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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31 #include "wine/port.h"
  32
  33 #include <math.h>
  34 #include <stdio.h>
  35
  36 #include "wined3d_private.h"
  37
  38 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 static BOOL shader_is_pshader_version(enum wined3d_shader_type type)
  43 {
  44     return type == WINED3D_SHADER_TYPE_PIXEL;
  45 }
  46
  47 static BOOL shader_is_vshader_version(enum wined3d_shader_type type)
  48 {
  49     return type == WINED3D_SHADER_TYPE_VERTEX;
  50 }
  51
  52 /* Extract a line. Note that this modifies the source string. */
  53 static char *get_line(char **ptr)
  54 {
  55     char *p, *q;
  56
  57     p = *ptr;
  58     if (!(q = strstr(p, "\n")))
  59     {
  60         if (!*p) return NULL;
  61         *ptr += strlen(p);
  62         return p;
  63     }
  64     *q = '\0';
  65     *ptr = q + 1;
  66
  67     return p;
  68 }
  69
  70 static void shader_arb_dump_program_source(const char *source)
  71 {
  72     ULONG source_size;
  73     char *ptr, *line, *tmp;
  74
  75     source_size = strlen(source) + 1;
  76     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
  77     if (!tmp)
  78     {
  79         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
  80         return;
  81     }
  82     memcpy(tmp, source, source_size);
  83
  84     ptr = tmp;
  85     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
  86     FIXME("\n");
  87
  88     HeapFree(GetProcessHeap(), 0, tmp);
  89 }
  90
  91 enum arb_helper_value
  92 {
  93     ARB_ZERO,
  94     ARB_ONE,
  95     ARB_TWO,
  96     ARB_0001,
  97     ARB_EPS,
  98
  99     ARB_VS_REL_OFFSET
 100 };
 101
 102 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
 103 {
 104     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
 105     {
 106         ERR("Geometry shaders are unsupported\n");
 107         return "bad";
 108     }
 109
 110     if (shader == WINED3D_SHADER_TYPE_PIXEL)
 111     {
 112         switch (value)
 113         {
 114             case ARB_ZERO: return "ps_helper_const.x";
 115             case ARB_ONE: return "ps_helper_const.y";
 116             case ARB_TWO: return "coefmul.x";
 117             case ARB_0001: return "ps_helper_const.xxxy";
 118             case ARB_EPS: return "ps_helper_const.z";
 119             default: break;
 120         }
 121     }
 122     else
 123     {
 124         switch (value)
 125         {
 126             case ARB_ZERO: return "helper_const.x";
 127             case ARB_ONE: return "helper_const.y";
 128             case ARB_TWO: return "helper_const.z";
 129             case ARB_EPS: return "helper_const.w";
 130             case ARB_0001: return "helper_const.xxxy";
 131             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
 132         }
 133     }
 134     FIXME("Unmanaged %s shader helper constant requested: %u\n",
 135           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
 136     switch (value)
 137     {
 138         case ARB_ZERO: return "0.0";
 139         case ARB_ONE: return "1.0";
 140         case ARB_TWO: return "2.0";
 141         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
 142         case ARB_EPS: return "1e-8";
 143         default: return "bad";
 144     }
 145 }
 146
 147 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
 148 {
 149     return state->lowest_disabled_stage < 7;
 150 }
 151
 152 /* ARB_program_shader private data */
 153
 154 struct control_frame
 155 {
 156     struct                          list entry;
 157     enum
 158     {
 159         IF,
 160         IFC,
 161         LOOP,
 162         REP
 163     } type;
 164     BOOL                            muting;
 165     BOOL                            outer_loop;
 166     union
 167     {
 168         unsigned int                loop;
 169         unsigned int                ifc;
 170     } no;
 171     struct wined3d_shader_loop_control loop_control;
 172     BOOL                            had_else;
 173 };
 174
 175 struct arb_ps_np2fixup_info
 176 {
 177     struct ps_np2fixup_info         super;
 178     /* For ARB we need a offset value:
 179      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 180      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 181      * array we need an offset to the index inside the program local parameter array. */
 182     UINT                            offset;
 183 };
 184
 185 struct arb_ps_compile_args
 186 {
 187     struct ps_compile_args          super;
 188     WORD                            bools;
 189     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 190     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 191 };
 192
 193 struct stb_const_desc
 194 {
 195     unsigned char           texunit;
 196     UINT                    const_num;
 197 };
 198
 199 struct arb_ps_compiled_shader
 200 {
 201     struct arb_ps_compile_args      args;
 202     struct arb_ps_np2fixup_info     np2fixup_info;
 203     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 204     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 205     UINT                            int_consts[MAX_CONST_I];
 206     GLuint                          prgId;
 207     UINT                            ycorrection;
 208     unsigned char                   numbumpenvmatconsts;
 209     char                            num_int_consts;
 210 };
 211
 212 struct arb_vs_compile_args
 213 {
 214     struct vs_compile_args          super;
 215     union
 216     {
 217         struct
 218         {
 219             WORD                    bools;
 220             unsigned char           clip_texcoord;
 221             unsigned char           clipplane_mask;
 222         }                           boolclip;
 223         DWORD                       boolclip_compare;
 224     } clip;
 225     DWORD                           ps_signature;
 226     union
 227     {
 228         unsigned char               samplers[4];
 229         DWORD                       samplers_compare;
 230     } vertex;
 231     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 232 };
 233
 234 struct arb_vs_compiled_shader
 235 {
 236     struct arb_vs_compile_args      args;
 237     GLuint                          prgId;
 238     UINT                            int_consts[MAX_CONST_I];
 239     char                            num_int_consts;
 240     char                            need_color_unclamp;
 241     UINT                            pos_fixup;
 242 };
 243
 244 struct recorded_instruction
 245 {
 246     struct wined3d_shader_instruction ins;
 247     struct list entry;
 248 };
 249
 250 struct shader_arb_ctx_priv
 251 {
 252     char addr_reg[20];
 253     enum
 254     {
 255         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 256         ARB,
 257         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 258         NV2,
 259         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 260         NV3
 261     } target_version;
 262
 263     const struct arb_vs_compile_args    *cur_vs_args;
 264     const struct arb_ps_compile_args    *cur_ps_args;
 265     const struct arb_ps_compiled_shader *compiled_fprog;
 266     const struct arb_vs_compiled_shader *compiled_vprog;
 267     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 268     struct list                         control_frames;
 269     struct list                         record;
 270     BOOL                                recording;
 271     BOOL                                muted;
 272     unsigned int                        num_loops, loop_depth, num_ifcs;
 273     int                                 aL;
 274
 275     unsigned int                        vs_clipplanes;
 276     BOOL                                footer_written;
 277     BOOL                                in_main_func;
 278
 279     /* For 3.0 vertex shaders */
 280     const char                          *vs_output[MAX_REG_OUTPUT];
 281     /* For 2.x and earlier vertex shaders */
 282     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 283
 284     /* 3.0 pshader input for compatibility with fixed function */
 285     const char                          *ps_input[MAX_REG_INPUT];
 286 };
 287
 288 struct ps_signature
 289 {
 290     struct wined3d_shader_signature_element *sig;
 291     DWORD                               idx;
 292     struct wine_rb_entry                entry;
 293 };
 294
 295 struct arb_pshader_private {
 296     struct arb_ps_compiled_shader   *gl_shaders;
 297     UINT                            num_gl_shaders, shader_array_size;
 298     DWORD                           input_signature_idx;
 299     DWORD                           clipplane_emulation;
 300     BOOL                            clamp_consts;
 301 };
 302
 303 struct arb_vshader_private {
 304     struct arb_vs_compiled_shader   *gl_shaders;
 305     UINT                            num_gl_shaders, shader_array_size;
 306     UINT rel_offset;
 307 };
 308
 309 struct shader_arb_priv
 310 {
 311     GLuint                  current_vprogram_id;
 312     GLuint                  current_fprogram_id;
 313     const struct arb_ps_compiled_shader *compiled_fprog;
 314     const struct arb_vs_compiled_shader *compiled_vprog;
 315     GLuint                  depth_blt_vprogram_id;
 316     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 317     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 318     BOOL                    use_arbfp_fixed_func;
 319     struct wine_rb_tree     fragment_shaders;
 320     BOOL                    last_ps_const_clamped;
 321     BOOL                    last_vs_color_unclamp;
 322
 323     struct wine_rb_tree     signature_tree;
 324     DWORD ps_sig_number;
 325
 326     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
 327     char *vshader_const_dirty, *pshader_const_dirty;
 328     const struct wined3d_context *last_context;
 329
 330     const struct fragment_pipeline *fragment_pipe;
 331     BOOL ffp_proj_control;
 332 };
 333
 334 /* Context activation for state handlers is done by the caller. */
 335
 336 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 337         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 338 {
 339     if (shader_data->rel_offset) return TRUE;
 340     if (!reg_maps->usesmova) return FALSE;
 341     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 342 }
 343
 344 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 345 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 346 {
 347     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 348             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 349 }
 350
 351 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 352         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 353 {
 354     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 355     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 356     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 357     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 358     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 359     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 360     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 361     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 362     return FALSE;
 363 }
 364
 365 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 366         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 367 {
 368     unsigned int ret = 1;
 369     /* We use one PARAM for the pos fixup, and in some cases one to load
 370      * some immediate values into the shader. */
 371     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 372     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 373     return ret;
 374 }
 375
 376 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 377  * When constant_list == NULL, it will load all the constants.
 378  *
 379  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 380  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 381  */
 382 /* Context activation is done by the caller. */
 383 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 384         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 385         const float *constants, char *dirty_consts)
 386 {
 387     struct wined3d_shader_lconst *lconst;
 388     DWORD i, j;
 389     unsigned int ret;
 390
 391     if (TRACE_ON(d3d_constants))
 392     {
 393         for(i = 0; i < max_constants; i++) {
 394             if(!dirty_consts[i]) continue;
 395             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 396                         constants[i * 4 + 0], constants[i * 4 + 1],
 397                         constants[i * 4 + 2], constants[i * 4 + 3]);
 398         }
 399     }
 400
 401     i = 0;
 402
 403     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 404     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 405     {
 406         float lcl_const[4];
 407         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 408          * shaders, the first 8 constants are marked dirty for reload
 409          */
 410         for(; i < min(8, max_constants); i++) {
 411             if(!dirty_consts[i]) continue;
 412             dirty_consts[i] = 0;
 413
 414             j = 4 * i;
 415             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 416             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 417             else lcl_const[0] = constants[j + 0];
 418
 419             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 420             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 421             else lcl_const[1] = constants[j + 1];
 422
 423             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 424             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 425             else lcl_const[2] = constants[j + 2];
 426
 427             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 428             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 429             else lcl_const[3] = constants[j + 3];
 430
 431             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 432         }
 433
 434         /* If further constants are dirty, reload them without clamping.
 435          *
 436          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 437          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 438          * above would always re-check the first 8 constants since max_constant remains at the init
 439          * value
 440          */
 441     }
 442
 443     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 444     {
 445         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 446          * or just reloading *all* constants at once
 447          *
 448         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 449          */
 450         for(; i < max_constants; i++) {
 451             if(!dirty_consts[i]) continue;
 452
 453             /* Find the next block of dirty constants */
 454             dirty_consts[i] = 0;
 455             j = i;
 456             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 457                 dirty_consts[i] = 0;
 458             }
 459
 460             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 461         }
 462     } else {
 463         for(; i < max_constants; i++) {
 464             if(dirty_consts[i]) {
 465                 dirty_consts[i] = 0;
 466                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 467             }
 468         }
 469     }
 470     checkGLcall("glProgramEnvParameter4fvARB()");
 471
 472     /* Load immediate constants */
 473     if (shader->load_local_constsF)
 474     {
 475         if (TRACE_ON(d3d_shader))
 476         {
 477             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 478             {
 479                 GLfloat* values = (GLfloat*)lconst->value;
 480                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 481                         values[0], values[1], values[2], values[3]);
 482             }
 483         }
 484         /* Immediate constants are clamped for 1.X shaders at loading times */
 485         ret = 0;
 486         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 487         {
 488             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 489             ret = max(ret, lconst->idx + 1);
 490             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 491         }
 492         checkGLcall("glProgramEnvParameter4fvARB()");
 493         return ret; /* The loaded immediate constants need reloading for the next shader */
 494     } else {
 495         return 0; /* No constants are dirty now */
 496     }
 497 }
 498
 499 /**
 500  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 501  */
 502 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 503         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 504 {
 505     const struct shader_arb_priv * priv = shader_priv;
 506
 507     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 508     if (!use_ps(state)) return;
 509
 510     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 511         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 512         UINT i;
 513         WORD active = fixup->super.active;
 514         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 515
 516         for (i = 0; active; active >>= 1, ++i)
 517         {
 518             const struct wined3d_texture *tex = state->textures[i];
 519             const unsigned char idx = fixup->super.idx[i];
 520             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 521
 522             if (!(active & 1)) continue;
 523
 524             if (!tex) {
 525                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 526                 continue;
 527             }
 528
 529             if (idx % 2)
 530             {
 531                 tex_dim[2] = tex->pow2_matrix[0];
 532                 tex_dim[3] = tex->pow2_matrix[5];
 533             }
 534             else
 535             {
 536                 tex_dim[0] = tex->pow2_matrix[0];
 537                 tex_dim[1] = tex->pow2_matrix[5];
 538             }
 539         }
 540
 541         for (i = 0; i < fixup->super.num_consts; ++i) {
 542             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 543                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 544         }
 545     }
 546 }
 547
 548 /* Context activation is done by the caller. */
 549 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 550         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 551 {
 552     const struct wined3d_gl_info *gl_info = context->gl_info;
 553     unsigned char i;
 554
 555     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 556     {
 557         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 558
 559         /* The state manager takes care that this function is always called if the bump env matrix changes */
 560         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 561         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 562                 gl_shader->bumpenvmatconst[i].const_num, data));
 563
 564         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 565         {
 566             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 567              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 568              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 569              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 570             */
 571             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 572             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 573                     gl_shader->luminanceconst[i].const_num, scale));
 574         }
 575     }
 576     checkGLcall("Load bumpmap consts");
 577
 578     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 579     {
 580         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 581         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 582         * ycorrection.z: 1.0
 583         * ycorrection.w: 0.0
 584         */
 585         float val[4];
 586         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 587         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 588         val[2] = 1.0f;
 589         val[3] = 0.0f;
 590         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 591         checkGLcall("y correction loading");
 592     }
 593
 594     if (!gl_shader->num_int_consts) return;
 595
 596     for(i = 0; i < MAX_CONST_I; i++)
 597     {
 598         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 599         {
 600             float val[4];
 601             val[0] = (float)state->ps_consts_i[4 * i];
 602             val[1] = (float)state->ps_consts_i[4 * i + 1];
 603             val[2] = (float)state->ps_consts_i[4 * i + 2];
 604             val[3] = -1.0f;
 605
 606             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 607         }
 608     }
 609     checkGLcall("Load ps int consts");
 610 }
 611
 612 /* Context activation is done by the caller. */
 613 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 614         const struct wined3d_context *context, const struct wined3d_state *state)
 615 {
 616     const struct wined3d_gl_info *gl_info = context->gl_info;
 617     float position_fixup[4];
 618     unsigned char i;
 619
 620     /* Upload the position fixup */
 621     shader_get_position_fixup(context, state, position_fixup);
 622     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 623
 624     if (!gl_shader->num_int_consts) return;
 625
 626     for(i = 0; i < MAX_CONST_I; i++)
 627     {
 628         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 629         {
 630             float val[4];
 631             val[0] = (float)state->vs_consts_i[4 * i];
 632             val[1] = (float)state->vs_consts_i[4 * i + 1];
 633             val[2] = (float)state->vs_consts_i[4 * i + 2];
 634             val[3] = -1.0f;
 635
 636             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 637         }
 638     }
 639     checkGLcall("Load vs int consts");
 640 }
 641
 642 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
 643         enum wined3d_shader_mode fragment_mode);
 644
 645 /**
 646  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 647  *
 648  * We only support float constants in ARB at the moment, so don't
 649  * worry about the Integers or Booleans
 650  */
 651 /* Context activation is done by the caller (state handler). */
 652 static void shader_arb_load_constants_internal(const struct wined3d_context *context,
 653         BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
 654 {
 655     struct wined3d_device *device = context->swapchain->device;
 656     const struct wined3d_stateblock *stateblock = device->stateBlock;
 657     const struct wined3d_state *state = &stateblock->state;
 658     const struct wined3d_gl_info *gl_info = context->gl_info;
 659     struct shader_arb_priv *priv = device->shader_priv;
 660
 661     if (!from_shader_select)
 662     {
 663         const struct wined3d_shader *vshader = state->vertex_shader, *pshader = state->pixel_shader;
 664         if (vshader
 665                 && (stateblock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants
 666                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 667                 && (stateblock->changed.vertexShaderConstantsI
 668                 & vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
 669         {
 670             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
 671             shader_arb_select(context,
 672                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 673                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 674         }
 675         else if (pshader
 676                 && (stateblock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants
 677                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
 678                 && (stateblock->changed.pixelShaderConstantsI
 679                 & pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
 680         {
 681             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
 682             shader_arb_select(context,
 683                     useVertexShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP,
 684                     usePixelShader ? WINED3D_SHADER_MODE_SHADER : WINED3D_SHADER_MODE_FFP);
 685         }
 686     }
 687
 688     if (context != priv->last_context)
 689     {
 690         memset(priv->vshader_const_dirty, 1,
 691                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 692         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 693
 694         memset(priv->pshader_const_dirty, 1,
 695                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 696         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 697
 698         priv->last_context = context;
 699     }
 700
 701     if (useVertexShader)
 702     {
 703         struct wined3d_shader *vshader = state->vertex_shader;
 704         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 705
 706         /* Load DirectX 9 float constants for vertex shader */
 707         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 708                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 709         shader_arb_vs_local_constants(gl_shader, context, state);
 710     }
 711
 712     if (usePixelShader)
 713     {
 714         struct wined3d_shader *pshader = state->pixel_shader;
 715         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 716         UINT rt_height = state->fb->render_targets[0]->resource.height;
 717
 718         /* Load DirectX 9 float constants for pixel shader */
 719         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 720                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 721         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 722     }
 723 }
 724
 725 static void shader_arb_load_constants(const struct wined3d_context *context, BOOL ps, BOOL vs)
 726 {
 727     shader_arb_load_constants_internal(context, ps, vs, FALSE);
 728 }
 729
 730 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 731 {
 732     struct wined3d_context *context = context_get_current();
 733     struct shader_arb_priv *priv = device->shader_priv;
 734
 735     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 736      * context. On a context switch the old context will be fully dirtified */
 737     if (!context || context->swapchain->device != device) return;
 738
 739     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 740     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 741 }
 742
 743 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 744 {
 745     struct wined3d_context *context = context_get_current();
 746     struct shader_arb_priv *priv = device->shader_priv;
 747
 748     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 749      * context. On a context switch the old context will be fully dirtified */
 750     if (!context || context->swapchain->device != device) return;
 751
 752     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 753     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 754 }
 755
 756 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 757 {
 758     const struct wined3d_shader_lconst *lconst;
 759     DWORD *ret;
 760     DWORD idx = 0;
 761
 762     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 763         return NULL;
 764
 765     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 766     if (!ret)
 767     {
 768         ERR("Out of memory\n");
 769         return NULL;
 770     }
 771
 772     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 773     {
 774         ret[lconst->idx] = idx++;
 775     }
 776     return ret;
 777 }
 778
 779 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 780 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 781         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 782         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 783         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 784 {
 785     DWORD i, next_local = 0;
 786     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 787     const struct wined3d_shader_lconst *lconst;
 788     unsigned max_constantsF;
 789     DWORD map;
 790
 791     /* In pixel shaders, all private constants are program local, we don't need anything
 792      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 793      * If we need a private constant the GL implementation will squeeze it in somewhere
 794      *
 795      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 796      * immediate values. The posFixup is loaded using program.env for now, so always
 797      * subtract one from the number of constants. If the shader uses indirect addressing,
 798      * account for the helper const too because we have to declare all available d3d constants
 799      * and don't know which are actually used.
