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_full[tex_type_count];
  93     GLhandleARB depth_blt_program_masked[tex_type_count];
  94     UINT next_constant_version;
  95 };
  96
  97 /* Struct to maintain data about a linked GLSL program */
  98 struct glsl_shader_prog_link {
  99     struct wine_rb_entry        program_lookup_entry;
 100     struct list                 vshader_entry;
 101     struct list                 pshader_entry;
 102     GLhandleARB                 programId;
 103     GLint                       *vuniformF_locations;
 104     GLint                       *puniformF_locations;
 105     GLint                       vuniformI_locations[MAX_CONST_I];
 106     GLint                       puniformI_locations[MAX_CONST_I];
 107     GLint                       posFixup_location;
 108     GLint                       np2Fixup_location;
 109     GLint                       bumpenvmat_location[MAX_TEXTURES];
 110     GLint                       luminancescale_location[MAX_TEXTURES];
 111     GLint                       luminanceoffset_location[MAX_TEXTURES];
 112     GLint                       ycorrection_location;
 113     GLenum                      vertex_color_clamp;
 114     IWineD3DVertexShader        *vshader;
 115     IWineD3DPixelShader         *pshader;
 116     struct vs_compile_args      vs_args;
 117     struct ps_compile_args      ps_args;
 118     UINT                        constant_version;
 119     const struct ps_np2fixup_info *np2Fixup_info;
 120 };
 121
 122 typedef struct {
 123     IWineD3DVertexShader        *vshader;
 124     IWineD3DPixelShader         *pshader;
 125     struct ps_compile_args      ps_args;
 126     struct vs_compile_args      vs_args;
 127 } glsl_program_key_t;
 128
 129 struct shader_glsl_ctx_priv {
 130     const struct vs_compile_args    *cur_vs_args;
 131     const struct ps_compile_args    *cur_ps_args;
 132     struct ps_np2fixup_info         *cur_np2fixup_info;
 133 };
 134
 135 struct glsl_ps_compiled_shader
 136 {
 137     struct ps_compile_args          args;
 138     struct ps_np2fixup_info         np2fixup;
 139     GLhandleARB                     prgId;
 140 };
 141
 142 struct glsl_pshader_private
 143 {
 144     struct glsl_ps_compiled_shader  *gl_shaders;
 145     UINT                            num_gl_shaders, shader_array_size;
 146 };
 147
 148 struct glsl_vs_compiled_shader
 149 {
 150     struct vs_compile_args          args;
 151     GLhandleARB                     prgId;
 152 };
 153
 154 struct glsl_vshader_private
 155 {
 156     struct glsl_vs_compiled_shader  *gl_shaders;
 157     UINT                            num_gl_shaders, shader_array_size;
 158 };
 159
 160 static const char *debug_gl_shader_type(GLenum type)
 161 {
 162     switch (type)
 163     {
 164 #define WINED3D_TO_STR(u) case u: return #u
 165         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
 166         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
 167         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
 168 #undef WINED3D_TO_STR
 169         default:
 170             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
 171     }
 172 }
 173
 174 /* Extract a line from the info log.
 175  * Note that this modifies the source string. */
 176 static char *get_info_log_line(char **ptr)
 177 {
 178     char *p, *q;
 179
 180     p = *ptr;
 181     if (!(q = strstr(p, "\n")))
 182     {
 183         if (!*p) return NULL;
 184         *ptr += strlen(p);
 185         return p;
 186     }
 187     *q = '\0';
 188     *ptr = q + 1;
 189
 190     return p;
 191 }
 192
 193 /** Prints the GLSL info log which will contain error messages if they exist */
 194 /* GL locking is done by the caller */
 195 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
 196 {
 197     int infologLength = 0;
 198     char *infoLog;
 199     unsigned int i;
 200     BOOL is_spam;
 201
 202     static const char * const spam[] =
 203     {
 204         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
 205         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx, with \n */
 206         "Fragment shader was successfully compiled to run on hardware.",    /* fglrx, no \n   */
 207         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
 208         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
 209         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
 210         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
 211         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
 212         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
 213     };
 214
 215     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 216
 217     GL_EXTCALL(glGetObjectParameterivARB(obj,
 218                GL_OBJECT_INFO_LOG_LENGTH_ARB,
 219                &infologLength));
 220
 221     /* A size of 1 is just a null-terminated string, so the log should be bigger than
 222      * that if there are errors. */
 223     if (infologLength > 1)
 224     {
 225         char *ptr, *line;
 226
 227         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
 228          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
 229          */
 230         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
 231         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
 232         is_spam = FALSE;
 233
 234         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
 235             if(strcmp(infoLog, spam[i]) == 0) {
 236                 is_spam = TRUE;
 237                 break;
 238             }
 239         }
 240
 241         ptr = infoLog;
 242         if (is_spam)
 243         {
 244             TRACE("Spam received from GLSL shader #%u:\n", obj);
 245             while ((line = get_info_log_line(&ptr))) TRACE("    %s\n", line);
 246         }
 247         else
 248         {
 249             FIXME("Error received from GLSL shader #%u:\n", obj);
 250             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 251         }
 252         HeapFree(GetProcessHeap(), 0, infoLog);
 253     }
 254 }
 255
 256 /* GL locking is done by the caller. */
 257 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 258 {
 259     GLint i, object_count, source_size = -1;
 260     GLhandleARB *objects;
 261     char *source = NULL;
 262
 263     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
 264     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
 265     if (!objects)
 266     {
 267         ERR("Failed to allocate object array memory.\n");
 268         return;
 269     }
 270
 271     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
 272     for (i = 0; i < object_count; ++i)
 273     {
 274         char *ptr, *line;
 275         GLint tmp;
 276
 277         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
 278
 279         if (source_size < tmp)
 280         {
 281             HeapFree(GetProcessHeap(), 0, source);
 282
 283             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
 284             if (!source)
 285             {
 286                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
 287                 HeapFree(GetProcessHeap(), 0, objects);
 288                 return;
 289             }
 290             source_size = tmp;
 291         }
 292
 293         FIXME("Object %u:\n", objects[i]);
 294         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
 295         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
 296         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
 297         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
 298         FIXME("\n");
 299
 300         ptr = source;
 301         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
 302         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
 303         FIXME("\n");
 304     }
 305
 306     HeapFree(GetProcessHeap(), 0, source);
 307     HeapFree(GetProcessHeap(), 0, objects);
 308 }
 309
 310 /* GL locking is done by the caller. */
 311 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
 312 {
 313     GLint tmp;
 314
 315     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
 316
 317     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
 318     if (tmp == GL_PROGRAM_OBJECT_ARB)
 319     {
 320         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
 321         if (!tmp)
 322         {
 323             FIXME("Program %u link status invalid.\n", program);
 324             shader_glsl_dump_program_source(gl_info, program);
 325         }
 326     }
 327
 328     print_glsl_info_log(gl_info, program);
 329 }
 330
 331 /**
 332  * Loads (pixel shader) samplers
 333  */
 334 /* GL locking is done by the caller */
 335 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
 336         DWORD *tex_unit_map, GLhandleARB programId)
 337 {
 338     GLint name_loc;
 339     int i;
 340     char sampler_name[20];
 341
 342     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
 343         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
 344         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 345         if (name_loc != -1) {
 346             DWORD mapped_unit = tex_unit_map[i];
 347             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
 348             {
 349                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 350                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 351                 checkGLcall("glUniform1iARB");
 352             } else {
 353                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 354             }
 355         }
 356     }
 357 }
 358
 359 /* GL locking is done by the caller */
 360 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
 361         DWORD *tex_unit_map, GLhandleARB programId)
 362 {
 363     GLint name_loc;
 364     char sampler_name[20];
 365     int i;
 366
 367     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
 368         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
 369         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
 370         if (name_loc != -1) {
 371             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
 372             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
 373             {
 374                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
 375                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
 376                 checkGLcall("glUniform1iARB");
 377             } else {
 378                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
 379             }
 380         }
 381     }
 382 }
 383
 384 /* GL locking is done by the caller */
 385 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
 386         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 387 {
 388     int stack_idx = 0;
 389     unsigned int heap_idx = 1;
 390     unsigned int idx;
 391
 392     if (heap->entries[heap_idx].version <= version) return;
 393
 394     idx = heap->entries[heap_idx].idx;
 395     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 396     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 397
 398     while (stack_idx >= 0)
 399     {
 400         /* Note that we fall through to the next case statement. */
 401         switch(stack[stack_idx])
 402         {
 403             case HEAP_NODE_TRAVERSE_LEFT:
 404             {
 405                 unsigned int left_idx = heap_idx << 1;
 406                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 407                 {
 408                     heap_idx = left_idx;
 409                     idx = heap->entries[heap_idx].idx;
 410                     if (constant_locations[idx] != -1)
 411                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 412
 413                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 414                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 415                     break;
 416                 }
 417             }
 418
 419             case HEAP_NODE_TRAVERSE_RIGHT:
 420             {
 421                 unsigned int right_idx = (heap_idx << 1) + 1;
 422                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 423                 {
 424                     heap_idx = right_idx;
 425                     idx = heap->entries[heap_idx].idx;
 426                     if (constant_locations[idx] != -1)
 427                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
 428
 429                     stack[stack_idx++] = HEAP_NODE_POP;
 430                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 431                     break;
 432                 }
 433             }
 434
 435             case HEAP_NODE_POP:
 436             {
 437                 heap_idx >>= 1;
 438                 --stack_idx;
 439                 break;
 440             }
 441         }
 442     }
 443     checkGLcall("walk_constant_heap()");
 444 }
 445
 446 /* GL locking is done by the caller */
 447 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
 448 {
 449     GLfloat clamped_constant[4];
 450
 451     if (location == -1) return;
 452
 453     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
 454     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
 455     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
 456     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
 457
 458     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
 459 }
 460
 461 /* GL locking is done by the caller */
 462 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
 463         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
 464 {
 465     int stack_idx = 0;
 466     unsigned int heap_idx = 1;
 467     unsigned int idx;
 468
 469     if (heap->entries[heap_idx].version <= version) return;
 470
 471     idx = heap->entries[heap_idx].idx;
 472     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 473     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 474
 475     while (stack_idx >= 0)
 476     {
 477         /* Note that we fall through to the next case statement. */
 478         switch(stack[stack_idx])
 479         {
 480             case HEAP_NODE_TRAVERSE_LEFT:
 481             {
 482                 unsigned int left_idx = heap_idx << 1;
 483                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
 484                 {
 485                     heap_idx = left_idx;
 486                     idx = heap->entries[heap_idx].idx;
 487                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 488
 489                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
 490                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 491                     break;
 492                 }
 493             }
 494
 495             case HEAP_NODE_TRAVERSE_RIGHT:
 496             {
 497                 unsigned int right_idx = (heap_idx << 1) + 1;
 498                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
 499                 {
 500                     heap_idx = right_idx;
 501                     idx = heap->entries[heap_idx].idx;
 502                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
 503
 504                     stack[stack_idx++] = HEAP_NODE_POP;
 505                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
 506                     break;
 507                 }
 508             }
 509
 510             case HEAP_NODE_POP:
 511             {
 512                 heap_idx >>= 1;
 513                 --stack_idx;
 514                 break;
 515             }
 516         }
 517     }
 518     checkGLcall("walk_constant_heap_clamped()");
 519 }
 520
 521 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
 522 /* GL locking is done by the caller */
 523 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 524         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
 525         unsigned char *stack, UINT version)
 526 {
 527     const local_constant *lconst;
 528
 529     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
 530     if (This->baseShader.reg_maps.shader_version.major == 1
 531             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
 532         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
 533     else
 534         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
 535
 536     if (!This->baseShader.load_local_constsF)
 537     {
 538         TRACE("No need to load local float constants for this shader\n");
 539         return;
 540     }
 541
 542     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
 543     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
 544     {
 545         GLint location = constant_locations[lconst->idx];
 546         /* We found this uniform name in the program - go ahead and send the data */
 547         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
 548     }
 549     checkGLcall("glUniform4fvARB()");
 550 }
 551
 552 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
 553 /* GL locking is done by the caller */
 554 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 555         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
 556 {
 557     unsigned int i;
 558     struct list* ptr;
 559
 560     for (i = 0; constants_set; constants_set >>= 1, ++i)
 561     {
 562         if (!(constants_set & 1)) continue;
 563
 564         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
 565                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
 566
 567         /* We found this uniform name in the program - go ahead and send the data */
 568         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
 569         checkGLcall("glUniform4ivARB");
 570     }
 571
 572     /* Load immediate constants */
 573     ptr = list_head(&This->baseShader.constantsI);
 574     while (ptr) {
 575         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 576         unsigned int idx = lconst->idx;
 577         const GLint *values = (const GLint *)lconst->value;
 578
 579         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
 580             values[0], values[1], values[2], values[3]);
 581
 582         /* We found this uniform name in the program - go ahead and send the data */
 583         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
 584         checkGLcall("glUniform4ivARB");
 585         ptr = list_next(&This->baseShader.constantsI, ptr);
 586     }
 587 }
 588
 589 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
 590 /* GL locking is done by the caller */
 591 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 592         GLhandleARB programId, const BOOL *constants, WORD constants_set)
 593 {
 594     GLint tmp_loc;
 595     unsigned int i;
 596     char tmp_name[8];
 597     const char *prefix;
 598     struct list* ptr;
 599
 600     switch (This->baseShader.reg_maps.shader_version.type)
 601     {
 602         case WINED3D_SHADER_TYPE_VERTEX:
 603             prefix = "VB";
 604             break;
 605
 606         case WINED3D_SHADER_TYPE_GEOMETRY:
 607             prefix = "GB";
 608             break;
 609
 610         case WINED3D_SHADER_TYPE_PIXEL:
 611             prefix = "PB";
 612             break;
 613
 614         default:
 615             FIXME("Unknown shader type %#x.\n",
 616                     This->baseShader.reg_maps.shader_version.type);
 617             prefix = "UB";
 618             break;
 619     }
 620
 621     /* TODO: Benchmark and see if it would be beneficial to store the
 622      * locations of the constants to avoid looking up each time */
 623     for (i = 0; constants_set; constants_set >>= 1, ++i)
 624     {
 625         if (!(constants_set & 1)) continue;
 626
 627         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
 628
 629         /* TODO: Benchmark and see if it would be beneficial to store the
 630          * locations of the constants to avoid looking up each time */
 631         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
 632         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 633         if (tmp_loc != -1)
 634         {
 635             /* We found this uniform name in the program - go ahead and send the data */
 636             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
 637             checkGLcall("glUniform1ivARB");
 638         }
 639     }
 640
 641     /* Load immediate constants */
 642     ptr = list_head(&This->baseShader.constantsB);
 643     while (ptr) {
 644         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
 645         unsigned int idx = lconst->idx;
 646         const GLint *values = (const GLint *)lconst->value;
 647
 648         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
 649
 650         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
 651         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
 652         if (tmp_loc != -1) {
 653             /* We found this uniform name in the program - go ahead and send the data */
 654             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
 655             checkGLcall("glUniform1ivARB");
 656         }
 657         ptr = list_next(&This->baseShader.constantsB, ptr);
 658     }
 659 }
 660
 661 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
 662 {
 663     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
 664 }
 665
 666 /**
 667  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
 668  */
 669 /* GL locking is done by the caller (state handler) */
 670 static void shader_glsl_load_np2fixup_constants(
 671     IWineD3DDevice* device,
 672     char usePixelShader,
 673     char useVertexShader) {
 674
 675     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
 676     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
 677
 678     if (!prog) {
 679         /* No GLSL program set - nothing to do. */
 680         return;
 681     }
 682
 683     if (!usePixelShader) {
 684         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 685         return;
 686     }
 687
 688     if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
 689         const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 690         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
 691         UINT i;
 692         UINT fixup = prog->ps_args.np2_fixup;
 693         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 694
 695         for (i = 0; fixup; fixup >>= 1, ++i) {
 696             const unsigned char idx = prog->np2Fixup_info->idx[i];
 697             const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
 698             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 699
 700             if (!tex) {
 701                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 702                 continue;
 703             }
 704
 705             if (idx % 2) {
 706                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 707             } else {
 708                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 709             }
 710         }
 711
 712         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
 713     }
 714 }
 715
 716 /**
 717  * Loads the app-supplied constants into the currently set GLSL program.
