsl_dst;
1585     DWORD mask;
1586
1587     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1588     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1589
1590     return mask;
1591 }
1592
1593 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1594 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1595 {
1596     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1597 }
1598
1599 /** Process GLSL instruction modifiers */
1600 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1601 {
1602     glsl_dst_param_t dst_param;
1603     DWORD modifiers;
1604
1605     if (!ins->dst_count) return;
1606
1607     modifiers = ins->dst[0].modifiers;
1608     if (!modifiers) return;
1609
1610     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1611
1612     if (modifiers & WINED3DSPDM_SATURATE)
1613     {
1614         /* _SAT means to clamp the value of the register to between 0 and 1 */
1615         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1616                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1617     }
1618
1619     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1620     {
1621         FIXME("_centroid modifier not handled\n");
1622     }
1623
1624     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1625     {
1626         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1627     }
1628 }
1629
1630 static inline const char *shader_get_comp_op(DWORD op)
1631 {
1632     switch (op) {
1633         case COMPARISON_GT: return ">";
1634         case COMPARISON_EQ: return "==";
1635         case COMPARISON_GE: return ">=";
1636         case COMPARISON_LT: return "<";
1637         case COMPARISON_NE: return "!=";
1638         case COMPARISON_LE: return "<=";
1639         default:
1640             FIXME("Unrecognized comparison value: %u\n", op);
1641             return "(\?\?)";
1642     }
1643 }
1644
1645 static void shader_glsl_get_sample_function(const struct wined3d_gl_info *gl_info,
1646         DWORD sampler_type, DWORD flags, glsl_sample_function_t *sample_function)
1647 {
1648     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1649     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1650     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1651     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1652
1653     /* Note that there's no such thing as a projected cube texture. */
1654     switch(sampler_type) {
1655         case WINED3DSTT_1D:
1656             if(lod) {
1657                 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1658             }
1659             else  if (grad)
1660             {
1661                 if (gl_info->supported[EXT_GPU_SHADER4])
1662                     sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1663                 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1664                     sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1665                 else
1666                 {
1667                     FIXME("Unsupported 1D grad function.\n");
1668                     sample_function->name = "unsupported1DGrad";
1669                 }
1670             }
1671             else
1672             {
1673                 sample_function->name = projected ? "texture1DProj" : "texture1D";
1674             }
1675             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1676             break;
1677         case WINED3DSTT_2D:
1678             if(texrect) {
1679                 if(lod) {
1680                     sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1681                 }
1682                 else  if (grad)
1683                 {
1684                     if (gl_info->supported[EXT_GPU_SHADER4])
1685                         sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1686                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1687                         sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1688                     else
1689                     {
1690                         FIXME("Unsupported RECT grad function.\n");
1691                         sample_function->name = "unsupported2DRectGrad";
1692                     }
1693                 }
1694                 else
1695                 {
1696                     sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1697                 }
1698             } else {
1699                 if(lod) {
1700                     sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1701                 }
1702                 else  if (grad)
1703                 {
1704                     if (gl_info->supported[EXT_GPU_SHADER4])
1705                         sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1706                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1707                         sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1708                     else
1709                     {
1710                         FIXME("Unsupported 2D grad function.\n");
1711                         sample_function->name = "unsupported2DGrad";
1712                     }
1713                 }
1714                 else
1715                 {
1716                     sample_function->name = projected ? "texture2DProj" : "texture2D";
1717                 }
1718             }
1719             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1720             break;
1721         case WINED3DSTT_CUBE:
1722             if(lod) {
1723                 sample_function->name = "textureCubeLod";
1724             }
1725             else if (grad)
1726             {
1727                 if (gl_info->supported[EXT_GPU_SHADER4])
1728                     sample_function->name = "textureCubeGrad";
1729                 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1730                     sample_function->name = "textureCubeGradARB";
1731                 else
1732                 {
1733                     FIXME("Unsupported Cube grad function.\n");
1734                     sample_function->name = "unsupportedCubeGrad";
1735                 }
1736             }
1737             else
1738             {
1739                 sample_function->name = "textureCube";
1740             }
1741             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1742             break;
1743         case WINED3DSTT_VOLUME:
1744             if(lod) {
1745                 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1746             }
1747             else  if (grad)
1748             {
1749                 if (gl_info->supported[EXT_GPU_SHADER4])
1750                     sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1751                 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1752                     sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1753                 else
1754                 {
1755                     FIXME("Unsupported 3D grad function.\n");
1756                     sample_function->name = "unsupported3DGrad";
1757                 }
1758             }
1759             else
1760             {
1761                 sample_function->name = projected ? "texture3DProj" : "texture3D";
1762             }
1763             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1764             break;
1765         default:
1766             sample_function->name = "";
1767             sample_function->coord_mask = 0;
1768             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1769             break;
1770     }
1771 }
1772
1773 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1774         BOOL sign_fixup, enum fixup_channel_source channel_source)
1775 {
1776     switch(channel_source)
1777     {
1778         case CHANNEL_SOURCE_ZERO:
1779             strcat(arguments, "0.0");
1780             break;
1781
1782         case CHANNEL_SOURCE_ONE:
1783             strcat(arguments, "1.0");
1784             break;
1785
1786         case CHANNEL_SOURCE_X:
1787             strcat(arguments, reg_name);
1788             strcat(arguments, ".x");
1789             break;
1790
1791         case CHANNEL_SOURCE_Y:
1792             strcat(arguments, reg_name);
1793             strcat(arguments, ".y");
1794             break;
1795
1796         case CHANNEL_SOURCE_Z:
1797             strcat(arguments, reg_name);
1798             strcat(arguments, ".z");
1799             break;
1800
1801         case CHANNEL_SOURCE_W:
1802             strcat(arguments, reg_name);
1803             strcat(arguments, ".w");
1804             break;
1805
1806         default:
1807             FIXME("Unhandled channel source %#x\n", channel_source);
1808             strcat(arguments, "undefined");
1809             break;
1810     }
1811
1812     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1813 }
1814
1815 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1816 {
1817     struct wined3d_shader_dst_param dst;
1818     unsigned int mask_size, remaining;
1819     glsl_dst_param_t dst_param;
1820     char arguments[256];
1821     DWORD mask;
1822
1823     mask = 0;
1824     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1825     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1826     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1827     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1828     mask &= ins->dst[0].write_mask;
1829
1830     if (!mask) return; /* Nothing to do */
1831
1832     if (is_complex_fixup(fixup))
1833     {
1834         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1835         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1836         return;
1837     }
1838
1839     mask_size = shader_glsl_get_write_mask_size(mask);
1840
1841     dst = ins->dst[0];
1842     dst.write_mask = mask;
1843     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1844
1845     arguments[0] = '\0';
1846     remaining = mask_size;
1847     if (mask & WINED3DSP_WRITEMASK_0)
1848     {
1849         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1850         if (--remaining) strcat(arguments, ", ");
1851     }
1852     if (mask & WINED3DSP_WRITEMASK_1)
1853     {
1854         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1855         if (--remaining) strcat(arguments, ", ");
1856     }
1857     if (mask & WINED3DSP_WRITEMASK_2)
1858     {
1859         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1860         if (--remaining) strcat(arguments, ", ");
1861     }
1862     if (mask & WINED3DSP_WRITEMASK_3)
1863     {
1864         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1865         if (--remaining) strcat(arguments, ", ");
1866     }
1867
1868     if (mask_size > 1)
1869     {
1870         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1871                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1872     }
1873     else
1874     {
1875         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1876     }
1877 }
1878
1879 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
1880         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
1881         const char *dx, const char *dy,
1882         const char *bias, const char *coord_reg_fmt, ...)