 800      */
 801     if (pshader)
 802     {
 803         max_constantsF = gl_info->limits.arb_ps_native_constants;
 804         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 805         if (max_constantsF < 24)
 806             max_constantsF = gl_info->limits.arb_ps_float_constants;
 807     }
 808     else
 809     {
 810         const struct arb_vshader_private *shader_data = shader->backend_data;
 811         max_constantsF = gl_info->limits.arb_vs_native_constants;
 812         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 813          * Also prevents max_constantsF from becoming less than 0 and
 814          * wrapping . */
 815         if (max_constantsF < 96)
 816             max_constantsF = gl_info->limits.arb_vs_float_constants;
 817
 818         if (reg_maps->usesrelconstF)
 819         {
 820             DWORD highest_constf = 0, clip_limit;
 821
 822             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 823             max_constantsF -= count_bits(reg_maps->integer_constants);
 824             max_constantsF -= gl_info->reserved_arb_constants;
 825
 826             for (i = 0; i < shader->limits.constant_float; ++i)
 827             {
 828                 DWORD idx = i >> 5;
 829                 DWORD shift = i & 0x1f;
 830                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 831             }
 832
 833             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 834             {
 835                 if(ctx->cur_vs_args->super.clip_enabled)
 836                     clip_limit = gl_info->limits.clipplanes;
 837                 else
 838                     clip_limit = 0;
 839             }
 840             else
 841             {
 842                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 843                 clip_limit = min(count_bits(mask), 4);
 844             }
 845             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 846             max_constantsF -= *num_clipplanes;
 847             if(*num_clipplanes < clip_limit)
 848             {
 849                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 850             }
 851         }
 852         else
 853         {
 854             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 855             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 856         }
 857     }
 858
 859     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 860     {
 861         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 862     }
 863
 864     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 865     {
 866         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 867     }
 868
 869     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 870     {
 871         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 872         {
 873             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 874         }
 875     }
 876
 877     /* Load local constants using the program-local space,
 878      * this avoids reloading them each time the shader is used
 879      */
 880     if (lconst_map)
 881     {
 882         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 883         {
 884             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 885                            lconst_map[lconst->idx]);
 886             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 887         }
 888     }
 889
 890     /* After subtracting privately used constants from the hardware limit(they are loaded as
 891      * local constants), make sure the shader doesn't violate the env constant limit
 892      */
 893     if(pshader)
 894     {
 895         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 896     }
 897     else
 898     {
 899         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 900     }
 901
 902     /* Avoid declaring more constants than needed */
 903     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 904
 905     /* we use the array-based constants array if the local constants are marked for loading,
 906      * because then we use indirect addressing, or when the local constant list is empty,
 907      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 908      * local constants do not declare the loaded constants as an array because ARB compilers usually
 909      * do not optimize unused constants away
 910      */
 911     if (reg_maps->usesrelconstF)
 912     {
 913         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 914         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 915                     max_constantsF, max_constantsF - 1);
 916     } else {
 917         for(i = 0; i < max_constantsF; i++) {
 918             DWORD idx, mask;
 919             idx = i >> 5;
 920             mask = 1 << (i & 0x1f);
 921             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 922             {
 923                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 924             }
 925         }
 926     }
 927
 928     return next_local;
 929 }
 930
 931 static const char * const shift_tab[] = {
 932     "dummy",     /*  0 (none) */
 933     "coefmul.x", /*  1 (x2)   */
 934     "coefmul.y", /*  2 (x4)   */
 935     "coefmul.z", /*  3 (x8)   */
 936     "coefmul.w", /*  4 (x16)  */
 937     "dummy",     /*  5 (x32)  */
 938     "dummy",     /*  6 (x64)  */
 939     "dummy",     /*  7 (x128) */
 940     "dummy",     /*  8 (d256) */
 941     "dummy",     /*  9 (d128) */
 942     "dummy",     /* 10 (d64)  */
 943     "dummy",     /* 11 (d32)  */
 944     "coefdiv.w", /* 12 (d16)  */
 945     "coefdiv.z", /* 13 (d8)   */
 946     "coefdiv.y", /* 14 (d4)   */
 947     "coefdiv.x"  /* 15 (d2)   */
 948 };
 949
 950 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 951         const struct wined3d_shader_dst_param *dst, char *write_mask)
 952 {
 953     char *ptr = write_mask;
 954
 955     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 956     {
 957         *ptr++ = '.';
 958         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 959         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 960         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 961         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 962     }
 963
 964     *ptr = '\0';
 965 }
 966
 967 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 968 {
 969     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 970      * but addressed as "rgba". To fix this we need to swap the register's x
 971      * and z components. */
 972     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 973     char *ptr = swizzle_str;
 974
 975     /* swizzle bits fields: wwzzyyxx */
 976     DWORD swizzle = param->swizzle;
 977     DWORD swizzle_x = swizzle & 0x03;
 978     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 979     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 980     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 981
 982     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 983      * generate a swizzle string. Unless we need to our own swizzling. */
 984     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 985     {
 986         *ptr++ = '.';
 987         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 988             *ptr++ = swizzle_chars[swizzle_x];
 989         } else {
 990             *ptr++ = swizzle_chars[swizzle_x];
 991             *ptr++ = swizzle_chars[swizzle_y];
 992             *ptr++ = swizzle_chars[swizzle_z];
 993             *ptr++ = swizzle_chars[swizzle_w];
 994         }
 995     }
 996
 997     *ptr = '\0';
 998 }
 999
1000 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
1001 {
1002     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1003     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1004
1005     if (!strcmp(priv->addr_reg, src)) return;
1006
1007     strcpy(priv->addr_reg, src);
1008     shader_addline(buffer, "ARL A0.x, %s;\n", src);
1009 }
1010
1011 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1012         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
1013
1014 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
1015         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
1016 {
1017     /* oPos, oFog and oPts in D3D */
1018     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
1019     const struct wined3d_shader *shader = ins->ctx->shader;
1020     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1021     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1022     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1023
1024     *is_color = FALSE;
1025
1026     switch (reg->type)
1027     {
1028         case WINED3DSPR_TEMP:
1029             sprintf(register_name, "R%u", reg->idx[0].offset);
1030             break;
1031
1032         case WINED3DSPR_INPUT:
1033             if (pshader)
1034             {
1035                 if (reg_maps->shader_version.major < 3)
1036                 {
1037                     if (!reg->idx[0].offset)
1038                         strcpy(register_name, "fragment.color.primary");
1039                     else
1040                         strcpy(register_name, "fragment.color.secondary");
1041                 }
1042                 else
1043                 {
1044                     if (reg->idx[0].rel_addr)
1045                     {
1046                         char rel_reg[50];
1047                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1048
1049                         if (!strcmp(rel_reg, "**aL_emul**"))
1050                         {
1051                             DWORD idx = ctx->aL + reg->idx[0].offset;
1052                             if(idx < MAX_REG_INPUT)
1053                             {
1054                                 strcpy(register_name, ctx->ps_input[idx]);
1055                             }
1056                             else
1057                             {
1058                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1059                                 sprintf(register_name, "out_of_bounds_%u", idx);
1060                             }
1061                         }
1062                         else if (reg_maps->input_registers & 0x0300)
1063                         {
1064                             /* There are two ways basically:
1065                              *
1066                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1067                              *    That means trouble if the loop also contains a breakc or if the control values
1068                              *    aren't local constants.
1069                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1070                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1071                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1072                              *    ADAC to load the condition code register and pop it again afterwards
1073                              */
1074                             FIXME("Relative input register addressing with more than 8 registers\n");
1075
1076                             /* This is better than nothing for now */
1077                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1078                         }
1079                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1080                         {
1081                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1082                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1083                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1084                              * proper varyings, or loop unrolling
1085                              *
1086                              * For now use the texcoords and hope for the best
1087                              */
1088                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1089                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1090                         }
1091                         else
1092                         {
1093                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1094                              * pulls GL_NV_fragment_program2 in
1095                              */
1096                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1097                         }
1098                     }
1099                     else
1100                     {
1101                         if (reg->idx[0].offset < MAX_REG_INPUT)
1102                         {
1103                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1104                         }
1105                         else
1106                         {
1107                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1108                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1109                         }
1110                     }
1111                 }
1112             }
1113             else
1114             {
1115                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx[0].offset))
1116                     *is_color = TRUE;
1117                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1118             }
1119             break;
1120
1121         case WINED3DSPR_CONST:
1122             if (!pshader && reg->idx[0].rel_addr)
1123             {
1124                 const struct arb_vshader_private *shader_data = shader->backend_data;
1125                 UINT rel_offset = shader_data->rel_offset;
1126                 BOOL aL = FALSE;
1127                 char rel_reg[50];
1128                 if (reg_maps->shader_version.major < 2)
1129                 {
1130                     sprintf(rel_reg, "A0.x");
1131                 }
1132                 else
1133                 {
1134                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1135                     if (ctx->target_version == ARB)
1136                     {
1137                         if (!strcmp(rel_reg, "**aL_emul**"))
1138                         {
1139                             aL = TRUE;
1140                         } else {
1141                             shader_arb_request_a0(ins, rel_reg);
1142                             sprintf(rel_reg, "A0.x");
1143                         }
1144                     }
1145                 }
1146                 if (aL)
1147                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1148                 else if (reg->idx[0].offset >= rel_offset)
1149                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1150                 else
1151                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1152             }
1153             else
1154             {
1155                 if (reg_maps->usesrelconstF)
1156                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1157                 else
1158                     sprintf(register_name, "C%u", reg->idx[0].offset);
1159             }
1160             break;
1161
1162         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1163             if (pshader)
1164             {
1165                 if (reg_maps->shader_version.major == 1
1166                         && reg_maps->shader_version.minor <= 3)
1167                     /* In ps <= 1.3, Tx is a temporary register as destination
1168                      * to all instructions, and as source to most instructions.
1169                      * For some instructions it is the texcoord input. Those
1170                      * instructions know about the special use. */
1171                     sprintf(register_name, "T%u", reg->idx[0].offset);
1172                 else
1173                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1174                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1175             }
1176             else
1177             {
1178                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1179                     sprintf(register_name, "A%u", reg->idx[0].offset);
1180                 else
1181                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1182             }
1183             break;
1184
1185         case WINED3DSPR_COLOROUT:
1186             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx[0].offset)
1187             {
1188                 strcpy(register_name, "TMP_COLOR");
1189             }
1190             else
1191             {
1192                 if (ctx->cur_ps_args->super.srgb_correction)
1193                     FIXME("sRGB correction on higher render targets.\n");
1194                 if (reg_maps->rt_mask > 1)
1195                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1196                 else
1197                     strcpy(register_name, "result.color");
1198             }
1199             break;
1200
1201         case WINED3DSPR_RASTOUT:
1202             if (reg->idx[0].offset == 1)
1203                 sprintf(register_name, "%s", ctx->fog_output);
1204             else
1205                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1206             break;
1207
1208         case WINED3DSPR_DEPTHOUT:
1209             strcpy(register_name, "result.depth");
1210             break;
1211
1212         case WINED3DSPR_ATTROUT:
1213         /* case WINED3DSPR_OUTPUT: */
1214             if (pshader)
1215                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1216             else
1217                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1218             break;
1219
1220         case WINED3DSPR_TEXCRDOUT:
1221             if (pshader)
1222                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1223             else if (reg_maps->shader_version.major < 3)
1224                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1225             else
1226                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1227             break;
1228
1229         case WINED3DSPR_LOOP:
1230             if(ctx->target_version >= NV2)
1231             {
1232                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1233                 if(pshader) sprintf(register_name, "A0.x");
1234                 else sprintf(register_name, "aL.y");
1235             }
1236             else
1237             {
1238                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1239                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1240                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1241                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1242                  * indexing
1243                  */
1244                 sprintf(register_name, "**aL_emul**");
1245             }
1246
1247             break;
1248
1249         case WINED3DSPR_CONSTINT:
1250             sprintf(register_name, "I%u", reg->idx[0].offset);
1251             break;
1252
1253         case WINED3DSPR_MISCTYPE:
1254             if (!reg->idx[0].offset)
1255                 sprintf(register_name, "vpos");
1256             else if (reg->idx[0].offset == 1)
1257                 sprintf(register_name, "fragment.facing.x");
1258             else
1259                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1260             break;
1261
1262         default:
1263             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1264             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1265             break;
1266     }
1267 }
1268
1269 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1270         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1271 {
1272     char register_name[255];
1273     char write_mask[6];
1274     BOOL is_color;
1275
1276     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1277     strcpy(str, register_name);
1278
1279     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1280     strcat(str, write_mask);
1281 }
1282
1283 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1284 {
1285     switch(channel_source)
1286     {
1287         case CHANNEL_SOURCE_ZERO: return "0";
1288         case CHANNEL_SOURCE_ONE: return "1";
1289         case CHANNEL_SOURCE_X: return "x";
1290         case CHANNEL_SOURCE_Y: return "y";
1291         case CHANNEL_SOURCE_Z: return "z";
1292         case CHANNEL_SOURCE_W: return "w";
1293         default:
1294             FIXME("Unhandled channel source %#x\n", channel_source);
1295             return "undefined";
1296     }
1297 }
1298
1299 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1300         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1301 {
1302     DWORD mask;
1303
1304     if (is_complex_fixup(fixup))
1305     {
1306         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1307         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1308         return;
1309     }
1310
1311     mask = 0;
1312     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1313     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1314     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1315     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1316     mask &= dst_mask;
1317
1318     if (mask)
1319     {
1320         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1321                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1322                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1323     }
1324
1325     mask = 0;
1326     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1327     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1328     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1329     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1330     mask &= dst_mask;
1331
1332     if (mask)
1333     {
1334         char reg_mask[6];
1335         char *ptr = reg_mask;
1336
1337         if (mask != WINED3DSP_WRITEMASK_ALL)
1338         {
1339             *ptr++ = '.';
1340             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1341             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1342             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1343             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1344         }
1345         *ptr = '\0';
1346
1347         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1348     }
1349 }
1350
1351 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1352 {
1353     DWORD mod;
1354     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1355     if (!ins->dst_count) return "";
1356
1357     mod = ins->dst[0].modifiers;
1358
1359     /* Silently ignore PARTIALPRECISION if its not supported */
1360     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1361
1362     if(mod & WINED3DSPDM_MSAMPCENTROID)
1363     {
1364         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1365         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1366     }
1367
1368     switch(mod)
1369     {
1370         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1371             return "H_SAT";
1372
1373         case WINED3DSPDM_SATURATE:
1374             return "_SAT";
1375
1376         case WINED3DSPDM_PARTIALPRECISION:
1377             return "H";
1378
1379         case 0:
1380             return "";
1381
1382         default:
1383             FIXME("Unknown modifiers 0x%08x\n", mod);
1384             return "";
1385     }
1386 }
1387
1388 #define TEX_PROJ        0x1
1389 #define TEX_BIAS        0x2
1390 #define TEX_LOD         0x4
1391 #define TEX_DERIV       0x10
1392
1393 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1394         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1395 {
1396     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1397     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1398     const struct wined3d_shader *shader = ins->ctx->shader;
1399     const struct wined3d_texture *texture;
1400     const char *tex_type;
1401     BOOL np2_fixup = FALSE;
1402     struct wined3d_device *device = shader->device;
1403     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1404     const char *mod;
1405     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1406
1407     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1408     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1409
1410     switch(sampler_type) {
1411         case WINED3DSTT_1D:
1412             tex_type = "1D";
1413             break;
1414
1415         case WINED3DSTT_2D:
1416             texture = device->stateBlock->state.textures[sampler_idx];
1417             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1418             {
1419                 tex_type = "RECT";
1420             } else {
1421                 tex_type = "2D";
1422             }
1423             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1424             {
1425                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1426                 {
1427                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1428                     else np2_fixup = TRUE;
1429                 }
1430             }
1431             break;
1432
1433         case WINED3DSTT_VOLUME:
1434             tex_type = "3D";
1435             break;
1436
1437         case WINED3DSTT_CUBE:
1438             tex_type = "CUBE";
1439             break;
1440
1441         default:
1442             ERR("Unexpected texture type %d\n", sampler_type);
1443             tex_type = "";
1444     }
1445
1446     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1447      * so don't use shader_arb_get_modifier
1448      */
1449     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1450     else mod = "";
1451
1452     /* Fragment samplers always have indentity mapping */
1453     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1454     {
1455         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1456     }
1457
1458     if (flags & TEX_DERIV)
1459     {
1460         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1461         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1462         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1463                        dsx, dsy,sampler_idx, tex_type);
1464     }
1465     else if(flags & TEX_LOD)
1466     {
1467         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1468         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1469         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1470                        sampler_idx, tex_type);
1471     }
1472     else if (flags & TEX_BIAS)
1473     {
1474         /* Shouldn't be possible, but let's check for it */
1475         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1476         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1477         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1478     }
1479     else if (flags & TEX_PROJ)
1480     {
1481         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1482     }
1483     else
1484     {
1485         if (np2_fixup)
1486         {
1487             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1488             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1489                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1490
1491             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1492         }
1493         else
1494             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1495     }
1496
1497     if (pshader)
1498     {
1499         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1500                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1501                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1502                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1503     }
1504 }
1505
1506 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1507         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1508 {
1509     /* Generate a line that does the input modifier computation and return the input register to use */
1510     BOOL is_color = FALSE;
1511     char regstr[256];
1512     char swzstr[20];
1513     int insert_line;
1514     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1515     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1516     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1517     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1518
1519     /* Assume a new line will be added */
1520     insert_line = 1;
1521
1522     /* Get register name */
1523     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1524     shader_arb_get_swizzle(src, is_color, swzstr);
1525
1526     switch (src->modifiers)
1527     {
1528     case WINED3DSPSM_NONE:
1529         sprintf(outregstr, "%s%s", regstr, swzstr);
1530         insert_line = 0;
1531         break;
1532     case WINED3DSPSM_NEG:
1533         sprintf(outregstr, "-%s%s", regstr, swzstr);
1534         insert_line = 0;
1535         break;
1536     case WINED3DSPSM_BIAS:
1537         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1538         break;
1539     case WINED3DSPSM_BIASNEG:
1540         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1541         break;
1542     case WINED3DSPSM_SIGN:
1543         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1544         break;
1545     case WINED3DSPSM_SIGNNEG:
1546         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1547         break;
1548     case WINED3DSPSM_COMP:
1549         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1550         break;
1551     case WINED3DSPSM_X2:
1552         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1553         break;
1554     case WINED3DSPSM_X2NEG:
1555         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1556         break;
1557     case WINED3DSPSM_DZ:
1558         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1559         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1560         break;
1561     case WINED3DSPSM_DW:
1562         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1563         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1564         break;
1565     case WINED3DSPSM_ABS:
1566         if(ctx->target_version >= NV2) {
1567             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1568             insert_line = 0;
1569         } else {
1570             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1571         }
1572         break;
1573     case WINED3DSPSM_ABSNEG:
1574         if(ctx->target_version >= NV2) {
1575             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1576         } else {
1577             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1578             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1579         }
1580         insert_line = 0;
1581         break;
1582     default:
1583         sprintf(outregstr, "%s%s", regstr, swzstr);
1584         insert_line = 0;
1585     }
1586
1587     /* Return modified or original register, with swizzle */
1588     if (insert_line)
1589         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1590 }
1591
1592 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1593 {
1594     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1595     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1596     DWORD sampler_code = dst->reg.idx[0].offset;
1597     char dst_name[50];
1598     char src_name[2][50];
1599
1600     shader_arb_get_dst_param(ins, dst, dst_name);
1601
1602     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1603      *
1604      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1605      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1606      * temps is done.