 718  */
 719 /* GL locking is done by the caller (state handler) */
 720 static void shader_glsl_load_constants(const struct wined3d_context *context,
 721         char usePixelShader, char useVertexShader)
 722 {
 723     const struct wined3d_gl_info *gl_info = context->gl_info;
 724     IWineD3DDeviceImpl *device = context->swapchain->device;
 725     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 726     struct shader_glsl_priv *priv = device->shader_priv;
 727
 728     GLhandleARB programId;
 729     struct glsl_shader_prog_link *prog = priv->glsl_program;
 730     UINT constant_version;
 731     int i;
 732
 733     if (!prog) {
 734         /* No GLSL program set - nothing to do. */
 735         return;
 736     }
 737     programId = prog->programId;
 738     constant_version = prog->constant_version;
 739
 740     if (useVertexShader) {
 741         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 742
 743         /* Load DirectX 9 float constants/uniforms for vertex shader */
 744         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
 745                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
 746
 747         /* Load DirectX 9 integer constants/uniforms for vertex shader */
 748         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
 749                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
 750
 751         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
 752         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
 753                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
 754
 755         /* Upload the position fixup params */
 756         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
 757         checkGLcall("glUniform4fvARB");
 758     }
 759
 760     if (usePixelShader) {
 761
 762         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 763
 764         /* Load DirectX 9 float constants/uniforms for pixel shader */
 765         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
 766                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
 767
 768         /* Load DirectX 9 integer constants/uniforms for pixel shader */
 769         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
 770                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
 771
 772         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
 773         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
 774                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
 775
 776         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
 777          * It can't be 0 for a valid texbem instruction.
 778          */
 779         for(i = 0; i < MAX_TEXTURES; i++) {
 780             const float *data;
 781
 782             if(prog->bumpenvmat_location[i] == -1) continue;
 783
 784             data = (const float *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVMAT00];
 785             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
 786             checkGLcall("glUniformMatrix2fvARB");
 787
 788             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
 789              * is set too, so we can check that in the needsbumpmat check
 790              */
 791             if(prog->luminancescale_location[i] != -1) {
 792                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLSCALE];
 793                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLOFFSET];
 794
 795                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
 796                 checkGLcall("glUniform1fvARB");
 797                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
 798                 checkGLcall("glUniform1fvARB");
 799             }
 800         }
 801
 802         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
 803             float correction_params[4];
 804
 805             if (context->render_offscreen)
 806             {
 807                 correction_params[0] = 0.0f;
 808                 correction_params[1] = 1.0f;
 809             } else {
 810                 /* position is window relative, not viewport relative */
 811                 correction_params[0] = context->current_rt->currentDesc.Height;
 812                 correction_params[1] = -1.0f;
 813             }
 814             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
 815         }
 816     }
 817
 818     if (priv->next_constant_version == UINT_MAX)
 819     {
 820         TRACE("Max constant version reached, resetting to 0.\n");
 821         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
 822         priv->next_constant_version = 1;
 823     }
 824     else
 825     {
 826         prog->constant_version = priv->next_constant_version++;
 827     }
 828 }
 829
 830 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
 831         unsigned int heap_idx, DWORD new_version)
 832 {
 833     struct constant_entry *entries = heap->entries;
 834     unsigned int *positions = heap->positions;
 835     unsigned int parent_idx;
 836
 837     while (heap_idx > 1)
 838     {
 839         parent_idx = heap_idx >> 1;
 840
 841         if (new_version <= entries[parent_idx].version) break;
 842
 843         entries[heap_idx] = entries[parent_idx];
 844         positions[entries[parent_idx].idx] = heap_idx;
 845         heap_idx = parent_idx;
 846     }
 847
 848     entries[heap_idx].version = new_version;
 849     entries[heap_idx].idx = idx;
 850     positions[idx] = heap_idx;
 851 }
 852
 853 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 854 {
 855     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 856     struct shader_glsl_priv *priv = This->shader_priv;
 857     struct constant_heap *heap = &priv->vconst_heap;
 858     UINT i;
 859
 860     for (i = start; i < count + start; ++i)
 861     {
 862         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
 863             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 864         else
 865             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 866     }
 867 }
 868
 869 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 870 {
 871     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 872     struct shader_glsl_priv *priv = This->shader_priv;
 873     struct constant_heap *heap = &priv->pconst_heap;
 874     UINT i;
 875
 876     for (i = start; i < count + start; ++i)
 877     {
 878         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
 879             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
 880         else
 881             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
 882     }
 883 }
 884
 885 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
 886 {
 887     unsigned int ret = gl_info->limits.glsl_varyings / 4;
 888     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
 889     if(shader_major > 3) return ret;
 890
 891     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
 892     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
 893     return ret;
 894 }
 895
 896 /** Generate the variable & register declarations for the GLSL output target */
 897 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
 898         struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
 899         const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
 900 {
 901     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 902     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
 903     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
 904     const struct wined3d_gl_info *gl_info = context->gl_info;
 905     unsigned int i, extra_constants_needed = 0;
 906     const local_constant *lconst;
 907     DWORD map;
 908
 909     /* There are some minor differences between pixel and vertex shaders */
 910     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 911     char prefix = pshader ? 'P' : 'V';
 912
 913     /* Prototype the subroutines */
 914     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
 915     {
 916         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
 917     }
 918
 919     /* Declare the constants (aka uniforms) */
 920     if (This->baseShader.limits.constant_float > 0) {
 921         unsigned max_constantsF;
 922         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
 923          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
 924          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
 925          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
 926          * a dx9 card, as long as it doesn't also use all the other constants.
 927          *
 928          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
 929          * declare only the amount that we're assured to have.
 930          *
 931          * Thus we run into problems in these two cases:
 932          * 1) The shader really uses more uniforms than supported
 933          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
 934          */
 935         if (pshader)
 936         {
 937             /* No indirect addressing here. */
 938             max_constantsF = gl_info->limits.glsl_ps_float_constants;
 939         }
 940         else
 941         {
 942             if(This->baseShader.reg_maps.usesrelconstF) {
 943                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
 944                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
 945                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
 946                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
 947                  *
 948                  * Writing gl_ClipVertex requires one uniform for each clipplane as well.
 949                  */
 950                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
 951                 if(ctx_priv->cur_vs_args->clip_enabled)
 952                 {
 953                     max_constantsF -= gl_info->limits.clipplanes;
 954                 }
 955                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 956                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
 957                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
 958                  * for now take this into account when calculating the number of available constants
 959                  */
 960                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
 961                 /* Set by driver quirks in directx.c */
 962                 max_constantsF -= gl_info->reserved_glsl_constants;
 963             }
 964             else
 965             {
 966                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
 967             }
 968         }
 969         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
 970         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
 971     }
 972
 973     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
 974      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
 975      */
 976     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
 977         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
 978
 979     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
 980         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
 981
 982     if(!pshader) {
 983         shader_addline(buffer, "uniform vec4 posFixup;\n");
 984         /* Predeclaration; This function is added at link time based on the pixel shader.
 985          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
 986          * that. We know the input to the reorder function at vertex shader compile time, so
 987          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
 988          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
 989          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
 990          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
 991          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
 992          * inout.
 993          */
 994         if (reg_maps->shader_version.major >= 3)
 995         {
 996             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
 997         } else {
 998             shader_addline(buffer, "void order_ps_input();\n");
 999         }
1000     } else {
1001         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1002         {
1003             if (!(map & 1)) continue;
1004
1005             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1006
1007             if (reg_maps->luminanceparams & (1 << i))
1008             {
1009                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1010                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1011                 extra_constants_needed++;
1012             }
1013
1014             extra_constants_needed++;
1015         }
1016
1017         if (ps_args->srgb_correction)
1018         {
1019             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1020                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1021             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1022                     srgb_cmp);
1023         }
1024         if (reg_maps->vpos || reg_maps->usesdsy)
1025         {
1026             if (This->baseShader.limits.constant_float + extra_constants_needed
1027                     + 1 < gl_info->limits.glsl_ps_float_constants)
1028             {
1029                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1030                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1031                 extra_constants_needed++;
1032             } else {
1033                 /* This happens because we do not have proper tracking of the constant registers that are
1034                  * actually used, only the max limit of the shader version
1035                  */
1036                 FIXME("Cannot find a free uniform for vpos correction params\n");
1037                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1038                         context->render_offscreen ? 0.0f : device->render_targets[0]->currentDesc.Height,
1039                         context->render_offscreen ? 1.0f : -1.0f);
1040             }
1041             shader_addline(buffer, "vec4 vpos;\n");
1042         }
1043     }
1044
1045     /* Declare texture samplers */
1046     for (i = 0; i < This->baseShader.limits.sampler; i++) {
1047         if (reg_maps->sampler_type[i])
1048         {
1049             switch (reg_maps->sampler_type[i])
1050             {
1051                 case WINED3DSTT_1D:
1052                     if (pshader && ps_args->shadow & (1 << i))
1053                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1054                     else
1055                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1056                     break;
1057                 case WINED3DSTT_2D:
1058                     if (pshader && ps_args->shadow & (1 << i))
1059                     {
1060                         if (device->stateBlock->textures[i]
1061                                 && IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i])
1062                                 == GL_TEXTURE_RECTANGLE_ARB)
1063                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1064                         else
1065                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1066                     }
1067                     else
1068                     {
1069                         if (device->stateBlock->textures[i]
1070                                 && IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i])
1071                                 == GL_TEXTURE_RECTANGLE_ARB)
1072                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1073                         else
1074                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1075                     }
1076                     break;
1077                 case WINED3DSTT_CUBE:
1078                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1079                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1080                     break;
1081                 case WINED3DSTT_VOLUME:
1082                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1083                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1084                     break;
1085                 default:
1086                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1087                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1088                     break;
1089             }
1090         }
1091     }
1092
1093     /* Declare uniforms for NP2 texcoord fixup:
1094      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1095      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1096      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1097     if (pshader && ps_args->np2_fixup) {
1098
1099         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1100         UINT cur = 0;
1101
1102         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1103          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1104          * samplerNP2Fixup stores texture dimensions and is updated through
1105          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1106
1107         for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1108             if (reg_maps->sampler_type[i]) {
1109                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1110
1111                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1112                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1113                     continue;
1114                 }
1115
1116                 fixup->idx[i] = cur++;
1117             }
1118         }
1119
1120         fixup->num_consts = (cur + 1) >> 1;
1121         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1122     }
1123
1124     /* Declare address variables */
1125     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1126     {
1127         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1128     }
1129
1130     /* Declare texture coordinate temporaries and initialize them */
1131     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1132     {
1133         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1134     }
1135
1136     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1137      * helper function shader that is linked in at link time
1138      */
1139     if (pshader && reg_maps->shader_version.major >= 3)
1140     {
1141         if (use_vs(device->stateBlock))
1142         {
1143             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1144         } else {
1145             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1146              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1147              * pixel shader that reads the fixed function color into the packed input registers.
1148              */
1149             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1150         }
1151     }
1152
1153     /* Declare output register temporaries */
1154     if(This->baseShader.limits.packed_output) {
1155         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1156     }
1157
1158     /* Declare temporary variables */
1159     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1160     {
1161         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1162     }
1163
1164     /* Declare attributes */
1165     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1166     {
1167         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1168         {
1169             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1170         }
1171     }
1172
1173     /* Declare loop registers aLx */
1174     for (i = 0; i < reg_maps->loop_depth; i++) {
1175         shader_addline(buffer, "int aL%u;\n", i);
1176         shader_addline(buffer, "int tmpInt%u;\n", i);
1177     }
1178
1179     /* Temporary variables for matrix operations */
1180     shader_addline(buffer, "vec4 tmp0;\n");
1181     shader_addline(buffer, "vec4 tmp1;\n");
1182
1183     /* Local constants use a different name so they can be loaded once at shader link time
1184      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1185      * float -> string conversion can cause precision loss.
1186      */
1187     if(!This->baseShader.load_local_constsF) {
1188         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1189             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1190         }
1191     }
1192
1193     shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
1194
1195     /* Start the main program */
1196     shader_addline(buffer, "void main() {\n");
1197     if(pshader && reg_maps->vpos) {
1198         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1199          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1200          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1201          * precision troubles when we just substract 0.5.
1202          *
1203          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1204          *
1205          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1206          *
1207          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1208          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1209          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1210          * correctly on drivers that returns integer values.