1883 {
1884     const char *sampler_base;
1885     char dst_swizzle[6];
1886     struct color_fixup_desc fixup;
1887     BOOL np2_fixup = FALSE;
1888     va_list args;
1889
1890     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
1891
1892     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1893     {
1894         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1895         fixup = priv->cur_ps_args->color_fixup[sampler];
1896         sampler_base = "Psampler";
1897
1898         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
1899             if(bias) {
1900                 FIXME("Biased sampling from NP2 textures is unsupported\n");
1901             } else {
1902                 np2_fixup = TRUE;
1903             }
1904         }
1905     } else {
1906         sampler_base = "Vsampler";
1907         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
1908     }
1909
1910     shader_glsl_append_dst(ins->ctx->buffer, ins);
1911
1912     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
1913
1914     va_start(args, coord_reg_fmt);
1915     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
1916     va_end(args);
1917
1918     if(bias) {
1919         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
1920     } else {
1921         if (np2_fixup) {
1922             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1923             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
1924
1925             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
1926                            (idx % 2) ? "zw" : "xy", dst_swizzle);
1927         } else if(dx && dy) {
1928             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
1929         } else {
1930             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
1931         }
1932     }
1933
1934     if(!is_identity_fixup(fixup)) {
1935         shader_glsl_color_correction(ins, fixup);
1936     }
1937 }
1938
1939 /*****************************************************************************
1940  * Begin processing individual instruction opcodes
1941  ****************************************************************************/
1942
1943 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1944 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
1945 {
1946     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1947     glsl_src_param_t src0_param;
1948     glsl_src_param_t src1_param;
1949     DWORD write_mask;
1950     char op;
1951
1952     /* Determine the GLSL operator to use based on the opcode */
1953     switch (ins->handler_idx)
1954     {
1955         case WINED3DSIH_MUL: op = '*'; break;
1956         case WINED3DSIH_ADD: op = '+'; break;
1957         case WINED3DSIH_SUB: op = '-'; break;
1958         default:
1959             op = ' ';
1960             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1961             break;
1962     }
1963
1964     write_mask = shader_glsl_append_dst(buffer, ins);
1965     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1966     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
1967     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1968 }
1969
1970 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1971 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
1972 {
1973     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1974     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1975     glsl_src_param_t src0_param;
1976     DWORD write_mask;
1977
1978     write_mask = shader_glsl_append_dst(buffer, ins);
1979     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1980
1981     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1982      * shader versions WINED3DSIO_MOVA is used for this. */
1983     if (ins->ctx->reg_maps->shader_version.major == 1
1984             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1985             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1986     {
1987         /* This is a simple floor() */
1988         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1989         if (mask_size > 1) {
1990             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1991         } else {
1992             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1993         }
1994     }
1995     else if(ins->handler_idx == WINED3DSIH_MOVA)
1996     {
1997         /* We need to *round* to the nearest int here. */
1998         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1999
2000         if (gl_info->supported[EXT_GPU_SHADER4])
2001         {
2002             if (mask_size > 1)
2003                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2004             else
2005                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2006         }
2007         else
2008         {
2009             if (mask_size > 1)
2010                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2011                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2012             else
2013                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2014                         src0_param.param_str, src0_param.param_str);
2015         }
2016     }
2017     else
2018     {
2019         shader_addline(buffer, "%s);\n", src0_param.param_str);
2020     }
2021 }
2022
2023 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2024 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2025 {
2026     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2027     glsl_src_param_t src0_param;
2028     glsl_src_param_t src1_param;
2029     DWORD dst_write_mask, src_write_mask;
2030     unsigned int dst_size = 0;
2031
2032     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2033     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2034
2035     /* dp3 works on vec3, dp4 on vec4 */
2036     if (ins->handler_idx == WINED3DSIH_DP4)
2037     {
2038         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2039     } else {
2040         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2041     }
2042
2043     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2044     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2045
2046     if (dst_size > 1) {
2047         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2048     } else {
2049         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2050     }
2051 }
2052
2053 /* Note that this instruction has some restrictions. The destination write mask
2054  * can't contain the w component, and the source swizzles have to be .xyzw */
2055 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2056 {
2057     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2058     glsl_src_param_t src0_param;
2059     glsl_src_param_t src1_param;
2060     char dst_mask[6];
2061
2062     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2063     shader_glsl_append_dst(ins->ctx->buffer, ins);
2064     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2065     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2066     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2067 }
2068
2069 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2070  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2071  * GLSL uses the value as-is. */
2072 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2073 {
2074     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2075     glsl_src_param_t src0_param;
2076     glsl_src_param_t src1_param;
2077     DWORD dst_write_mask;
2078     unsigned int dst_size;
2079
2080     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2081     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2082
2083     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2084     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2085
2086     if (dst_size > 1) {
2087         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2088     } else {
2089         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
2090     }
2091 }
2092
2093 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2094  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2095  * GLSL uses the value as-is. */
2096 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2097 {
2098     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2099     glsl_src_param_t src0_param;
2100     DWORD dst_write_mask;
2101     unsigned int dst_size;
2102
2103     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2104     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2105
2106     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2107
2108     if (dst_size > 1)
2109     {
2110         shader_addline(buffer, "vec%d(%s == 0.0 ? -FLT_MAX : log2(abs(%s))));\n",
2111                 dst_size, src0_param.param_str, src0_param.param_str);
2112     }
2113     else
2114     {
2115         shader_addline(buffer, "%s == 0.0 ? -FLT_MAX : log2(abs(%s)));\n",
2116                 src0_param.param_str, src0_param.param_str);
2117     }
2118 }
2119
2120 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2121 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2122 {
2123     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2124     glsl_src_param_t src_param;
2125     const char *instruction;
2126     DWORD write_mask;
2127     unsigned i;
2128
2129     /* Determine the GLSL function to use based on the opcode */
2130     /* TODO: Possibly make this a table for faster lookups */
2131     switch (ins->handler_idx)
2132     {
2133         case WINED3DSIH_MIN: instruction = "min"; break;
2134         case WINED3DSIH_MAX: instruction = "max"; break;
2135         case WINED3DSIH_ABS: instruction = "abs"; break;
2136         case WINED3DSIH_FRC: instruction = "fract"; break;
2137         case WINED3DSIH_EXP: instruction = "exp2"; break;
2138         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2139         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2140         default: instruction = "";
2141             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2142             break;
2143     }
2144
2145     write_mask = shader_glsl_append_dst(buffer, ins);
2146
2147     shader_addline(buffer, "%s(", instruction);
2148
2149     if (ins->src_count)
2150     {
2151         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2152         shader_addline(buffer, "%s", src_param.param_str);
2153         for (i = 1; i < ins->src_count; ++i)
2154         {
2155             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2156             shader_addline(buffer, ", %s", src_param.param_str);
2157         }
2158     }
2159
2160     shader_addline(buffer, "));\n");
2161 }
2162
2163 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2164 {
2165     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2166     glsl_src_param_t src_param;
2167     DWORD write_mask;
2168     char dst_mask[6];
2169
2170     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2171     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2172
2173     shader_addline(buffer, "tmp0.x = length(%s);\n", src_param.param_str);
2174     shader_glsl_append_dst(buffer, ins);
2175     shader_addline(buffer, "tmp0.x == 0.0 ? (%s * FLT_MAX) : (%s / tmp0.x));",
2176             src_param.param_str, src_param.param_str);
2177 }
2178
2179 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2180  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2181  *   dst.x = 2^(floor(src))
2182  *   dst.y = src - floor(src)
2183  *   dst.z = 2^src   (partial precision is allowed, but optional)
2184  *   dst.w = 1.0;
2185  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2186  *   dst = 2^src;    (partial precision is allowed, but optional)
2187  */
2188 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2189 {
2190     glsl_src_param_t src_param;
2191
2192     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2193
2194     if (ins->ctx->reg_maps->shader_version.major < 2)
2195     {
2196         char dst_mask[6];
2197
2198         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2199         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2200         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2201         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2202
2203         shader_glsl_append_dst(ins->ctx->buffer, ins);
2204         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2205         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2206     } else {
2207         DWORD write_mask;
2208         unsigned int mask_size;
2209
2210         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2211         mask_size = shader_glsl_get_write_mask_size(write_mask);
2212
2213         if (mask_size > 1) {
2214             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2215         } else {
2216             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2217         }
2218     }
2219 }
2220
2221 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2222 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2223 {
2224     glsl_src_param_t src_param;
2225     DWORD write_mask;
2226     unsigned int mask_size;
2227
2228     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2229     mask_size = shader_glsl_get_write_mask_size(write_mask);
2230     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2231
2232     if (mask_size > 1)
2233     {
2234         shader_addline(ins->ctx->buffer, "vec%d(%s == 0.0 ? FLT_MAX : 1.0 / %s));\n",
2235                 mask_size, src_param.param_str, src_param.param_str);
2236     }
2237     else
2238     {
2239         shader_addline(ins->ctx->buffer, "%s == 0.0 ? FLT_MAX : 1.0 / %s);\n",
2240                 src_param.param_str, src_param.param_str);
2241     }
2242 }
2243
2244 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2245 {
2246     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2247     glsl_src_param_t src_param;
2248     DWORD write_mask;
2249     unsigned int mask_size;
2250
2251     write_mask = shader_glsl_append_dst(buffer, ins);
2252     mask_size = shader_glsl_get_write_mask_size(write_mask);
2253
2254     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2255
2256     if (mask_size > 1)
2257     {
2258         shader_addline(buffer, "vec%d(%s == 0.0 ? FLT_MAX : inversesqrt(abs(%s))));\n",
2259                 mask_size, src_param.param_str, src_param.param_str);
2260     }
2261     else
2262     {
2263         shader_addline(buffer, "%s == 0.0 ? FLT_MAX : inversesqrt(abs(%s)));\n",
2264                 src_param.param_str, src_param.param_str);
2265     }
2266 }
2267
2268 /** Process signed comparison opcodes in GLSL. */
2269 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2270 {
2271     glsl_src_param_t src0_param;
2272     glsl_src_param_t src1_param;
2273     DWORD write_mask;
2274     unsigned int mask_size;
2275
2276     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2277     mask_size = shader_glsl_get_write_mask_size(write_mask);
2278     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2279     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2280
2281     if (mask_size > 1) {
2282         const char *compare;
2283
2284         switch(ins->handler_idx)
2285         {
2286             case WINED3DSIH_SLT: compare = "lessThan"; break;
2287             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2288             default: compare = "";
2289                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2290         }
2291
2292         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2293                 src0_param.param_str, src1_param.param_str);
2294     } else {
2295         switch(ins->handler_idx)
2296         {
2297             case WINED3DSIH_SLT:
2298                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2299                  * to return 0.0 but step returns 1.0 because step is not < x
2300                  * An alternative is a bvec compare padded with an unused second component.
2301                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2302                  * issue. Playing with not() is not possible either because not() does not accept
2303                  * a scalar.
2304                  */
2305                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2306                         src0_param.param_str, src1_param.param_str);
2307                 break;
2308             case WINED3DSIH_SGE:
2309                 /* Here we can use the step() function and safe a conditional */
2310                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2311                 break;
2312             default:
2313                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2314         }
2315
2316     }
2317 }
2318
2319 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2320 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2321 {
2322     glsl_src_param_t src0_param;
2323     glsl_src_param_t src1_param;
2324     glsl_src_param_t src2_param;
2325     DWORD write_mask, cmp_channel = 0;
2326     unsigned int i, j;
2327     char mask_char[6];
2328     BOOL temp_destination = FALSE;
2329
2330     if (shader_is_scalar(&ins->src[0].reg))
2331     {
2332         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2333
2334         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2335         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2336         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2337
2338         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2339                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2340     } else {
2341         DWORD dst_mask = ins->dst[0].write_mask;
2342         struct wined3d_shader_dst_param dst = ins->dst[0];
2343
2344         /* Cycle through all source0 channels */
2345         for (i=0; i<4; i++) {
2346             write_mask = 0;
2347             /* Find the destination channels which use the current source0 channel */
2348             for (j=0; j<4; j++) {
2349                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2350                 {
2351                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2352                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2353                 }
2354             }
2355             dst.write_mask = dst_mask & write_mask;
2356
2357             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2358             * The first lines may overwrite source parameters of the following lines.