1607      */
1608     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1609     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1610     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1611     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1612     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1613
1614     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1615     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1616 }
1617
1618 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1619 {
1620     *extra_char = ' ';
1621     switch(mod)
1622     {
1623         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1624         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1625         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1626         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1627         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1628         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1629         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1630         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1631         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1632         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1633         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1634         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1635         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1636     }
1637     FIXME("Unknown modifier %u\n", mod);
1638     return mod;
1639 }
1640
1641 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1642 {
1643     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1644     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1645     char dst_name[50];
1646     char src_name[3][50];
1647     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1648             ins->ctx->reg_maps->shader_version.minor);
1649
1650     shader_arb_get_dst_param(ins, dst, dst_name);
1651     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1652
1653     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1654     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1655     {
1656         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1657     }
1658     else
1659     {
1660         struct wined3d_shader_src_param src0_copy = ins->src[0];
1661         char extra_neg;
1662
1663         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1664         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1665
1666         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1667         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1668         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1669         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1670                 dst_name, src_name[1], src_name[2]);
1671     }
1672 }
1673
1674 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1675 {
1676     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1677     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1678     char dst_name[50];
1679     char src_name[3][50];
1680
1681     shader_arb_get_dst_param(ins, dst, dst_name);
1682
1683     /* Generate input register names (with modifiers) */
1684     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1685     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1686     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1687
1688     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1689             dst_name, src_name[0], src_name[2], src_name[1]);
1690 }
1691
1692 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1693  * dst = dot2(src0, src1) + src2 */
1694 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1695 {
1696     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1697     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1698     char dst_name[50];
1699     char src_name[3][50];
1700     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1701
1702     shader_arb_get_dst_param(ins, dst, dst_name);
1703     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1704     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1705
1706     if(ctx->target_version >= NV3)
1707     {
1708         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1709         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1710         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1711                        dst_name, src_name[0], src_name[1], src_name[2]);
1712     }
1713     else if(ctx->target_version >= NV2)
1714     {
1715         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1716          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1717          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1718          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1719          *
1720          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1721          *
1722          * .xyxy and other swizzles that we could get with this are not valid in
1723          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1724          */
1725         struct wined3d_shader_src_param tmp_param = ins->src[1];
1726         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1727         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1728
1729         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1730
1731         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1732                        dst_name, src_name[2], src_name[0], src_name[1]);
1733     }
1734     else
1735     {
1736         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1737         /* Emulate a DP2 with a DP3 and 0.0. Don't use the dest as temp register, it could be src[1] or src[2]
1738         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1739         */
1740         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1741         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1742         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1743         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1744     }
1745 }
1746
1747 /* Map the opcode 1-to-1 to the GL code */
1748 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1749 {
1750     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1751     const char *instruction;
1752     char arguments[256], dst_str[50];
1753     unsigned int i;
1754     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1755
1756     switch (ins->handler_idx)
1757     {
1758         case WINED3DSIH_ABS: instruction = "ABS"; break;
1759         case WINED3DSIH_ADD: instruction = "ADD"; break;
1760         case WINED3DSIH_CRS: instruction = "XPD"; break;
1761         case WINED3DSIH_DP3: instruction = "DP3"; break;
1762         case WINED3DSIH_DP4: instruction = "DP4"; break;
1763         case WINED3DSIH_DST: instruction = "DST"; break;
1764         case WINED3DSIH_FRC: instruction = "FRC"; break;
1765         case WINED3DSIH_LIT: instruction = "LIT"; break;
1766         case WINED3DSIH_LRP: instruction = "LRP"; break;
1767         case WINED3DSIH_MAD: instruction = "MAD"; break;
1768         case WINED3DSIH_MAX: instruction = "MAX"; break;
1769         case WINED3DSIH_MIN: instruction = "MIN"; break;
1770         case WINED3DSIH_MOV: instruction = "MOV"; break;
1771         case WINED3DSIH_MUL: instruction = "MUL"; break;
1772         case WINED3DSIH_SGE: instruction = "SGE"; break;
1773         case WINED3DSIH_SLT: instruction = "SLT"; break;
1774         case WINED3DSIH_SUB: instruction = "SUB"; break;
1775         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1776         case WINED3DSIH_DSX: instruction = "DDX"; break;
1777         default: instruction = "";
1778             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1779             break;
1780     }
1781
1782     /* Note that shader_arb_add_dst_param() adds spaces. */
1783     arguments[0] = '\0';
1784     shader_arb_get_dst_param(ins, dst, dst_str);
1785     for (i = 0; i < ins->src_count; ++i)
1786     {
1787         char operand[100];
1788         strcat(arguments, ", ");
1789         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1790         strcat(arguments, operand);
1791     }
1792     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1793 }
1794
1795 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1796
1797 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1798 {
1799     const struct wined3d_shader *shader = ins->ctx->shader;
1800     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1801     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1802     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1803     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1804     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1805     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1806
1807     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1808     char src0_param[256];
1809
1810     if (ins->handler_idx == WINED3DSIH_MOVA)
1811     {
1812         const struct arb_vshader_private *shader_data = shader->backend_data;
1813         char write_mask[6];
1814         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1815
1816         if(ctx->target_version >= NV2) {
1817             shader_hw_map2gl(ins);
1818             return;
1819         }
1820         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1821         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1822
1823         /* This implements the mova formula used in GLSL. The first two instructions
1824          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1825          * in this case:
1826          * mova A0.x, 0.0
1827          *
1828          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1829          *
1830          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1831          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1832          */
1833         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1834         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1835
1836         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1837         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1838         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1839         if (shader_data->rel_offset)
1840         {
1841             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1842         }
1843         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1844
1845         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1846     }
1847     else if (reg_maps->shader_version.major == 1
1848           && !shader_is_pshader_version(reg_maps->shader_version.type)
1849           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1850     {
1851         const struct arb_vshader_private *shader_data = shader->backend_data;
1852         src0_param[0] = '\0';
1853
1854         if (shader_data->rel_offset)
1855         {
1856             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1857             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1858             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1859             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1860         }
1861         else
1862         {
1863             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1864              * with more than one component. Thus replicate the first source argument over all
1865              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1866             struct wined3d_shader_src_param tmp_src = ins->src[0];
1867             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1868             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1869             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1870         }
1871     }
1872     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1873     {
1874         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1875         {
1876             shader_addline(buffer, "#mov handled in srgb write code\n");
1877             return;
1878         }
1879         shader_hw_map2gl(ins);
1880     }
1881     else
1882     {
1883         shader_hw_map2gl(ins);
1884     }
1885 }
1886
1887 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1888 {
1889     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1890     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1891     char reg_dest[40];
1892
1893     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1894      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1895      */
1896     shader_arb_get_dst_param(ins, dst, reg_dest);
1897
1898     if (ins->ctx->reg_maps->shader_version.major >= 2)
1899     {
1900         const char *kilsrc = "TA";
1901         BOOL is_color;
1902
1903         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1904         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1905         {
1906             kilsrc = reg_dest;
1907         }
1908         else
1909         {
1910             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1911              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1912              * masked out components to 0(won't kill)
1913              */
1914             char x = '0', y = '0', z = '0', w = '0';
1915             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1916             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1917             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1918             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1919             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1920         }
1921         shader_addline(buffer, "KIL %s;\n", kilsrc);
1922     }
1923     else
1924     {
1925         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1926          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1927          *
1928          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1929          * or pass in any temporary register(in shader phase 2)
1930          */
1931         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1932             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1933         else
1934             shader_arb_get_dst_param(ins, dst, reg_dest);
1935         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1936         shader_addline(buffer, "KIL TA;\n");
1937     }
1938 }
1939
1940 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1941 {
1942     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1943     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1944     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1945             ins->ctx->reg_maps->shader_version.minor);
1946     struct wined3d_shader_src_param src;
1947
1948     char reg_dest[40];
1949     char reg_coord[40];
1950     DWORD reg_sampler_code;
1951     WORD myflags = 0;
1952     BOOL swizzle_coord = FALSE;
1953
1954     /* All versions have a destination register */
1955     shader_arb_get_dst_param(ins, dst, reg_dest);
1956
1957     /* 1.0-1.4: Use destination register number as texture code.
1958        2.0+: Use provided sampler number as texure code. */
1959     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1960         reg_sampler_code = dst->reg.idx[0].offset;
1961     else
1962         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1963
1964     /* 1.0-1.3: Use the texcoord varying.
1965        1.4+: Use provided coordinate source register. */
1966     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1967         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1968     else {
1969         /* TEX is the only instruction that can handle DW and DZ natively */
1970         src = ins->src[0];
1971         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1972         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1973         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1974     }
1975
1976     /* projection flag:
1977      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1978      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1979      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1980      */
1981     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1982     {
1983         DWORD flags = 0;
1984         if (reg_sampler_code < MAX_TEXTURES)
1985             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1986         if (flags & WINED3D_PSARGS_PROJECTED)
1987         {
1988             myflags |= TEX_PROJ;
1989             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1990                 swizzle_coord = TRUE;
1991         }
1992     }
1993     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1994     {
1995         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1996         if (src_mod == WINED3DSPSM_DZ)
1997         {
1998             swizzle_coord = TRUE;
1999             myflags |= TEX_PROJ;
2000         } else if(src_mod == WINED3DSPSM_DW) {
2001             myflags |= TEX_PROJ;
2002         }
2003     } else {
2004         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
2005         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
2006     }
2007
2008     if (swizzle_coord)
2009     {
2010         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2011          * reg_coord is a read-only varying register, so we need a temp reg */
2012         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2013         strcpy(reg_coord, "TA");
2014     }
2015
2016     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2017 }
2018
2019 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2020 {
2021     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2022     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2023     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2024             ins->ctx->reg_maps->shader_version.minor);
2025     char dst_str[50];
2026
2027     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2028     {
2029         DWORD reg = dst->reg.idx[0].offset;
2030
2031         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2032         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2033     } else {
2034         char reg_src[40];
2035
2036         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2037         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2038         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2039     }
2040 }
2041
2042 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2043 {
2044      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2045      DWORD flags = 0;
2046
2047      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2048      char dst_str[50];
2049      char src_str[50];
2050
2051      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2052      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2053      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2054      /* Move .x first in case src_str is "TA" */
2055      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2056      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2057      if (reg1 < MAX_TEXTURES)
2058      {
2059          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2060          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2061      }
2062      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2063 }
2064
2065 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2066 {
2067      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2068
2069      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2070      char dst_str[50];
2071      char src_str[50];
2072
2073      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2074      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2075      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2076      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2077      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2078      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2079 }
2080
2081 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2082 {
2083     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2084     char dst_str[50];
2085     char src_str[50];
2086
2087     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2088     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2089     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2090     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2091 }
2092
2093 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2094 {
2095     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2096     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2097     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2098     char reg_coord[40], dst_reg[50], src_reg[50];
2099     DWORD reg_dest_code;
2100
2101     /* All versions have a destination register. The Tx where the texture coordinates come
2102      * from is the varying incarnation of the texture register
2103      */
2104     reg_dest_code = dst->reg.idx[0].offset;
2105     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2106     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2107     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2108
2109     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2110      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2111      *
2112      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2113      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2114      *
2115      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2116      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2117      * extension.
2118      */
2119     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2120     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2121     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2122     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2123
2124     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2125      * so we can't let the GL handle this.
2126      */
2127     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2128             & WINED3D_PSARGS_PROJECTED)
2129     {
2130         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2131         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2132         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2133     } else {
2134         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2135     }
2136
2137     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2138
2139     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2140     {
2141         /* No src swizzles are allowed, so this is ok */
2142         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2143                        src_reg, reg_dest_code, reg_dest_code);
2144         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2145     }
2146 }
2147
2148 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2149 {
2150     DWORD reg = ins->dst[0].reg.idx[0].offset;
2151     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2152     char src0_name[50], dst_name[50];
2153     BOOL is_color;
2154     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2155
2156     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2157     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2158      * T<reg+1> register. Use this register to store the calculated vector
2159      */
2160     tmp_reg.idx[0].offset = reg + 1;
2161     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2162     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2163 }
2164
2165 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2166 {
2167     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2168     DWORD flags;
2169     DWORD reg = ins->dst[0].reg.idx[0].offset;
2170     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2171     char dst_str[50];
2172     char src0_name[50];
2173     char dst_reg[50];
2174     BOOL is_color;
2175
2176     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2177     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2178
2179     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2180     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2181     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2182     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2183     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2184 }
2185
2186 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2187 {
2188     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2189     DWORD reg = ins->dst[0].reg.idx[0].offset;
2190     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2191     char src0_name[50], dst_name[50];
2192     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2193     BOOL is_color;
2194
2195     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2196      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2197      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2198      */
2199     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2200     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2201
2202     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2203     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2204                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2205     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2206 }
2207
2208 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2209 {
2210     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2211     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2212     DWORD flags;
2213     DWORD reg = ins->dst[0].reg.idx[0].offset;
2214     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2215     char dst_str[50];
2216     char src0_name[50], dst_name[50];
2217     BOOL is_color;
2218
2219     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2220     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2221     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2222
2223     /* Sample the texture using the calculated coordinates */
2224     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2225     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2226     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2227     tex_mx->current_row = 0;
2228 }
2229
2230 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2231 {
2232     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2233     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2234     DWORD flags;
2235     DWORD reg = ins->dst[0].reg.idx[0].offset;
2236     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2237     char dst_str[50];
2238     char src0_name[50];
2239     char dst_reg[50];
2240     BOOL is_color;
2241
2242     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2243      * components for temporary data storage
2244      */
2245     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2246     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2247     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2248
2249     /* Construct the eye-ray vector from w coordinates */
2250     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2251     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2252     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2253
2254     /* Calculate reflection vector
2255      */
2256     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2257     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2258     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2259     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2260     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2261     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2262     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2263
2264     /* Sample the texture using the calculated coordinates */
2265     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2266     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2267     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2268     tex_mx->current_row = 0;
2269 }
2270
2271 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2272 {
2273     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2274     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2275     DWORD flags;
2276     DWORD reg = ins->dst[0].reg.idx[0].offset;
2277     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2278     char dst_str[50];
2279     char src0_name[50];
2280     char src1_name[50];
2281     char dst_reg[50];
2282     BOOL is_color;
2283
2284     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2285     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2286     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2287     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2288     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2289
2290     /* Calculate reflection vector.
2291      *
2292      *                   dot(N, E)
2293      * dst_reg.xyz = 2 * --------- * N - E
2294      *                   dot(N, N)
2295      *
2296      * Which normalizes the normal vector
2297      */
2298     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2299     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2300     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2301     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2302     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2303     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2304
2305     /* Sample the texture using the calculated coordinates */
2306     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2307     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2308     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2309     tex_mx->current_row = 0;
2310 }
2311
2312 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2313 {
2314     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2315     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2316     char dst_name[50];
2317     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2318     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2319
2320     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2321      * which is essentially an input, is the destination register because it is the first
2322      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2323      * here(writemasks/swizzles are not valid on texdepth)
2324      */
2325     shader_arb_get_dst_param(ins, dst, dst_name);
2326
2327     /* According to the msdn, the source register(must be r5) is unusable after
2328      * the texdepth instruction, so we're free to modify it
2329      */
2330     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2331
2332     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2333      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2334      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2335      */
2336     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2337     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2338     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2339     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2340 }
2341
2342 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2343  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2344  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2345 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2346 {
2347     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2348     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2349     char src0[50];
2350     char dst_str[50];
2351
2352     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2353     shader_addline(buffer, "MOV TB, 0.0;\n");
2354     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2355
2356     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2357     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2358 }
2359
2360 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2361  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2362 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2363 {
2364     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2365     char src0[50];
2366     char dst_str[50];
2367     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2368
2369     /* Handle output register */
2370     shader_arb_get_dst_param(ins, dst, dst_str);
2371     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2372     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2373 }
2374
2375 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2376  * Perform the 3rd row of a 3x3 matrix multiply */
2377 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2378 {
2379     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2380     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2381     char dst_str[50], dst_name[50];
2382     char src0[50];
2383     BOOL is_color;
2384
2385     shader_arb_get_dst_param(ins, dst, dst_str);
2386     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2387     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2388     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2389     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2390 }
2391
2392 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2393  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2394  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2395  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2396  */
2397 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2398 {
2399     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2400     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2401     char src0[50], dst_name[50];
2402     BOOL is_color;
2403     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2404     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2405
2406     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2407     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2408     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2409
2410     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2411      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2412      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2413      */
2414     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2415     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2416     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2417     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2418 }
2419
2420 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2421     Vertex/Pixel shaders to ARB_vertex_program codes */
2422 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2423 {
2424     int i;
2425     int nComponents = 0;
2426     struct wined3d_shader_dst_param tmp_dst = {{0}};
2427     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2428     struct wined3d_shader_instruction tmp_ins;
2429
2430     memset(&tmp_ins, 0, sizeof(tmp_ins));
2431
2432     /* Set constants for the temporary argument */
2433     tmp_ins.ctx = ins->ctx;
2434     tmp_ins.dst_count = 1;
2435     tmp_ins.dst = &tmp_dst;
2436     tmp_ins.src_count = 2;
2437     tmp_ins.src = tmp_src;
2438
2439     switch(ins->handler_idx)
2440     {
2441         case WINED3DSIH_M4x4:
2442             nComponents = 4;
2443             tmp_ins.handler_idx = WINED3DSIH_DP4;
2444             break;
2445         case WINED3DSIH_M4x3:
2446             nComponents = 3;
2447             tmp_ins.handler_idx = WINED3DSIH_DP4;
2448             break;
2449         case WINED3DSIH_M3x4:
2450             nComponents = 4;
2451             tmp_ins.handler_idx = WINED3DSIH_DP3;
2452             break;
2453         case WINED3DSIH_M3x3:
2454             nComponents = 3;
2455             tmp_ins.handler_idx = WINED3DSIH_DP3;
2456             break;
2457         case WINED3DSIH_M3x2:
2458             nComponents = 2;
2459             tmp_ins.handler_idx = WINED3DSIH_DP3;
2460             break;
2461         default:
2462             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2463             break;
2464     }
2465
2466     tmp_dst = ins->dst[0];
2467     tmp_src[0] = ins->src[0];
2468     tmp_src[1] = ins->src[1];
2469     for (i = 0; i < nComponents; ++i)
2470     {
2471         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2472         shader_hw_map2gl(&tmp_ins);
2473         ++tmp_src[1].reg.idx[0].offset;
2474     }
2475 }
2476
2477 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2478 {
2479     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2480
2481     char dst[50];
2482     char src[50];
2483
2484     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2485     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2486     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2487     {
2488         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2489          * .w is used
2490          */
2491         strcat(src, ".w");
2492     }
2493
2494     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2495 }
2496
2497 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2498 {
2499     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2500     const char *instruction;
2501
2502     char dst[50];
2503     char src[50];
2504
2505     switch(ins->handler_idx)
2506     {
2507         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2508         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2509         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2510         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2511         default: instruction = "";
2512             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2513             break;
2514     }
2515
2516     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2517     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2518     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2519     {
2520         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2521          * .w is used
2522          */
2523         strcat(src, ".w");
2524     }
2525
2526     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2527 }
2528
2529 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2530 {
2531     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2532     char dst_name[50];
2533     char src_name[50];
2534     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2535     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2536     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2537
2538     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2539     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2540
2541     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2542      * otherwise NRM or RSQ would return NaN */
2543     if(pshader && priv->target_version >= NV3)
2544     {
2545         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2546          *
2547          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2548          */
2549         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2550         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2551         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2552     }
2553     else if(priv->target_version >= NV2)
2554     {
2555         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2556         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2557         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2558                        src_name);
2559     }
2560     else
2561     {
2562         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2563
2564         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2565         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2566          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2567          */
2568         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2569         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2570
2571         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2572         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2573         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2574                     src_name);
2575     }
2576 }
2577
2578 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2579 {
2580     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2581     char dst_name[50];
2582     char src_name[3][50];
2583
2584     /* ARB_fragment_program has a convenient LRP instruction */
2585     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2586         shader_hw_map2gl(ins);
2587         return;
2588     }
2589
2590     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2591     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2592     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2593     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2594
2595     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2596     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2597                    dst_name, src_name[0], src_name[2]);
2598 }
2599
2600 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2601 {
2602     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2603      * must contain fixed constants. So we need a separate function to filter those constants and
2604      * can't use map2gl
2605      */
2606     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2607     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2608     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2609     char dst_name[50];
2610     char src_name0[50], src_name1[50], src_name2[50];
2611     BOOL is_color;
2612
2613     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2614     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2615         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2616         /* No modifiers are supported on SCS */
2617         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2618
2619         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2620         {
2621             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2622             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2623         }
2624     } else if(priv->target_version >= NV2) {
2625         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2626
2627         /* Sincos writemask must be .x, .y or .xy */
2628         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2629             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2630         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2631             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2632     } else {
2633         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2634          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2635          *
2636          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2637          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2638          *
2639          * The constants we get are:
2640          *
2641          *  +1   +1,     -1     -1     +1      +1      -1       -1
2642          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2643          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2644          *
2645          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2646          *
2647          * (x/2)^2 = x^2 / 4
2648          * (x/2)^3 = x^3 / 8
2649          * (x/2)^4 = x^4 / 16
2650          * (x/2)^5 = x^5 / 32
2651          * etc
2652          *
2653          * To get the final result:
2654          * sin(x) = 2 * sin(x/2) * cos(x/2)
2655          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2656          * (from sin(x+y) and cos(x+y) rules)
2657          *
2658          * As per MSDN, dst.z is undefined after the operation, and so is
2659          * dst.x and dst.y if they're masked out by the writemask. Ie
2660          * sincos dst.y, src1, c0, c1
2661          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2662          * vsa.exe also stops with an error if the dest register is the same register as the source
2663          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2664          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2665          */
2666         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2667         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2668         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2669
2670         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2671         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2672         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2673         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2674         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2675         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2676
2677         /* sin(x/2)
2678          *
2679          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2680          * properly merge that with MULs in the code above?
2681          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2682          * we can merge the sine and cosine MAD rows to calculate them together.