1211          */
1212         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1213     }
1214 }
1215
1216 /*****************************************************************************
1217  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1218  *
1219  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1220  ****************************************************************************/
1221
1222 /* Prototypes */
1223 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1224         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1225
1226 /** Used for opcode modifiers - They multiply the result by the specified amount */
1227 static const char * const shift_glsl_tab[] = {
1228     "",           /*  0 (none) */
1229     "2.0 * ",     /*  1 (x2)   */
1230     "4.0 * ",     /*  2 (x4)   */
1231     "8.0 * ",     /*  3 (x8)   */
1232     "16.0 * ",    /*  4 (x16)  */
1233     "32.0 * ",    /*  5 (x32)  */
1234     "",           /*  6 (x64)  */
1235     "",           /*  7 (x128) */
1236     "",           /*  8 (d256) */
1237     "",           /*  9 (d128) */
1238     "",           /* 10 (d64)  */
1239     "",           /* 11 (d32)  */
1240     "0.0625 * ",  /* 12 (d16)  */
1241     "0.125 * ",   /* 13 (d8)   */
1242     "0.25 * ",    /* 14 (d4)   */
1243     "0.5 * "      /* 15 (d2)   */
1244 };
1245
1246 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1247 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1248 {
1249     out_str[0] = 0;
1250
1251     switch (src_modifier)
1252     {
1253     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1254     case WINED3DSPSM_DW:
1255     case WINED3DSPSM_NONE:
1256         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1257         break;
1258     case WINED3DSPSM_NEG:
1259         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1260         break;
1261     case WINED3DSPSM_NOT:
1262         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1263         break;
1264     case WINED3DSPSM_BIAS:
1265         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1266         break;
1267     case WINED3DSPSM_BIASNEG:
1268         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1269         break;
1270     case WINED3DSPSM_SIGN:
1271         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1272         break;
1273     case WINED3DSPSM_SIGNNEG:
1274         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1275         break;
1276     case WINED3DSPSM_COMP:
1277         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1278         break;
1279     case WINED3DSPSM_X2:
1280         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1281         break;
1282     case WINED3DSPSM_X2NEG:
1283         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1284         break;
1285     case WINED3DSPSM_ABS:
1286         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1287         break;
1288     case WINED3DSPSM_ABSNEG:
1289         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1290         break;
1291     default:
1292         FIXME("Unhandled modifier %u\n", src_modifier);
1293         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1294     }
1295 }
1296
1297 /** Writes the GLSL variable name that corresponds to the register that the
1298  * DX opcode parameter is trying to access */
1299 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1300         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1301 {
1302     /* oPos, oFog and oPts in D3D */
1303     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
1304
1305     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1306     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1307     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1308
1309     *is_color = FALSE;
1310
1311     switch (reg->type)
1312     {
1313         case WINED3DSPR_TEMP:
1314             sprintf(register_name, "R%u", reg->idx);
1315             break;
1316
1317         case WINED3DSPR_INPUT:
1318             /* vertex shaders */
1319             if (!pshader)
1320             {
1321                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1322                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1323                 sprintf(register_name, "attrib%u", reg->idx);
1324                 break;
1325             }
1326
1327             /* pixel shaders >= 3.0 */
1328             if (This->baseShader.reg_maps.shader_version.major >= 3)
1329             {
1330                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1331                 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1332
1333                 if (reg->rel_addr)
1334                 {
1335                     glsl_src_param_t rel_param;
1336
1337                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1338
1339                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1340                      * operation there */
1341                     if (idx)
1342                     {
1343                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1344                         {
1345                             sprintf(register_name,
1346                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1347                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1348                                     rel_param.param_str, idx);
1349                         }
1350                         else
1351                         {
1352                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1353                         }
1354                     }
1355                     else
1356                     {
1357                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1358                         {
1359                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1360                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1361                                     rel_param.param_str);
1362                         }
1363                         else
1364                         {
1365                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1366                         }
1367                     }
1368                 }
1369                 else
1370                 {
1371                     if (idx == in_count) sprintf(register_name, "gl_Color");
1372                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1373                     else sprintf(register_name, "IN[%u]", idx);
1374                 }
1375             }
1376             else
1377             {
1378                 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1379                 else strcpy(register_name, "gl_SecondaryColor");
1380                 break;
1381             }
1382             break;
1383
1384         case WINED3DSPR_CONST:
1385             {
1386                 const char prefix = pshader ? 'P' : 'V';
1387
1388                 /* Relative addressing */
1389                 if (reg->rel_addr)
1390                 {
1391                     glsl_src_param_t rel_param;
1392                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1393                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1394                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1395                 }
1396                 else
1397                 {
1398                     if (shader_constant_is_local(This, reg->idx))
1399                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1400                     else
1401                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1402                 }
1403             }
1404             break;
1405
1406         case WINED3DSPR_CONSTINT:
1407             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1408             else sprintf(register_name, "VI[%u]", reg->idx);
1409             break;
1410
1411         case WINED3DSPR_CONSTBOOL:
1412             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1413             else sprintf(register_name, "VB[%u]", reg->idx);
1414             break;
1415
1416         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1417             if (pshader) sprintf(register_name, "T%u", reg->idx);
1418             else sprintf(register_name, "A%u", reg->idx);
1419             break;
1420
1421         case WINED3DSPR_LOOP:
1422             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1423             break;
1424
1425         case WINED3DSPR_SAMPLER:
1426             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1427             else sprintf(register_name, "Vsampler%u", reg->idx);
1428             break;
1429
1430         case WINED3DSPR_COLOROUT:
1431             if (reg->idx >= gl_info->limits.buffers)
1432                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1433
1434             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1435             break;
1436
1437         case WINED3DSPR_RASTOUT:
1438             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1439             break;
1440
1441         case WINED3DSPR_DEPTHOUT:
1442             sprintf(register_name, "gl_FragDepth");
1443             break;
1444
1445         case WINED3DSPR_ATTROUT:
1446             if (reg->idx == 0) sprintf(register_name, "gl_FrontColor");
1447             else sprintf(register_name, "gl_FrontSecondaryColor");
1448             break;
1449
1450         case WINED3DSPR_TEXCRDOUT:
1451             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1452             if (This->baseShader.reg_maps.shader_version.major >= 3) sprintf(register_name, "OUT[%u]", reg->idx);
1453             else sprintf(register_name, "gl_TexCoord[%u]", reg->idx);
1454             break;
1455
1456         case WINED3DSPR_MISCTYPE:
1457             if (reg->idx == 0)
1458             {
1459                 /* vPos */
1460                 sprintf(register_name, "vpos");
1461             }
1462             else if (reg->idx == 1)
1463             {
1464                 /* Note that gl_FrontFacing is a bool, while vFace is
1465                  * a float for which the sign determines front/back */
1466                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1467             }
1468             else
1469             {
1470                 FIXME("Unhandled misctype register %d\n", reg->idx);
1471                 sprintf(register_name, "unrecognized_register");
1472             }
1473             break;
1474
1475         case WINED3DSPR_IMMCONST:
1476             switch (reg->immconst_type)
1477             {
1478                 case WINED3D_IMMCONST_FLOAT:
1479                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1480                     break;
1481
1482                 case WINED3D_IMMCONST_FLOAT4:
1483                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1484                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1485                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1486                     break;
1487
1488                 default:
1489                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1490                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1491             }
1492             break;
1493
1494         default:
1495             FIXME("Unhandled register name Type(%d)\n", reg->type);
1496             sprintf(register_name, "unrecognized_register");
1497             break;
1498     }
1499 }
1500
1501 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1502 {
1503     *str++ = '.';
1504     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1505     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1506     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1507     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1508     *str = '\0';
1509 }
1510
1511 /* Get the GLSL write mask for the destination register */
1512 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1513 {
1514     DWORD mask = param->write_mask;
1515
1516     if (shader_is_scalar(&param->reg))
1517     {
1518         mask = WINED3DSP_WRITEMASK_0;
1519         *write_mask = '\0';
1520     }
1521     else
1522     {
1523         shader_glsl_write_mask_to_str(mask, write_mask);
1524     }
1525
1526     return mask;
1527 }
1528
1529 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1530     unsigned int size = 0;
1531
1532     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1533     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1534     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1535     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1536
1537     return size;
1538 }
1539
1540 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1541 {
1542     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1543      * but addressed as "rgba". To fix this we need to swap the register's x
1544      * and z components. */
1545     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1546
1547     *str++ = '.';
1548     /* swizzle bits fields: wwzzyyxx */
1549     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1550     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1551     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1552     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1553     *str = '\0';
1554 }
1555
1556 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1557         BOOL fixup, DWORD mask, char *swizzle_str)
1558 {
1559     if (shader_is_scalar(&param->reg))
1560         *swizzle_str = '\0';
1561     else
1562         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_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 void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1569         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1570 {
1571     BOOL is_color = FALSE;
1572     char swizzle_str[6];
1573
1574     glsl_src->reg_name[0] = '\0';
1575     glsl_src->param_str[0] = '\0';
1576     swizzle_str[0] = '\0';
1577
1578     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1579     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1580     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1581 }
1582
1583 /* From a given parameter token, generate the corresponding GLSL string.
1584  * Also, return the actual register name and swizzle in case the
1585  * caller needs this information as well. */
1586 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1587         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1588 {
1589     BOOL is_color = FALSE;
1590
1591     glsl_dst->mask_str[0] = '\0';
1592     glsl_dst->reg_name[0] = '\0';
1593
1594     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1595     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1596 }
1597
1598 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1599 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1600         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1601 {
1602     glsl_dst_param_t glsl_dst;
1603     DWORD mask;
1604
1605     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1606     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1607
1608     return mask;
1609 }
1610
1611 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1612 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1613 {
1614     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1615 }
1616
1617 /** Process GLSL instruction modifiers */
1618 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1619 {
1620     glsl_dst_param_t dst_param;
1621     DWORD modifiers;
1622
1623     if (!ins->dst_count) return;
1624
1625     modifiers = ins->dst[0].modifiers;
1626     if (!modifiers) return;
1627
1628     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1629
1630     if (modifiers & WINED3DSPDM_SATURATE)
1631     {
1632         /* _SAT means to clamp the value of the register to between 0 and 1 */
1633         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1634                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1635     }
1636
1637     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1638     {
1639         FIXME("_centroid modifier not handled\n");
1640     }
1641
1642     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1643     {
1644         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1645     }
1646 }
1647
1648 static inline const char *shader_get_comp_op(DWORD op)
1649 {
1650     switch (op) {
1651         case COMPARISON_GT: return ">";
1652         case COMPARISON_EQ: return "==";
1653         case COMPARISON_GE: return ">=";
1654         case COMPARISON_LT: return "<";
1655         case COMPARISON_NE: return "!=";
1656         case COMPARISON_LE: return "<=";
1657         default:
1658             FIXME("Unrecognized comparison value: %u\n", op);
1659             return "(\?\?)";
1660     }
1661 }
1662
1663 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1664         DWORD sampler_idx, DWORD flags, glsl_sample_function_t *sample_function)
1665 {
1666     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1667     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1668     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1669             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1670     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1671     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1672     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1673     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1674
1675     /* Note that there's no such thing as a projected cube texture. */
1676     switch(sampler_type) {
1677         case WINED3DSTT_1D:
1678             if (shadow)
1679             {
1680                 if (lod)
1681                 {
1682                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1683                 }
1684                 else if (grad)
1685                 {
1686                     if (gl_info->supported[EXT_GPU_SHADER4])
1687                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1688                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1689                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1690                     else
1691                     {
1692                         FIXME("Unsupported 1D shadow grad function.\n");
1693                         sample_function->name = "unsupported1DGrad";
1694                     }
1695                 }
1696                 else
1697                 {
1698                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1699                 }
1700                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1701             }
1702             else
1703             {
1704                 if (lod)
1705                 {
1706                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1707                 }
1708                 else if (grad)
1709                 {
1710                     if (gl_info->supported[EXT_GPU_SHADER4])
1711                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1712                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1713                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1714                     else
1715                     {
1716                         FIXME("Unsupported 1D grad function.\n");
1717                         sample_function->name = "unsupported1DGrad";
1718                     }
1719                 }
1720                 else
1721                 {
1722                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1723                 }
1724                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1725             }
1726             break;
1727
1728         case WINED3DSTT_2D:
1729             if (shadow)
1730             {
1731                 if (texrect)
1732                 {
1733                     if (lod)
1734                     {
1735                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1736                     }
1737                     else if (grad)
1738                     {
1739                         if (gl_info->supported[EXT_GPU_SHADER4])
1740                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1741                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1742                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1743                         else
1744                         {
1745                             FIXME("Unsupported RECT shadow grad function.\n");
1746                             sample_function->name = "unsupported2DRectGrad";
1747                         }
1748                     }
1749                     else
1750                     {
1751                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1752                     }
1753                 }
1754                 else
1755                 {
1756                     if (lod)
1757                     {
1758                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1759                     }
1760                     else if (grad)
1761                     {
1762                         if (gl_info->supported[EXT_GPU_SHADER4])
1763                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1764                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1765                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1766                         else
1767                         {
1768                             FIXME("Unsupported 2D shadow grad function.\n");
1769                             sample_function->name = "unsupported2DGrad";
1770                         }
1771                     }
1772                     else
1773                     {
1774                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1775                     }
1776                 }
1777                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1778             }
1779             else
1780             {
1781                 if (texrect)
1782                 {
1783                     if (lod)
1784                     {
1785                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1786                     }
1787                     else if (grad)
1788                     {
1789                         if (gl_info->supported[EXT_GPU_SHADER4])
1790                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1791                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1792                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1793                         else
1794                         {
1795                             FIXME("Unsupported RECT grad function.\n");
1796                             sample_function->name = "unsupported2DRectGrad";
1797                         }
1798                     }
1799                     else
1800                     {
1801                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1802                     }
1803                 }
1804                 else
1805                 {
1806                     if (lod)
1807                     {
1808                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1809                     }
1810                     else if (grad)
1811                     {
1812                         if (gl_info->supported[EXT_GPU_SHADER4])
1813                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1814                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1815                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1816                         else
1817                         {
1818                             FIXME("Unsupported 2D grad function.\n");
1819                             sample_function->name = "unsupported2DGrad";
1820                         }
1821                     }
1822                     else
1823                     {
1824                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1825                     }
1826                 }
1827                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1828             }
1829             break;
1830
1831         case WINED3DSTT_CUBE:
1832             if (shadow)
1833             {
1834                 FIXME("Unsupported Cube shadow function.\n ");
1835                 sample_function->name = "unsupportedCubeShadow";
1836                 sample_function->coord_mask = 0;
1837             }
1838             else
1839             {
1840                 if (lod)
1841                 {
1842                     sample_function->name = "textureCubeLod";
1843                 }
1844                 else if (grad)
1845                 {
1846                     if (gl_info->supported[EXT_GPU_SHADER4])
1847                         sample_function->name = "textureCubeGrad";
1848                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1849                         sample_function->name = "textureCubeGradARB";
1850                     else
1851                     {
1852                         FIXME("Unsupported Cube grad function.\n");
1853                         sample_function->name = "unsupportedCubeGrad";
1854                     }
1855                 }
1856                 else
1857                 {
1858                     sample_function->name = "textureCube";
1859                 }
1860                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1861             }
1862             break;
1863
1864         case WINED3DSTT_VOLUME:
1865             if (shadow)
1866             {
1867                 FIXME("Unsupported 3D shadow function.\n ");
1868                 sample_function->name = "unsupported3DShadow";
1869                 sample_function->coord_mask = 0;
1870             }
1871             else
1872             {
1873                 if (lod)
1874                 {
1875                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1876                 }
1877                 else  if (grad)
1878                 {
1879                     if (gl_info->supported[EXT_GPU_SHADER4])
1880                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1881                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1882                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1883                     else
1884                     {
1885                         FIXME("Unsupported 3D grad function.\n");
1886                         sample_function->name = "unsupported3DGrad";
1887                     }
1888                 }
1889                 else
1890                 {
1891                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1892                 }
1893                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1894             }
1895             break;
1896
1897         default:
1898             sample_function->name = "";
1899             sample_function->coord_mask = 0;
1900             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1901             break;
1902     }
1903 }
1904
1905 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1906         BOOL sign_fixup, enum fixup_channel_source channel_source)
1907 {
1908     switch(channel_source)
1909     {
1910         case CHANNEL_SOURCE_ZERO:
1911             strcat(arguments, "0.0");
1912             break;
1913
1914         case CHANNEL_SOURCE_ONE:
1915             strcat(arguments, "1.0");
1916             break;
1917
1918         case CHANNEL_SOURCE_X:
1919             strcat(arguments, reg_name);
1920             strcat(arguments, ".x");
1921             break;
1922
1923         case CHANNEL_SOURCE_Y:
1924             strcat(arguments, reg_name);
1925             strcat(arguments, ".y");
1926             break;
1927
1928         case CHANNEL_SOURCE_Z:
1929             strcat(arguments, reg_name);
1930             strcat(arguments, ".z");
1931             break;
1932
1933         case CHANNEL_SOURCE_W:
1934             strcat(arguments, reg_name);
1935             strcat(arguments, ".w");
1936             break;
1937
1938         default:
1939             FIXME("Unhandled channel source %#x\n", channel_source);
1940             strcat(arguments, "undefined");
1941             break;
1942     }
1943
1944     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1945 }
1946
1947 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1948 {
1949     struct wined3d_shader_dst_param dst;
1950     unsigned int mask_size, remaining;
1951     glsl_dst_param_t dst_param;
1952     char arguments[256];
1953     DWORD mask;
1954
1955     mask = 0;
1956     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1957     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1958     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1959     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1960     mask &= ins->dst[0].write_mask;
1961
1962     if (!mask) return; /* Nothing to do */
1963
1964     if (is_complex_fixup(fixup))
1965     {
1966         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1967         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1968         return;
1969     }
1970
1971     mask_size = shader_glsl_get_write_mask_size(mask);
1972
1973     dst = ins->dst[0];
1974     dst.write_mask = mask;
1975     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1976
1977     arguments[0] = '\0';
1978     remaining = mask_size;
1979     if (mask & WINED3DSP_WRITEMASK_0)
1980     {
1981         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1982         if (--remaining) strcat(arguments, ", ");
1983     }
1984     if (mask & WINED3DSP_WRITEMASK_1)
1985     {
1986         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1987         if (--remaining) strcat(arguments, ", ");
1988     }
1989     if (mask & WINED3DSP_WRITEMASK_2)
1990     {
1991         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1992         if (--remaining) strcat(arguments, ", ");
1993     }
1994     if (mask & WINED3DSP_WRITEMASK_3)
1995     {
1996         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1997         if (--remaining) strcat(arguments, ", ");
1998     }
1999
2000     if (mask_size > 1)
2001     {
2002         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
2003                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
2004     }
2005     else
2006     {
2007         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2008     }
2009 }
2010
2011 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2012         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
2013         const char *dx, const char *dy,
2014         const char *bias, const char *coord_reg_fmt, ...)