2359             * Deal with that by using a temporary destination register if needed
2360             */
2361             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2362                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2363                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2364                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2365                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2366                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2367             {
2368                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2369                 if (!write_mask) continue;
2370                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2371                 temp_destination = TRUE;
2372             } else {
2373                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2374                 if (!write_mask) continue;
2375             }
2376
2377             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2378             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2379             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2380
2381             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2382                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2383         }
2384
2385         if(temp_destination) {
2386             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2387             shader_glsl_append_dst(ins->ctx->buffer, ins);
2388             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2389         }
2390     }
2391
2392 }
2393
2394 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2395 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2396  * the compare is done per component of src0. */
2397 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2398 {
2399     struct wined3d_shader_dst_param dst;
2400     glsl_src_param_t src0_param;
2401     glsl_src_param_t src1_param;
2402     glsl_src_param_t src2_param;
2403     DWORD write_mask, cmp_channel = 0;
2404     unsigned int i, j;
2405     DWORD dst_mask;
2406     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2407             ins->ctx->reg_maps->shader_version.minor);
2408
2409     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2410     {
2411         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2412         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2413         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2414         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2415
2416         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2417         if (ins->coissue)
2418         {
2419             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2420         } else {
2421             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2422                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2423         }
2424         return;
2425     }
2426     /* Cycle through all source0 channels */
2427     dst_mask = ins->dst[0].write_mask;
2428     dst = ins->dst[0];
2429     for (i=0; i<4; i++) {
2430         write_mask = 0;
2431         /* Find the destination channels which use the current source0 channel */
2432         for (j=0; j<4; j++) {
2433             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2434             {
2435                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2436                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2437             }
2438         }
2439
2440         dst.write_mask = dst_mask & write_mask;
2441         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2442         if (!write_mask) continue;
2443
2444         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2445         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2446         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2447
2448         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2449                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2450     }
2451 }
2452
2453 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2454 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2455 {
2456     glsl_src_param_t src0_param;
2457     glsl_src_param_t src1_param;
2458     glsl_src_param_t src2_param;
2459     DWORD write_mask;
2460
2461     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2462     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2463     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2464     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2465     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2466             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2467 }
2468
2469 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2470    Vertex shaders to GLSL codes */
2471 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2472 {
2473     int i;
2474     int nComponents = 0;
2475     struct wined3d_shader_dst_param tmp_dst = {{0}};
2476     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2477     struct wined3d_shader_instruction tmp_ins;
2478
2479     memset(&tmp_ins, 0, sizeof(tmp_ins));
2480
2481     /* Set constants for the temporary argument */
2482     tmp_ins.ctx = ins->ctx;
2483     tmp_ins.dst_count = 1;
2484     tmp_ins.dst = &tmp_dst;
2485     tmp_ins.src_count = 2;
2486     tmp_ins.src = tmp_src;
2487
2488     switch(ins->handler_idx)
2489     {
2490         case WINED3DSIH_M4x4:
2491             nComponents = 4;
2492             tmp_ins.handler_idx = WINED3DSIH_DP4;
2493             break;
2494         case WINED3DSIH_M4x3:
2495             nComponents = 3;
2496             tmp_ins.handler_idx = WINED3DSIH_DP4;
2497             break;
2498         case WINED3DSIH_M3x4:
2499             nComponents = 4;
2500             tmp_ins.handler_idx = WINED3DSIH_DP3;
2501             break;
2502         case WINED3DSIH_M3x3:
2503             nComponents = 3;
2504             tmp_ins.handler_idx = WINED3DSIH_DP3;
2505             break;
2506         case WINED3DSIH_M3x2:
2507             nComponents = 2;
2508             tmp_ins.handler_idx = WINED3DSIH_DP3;
2509             break;
2510         default:
2511             break;
2512     }
2513
2514     tmp_dst = ins->dst[0];
2515     tmp_src[0] = ins->src[0];
2516     tmp_src[1] = ins->src[1];
2517     for (i = 0; i < nComponents; ++i)
2518     {
2519         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2520         shader_glsl_dot(&tmp_ins);
2521         ++tmp_src[1].reg.idx;
2522     }
2523 }
2524
2525 /**
2526     The LRP instruction performs a component-wise linear interpolation
2527     between the second and third operands using the first operand as the
2528     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2529     This is equivalent to mix(src2, src1, src0);
2530 */
2531 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2532 {
2533     glsl_src_param_t src0_param;
2534     glsl_src_param_t src1_param;
2535     glsl_src_param_t src2_param;
2536     DWORD write_mask;
2537
2538     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2539
2540     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2541     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2542     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2543
2544     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2545             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2546 }
2547
2548 /** Process the WINED3DSIO_LIT instruction in GLSL:
2549  * dst.x = dst.w = 1.0
2550  * dst.y = (src0.x > 0) ? src0.x
2551  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2552  *                                        where src.w is clamped at +- 128
2553  */
2554 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2555 {
2556     glsl_src_param_t src0_param;
2557     glsl_src_param_t src1_param;
2558     glsl_src_param_t src3_param;
2559     char dst_mask[6];
2560
2561     shader_glsl_append_dst(ins->ctx->buffer, ins);
2562     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2563
2564     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2565     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2566     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2567
2568     /* The sdk specifies the instruction like this
2569      * dst.x = 1.0;
2570      * if(src.x > 0.0) dst.y = src.x
2571      * else dst.y = 0.0.
2572      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2573      * else dst.z = 0.0;
2574      * dst.w = 1.0;
2575      *
2576      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2577      * dst.x = 1.0                                  ... No further explanation needed
2578      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2579      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2580      * dst.w = 1.0.                                 ... Nothing fancy.
2581      *
2582      * So we still have one conditional in there. So do this:
2583      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2584      *
2585      * 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),
2586      * 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.
2587      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2588      */
2589     shader_addline(ins->ctx->buffer,
2590             "vec4(1.0, max(%s, 0.0), pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
2591             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2592 }
2593
2594 /** Process the WINED3DSIO_DST instruction in GLSL:
2595  * dst.x = 1.0
2596  * dst.y = src0.x * src0.y
2597  * dst.z = src0.z
2598  * dst.w = src1.w
2599  */
2600 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2601 {
2602     glsl_src_param_t src0y_param;
2603     glsl_src_param_t src0z_param;
2604     glsl_src_param_t src1y_param;
2605     glsl_src_param_t src1w_param;
2606     char dst_mask[6];
2607
2608     shader_glsl_append_dst(ins->ctx->buffer, ins);
2609     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2610
2611     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2612     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2613     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2614     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2615
2616     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2617             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2618 }
2619
2620 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2621  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2622  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2623  *
2624  * dst.x = cos(src0.?)
2625  * dst.y = sin(src0.?)
2626  * dst.z = dst.z
2627  * dst.w = dst.w
2628  */
2629 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2630 {
2631     glsl_src_param_t src0_param;
2632     DWORD write_mask;
2633
2634     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2635     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2636
2637     switch (write_mask) {
2638         case WINED3DSP_WRITEMASK_0:
2639             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2640             break;
2641
2642         case WINED3DSP_WRITEMASK_1:
2643             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2644             break;
2645
2646         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2647             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2648             break;
2649
2650         default:
2651             ERR("Write mask should be .x, .y or .xy\n");
2652             break;
2653     }
2654 }
2655
2656 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2657  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2658  * generate invalid code
2659  */
2660 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2661 {
2662     glsl_src_param_t src0_param;
2663     DWORD write_mask;
2664
2665     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2666     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2667
2668     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2669 }
2670
2671 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2672  * Start a for() loop where src1.y is the initial value of aL,
2673  *  increment aL by src1.z for a total of src1.x iterations.
2674  *  Need to use a temporary variable for this operation.
2675  */
2676 /* FIXME: I don't think nested loops will work correctly this way. */
2677 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2678 {
2679     glsl_src_param_t src1_param;
2680     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2681     const DWORD *control_values = NULL;
2682     const local_constant *constant;
2683
2684     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2685
2686     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2687      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2688      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2689      * addressing.
2690      */
2691     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2692     {
2693         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2694             if (constant->idx == ins->src[1].reg.idx)
2695             {
2696                 control_values = constant->value;
2697                 break;
2698             }
2699         }
2700     }
2701
2702     if (control_values)
2703     {
2704         struct wined3d_shader_loop_control loop_control;
2705         loop_control.count = control_values[0];
2706         loop_control.start = control_values[1];
2707         loop_control.step = (int)control_values[2];
2708
2709         if (loop_control.step > 0)
2710         {
2711             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2712                     shader->baseShader.cur_loop_depth, loop_control.start,
2713                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2714                     shader->baseShader.cur_loop_depth, loop_control.step);
2715         }
2716         else if (loop_control.step < 0)
2717         {
2718             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2719                     shader->baseShader.cur_loop_depth, loop_control.start,
2720                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2721                     shader->baseShader.cur_loop_depth, loop_control.step);
2722         }
2723         else
2724         {
2725             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2726                     shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2727                     shader->baseShader.cur_loop_depth, loop_control.count,
2728                     shader->baseShader.cur_loop_depth);
2729         }
2730     } else {
2731         shader_addline(ins->ctx->buffer,
2732                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2733                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2734                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2735                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2736     }
2737
2738     shader->baseShader.cur_loop_depth++;
2739     shader->baseShader.cur_loop_regno++;
2740 }
2741
2742 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2743 {
2744     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2745
2746     shader_addline(ins->ctx->buffer, "}\n");
2747
2748     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2749     {
2750         shader->baseShader.cur_loop_depth--;
2751         shader->baseShader.cur_loop_regno--;
2752     }
2753
2754     if (ins->handler_idx == WINED3DSIH_ENDREP)
2755     {
2756         shader->baseShader.cur_loop_depth--;
2757     }
2758 }
2759
2760 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2761 {
2762     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2763     glsl_src_param_t src0_param;
2764     const DWORD *control_values = NULL;
2765     const local_constant *constant;
2766
2767     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2768     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2769     {
2770         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2771         {
2772             if (constant->idx == ins->src[0].reg.idx)
2773             {
2774                 control_values = constant->value;
2775                 break;
2776             }
2777         }
2778     }
2779
2780     if(control_values) {
2781         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2782                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2783                        control_values[0], shader->baseShader.cur_loop_depth);
2784     } else {
2785         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2786         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2787                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2788                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2789     }
2790     shader->baseShader.cur_loop_depth++;
2791 }
2792
2793 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2794 {
2795     glsl_src_param_t src0_param;
2796
2797     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2798     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2799 }
2800
2801 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2802 {
2803     glsl_src_param_t src0_param;
2804     glsl_src_param_t src1_param;
2805
2806     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2807     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2808
2809     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2810             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2811 }
2812
2813 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2814 {
2815     shader_addline(ins->ctx->buffer, "} else {\n");
2816 }
2817
2818 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2819 {
2820     shader_addline(ins->ctx->buffer, "break;\n");
2821 }
2822
2823 /* FIXME: According to MSDN the compare is done per component. */
2824 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2825 {
2826     glsl_src_param_t src0_param;
2827     glsl_src_param_t src1_param;
2828
2829     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2830     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2831
2832     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2833             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2834 }
2835
2836 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2837 {
2838     shader_addline(ins->ctx->buffer, "}\n");
2839     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2840 }
2841
2842 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2843 {
2844     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2845 }
2846
2847 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2848 {
2849     glsl_src_param_t src1_param;
2850
2851     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2852     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
2853 }
2854
2855 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
2856 {
2857     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
2858      * function only suppresses the unhandled instruction warning
2859      */
2860 }
2861
2862 /*********************************************
2863  * Pixel Shader Specific Code begins here
2864  ********************************************/
2865 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
2866 {
2867     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2868     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2869     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2870             ins->ctx->reg_maps->shader_version.minor);
2871     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2872     glsl_sample_function_t sample_function;
2873     DWORD sample_flags = 0;
2874     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
2875     DWORD sampler_idx;
2876     DWORD mask = 0, swizzle;
2877
2878     /* 1.0-1.4: Use destination register as sampler source.