2683          */
2684         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2685         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2686         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2687         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2688
2689         /* cos(x/2) */
2690         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2691         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2692         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2693
2694         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2695             /* cos x */
2696             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2697             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2698         }
2699         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2700             /* sin x */
2701             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2702             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2703         }
2704     }
2705 }
2706
2707 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2708 {
2709     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2710     char dst_name[50];
2711     char src_name[50];
2712     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2713
2714     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2715     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2716
2717     /* SGN is only valid in vertex shaders */
2718     if(ctx->target_version >= NV2) {
2719         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2720         return;
2721     }
2722
2723     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2724      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2725      */
2726     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2727         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2728     } else {
2729         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2730          * Then use TA, and calculate the final result
2731          *
2732          * Not reading from TA? Store the first result in TA to avoid overwriting the
2733          * destination if src reg = dst reg
2734          */
2735         if(strstr(src_name, "TA"))
2736         {
2737             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2738             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2739             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2740         }
2741         else
2742         {
2743             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2744             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2745             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2746         }
2747     }
2748 }
2749
2750 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2751 {
2752     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2753     char src[50];
2754     char dst[50];
2755     char dst_name[50];
2756     BOOL is_color;
2757
2758     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2759     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2760     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2761
2762     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2763     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2764 }
2765
2766 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2767 {
2768     *need_abs = FALSE;
2769
2770     switch(mod)
2771     {
2772         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2773         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2774         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2775         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2776         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2777         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2778         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2779         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2780         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2781         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2782         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2783         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2784         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2785     }
2786     FIXME("Unknown modifier %u\n", mod);
2787     return mod;
2788 }
2789
2790 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2791 {
2792     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2793     char src0[50], dst[50];
2794     struct wined3d_shader_src_param src0_copy = ins->src[0];
2795     BOOL need_abs = FALSE;
2796     const char *instr;
2797
2798     switch(ins->handler_idx)
2799     {
2800         case WINED3DSIH_LOG:  instr = "LG2"; break;
2801         case WINED3DSIH_LOGP: instr = "LOG"; break;
2802         default:
2803             ERR("Unexpected instruction %d\n", ins->handler_idx);
2804             return;
2805     }
2806
2807     /* LOG and LOGP operate on the absolute value of the input */
2808     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2809
2810     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2811     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2812
2813     if(need_abs)
2814     {
2815         shader_addline(buffer, "ABS TA, %s;\n", src0);
2816         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2817     }
2818     else
2819     {
2820         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2821     }
2822 }
2823
2824 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2825 {
2826     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2827     char src0[50], src1[50], dst[50];
2828     struct wined3d_shader_src_param src0_copy = ins->src[0];
2829     BOOL need_abs = FALSE;
2830     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2831     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2832
2833     /* POW operates on the absolute value of the input */
2834     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2835
2836     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2837     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2838     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2839
2840     if (need_abs)
2841         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2842     else
2843         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2844
2845     if (priv->target_version >= NV2)
2846     {
2847         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2848         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2849         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2850     }
2851     else
2852     {
2853         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2854         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2855
2856         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2857         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2858         /* Possibly add flt_eps to avoid getting float special values */
2859         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2860         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2861         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2862         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2863     }
2864 }
2865
2866 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2867 {
2868     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2869     char src_name[50];
2870     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2871
2872     /* src0 is aL */
2873     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2874
2875     if(vshader)
2876     {
2877         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2878         struct list *e = list_head(&priv->control_frames);
2879         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2880
2881         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2882         /* The constant loader makes sure to load -1 into iX.w */
2883         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2884         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2885         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2886     }
2887     else
2888     {
2889         shader_addline(buffer, "LOOP %s;\n", src_name);
2890     }
2891 }
2892
2893 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2894 {
2895     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2896     char src_name[50];
2897     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2898
2899     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2900
2901     /* The constant loader makes sure to load -1 into iX.w */
2902     if(vshader)
2903     {
2904         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2905         struct list *e = list_head(&priv->control_frames);
2906         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2907
2908         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2909
2910         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2911         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2912         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2913     }
2914     else
2915     {
2916         shader_addline(buffer, "REP %s;\n", src_name);
2917     }
2918 }
2919
2920 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2921 {
2922     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2923     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2924
2925     if(vshader)
2926     {
2927         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2928         struct list *e = list_head(&priv->control_frames);
2929         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2930
2931         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2932         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2933         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2934
2935         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2936     }
2937     else
2938     {
2939         shader_addline(buffer, "ENDLOOP;\n");
2940     }
2941 }
2942
2943 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2944 {
2945     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2946     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2947
2948     if(vshader)
2949     {
2950         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2951         struct list *e = list_head(&priv->control_frames);
2952         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2953
2954         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2955         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2956         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2957
2958         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2959     }
2960     else
2961     {
2962         shader_addline(buffer, "ENDREP;\n");
2963     }
2964 }
2965
2966 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2967 {
2968     struct control_frame *control_frame;
2969
2970     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2971     {
2972         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2973     }
2974     ERR("Could not find loop for break\n");
2975     return NULL;
2976 }
2977
2978 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2979 {
2980     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2981     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2982     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2983
2984     if(vshader)
2985     {
2986         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2987     }
2988     else
2989     {
2990         shader_addline(buffer, "BRK;\n");
2991     }
2992 }
2993
2994 static const char *get_compare(enum wined3d_shader_rel_op op)
2995 {
2996     switch (op)
2997     {
2998         case WINED3D_SHADER_REL_OP_GT: return "GT";
2999         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
3000         case WINED3D_SHADER_REL_OP_GE: return "GE";
3001         case WINED3D_SHADER_REL_OP_LT: return "LT";
3002         case WINED3D_SHADER_REL_OP_NE: return "NE";
3003         case WINED3D_SHADER_REL_OP_LE: return "LE";
3004         default:
3005             FIXME("Unrecognized operator %#x.\n", op);
3006             return "(\?\?)";
3007     }
3008 }
3009
3010 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3011 {
3012     switch (op)
3013     {
3014         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3015         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3016         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3017         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3018         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3019         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3020         default:
3021             FIXME("Unrecognized operator %#x.\n", op);
3022             return -1;
3023     }
3024 }
3025
3026 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3027 {
3028     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3029     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3030     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3031     char src_name0[50];
3032     char src_name1[50];
3033     const char *comp = get_compare(ins->flags);
3034
3035     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3036     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3037
3038     if(vshader)
3039     {
3040         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3041          * away the subtraction result
3042          */
3043         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3044         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3045     }
3046     else
3047     {
3048         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3049         shader_addline(buffer, "BRK (%s.x);\n", comp);
3050     }
3051 }
3052
3053 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3054 {
3055     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3056     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3057     struct list *e = list_head(&priv->control_frames);
3058     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3059     const char *comp;
3060     char src_name0[50];
3061     char src_name1[50];
3062     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3063
3064     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3065     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3066
3067     if(vshader)
3068     {
3069         /* Invert the flag. We jump to the else label if the condition is NOT true */
3070         comp = get_compare(invert_compare(ins->flags));
3071         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3072         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3073     }
3074     else
3075     {
3076         comp = get_compare(ins->flags);
3077         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3078         shader_addline(buffer, "IF %s.x;\n", comp);
3079     }
3080 }
3081
3082 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3083 {
3084     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3085     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3086     struct list *e = list_head(&priv->control_frames);
3087     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3088     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3089
3090     if(vshader)
3091     {
3092         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3093         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3094         control_frame->had_else = TRUE;
3095     }
3096     else
3097     {
3098         shader_addline(buffer, "ELSE;\n");
3099     }
3100 }
3101
3102 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3103 {
3104     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3105     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3106     struct list *e = list_head(&priv->control_frames);
3107     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3108     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3109
3110     if(vshader)
3111     {
3112         if(control_frame->had_else)
3113         {
3114             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3115         }
3116         else
3117         {
3118             shader_addline(buffer, "#No else branch. else is endif\n");
3119             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3120         }
3121     }
3122     else
3123     {
3124         shader_addline(buffer, "ENDIF;\n");
3125     }
3126 }
3127
3128 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3129 {
3130     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3131     char reg_dest[40];
3132     char reg_src[3][40];
3133     WORD flags = TEX_DERIV;
3134
3135     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3136     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3137     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3138     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3139
3140     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3141     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3142
3143     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3144 }
3145
3146 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3147 {
3148     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3149     char reg_dest[40];
3150     char reg_coord[40];
3151     WORD flags = TEX_LOD;
3152
3153     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3154     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3155
3156     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3157     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3158
3159     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3160 }
3161
3162 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3163 {
3164     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3165     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3166
3167     priv->in_main_func = FALSE;
3168     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3169      * subroutine, don't generate a label that will make GL complain
3170      */
3171     if(priv->target_version == ARB) return;
3172
3173     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3174 }
3175
3176 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3177         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3178         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3179         struct wined3d_shader_buffer *buffer)
3180 {
3181     unsigned int i;
3182
3183     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3184      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3185      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3186      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3187      */
3188     if (args->super.fog_src == VS_FOG_Z)
3189     {
3190         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3191     }
3192     else
3193     {
3194         if (!reg_maps->fog)
3195         {
3196             /* posFixup.x is always 1.0, so we can safely use it */
3197             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3198         }
3199         else
3200         {
3201             /* Clamp fogcoord */
3202             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3203             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3204
3205             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3206             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3207         }
3208     }
3209
3210     /* Clipplanes are always stored without y inversion */
3211     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3212     {
3213         if (args->super.clip_enabled)
3214         {
3215             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3216             {
3217                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3218             }
3219         }
3220     }
3221     else if (args->clip.boolclip.clip_texcoord)
3222     {
3223         unsigned int cur_clip = 0;
3224         char component[4] = {'x', 'y', 'z', 'w'};
3225         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3226
3227         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3228         {
3229             if (args->clip.boolclip.clipplane_mask & (1 << i))
3230             {
3231                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3232                                component[cur_clip++], i);
3233             }
3234         }
3235         switch (cur_clip)
3236         {
3237             case 0:
3238                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3239                 break;
3240             case 1:
3241                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3242                 break;
3243             case 2:
3244                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3245                 break;
3246             case 3:
3247                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3248                 break;
3249         }
3250         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3251                        args->clip.boolclip.clip_texcoord - 1);
3252     }
3253
3254     /* Write the final position.
3255      *
3256      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3257      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3258      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3259      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3260      */
3261     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3262     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3263     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3264
3265     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3266      * and the glsl equivalent
3267      */
3268     if (need_helper_const(shader_data, reg_maps, gl_info))
3269     {
3270         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3271         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3272     }
3273     else
3274     {
3275         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3276         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3277     }
3278
3279     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3280
3281     priv_ctx->footer_written = TRUE;
3282 }
3283
3284 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3285 {
3286     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3287     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3288     const struct wined3d_shader *shader = ins->ctx->shader;
3289     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3290
3291     if(priv->target_version == ARB) return;
3292
3293     if(vshader)
3294     {
3295         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3296                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3297     }
3298
3299     shader_addline(buffer, "RET;\n");
3300 }
3301
3302 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3303 {
3304     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3305     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3306 }
3307
3308 /* Context activation is done by the caller. */
3309 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3310 {
3311     GLuint program_id = 0;
3312     GLint pos;
3313
3314     const char *blt_vprogram =
3315         "!!ARBvp1.0\n"
3316         "PARAM c[1] = { { 1, 0.5 } };\n"
3317         "MOV result.position, vertex.position;\n"
3318         "MOV result.color, c[0].x;\n"
3319         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3320         "END\n";
3321
3322     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3323     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3324     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3325             strlen(blt_vprogram), blt_vprogram));
3326     checkGLcall("glProgramStringARB()");
3327
3328     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3329     if (pos != -1)
3330     {
3331         FIXME("Vertex program error at position %d: %s\n\n", pos,
3332             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3333         shader_arb_dump_program_source(blt_vprogram);
3334     }
3335     else
3336     {
3337         GLint native;
3338
3339         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3340         checkGLcall("glGetProgramivARB()");
3341         if (!native) WARN("Program exceeds native resource limits.\n");
3342     }
3343
3344     return program_id;
3345 }
3346
3347 /* Context activation is done by the caller. */
3348 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3349         enum tex_types tex_type, BOOL masked)
3350 {
3351     GLuint program_id = 0;
3352     const char *fprogram;
3353     GLint pos;
3354
3355     static const char * const blt_fprograms_full[tex_type_count] =
3356     {
3357         /* tex_1d */
3358         NULL,
3359         /* tex_2d */
3360         "!!ARBfp1.0\n"
3361         "TEMP R0;\n"
3362         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3363         "MOV result.depth.z, R0.x;\n"
3364         "END\n",
3365         /* tex_3d */
3366         NULL,
3367         /* tex_cube */
3368         "!!ARBfp1.0\n"
3369         "TEMP R0;\n"
3370         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3371         "MOV result.depth.z, R0.x;\n"
3372         "END\n",
3373         /* tex_rect */
3374         "!!ARBfp1.0\n"
3375         "TEMP R0;\n"
3376         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3377         "MOV result.depth.z, R0.x;\n"
3378         "END\n",
3379     };
3380
3381     static const char * const blt_fprograms_masked[tex_type_count] =
3382     {
3383         /* tex_1d */
3384         NULL,
3385         /* tex_2d */
3386         "!!ARBfp1.0\n"
3387         "PARAM mask = program.local[0];\n"
3388         "TEMP R0;\n"
3389         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3390         "MUL R0.x, R0.x, R0.y;\n"
3391         "KIL -R0.x;\n"
3392         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3393         "MOV result.depth.z, R0.x;\n"
3394         "END\n",
3395         /* tex_3d */
3396         NULL,
3397         /* tex_cube */
3398         "!!ARBfp1.0\n"
3399         "PARAM mask = program.local[0];\n"
3400         "TEMP R0;\n"
3401         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3402         "MUL R0.x, R0.x, R0.y;\n"
3403         "KIL -R0.x;\n"
3404         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3405         "MOV result.depth.z, R0.x;\n"
3406         "END\n",
3407         /* tex_rect */
3408         "!!ARBfp1.0\n"
3409         "PARAM mask = program.local[0];\n"
3410         "TEMP R0;\n"
3411         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3412         "MUL R0.x, R0.x, R0.y;\n"
3413         "KIL -R0.x;\n"
3414         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3415         "MOV result.depth.z, R0.x;\n"
3416         "END\n",
3417     };
3418
3419     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3420     if (!fprogram)
3421     {
3422         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3423         tex_type = tex_2d;
3424         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3425     }
3426
3427     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3428     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3429     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3430     checkGLcall("glProgramStringARB()");
3431
3432     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3433     if (pos != -1)
3434     {
3435         FIXME("Fragment program error at position %d: %s\n\n", pos,
3436             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3437         shader_arb_dump_program_source(fprogram);
3438     }
3439     else
3440     {
3441         GLint native;
3442
3443         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3444         checkGLcall("glGetProgramivARB()");
3445         if (!native) WARN("Program exceeds native resource limits.\n");
3446     }
3447
3448     return program_id;
3449 }
3450
3451 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3452         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3453 {
3454     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3455
3456     if(condcode)
3457     {
3458         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3459         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3460         /* Calculate the > 0.0031308 case */
3461         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3462         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3463         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3464         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3465         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3466         /* Calculate the < case */
3467         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3468     }
3469     else
3470     {
3471         /* Calculate the > 0.0031308 case */
3472         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3473         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3474         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3475         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3476         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3477         /* Calculate the < case */
3478         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3479         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3480         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3481         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3482         /* Store the components > 0.0031308 in the destination */
3483         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3484         /* Add the components that are < 0.0031308 */
3485         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3486         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3487         * result.color writes(.rgb first, then .a), or handle overwriting already written
3488         * components. The assembler uses a temporary register in this case, which is usually
3489         * not allocated from one of our registers that were used earlier.
3490         */
3491     }
3492     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3493 }
3494
3495 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3496 {
3497     const struct wined3d_shader_lconst *constant;
3498
3499     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3500     {
3501         if (constant->idx == idx)
3502         {
3503             return constant->value;
3504         }
3505     }
3506     return NULL;
3507 }
3508
3509 static void init_ps_input(const struct wined3d_shader *shader,
3510         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3511 {
3512     static const char * const texcoords[8] =
3513     {
3514         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3515         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3516     };
3517     unsigned int i;
3518     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3519     const char *semantic_name;
3520     DWORD semantic_idx;
3521
3522     switch(args->super.vp_mode)
3523     {
3524         case pretransformed:
3525         case fixedfunction:
3526             /* The pixelshader has to collect the varyings on its own. In any case properly load
3527              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3528              * other attribs to 0.0.
3529              *
3530              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3531              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3532              * load the texcoord attrib pointers to match the pixel shader signature
3533              */
3534             for(i = 0; i < MAX_REG_INPUT; i++)
3535             {
3536                 semantic_name = sig[i].semantic_name;
3537                 semantic_idx = sig[i].semantic_idx;
3538                 if (!semantic_name) continue;
3539
3540                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3541                 {
3542                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3543                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3544                     else priv->ps_input[i] = "0.0";
3545                 }
3546                 else if(args->super.vp_mode == fixedfunction)
3547                 {
3548                     priv->ps_input[i] = "0.0";
3549                 }
3550                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3551                 {
3552                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3553                     else priv->ps_input[i] = "0.0";
3554                 }
3555                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3556                 {
3557                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3558                     else priv->ps_input[i] = "0.0";
3559                 }
3560                 else
3561                 {
3562                     priv->ps_input[i] = "0.0";
3563                 }
3564
3565                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3566             }
3567             break;
3568
3569         case vertexshader:
3570             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3571              * fragment.color
3572              */
3573             for(i = 0; i < 8; i++)
3574             {
3575                 priv->ps_input[i] = texcoords[i];
3576             }
3577             priv->ps_input[8] = "fragment.color.primary";
3578             priv->ps_input[9] = "fragment.color.secondary";
3579             break;
3580     }
3581 }
3582
3583 /* Context activation is done by the caller. */
3584 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3585         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3586         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3587 {
3588     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3589     const struct wined3d_shader_lconst *lconst;
3590     const DWORD *function = shader->function;
3591     GLuint retval;
3592     char fragcolor[16];
3593     DWORD *lconst_map = local_const_mapping(shader), next_local;
3594     struct shader_arb_ctx_priv priv_ctx;
3595     BOOL dcl_td = FALSE;
3596     BOOL want_nv_prog = FALSE;
3597     struct arb_pshader_private *shader_priv = shader->backend_data;
3598     GLint errPos;
3599     DWORD map;
3600
3601     char srgbtmp[4][4];
3602     unsigned int i, found = 0;
3603
3604     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3605     {
3606         if (!(map & 1)
3607                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3608                 || (reg_maps->shader_version.major < 2 && !i))
3609             continue;
3610
3611         sprintf(srgbtmp[found], "R%u", i);
3612         ++found;
3613         if (found == 4) break;
3614     }
3615
3616     switch(found) {
3617         case 0:
3618             sprintf(srgbtmp[0], "TA");
3619             sprintf(srgbtmp[1], "TB");
3620             sprintf(srgbtmp[2], "TC");
3621             sprintf(srgbtmp[3], "TD");
3622             dcl_td = TRUE;
3623             break;
3624         case 1:
3625             sprintf(srgbtmp[1], "TA");
3626             sprintf(srgbtmp[2], "TB");
3627             sprintf(srgbtmp[3], "TC");
3628             break;
3629         case 2:
3630             sprintf(srgbtmp[2], "TA");
3631             sprintf(srgbtmp[3], "TB");
3632             break;
3633         case 3:
3634             sprintf(srgbtmp[3], "TA");
3635             break;
3636         case 4:
3637             break;
3638     }
3639
3640     /*  Create the hw ARB shader */
3641     memset(&priv_ctx, 0, sizeof(priv_ctx));
3642     priv_ctx.cur_ps_args = args;
3643     priv_ctx.compiled_fprog = compiled;
3644     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3645     init_ps_input(shader, args, &priv_ctx);
3646     list_init(&priv_ctx.control_frames);
3647
3648     /* Avoid enabling NV_fragment_program* if we do not need it.
3649      *
3650      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3651      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3652      * is faster than what we gain from using higher native instructions. There are some things though
3653      * that cannot be emulated. In that case enable the extensions.
3654      * If the extension is enabled, instruction handlers that support both ways will use it.
3655      *
3656      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3657      * So enable the best we can get.
3658      */
3659     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3660        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3661     {
3662         want_nv_prog = TRUE;
3663     }
3664
3665     shader_addline(buffer, "!!ARBfp1.0\n");
3666     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3667     {
3668         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3669         priv_ctx.target_version = NV3;
3670     }
3671     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3672     {
3673         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3674         priv_ctx.target_version = NV2;
3675     } else {
3676         if(want_nv_prog)
3677         {
3678             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3679              * limits properly
3680              */
3681             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3682             ERR("Try GLSL\n");
3683         }
3684         priv_ctx.target_version = ARB;
3685     }
3686
3687     if (reg_maps->rt_mask > 1)
3688     {
3689         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3690     }
3691
3692     if (reg_maps->shader_version.major < 3)
3693     {
3694         switch(args->super.fog) {
3695             case FOG_OFF:
3696                 break;
3697             case FOG_LINEAR:
3698                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3699                 break;
3700             case FOG_EXP:
3701                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3702                 break;
3703             case FOG_EXP2:
3704                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3705                 break;
3706         }
3707     }
3708
3709     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3710      * unused temps away(but occupies them for the whole shader if they're used once). Always
3711      * declaring them avoids tricky bookkeeping work
3712      */
3713     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3714     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3715     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3716     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3717     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3718     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3719     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3720
3721     if (reg_maps->shader_version.major < 2)
3722     {
3723         strcpy(fragcolor, "R0");
3724     }
3725     else
3726     {
3727         if (args->super.srgb_correction)
3728         {
3729             if (shader->u.ps.color0_mov)
3730             {
3731                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3732             }
3733             else
3734             {
3735                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3736                 strcpy(fragcolor, "TMP_COLOR");
3737             }
3738         } else {
3739             strcpy(fragcolor, "result.color");
3740         }
3741     }
3742
3743     if(args->super.srgb_correction) {
3744         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3745                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3746         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3747                        srgb_sub_high, 0.0, 0.0, 0.0);
3748     }
3749
3750     /* Base Declarations */
3751     next_local = shader_generate_arb_declarations(shader, reg_maps,
3752             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3753
3754     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3755     {
3756         unsigned char bump_const;
3757
3758         if (!(map & 1)) continue;
3759
3760         bump_const = compiled->numbumpenvmatconsts;
3761         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3762         compiled->bumpenvmatconst[bump_const].texunit = i;
3763         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3764         compiled->luminanceconst[bump_const].texunit = i;
3765
3766         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3767          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3768          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3769          * textures due to conditional NP2 restrictions)
3770          *
3771          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3772          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3773          * their location is shader dependent anyway and they cannot be loaded globally.