2015 {
2016     const char *sampler_base;
2017     char dst_swizzle[6];
2018     struct color_fixup_desc fixup;
2019     BOOL np2_fixup = FALSE;
2020     va_list args;
2021
2022     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2023
2024     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2025     {
2026         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2027         fixup = priv->cur_ps_args->color_fixup[sampler];
2028         sampler_base = "Psampler";
2029
2030         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2031             if(bias) {
2032                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2033             } else {
2034                 np2_fixup = TRUE;
2035             }
2036         }
2037     } else {
2038         sampler_base = "Vsampler";
2039         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2040     }
2041
2042     shader_glsl_append_dst(ins->ctx->buffer, ins);
2043
2044     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2045
2046     va_start(args, coord_reg_fmt);
2047     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2048     va_end(args);
2049
2050     if(bias) {
2051         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2052     } else {
2053         if (np2_fixup) {
2054             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2055             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2056
2057             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2058                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2059         } else if(dx && dy) {
2060             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2061         } else {
2062             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2063         }
2064     }
2065
2066     if(!is_identity_fixup(fixup)) {
2067         shader_glsl_color_correction(ins, fixup);
2068     }
2069 }
2070
2071 /*****************************************************************************
2072  * Begin processing individual instruction opcodes
2073  ****************************************************************************/
2074
2075 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2076 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2077 {
2078     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2079     glsl_src_param_t src0_param;
2080     glsl_src_param_t src1_param;
2081     DWORD write_mask;
2082     char op;
2083
2084     /* Determine the GLSL operator to use based on the opcode */
2085     switch (ins->handler_idx)
2086     {
2087         case WINED3DSIH_MUL: op = '*'; break;
2088         case WINED3DSIH_ADD: op = '+'; break;
2089         case WINED3DSIH_SUB: op = '-'; break;
2090         default:
2091             op = ' ';
2092             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2093             break;
2094     }
2095
2096     write_mask = shader_glsl_append_dst(buffer, ins);
2097     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2098     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2099     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2100 }
2101
2102 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2103 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2104 {
2105     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2106     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2107     glsl_src_param_t src0_param;
2108     DWORD write_mask;
2109
2110     write_mask = shader_glsl_append_dst(buffer, ins);
2111     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2112
2113     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2114      * shader versions WINED3DSIO_MOVA is used for this. */
2115     if (ins->ctx->reg_maps->shader_version.major == 1
2116             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2117             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2118     {
2119         /* This is a simple floor() */
2120         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2121         if (mask_size > 1) {
2122             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2123         } else {
2124             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2125         }
2126     }
2127     else if(ins->handler_idx == WINED3DSIH_MOVA)
2128     {
2129         /* We need to *round* to the nearest int here. */
2130         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2131
2132         if (gl_info->supported[EXT_GPU_SHADER4])
2133         {
2134             if (mask_size > 1)
2135                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2136             else
2137                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2138         }
2139         else
2140         {
2141             if (mask_size > 1)
2142                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2143                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2144             else
2145                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2146                         src0_param.param_str, src0_param.param_str);
2147         }
2148     }
2149     else
2150     {
2151         shader_addline(buffer, "%s);\n", src0_param.param_str);
2152     }
2153 }
2154
2155 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2156 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2157 {
2158     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2159     glsl_src_param_t src0_param;
2160     glsl_src_param_t src1_param;
2161     DWORD dst_write_mask, src_write_mask;
2162     unsigned int dst_size = 0;
2163
2164     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2165     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2166
2167     /* dp3 works on vec3, dp4 on vec4 */
2168     if (ins->handler_idx == WINED3DSIH_DP4)
2169     {
2170         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2171     } else {
2172         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2173     }
2174
2175     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2176     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2177
2178     if (dst_size > 1) {
2179         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2180     } else {
2181         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2182     }
2183 }
2184
2185 /* Note that this instruction has some restrictions. The destination write mask
2186  * can't contain the w component, and the source swizzles have to be .xyzw */
2187 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2188 {
2189     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2190     glsl_src_param_t src0_param;
2191     glsl_src_param_t src1_param;
2192     char dst_mask[6];
2193
2194     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2195     shader_glsl_append_dst(ins->ctx->buffer, ins);
2196     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2197     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2198     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2199 }
2200
2201 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2202  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2203  * GLSL uses the value as-is. */
2204 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2205 {
2206     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2207     glsl_src_param_t src0_param;
2208     glsl_src_param_t src1_param;
2209     DWORD dst_write_mask;
2210     unsigned int dst_size;
2211
2212     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2213     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2214
2215     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2216     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2217
2218     if (dst_size > 1) {
2219         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2220     } else {
2221         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
2222     }
2223 }
2224
2225 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2226  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2227  * GLSL uses the value as-is. */
2228 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2229 {
2230     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2231     glsl_src_param_t src0_param;
2232     DWORD dst_write_mask;
2233     unsigned int dst_size;
2234
2235     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2236     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2237
2238     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2239
2240     if (dst_size > 1)
2241     {
2242         shader_addline(buffer, "vec%d(%s == 0.0 ? -FLT_MAX : log2(abs(%s))));\n",
2243                 dst_size, src0_param.param_str, src0_param.param_str);
2244     }
2245     else
2246     {
2247         shader_addline(buffer, "%s == 0.0 ? -FLT_MAX : log2(abs(%s)));\n",
2248                 src0_param.param_str, src0_param.param_str);
2249     }
2250 }
2251
2252 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2253 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2254 {
2255     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2256     glsl_src_param_t src_param;
2257     const char *instruction;
2258     DWORD write_mask;
2259     unsigned i;
2260
2261     /* Determine the GLSL function to use based on the opcode */
2262     /* TODO: Possibly make this a table for faster lookups */
2263     switch (ins->handler_idx)
2264     {
2265         case WINED3DSIH_MIN: instruction = "min"; break;
2266         case WINED3DSIH_MAX: instruction = "max"; break;
2267         case WINED3DSIH_ABS: instruction = "abs"; break;
2268         case WINED3DSIH_FRC: instruction = "fract"; break;
2269         case WINED3DSIH_EXP: instruction = "exp2"; break;
2270         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2271         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2272         default: instruction = "";
2273             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2274             break;
2275     }
2276
2277     write_mask = shader_glsl_append_dst(buffer, ins);
2278
2279     shader_addline(buffer, "%s(", instruction);
2280
2281     if (ins->src_count)
2282     {
2283         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2284         shader_addline(buffer, "%s", src_param.param_str);
2285         for (i = 1; i < ins->src_count; ++i)
2286         {
2287             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2288             shader_addline(buffer, ", %s", src_param.param_str);
2289         }
2290     }
2291
2292     shader_addline(buffer, "));\n");
2293 }
2294
2295 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2296 {
2297     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2298     glsl_src_param_t src_param;
2299     unsigned int mask_size;
2300     DWORD write_mask;
2301     char dst_mask[6];
2302
2303     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2304     mask_size = shader_glsl_get_write_mask_size(write_mask);
2305     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2306
2307     shader_addline(buffer, "tmp0.x = length(%s);\n", src_param.param_str);
2308     shader_glsl_append_dst(buffer, ins);
2309     if (mask_size > 1)
2310     {
2311         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s / tmp0.x));\n",
2312                 mask_size, src_param.param_str);
2313     }
2314     else
2315     {
2316         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s / tmp0.x));\n",
2317                 src_param.param_str);
2318     }
2319 }
2320
2321 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2322  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2323  *   dst.x = 2^(floor(src))
2324  *   dst.y = src - floor(src)
2325  *   dst.z = 2^src   (partial precision is allowed, but optional)
2326  *   dst.w = 1.0;
2327  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2328  *   dst = 2^src;    (partial precision is allowed, but optional)
2329  */
2330 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2331 {
2332     glsl_src_param_t src_param;
2333
2334     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2335
2336     if (ins->ctx->reg_maps->shader_version.major < 2)
2337     {
2338         char dst_mask[6];
2339
2340         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2341         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2342         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2343         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2344
2345         shader_glsl_append_dst(ins->ctx->buffer, ins);
2346         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2347         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2348     } else {
2349         DWORD write_mask;
2350         unsigned int mask_size;
2351
2352         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2353         mask_size = shader_glsl_get_write_mask_size(write_mask);
2354
2355         if (mask_size > 1) {
2356             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2357         } else {
2358             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2359         }
2360     }
2361 }
2362
2363 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2364 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2365 {
2366     glsl_src_param_t src_param;
2367     DWORD write_mask;
2368     unsigned int mask_size;
2369
2370     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2371     mask_size = shader_glsl_get_write_mask_size(write_mask);
2372     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2373
2374     if (mask_size > 1)
2375     {
2376         shader_addline(ins->ctx->buffer, "vec%d(%s == 0.0 ? FLT_MAX : 1.0 / %s));\n",
2377                 mask_size, src_param.param_str, src_param.param_str);
2378     }
2379     else
2380     {
2381         shader_addline(ins->ctx->buffer, "%s == 0.0 ? FLT_MAX : 1.0 / %s);\n",
2382                 src_param.param_str, src_param.param_str);
2383     }
2384 }
2385
2386 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2387 {
2388     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2389     glsl_src_param_t src_param;
2390     DWORD write_mask;
2391     unsigned int mask_size;
2392
2393     write_mask = shader_glsl_append_dst(buffer, ins);
2394     mask_size = shader_glsl_get_write_mask_size(write_mask);
2395
2396     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2397
2398     if (mask_size > 1)
2399     {
2400         shader_addline(buffer, "vec%d(%s == 0.0 ? FLT_MAX : inversesqrt(abs(%s))));\n",
2401                 mask_size, src_param.param_str, src_param.param_str);
2402     }
2403     else
2404     {
2405         shader_addline(buffer, "%s == 0.0 ? FLT_MAX : inversesqrt(abs(%s)));\n",
2406                 src_param.param_str, src_param.param_str);
2407     }
2408 }
2409
2410 /** Process signed comparison opcodes in GLSL. */
2411 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2412 {
2413     glsl_src_param_t src0_param;
2414     glsl_src_param_t src1_param;
2415     DWORD write_mask;
2416     unsigned int mask_size;
2417
2418     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2419     mask_size = shader_glsl_get_write_mask_size(write_mask);
2420     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2421     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2422
2423     if (mask_size > 1) {
2424         const char *compare;
2425
2426         switch(ins->handler_idx)
2427         {
2428             case WINED3DSIH_SLT: compare = "lessThan"; break;
2429             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2430             default: compare = "";
2431                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2432         }
2433
2434         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2435                 src0_param.param_str, src1_param.param_str);
2436     } else {
2437         switch(ins->handler_idx)
2438         {
2439             case WINED3DSIH_SLT:
2440                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2441                  * to return 0.0 but step returns 1.0 because step is not < x
2442                  * An alternative is a bvec compare padded with an unused second component.
2443                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2444                  * issue. Playing with not() is not possible either because not() does not accept
2445                  * a scalar.
2446                  */
2447                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2448                         src0_param.param_str, src1_param.param_str);
2449                 break;
2450             case WINED3DSIH_SGE:
2451                 /* Here we can use the step() function and safe a conditional */
2452                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2453                 break;
2454             default:
2455                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2456         }
2457
2458     }
2459 }
2460
2461 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2462 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2463 {
2464     glsl_src_param_t src0_param;
2465     glsl_src_param_t src1_param;
2466     glsl_src_param_t src2_param;
2467     DWORD write_mask, cmp_channel = 0;
2468     unsigned int i, j;
2469     char mask_char[6];
2470     BOOL temp_destination = FALSE;
2471
2472     if (shader_is_scalar(&ins->src[0].reg))
2473     {
2474         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2475
2476         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2477         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2478         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2479
2480         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2481                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2482     } else {
2483         DWORD dst_mask = ins->dst[0].write_mask;
2484         struct wined3d_shader_dst_param dst = ins->dst[0];
2485
2486         /* Cycle through all source0 channels */
2487         for (i=0; i<4; i++) {
2488             write_mask = 0;
2489             /* Find the destination channels which use the current source0 channel */
2490             for (j=0; j<4; j++) {
2491                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2492                 {
2493                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2494                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2495                 }
2496             }
2497             dst.write_mask = dst_mask & write_mask;
2498
2499             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2500             * The first lines may overwrite source parameters of the following lines.
2501             * Deal with that by using a temporary destination register if needed
2502             */
2503             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2504                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2505                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2506                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2507                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2508                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2509             {
2510                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2511                 if (!write_mask) continue;
2512                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2513                 temp_destination = TRUE;
2514             } else {
2515                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2516                 if (!write_mask) continue;
2517             }
2518
2519             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2520             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2521             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2522
2523             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2524                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2525         }
2526
2527         if(temp_destination) {
2528             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2529             shader_glsl_append_dst(ins->ctx->buffer, ins);
2530             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2531         }
2532     }
2533
2534 }
2535
2536 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2537 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2538  * the compare is done per component of src0. */
2539 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2540 {
2541     struct wined3d_shader_dst_param dst;
2542     glsl_src_param_t src0_param;
2543     glsl_src_param_t src1_param;
2544     glsl_src_param_t src2_param;
2545     DWORD write_mask, cmp_channel = 0;
2546     unsigned int i, j;
2547     DWORD dst_mask;
2548     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2549             ins->ctx->reg_maps->shader_version.minor);
2550
2551     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2552     {
2553         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2554         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2555         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2556         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2557
2558         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2559         if (ins->coissue)
2560         {
2561             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2562         } else {
2563             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2564                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2565         }
2566         return;
2567     }
2568     /* Cycle through all source0 channels */
2569     dst_mask = ins->dst[0].write_mask;
2570     dst = ins->dst[0];
2571     for (i=0; i<4; i++) {
2572         write_mask = 0;
2573         /* Find the destination channels which use the current source0 channel */
2574         for (j=0; j<4; j++) {
2575             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2576             {
2577                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2578                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2579             }
2580         }
2581
2582         dst.write_mask = dst_mask & write_mask;
2583         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2584         if (!write_mask) continue;
2585
2586         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2587         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2588         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2589
2590         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2591                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2592     }
2593 }
2594
2595 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2596 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2597 {
2598     glsl_src_param_t src0_param;
2599     glsl_src_param_t src1_param;
2600     glsl_src_param_t src2_param;
2601     DWORD write_mask;
2602
2603     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2604     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2605     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2606     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2607     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2608             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2609 }
2610
2611 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2612    Vertex shaders to GLSL codes */
2613 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2614 {
2615     int i;
2616     int nComponents = 0;
2617     struct wined3d_shader_dst_param tmp_dst = {{0}};
2618     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2619     struct wined3d_shader_instruction tmp_ins;
2620
2621     memset(&tmp_ins, 0, sizeof(tmp_ins));
2622
2623     /* Set constants for the temporary argument */
2624     tmp_ins.ctx = ins->ctx;
2625     tmp_ins.dst_count = 1;
2626     tmp_ins.dst = &tmp_dst;
2627     tmp_ins.src_count = 2;
2628     tmp_ins.src = tmp_src;
2629
2630     switch(ins->handler_idx)
2631     {
2632         case WINED3DSIH_M4x4:
2633             nComponents = 4;
2634             tmp_ins.handler_idx = WINED3DSIH_DP4;
2635             break;
2636         case WINED3DSIH_M4x3:
2637             nComponents = 3;
2638             tmp_ins.handler_idx = WINED3DSIH_DP4;
2639             break;
2640         case WINED3DSIH_M3x4:
2641             nComponents = 4;
2642             tmp_ins.handler_idx = WINED3DSIH_DP3;
2643             break;
2644         case WINED3DSIH_M3x3:
2645             nComponents = 3;
2646             tmp_ins.handler_idx = WINED3DSIH_DP3;
2647             break;
2648         case WINED3DSIH_M3x2:
2649             nComponents = 2;
2650             tmp_ins.handler_idx = WINED3DSIH_DP3;
2651             break;
2652         default:
2653             break;
2654     }
2655
2656     tmp_dst = ins->dst[0];
2657     tmp_src[0] = ins->src[0];
2658     tmp_src[1] = ins->src[1];
2659     for (i = 0; i < nComponents; ++i)
2660     {
2661         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2662         shader_glsl_dot(&tmp_ins);
2663         ++tmp_src[1].reg.idx;
2664     }
2665 }
2666
2667 /**
2668     The LRP instruction performs a component-wise linear interpolation
2669     between the second and third operands using the first operand as the
2670     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2671     This is equivalent to mix(src2, src1, src0);
2672 */
2673 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2674 {
2675     glsl_src_param_t src0_param;
2676     glsl_src_param_t src1_param;
2677     glsl_src_param_t src2_param;
2678     DWORD write_mask;
2679
2680     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2681
2682     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2683     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2684     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2685
2686     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2687             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2688 }
2689
2690 /** Process the WINED3DSIO_LIT instruction in GLSL:
2691  * dst.x = dst.w = 1.0
2692  * dst.y = (src0.x > 0) ? src0.x
2693  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2694  *                                        where src.w is clamped at +- 128
2695  */
2696 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2697 {
2698     glsl_src_param_t src0_param;
2699     glsl_src_param_t src1_param;
2700     glsl_src_param_t src3_param;
2701     char dst_mask[6];
2702
2703     shader_glsl_append_dst(ins->ctx->buffer, ins);
2704     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2705
2706     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2707     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2708     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2709
2710     /* The sdk specifies the instruction like this
2711      * dst.x = 1.0;
2712      * if(src.x > 0.0) dst.y = src.x
2713      * else dst.y = 0.0.