2879      * 2.0+: Use provided sampler source. */
2880     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
2881     else sampler_idx = ins->src[1].reg.idx;
2882     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2883
2884     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2885     {
2886         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2887
2888         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2889         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2890             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2891             switch (flags & ~WINED3DTTFF_PROJECTED) {
2892                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2893                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2894                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2895                 case WINED3DTTFF_COUNT4:
2896                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2897             }
2898         }
2899     }
2900     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
2901     {
2902         DWORD src_mod = ins->src[0].modifiers;
2903
2904         if (src_mod == WINED3DSPSM_DZ) {
2905             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2906             mask = WINED3DSP_WRITEMASK_2;
2907         } else if (src_mod == WINED3DSPSM_DW) {
2908             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2909             mask = WINED3DSP_WRITEMASK_3;
2910         }
2911     } else {
2912         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
2913         {
2914             /* ps 2.0 texldp instruction always divides by the fourth component. */
2915             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2916             mask = WINED3DSP_WRITEMASK_3;
2917         }
2918     }
2919
2920     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2921        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2922         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2923     }
2924
2925     shader_glsl_get_sample_function(gl_info, sampler_type, sample_flags, &sample_function);
2926     mask |= sample_function.coord_mask;
2927
2928     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
2929     else swizzle = ins->src[1].swizzle;
2930
2931     /* 1.0-1.3: Use destination register as coordinate source.
2932        1.4+: Use provided coordinate source register. */
2933     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2934     {
2935         char coord_mask[6];
2936         shader_glsl_write_mask_to_str(mask, coord_mask);
2937         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2938                 "T%u%s", sampler_idx, coord_mask);
2939     } else {
2940         glsl_src_param_t coord_param;
2941         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
2942         if (ins->flags & WINED3DSI_TEXLD_BIAS)
2943         {
2944             glsl_src_param_t bias;
2945             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
2946             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
2947                     "%s", coord_param.param_str);
2948         } else {
2949             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2950                     "%s", coord_param.param_str);
2951         }
2952     }
2953 }
2954
2955 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
2956 {
2957     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2958     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2959     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2960     glsl_sample_function_t sample_function;
2961     glsl_src_param_t coord_param, dx_param, dy_param;
2962     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
2963     DWORD sampler_type;
2964     DWORD sampler_idx;
2965     DWORD swizzle = ins->src[1].swizzle;
2966
2967     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
2968     {
2969         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
2970         return shader_glsl_tex(ins);
2971     }
2972
2973     sampler_idx = ins->src[1].reg.idx;
2974     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2975     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2976        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2977         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2978     }
2979
2980     shader_glsl_get_sample_function(gl_info, sampler_type, sample_flags, &sample_function);
2981     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2982     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
2983     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
2984
2985     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
2986                                 "%s", coord_param.param_str);
2987 }
2988
2989 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
2990 {
2991     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2992     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2993     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2994     glsl_sample_function_t sample_function;
2995     glsl_src_param_t coord_param, lod_param;
2996     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
2997     DWORD sampler_type;
2998     DWORD sampler_idx;
2999     DWORD swizzle = ins->src[1].swizzle;
3000
3001     sampler_idx = ins->src[1].reg.idx;
3002     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3003     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3004        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3005         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3006     }
3007     shader_glsl_get_sample_function(gl_info, sampler_type, sample_flags, &sample_function);
3008     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3009
3010     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3011
3012     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3013             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3014     {
3015         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3016          * However, they seem to work just fine in fragment shaders as well. */
3017         WARN("Using %s in fragment shader.\n", sample_function.name);
3018     }
3019     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3020             "%s", coord_param.param_str);
3021 }
3022
3023 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3024 {
3025     /* FIXME: Make this work for more than just 2D textures */
3026     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3027     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3028
3029     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3030     {
3031         char dst_mask[6];
3032
3033         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3034         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3035                 ins->dst[0].reg.idx, dst_mask);
3036     } else {
3037         DWORD reg = ins->src[0].reg.idx;
3038         DWORD src_mod = ins->src[0].modifiers;
3039         char dst_swizzle[6];
3040
3041         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3042
3043         if (src_mod == WINED3DSPSM_DZ) {
3044             glsl_src_param_t div_param;
3045             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3046             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3047
3048             if (mask_size > 1) {
3049                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3050             } else {
3051                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3052             }
3053         } else if (src_mod == WINED3DSPSM_DW) {
3054             glsl_src_param_t div_param;
3055             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3056             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3057
3058             if (mask_size > 1) {
3059                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3060             } else {
3061                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3062             }
3063         } else {
3064             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3065         }
3066     }
3067 }
3068
3069 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3070  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3071  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3072 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3073 {
3074     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3075     glsl_src_param_t src0_param;
3076     glsl_sample_function_t sample_function;
3077     DWORD sampler_idx = ins->dst[0].reg.idx;
3078     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3079     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3080     UINT mask_size;
3081
3082     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3083
3084     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3085      * scalar, and projected sampling would require 4.
3086      *
3087      * It is a dependent read - not valid with conditional NP2 textures
3088      */
3089     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3090     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3091
3092     switch(mask_size)
3093     {
3094         case 1:
3095             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3096                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3097             break;
3098
3099         case 2:
3100             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3101                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3102             break;
3103
3104         case 3:
3105             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3106                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3107             break;
3108
3109         default:
3110             FIXME("Unexpected mask size %u\n", mask_size);
3111             break;
3112     }
3113 }
3114
3115 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3116  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3117 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3118 {
3119     glsl_src_param_t src0_param;
3120     DWORD dstreg = ins->dst[0].reg.idx;
3121     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3122     DWORD dst_mask;
3123     unsigned int mask_size;
3124
3125     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3126     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3127     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3128
3129     if (mask_size > 1) {
3130         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3131     } else {
3132         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3133     }
3134 }
3135
3136 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3137  * Calculate the depth as dst.x / dst.y   */
3138 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3139 {
3140     glsl_dst_param_t dst_param;
3141
3142     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3143
3144     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3145      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3146      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3147      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3148      * >= 1.0 or < 0.0
3149      */
3150     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3151             dst_param.reg_name, dst_param.reg_name);
3152 }
3153
3154 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3155  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3156  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3157  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3158  */
3159 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3160 {
3161     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3162     DWORD dstreg = ins->dst[0].reg.idx;
3163     glsl_src_param_t src0_param;
3164
3165     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3166
3167     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3168     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3169 }
3170
3171 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3172  * Calculate the 1st of a 2-row matrix multiplication. */
3173 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3174 {
3175     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3176     DWORD reg = ins->dst[0].reg.idx;
3177     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3178     glsl_src_param_t src0_param;
3179
3180     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3181     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3182 }
3183
3184 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3185  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3186 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3187 {
3188     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3189     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3190     DWORD reg = ins->dst[0].reg.idx;
3191     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3192     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3193     glsl_src_param_t src0_param;
3194
3195     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3196     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3197     current_state->texcoord_w[current_state->current_row++] = reg;
3198 }
3199
3200 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3201 {
3202     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3203     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3204     DWORD reg = ins->dst[0].reg.idx;
3205     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3206     glsl_src_param_t src0_param;
3207     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3208     glsl_sample_function_t sample_function;
3209
3210     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3211     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3212
3213     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3214
3215     /* Sample the texture using the calculated coordinates */
3216     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3217 }
3218
3219 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3220  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3221 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3222 {
3223     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3224     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3225     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3226     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3227     glsl_src_param_t src0_param;
3228     DWORD reg = ins->dst[0].reg.idx;
3229     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3230     glsl_sample_function_t sample_function;
3231
3232     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3233     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3234
3235     /* Dependent read, not valid with conditional NP2 */
3236     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3237
3238     /* Sample the texture using the calculated coordinates */
3239     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3240
3241     current_state->current_row = 0;
3242 }
3243
3244 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3245  * Perform the 3rd row of a 3x3 matrix multiply */
3246 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3247 {
3248     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3249     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3250     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3251     glsl_src_param_t src0_param;
3252     char dst_mask[6];
3253     DWORD reg = ins->dst[0].reg.idx;
3254
3255     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3256
3257     shader_glsl_append_dst(ins->ctx->buffer, ins);
3258     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3259     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3260
3261     current_state->current_row = 0;
3262 }
3263
3264 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3265  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3266 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3267 {
3268     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3269     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3270     DWORD reg = ins->dst[0].reg.idx;
3271     glsl_src_param_t src0_param;
3272     glsl_src_param_t src1_param;
3273     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3274     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3275     WINED3DSAMPLER_TEXTURE_TYPE stype = ins->ctx->reg_maps->sampler_type[reg];
3276     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3277     glsl_sample_function_t sample_function;
3278
3279     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3280     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3281
3282     /* Perform the last matrix multiply operation */
3283     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3284     /* Reflection calculation */
3285     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3286
3287     /* Dependent read, not valid with conditional NP2 */
3288     shader_glsl_get_sample_function(gl_info, stype, 0, &sample_function);
3289
3290     /* Sample the texture */
3291     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3292
3293     current_state->current_row = 0;
3294 }
3295
3296 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3297  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3298 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3299 {
3300     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3301     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3302     DWORD reg = ins->dst[0].reg.idx;
3303     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3304     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3305     glsl_src_param_t src0_param;
3306     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3307     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3308     glsl_sample_function_t sample_function;
3309
3310     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3311
3312     /* Perform the last matrix multiply operation */
3313     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3314
3315     /* Construct the eye-ray vector from w coordinates */
3316     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3317             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3318     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3319
3320     /* Dependent read, not valid with conditional NP2 */
3321     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3322
3323     /* Sample the texture using the calculated coordinates */
3324     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3325
3326     current_state->current_row = 0;
3327 }
3328
3329 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3330  * Apply a fake bump map transform.
3331  * texbem is pshader <= 1.3 only, this saves a few version checks
3332  */
3333 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3334 {
3335     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3336     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3337     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3338     glsl_sample_function_t sample_function;
3339     glsl_src_param_t coord_param;
3340     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
3341     DWORD sampler_idx;
3342     DWORD mask;
3343     DWORD flags;
3344     char coord_mask[6];
3345
3346     sampler_idx = ins->dst[0].reg.idx;
3347     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3348
3349     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3350     /* Dependent read, not valid with conditional NP2 */
3351     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3352     mask = sample_function.coord_mask;
3353
3354     shader_glsl_write_mask_to_str(mask, coord_mask);
3355
3356     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3357          * so we can't let the GL handle this.