3774          */
3775         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3776         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3777                        i, compiled->bumpenvmatconst[bump_const].const_num);
3778         compiled->numbumpenvmatconsts = bump_const + 1;
3779
3780         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3781
3782         compiled->luminanceconst[bump_const].const_num = next_local++;
3783         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3784                        i, compiled->luminanceconst[bump_const].const_num);
3785     }
3786
3787     for(i = 0; i < MAX_CONST_I; i++)
3788     {
3789         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3790         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3791         {
3792             const DWORD *control_values = find_loop_control_values(shader, i);
3793
3794             if(control_values)
3795             {
3796                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3797                                 control_values[0], control_values[1], control_values[2]);
3798             }
3799             else
3800             {
3801                 compiled->int_consts[i] = next_local;
3802                 compiled->num_int_consts++;
3803                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3804             }
3805         }
3806     }
3807
3808     if(reg_maps->vpos || reg_maps->usesdsy)
3809     {
3810         compiled->ycorrection = next_local;
3811         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3812
3813         if(reg_maps->vpos)
3814         {
3815             shader_addline(buffer, "TEMP vpos;\n");
3816             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3817              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3818              * ycorrection.z: 1.0
3819              * ycorrection.w: 0.0
3820              */
3821             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3822             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3823         }
3824     }
3825     else
3826     {
3827         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3828     }
3829
3830     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3831      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3832      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3833      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3834      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3835      * shader compilation errors and the subsequent errors when drawing with this shader. */
3836     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3837         unsigned char cur_fixup_sampler = 0;
3838
3839         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3840         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3841         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3842
3843         fixup->offset = next_local;
3844         fixup->super.active = 0;
3845
3846         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3847             if (!(map & (1 << i))) continue;
3848
3849             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3850                 fixup->super.active |= (1 << i);
3851                 fixup->super.idx[i] = cur_fixup_sampler++;
3852             } else {
3853                 FIXME("No free constant found to load NP2 fixup data into shader. "
3854                       "Sampling from this texture will probably look wrong.\n");
3855                 break;
3856             }
3857         }
3858
3859         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3860         if (fixup->super.num_consts) {
3861             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3862                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3863         }
3864     }
3865
3866     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3867     {
3868         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3869     }
3870
3871     /* Base Shader Body */
3872     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3873
3874     if(args->super.srgb_correction) {
3875         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3876                                   priv_ctx.target_version >= NV2);
3877     }
3878
3879     if(strcmp(fragcolor, "result.color")) {
3880         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3881     }
3882     shader_addline(buffer, "END\n");
3883
3884     /* TODO: change to resource.glObjectHandle or something like that */
3885     GL_EXTCALL(glGenProgramsARB(1, &retval));
3886
3887     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3888     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3889
3890     TRACE("Created hw pixel shader, prg=%d\n", retval);
3891     /* Create the program and check for errors */
3892     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3893                buffer->content_size, buffer->buffer));
3894     checkGLcall("glProgramStringARB()");
3895
3896     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3897     if (errPos != -1)
3898     {
3899         FIXME("HW PixelShader Error at position %d: %s\n\n",
3900               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3901         shader_arb_dump_program_source(buffer->buffer);
3902         retval = 0;
3903     }
3904     else
3905     {
3906         GLint native;
3907
3908         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3909         checkGLcall("glGetProgramivARB()");
3910         if (!native) WARN("Program exceeds native resource limits.\n");
3911     }
3912
3913     /* Load immediate constants */
3914     if (lconst_map)
3915     {
3916         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3917         {
3918             const float *value = (const float *)lconst->value;
3919             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3920             checkGLcall("glProgramLocalParameter4fvARB");
3921         }
3922         HeapFree(GetProcessHeap(), 0, lconst_map);
3923     }
3924
3925     return retval;
3926 }
3927
3928 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3929 {
3930     unsigned int i;
3931     int ret;
3932
3933     for(i = 0; i < MAX_REG_INPUT; i++)
3934     {
3935         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3936         {
3937             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3938             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3939             continue;
3940         }
3941
3942         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3943         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3944         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3945         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3946         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3947         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3948     }
3949     return 0;
3950 }
3951
3952 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3953 {
3954     struct wined3d_shader_signature_element *new;
3955     int i;
3956     char *name;
3957
3958     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3959     for(i = 0; i < MAX_REG_INPUT; i++)
3960     {
3961         if (!sig[i].semantic_name) continue;
3962
3963         new[i] = sig[i];
3964         /* Clone the semantic string */
3965         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3966         strcpy(name, sig[i].semantic_name);
3967         new[i].semantic_name = name;
3968     }
3969     return new;
3970 }
3971
3972 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3973 {
3974     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3975     struct ps_signature *found_sig;
3976
3977     if (entry)
3978     {
3979         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3980         TRACE("Found existing signature %u\n", found_sig->idx);
3981         return found_sig->idx;
3982     }
3983     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3984     found_sig->sig = clone_sig(sig);
3985     found_sig->idx = priv->ps_sig_number++;
3986     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3987     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3988     {
3989         ERR("Failed to insert program entry.\n");
3990     }
3991     return found_sig->idx;
3992 }
3993
3994 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3995         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3996 {
3997     unsigned int i, j;
3998     static const char * const texcoords[8] =
3999     {
4000         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4001         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
4002     };
4003     struct wined3d_device *device = shader->device;
4004     const struct wined3d_shader_signature_element *sig;
4005     const char *semantic_name;
4006     DWORD semantic_idx, reg_idx;
4007
4008     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
4009      * and varying 9 to result.color.secondary
4010      */
4011     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
4012     {
4013         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4014         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
4015         "result.color.primary", "result.color.secondary"
4016     };
4017
4018     if(sig_num == ~0)
4019     {
4020         TRACE("Pixel shader uses builtin varyings\n");
4021         /* Map builtins to builtins */
4022         for(i = 0; i < 8; i++)
4023         {
4024             priv_ctx->texcrd_output[i] = texcoords[i];
4025         }
4026         priv_ctx->color_output[0] = "result.color.primary";
4027         priv_ctx->color_output[1] = "result.color.secondary";
4028         priv_ctx->fog_output = "TMP_FOGCOORD";
4029
4030         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
4031         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4032         {
4033             semantic_name = shader->output_signature[i].semantic_name;
4034             if (!semantic_name) continue;
4035
4036             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4037             {
4038                 TRACE("o%u is TMP_OUT\n", i);
4039                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4040                 else priv_ctx->vs_output[i] = "TA";
4041             }
4042             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4043             {
4044                 TRACE("o%u is result.pointsize\n", i);
4045                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4046                 else priv_ctx->vs_output[i] = "TA";
4047             }
4048             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4049             {
4050                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4051                 if (!shader->output_signature[i].semantic_idx)
4052                     priv_ctx->vs_output[i] = "result.color.primary";
4053                 else if (shader->output_signature[i].semantic_idx == 1)
4054                     priv_ctx->vs_output[i] = "result.color.secondary";
4055                 else priv_ctx->vs_output[i] = "TA";
4056             }
4057             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4058             {
4059                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4060                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4061                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4062             }
4063             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4064             {
4065                 TRACE("o%u is result.fogcoord\n", i);
4066                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4067                 else priv_ctx->vs_output[i] = "result.fogcoord";
4068             }
4069             else
4070             {
4071                 priv_ctx->vs_output[i] = "TA";
4072             }
4073         }
4074         return;
4075     }
4076
4077     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4078      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4079      */
4080     sig = device->stateBlock->state.pixel_shader->input_signature;
4081     TRACE("Pixel shader uses declared varyings\n");
4082
4083     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4084     for(i = 0; i < 8; i++)
4085     {
4086         priv_ctx->texcrd_output[i] = "TA";
4087     }
4088     priv_ctx->color_output[0] = "TA";
4089     priv_ctx->color_output[1] = "TA";
4090     priv_ctx->fog_output = "TA";
4091
4092     for(i = 0; i < MAX_REG_INPUT; i++)
4093     {
4094         semantic_name = sig[i].semantic_name;
4095         semantic_idx = sig[i].semantic_idx;
4096         reg_idx = sig[i].register_idx;
4097         if (!semantic_name) continue;
4098
4099         /* If a declared input register is not written by builtin arguments, don't write to it.
4100          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4101          *
4102          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4103          * to TMP_OUT in any case
4104          */
4105         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4106         {
4107             if (semantic_idx < 8)
4108                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4109         }
4110         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4111         {
4112             if (semantic_idx < 2)
4113                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4114         }
4115         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4116         {
4117             if (!semantic_idx)
4118                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4119         }
4120         else
4121         {
4122             continue;
4123         }
4124
4125         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4126                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4127         {
4128             compiled->need_color_unclamp = TRUE;
4129         }
4130     }
4131
4132     /* Map declared to declared */
4133     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4134     {
4135         /* Write unread output to TA to throw them away */
4136         priv_ctx->vs_output[i] = "TA";
4137         semantic_name = shader->output_signature[i].semantic_name;
4138         if (!semantic_name) continue;
4139
4140         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4141                 && !shader->output_signature[i].semantic_idx)
4142         {
4143             priv_ctx->vs_output[i] = "TMP_OUT";
4144             continue;
4145         }
4146         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4147                 && !shader->output_signature[i].semantic_idx)
4148         {
4149             priv_ctx->vs_output[i] = "result.pointsize";
4150             continue;
4151         }
4152
4153         for(j = 0; j < MAX_REG_INPUT; j++)
4154         {
4155             if (!sig[j].semantic_name) continue;
4156
4157             if (!strcmp(sig[j].semantic_name, semantic_name)
4158                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4159             {
4160                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4161
4162                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4163                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4164                 {
4165                     compiled->need_color_unclamp = TRUE;
4166                 }
4167             }
4168         }
4169     }
4170 }
4171
4172 /* Context activation is done by the caller. */
4173 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4174         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4175         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4176 {
4177     const struct arb_vshader_private *shader_data = shader->backend_data;
4178     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4179     struct shader_arb_priv *priv = shader->device->shader_priv;
4180     const struct wined3d_shader_lconst *lconst;
4181     const DWORD *function = shader->function;
4182     GLuint ret;
4183     DWORD next_local, *lconst_map = local_const_mapping(shader);
4184     struct shader_arb_ctx_priv priv_ctx;
4185     unsigned int i;
4186     GLint errPos;
4187
4188     memset(&priv_ctx, 0, sizeof(priv_ctx));
4189     priv_ctx.cur_vs_args = args;
4190     list_init(&priv_ctx.control_frames);
4191     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4192
4193     /*  Create the hw ARB shader */
4194     shader_addline(buffer, "!!ARBvp1.0\n");
4195
4196     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4197      * mesurable performance penalty, and we can always make use of it for clipplanes.
4198      */
4199     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4200     {
4201         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4202         priv_ctx.target_version = NV3;
4203         shader_addline(buffer, "ADDRESS aL;\n");
4204     }
4205     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4206     {
4207         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4208         priv_ctx.target_version = NV2;
4209         shader_addline(buffer, "ADDRESS aL;\n");
4210     } else {
4211         priv_ctx.target_version = ARB;
4212     }
4213
4214     shader_addline(buffer, "TEMP TMP_OUT;\n");
4215     if (reg_maps->fog)
4216         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4217     if (need_helper_const(shader_data, reg_maps, gl_info))
4218     {
4219         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4220     }
4221     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4222     {
4223         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4224         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4225     }
4226
4227     shader_addline(buffer, "TEMP TA;\n");
4228     shader_addline(buffer, "TEMP TB;\n");
4229
4230     /* Base Declarations */
4231     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4232             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4233
4234     for(i = 0; i < MAX_CONST_I; i++)
4235     {
4236         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4237         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4238         {
4239             const DWORD *control_values = find_loop_control_values(shader, i);
4240
4241             if(control_values)
4242             {
4243                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4244                                 control_values[0], control_values[1], control_values[2]);
4245             }
4246             else
4247             {
4248                 compiled->int_consts[i] = next_local;
4249                 compiled->num_int_consts++;
4250                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4251             }
4252         }
4253     }
4254
4255     /* We need a constant to fixup the final position */
4256     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4257     compiled->pos_fixup = next_local++;
4258
4259     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4260      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4261      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4262      * a replacement shader depend on the texcoord.w being set properly.
4263      *
4264      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4265      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4266      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4267      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4268      * this can eat a number of instructions, so skip it unless this cap is set as well
4269      */
4270     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4271     {
4272         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4273         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4274
4275         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->ffp_proj_control)
4276         {
4277             int i;
4278             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4279             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4280             {
4281                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4282                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4283             }
4284         }
4285     }
4286
4287     /* The shader starts with the main function */
4288     priv_ctx.in_main_func = TRUE;
4289     /* Base Shader Body */
4290     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4291
4292     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4293             shader_data, args, reg_maps, gl_info, buffer);
4294
4295     shader_addline(buffer, "END\n");
4296
4297     /* TODO: change to resource.glObjectHandle or something like that */
4298     GL_EXTCALL(glGenProgramsARB(1, &ret));
4299
4300     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4301     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4302
4303     TRACE("Created hw vertex shader, prg=%d\n", ret);
4304     /* Create the program and check for errors */
4305     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4306                buffer->content_size, buffer->buffer));
4307     checkGLcall("glProgramStringARB()");
4308
4309     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4310     if (errPos != -1)
4311     {
4312         FIXME("HW VertexShader Error at position %d: %s\n\n",
4313               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4314         shader_arb_dump_program_source(buffer->buffer);
4315         ret = -1;
4316     }
4317     else
4318     {
4319         GLint native;
4320
4321         GL_EXTCALL(glGetProgramivARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4322         checkGLcall("glGetProgramivARB()");
4323         if (!native) WARN("Program exceeds native resource limits.\n");
4324
4325         /* Load immediate constants */
4326         if (lconst_map)
4327         {
4328             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4329             {
4330                 const float *value = (const float *)lconst->value;
4331                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4332             }
4333         }
4334     }
4335     HeapFree(GetProcessHeap(), 0, lconst_map);
4336
4337     return ret;
4338 }
4339
4340 /* Context activation is done by the caller. */
4341 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4342         const struct arb_ps_compile_args *args)
4343 {
4344     struct wined3d_device *device = shader->device;
4345     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4346     UINT i;
4347     DWORD new_size;
4348     struct arb_ps_compiled_shader *new_array;
4349     struct wined3d_shader_buffer buffer;
4350     struct arb_pshader_private *shader_data;
4351     GLuint ret;
4352
4353     if (!shader->backend_data)
4354     {
4355         struct shader_arb_priv *priv = device->shader_priv;
4356
4357         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4358         shader_data = shader->backend_data;
4359         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4360
4361         if (shader->reg_maps.shader_version.major < 3)
4362             shader_data->input_signature_idx = ~0;
4363         else
4364             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4365
4366         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4367
4368         if (!device->vs_clipping)
4369             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4370                     gl_info->limits.texture_stages - 1);
4371         else
4372             shader_data->clipplane_emulation = ~0U;
4373     }
4374     shader_data = shader->backend_data;
4375
4376     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4377      * so a linear search is more performant than a hashmap or a binary search
4378      * (cache coherency etc)
4379      */
4380     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4381     {
4382         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4383             return &shader_data->gl_shaders[i];
4384     }
4385
4386     TRACE("No matching GL shader found, compiling a new shader\n");
4387     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4388         if (shader_data->num_gl_shaders)
4389         {
4390             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4391             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4392                                     new_size * sizeof(*shader_data->gl_shaders));
4393         } else {
4394             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4395             new_size = 1;
4396         }
4397
4398         if(!new_array) {
4399             ERR("Out of memory\n");
4400             return 0;
4401         }
4402         shader_data->gl_shaders = new_array;
4403         shader_data->shader_array_size = new_size;
4404     }
4405
4406     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4407
4408     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4409
4410     if (!shader_buffer_init(&buffer))
4411     {
4412         ERR("Failed to initialize shader buffer.\n");
4413         return 0;
4414     }
4415
4416     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4417             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4418     shader_buffer_free(&buffer);
4419     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4420
4421     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4422 }
4423
4424 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4425                                  const DWORD use_map, BOOL skip_int) {
4426     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4427     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4428     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4429     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4430     if(stored->ps_signature != new->ps_signature) return FALSE;
4431     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4432     if(skip_int) return TRUE;
4433
4434     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4435 }
4436
4437 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4438         const struct arb_vs_compile_args *args)
4439 {
4440     struct wined3d_device *device = shader->device;
4441     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4442     DWORD use_map = device->strided_streams.use_map;
4443     UINT i;
4444     DWORD new_size;
4445     struct arb_vs_compiled_shader *new_array;
4446     struct wined3d_shader_buffer buffer;
4447     struct arb_vshader_private *shader_data;
4448     GLuint ret;
4449
4450     if (!shader->backend_data)
4451     {
4452         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4453
4454         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4455         shader_data = shader->backend_data;
4456
4457         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4458                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4459         {
4460             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4461             {
4462                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4463                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4464                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4465             }
4466             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4467                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4468             else if (reg_maps->max_rel_offset > 63)
4469                 shader_data->rel_offset = reg_maps->min_rel_offset;
4470         }
4471     }
4472     shader_data = shader->backend_data;
4473
4474     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4475      * so a linear search is more performant than a hashmap or a binary search
4476      * (cache coherency etc)
4477      */
4478     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4479         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4480                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4481         {
4482             return &shader_data->gl_shaders[i];
4483         }
4484     }
4485
4486     TRACE("No matching GL shader found, compiling a new shader\n");
4487
4488     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4489         if (shader_data->num_gl_shaders)
4490         {
4491             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4492             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4493                                     new_size * sizeof(*shader_data->gl_shaders));
4494         } else {
4495             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4496             new_size = 1;
4497         }
4498
4499         if(!new_array) {
4500             ERR("Out of memory\n");
4501             return 0;
4502         }
4503         shader_data->gl_shaders = new_array;
4504         shader_data->shader_array_size = new_size;
4505     }
4506
4507     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4508
4509     if (!shader_buffer_init(&buffer))
4510     {
4511         ERR("Failed to initialize shader buffer.\n");
4512         return 0;
4513     }
4514
4515     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4516             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4517     shader_buffer_free(&buffer);
4518     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4519
4520     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4521 }
4522
4523 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4524         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4525 {
4526     struct wined3d_device *device = shader->device;
4527     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4528     int i;
4529     WORD int_skip;
4530
4531     find_ps_compile_args(state, shader, &args->super);
4532
4533     /* This forces all local boolean constants to 1 to make them stateblock independent */
4534     args->bools = shader->reg_maps.local_bool_consts;
4535
4536     for(i = 0; i < MAX_CONST_B; i++)
4537     {
4538         if (state->ps_consts_b[i])
4539             args->bools |= ( 1 << i);
4540     }
4541
4542     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4543      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4544      * duplicate the shader than have a no-op KIL instruction in every shader
4545      */
4546     if (!device->vs_clipping && use_vs(state)
4547             && state->render_states[WINED3D_RS_CLIPPING]
4548             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4549         args->clip = 1;
4550     else
4551         args->clip = 0;
4552
4553     /* Skip if unused or local, or supported natively */
4554     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4555     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4556     {
4557         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4558         return;
4559     }
4560
4561     for(i = 0; i < MAX_CONST_I; i++)
4562     {
4563         if(int_skip & (1 << i))
4564         {
4565             args->loop_ctrl[i][0] = 0;
4566             args->loop_ctrl[i][1] = 0;
4567             args->loop_ctrl[i][2] = 0;
4568         }
4569         else
4570         {
4571             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4572             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4573             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4574         }
4575     }
4576 }
4577
4578 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4579         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4580 {
4581     struct wined3d_device *device = shader->device;
4582     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4583     int i;
4584     WORD int_skip;
4585
4586     find_vs_compile_args(state, shader, &args->super);
4587
4588     args->clip.boolclip_compare = 0;
4589     if (use_ps(state))
4590     {
4591         const struct wined3d_shader *ps = state->pixel_shader;
4592         const struct arb_pshader_private *shader_priv = ps->backend_data;
4593         args->ps_signature = shader_priv->input_signature_idx;
4594
4595         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4596     }
4597     else
4598     {
4599         args->ps_signature = ~0;
4600         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4601         {
4602             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4603         }
4604         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4605     }
4606
4607     if (args->clip.boolclip.clip_texcoord)
4608     {
4609         if (state->render_states[WINED3D_RS_CLIPPING])
4610             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4611         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4612     }
4613
4614     /* This forces all local boolean constants to 1 to make them stateblock independent */
4615     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4616     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4617     for(i = 0; i < MAX_CONST_B; i++)
4618     {
4619         if (state->vs_consts_b[i])
4620             args->clip.boolclip.bools |= ( 1 << i);
4621     }
4622
4623     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4624     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4625     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4626     args->vertex.samplers[3] = 0;
4627
4628     /* Skip if unused or local */
4629     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4630     /* This is about flow control, not clipping. */
4631     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4632     {
4633         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4634         return;
4635     }
4636
4637     for(i = 0; i < MAX_CONST_I; i++)
4638     {
4639         if(int_skip & (1 << i))
4640         {
4641             args->loop_ctrl[i][0] = 0;
4642             args->loop_ctrl[i][1] = 0;
4643             args->loop_ctrl[i][2] = 0;
4644         }
4645         else
4646         {
4647             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4648             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4649             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4650         }
4651     }
4652 }
4653
4654 /* Context activation is done by the caller. */
4655 static void shader_arb_select(const struct wined3d_context *context, enum wined3d_shader_mode vertex_mode,
4656         enum wined3d_shader_mode fragment_mode)
4657 {
4658     struct wined3d_device *device = context->swapchain->device;
4659     struct shader_arb_priv *priv = device->shader_priv;
4660     const struct wined3d_gl_info *gl_info = context->gl_info;
4661     const struct wined3d_state *state = &device->stateBlock->state;
4662     int i;
4663
4664     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4665     if (fragment_mode == WINED3D_SHADER_MODE_SHADER)
4666     {
4667         struct wined3d_shader *ps = state->pixel_shader;
4668         struct arb_ps_compile_args compile_args;
4669         struct arb_ps_compiled_shader *compiled;
4670
4671         TRACE("Using pixel shader %p.\n", ps);
4672         find_arb_ps_compile_args(state, ps, &compile_args);
4673         compiled = find_arb_pshader(ps, &compile_args);
4674         priv->current_fprogram_id = compiled->prgId;
4675         priv->compiled_fprog = compiled;
4676
4677         /* Bind the fragment program */
4678         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4679         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4680
4681         if (!priv->use_arbfp_fixed_func)
4682             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4683
4684         /* Enable OpenGL fragment programs. */
4685         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4686         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4687
4688         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4689                 device, priv->current_fprogram_id);
4690