2714      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2715      * else dst.z = 0.0;
2716      * dst.w = 1.0;
2717      *
2718      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2719      * dst.x = 1.0                                  ... No further explanation needed
2720      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2721      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2722      * dst.w = 1.0.                                 ... Nothing fancy.
2723      *
2724      * So we still have one conditional in there. So do this:
2725      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2726      *
2727      * 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),
2728      * 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.
2729      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2730      */
2731     shader_addline(ins->ctx->buffer,
2732             "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",
2733             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2734 }
2735
2736 /** Process the WINED3DSIO_DST instruction in GLSL:
2737  * dst.x = 1.0
2738  * dst.y = src0.x * src0.y
2739  * dst.z = src0.z
2740  * dst.w = src1.w
2741  */
2742 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2743 {
2744     glsl_src_param_t src0y_param;
2745     glsl_src_param_t src0z_param;
2746     glsl_src_param_t src1y_param;
2747     glsl_src_param_t src1w_param;
2748     char dst_mask[6];
2749
2750     shader_glsl_append_dst(ins->ctx->buffer, ins);
2751     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2752
2753     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2754     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2755     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2756     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2757
2758     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2759             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2760 }
2761
2762 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2763  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2764  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2765  *
2766  * dst.x = cos(src0.?)
2767  * dst.y = sin(src0.?)
2768  * dst.z = dst.z
2769  * dst.w = dst.w
2770  */
2771 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2772 {
2773     glsl_src_param_t src0_param;
2774     DWORD write_mask;
2775
2776     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2777     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2778
2779     switch (write_mask) {
2780         case WINED3DSP_WRITEMASK_0:
2781             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2782             break;
2783
2784         case WINED3DSP_WRITEMASK_1:
2785             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2786             break;
2787
2788         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2789             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2790             break;
2791
2792         default:
2793             ERR("Write mask should be .x, .y or .xy\n");
2794             break;
2795     }
2796 }
2797
2798 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2799  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2800  * generate invalid code
2801  */
2802 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2803 {
2804     glsl_src_param_t src0_param;
2805     DWORD write_mask;
2806
2807     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2808     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2809
2810     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2811 }
2812
2813 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2814  * Start a for() loop where src1.y is the initial value of aL,
2815  *  increment aL by src1.z for a total of src1.x iterations.
2816  *  Need to use a temporary variable for this operation.
2817  */
2818 /* FIXME: I don't think nested loops will work correctly this way. */
2819 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2820 {
2821     glsl_src_param_t src1_param;
2822     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2823     const DWORD *control_values = NULL;
2824     const local_constant *constant;
2825
2826     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2827
2828     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2829      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2830      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2831      * addressing.
2832      */
2833     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2834     {
2835         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2836             if (constant->idx == ins->src[1].reg.idx)
2837             {
2838                 control_values = constant->value;
2839                 break;
2840             }
2841         }
2842     }
2843
2844     if (control_values)
2845     {
2846         struct wined3d_shader_loop_control loop_control;
2847         loop_control.count = control_values[0];
2848         loop_control.start = control_values[1];
2849         loop_control.step = (int)control_values[2];
2850
2851         if (loop_control.step > 0)
2852         {
2853             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2854                     shader->baseShader.cur_loop_depth, loop_control.start,
2855                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2856                     shader->baseShader.cur_loop_depth, loop_control.step);
2857         }
2858         else if (loop_control.step < 0)
2859         {
2860             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2861                     shader->baseShader.cur_loop_depth, loop_control.start,
2862                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2863                     shader->baseShader.cur_loop_depth, loop_control.step);
2864         }
2865         else
2866         {
2867             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2868                     shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2869                     shader->baseShader.cur_loop_depth, loop_control.count,
2870                     shader->baseShader.cur_loop_depth);
2871         }
2872     } else {
2873         shader_addline(ins->ctx->buffer,
2874                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2875                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2876                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2877                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2878     }
2879
2880     shader->baseShader.cur_loop_depth++;
2881     shader->baseShader.cur_loop_regno++;
2882 }
2883
2884 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2885 {
2886     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2887
2888     shader_addline(ins->ctx->buffer, "}\n");
2889
2890     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2891     {
2892         shader->baseShader.cur_loop_depth--;
2893         shader->baseShader.cur_loop_regno--;
2894     }
2895
2896     if (ins->handler_idx == WINED3DSIH_ENDREP)
2897     {
2898         shader->baseShader.cur_loop_depth--;
2899     }
2900 }
2901
2902 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2903 {
2904     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2905     glsl_src_param_t src0_param;
2906     const DWORD *control_values = NULL;
2907     const local_constant *constant;
2908
2909     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2910     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2911     {
2912         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2913         {
2914             if (constant->idx == ins->src[0].reg.idx)
2915             {
2916                 control_values = constant->value;
2917                 break;
2918             }
2919         }
2920     }
2921
2922     if(control_values) {
2923         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2924                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2925                        control_values[0], shader->baseShader.cur_loop_depth);
2926     } else {
2927         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2928         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2929                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2930                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2931     }
2932     shader->baseShader.cur_loop_depth++;
2933 }
2934
2935 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2936 {
2937     glsl_src_param_t src0_param;
2938
2939     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2940     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2941 }
2942
2943 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2944 {
2945     glsl_src_param_t src0_param;
2946     glsl_src_param_t src1_param;
2947
2948     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2949     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2950
2951     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2952             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2953 }
2954
2955 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2956 {
2957     shader_addline(ins->ctx->buffer, "} else {\n");
2958 }
2959
2960 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2961 {
2962     shader_addline(ins->ctx->buffer, "break;\n");
2963 }
2964
2965 /* FIXME: According to MSDN the compare is done per component. */
2966 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2967 {
2968     glsl_src_param_t src0_param;
2969     glsl_src_param_t src1_param;
2970
2971     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2972     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2973
2974     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2975             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2976 }
2977
2978 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2979 {
2980     shader_addline(ins->ctx->buffer, "}\n");
2981     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2982 }
2983
2984 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2985 {
2986     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2987 }
2988
2989 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2990 {
2991     glsl_src_param_t src1_param;
2992
2993     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2994     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
2995 }
2996
2997 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
2998 {
2999     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3000      * function only suppresses the unhandled instruction warning
3001      */
3002 }
3003
3004 /*********************************************
3005  * Pixel Shader Specific Code begins here
3006  ********************************************/
3007 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3008 {
3009     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3010     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3011     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3012             ins->ctx->reg_maps->shader_version.minor);
3013     glsl_sample_function_t sample_function;
3014     DWORD sample_flags = 0;
3015     DWORD sampler_idx;
3016     DWORD mask = 0, swizzle;
3017
3018     /* 1.0-1.4: Use destination register as sampler source.
3019      * 2.0+: Use provided sampler source. */
3020     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3021     else sampler_idx = ins->src[1].reg.idx;
3022
3023     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3024     {
3025         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3026         WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3027
3028         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3029         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3030             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3031             switch (flags & ~WINED3DTTFF_PROJECTED) {
3032                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3033                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3034                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3035                 case WINED3DTTFF_COUNT4:
3036                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3037             }
3038         }
3039     }
3040     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3041     {
3042         DWORD src_mod = ins->src[0].modifiers;
3043
3044         if (src_mod == WINED3DSPSM_DZ) {
3045             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3046             mask = WINED3DSP_WRITEMASK_2;
3047         } else if (src_mod == WINED3DSPSM_DW) {
3048             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3049             mask = WINED3DSP_WRITEMASK_3;
3050         }
3051     } else {
3052         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3053         {
3054             /* ps 2.0 texldp instruction always divides by the fourth component. */
3055             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3056             mask = WINED3DSP_WRITEMASK_3;
3057         }
3058     }
3059
3060     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3061        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3062         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3063     }
3064
3065     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3066     mask |= sample_function.coord_mask;
3067
3068     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3069     else swizzle = ins->src[1].swizzle;
3070
3071     /* 1.0-1.3: Use destination register as coordinate source.
3072        1.4+: Use provided coordinate source register. */
3073     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3074     {
3075         char coord_mask[6];
3076         shader_glsl_write_mask_to_str(mask, coord_mask);
3077         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3078                 "T%u%s", sampler_idx, coord_mask);
3079     } else {
3080         glsl_src_param_t coord_param;
3081         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3082         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3083         {
3084             glsl_src_param_t bias;
3085             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3086             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3087                     "%s", coord_param.param_str);
3088         } else {
3089             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3090                     "%s", coord_param.param_str);
3091         }
3092     }
3093 }
3094
3095 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3096 {
3097     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3098     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3099     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3100     glsl_sample_function_t sample_function;
3101     glsl_src_param_t coord_param, dx_param, dy_param;
3102     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3103     DWORD sampler_idx;
3104     DWORD swizzle = ins->src[1].swizzle;
3105
3106     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3107     {
3108         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3109         return shader_glsl_tex(ins);
3110     }
3111
3112     sampler_idx = ins->src[1].reg.idx;
3113     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3114        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3115         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3116     }
3117
3118     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3119     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3120     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3121     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3122
3123     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3124                                 "%s", coord_param.param_str);
3125 }
3126
3127 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3128 {
3129     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3130     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3131     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3132     glsl_sample_function_t sample_function;
3133     glsl_src_param_t coord_param, lod_param;
3134     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3135     DWORD sampler_idx;
3136     DWORD swizzle = ins->src[1].swizzle;
3137
3138     sampler_idx = ins->src[1].reg.idx;
3139     if(deviceImpl->stateBlock->textures[sampler_idx] &&
3140        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3141         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3142     }
3143     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3144     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3145
3146     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3147
3148     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3149             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3150     {
3151         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3152          * However, they seem to work just fine in fragment shaders as well. */
3153         WARN("Using %s in fragment shader.\n", sample_function.name);
3154     }
3155     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3156             "%s", coord_param.param_str);
3157 }
3158
3159 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3160 {
3161     /* FIXME: Make this work for more than just 2D textures */
3162     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3163     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3164
3165     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3166     {
3167         char dst_mask[6];
3168
3169         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3170         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3171                 ins->dst[0].reg.idx, dst_mask);
3172     } else {
3173         DWORD reg = ins->src[0].reg.idx;
3174         DWORD src_mod = ins->src[0].modifiers;
3175         char dst_swizzle[6];
3176
3177         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3178
3179         if (src_mod == WINED3DSPSM_DZ) {
3180             glsl_src_param_t div_param;
3181             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3182             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3183
3184             if (mask_size > 1) {
3185                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3186             } else {
3187                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3188             }
3189         } else if (src_mod == WINED3DSPSM_DW) {
3190             glsl_src_param_t div_param;
3191             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3192             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3193
3194             if (mask_size > 1) {
3195                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3196             } else {
3197                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3198             }
3199         } else {
3200             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3201         }
3202     }
3203 }
3204
3205 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3206  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3207  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3208 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3209 {
3210     glsl_src_param_t src0_param;
3211     glsl_sample_function_t sample_function;
3212     DWORD sampler_idx = ins->dst[0].reg.idx;
3213     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3214     UINT mask_size;
3215
3216     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3217
3218     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3219      * scalar, and projected sampling would require 4.