3358          */
3359     if (flags & WINED3DTTFF_PROJECTED) {
3360         DWORD div_mask=0;
3361         char coord_div_mask[3];
3362         switch (flags & ~WINED3DTTFF_PROJECTED) {
3363             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3364             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3365             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3366             case WINED3DTTFF_COUNT4:
3367             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3368         }
3369         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3370         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3371     }
3372
3373     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3374
3375     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3376             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3377             coord_param.param_str, coord_mask);
3378
3379     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3380     {
3381         glsl_src_param_t luminance_param;
3382         glsl_dst_param_t dst_param;
3383
3384         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3385         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3386
3387         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3388                 dst_param.reg_name, dst_param.mask_str,
3389                 luminance_param.param_str, sampler_idx, sampler_idx);
3390     }
3391 }
3392
3393 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3394 {
3395     glsl_src_param_t src0_param, src1_param;
3396     DWORD sampler_idx = ins->dst[0].reg.idx;
3397
3398     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3399     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3400
3401     shader_glsl_append_dst(ins->ctx->buffer, ins);
3402     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3403             src0_param.param_str, sampler_idx, src1_param.param_str);
3404 }
3405
3406 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3407  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3408 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3409 {
3410     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3411     glsl_src_param_t src0_param;
3412     DWORD sampler_idx = ins->dst[0].reg.idx;
3413     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3414     glsl_sample_function_t sample_function;
3415
3416     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3417
3418     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3419     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3420             "%s.wx", src0_param.reg_name);
3421 }
3422
3423 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3424  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3425 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3426 {
3427     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3428     glsl_src_param_t src0_param;
3429     DWORD sampler_idx = ins->dst[0].reg.idx;
3430     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3431     glsl_sample_function_t sample_function;
3432
3433     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3434
3435     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3436     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3437             "%s.yz", src0_param.reg_name);
3438 }
3439
3440 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3441  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3442 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3443 {
3444     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3445     glsl_src_param_t src0_param;
3446     DWORD sampler_idx = ins->dst[0].reg.idx;
3447     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3448     glsl_sample_function_t sample_function;
3449
3450     /* Dependent read, not valid with conditional NP2 */
3451     shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3452     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3453
3454     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3455             "%s", src0_param.param_str);
3456 }
3457
3458 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3459  * If any of the first 3 components are < 0, discard this pixel */
3460 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3461 {
3462     glsl_dst_param_t dst_param;
3463
3464     /* The argument is a destination parameter, and no writemasks are allowed */
3465     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3466     if (ins->ctx->reg_maps->shader_version.major >= 2)
3467     {
3468         /* 2.0 shaders compare all 4 components in texkill */
3469         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3470     } else {
3471         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3472          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3473          * 4 components are defined, only the first 3 are used
3474          */
3475         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3476     }
3477 }
3478
3479 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3480  * dst = dot2(src0, src1) + src2 */
3481 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3482 {
3483     glsl_src_param_t src0_param;
3484     glsl_src_param_t src1_param;
3485     glsl_src_param_t src2_param;
3486     DWORD write_mask;
3487     unsigned int mask_size;
3488
3489     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3490     mask_size = shader_glsl_get_write_mask_size(write_mask);
3491
3492     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3493     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3494     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3495
3496     if (mask_size > 1) {
3497         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3498                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3499     } else {
3500         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3501                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3502     }
3503 }
3504
3505 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3506         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3507         enum vertexprocessing_mode vertexprocessing)
3508 {
3509     unsigned int i;
3510     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3511     WORD map = reg_maps->input_registers;
3512
3513     for (i = 0; map; map >>= 1, ++i)
3514     {
3515         const char *semantic_name;
3516         UINT semantic_idx;
3517         char reg_mask[6];
3518
3519         /* Unused */
3520         if (!(map & 1)) continue;
3521
3522         semantic_name = input_signature[i].semantic_name;
3523         semantic_idx = input_signature[i].semantic_idx;
3524         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3525
3526         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3527         {
3528             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3529                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3530                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3531             else
3532                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3533                         This->input_reg_map[i], reg_mask, reg_mask);
3534         }
3535         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3536         {
3537             if (semantic_idx == 0)
3538                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3539                         This->input_reg_map[i], reg_mask, reg_mask);
3540             else if (semantic_idx == 1)
3541                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3542                         This->input_reg_map[i], reg_mask, reg_mask);
3543             else
3544                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3545                         This->input_reg_map[i], reg_mask, reg_mask);
3546         }
3547         else
3548         {
3549             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3550                     This->input_reg_map[i], reg_mask, reg_mask);
3551         }
3552     }
3553 }
3554
3555 /*********************************************
3556  * Vertex Shader Specific Code begins here
3557  ********************************************/
3558
3559 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3560     glsl_program_key_t key;
3561
3562     key.vshader = entry->vshader;
3563     key.pshader = entry->pshader;
3564     key.vs_args = entry->vs_args;
3565     key.ps_args = entry->ps_args;
3566
3567     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3568     {
3569         ERR("Failed to insert program entry.\n");
3570     }
3571 }
3572
3573 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3574         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3575         struct ps_compile_args *ps_args) {
3576     struct wine_rb_entry *entry;
3577     glsl_program_key_t key;
3578
3579     key.vshader = vshader;
3580     key.pshader = pshader;
3581     key.vs_args = *vs_args;
3582     key.ps_args = *ps_args;
3583
3584     entry = wine_rb_get(&priv->program_lookup, &key);
3585     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3586 }
3587
3588 /* GL locking is done by the caller */
3589 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3590         struct glsl_shader_prog_link *entry)
3591 {
3592     glsl_program_key_t key;
3593
3594     key.vshader = entry->vshader;
3595     key.pshader = entry->pshader;
3596     key.vs_args = entry->vs_args;
3597     key.ps_args = entry->ps_args;
3598     wine_rb_remove(&priv->program_lookup, &key);
3599
3600     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3601     if (entry->vshader) list_remove(&entry->vshader_entry);
3602     if (entry->pshader) list_remove(&entry->pshader_entry);
3603     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3604     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3605     HeapFree(GetProcessHeap(), 0, entry);
3606 }
3607
3608 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3609         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3610         const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3611 {
3612     unsigned int i, j;
3613     const char *semantic_name_in, *semantic_name_out;
3614     UINT semantic_idx_in, semantic_idx_out;
3615     DWORD *set;
3616     DWORD in_idx;
3617     unsigned int in_count = vec4_varyings(3, gl_info);
3618     char reg_mask[6], reg_mask_out[6];
3619     char destination[50];
3620     WORD input_map, output_map;
3621
3622     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3623
3624     if (!output_signature)
3625     {
3626         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3627         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3628         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3629     }
3630
3631     input_map = reg_maps_in->input_registers;
3632     for (i = 0; input_map; input_map >>= 1, ++i)
3633     {
3634         if (!(input_map & 1)) continue;
3635
3636         in_idx = map[i];
3637         if (in_idx >= (in_count + 2)) {
3638             FIXME("More input varyings declared than supported, expect issues\n");
3639             continue;
3640         }
3641         else if (map[i] == ~0U)
3642         {
3643             /* Declared, but not read register */
3644             continue;
3645         }
3646
3647         if (in_idx == in_count) {
3648             sprintf(destination, "gl_FrontColor");
3649         } else if (in_idx == in_count + 1) {
3650             sprintf(destination, "gl_FrontSecondaryColor");
3651         } else {
3652             sprintf(destination, "IN[%u]", in_idx);
3653         }
3654
3655         semantic_name_in = input_signature[i].semantic_name;
3656         semantic_idx_in = input_signature[i].semantic_idx;
3657         set[map[i]] = input_signature[i].mask;
3658         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3659
3660         if (!output_signature)
3661         {
3662             if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_COLOR))
3663             {
3664                 if (semantic_idx_in == 0)
3665                     shader_addline(buffer, "%s%s = front_color%s;\n",
3666                             destination, reg_mask, reg_mask);
3667                 else if (semantic_idx_in == 1)
3668                     shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3669                             destination, reg_mask, reg_mask);
3670                 else
3671                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3672                             destination, reg_mask, reg_mask);
3673             }
3674             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_TEXCOORD))
3675             {
3676                 if (semantic_idx_in < 8)
3677                 {
3678                     shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3679                             destination, reg_mask, semantic_idx_in, reg_mask);
3680                 }
3681                 else
3682                 {
3683                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3684                             destination, reg_mask, reg_mask);
3685                 }
3686             }
3687             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_FOG))
3688             {
3689                 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3690                         destination, reg_mask, reg_mask);
3691             }
3692             else
3693             {
3694                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3695                         destination, reg_mask, reg_mask);
3696             }
3697         } else {
3698             BOOL found = FALSE;
3699
3700             output_map = reg_maps_out->output_registers;
3701             for (j = 0; output_map; output_map >>= 1, ++j)
3702             {
3703                 if (!(output_map & 1)) continue;
3704
3705                 semantic_name_out = output_signature[j].semantic_name;
3706                 semantic_idx_out = output_signature[j].semantic_idx;
3707                 shader_glsl_write_mask_to_str(output_signature[j].mask, reg_mask_out);
3708
3709                 if (semantic_idx_in == semantic_idx_out
3710                         && !strcmp(semantic_name_in, semantic_name_out))
3711                 {
3712                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3713                             destination, reg_mask, j, reg_mask);
3714                     found = TRUE;
3715                 }
3716             }
3717             if(!found) {
3718                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3719                                destination, reg_mask, reg_mask);
3720             }
3721         }
3722     }
3723
3724     /* This is solely to make the compiler / linker happy and avoid warning about undefined
3725      * varyings. It shouldn't result in any real code executed on the GPU, since all read
3726      * input varyings are assigned above, if the optimizer works properly.