4691         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4692          * a 1.x and newer shader, reload the first 8 constants
4693          */
4694         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4695         {
4696             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4697             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4698             for(i = 0; i < 8; i++)
4699             {
4700                 priv->pshader_const_dirty[i] = 1;
4701             }
4702             /* Also takes care of loading local constants */
4703             shader_arb_load_constants_internal(context, TRUE, FALSE, TRUE);
4704         }
4705         else
4706         {
4707             UINT rt_height = state->fb->render_targets[0]->resource.height;
4708             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4709         }
4710
4711         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4712         if (compiled->np2fixup_info.super.active)
4713             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4714     }
4715     else
4716     {
4717         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4718         {
4719             /* Disable only if we're not using arbfp fixed function fragment
4720              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4721              * enabled, and the fixed function pipeline will bind the fixed
4722              * function replacement shader. */
4723             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4724             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4725             priv->current_fprogram_id = 0;
4726         }
4727         priv->fragment_pipe->enable_extension(gl_info, fragment_mode == WINED3D_SHADER_MODE_FFP);
4728     }
4729
4730     if (vertex_mode == WINED3D_SHADER_MODE_SHADER)
4731     {
4732         struct wined3d_shader *vs = state->vertex_shader;
4733         struct arb_vs_compile_args compile_args;
4734         struct arb_vs_compiled_shader *compiled;
4735
4736         TRACE("Using vertex shader %p\n", vs);
4737         find_arb_vs_compile_args(state, vs, &compile_args);
4738         compiled = find_arb_vshader(vs, &compile_args);
4739         priv->current_vprogram_id = compiled->prgId;
4740         priv->compiled_vprog = compiled;
4741
4742         /* Bind the vertex program */
4743         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4744         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4745
4746         /* Enable OpenGL vertex programs */
4747         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4748         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4749         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4750         shader_arb_vs_local_constants(compiled, context, state);
4751
4752         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4753             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4754
4755             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4756             {
4757                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4758                 checkGLcall("glClampColorARB");
4759             } else {
4760                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4761             }
4762         }
4763     }
4764     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4765     {
4766         priv->current_vprogram_id = 0;
4767         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4768         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4769     }
4770 }
4771
4772 /* Context activation is done by the caller. */
4773 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4774         enum tex_types tex_type, const SIZE *ds_mask_size)
4775 {
4776     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4777     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4778     struct shader_arb_priv *priv = shader_priv;
4779     GLuint *blt_fprogram;
4780
4781     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4782     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4783     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4784
4785     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4786     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4787     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4788     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4789     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4790 }
4791
4792 /* Context activation is done by the caller. */
4793 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4794 {
4795     struct shader_arb_priv *priv = shader_priv;
4796
4797     if (priv->current_vprogram_id) {
4798         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4799         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4800
4801         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4802     }
4803     else
4804     {
4805         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4806         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4807     }
4808
4809     if (priv->current_fprogram_id) {
4810         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4811         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4812
4813         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4814     }
4815     else if(!priv->use_arbfp_fixed_func)
4816     {
4817         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4818         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4819     }
4820 }
4821
4822 static void shader_arb_destroy(struct wined3d_shader *shader)
4823 {
4824     struct wined3d_device *device = shader->device;
4825     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4826
4827     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4828     {
4829         struct arb_pshader_private *shader_data = shader->backend_data;
4830         UINT i;
4831
4832         if(!shader_data) return; /* This can happen if a shader was never compiled */
4833
4834         if (shader_data->num_gl_shaders)
4835         {
4836             struct wined3d_context *context = context_acquire(device, NULL);
4837
4838             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4839             {
4840                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4841                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4842             }
4843
4844             context_release(context);
4845         }
4846
4847         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4848         HeapFree(GetProcessHeap(), 0, shader_data);
4849         shader->backend_data = NULL;
4850     }
4851     else
4852     {
4853         struct arb_vshader_private *shader_data = shader->backend_data;
4854         UINT i;
4855
4856         if(!shader_data) return; /* This can happen if a shader was never compiled */
4857
4858         if (shader_data->num_gl_shaders)
4859         {
4860             struct wined3d_context *context = context_acquire(device, NULL);
4861
4862             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4863             {
4864                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4865                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4866             }
4867
4868             context_release(context);
4869         }
4870
4871         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4872         HeapFree(GetProcessHeap(), 0, shader_data);
4873         shader->backend_data = NULL;
4874     }
4875 }
4876
4877 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4878 {
4879     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4880     return compare_sig(key, e->sig);
4881 }
4882
4883 static const struct wine_rb_functions sig_tree_functions =
4884 {
4885     wined3d_rb_alloc,
4886     wined3d_rb_realloc,
4887     wined3d_rb_free,
4888     sig_tree_compare
4889 };
4890
4891 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct fragment_pipeline *fragment_pipe)
4892 {
4893     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4894     struct fragment_caps fragment_caps;
4895     void *fragment_priv;
4896
4897     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4898     {
4899         ERR("Failed to initialize fragment pipe.\n");
4900         HeapFree(GetProcessHeap(), 0, priv);
4901         return E_FAIL;
4902     }
4903
4904     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4905             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4906     if (!priv->vshader_const_dirty)
4907         goto fail;
4908     memset(priv->vshader_const_dirty, 1,
4909            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4910
4911     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4912             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4913     if (!priv->pshader_const_dirty)
4914         goto fail;
4915     memset(priv->pshader_const_dirty, 1,
4916             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4917
4918     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4919     {
4920         ERR("RB tree init failed\n");
4921         goto fail;
4922     }
4923
4924     fragment_pipe->get_caps(&device->adapter->gl_info, &fragment_caps);
4925     priv->ffp_proj_control = fragment_caps.wined3d_caps & WINED3D_FRAGMENT_CAP_PROJ_CONTROL;
4926     device->fragment_priv = fragment_priv;
4927     priv->fragment_pipe = fragment_pipe;
4928     device->shader_priv = priv;
4929     return WINED3D_OK;
4930
4931 fail:
4932     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4933     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4934     fragment_pipe->free_private(device);
4935     HeapFree(GetProcessHeap(), 0, priv);
4936     return E_OUTOFMEMORY;
4937 }
4938
4939 static void release_signature(struct wine_rb_entry *entry, void *context)
4940 {
4941     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4942     int i;
4943     for(i = 0; i < MAX_REG_INPUT; i++)
4944     {
4945         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4946     }
4947     HeapFree(GetProcessHeap(), 0, sig->sig);
4948     HeapFree(GetProcessHeap(), 0, sig);
4949 }
4950
4951 /* Context activation is done by the caller. */
4952 static void shader_arb_free(struct wined3d_device *device)
4953 {
4954     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4955     struct shader_arb_priv *priv = device->shader_priv;
4956     int i;
4957
4958     if (priv->depth_blt_vprogram_id)
4959         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4960
4961     for (i = 0; i < tex_type_count; ++i)
4962     {
4963         if (priv->depth_blt_fprogram_id_full[i])
4964         {
4965             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4966         }
4967         if (priv->depth_blt_fprogram_id_masked[i])
4968         {
4969             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4970         }
4971     }
4972
4973     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4974     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4975     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4976     priv->fragment_pipe->free_private(device);
4977     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4978 }
4979
4980 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4981 {
4982     struct shader_arb_priv *priv = shader_priv;
4983
4984     if (priv->last_context == context)
4985         priv->last_context = NULL;
4986 }
4987
4988 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4989 {
4990     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4991     {
4992         DWORD vs_consts;
4993         UINT vs_version;
4994
4995         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4996          * for vertex programs. If the native limit is less than that it's
4997          * not very useful, and e.g. Mesa swrast returns 0, probably to
4998          * indicate it's a software implementation. */
4999         if (gl_info->limits.arb_vs_native_constants < 96)
5000             vs_consts = gl_info->limits.arb_vs_float_constants;
5001         else
5002             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
5003
5004         if (gl_info->supported[NV_VERTEX_PROGRAM3])
5005         {
5006             vs_version = 3;
5007             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
5008         }
5009         else if (vs_consts >= 256)
5010         {
5011             /* Shader Model 2.0 requires at least 256 vertex shader constants */
5012             vs_version = 2;
5013             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
5014         }
5015         else
5016         {
5017             vs_version = 1;
5018             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
5019         }
5020         caps->vs_version = min(wined3d_settings.max_sm_vs, vs_version);
5021         caps->vs_uniform_count = vs_consts;
5022     }
5023     else
5024     {
5025         caps->vs_version = 0;
5026         caps->vs_uniform_count = 0;
5027     }
5028
5029     caps->gs_version = 0;
5030
5031     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
5032     {
5033         DWORD ps_consts;
5034         UINT ps_version;
5035
5036         /* Similar as above for vertex programs, but the minimum for fragment
5037          * programs is 24. */
5038         if (gl_info->limits.arb_ps_native_constants < 24)
5039             ps_consts = gl_info->limits.arb_ps_float_constants;
5040         else
5041             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5042
5043         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5044         {
5045             ps_version = 3;
5046             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5047         }
5048         else if (ps_consts >= 32)
5049         {
5050             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5051             ps_version = 2;
5052             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5053         }
5054         else
5055         {
5056             ps_version = 1;
5057             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5058         }
5059         caps->ps_version = min(wined3d_settings.max_sm_ps, ps_version);
5060         caps->ps_uniform_count = ps_consts;
5061         caps->ps_1x_max_value = 8.0f;
5062     }
5063     else
5064     {
5065         caps->ps_version = 0;
5066         caps->ps_uniform_count = 0;
5067         caps->ps_1x_max_value = 0.0f;
5068     }
5069
5070     caps->wined3d_caps = WINED3D_SHADER_CAP_SRGB_WRITE;
5071     if (use_nv_clip(gl_info))
5072         caps->wined3d_caps |= WINED3D_SHADER_CAP_VS_CLIPPING;
5073 }
5074
5075 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5076 {
5077     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5078     {
5079         TRACE("Checking support for color_fixup:\n");
5080         dump_color_fixup_desc(fixup);
5081     }
5082
5083     /* We support everything except complex conversions. */
5084     if (!is_complex_fixup(fixup))
5085     {
5086         TRACE("[OK]\n");
5087         return TRUE;
5088     }
5089
5090     TRACE("[FAILED]\n");
5091     return FALSE;
5092 }
5093
5094 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5095     DWORD shift;
5096     char write_mask[20], regstr[50];
5097     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5098     BOOL is_color = FALSE;
5099     const struct wined3d_shader_dst_param *dst;
5100
5101     if (!ins->dst_count) return;
5102
5103     dst = &ins->dst[0];
5104     shift = dst->shift;
5105     if (!shift) return; /* Saturate alone is handled by the instructions */
5106
5107     shader_arb_get_write_mask(ins, dst, write_mask);
5108     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5109
5110     /* Generate a line that does the output modifier computation
5111      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5112      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5113      */
5114     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5115                    regstr, write_mask, regstr, shift_tab[shift]);
5116 }
5117
5118 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5119 {
5120     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5121     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5122     /* WINED3DSIH_AND                   */ NULL,
5123     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5124     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5125     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5126     /* WINED3DSIH_BREAKP                */ NULL,
5127     /* WINED3DSIH_CALL                  */ shader_hw_call,
5128     /* WINED3DSIH_CALLNZ                */ NULL,
5129     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5130     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5131     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5132     /* WINED3DSIH_CUT                   */ NULL,
5133     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5134     /* WINED3DSIH_DCL_CONSTANT_BUFFER   */ shader_hw_nop,
5135     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5136     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5137     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5138     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5139     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5140     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5141     /* WINED3DSIH_DIV                   */ NULL,
5142     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5143     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5144     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5145     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5146     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5147     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5148     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5149     /* WINED3DSIH_EMIT                  */ NULL,
5150     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5151     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5152     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5153     /* WINED3DSIH_EQ                    */ NULL,
5154     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5155     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5156     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5157     /* WINED3DSIH_FTOI                  */ NULL,
5158     /* WINED3DSIH_GE                    */ NULL,
5159     /* WINED3DSIH_IADD                  */ NULL,
5160     /* WINED3DSIH_IEQ                   */ NULL,
5161     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5162     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5163     /* WINED3DSIH_IGE                   */ NULL,
5164     /* WINED3DSIH_IMUL                  */ NULL,
5165     /* WINED3DSIH_ITOF                  */ NULL,
5166     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5167     /* WINED3DSIH_LD                    */ NULL,
5168     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5169     /* WINED3DSIH_LOG                   */ shader_hw_log,
5170     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5171     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5172     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5173     /* WINED3DSIH_LT                    */ NULL,
5174     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5175     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5176     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5177     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5178     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5179     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5180     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5181     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5182     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5183     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5184     /* WINED3DSIH_MOVC                  */ NULL,
5185     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5186     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5187     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5188     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5189     /* WINED3DSIH_POW                   */ shader_hw_pow,
5190     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5191     /* WINED3DSIH_REP                   */ shader_hw_rep,
5192     /* WINED3DSIH_RET                   */ shader_hw_ret,
5193     /* WINED3DSIH_ROUND_NI              */ NULL,
5194     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5195     /* WINED3DSIH_SAMPLE                */ NULL,
5196     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5197     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5198     /* WINED3DSIH_SETP                  */ NULL,
5199     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5200     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5201     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5202     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5203     /* WINED3DSIH_SQRT                  */ NULL,
5204     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5205     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5206     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5207     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5208     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5209     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5210     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5211     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5212     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5213     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5214     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5215     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5216     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5217     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5218     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5219     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5220     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5221     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5222     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5223     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5224     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5225     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5226     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5227     /* WINED3DSIH_UDIV                  */ NULL,
5228     /* WINED3DSIH_USHR                  */ NULL,
5229     /* WINED3DSIH_UTOF                  */ NULL,
5230     /* WINED3DSIH_XOR                   */ NULL,
5231 };
5232
5233 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5234         const struct wined3d_shader *shader, DWORD idx)
5235 {
5236     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5237     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5238     const struct wined3d_shader_lconst *constant;
5239     WORD bools = 0;
5240     WORD flag = (1 << idx);
5241     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5242
5243     if (reg_maps->local_bool_consts & flag)
5244     {
5245         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5246         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5247         {
5248             if (constant->idx == idx)
5249             {
5250                 return constant->value[0];
5251             }
5252         }
5253         ERR("Local constant not found\n");
5254         return FALSE;
5255     }
5256     else
5257     {
5258         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5259         else bools = priv->cur_ps_args->bools;
5260         return bools & flag;
5261     }
5262 }
5263
5264 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5265         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5266 {
5267     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5268     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5269
5270     /* Integer constants can either be a local constant, or they can be stored in the shader
5271      * type specific compile args. */
5272     if (reg_maps->local_int_consts & (1 << idx))
5273     {
5274         const struct wined3d_shader_lconst *constant;
5275
5276         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5277         {
5278             if (constant->idx == idx)
5279             {
5280                 loop_control->count = constant->value[0];
5281                 loop_control->start = constant->value[1];
5282                 /* Step is signed. */
5283                 loop_control->step = (int)constant->value[2];
5284                 return;
5285             }
5286         }
5287         /* If this happens the flag was set incorrectly */
5288         ERR("Local constant not found\n");
5289         loop_control->count = 0;
5290         loop_control->start = 0;
5291         loop_control->step = 0;
5292         return;
5293     }
5294
5295     switch (reg_maps->shader_version.type)
5296     {
5297         case WINED3D_SHADER_TYPE_VERTEX:
5298             /* Count and aL start value are unsigned */
5299             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5300             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5301             /* Step is signed. */
5302             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5303             break;
5304
5305         case WINED3D_SHADER_TYPE_PIXEL:
5306             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5307             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5308             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5309             break;
5310
5311         default:
5312             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5313             break;
5314     }
5315 }
5316
5317 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5318 {
5319     unsigned int i;
5320     struct wined3d_shader_dst_param *dst_param = NULL;
5321     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5322     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5323     if(!rec)
5324     {
5325         ERR("Out of memory\n");
5326         return;
5327     }
5328
5329     rec->ins = *ins;
5330     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5331     if(!dst_param) goto free;
5332     *dst_param = *ins->dst;
5333     if (ins->dst->reg.idx[0].rel_addr)
5334     {
5335         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5336         if (!rel_addr)
5337             goto free;
5338         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5339         dst_param->reg.idx[0].rel_addr = rel_addr;
5340     }
5341     rec->ins.dst = dst_param;
5342
5343     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5344     if (!src_param)
5345         goto free;
5346     for (i = 0; i < ins->src_count; ++i)
5347     {
5348         src_param[i] = ins->src[i];
5349         if (ins->src[i].reg.idx[0].rel_addr)
5350         {
5351             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5352             if (!rel_addr)
5353                 goto free;
5354             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5355             src_param[i].reg.idx[0].rel_addr = rel_addr;
5356         }
5357     }
5358     rec->ins.src = src_param;
5359     list_add_tail(list, &rec->entry);
5360     return;
5361
5362 free:
5363     ERR("Out of memory\n");
5364     if(dst_param)
5365     {
5366         HeapFree(GetProcessHeap(), 0, (void *)dst_param->reg.idx[0].rel_addr);
5367         HeapFree(GetProcessHeap(), 0, dst_param);
5368     }
5369     if(src_param)
5370     {
5371         for(i = 0; i < ins->src_count; i++)
5372         {
5373             HeapFree(GetProcessHeap(), 0, (void *)src_param[i].reg.idx[0].rel_addr);
5374         }
5375         HeapFree(GetProcessHeap(), 0, src_param);
5376     }
5377     HeapFree(GetProcessHeap(), 0, rec);
5378 }
5379
5380 static void free_recorded_instruction(struct list *list)
5381 {
5382     struct recorded_instruction *rec_ins, *entry2;
5383     unsigned int i;
5384
5385     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5386     {
5387         list_remove(&rec_ins->entry);
5388         if (rec_ins->ins.dst)
5389         {
5390             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5391             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst);
5392         }
5393         if (rec_ins->ins.src)
5394         {
5395             for (i = 0; i < rec_ins->ins.src_count; ++i)
5396             {
5397                 HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5398             }
5399             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src);
5400         }
5401         HeapFree(GetProcessHeap(), 0, rec_ins);
5402     }
5403 }
5404
5405 static void pop_control_frame(const struct wined3d_shader_instruction *ins)
5406 {
5407     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5408     struct control_frame *control_frame;
5409
5410     if (ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5411     {
5412         struct list *e = list_head(&priv->control_frames);
5413         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5414         list_remove(&control_frame->entry);
5415         HeapFree(GetProcessHeap(), 0, control_frame);
5416         priv->loop_depth--;
5417     }
5418     else if (ins->handler_idx == WINED3DSIH_ENDIF)
5419     {
5420         /* Non-ifc ENDIFs were already handled previously. */
5421         struct list *e = list_head(&priv->control_frames);
5422         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5423         list_remove(&control_frame->entry);
5424         HeapFree(GetProcessHeap(), 0, control_frame);
5425     }
5426 }
5427
5428 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5429     SHADER_HANDLER hw_fct;
5430     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5431     const struct wined3d_shader *shader = ins->ctx->shader;
5432     struct control_frame *control_frame;
5433     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5434     BOOL bool_const;
5435
5436     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5437     {
5438         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5439         list_add_head(&priv->control_frames, &control_frame->entry);
5440
5441         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5442         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5443
5444         if(priv->target_version >= NV2)
5445         {
5446             control_frame->no.loop = priv->num_loops++;
5447             priv->loop_depth++;
5448         }
5449         else
5450         {
5451             /* Don't bother recording when we're in a not used if branch */
5452             if(priv->muted)
5453             {
5454                 return;
5455             }
5456
5457             if(!priv->recording)
5458             {
5459                 list_init(&priv->record);
5460                 priv->recording = TRUE;
5461                 control_frame->outer_loop = TRUE;
5462                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5463                 return; /* Instruction is handled */
5464             }
5465             /* Record this loop in the outer loop's recording */
5466         }
5467     }
5468     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5469     {
5470         if(priv->target_version >= NV2)
5471         {
5472             /* Nothing to do. The control frame is popped after the HW instr handler */
5473         }
5474         else
5475         {
5476             struct list *e = list_head(&priv->control_frames);
5477             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5478             list_remove(&control_frame->entry);
5479
5480             if(control_frame->outer_loop)
5481             {
5482                 unsigned int iteration;
5483                 int aL = 0;
5484                 struct list copy;
5485
5486                 /* Turn off recording before playback */
5487                 priv->recording = FALSE;
5488
5489                 /* Move the recorded instructions to a separate list and get them out of the private data
5490                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5491                  * be recorded again, thus priv->record might be overwritten
5492                  */
5493                 list_init(&copy);
5494                 list_move_tail(&copy, &priv->record);
5495                 list_init(&priv->record);
5496
5497                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5498                 {
5499                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5500                                    control_frame->loop_control.count, control_frame->loop_control.start,
5501                                    control_frame->loop_control.step);
5502                     aL = control_frame->loop_control.start;
5503                 }
5504                 else
5505                 {
5506                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5507                 }
5508
5509                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5510                 {
5511                     struct recorded_instruction *rec_ins;
5512                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5513                     {
5514                         priv->aL = aL;
5515                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5516                     }
5517                     else
5518                     {
5519                         shader_addline(buffer, "#Iteration %u\n", iteration);
5520                     }
5521
5522                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5523                     {
5524                         shader_arb_handle_instruction(&rec_ins->ins);
5525                     }
5526
5527                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5528                     {
5529                         aL += control_frame->loop_control.step;
5530                     }
5531                 }
5532                 shader_addline(buffer, "#end loop/rep\n");
5533
5534                 free_recorded_instruction(&copy);
5535                 HeapFree(GetProcessHeap(), 0, control_frame);
5536                 return; /* Instruction is handled */
5537             }
5538             else
5539             {
5540                 /* This is a nested loop. Proceed to the normal recording function */
5541                 HeapFree(GetProcessHeap(), 0, control_frame);
5542             }
5543         }
5544     }
5545
5546     if(priv->recording)
5547     {
5548         record_instruction(&priv->record, ins);
5549         return;
5550     }
5551
5552     /* boolean if */
5553     if(ins->handler_idx == WINED3DSIH_IF)
5554     {
5555         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5556         list_add_head(&priv->control_frames, &control_frame->entry);
5557         control_frame->type = IF;
5558
5559         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5560         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5561             bool_const = !bool_const;
5562         if (!priv->muted && !bool_const)
5563         {
5564             shader_addline(buffer, "#if(FALSE){\n");
5565             priv->muted = TRUE;
5566             control_frame->muting = TRUE;
5567         }
5568         else shader_addline(buffer, "#if(TRUE) {\n");
5569
5570         return; /* Instruction is handled */
5571     }
5572     else if(ins->handler_idx == WINED3DSIH_IFC)
5573     {
5574         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5575         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5576         control_frame->type = IFC;
5577         control_frame->no.ifc = priv->num_ifcs++;
5578         list_add_head(&priv->control_frames, &control_frame->entry);
5579     }
5580     else if(ins->handler_idx == WINED3DSIH_ELSE)
5581     {
5582         struct list *e = list_head(&priv->control_frames);