3220      *
3221      * It is a dependent read - not valid with conditional NP2 textures
3222      */
3223     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3224     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3225
3226     switch(mask_size)
3227     {
3228         case 1:
3229             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3230                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3231             break;
3232
3233         case 2:
3234             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3235                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3236             break;
3237
3238         case 3:
3239             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3240                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3241             break;
3242
3243         default:
3244             FIXME("Unexpected mask size %u\n", mask_size);
3245             break;
3246     }
3247 }
3248
3249 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3250  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3251 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3252 {
3253     glsl_src_param_t src0_param;
3254     DWORD dstreg = ins->dst[0].reg.idx;
3255     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3256     DWORD dst_mask;
3257     unsigned int mask_size;
3258
3259     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3260     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3261     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3262
3263     if (mask_size > 1) {
3264         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3265     } else {
3266         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3267     }
3268 }
3269
3270 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3271  * Calculate the depth as dst.x / dst.y   */
3272 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3273 {
3274     glsl_dst_param_t dst_param;
3275
3276     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3277
3278     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3279      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3280      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3281      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3282      * >= 1.0 or < 0.0
3283      */
3284     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3285             dst_param.reg_name, dst_param.reg_name);
3286 }
3287
3288 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3289  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3290  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3291  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3292  */
3293 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3294 {
3295     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3296     DWORD dstreg = ins->dst[0].reg.idx;
3297     glsl_src_param_t src0_param;
3298
3299     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3300
3301     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3302     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3303 }
3304
3305 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3306  * Calculate the 1st of a 2-row matrix multiplication. */
3307 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3308 {
3309     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3310     DWORD reg = ins->dst[0].reg.idx;
3311     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3312     glsl_src_param_t src0_param;
3313
3314     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3315     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3316 }
3317
3318 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3319  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3320 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3321 {
3322     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3323     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3324     DWORD reg = ins->dst[0].reg.idx;
3325     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3326     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3327     glsl_src_param_t src0_param;
3328
3329     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3330     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3331     current_state->texcoord_w[current_state->current_row++] = reg;
3332 }
3333
3334 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3335 {
3336     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3337     DWORD reg = ins->dst[0].reg.idx;
3338     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3339     glsl_src_param_t src0_param;
3340     glsl_sample_function_t sample_function;
3341
3342     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3343     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3344
3345     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3346
3347     /* Sample the texture using the calculated coordinates */
3348     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3349 }
3350
3351 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3352  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3353 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3354 {
3355     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3356     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3357     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3358     glsl_src_param_t src0_param;
3359     DWORD reg = ins->dst[0].reg.idx;
3360     glsl_sample_function_t sample_function;
3361
3362     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3363     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3364
3365     /* Dependent read, not valid with conditional NP2 */
3366     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3367
3368     /* Sample the texture using the calculated coordinates */
3369     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3370
3371     current_state->current_row = 0;
3372 }
3373
3374 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3375  * Perform the 3rd row of a 3x3 matrix multiply */
3376 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3377 {
3378     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3379     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3380     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3381     glsl_src_param_t src0_param;
3382     char dst_mask[6];
3383     DWORD reg = ins->dst[0].reg.idx;
3384
3385     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3386
3387     shader_glsl_append_dst(ins->ctx->buffer, ins);
3388     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3389     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3390
3391     current_state->current_row = 0;
3392 }
3393
3394 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3395  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3396 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3397 {
3398     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3399     DWORD reg = ins->dst[0].reg.idx;
3400     glsl_src_param_t src0_param;
3401     glsl_src_param_t src1_param;
3402     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3403     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3404     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3405     glsl_sample_function_t sample_function;
3406
3407     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3408     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3409
3410     /* Perform the last matrix multiply operation */
3411     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3412     /* Reflection calculation */
3413     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3414
3415     /* Dependent read, not valid with conditional NP2 */
3416     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3417
3418     /* Sample the texture */
3419     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3420
3421     current_state->current_row = 0;
3422 }
3423
3424 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3425  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3426 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3427 {
3428     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3429     DWORD reg = ins->dst[0].reg.idx;
3430     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3431     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3432     glsl_src_param_t src0_param;
3433     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3434     glsl_sample_function_t sample_function;
3435
3436     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3437
3438     /* Perform the last matrix multiply operation */
3439     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3440
3441     /* Construct the eye-ray vector from w coordinates */
3442     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3443             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3444     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3445
3446     /* Dependent read, not valid with conditional NP2 */
3447     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3448
3449     /* Sample the texture using the calculated coordinates */
3450     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3451
3452     current_state->current_row = 0;
3453 }
3454
3455 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3456  * Apply a fake bump map transform.
3457  * texbem is pshader <= 1.3 only, this saves a few version checks
3458  */
3459 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3460 {
3461     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3462     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3463     glsl_sample_function_t sample_function;
3464     glsl_src_param_t coord_param;
3465     DWORD sampler_idx;
3466     DWORD mask;
3467     DWORD flags;
3468     char coord_mask[6];
3469
3470     sampler_idx = ins->dst[0].reg.idx;
3471     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3472
3473     /* Dependent read, not valid with conditional NP2 */
3474     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3475     mask = sample_function.coord_mask;
3476
3477     shader_glsl_write_mask_to_str(mask, coord_mask);
3478
3479     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3480          * so we can't let the GL handle this.
3481          */
3482     if (flags & WINED3DTTFF_PROJECTED) {
3483         DWORD div_mask=0;
3484         char coord_div_mask[3];
3485         switch (flags & ~WINED3DTTFF_PROJECTED) {
3486             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3487             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3488             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3489             case WINED3DTTFF_COUNT4:
3490             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3491         }
3492         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3493         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3494     }
3495
3496     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3497
3498     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3499             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3500             coord_param.param_str, coord_mask);
3501
3502     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3503     {
3504         glsl_src_param_t luminance_param;
3505         glsl_dst_param_t dst_param;
3506
3507         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3508         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3509
3510         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3511                 dst_param.reg_name, dst_param.mask_str,
3512                 luminance_param.param_str, sampler_idx, sampler_idx);
3513     }
3514 }
3515
3516 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3517 {
3518     glsl_src_param_t src0_param, src1_param;
3519     DWORD sampler_idx = ins->dst[0].reg.idx;
3520
3521     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3522     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3523
3524     shader_glsl_append_dst(ins->ctx->buffer, ins);
3525     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3526             src0_param.param_str, sampler_idx, src1_param.param_str);
3527 }
3528
3529 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3530  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3531 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3532 {
3533     glsl_src_param_t src0_param;
3534     DWORD sampler_idx = ins->dst[0].reg.idx;
3535     glsl_sample_function_t sample_function;
3536
3537     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3538
3539     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3540     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3541             "%s.wx", src0_param.reg_name);
3542 }
3543
3544 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3545  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3546 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3547 {
3548     glsl_src_param_t src0_param;
3549     DWORD sampler_idx = ins->dst[0].reg.idx;
3550     glsl_sample_function_t sample_function;
3551
3552     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3553
3554     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3555     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3556             "%s.yz", src0_param.reg_name);
3557 }
3558
3559 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3560  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3561 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3562 {
3563     glsl_src_param_t src0_param;
3564     DWORD sampler_idx = ins->dst[0].reg.idx;
3565     glsl_sample_function_t sample_function;
3566
3567     /* Dependent read, not valid with conditional NP2 */
3568     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3569     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3570
3571     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3572             "%s", src0_param.param_str);
3573 }
3574
3575 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3576  * If any of the first 3 components are < 0, discard this pixel */
3577 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3578 {
3579     glsl_dst_param_t dst_param;
3580
3581     /* The argument is a destination parameter, and no writemasks are allowed */
3582     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3583     if (ins->ctx->reg_maps->shader_version.major >= 2)
3584     {
3585         /* 2.0 shaders compare all 4 components in texkill */
3586         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3587     } else {
3588         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3589          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3590          * 4 components are defined, only the first 3 are used
3591          */
3592         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3593     }
3594 }
3595
3596 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3597  * dst = dot2(src0, src1) + src2 */
3598 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3599 {
3600     glsl_src_param_t src0_param;
3601     glsl_src_param_t src1_param;
3602     glsl_src_param_t src2_param;
3603     DWORD write_mask;
3604     unsigned int mask_size;
3605
3606     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3607     mask_size = shader_glsl_get_write_mask_size(write_mask);
3608
3609     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3610     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3611     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3612
3613     if (mask_size > 1) {
3614         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3615                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3616     } else {
3617         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3618                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3619     }
3620 }
3621
3622 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3623         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3624         enum vertexprocessing_mode vertexprocessing)
3625 {
3626     unsigned int i;
3627     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3628     WORD map = reg_maps->input_registers;
3629
3630     for (i = 0; map; map >>= 1, ++i)
3631     {
3632         const char *semantic_name;
3633         UINT semantic_idx;
3634         char reg_mask[6];
3635
3636         /* Unused */
3637         if (!(map & 1)) continue;
3638
3639         semantic_name = input_signature[i].semantic_name;
3640         semantic_idx = input_signature[i].semantic_idx;
3641         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3642
3643         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3644         {
3645             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3646                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3647                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3648             else
3649                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3650                         This->input_reg_map[i], reg_mask, reg_mask);
3651         }
3652         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3653         {
3654             if (semantic_idx == 0)
3655                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3656                         This->input_reg_map[i], reg_mask, reg_mask);
3657             else if (semantic_idx == 1)
3658                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3659                         This->input_reg_map[i], reg_mask, reg_mask);
3660             else
3661                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3662                         This->input_reg_map[i], reg_mask, reg_mask);
3663         }
3664         else
3665         {
3666             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3667                     This->input_reg_map[i], reg_mask, reg_mask);
3668         }
3669     }
3670 }
3671
3672 /*********************************************
3673  * Vertex Shader Specific Code begins here
3674  ********************************************/
3675
3676 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3677     glsl_program_key_t key;
3678
3679     key.vshader = entry->vshader;
3680     key.pshader = entry->pshader;
3681     key.vs_args = entry->vs_args;
3682     key.ps_args = entry->ps_args;
3683
3684     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3685     {
3686         ERR("Failed to insert program entry.\n");
3687     }
3688 }
3689
3690 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3691         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3692         struct ps_compile_args *ps_args) {
3693     struct wine_rb_entry *entry;
3694     glsl_program_key_t key;
3695
3696     key.vshader = vshader;
3697     key.pshader = pshader;
3698     key.vs_args = *vs_args;
3699     key.ps_args = *ps_args;
3700
3701     entry = wine_rb_get(&priv->program_lookup, &key);
3702     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3703 }
3704
3705 /* GL locking is done by the caller */
3706 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3707         struct glsl_shader_prog_link *entry)
3708 {
3709     glsl_program_key_t key;
3710
3711     key.vshader = entry->vshader;
3712     key.pshader = entry->pshader;
3713     key.vs_args = entry->vs_args;
3714     key.ps_args = entry->ps_args;
3715     wine_rb_remove(&priv->program_lookup, &key);
3716
3717     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3718     if (entry->vshader) list_remove(&entry->vshader_entry);
3719     if (entry->pshader) list_remove(&entry->pshader_entry);
3720     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3721     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3722     HeapFree(GetProcessHeap(), 0, entry);
3723 }
3724
3725 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3726         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3727         const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3728 {
3729     unsigned int i, j;
3730     const char *semantic_name_in, *semantic_name_out;
3731     UINT semantic_idx_in, semantic_idx_out;
3732     DWORD *set;
3733     DWORD in_idx;
3734     unsigned int in_count = vec4_varyings(3, gl_info);
3735     char reg_mask[6], reg_mask_out[6];
3736     char destination[50];
3737     WORD input_map, output_map;
3738
3739     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3740
3741     if (!output_signature)
3742     {
3743         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3744         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3745         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3746     }
3747
3748     input_map = reg_maps_in->input_registers;
3749     for (i = 0; input_map; input_map >>= 1, ++i)
3750     {
3751         if (!(input_map & 1)) continue;
3752
3753         in_idx = map[i];
3754         if (in_idx >= (in_count + 2)) {
3755             FIXME("More input varyings declared than supported, expect issues\n");
3756             continue;
3757         }
3758         else if (map[i] == ~0U)
3759         {
3760             /* Declared, but not read register */
3761             continue;
3762         }
3763
3764         if (in_idx == in_count) {
3765             sprintf(destination, "gl_FrontColor");
3766         } else if (in_idx == in_count + 1) {
3767             sprintf(destination, "gl_FrontSecondaryColor");
3768         } else {
3769             sprintf(destination, "IN[%u]", in_idx);
3770         }
3771
3772         semantic_name_in = input_signature[i].semantic_name;
3773         semantic_idx_in = input_signature[i].semantic_idx;
3774         set[map[i]] = input_signature[i].mask;
3775         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3776
3777         if (!output_signature)
3778         {
3779             if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_COLOR))
3780             {
3781                 if (semantic_idx_in == 0)
3782                     shader_addline(buffer, "%s%s = front_color%s;\n",
3783                             destination, reg_mask, reg_mask);
3784                 else if (semantic_idx_in == 1)
3785                     shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3786                             destination, reg_mask, reg_mask);
3787                 else
3788                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3789                             destination, reg_mask, reg_mask);
3790             }
3791             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_TEXCOORD))
3792             {
3793                 if (semantic_idx_in < 8)
3794                 {
3795                     shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3796                             destination, reg_mask, semantic_idx_in, reg_mask);
3797                 }
3798                 else
3799                 {
3800                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3801                             destination, reg_mask, reg_mask);
3802                 }
3803             }
3804             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_FOG))
3805             {
3806                 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3807                         destination, reg_mask, reg_mask);
3808             }
3809             else
3810             {
3811                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3812                         destination, reg_mask, reg_mask);
3813             }
3814         } else {
3815             BOOL found = FALSE;
3816
3817             output_map = reg_maps_out->output_registers;
3818             for (j = 0; output_map; output_map >>= 1, ++j)
3819             {
3820                 if (!(output_map & 1)) continue;
3821
3822                 semantic_name_out = output_signature[j].semantic_name;
3823                 semantic_idx_out = output_signature[j].semantic_idx;
3824                 shader_glsl_write_mask_to_str(output_signature[j].mask, reg_mask_out);
3825
3826                 if (semantic_idx_in == semantic_idx_out
3827                         && !strcmp(semantic_name_in, semantic_name_out))
3828                 {
3829                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3830                             destination, reg_mask, j, reg_mask);
3831                     found = TRUE;
3832                 }
3833             }
3834             if(!found) {
3835                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3836                                destination, reg_mask, reg_mask);
3837             }
3838         }
3839     }
3840
3841     /* This is solely to make the compiler / linker happy and avoid warning about undefined
3842      * varyings. It shouldn't result in any real code executed on the GPU, since all read
3843      * input varyings are assigned above, if the optimizer works properly.