3727      */
3728     for(i = 0; i < in_count + 2; i++) {
3729         if (set[i] && set[i] != WINED3DSP_WRITEMASK_ALL)
3730         {
3731             unsigned int size = 0;
3732             memset(reg_mask, 0, sizeof(reg_mask));
3733             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3734                 reg_mask[size] = 'x';
3735                 size++;
3736             }
3737             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3738                 reg_mask[size] = 'y';
3739                 size++;
3740             }
3741             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3742                 reg_mask[size] = 'z';
3743                 size++;
3744             }
3745             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3746                 reg_mask[size] = 'w';
3747                 size++;
3748             }
3749
3750             if (i == in_count) {
3751                 sprintf(destination, "gl_FrontColor");
3752             } else if (i == in_count + 1) {
3753                 sprintf(destination, "gl_FrontSecondaryColor");
3754             } else {
3755                 sprintf(destination, "IN[%u]", i);
3756             }
3757
3758             if (size == 1) {
3759                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3760             } else {
3761                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3762             }
3763         }
3764     }
3765
3766     HeapFree(GetProcessHeap(), 0, set);
3767 }
3768
3769 /* GL locking is done by the caller */
3770 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3771         IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3772 {
3773     GLhandleARB ret = 0;
3774     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3775     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3776     IWineD3DDeviceImpl *device;
3777     DWORD vs_major = vs->baseShader.reg_maps.shader_version.major;
3778     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3779     unsigned int i;
3780     const char *semantic_name;
3781     UINT semantic_idx;
3782     char reg_mask[6];
3783     const struct wined3d_shader_signature_element *output_signature;
3784
3785     shader_buffer_clear(buffer);
3786
3787     shader_addline(buffer, "#version 120\n");
3788
3789     if(vs_major < 3 && ps_major < 3) {
3790         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3791          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3792          */
3793         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3794         if ((gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W)
3795                 && ps_major == 0 && vs_major > 0 && !device->frag_pipe->ffp_proj_control)
3796         {
3797             shader_addline(buffer, "void order_ps_input() {\n");
3798             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3799                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3800                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3801                     shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3802                 }
3803             }
3804             shader_addline(buffer, "}\n");
3805         } else {
3806             shader_addline(buffer, "void order_ps_input() { /* do nothing */ }\n");
3807         }
3808     } else if(ps_major < 3 && vs_major >= 3) {
3809         WORD map = vs->baseShader.reg_maps.output_registers;
3810
3811         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3812         output_signature = vs->baseShader.output_signature;
3813
3814         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3815         for (i = 0; map; map >>= 1, ++i)
3816         {
3817             DWORD write_mask;
3818
3819             if (!(map & 1)) continue;
3820
3821             semantic_name = output_signature[i].semantic_name;
3822             semantic_idx = output_signature[i].semantic_idx;
3823             write_mask = output_signature[i].mask;
3824             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3825
3826             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3827             {
3828                 if (semantic_idx == 0)
3829                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3830                 else if (semantic_idx == 1)
3831                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3832             }
3833             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3834             {
3835                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3836             }
3837             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3838             {
3839                 if (semantic_idx < 8)
3840                 {
3841                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3842                         write_mask |= WINED3DSP_WRITEMASK_3;
3843
3844                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3845                             semantic_idx, reg_mask, i, reg_mask);
3846                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3847                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3848                 }
3849             }
3850             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3851             {
3852                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3853             }
3854             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3855             {
3856                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3857             }
3858         }
3859         shader_addline(buffer, "}\n");
3860
3861     } else if(ps_major >= 3 && vs_major >= 3) {
3862         WORD map = vs->baseShader.reg_maps.output_registers;
3863
3864         output_signature = vs->baseShader.output_signature;
3865
3866         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3867         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3868         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3869
3870         /* First, sort out position and point size. Those are not passed to the pixel shader */
3871         for (i = 0; map; map >>= 1, ++i)
3872         {
3873             if (!(map & 1)) continue;
3874
3875             semantic_name = output_signature[i].semantic_name;
3876             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3877
3878             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3879             {
3880                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3881             }
3882             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3883             {
3884                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3885             }
3886         }
3887
3888         /* Then, fix the pixel shader input */
3889         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3890                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3891
3892         shader_addline(buffer, "}\n");
3893     } else if(ps_major >= 3 && vs_major < 3) {
3894         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3895         shader_addline(buffer, "void order_ps_input() {\n");
3896         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3897          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3898          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3899          */
3900         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3901                 &ps->baseShader.reg_maps, NULL, NULL);
3902         shader_addline(buffer, "}\n");
3903     } else {
3904         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3905     }
3906
3907     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3908     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3909     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer->buffer, NULL));
3910     checkGLcall("glShaderSourceARB(ret, 1, &buffer->buffer, NULL)");
3911     GL_EXTCALL(glCompileShaderARB(ret));
3912     checkGLcall("glCompileShaderARB(ret)");
3913
3914     return ret;
3915 }
3916
3917 /* GL locking is done by the caller */
3918 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3919         GLhandleARB programId, char prefix)
3920 {
3921     const local_constant *lconst;
3922     GLint tmp_loc;
3923     const float *value;
3924     char glsl_name[8];
3925
3926     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3927         value = (const float *)lconst->value;
3928         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3929         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3930         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3931     }
3932     checkGLcall("Hardcoding local constants");
3933 }
3934
3935 /* GL locking is done by the caller */
3936 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3937         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3938         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3939 {
3940     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3941     const struct wined3d_gl_info *gl_info = context->gl_info;
3942     CONST DWORD *function = This->baseShader.function;
3943     struct shader_glsl_ctx_priv priv_ctx;
3944
3945     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3946     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3947
3948     memset(&priv_ctx, 0, sizeof(priv_ctx));
3949     priv_ctx.cur_ps_args = args;
3950     priv_ctx.cur_np2fixup_info = np2fixup_info;
3951
3952     shader_addline(buffer, "#version 120\n");
3953
3954     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3955     {
3956         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3957     }
3958     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3959     {
3960         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3961          * drivers write a warning if we don't do so
3962          */
3963         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3964     }
3965     if (gl_info->supported[EXT_GPU_SHADER4])
3966     {
3967         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3968     }
3969
3970     /* Base Declarations */
3971     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3972
3973     /* Pack 3.0 inputs */
3974     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3975     {
3976         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
3977                 This->baseShader.input_signature, reg_maps, args->vp_mode);
3978     }
3979
3980     /* Base Shader Body */
3981     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3982
3983     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3984     if (reg_maps->shader_version.major < 2)
3985     {
3986         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3987         shader_addline(buffer, "gl_FragData[0] = R0;\n");
3988     }
3989
3990     if (args->srgb_correction)
3991     {
3992         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
3993         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
3994         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
3995         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
3996         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
3997         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
3998     }
3999     /* Pixel shader < 3.0 do not replace the fog stage.
4000      * This implements linear fog computation and blending.
4001      * TODO: non linear fog
4002      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4003      * -1/(e-s) and e/(e-s) respectively.
4004      */
4005     if (reg_maps->shader_version.major < 3)
4006     {
4007         switch(args->fog) {
4008             case FOG_OFF: break;
4009             case FOG_LINEAR:
4010                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4011                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4012                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4013                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4014                 break;
4015             case FOG_EXP:
4016                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4017                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4018                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4019                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4020                 break;
4021             case FOG_EXP2:
4022                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4023                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4024                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4025                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4026                 break;
4027         }
4028     }
4029
4030     shader_addline(buffer, "}\n");
4031
4032     TRACE("Compiling shader object %u\n", shader_obj);
4033     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4034     GL_EXTCALL(glCompileShaderARB(shader_obj));
4035     print_glsl_info_log(gl_info, shader_obj);
4036
4037     /* Store the shader object */
4038     return shader_obj;
4039 }
4040
4041 /* GL locking is done by the caller */
4042 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4043         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4044         const struct vs_compile_args *args)
4045 {
4046     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4047     const struct wined3d_gl_info *gl_info = context->gl_info;
4048     CONST DWORD *function = This->baseShader.function;
4049     struct shader_glsl_ctx_priv priv_ctx;
4050
4051     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4052     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4053
4054     shader_addline(buffer, "#version 120\n");
4055
4056     if (gl_info->supported[EXT_GPU_SHADER4])
4057     {
4058         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4059     }
4060
4061     memset(&priv_ctx, 0, sizeof(priv_ctx));
4062     priv_ctx.cur_vs_args = args;
4063
4064     /* Base Declarations */
4065     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4066
4067     /* Base Shader Body */
4068     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4069
4070     /* Unpack 3.0 outputs */
4071     if (reg_maps->shader_version.major >= 3) shader_addline(buffer, "order_ps_input(OUT);\n");
4072     else shader_addline(buffer, "order_ps_input();\n");
4073
4074     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4075      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4076      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4077      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4078      */
4079     if(args->fog_src == VS_FOG_Z) {
4080         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4081     } else if (!reg_maps->fog) {
4082         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4083     }
4084
4085     /* Write the final position.
4086      *
4087      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4088      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4089      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4090      * contains 1.0 to allow a mad.
4091      */
4092     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4093     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4094     if(args->clip_enabled) {
4095         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4096     }
4097
4098     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4099      *
4100      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4101      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4102      * which is the same as z = z * 2 - w.
4103      */
4104     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4105
4106     shader_addline(buffer, "}\n");
4107
4108     TRACE("Compiling shader object %u\n", shader_obj);
4109     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4110     GL_EXTCALL(glCompileShaderARB(shader_obj));
4111     print_glsl_info_log(gl_info, shader_obj);
4112
4113     return shader_obj;
4114 }
4115
4116 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4117         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4118         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4119 {
4120     UINT i;
4121     DWORD new_size;
4122     struct glsl_ps_compiled_shader *new_array;
4123     struct glsl_pshader_private    *shader_data;
4124     struct ps_np2fixup_info        *np2fixup = NULL;
4125     GLhandleARB ret;
4126
4127     if (!shader->baseShader.backend_data)
4128     {
4129         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4130         if (!shader->baseShader.backend_data)
4131         {
4132             ERR("Failed to allocate backend data.\n");
4133             return 0;
4134         }
4135     }
4136     shader_data = shader->baseShader.backend_data;
4137
4138     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4139      * so a linear search is more performant than a hashmap or a binary search
4140      * (cache coherency etc)
4141      */
4142     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4143         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4144             if(args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4145             return shader_data->gl_shaders[i].prgId;
4146         }
4147     }
4148
4149     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4150     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4151         if (shader_data->num_gl_shaders)
4152         {
4153             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4154             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4155                                     new_size * sizeof(*shader_data->gl_shaders));
4156         } else {
4157             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4158             new_size = 1;
4159         }
4160
4161         if(!new_array) {
4162             ERR("Out of memory\n");
4163             return 0;
4164         }
4165         shader_data->gl_shaders = new_array;
4166         shader_data->shader_array_size = new_size;
4167     }
4168
4169     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4170
4171     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4172     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4173
4174     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4175             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4176
4177     shader_buffer_clear(buffer);
4178     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4179     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4180     *np2fixup_info = np2fixup;
4181
4182     return ret;
4183 }
4184
4185 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4186                                  const DWORD use_map) {
4187     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4188     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4189     return stored->fog_src == new->fog_src;
4190 }
4191
4192 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4193         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4194         const struct vs_compile_args *args)
4195 {
4196     UINT i;
4197     DWORD new_size;
4198     struct glsl_vs_compiled_shader *new_array;
4199     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4200     struct glsl_vshader_private *shader_data;
4201     GLhandleARB ret;
4202
4203     if (!shader->baseShader.backend_data)
4204     {
4205         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4206         if (!shader->baseShader.backend_data)
4207         {
4208             ERR("Failed to allocate backend data.\n");
4209             return 0;
4210         }
4211     }
4212     shader_data = shader->baseShader.backend_data;
4213
4214     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4215      * so a linear search is more performant than a hashmap or a binary search
4216      * (cache coherency etc)
4217      */
4218     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4219         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4220             return shader_data->gl_shaders[i].prgId;
4221         }
4222     }
4223
4224     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4225
4226     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4227         if (shader_data->num_gl_shaders)
4228         {
4229             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4230             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4231                                     new_size * sizeof(*shader_data->gl_shaders));
4232         } else {
4233             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4234             new_size = 1;
4235         }
4236
4237         if(!new_array) {
4238             ERR("Out of memory\n");
4239             return 0;
4240         }
4241         shader_data->gl_shaders = new_array;
4242         shader_data->shader_array_size = new_size;
4243     }
4244
4245     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4246
4247     shader_buffer_clear(buffer);
4248     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4249     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4250
4251     return ret;
4252 }
4253
4254 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4255  * It sets the programId on the current StateBlock (because it should be called
4256  * inside of the DrawPrimitive() part of the render loop).