5583         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5584
5585         if(control_frame->type == IF)
5586         {
5587             shader_addline(buffer, "#} else {\n");
5588             if(!priv->muted && !control_frame->muting)
5589             {
5590                 priv->muted = TRUE;
5591                 control_frame->muting = TRUE;
5592             }
5593             else if(control_frame->muting) priv->muted = FALSE;
5594             return; /* Instruction is handled. */
5595         }
5596         /* In case of an ifc, generate a HW shader instruction */
5597         if (control_frame->type != IFC)
5598             ERR("Control frame does not match.\n");
5599     }
5600     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5601     {
5602         struct list *e = list_head(&priv->control_frames);
5603         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5604
5605         if(control_frame->type == IF)
5606         {
5607             shader_addline(buffer, "#} endif\n");
5608             if(control_frame->muting) priv->muted = FALSE;
5609             list_remove(&control_frame->entry);
5610             HeapFree(GetProcessHeap(), 0, control_frame);
5611             return; /* Instruction is handled */
5612         }
5613         /* In case of an ifc, generate a HW shader instruction */
5614         if (control_frame->type != IFC)
5615             ERR("Control frame does not match.\n");
5616     }
5617
5618     if(priv->muted)
5619     {
5620         pop_control_frame(ins);
5621         return;
5622     }
5623
5624     /* Select handler */
5625     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5626
5627     /* Unhandled opcode */
5628     if (!hw_fct)
5629     {
5630         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5631         return;
5632     }
5633     hw_fct(ins);
5634
5635     pop_control_frame(ins);
5636
5637     shader_arb_add_instruction_modifiers(ins);
5638 }
5639
5640 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5641 {
5642     struct shader_arb_priv *priv = shader_priv;
5643
5644     return priv->ffp_proj_control;
5645 }
5646
5647 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5648 {
5649     shader_arb_handle_instruction,
5650     shader_arb_select,
5651     shader_arb_select_depth_blt,
5652     shader_arb_deselect_depth_blt,
5653     shader_arb_update_float_vertex_constants,
5654     shader_arb_update_float_pixel_constants,
5655     shader_arb_load_constants,
5656     shader_arb_load_np2fixup_constants,
5657     shader_arb_destroy,
5658     shader_arb_alloc,
5659     shader_arb_free,
5660     shader_arb_context_destroyed,
5661     shader_arb_get_caps,
5662     shader_arb_color_fixup_supported,
5663     shader_arb_has_ffp_proj_control,
5664 };
5665
5666 /* ARB_fragment_program fixed function pipeline replacement definitions */
5667 #define ARB_FFP_CONST_TFACTOR           0
5668 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5669 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5670 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5671 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5672
5673 struct arbfp_ffp_desc
5674 {
5675     struct ffp_frag_desc parent;
5676     GLuint shader;
5677 };
5678
5679 /* Context activation is done by the caller. */
5680 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5681 {
5682     if (enable)
5683     {
5684         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5685         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5686     }
5687     else
5688     {
5689         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5690         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5691     }
5692 }
5693
5694 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5695 {
5696     struct shader_arb_priv *priv;
5697
5698     /* Share private data between the shader backend and the pipeline
5699      * replacement, if both are the arb implementation. This is needed to
5700      * figure out whether ARBfp should be disabled if no pixel shader is bound
5701      * or not. */
5702     if (shader_backend == &arb_program_shader_backend)
5703         priv = shader_priv;
5704     else if (!(priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv))))
5705         return NULL;
5706
5707     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5708     {
5709         ERR("Failed to initialize rbtree.\n");
5710         if (priv != shader_priv)
5711             HeapFree(GetProcessHeap(), 0, priv);
5712         return NULL;
5713     }
5714     priv->use_arbfp_fixed_func = TRUE;
5715
5716     return priv;
5717 }
5718
5719 /* Context activation is done by the caller. */
5720 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5721 {
5722     const struct wined3d_gl_info *gl_info = context;
5723     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5724
5725     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5726     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5727     HeapFree(GetProcessHeap(), 0, entry_arb);
5728 }
5729
5730 /* Context activation is done by the caller. */
5731 static void arbfp_free(struct wined3d_device *device)
5732 {
5733     struct shader_arb_priv *priv = device->fragment_priv;
5734
5735     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5736     priv->use_arbfp_fixed_func = FALSE;
5737
5738     if (device->shader_backend != &arb_program_shader_backend)
5739     {
5740         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5741     }
5742 }
5743
5744 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5745 {
5746     caps->wined3d_caps = WINED3D_FRAGMENT_CAP_PROJ_CONTROL
5747             | WINED3D_FRAGMENT_CAP_SRGB_WRITE;
5748     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5749     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5750                            WINED3DTEXOPCAPS_SELECTARG1                  |
5751                            WINED3DTEXOPCAPS_SELECTARG2                  |
5752                            WINED3DTEXOPCAPS_MODULATE4X                  |
5753                            WINED3DTEXOPCAPS_MODULATE2X                  |
5754                            WINED3DTEXOPCAPS_MODULATE                    |
5755                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5756                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5757                            WINED3DTEXOPCAPS_ADD                         |
5758                            WINED3DTEXOPCAPS_SUBTRACT                    |
5759                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5760                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5761                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5762                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5763                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5764                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5765                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5766                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5767                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5768                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5769                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5770                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5771                            WINED3DTEXOPCAPS_LERP                        |
5772                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5773                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5774
5775     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5776
5777     caps->MaxTextureBlendStages   = 8;
5778     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5779 }
5780
5781 static void state_texfactor_arbfp(struct wined3d_context *context,
5782         const struct wined3d_state *state, DWORD state_id)
5783 {
5784     struct wined3d_device *device = context->swapchain->device;
5785     const struct wined3d_gl_info *gl_info = context->gl_info;
5786     float col[4];
5787
5788     /* Don't load the parameter if we're using an arbfp pixel shader,
5789      * otherwise we'll overwrite application provided constants. */
5790     if (device->shader_backend == &arb_program_shader_backend)
5791     {
5792         struct shader_arb_priv *priv;
5793
5794         if (use_ps(state)) return;
5795
5796         priv = device->shader_priv;
5797         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5798         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5799     }
5800
5801     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5802     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5803     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5804 }
5805
5806 static void state_arb_specularenable(struct wined3d_context *context,
5807         const struct wined3d_state *state, DWORD state_id)
5808 {
5809     struct wined3d_device *device = context->swapchain->device;
5810     const struct wined3d_gl_info *gl_info = context->gl_info;
5811     float col[4];
5812
5813     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5814      * application provided constants
5815      */
5816     if (device->shader_backend == &arb_program_shader_backend)
5817     {
5818         struct shader_arb_priv *priv;
5819
5820         if (use_ps(state)) return;
5821
5822         priv = device->shader_priv;
5823         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5824         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5825     }
5826
5827     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5828     {
5829         /* The specular color has no alpha */
5830         col[0] = 1.0f; col[1] = 1.0f;
5831         col[2] = 1.0f; col[3] = 0.0f;
5832     } else {
5833         col[0] = 0.0f; col[1] = 0.0f;
5834         col[2] = 0.0f; col[3] = 0.0f;
5835     }
5836     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5837     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5838 }
5839
5840 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5841 {
5842     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5843     struct wined3d_device *device = context->swapchain->device;
5844     const struct wined3d_gl_info *gl_info = context->gl_info;
5845     float mat[2][2];
5846
5847     if (use_ps(state))
5848     {
5849         /* The pixel shader has to know the bump env matrix. Do a constants
5850          * update. */
5851         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5852             context->load_constants = 1;
5853
5854         if(device->shader_backend == &arb_program_shader_backend) {
5855             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5856             return;
5857         }
5858     }
5859     else if (device->shader_backend == &arb_program_shader_backend)
5860     {
5861         struct shader_arb_priv *priv = device->shader_priv;
5862         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5863         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5864     }
5865
5866     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5867     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5868     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5869     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5870
5871     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5872     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5873 }
5874
5875 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5876         const struct wined3d_state *state, DWORD state_id)
5877 {
5878     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5879     struct wined3d_device *device = context->swapchain->device;
5880     const struct wined3d_gl_info *gl_info = context->gl_info;
5881     float param[4];
5882
5883     if (use_ps(state))
5884     {
5885         /* The pixel shader has to know the luminance offset. Do a constants
5886          * update. */
5887         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5888             context->load_constants = 1;
5889
5890         if(device->shader_backend == &arb_program_shader_backend) {
5891             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5892             return;
5893         }
5894     }
5895     else if (device->shader_backend == &arb_program_shader_backend)
5896     {
5897         struct shader_arb_priv *priv = device->shader_priv;
5898         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5899         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5900     }
5901
5902     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5903     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5904     param[2] = 0.0f;
5905     param[3] = 0.0f;
5906
5907     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5908     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5909 }
5910
5911 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5912 {
5913     const char *ret;
5914
5915     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5916
5917     switch(arg & WINED3DTA_SELECTMASK) {
5918         case WINED3DTA_DIFFUSE:
5919             ret = "fragment.color.primary"; break;
5920
5921         case WINED3DTA_CURRENT:
5922             if (!stage) ret = "fragment.color.primary";
5923             else ret = "ret";
5924             break;
5925
5926         case WINED3DTA_TEXTURE:
5927             switch(stage) {
5928                 case 0: ret = "tex0"; break;
5929                 case 1: ret = "tex1"; break;
5930                 case 2: ret = "tex2"; break;
5931                 case 3: ret = "tex3"; break;
5932                 case 4: ret = "tex4"; break;
5933                 case 5: ret = "tex5"; break;
5934                 case 6: ret = "tex6"; break;
5935                 case 7: ret = "tex7"; break;
5936                 default: ret = "unknown texture";
5937             }
5938             break;
5939
5940         case WINED3DTA_TFACTOR:
5941             ret = "tfactor"; break;
5942
5943         case WINED3DTA_SPECULAR:
5944             ret = "fragment.color.secondary"; break;
5945
5946         case WINED3DTA_TEMP:
5947             ret = "tempreg"; break;
5948
5949         case WINED3DTA_CONSTANT:
5950             FIXME("Implement perstage constants\n");
5951             switch(stage) {
5952                 case 0: ret = "const0"; break;
5953                 case 1: ret = "const1"; break;
5954                 case 2: ret = "const2"; break;
5955                 case 3: ret = "const3"; break;
5956                 case 4: ret = "const4"; break;
5957                 case 5: ret = "const5"; break;
5958                 case 6: ret = "const6"; break;
5959                 case 7: ret = "const7"; break;
5960                 default: ret = "unknown constant";
5961             }
5962             break;
5963
5964         default:
5965             return "unknown";
5966     }
5967
5968     if(arg & WINED3DTA_COMPLEMENT) {
5969         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5970         if(argnum == 0) ret = "arg0";
5971         if(argnum == 1) ret = "arg1";
5972         if(argnum == 2) ret = "arg2";
5973     }
5974     if(arg & WINED3DTA_ALPHAREPLICATE) {
5975         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5976         if(argnum == 0) ret = "arg0";
5977         if(argnum == 1) ret = "arg1";
5978         if(argnum == 2) ret = "arg2";
5979     }
5980     return ret;
5981 }
5982
5983 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5984         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5985 {
5986     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5987     unsigned int mul = 1;
5988
5989     if(color && alpha) dstmask = "";
5990     else if(color) dstmask = ".xyz";
5991     else dstmask = ".w";
5992
5993     if(dst == tempreg) dstreg = "tempreg";
5994     else dstreg = "ret";
5995
5996     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5997     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5998     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5999
6000     switch (op)
6001     {
6002         case WINED3D_TOP_DISABLE:
6003             if (!stage)
6004                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
6005             break;
6006
6007         case WINED3D_TOP_SELECT_ARG2:
6008             arg1 = arg2;
6009             /* FALLTHROUGH */
6010         case WINED3D_TOP_SELECT_ARG1:
6011             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
6012             break;
6013
6014         case WINED3D_TOP_MODULATE_4X:
6015             mul = 2;
6016             /* FALLTHROUGH */
6017         case WINED3D_TOP_MODULATE_2X:
6018             mul *= 2;
6019             /* FALLTHROUGH */
6020         case WINED3D_TOP_MODULATE:
6021             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6022             break;
6023
6024         case WINED3D_TOP_ADD_SIGNED_2X:
6025             mul = 2;
6026             /* FALLTHROUGH */
6027         case WINED3D_TOP_ADD_SIGNED:
6028             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6029             arg2 = "arg2";
6030             /* FALLTHROUGH */
6031         case WINED3D_TOP_ADD:
6032             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6033             break;
6034
6035         case WINED3D_TOP_SUBTRACT:
6036             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6037             break;
6038
6039         case WINED3D_TOP_ADD_SMOOTH:
6040             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6041             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6042             break;
6043
6044         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6045             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6046             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6047             break;
6048         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6049             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6050             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6051             break;
6052         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6053             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6054             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6055             break;
6056         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6057             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6058             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6059             break;
6060
6061         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6062             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6063             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6064             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6065             break;
6066
6067         /* D3DTOP_PREMODULATE ???? */
6068
6069         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6070             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6071             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6072             break;
6073         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6074             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6075             break;
6076         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6077             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6078             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6079             break;
6080         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6081             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6082             break;
6083
6084         case WINED3D_TOP_DOTPRODUCT3:
6085             mul = 4;
6086             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6087             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6088             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6089             break;
6090
6091         case WINED3D_TOP_MULTIPLY_ADD:
6092             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6093             break;
6094
6095         case WINED3D_TOP_LERP:
6096             /* The msdn is not quite right here */
6097             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6098             break;
6099
6100         case WINED3D_TOP_BUMPENVMAP:
6101         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6102             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6103             break;
6104
6105         default:
6106             FIXME("Unhandled texture op %08x\n", op);
6107     }
6108
6109     if (mul == 2)
6110         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6111     else if (mul == 4)
6112         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6113 }
6114
6115 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6116 {
6117     unsigned int stage;
6118     struct wined3d_shader_buffer buffer;
6119     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6120     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6121     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6122     UINT lowest_disabled_stage;
6123     const char *textype;
6124     const char *instr, *sat;
6125     char colorcor_dst[8];
6126     GLuint ret;
6127     DWORD arg0, arg1, arg2;
6128     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6129     BOOL op_equal;
6130     const char *final_combiner_src = "ret";
6131     GLint pos;
6132
6133     /* Find out which textures are read */
6134     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6135     {
6136         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6137             break;
6138         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6139         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6140         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6141         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6142         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6143         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6144
6145         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6146             tex_read[stage] = TRUE;
6147         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6148             tex_read[stage] = TRUE;
6149         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6150         {
6151             bump_used[stage] = TRUE;
6152             tex_read[stage] = TRUE;
6153         }
6154         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6155         {
6156             bump_used[stage] = TRUE;
6157             tex_read[stage] = TRUE;
6158             luminance_used[stage] = TRUE;
6159         }
6160         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6161         {
6162             tfactor_used = TRUE;
6163         }
6164
6165         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6166             tfactor_used = TRUE;
6167         }
6168
6169         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6170         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6171             tempreg_used = TRUE;
6172         }
6173
6174         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6175             continue;
6176         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6177         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6178         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6179         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6180         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6181         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6182
6183         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6184             tempreg_used = TRUE;
6185         }
6186         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6187             tfactor_used = TRUE;
6188         }
6189     }
6190     lowest_disabled_stage = stage;
6191
6192     /* Shader header */
6193     if (!shader_buffer_init(&buffer))
6194     {
6195         ERR("Failed to initialize shader buffer.\n");
6196         return 0;
6197     }
6198
6199     shader_addline(&buffer, "!!ARBfp1.0\n");
6200
6201     switch(settings->fog) {
6202         case FOG_OFF:                                                         break;
6203         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6204         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6205         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6206         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6207     }
6208
6209     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6210     shader_addline(&buffer, "TEMP TMP;\n");
6211     shader_addline(&buffer, "TEMP ret;\n");
6212     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6213     shader_addline(&buffer, "TEMP arg0;\n");
6214     shader_addline(&buffer, "TEMP arg1;\n");
6215     shader_addline(&buffer, "TEMP arg2;\n");
6216     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6217         if(!tex_read[stage]) continue;
6218         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6219         if(!bump_used[stage]) continue;
6220         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6221         if(!luminance_used[stage]) continue;
6222         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6223     }
6224     if(tfactor_used) {
6225         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6226     }
6227     shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6228
6229     if(settings->sRGB_write) {
6230         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6231                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6232         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6233                        srgb_sub_high, 0.0, 0.0, 0.0);
6234     }
6235
6236     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6237         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6238
6239     /* Generate texture sampling instructions) */
6240     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6241     {
6242         if (!tex_read[stage])
6243             continue;
6244
6245         switch(settings->op[stage].tex_type) {
6246             case tex_1d:                    textype = "1D";     break;
6247             case tex_2d:                    textype = "2D";     break;
6248             case tex_3d:                    textype = "3D";     break;
6249             case tex_cube:                  textype = "CUBE";   break;
6250             case tex_rect:                  textype = "RECT";   break;
6251             default: textype = "unexpected_textype";   break;
6252         }
6253
6254         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6255                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6256             sat = "";
6257         else
6258             sat = "_SAT";
6259
6260         if(settings->op[stage].projected == proj_none) {
6261             instr = "TEX";
6262         } else if(settings->op[stage].projected == proj_count4 ||
6263                   settings->op[stage].projected == proj_count3) {
6264             instr = "TXP";
6265         } else {
6266             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6267             instr = "TXP";
6268         }
6269
6270         if (stage > 0
6271                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6272                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6273         {
6274             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6275             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6276             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6277             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6278
6279             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6280              * so multiply the displacement with the dividing parameter before passing it to TXP
6281              */
6282             if (settings->op[stage].projected != proj_none) {
6283                 if(settings->op[stage].projected == proj_count4) {
6284                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6285                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6286                 } else {
6287                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6288                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6289                 }
6290             } else {
6291                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6292             }
6293
6294             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6295                     instr, sat, stage, stage, textype);
6296             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6297             {
6298                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6299                                stage - 1, stage - 1, stage - 1);
6300                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6301             }
6302         } else if(settings->op[stage].projected == proj_count3) {
6303             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6304             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6305             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6306                             instr, sat, stage, stage, textype);
6307         } else {
6308             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6309                             instr, sat, stage, stage, stage, textype);
6310         }
6311
6312         sprintf(colorcor_dst, "tex%u", stage);
6313         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6314                 settings->op[stage].color_fixup);
6315     }
6316
6317     /* Generate the main shader */
6318     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6319     {
6320         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6321         {
6322             if (!stage)
6323                 final_combiner_src = "fragment.color.primary";
6324             break;
6325         }
6326
6327         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6328                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6329             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6330         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6331                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6332             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6333         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6334                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6335             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6336         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6337                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6338             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6339         else
6340             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6341                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6342                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6343                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6344
6345         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6346         {
6347             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6348                           settings->op[stage].cop, settings->op[stage].carg0,
6349                           settings->op[stage].carg1, settings->op[stage].carg2);
6350             if (!stage)
6351                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6352         }
6353         else if (op_equal)
6354         {
6355             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6356                           settings->op[stage].cop, settings->op[stage].carg0,
6357                           settings->op[stage].carg1, settings->op[stage].carg2);
6358         } else {
6359             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6360                           settings->op[stage].cop, settings->op[stage].carg0,
6361                           settings->op[stage].carg1, settings->op[stage].carg2);
6362             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6363                           settings->op[stage].aop, settings->op[stage].aarg0,
6364                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6365         }
6366     }
6367
6368     if(settings->sRGB_write) {
6369         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6370         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6371         shader_addline(&buffer, "MOV result.color, ret;\n");
6372     } else {
6373         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6374     }
6375
6376     /* Footer */
6377     shader_addline(&buffer, "END\n");
6378
6379     /* Generate the shader */
6380     GL_EXTCALL(glGenProgramsARB(1, &ret));
6381     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6382     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6383             strlen(buffer.buffer), buffer.buffer));
6384     checkGLcall("glProgramStringARB()");
6385
6386     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6387     if (pos != -1)
6388     {
6389         FIXME("Fragment program error at position %d: %s\n\n", pos,
6390               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6391         shader_arb_dump_program_source(buffer.buffer);
6392     }
6393     else
6394     {
6395         GLint native;
6396
6397         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6398         checkGLcall("glGetProgramivARB()");
6399         if (!native) WARN("Program exceeds native resource limits.\n");
6400     }
6401
6402     shader_buffer_free(&buffer);
6403     return ret;
6404 }
6405
6406 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6407 {
6408     const struct wined3d_device *device = context->swapchain->device;
6409     const struct wined3d_gl_info *gl_info = context->gl_info;
6410     struct shader_arb_priv *priv = device->fragment_priv;
6411     BOOL use_pshader = use_ps(state);
6412     struct ffp_frag_settings settings;
6413     const struct arbfp_ffp_desc *desc;
6414     unsigned int i;
6415
6416     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6417
6418     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6419     {
6420         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6421         {
6422             /* Reload fixed function constants since they collide with the
6423              * pixel shader constants. */
6424             for (i = 0; i < MAX_TEXTURES; ++i)
6425             {
6426                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6427             }
6428             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6429             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6430         }
6431         else if (use_pshader)
6432         {
6433             context->select_shader = 1;
6434         }
6435         return;
6436     }
6437
6438     if (!use_pshader)
6439     {
6440         /* Find or create a shader implementing the fixed function pipeline
6441          * settings, then activate it. */
6442         gen_ffp_frag_op(device, state, &settings, FALSE);
6443         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6444         if(!desc) {
6445             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6446             if (!new_desc)
6447             {
6448                 ERR("Out of memory\n");
6449                 return;
6450             }
6451
6452             new_desc->parent.settings = settings;
6453             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6454             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6455             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6456             desc = new_desc;
6457         }
6458
6459         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6460          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6461          * deactivate it.