3844      */
3845     for(i = 0; i < in_count + 2; i++) {
3846         if (set[i] && set[i] != WINED3DSP_WRITEMASK_ALL)
3847         {
3848             unsigned int size = 0;
3849             memset(reg_mask, 0, sizeof(reg_mask));
3850             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3851                 reg_mask[size] = 'x';
3852                 size++;
3853             }
3854             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3855                 reg_mask[size] = 'y';
3856                 size++;
3857             }
3858             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3859                 reg_mask[size] = 'z';
3860                 size++;
3861             }
3862             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3863                 reg_mask[size] = 'w';
3864                 size++;
3865             }
3866
3867             if (i == in_count) {
3868                 sprintf(destination, "gl_FrontColor");
3869             } else if (i == in_count + 1) {
3870                 sprintf(destination, "gl_FrontSecondaryColor");
3871             } else {
3872                 sprintf(destination, "IN[%u]", i);
3873             }
3874
3875             if (size == 1) {
3876                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3877             } else {
3878                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3879             }
3880         }
3881     }
3882
3883     HeapFree(GetProcessHeap(), 0, set);
3884 }
3885
3886 /* GL locking is done by the caller */
3887 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3888         IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3889 {
3890     GLhandleARB ret = 0;
3891     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3892     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3893     IWineD3DDeviceImpl *device;
3894     DWORD vs_major = vs->baseShader.reg_maps.shader_version.major;
3895     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3896     unsigned int i;
3897     const char *semantic_name;
3898     UINT semantic_idx;
3899     char reg_mask[6];
3900     const struct wined3d_shader_signature_element *output_signature;
3901
3902     shader_buffer_clear(buffer);
3903
3904     shader_addline(buffer, "#version 120\n");
3905
3906     if(vs_major < 3 && ps_major < 3) {
3907         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3908          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3909          */
3910         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3911         if ((gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W)
3912                 && ps_major == 0 && vs_major > 0 && !device->frag_pipe->ffp_proj_control)
3913         {
3914             shader_addline(buffer, "void order_ps_input() {\n");
3915             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3916                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3917                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3918                     shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3919                 }
3920             }
3921             shader_addline(buffer, "}\n");
3922         } else {
3923             shader_addline(buffer, "void order_ps_input() { /* do nothing */ }\n");
3924         }
3925     } else if(ps_major < 3 && vs_major >= 3) {
3926         WORD map = vs->baseShader.reg_maps.output_registers;
3927
3928         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3929         output_signature = vs->baseShader.output_signature;
3930
3931         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3932         for (i = 0; map; map >>= 1, ++i)
3933         {
3934             DWORD write_mask;
3935
3936             if (!(map & 1)) continue;
3937
3938             semantic_name = output_signature[i].semantic_name;
3939             semantic_idx = output_signature[i].semantic_idx;
3940             write_mask = output_signature[i].mask;
3941             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3942
3943             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3944             {
3945                 if (semantic_idx == 0)
3946                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3947                 else if (semantic_idx == 1)
3948                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3949             }
3950             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3951             {
3952                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3953             }
3954             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3955             {
3956                 if (semantic_idx < 8)
3957                 {
3958                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3959                         write_mask |= WINED3DSP_WRITEMASK_3;
3960
3961                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3962                             semantic_idx, reg_mask, i, reg_mask);
3963                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3964                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3965                 }
3966             }
3967             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3968             {
3969                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3970             }
3971             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3972             {
3973                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3974             }
3975         }
3976         shader_addline(buffer, "}\n");
3977
3978     } else if(ps_major >= 3 && vs_major >= 3) {
3979         WORD map = vs->baseShader.reg_maps.output_registers;
3980
3981         output_signature = vs->baseShader.output_signature;
3982
3983         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3984         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3985         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3986
3987         /* First, sort out position and point size. Those are not passed to the pixel shader */
3988         for (i = 0; map; map >>= 1, ++i)
3989         {
3990             if (!(map & 1)) continue;
3991
3992             semantic_name = output_signature[i].semantic_name;
3993             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3994
3995             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3996             {
3997                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3998             }
3999             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
4000             {
4001                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
4002             }
4003         }
4004
4005         /* Then, fix the pixel shader input */
4006         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
4007                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
4008
4009         shader_addline(buffer, "}\n");
4010     } else if(ps_major >= 3 && vs_major < 3) {
4011         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
4012         shader_addline(buffer, "void order_ps_input() {\n");
4013         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
4014          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
4015          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
4016          */
4017         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
4018                 &ps->baseShader.reg_maps, NULL, NULL);
4019         shader_addline(buffer, "}\n");
4020     } else {
4021         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
4022     }
4023
4024     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4025     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
4026     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer->buffer, NULL));
4027     checkGLcall("glShaderSourceARB(ret, 1, &buffer->buffer, NULL)");
4028     GL_EXTCALL(glCompileShaderARB(ret));
4029     checkGLcall("glCompileShaderARB(ret)");
4030
4031     return ret;
4032 }
4033
4034 /* GL locking is done by the caller */
4035 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
4036         GLhandleARB programId, char prefix)
4037 {
4038     const local_constant *lconst;
4039     GLint tmp_loc;
4040     const float *value;
4041     char glsl_name[8];
4042
4043     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
4044         value = (const float *)lconst->value;
4045         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
4046         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4047         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
4048     }
4049     checkGLcall("Hardcoding local constants");
4050 }
4051
4052 /* GL locking is done by the caller */
4053 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
4054         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
4055         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
4056 {
4057     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4058     const struct wined3d_gl_info *gl_info = context->gl_info;
4059     CONST DWORD *function = This->baseShader.function;
4060     struct shader_glsl_ctx_priv priv_ctx;
4061
4062     /* Create the hw GLSL shader object and assign it as the shader->prgId */
4063     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4064
4065     memset(&priv_ctx, 0, sizeof(priv_ctx));
4066     priv_ctx.cur_ps_args = args;
4067     priv_ctx.cur_np2fixup_info = np2fixup_info;
4068
4069     shader_addline(buffer, "#version 120\n");
4070
4071     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
4072     {
4073         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
4074     }
4075     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
4076     {
4077         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
4078          * drivers write a warning if we don't do so
4079          */
4080         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
4081     }
4082     if (gl_info->supported[EXT_GPU_SHADER4])
4083     {
4084         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4085     }
4086
4087     /* Base Declarations */
4088     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4089
4090     /* Pack 3.0 inputs */
4091     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4092     {
4093         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
4094                 This->baseShader.input_signature, reg_maps, args->vp_mode);
4095     }
4096
4097     /* Base Shader Body */
4098     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4099
4100     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4101     if (reg_maps->shader_version.major < 2)
4102     {
4103         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4104         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4105     }
4106
4107     if (args->srgb_correction)
4108     {
4109         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4110         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4111         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4112         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4113         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4114         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4115     }
4116     /* Pixel shader < 3.0 do not replace the fog stage.
4117      * This implements linear fog computation and blending.
4118      * TODO: non linear fog
4119      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4120      * -1/(e-s) and e/(e-s) respectively.
4121      */
4122     if (reg_maps->shader_version.major < 3)
4123     {
4124         switch(args->fog) {
4125             case FOG_OFF: break;
4126             case FOG_LINEAR:
4127                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4128                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4129                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4130                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4131                 break;
4132             case FOG_EXP:
4133                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4134                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4135                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4136                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4137                 break;
4138             case FOG_EXP2:
4139                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4140                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4141                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4142                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4143                 break;
4144         }
4145     }
4146
4147     shader_addline(buffer, "}\n");
4148
4149     TRACE("Compiling shader object %u\n", shader_obj);
4150     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4151     GL_EXTCALL(glCompileShaderARB(shader_obj));
4152     print_glsl_info_log(gl_info, shader_obj);
4153
4154     /* Store the shader object */
4155     return shader_obj;
4156 }
4157
4158 /* GL locking is done by the caller */
4159 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4160         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4161         const struct vs_compile_args *args)
4162 {
4163     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4164     const struct wined3d_gl_info *gl_info = context->gl_info;
4165     CONST DWORD *function = This->baseShader.function;
4166     struct shader_glsl_ctx_priv priv_ctx;
4167
4168     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4169     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4170
4171     shader_addline(buffer, "#version 120\n");
4172
4173     if (gl_info->supported[EXT_GPU_SHADER4])
4174     {
4175         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4176     }
4177
4178     memset(&priv_ctx, 0, sizeof(priv_ctx));
4179     priv_ctx.cur_vs_args = args;
4180
4181     /* Base Declarations */
4182     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4183
4184     /* Base Shader Body */
4185     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4186
4187     /* Unpack 3.0 outputs */
4188     if (reg_maps->shader_version.major >= 3) shader_addline(buffer, "order_ps_input(OUT);\n");
4189     else shader_addline(buffer, "order_ps_input();\n");
4190
4191     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4192      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4193      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4194      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4195      */
4196     if(args->fog_src == VS_FOG_Z) {
4197         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4198     } else if (!reg_maps->fog) {
4199         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4200     }
4201
4202     /* Write the final position.
4203      *
4204      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4205      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4206      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4207      * contains 1.0 to allow a mad.
4208      */
4209     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4210     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4211     if(args->clip_enabled) {
4212         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4213     }
4214
4215     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4216      *
4217      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4218      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4219      * which is the same as z = z * 2 - w.
4220      */
4221     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4222
4223     shader_addline(buffer, "}\n");
4224
4225     TRACE("Compiling shader object %u\n", shader_obj);
4226     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4227     GL_EXTCALL(glCompileShaderARB(shader_obj));
4228     print_glsl_info_log(gl_info, shader_obj);
4229
4230     return shader_obj;
4231 }
4232
4233 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4234         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4235         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4236 {
4237     UINT i;
4238     DWORD new_size;
4239     struct glsl_ps_compiled_shader *new_array;
4240     struct glsl_pshader_private    *shader_data;
4241     struct ps_np2fixup_info        *np2fixup = NULL;
4242     GLhandleARB ret;
4243
4244     if (!shader->baseShader.backend_data)
4245     {
4246         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4247         if (!shader->baseShader.backend_data)
4248         {
4249             ERR("Failed to allocate backend data.\n");
4250             return 0;
4251         }
4252     }
4253     shader_data = shader->baseShader.backend_data;
4254
4255     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4256      * so a linear search is more performant than a hashmap or a binary search
4257      * (cache coherency etc)
4258      */
4259     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4260         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4261             if(args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4262             return shader_data->gl_shaders[i].prgId;
4263         }
4264     }
4265
4266     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4267     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4268         if (shader_data->num_gl_shaders)
4269         {
4270             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4271             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4272                                     new_size * sizeof(*shader_data->gl_shaders));
4273         } else {
4274             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4275             new_size = 1;
4276         }
4277
4278         if(!new_array) {
4279             ERR("Out of memory\n");
4280             return 0;
4281         }
4282         shader_data->gl_shaders = new_array;
4283         shader_data->shader_array_size = new_size;
4284     }
4285
4286     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4287
4288     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4289     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4290
4291     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4292             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4293
4294     shader_buffer_clear(buffer);
4295     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4296     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4297     *np2fixup_info = np2fixup;
4298
4299     return ret;
4300 }
4301
4302 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4303                                  const DWORD use_map) {
4304     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4305     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4306     return stored->fog_src == new->fog_src;
4307 }
4308
4309 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4310         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4311         const struct vs_compile_args *args)
4312 {
4313     UINT i;
4314     DWORD new_size;
4315     struct glsl_vs_compiled_shader *new_array;
4316     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4317     struct glsl_vshader_private *shader_data;
4318     GLhandleARB ret;
4319
4320     if (!shader->baseShader.backend_data)
4321     {
4322         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4323         if (!shader->baseShader.backend_data)
4324         {
4325             ERR("Failed to allocate backend data.\n");
4326             return 0;
4327         }
4328     }
4329     shader_data = shader->baseShader.backend_data;
4330
4331     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4332      * so a linear search is more performant than a hashmap or a binary search
4333      * (cache coherency etc)
4334      */
4335     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4336         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4337             return shader_data->gl_shaders[i].prgId;
4338         }
4339     }
4340
4341     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4342
4343     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4344         if (shader_data->num_gl_shaders)
4345         {
4346             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4347             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4348                                     new_size * sizeof(*shader_data->gl_shaders));
4349         } else {
4350             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4351             new_size = 1;
4352         }
4353
4354         if(!new_array) {
4355             ERR("Out of memory\n");
4356             return 0;
4357         }
4358         shader_data->gl_shaders = new_array;
4359         shader_data->shader_array_size = new_size;
4360     }
4361
4362     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4363
4364     shader_buffer_clear(buffer);
4365     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4366     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4367
4368     return ret;
4369 }
4370
4371 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4372  * It sets the programId on the current StateBlock (because it should be called
4373  * inside of the DrawPrimitive() part of the render loop).
4374  *
4375  * If a program for the given combination does not exist, create one, and store
4376  * the program in the hash table.  If it creates a program, it will link the
4377  * given objects, too.
4378  */
4379
4380 /* GL locking is done by the caller */
4381 static void set_glsl_shader_program(const struct wined3d_context *context,
4382         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4383 {
4384     IWineD3DVertexShader *vshader = use_vs ? device->stateBlock->vertexShader : NULL;
4385     IWineD3DPixelShader *pshader = use_ps ? device->stateBlock->pixelShader : NULL;
4386     const struct wined3d_gl_info *gl_info = context->gl_info;
4387     struct shader_glsl_priv *priv = device->shader_priv;
4388     struct glsl_shader_prog_link *entry    = NULL;
4389     GLhandleARB programId                  = 0;
4390     GLhandleARB reorder_shader_id          = 0;
4391     unsigned int i;
4392     char glsl_name[8];
4393     struct ps_compile_args ps_compile_args;
4394     struct vs_compile_args vs_compile_args;
4395
4396     if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4397     if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4398
4399     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4400     if (entry) {
4401         priv->glsl_program = entry;
4402         return;
4403     }
4404
4405     /* If we get to this point, then no matching program exists, so we create one */
4406     programId = GL_EXTCALL(glCreateProgramObjectARB());
4407     TRACE("Created new GLSL shader program %u\n", programId);
4408
4409     /* Create the entry */
4410     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4411     entry->programId = programId;
4412     entry->vshader = vshader;
4413     entry->pshader = pshader;
4414     entry->vs_args = vs_compile_args;
4415     entry->ps_args = ps_compile_args;
4416     entry->constant_version = 0;
4417     entry->np2Fixup_info = NULL;
4418     /* Add the hash table entry */
4419     add_glsl_program_entry(priv, entry);
4420
4421     /* Set the current program */
4422     priv->glsl_program = entry;
4423
4424     /* Attach GLSL vshader */
4425     if (vshader)
4426     {
4427         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4428                 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4429         WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4430         char tmp_name[10];
4431
4432         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4433         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4434         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4435         checkGLcall("glAttachObjectARB");
4436         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4437          * is destroyed
4438          */
4439         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4440
4441         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4442         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4443         checkGLcall("glAttachObjectARB");
4444
4445         /* Bind vertex attributes to a corresponding index number to match
4446          * the same index numbers as ARB_vertex_programs (makes loading
4447          * vertex attributes simpler).  With this method, we can use the
4448          * exact same code to load the attributes later for both ARB and
4449          * GLSL shaders.
4450          *
4451          * We have to do this here because we need to know the Program ID
4452          * in order to make the bindings work, and it has to be done prior
4453          * to linking the GLSL program. */
4454         for (i = 0; map; map >>= 1, ++i)
4455         {
4456             if (!(map & 1)) continue;
4457
4458             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4459             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4460         }
4461         checkGLcall("glBindAttribLocationARB");
4462
4463         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4464     }
4465
4466     /* Attach GLSL pshader */
4467     if (pshader)
4468     {
4469         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4470                 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4471         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4472         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4473         checkGLcall("glAttachObjectARB");
4474
4475         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4476     }
4477
4478     /* Link the program */
4479     TRACE("Linking GLSL shader program %u\n", programId);
4480     GL_EXTCALL(glLinkProgramARB(programId));
4481     shader_glsl_validate_link(gl_info, programId);
4482
4483     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4484             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4485     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4486     {
4487         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4488         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4489     }
4490     for (i = 0; i < MAX_CONST_I; ++i)
4491     {
4492         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4493         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4494     }
4495     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4496             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4497     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4498     {
4499         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4500         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4501     }
4502     for (i = 0; i < MAX_CONST_I; ++i)
4503     {
4504         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4505         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4506     }
4507
4508     if(pshader) {
4509         char name[32];
4510
4511         for(i = 0; i < MAX_TEXTURES; i++) {
4512             sprintf(name, "bumpenvmat%u", i);
4513             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4514             sprintf(name, "luminancescale%u", i);
4515             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4516             sprintf(name, "luminanceoffset%u", i);
4517             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4518         }
4519
4520         if (ps_compile_args.np2_fixup) {
4521             if (entry->np2Fixup_info) {
4522                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4523             } else {
4524                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4525             }
4526         }
4527     }
4528
4529     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4530     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4531     checkGLcall("Find glsl program uniform locations");
4532
4533     if (pshader
4534             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4535             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4536     {
4537         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4538         entry->vertex_color_clamp = GL_FALSE;
4539     } else {
4540         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4541     }
4542
4543     /* Set the shader to allow uniform loading on it */
4544     GL_EXTCALL(glUseProgramObjectARB(programId));
4545     checkGLcall("glUseProgramObjectARB(programId)");
4546
4547     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4548      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4549      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4550      * vertex shader with fixed function pixel processing is used we make sure that the card
4551      * supports enough samplers to allow the max number of vertex samplers with all possible
4552      * fixed function fragment processing setups. So once the program is linked these samplers
4553      * won't change.