4257  *
4258  * If a program for the given combination does not exist, create one, and store
4259  * the program in the hash table.  If it creates a program, it will link the
4260  * given objects, too.
4261  */
4262
4263 /* GL locking is done by the caller */
4264 static void set_glsl_shader_program(const struct wined3d_context *context,
4265         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4266 {
4267     IWineD3DVertexShader *vshader = use_vs ? device->stateBlock->vertexShader : NULL;
4268     IWineD3DPixelShader *pshader = use_ps ? device->stateBlock->pixelShader : NULL;
4269     const struct wined3d_gl_info *gl_info = context->gl_info;
4270     struct shader_glsl_priv *priv = device->shader_priv;
4271     struct glsl_shader_prog_link *entry    = NULL;
4272     GLhandleARB programId                  = 0;
4273     GLhandleARB reorder_shader_id          = 0;
4274     unsigned int i;
4275     char glsl_name[8];
4276     struct ps_compile_args ps_compile_args;
4277     struct vs_compile_args vs_compile_args;
4278
4279     if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4280     if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4281
4282     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4283     if (entry) {
4284         priv->glsl_program = entry;
4285         return;
4286     }
4287
4288     /* If we get to this point, then no matching program exists, so we create one */
4289     programId = GL_EXTCALL(glCreateProgramObjectARB());
4290     TRACE("Created new GLSL shader program %u\n", programId);
4291
4292     /* Create the entry */
4293     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4294     entry->programId = programId;
4295     entry->vshader = vshader;
4296     entry->pshader = pshader;
4297     entry->vs_args = vs_compile_args;
4298     entry->ps_args = ps_compile_args;
4299     entry->constant_version = 0;
4300     entry->np2Fixup_info = NULL;
4301     /* Add the hash table entry */
4302     add_glsl_program_entry(priv, entry);
4303
4304     /* Set the current program */
4305     priv->glsl_program = entry;
4306
4307     /* Attach GLSL vshader */
4308     if (vshader)
4309     {
4310         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4311                 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4312         WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4313         char tmp_name[10];
4314
4315         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4316         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4317         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4318         checkGLcall("glAttachObjectARB");
4319         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4320          * is destroyed
4321          */
4322         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4323
4324         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4325         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4326         checkGLcall("glAttachObjectARB");
4327
4328         /* Bind vertex attributes to a corresponding index number to match
4329          * the same index numbers as ARB_vertex_programs (makes loading
4330          * vertex attributes simpler).  With this method, we can use the
4331          * exact same code to load the attributes later for both ARB and
4332          * GLSL shaders.
4333          *
4334          * We have to do this here because we need to know the Program ID
4335          * in order to make the bindings work, and it has to be done prior
4336          * to linking the GLSL program. */
4337         for (i = 0; map; map >>= 1, ++i)
4338         {
4339             if (!(map & 1)) continue;
4340
4341             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4342             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4343         }
4344         checkGLcall("glBindAttribLocationARB");
4345
4346         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4347     }
4348
4349     /* Attach GLSL pshader */
4350     if (pshader)
4351     {
4352         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4353                 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4354         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4355         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4356         checkGLcall("glAttachObjectARB");
4357
4358         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4359     }
4360
4361     /* Link the program */
4362     TRACE("Linking GLSL shader program %u\n", programId);
4363     GL_EXTCALL(glLinkProgramARB(programId));
4364     shader_glsl_validate_link(gl_info, programId);
4365
4366     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4367             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4368     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4369     {
4370         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4371         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4372     }
4373     for (i = 0; i < MAX_CONST_I; ++i)
4374     {
4375         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4376         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4377     }
4378     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4379             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4380     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4381     {
4382         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4383         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4384     }
4385     for (i = 0; i < MAX_CONST_I; ++i)
4386     {
4387         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4388         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4389     }
4390
4391     if(pshader) {
4392         char name[32];
4393
4394         for(i = 0; i < MAX_TEXTURES; i++) {
4395             sprintf(name, "bumpenvmat%u", i);
4396             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4397             sprintf(name, "luminancescale%u", i);
4398             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4399             sprintf(name, "luminanceoffset%u", i);
4400             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4401         }
4402
4403         if (ps_compile_args.np2_fixup) {
4404             if (entry->np2Fixup_info) {
4405                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4406             } else {
4407                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4408             }
4409         }
4410     }
4411
4412     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4413     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4414     checkGLcall("Find glsl program uniform locations");
4415
4416     if (pshader
4417             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4418             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4419     {
4420         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4421         entry->vertex_color_clamp = GL_FALSE;
4422     } else {
4423         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4424     }
4425
4426     /* Set the shader to allow uniform loading on it */
4427     GL_EXTCALL(glUseProgramObjectARB(programId));
4428     checkGLcall("glUseProgramObjectARB(programId)");
4429
4430     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4431      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4432      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4433      * vertex shader with fixed function pixel processing is used we make sure that the card
4434      * supports enough samplers to allow the max number of vertex samplers with all possible
4435      * fixed function fragment processing setups. So once the program is linked these samplers
4436      * won't change.
4437      */
4438     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4439     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4440
4441     /* If the local constants do not have to be loaded with the environment constants,
4442      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4443      * later
4444      */
4445     if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4446     {
4447         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4448     }
4449     if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4450     {
4451         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4452     }
4453 }
4454
4455 /* GL locking is done by the caller */
4456 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
4457 {
4458     GLhandleARB program_id;
4459     GLhandleARB vshader_id, pshader_id;
4460     static const char *blt_vshader[] =
4461     {
4462         "#version 120\n"
4463         "void main(void)\n"
4464         "{\n"
4465         "    gl_Position = gl_Vertex;\n"
4466         "    gl_FrontColor = vec4(1.0);\n"
4467         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4468         "}\n"
4469     };
4470
4471     static const char *blt_pshaders[tex_type_count] =
4472     {
4473         /* tex_1d */
4474         NULL,
4475         /* tex_2d */
4476         "#version 120\n"
4477         "uniform sampler2D sampler;\n"
4478         "void main(void)\n"
4479         "{\n"
4480         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4481         "}\n",
4482         /* tex_3d */
4483         NULL,
4484         /* tex_cube */
4485         "#version 120\n"
4486         "uniform samplerCube sampler;\n"
4487         "void main(void)\n"
4488         "{\n"
4489         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4490         "}\n",
4491         /* tex_rect */
4492         "#version 120\n"
4493         "#extension GL_ARB_texture_rectangle : enable\n"
4494         "uniform sampler2DRect sampler;\n"
4495         "void main(void)\n"
4496         "{\n"
4497         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4498         "}\n",
4499     };
4500
4501     if (!blt_pshaders[tex_type])
4502     {
4503         FIXME("tex_type %#x not supported\n", tex_type);
4504         tex_type = tex_2d;
4505     }
4506
4507     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4508     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
4509     GL_EXTCALL(glCompileShaderARB(vshader_id));
4510
4511     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4512     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
4513     GL_EXTCALL(glCompileShaderARB(pshader_id));
4514
4515     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4516     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4517     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4518     GL_EXTCALL(glLinkProgramARB(program_id));
4519
4520     shader_glsl_validate_link(gl_info, program_id);
4521
4522     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4523      * is destroyed
4524      */
4525     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4526     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4527     return program_id;
4528 }
4529
4530 /* GL locking is done by the caller */
4531 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4532 {
4533     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.device;
4534     const struct wined3d_gl_info *gl_info = context->gl_info;
4535     struct shader_glsl_priv *priv = device->shader_priv;
4536     GLhandleARB program_id = 0;
4537     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4538
4539     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4540
4541     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4542     else priv->glsl_program = NULL;
4543
4544     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4545
4546     if (old_vertex_color_clamp != current_vertex_color_clamp)
4547     {
4548         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4549         {
4550             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4551             checkGLcall("glClampColorARB");
4552         }
4553         else
4554         {
4555             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4556         }
4557     }
4558
4559     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4560     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4561     GL_EXTCALL(glUseProgramObjectARB(program_id));
4562     checkGLcall("glUseProgramObjectARB");
4563
4564     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4565      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4566      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4567     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4568     {
4569         shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4570     }
4571 }
4572
4573 /* GL locking is done by the caller */
4574 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4575     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4576     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4577     struct shader_glsl_priv *priv = This->shader_priv;
4578     GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
4579
4580     if (!*blt_program) {
4581         GLint loc;
4582         *blt_program = create_glsl_blt_shader(gl_info, tex_type);
4583         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4584         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4585         GL_EXTCALL(glUniform1iARB(loc, 0));
4586     } else {
4587         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4588     }
4589 }
4590
4591 /* GL locking is done by the caller */
4592 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4593     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4594     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4595     struct shader_glsl_priv *priv = This->shader_priv;
4596     GLhandleARB program_id;
4597
4598     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4599     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4600
4601     GL_EXTCALL(glUseProgramObjectARB(program_id));
4602     checkGLcall("glUseProgramObjectARB");
4603 }
4604
4605 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4606     const struct list *linked_programs;
4607     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4608     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4609     struct shader_glsl_priv *priv = device->shader_priv;
4610     const struct wined3d_gl_info *gl_info;
4611     struct wined3d_context *context;
4612
4613     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4614      * can be called from IWineD3DBaseShader::Release
4615      */
4616     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4617
4618     if(pshader) {
4619         struct glsl_pshader_private *shader_data;
4620         shader_data = This->baseShader.backend_data;
4621         if(!shader_data || shader_data->num_gl_shaders == 0)
4622         {
4623             HeapFree(GetProcessHeap(), 0, shader_data);
4624             This->baseShader.backend_data = NULL;
4625             return;
4626         }
4627
4628         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4629         gl_info = context->gl_info;
4630
4631         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4632         {
4633             ENTER_GL();
4634             shader_glsl_select(context, FALSE, FALSE);
4635             LEAVE_GL();
4636         }
4637     } else {
4638         struct glsl_vshader_private *shader_data;
4639         shader_data = This->baseShader.backend_data;
4640         if(!shader_data || shader_data->num_gl_shaders == 0)
4641         {
4642             HeapFree(GetProcessHeap(), 0, shader_data);
4643             This->baseShader.backend_data = NULL;
4644             return;
4645         }
4646
4647         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4648         gl_info = context->gl_info;
4649
4650         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4651         {