4554      */
4555     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4556     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4557
4558     /* If the local constants do not have to be loaded with the environment constants,
4559      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4560      * later
4561      */
4562     if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4563     {
4564         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4565     }
4566     if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4567     {
4568         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4569     }
4570 }
4571
4572 /* GL locking is done by the caller */
4573 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4574 {
4575     GLhandleARB program_id;
4576     GLhandleARB vshader_id, pshader_id;
4577     const char *blt_pshader;
4578
4579     static const char *blt_vshader[] =
4580     {
4581         "#version 120\n"
4582         "void main(void)\n"
4583         "{\n"
4584         "    gl_Position = gl_Vertex;\n"
4585         "    gl_FrontColor = vec4(1.0);\n"
4586         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4587         "}\n"
4588     };
4589
4590     static const char *blt_pshaders_full[tex_type_count] =
4591     {
4592         /* tex_1d */
4593         NULL,
4594         /* tex_2d */
4595         "#version 120\n"
4596         "uniform sampler2D sampler;\n"
4597         "void main(void)\n"
4598         "{\n"
4599         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4600         "}\n",
4601         /* tex_3d */
4602         NULL,
4603         /* tex_cube */
4604         "#version 120\n"
4605         "uniform samplerCube sampler;\n"
4606         "void main(void)\n"
4607         "{\n"
4608         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4609         "}\n",
4610         /* tex_rect */
4611         "#version 120\n"
4612         "#extension GL_ARB_texture_rectangle : enable\n"
4613         "uniform sampler2DRect sampler;\n"
4614         "void main(void)\n"
4615         "{\n"
4616         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4617         "}\n",
4618     };
4619
4620     static const char *blt_pshaders_masked[tex_type_count] =
4621     {
4622         /* tex_1d */
4623         NULL,
4624         /* tex_2d */
4625         "#version 120\n"
4626         "uniform sampler2D sampler;\n"
4627         "uniform vec4 mask;\n"
4628         "void main(void)\n"
4629         "{\n"
4630         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4631         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4632         "}\n",
4633         /* tex_3d */
4634         NULL,
4635         /* tex_cube */
4636         "#version 120\n"
4637         "uniform samplerCube sampler;\n"
4638         "uniform vec4 mask;\n"
4639         "void main(void)\n"
4640         "{\n"
4641         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4642         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4643         "}\n",
4644         /* tex_rect */
4645         "#version 120\n"
4646         "#extension GL_ARB_texture_rectangle : enable\n"
4647         "uniform sampler2DRect sampler;\n"
4648         "uniform vec4 mask;\n"
4649         "void main(void)\n"
4650         "{\n"
4651         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4652         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4653         "}\n",
4654     };
4655
4656     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4657     if (!blt_pshader)
4658     {
4659         FIXME("tex_type %#x not supported\n", tex_type);
4660         tex_type = tex_2d;
4661     }
4662
4663     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4664     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
4665     GL_EXTCALL(glCompileShaderARB(vshader_id));
4666
4667     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4668     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshader, NULL));
4669     GL_EXTCALL(glCompileShaderARB(pshader_id));
4670
4671     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4672     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4673     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4674     GL_EXTCALL(glLinkProgramARB(program_id));
4675
4676     shader_glsl_validate_link(gl_info, program_id);
4677
4678     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4679      * is destroyed
4680      */
4681     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4682     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4683     return program_id;
4684 }
4685
4686 /* GL locking is done by the caller */
4687 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4688 {
4689     const struct wined3d_gl_info *gl_info = context->gl_info;
4690     IWineD3DDeviceImpl *device = context->swapchain->device;
4691     struct shader_glsl_priv *priv = device->shader_priv;
4692     GLhandleARB program_id = 0;
4693     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4694
4695     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4696
4697     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4698     else priv->glsl_program = NULL;
4699
4700     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4701
4702     if (old_vertex_color_clamp != current_vertex_color_clamp)
4703     {
4704         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4705         {
4706             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4707             checkGLcall("glClampColorARB");
4708         }
4709         else
4710         {
4711             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4712         }
4713     }
4714
4715     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4716     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4717     GL_EXTCALL(glUseProgramObjectARB(program_id));
4718     checkGLcall("glUseProgramObjectARB");
4719
4720     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4721      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4722      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4723     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4724     {
4725         shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4726     }
4727 }
4728
4729 /* GL locking is done by the caller */
4730 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface,
4731         enum tex_types tex_type, const SIZE *ds_mask_size)
4732 {
4733     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4734     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4735     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4736     struct shader_glsl_priv *priv = This->shader_priv;
4737     GLhandleARB *blt_program;
4738     GLint loc;
4739
4740     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4741     if (!*blt_program)
4742     {
4743         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4744         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4745         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4746         GL_EXTCALL(glUniform1iARB(loc, 0));
4747     }
4748     else
4749     {
4750         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4751     }
4752
4753     if (masked)
4754     {
4755         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4756         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4757     }
4758 }
4759
4760 /* GL locking is done by the caller */
4761 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4762     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4763     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4764     struct shader_glsl_priv *priv = This->shader_priv;
4765     GLhandleARB program_id;
4766
4767     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4768     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4769
4770     GL_EXTCALL(glUseProgramObjectARB(program_id));
4771     checkGLcall("glUseProgramObjectARB");
4772 }
4773
4774 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4775     const struct list *linked_programs;
4776     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4777     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4778     struct shader_glsl_priv *priv = device->shader_priv;
4779     const struct wined3d_gl_info *gl_info;
4780     struct wined3d_context *context;
4781
4782     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4783      * can be called from IWineD3DBaseShader::Release
4784      */
4785     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4786
4787     if(pshader) {
4788         struct glsl_pshader_private *shader_data;
4789         shader_data = This->baseShader.backend_data;
4790         if(!shader_data || shader_data->num_gl_shaders == 0)
4791         {
4792             HeapFree(GetProcessHeap(), 0, shader_data);
4793             This->baseShader.backend_data = NULL;
4794             return;
4795         }
4796
4797         context = context_acquire(device, NULL);
4798         gl_info = context->gl_info;
4799
4800         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4801         {
4802             ENTER_GL();
4803             shader_glsl_select(context, FALSE, FALSE);
4804             LEAVE_GL();
4805         }
4806     } else {
4807         struct glsl_vshader_private *shader_data;
4808         shader_data = This->baseShader.backend_data;
4809         if(!shader_data || shader_data->num_gl_shaders == 0)
4810         {
4811             HeapFree(GetProcessHeap(), 0, shader_data);
4812             This->baseShader.backend_data = NULL;
4813             return;
4814         }
4815
4816         context = context_acquire(device, NULL);
4817         gl_info = context->gl_info;
4818
4819         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4820         {
4821             ENTER_GL();
4822             shader_glsl_select(context, FALSE, FALSE);
4823             LEAVE_GL();
4824         }
4825     }
4826
4827     linked_programs = &This->baseShader.linked_programs;
4828
4829     TRACE("Deleting linked programs\n");
4830     if (linked_programs->next) {
4831         struct glsl_shader_prog_link *entry, *entry2;
4832
4833         ENTER_GL();
4834         if(pshader) {
4835             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4836                 delete_glsl_program_entry(priv, gl_info, entry);
4837             }
4838         } else {
4839             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4840                 delete_glsl_program_entry(priv, gl_info, entry);
4841             }
4842         }
4843         LEAVE_GL();
4844     }
4845
4846     if(pshader) {
4847         UINT i;
4848         struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
4849
4850         ENTER_GL();
4851         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4852             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4853             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4854             checkGLcall("glDeleteObjectARB");
4855         }
4856         LEAVE_GL();
4857         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4858     }
4859     else
4860     {
4861         UINT i;
4862         struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
4863
4864         ENTER_GL();
4865         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4866             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4867             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4868             checkGLcall("glDeleteObjectARB");
4869         }
4870         LEAVE_GL();
4871         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4872     }
4873
4874     HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
4875     This->baseShader.backend_data = NULL;
4876
4877     context_release(context);
4878 }
4879
4880 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4881 {
4882     const glsl_program_key_t *k = key;
4883     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4884             const struct glsl_shader_prog_link, program_lookup_entry);
4885     int cmp;
4886
4887     if (k->vshader > prog->vshader) return 1;
4888     else if (k->vshader < prog->vshader) return -1;
4889
4890     if (k->pshader > prog->pshader) return 1;
4891     else if (k->pshader < prog->pshader) return -1;
4892
4893     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4894     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4895
4896     return 0;
4897 }
4898
4899 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4900 {
4901     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4902     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4903
4904     if (!mem)
4905     {
4906         ERR("Failed to allocate memory\n");
4907         return FALSE;
4908     }
4909
4910     heap->entries = mem;
4911     heap->entries[1].version = 0;
4912     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4913     heap->size = 1;
4914
4915     return TRUE;
4916 }
4917
4918 static void constant_heap_free(struct constant_heap *heap)
4919 {
4920     HeapFree(GetProcessHeap(), 0, heap->entries);
4921 }
4922
4923 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4924 {
4925     wined3d_rb_alloc,
4926     wined3d_rb_realloc,
4927     wined3d_rb_free,
4928     glsl_program_key_compare,
4929 };
4930
4931 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4932     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4933     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4934     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4935     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4936             gl_info->limits.glsl_ps_float_constants)) + 1;
4937
4938     if (!shader_buffer_init(&priv->shader_buffer))
4939     {
4940         ERR("Failed to initialize shader buffer.\n");
4941         goto fail;
4942     }
4943
4944     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4945     if (!priv->stack)
4946     {
4947         ERR("Failed to allocate memory.\n");
4948         goto fail;
4949     }
4950
4951     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4952     {
4953         ERR("Failed to initialize vertex shader constant heap\n");
4954         goto fail;
4955     }
4956
4957     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4958     {
4959         ERR("Failed to initialize pixel shader constant heap\n");
4960         goto fail;
4961     }
4962
4963     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4964     {
4965         ERR("Failed to initialize rbtree.\n");
4966         goto fail;
4967     }
4968
4969     priv->next_constant_version = 1;
4970
4971     This->shader_priv = priv;
4972     return WINED3D_OK;
4973
4974 fail:
4975     constant_heap_free(&priv->pconst_heap);
4976     constant_heap_free(&priv->vconst_heap);
4977     HeapFree(GetProcessHeap(), 0, priv->stack);
4978     shader_buffer_free(&priv->shader_buffer);
4979     HeapFree(GetProcessHeap(), 0, priv);
4980     return E_OUTOFMEMORY;
4981 }
4982
4983 /* Context activation is done by the caller. */
4984 static void shader_glsl_free(IWineD3DDevice *iface) {
4985     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4986     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4987     struct shader_glsl_priv *priv = This->shader_priv;
4988     int i;
4989
4990     ENTER_GL();
4991     for (i = 0; i < tex_type_count; ++i)
4992     {
4993         if (priv->depth_blt_program_full[i])
4994         {
4995             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4996         }
4997         if (priv->depth_blt_program_masked[i])
4998         {
4999             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
5000         }
5001     }
5002     LEAVE_GL();
5003
5004     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
5005     constant_heap_free(&priv->pconst_heap);
5006     constant_heap_free(&priv->vconst_heap);
5007     HeapFree(GetProcessHeap(), 0, priv->stack);
5008     shader_buffer_free(&priv->shader_buffer);
5009
5010     HeapFree(GetProcessHeap(), 0, This->shader_priv);
5011     This->shader_priv = NULL;
5012 }
5013
5014 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
5015     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
5016     return FALSE;
5017 }
5018
5019 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
5020 {
5021     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
5022      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
5023      * on the version of NV_vertex_program.
5024      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
5025      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
5026      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
5027      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
5028      */
5029     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
5030             || gl_info->limits.arb_ps_instructions <= 512)
5031         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
5032     else
5033         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
5034     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
5035     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
5036
5037     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
5038      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
5039      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
5040      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
5041      * in max native instructions. Intel and others also offer the info in this extension but they
5042      * don't support GLSL (at least on Windows).
5043      *
5044      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
5045      * of instructions is 512 or less we have to do with ps2.0 hardware.
5046      * 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.
5047      */
5048     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
5049             || gl_info->limits.arb_ps_instructions <= 512)
5050         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
5051     else
5052         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
5053
5054     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
5055
5056     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
5057      * Direct3D minimum requirement.
5058      *
5059      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
5060      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
5061      *
5062      * The problem is that the refrast clamps temporary results in the shader to
5063      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
5064      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
5065      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
5066      * offer a way to query this.
5067      */
5068     pCaps->PixelShader1xMaxValue = 8.0;
5069     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
5070
5071     pCaps->VSClipping = TRUE;
5072 }
5073
5074 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
5075 {
5076     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5077     {
5078         TRACE("Checking support for fixup:\n");
5079         dump_color_fixup_desc(fixup);
5080     }
5081
5082     /* We support everything except YUV conversions. */
5083     if (!is_complex_fixup(fixup))
5084     {
5085         TRACE("[OK]\n");
5086         return TRUE;
5087     }
5088
5089     TRACE("[FAILED]\n");
5090     return FALSE;
5091 }
5092
5093 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5094 {
5095     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
5096     /* WINED3DSIH_ADD           */ shader_glsl_arith,
5097     /* WINED3DSIH_BEM           */ shader_glsl_bem,
5098     /* WINED3DSIH_BREAK         */ shader_glsl_break,
5099     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
5100     /* WINED3DSIH_BREAKP        */ NULL,
5101     /* WINED3DSIH_CALL          */ shader_glsl_call,
5102     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
5103     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
5104     /* WINED3DSIH_CND           */ shader_glsl_cnd,
5105     /* WINED3DSIH_CRS           */ shader_glsl_cross,
5106     /* WINED3DSIH_CUT           */ NULL,
5107     /* WINED3DSIH_DCL           */ NULL,
5108     /* WINED3DSIH_DEF           */ NULL,
5109     /* WINED3DSIH_DEFB          */ NULL,
5110     /* WINED3DSIH_DEFI          */ NULL,
5111     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
5112     /* WINED3DSIH_DP3           */ shader_glsl_dot,
5113     /* WINED3DSIH_DP4           */ shader_glsl_dot,
5114     /* WINED3DSIH_DST           */ shader_glsl_dst,
5115     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5116     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5117     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5118     /* WINED3DSIH_EMIT          */ NULL,
5119     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5120     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5121     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5122     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5123     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5124     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5125     /* WINED3DSIH_IADD          */ NULL,
5126     /* WINED3DSIH_IF            */ shader_glsl_if,
5127     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5128     /* WINED3DSIH_IGE           */ NULL,
5129     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5130     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5131     /* WINED3DSIH_LOG           */ shader_glsl_log,
5132     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5133     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5134     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5135     /* WINED3DSIH_LT            */ NULL,
5136     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5137     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5138     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5139     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5140     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5141     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5142     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5143     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5144     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5145     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5146     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5147     /* WINED3DSIH_NOP           */ NULL,
5148     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5149     /* WINED3DSIH_PHASE         */ NULL,
5150     /* WINED3DSIH_POW           */ shader_glsl_pow,
5151     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5152     /* WINED3DSIH_REP           */ shader_glsl_rep,
5153     /* WINED3DSIH_RET           */ shader_glsl_ret,
5154     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5155     /* WINED3DSIH_SETP          */ NULL,
5156     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5157     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5158     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5159     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5160     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5161     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5162     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5163     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5164     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5165     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5166     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5167     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5168     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5169     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5170     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5171     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5172     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5173     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5174     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5175     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5176     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5177     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5178     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5179     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5180     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5181     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5182     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5183 };
5184
5185 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5186     SHADER_HANDLER hw_fct;
5187
5188     /* Select handler */
5189     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5190
5191     /* Unhandled opcode */
5192     if (!hw_fct)
5193     {
5194         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5195         return;
5196     }
5197     hw_fct(ins);
5198
5199     shader_glsl_add_instruction_modifiers(ins);
5200 }
5201
5202 const shader_backend_t glsl_shader_backend = {
5203     shader_glsl_handle_instruction,
5204     shader_glsl_select,
5205     shader_glsl_select_depth_blt,
5206     shader_glsl_deselect_depth_blt,
5207     shader_glsl_update_float_vertex_constants,
5208     shader_glsl_update_float_pixel_constants,
5209     shader_glsl_load_constants,
5210     shader_glsl_load_np2fixup_constants,
5211     shader_glsl_destroy,
5212     shader_glsl_alloc,
5213     shader_glsl_free,
5214     shader_glsl_dirty_const,
5215     shader_glsl_get_caps,
5216     shader_glsl_color_fixup_supported,
5217 };