4652             ENTER_GL();
4653             shader_glsl_select(context, FALSE, FALSE);
4654             LEAVE_GL();
4655         }
4656     }
4657
4658     linked_programs = &This->baseShader.linked_programs;
4659
4660     TRACE("Deleting linked programs\n");
4661     if (linked_programs->next) {
4662         struct glsl_shader_prog_link *entry, *entry2;
4663
4664         ENTER_GL();
4665         if(pshader) {
4666             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4667                 delete_glsl_program_entry(priv, gl_info, entry);
4668             }
4669         } else {
4670             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4671                 delete_glsl_program_entry(priv, gl_info, entry);
4672             }
4673         }
4674         LEAVE_GL();
4675     }
4676
4677     if(pshader) {
4678         UINT i;
4679         struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
4680
4681         ENTER_GL();
4682         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4683             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4684             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4685             checkGLcall("glDeleteObjectARB");
4686         }
4687         LEAVE_GL();
4688         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4689     }
4690     else
4691     {
4692         UINT i;
4693         struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
4694
4695         ENTER_GL();
4696         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4697             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4698             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4699             checkGLcall("glDeleteObjectARB");
4700         }
4701         LEAVE_GL();
4702         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4703     }
4704
4705     HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
4706     This->baseShader.backend_data = NULL;
4707
4708     context_release(context);
4709 }
4710
4711 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4712 {
4713     const glsl_program_key_t *k = key;
4714     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4715             const struct glsl_shader_prog_link, program_lookup_entry);
4716     int cmp;
4717
4718     if (k->vshader > prog->vshader) return 1;
4719     else if (k->vshader < prog->vshader) return -1;
4720
4721     if (k->pshader > prog->pshader) return 1;
4722     else if (k->pshader < prog->pshader) return -1;
4723
4724     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4725     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4726
4727     return 0;
4728 }
4729
4730 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4731 {
4732     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4733     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4734
4735     if (!mem)
4736     {
4737         ERR("Failed to allocate memory\n");
4738         return FALSE;
4739     }
4740
4741     heap->entries = mem;
4742     heap->entries[1].version = 0;
4743     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4744     heap->size = 1;
4745
4746     return TRUE;
4747 }
4748
4749 static void constant_heap_free(struct constant_heap *heap)
4750 {
4751     HeapFree(GetProcessHeap(), 0, heap->entries);
4752 }
4753
4754 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4755 {
4756     wined3d_rb_alloc,
4757     wined3d_rb_realloc,
4758     wined3d_rb_free,
4759     glsl_program_key_compare,
4760 };
4761
4762 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4763     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4764     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4765     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4766     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4767             gl_info->limits.glsl_ps_float_constants)) + 1;
4768
4769     if (!shader_buffer_init(&priv->shader_buffer))
4770     {
4771         ERR("Failed to initialize shader buffer.\n");
4772         goto fail;
4773     }
4774
4775     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4776     if (!priv->stack)
4777     {
4778         ERR("Failed to allocate memory.\n");
4779         goto fail;
4780     }
4781
4782     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4783     {
4784         ERR("Failed to initialize vertex shader constant heap\n");
4785         goto fail;
4786     }
4787
4788     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4789     {
4790         ERR("Failed to initialize pixel shader constant heap\n");
4791         goto fail;
4792     }
4793
4794     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4795     {
4796         ERR("Failed to initialize rbtree.\n");
4797         goto fail;
4798     }
4799
4800     priv->next_constant_version = 1;
4801
4802     This->shader_priv = priv;
4803     return WINED3D_OK;
4804
4805 fail:
4806     constant_heap_free(&priv->pconst_heap);
4807     constant_heap_free(&priv->vconst_heap);
4808     HeapFree(GetProcessHeap(), 0, priv->stack);
4809     shader_buffer_free(&priv->shader_buffer);
4810     HeapFree(GetProcessHeap(), 0, priv);
4811     return E_OUTOFMEMORY;
4812 }
4813
4814 /* Context activation is done by the caller. */
4815 static void shader_glsl_free(IWineD3DDevice *iface) {
4816     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4817     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4818     struct shader_glsl_priv *priv = This->shader_priv;
4819     int i;
4820
4821     ENTER_GL();
4822     for (i = 0; i < tex_type_count; ++i)
4823     {
4824         if (priv->depth_blt_program[i])
4825         {
4826             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
4827         }
4828     }
4829     LEAVE_GL();
4830
4831     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4832     constant_heap_free(&priv->pconst_heap);
4833     constant_heap_free(&priv->vconst_heap);
4834     HeapFree(GetProcessHeap(), 0, priv->stack);
4835     shader_buffer_free(&priv->shader_buffer);
4836
4837     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4838     This->shader_priv = NULL;
4839 }
4840
4841 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4842     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4843     return FALSE;
4844 }
4845
4846 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4847 {
4848     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4849      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4850      * on the version of NV_vertex_program.
4851      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4852      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4853      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4854      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4855      */
4856     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4857             || gl_info->limits.arb_ps_instructions <= 512)
4858         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4859     else
4860         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4861     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4862     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4863
4864     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4865      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4866      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4867      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4868      * in max native instructions. Intel and others also offer the info in this extension but they
4869      * don't support GLSL (at least on Windows).
4870      *
4871      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4872      * of instructions is 512 or less we have to do with ps2.0 hardware.
4873      * NOTE: ps3.0 hardware requires 512 or more instructions but ati and nvidia offer 'enough' (1024 vs 4096) on their most basic ps3.0 hardware.
4874      */
4875     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4876             || gl_info->limits.arb_ps_instructions <= 512)
4877         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4878     else
4879         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4880
4881     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4882
4883     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4884      * Direct3D minimum requirement.
4885      *
4886      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4887      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4888      *
4889      * The problem is that the refrast clamps temporary results in the shader to
4890      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4891      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4892      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4893      * offer a way to query this.
4894      */
4895     pCaps->PixelShader1xMaxValue = 8.0;
4896     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4897
4898     pCaps->VSClipping = TRUE;
4899 }
4900
4901 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4902 {
4903     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4904     {
4905         TRACE("Checking support for fixup:\n");
4906         dump_color_fixup_desc(fixup);
4907     }
4908
4909     /* We support everything except YUV conversions. */
4910     if (!is_complex_fixup(fixup))
4911     {
4912         TRACE("[OK]\n");
4913         return TRUE;
4914     }
4915
4916     TRACE("[FAILED]\n");
4917     return FALSE;
4918 }
4919
4920 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4921 {
4922     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4923     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4924     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4925     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4926     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4927     /* WINED3DSIH_BREAKP        */ NULL,
4928     /* WINED3DSIH_CALL          */ shader_glsl_call,
4929     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4930     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4931     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4932     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4933     /* WINED3DSIH_CUT           */ NULL,
4934     /* WINED3DSIH_DCL           */ NULL,
4935     /* WINED3DSIH_DEF           */ NULL,
4936     /* WINED3DSIH_DEFB          */ NULL,
4937     /* WINED3DSIH_DEFI          */ NULL,
4938     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
4939     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4940     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4941     /* WINED3DSIH_DST           */ shader_glsl_dst,
4942     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
4943     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
4944     /* WINED3DSIH_ELSE          */ shader_glsl_else,
4945     /* WINED3DSIH_EMIT          */ NULL,
4946     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
4947     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
4948     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
4949     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
4950     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
4951     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
4952     /* WINED3DSIH_IADD          */ NULL,
4953     /* WINED3DSIH_IF            */ shader_glsl_if,
4954     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
4955     /* WINED3DSIH_IGE           */ NULL,
4956     /* WINED3DSIH_LABEL         */ shader_glsl_label,
4957     /* WINED3DSIH_LIT           */ shader_glsl_lit,
4958     /* WINED3DSIH_LOG           */ shader_glsl_log,
4959     /* WINED3DSIH_LOGP          */ shader_glsl_log,
4960     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
4961     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
4962     /* WINED3DSIH_LT            */ NULL,
4963     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
4964     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
4965     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
4966     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
4967     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
4968     /* WINED3DSIH_MAD           */ shader_glsl_mad,
4969     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
4970     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
4971     /* WINED3DSIH_MOV           */ shader_glsl_mov,
4972     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
4973     /* WINED3DSIH_MUL           */ shader_glsl_arith,
4974     /* WINED3DSIH_NOP           */ NULL,
4975     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
4976     /* WINED3DSIH_PHASE         */ NULL,
4977     /* WINED3DSIH_POW           */ shader_glsl_pow,
4978     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
4979     /* WINED3DSIH_REP           */ shader_glsl_rep,
4980     /* WINED3DSIH_RET           */ shader_glsl_ret,
4981     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
4982     /* WINED3DSIH_SETP          */ NULL,
4983     /* WINED3DSIH_SGE           */ shader_glsl_compare,
4984     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
4985     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
4986     /* WINED3DSIH_SLT           */ shader_glsl_compare,
4987     /* WINED3DSIH_SUB           */ shader_glsl_arith,
4988     /* WINED3DSIH_TEX           */ shader_glsl_tex,
4989     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
4990     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
4991     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
4992     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
4993     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
4994     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
4995     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
4996     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
4997     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
4998     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
4999     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5000     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5001     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5002     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5003     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5004     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5005     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5006     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5007     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5008     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5009     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5010 };
5011
5012 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5013     SHADER_HANDLER hw_fct;
5014
5015     /* Select handler */
5016     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5017
5018     /* Unhandled opcode */
5019     if (!hw_fct)
5020     {
5021         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5022         return;
5023     }
5024     hw_fct(ins);
5025
5026     shader_glsl_add_instruction_modifiers(ins);
5027 }
5028
5029 const shader_backend_t glsl_shader_backend = {
5030     shader_glsl_handle_instruction,
5031     shader_glsl_select,
5032     shader_glsl_select_depth_blt,
5033     shader_glsl_deselect_depth_blt,
5034     shader_glsl_update_float_vertex_constants,
5035     shader_glsl_update_float_pixel_constants,
5036     shader_glsl_load_constants,
5037     shader_glsl_load_np2fixup_constants,
5038     shader_glsl_destroy,
5039     shader_glsl_alloc,
5040     shader_glsl_free,
5041     shader_glsl_dirty_const,
5042     shader_glsl_get_caps,
5043     shader_glsl_color_fixup_supported,
5044 };