6462          */
6463         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6464         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6465         priv->current_fprogram_id = desc->shader;
6466
6467         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6468         {
6469             /* Reload fixed function constants since they collide with the
6470              * pixel shader constants. */
6471             for (i = 0; i < MAX_TEXTURES; ++i)
6472             {
6473                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6474             }
6475             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6476             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6477         }
6478         context->last_was_pshader = FALSE;
6479     } else {
6480         context->last_was_pshader = TRUE;
6481     }
6482
6483     context->select_shader = 1;
6484     context->load_constants = 1;
6485 }
6486
6487 /* We can't link the fog states to the fragment state directly since the
6488  * vertex pipeline links them to FOGENABLE. A different linking in different
6489  * pipeline parts can't be expressed in the combined state table, so we need
6490  * to handle that with a forwarding function. The other invisible side effect
6491  * is that changing the fog start and fog end (which links to FOGENABLE in
6492  * vertex) results in the fragment_prog_arbfp function being called because
6493  * FOGENABLE is dirty, which calls this function here. */
6494 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6495 {
6496     enum fogsource new_source;
6497
6498     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6499
6500     if (!isStateDirty(context, STATE_PIXELSHADER))
6501         fragment_prog_arbfp(context, state, state_id);
6502
6503     if (!state->render_states[WINED3D_RS_FOGENABLE])
6504         return;
6505
6506     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6507     {
6508         if (use_vs(state))
6509         {
6510             new_source = FOGSOURCE_VS;
6511         }
6512         else
6513         {
6514             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6515                 new_source = FOGSOURCE_COORD;
6516             else
6517                 new_source = FOGSOURCE_FFP;
6518         }
6519     }
6520     else
6521     {
6522         new_source = FOGSOURCE_FFP;
6523     }
6524
6525     if (new_source != context->fog_source)
6526     {
6527         context->fog_source = new_source;
6528         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6529     }
6530 }
6531
6532 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6533 {
6534     if (!isStateDirty(context, STATE_PIXELSHADER))
6535         fragment_prog_arbfp(context, state, state_id);
6536 }
6537
6538 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6539 {
6540     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6636     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6637     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6639     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6640     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6641     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6642     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6643     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6652     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6653     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6654     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6655     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6656     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6657     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6658     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6659     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6660     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6661     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6662     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6663     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6664     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6665     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6666     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6667     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6668     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6669     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6670     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6671     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6672     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6673     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6674     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6675     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6676     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6677     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6678     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6679     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6680     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6681 };
6682
6683 const struct fragment_pipeline arbfp_fragment_pipeline = {
6684     arbfp_enable,
6685     arbfp_get_caps,
6686     arbfp_alloc,
6687     arbfp_free,
6688     shader_arb_color_fixup_supported,
6689     arbfp_fragmentstate_template,
6690 };
6691
6692 struct arbfp_blit_priv {
6693     GLenum yuy2_rect_shader, yuy2_2d_shader;
6694     GLenum uyvy_rect_shader, uyvy_2d_shader;
6695     GLenum yv12_rect_shader, yv12_2d_shader;
6696     GLenum p8_rect_shader, p8_2d_shader;
6697     GLuint palette_texture;
6698 };
6699
6700 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6701 {
6702     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6703     if(!device->blit_priv) {
6704         ERR("Out of memory\n");
6705         return E_OUTOFMEMORY;
6706     }
6707     return WINED3D_OK;
6708 }
6709
6710 /* Context activation is done by the caller. */
6711 static void arbfp_blit_free(struct wined3d_device *device)
6712 {
6713     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6714     struct arbfp_blit_priv *priv = device->blit_priv;
6715
6716     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6717     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6718     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6719     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6720     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6721     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6722     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6723     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6724     checkGLcall("Delete yuv and p8 programs");
6725
6726     if (priv->palette_texture)
6727         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6728
6729     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6730     device->blit_priv = NULL;
6731 }
6732
6733 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6734         GLenum textype, char *luminance)
6735 {
6736     char chroma;
6737     const char *tex, *texinstr;
6738
6739     if (fixup == COMPLEX_FIXUP_UYVY) {
6740         chroma = 'x';
6741         *luminance = 'w';
6742     } else {
6743         chroma = 'w';
6744         *luminance = 'x';
6745     }
6746     switch(textype) {
6747         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6748         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6749         default:
6750             /* This is more tricky than just replacing the texture type - we have to navigate
6751              * properly in the texture to find the correct chroma values
6752              */
6753             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6754             return FALSE;
6755     }
6756
6757     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6758      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6759      * filtering when we sample the texture.
6760      *
6761      * These are the rules for reading the chroma:
6762      *
6763      * Even pixel: Cr
6764      * Even pixel: U
6765      * Odd pixel: V
6766      *
6767      * So we have to get the sampling x position in non-normalized coordinates in integers
6768      */
6769     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6770         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6771         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6772     } else {
6773         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6774     }
6775     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6776      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6777      * 0.5, so add 0.5.
6778      */
6779     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6780     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6781
6782     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6783      * even and odd pixels respectively
6784      */
6785     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6786     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6787
6788     /* Sample Pixel 1 */
6789     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6790
6791     /* Put the value into either of the chroma values */
6792     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6793     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6794     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6795     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6796
6797     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6798      * the pixel right to the current one. Otherwise, sample the left pixel.
6799      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6800      */
6801     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6802     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6803     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6804
6805     /* Put the value into the other chroma */
6806     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6807     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6808     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6809     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6810
6811     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6812      * the current one and lerp the two U and V values
6813      */
6814
6815     /* This gives the correctly filtered luminance value */
6816     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6817
6818     return TRUE;
6819 }
6820
6821 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6822 {
6823     const char *tex;
6824
6825     switch(textype) {
6826         case GL_TEXTURE_2D:             tex = "2D";     break;
6827         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6828         default:
6829             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6830             return FALSE;
6831     }
6832
6833     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6834      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6835      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6836      * pitch of the luminance plane, the packing into the gl texture is a bit
6837      * unfortunate. If the whole texture is interpreted as luminance data it looks
6838      * approximately like this:
6839      *
6840      *        +----------------------------------+----
6841      *        |                                  |
6842      *        |                                  |
6843      *        |                                  |
6844      *        |                                  |
6845      *        |                                  |   2
6846      *        |            LUMINANCE             |   -
6847      *        |                                  |   3
6848      *        |                                  |
6849      *        |                                  |
6850      *        |                                  |
6851      *        |                                  |
6852      *        +----------------+-----------------+----
6853      *        |                |                 |
6854      *        |  U even rows   |  U odd rows     |
6855      *        |                |                 |   1
6856      *        +----------------+------------------   -
6857      *        |                |                 |   3
6858      *        |  V even rows   |  V odd rows     |
6859      *        |                |                 |
6860      *        +----------------+-----------------+----
6861      *        |                |                 |
6862      *        |     0.5        |       0.5       |
6863      *
6864      * So it appears as if there are 4 chroma images, but in fact the odd rows
6865      * in the chroma images are in the same row as the even ones. So its is
6866      * kinda tricky to read
6867      *
6868      * When reading from rectangle textures, keep in mind that the input y coordinates
6869      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6870      */
6871     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6872             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6873
6874     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6875     /* the chroma planes have only half the width */
6876     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6877
6878     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6879      * the coordinate. Also read the right side of the image when reading odd lines
6880      *
6881      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6882      * bleeding
6883      */
6884     if(textype == GL_TEXTURE_2D) {
6885
6886         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6887
6888         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6889
6890         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6891         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6892
6893         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6894         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6895         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6896         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6897         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6898
6899         /* clamp, keep the half pixel origin in mind */
6900         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6901         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6902         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6903         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6904     } else {
6905         /* Read from [size - size+size/4] */
6906         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6907         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6908
6909         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6910         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6911         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6912         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6913         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6914         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6915
6916         /* Make sure to read exactly from the pixel center */
6917         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6918         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6919
6920         /* Clamp */
6921         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6922         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6923         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6924         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6925         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6926     }
6927     /* Read the texture, put the result into the output register */
6928     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6929     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6930
6931     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6932      * No need to clamp because we're just reusing the already clamped value from above
6933      */
6934     if(textype == GL_TEXTURE_2D) {
6935         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6936     } else {
6937         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6938     }
6939     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6940     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6941
6942     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6943      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6944      * values due to filtering
6945      */
6946     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6947     if(textype == GL_TEXTURE_2D) {
6948         /* Multiply the y coordinate by 2/3 and clamp it */
6949         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6950         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6951         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6952         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6953     } else {
6954         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6955          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6956          * is bigger
6957          */
6958         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6959         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6960         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6961     }
6962     *luminance = 'a';
6963
6964     return TRUE;
6965 }
6966
6967 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6968         const struct wined3d_gl_info *gl_info, GLenum textype)
6969 {
6970     GLenum shader;
6971     struct wined3d_shader_buffer buffer;
6972     GLint pos;
6973
6974     /* Shader header */
6975     if (!shader_buffer_init(&buffer))
6976     {
6977         ERR("Failed to initialize shader buffer.\n");
6978         return 0;
6979     }
6980
6981     GL_EXTCALL(glGenProgramsARB(1, &shader));
6982     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6983     if (!shader)
6984     {
6985         shader_buffer_free(&buffer);
6986         return 0;
6987     }
6988
6989     shader_addline(&buffer, "!!ARBfp1.0\n");
6990     shader_addline(&buffer, "TEMP index;\n");
6991
6992     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6993     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6994
6995     /* The alpha-component contains the palette index */
6996     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6997         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6998     else
6999         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
7000
7001     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
7002     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
7003
7004     /* Use the alpha-component as an index in the palette to get the final color */
7005     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
7006     shader_addline(&buffer, "END\n");
7007
7008     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7009             strlen(buffer.buffer), buffer.buffer));
7010     checkGLcall("glProgramStringARB()");
7011
7012     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7013     if (pos != -1)
7014     {
7015         FIXME("Fragment program error at position %d: %s\n\n", pos,
7016               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7017         shader_arb_dump_program_source(buffer.buffer);
7018     }
7019
7020     if (textype == GL_TEXTURE_RECTANGLE_ARB)
7021         priv->p8_rect_shader = shader;
7022     else
7023         priv->p8_2d_shader = shader;
7024
7025     shader_buffer_free(&buffer);
7026
7027     return shader;
7028 }
7029
7030 /* Context activation is done by the caller. */
7031 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7032 {
7033     BYTE table[256][4];
7034     struct wined3d_device *device = surface->resource.device;
7035     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7036     struct arbfp_blit_priv *priv = device->blit_priv;
7037     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) != 0;
7038
7039     d3dfmt_p8_init_palette(surface, table, colorkey);
7040
7041     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7042     {
7043         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7044         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7045     }
7046
7047     if (!priv->palette_texture)
7048         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7049
7050     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7051     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7052
7053     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7054
7055     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7056     /* Make sure we have discrete color levels. */
7057     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7058     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7059     /* Upload the palette */
7060     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7061     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7062
7063     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7064     {
7065         gl_info->gl_ops.gl.p_glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7066         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7067     }
7068
7069     /* Switch back to unit 0 in which the 2D texture will be stored. */
7070     context_active_texture(context, gl_info, 0);
7071 }
7072
7073 /* Context activation is done by the caller. */
7074 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7075         enum complex_fixup yuv_fixup, GLenum textype)
7076 {
7077     GLenum shader;
7078     struct wined3d_shader_buffer buffer;
7079     char luminance_component;
7080     GLint pos;
7081
7082     /* Shader header */
7083     if (!shader_buffer_init(&buffer))
7084     {
7085         ERR("Failed to initialize shader buffer.\n");
7086         return 0;
7087     }
7088
7089     GL_EXTCALL(glGenProgramsARB(1, &shader));
7090     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7091     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7092     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7093     if (!shader)
7094     {
7095         shader_buffer_free(&buffer);
7096         return 0;
7097     }
7098
7099     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7100      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7101      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7102      * each single pixel it contains, and one U and one V value shared between both
7103      * pixels.
7104      *
7105      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7106      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7107      * take the format into account when generating the read swizzles
7108      *
7109      * Reading the Y value is straightforward - just sample the texture. The hardware
7110      * takes care of filtering in the horizontal and vertical direction.
7111      *
7112      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7113      * because that would mix the U and V values of one pixel or two adjacent pixels.
7114      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7115      * regardless of the filtering setting. Vertical filtering works automatically
7116      * though - the U and V values of two rows are mixed nicely.
7117      *
7118      * Apart of avoiding filtering issues, the code has to know which value it just
7119      * read, and where it can find the other one. To determine this, it checks if
7120      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7121      *
7122      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7123      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7124      *
7125      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7126      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7127      * in an unfiltered situation. Finding the luminance on the other hand requires
7128      * finding out if it is an odd or even pixel. The real drawback of this approach
7129      * is filtering. This would have to be emulated completely in the shader, reading
7130      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7131      * vertically. Beyond that it would require adjustments to the texture handling
7132      * code to deal with the width scaling
7133      */
7134     shader_addline(&buffer, "!!ARBfp1.0\n");
7135     shader_addline(&buffer, "TEMP luminance;\n");
7136     shader_addline(&buffer, "TEMP temp;\n");
7137     shader_addline(&buffer, "TEMP chroma;\n");
7138     shader_addline(&buffer, "TEMP texcrd;\n");
7139     shader_addline(&buffer, "TEMP texcrd2;\n");
7140     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7141     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7142     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7143
7144     switch (yuv_fixup)
7145     {
7146         case COMPLEX_FIXUP_UYVY:
7147         case COMPLEX_FIXUP_YUY2:
7148             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7149             {
7150                 shader_buffer_free(&buffer);
7151                 return 0;
7152             }
7153             break;
7154
7155         case COMPLEX_FIXUP_YV12:
7156             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7157             {
7158                 shader_buffer_free(&buffer);
7159                 return 0;
7160             }
7161             break;
7162
7163         default:
7164             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7165             shader_buffer_free(&buffer);
7166             return 0;
7167     }
7168
7169     /* Calculate the final result. Formula is taken from
7170      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7171      * ranges from -0.5 to 0.5
7172      */
7173     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7174
7175     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7176     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7177     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7178     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7179     shader_addline(&buffer, "END\n");
7180
7181     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7182             strlen(buffer.buffer), buffer.buffer));
7183     checkGLcall("glProgramStringARB()");
7184
7185     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7186     if (pos != -1)
7187     {
7188         FIXME("Fragment program error at position %d: %s\n\n", pos,
7189               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7190         shader_arb_dump_program_source(buffer.buffer);
7191     }
7192     else
7193     {
7194         GLint native;
7195
7196         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7197         checkGLcall("glGetProgramivARB()");
7198         if (!native) WARN("Program exceeds native resource limits.\n");
7199     }
7200
7201     shader_buffer_free(&buffer);
7202
7203     switch (yuv_fixup)
7204     {
7205         case COMPLEX_FIXUP_YUY2:
7206             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7207             else priv->yuy2_2d_shader = shader;
7208             break;
7209
7210         case COMPLEX_FIXUP_UYVY:
7211             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7212             else priv->uyvy_2d_shader = shader;
7213             break;
7214
7215         case COMPLEX_FIXUP_YV12:
7216             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7217             else priv->yv12_2d_shader = shader;
7218             break;
7219         default:
7220             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7221     }
7222
7223     return shader;
7224 }
7225
7226 /* Context activation is done by the caller. */
7227 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7228 {
7229     GLenum shader;
7230     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7231     struct arbfp_blit_priv *priv = blit_priv;
7232     enum complex_fixup fixup;
7233     const struct wined3d_gl_info *gl_info = context->gl_info;
7234     GLenum textype;
7235
7236     if (surface->container.type == WINED3D_CONTAINER_TEXTURE)
7237         textype = surface->container.u.texture->target;
7238     else
7239         textype = surface->texture_target;
7240
7241     if (surface->flags & SFLAG_CONVERTED)
7242     {
7243         gl_info->gl_ops.gl.p_glEnable(textype);
7244         checkGLcall("glEnable(textype)");
7245         return WINED3D_OK;
7246     }
7247
7248     if (!is_complex_fixup(surface->resource.format->color_fixup))
7249     {
7250         TRACE("Fixup:\n");
7251         dump_color_fixup_desc(surface->resource.format->color_fixup);
7252         /* Don't bother setting up a shader for unconverted formats */
7253         gl_info->gl_ops.gl.p_glEnable(textype);
7254         checkGLcall("glEnable(textype)");
7255         return WINED3D_OK;
7256     }
7257
7258     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7259
7260     switch(fixup)
7261     {
7262         case COMPLEX_FIXUP_YUY2:
7263             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7264             break;
7265
7266         case COMPLEX_FIXUP_UYVY:
7267             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7268             break;
7269
7270         case COMPLEX_FIXUP_YV12:
7271             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7272             break;
7273
7274         case COMPLEX_FIXUP_P8:
7275             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7276             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7277
7278             upload_palette(surface, context);
7279             break;
7280
7281         default:
7282             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7283             gl_info->gl_ops.gl.p_glEnable(textype);
7284             checkGLcall("glEnable(textype)");
7285             return E_NOTIMPL;
7286     }
7287
7288     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7289
7290     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7291     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7292     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7293     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7294     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7295     checkGLcall("glProgramLocalParameter4fvARB");
7296
7297     return WINED3D_OK;
7298 }
7299
7300 /* Context activation is done by the caller. */
7301 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7302 {
7303     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7304     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7305     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7306     checkGLcall("glDisable(GL_TEXTURE_2D)");
7307     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7308     {
7309         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7310         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7311     }
7312     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7313     {
7314         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7315         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7316     }
7317 }
7318
7319 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7320         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7321         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7322 {
7323     enum complex_fixup src_fixup;
7324
7325     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7326         return FALSE;
7327
7328     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7329     {
7330         TRACE("Unsupported blit_op=%d\n", blit_op);
7331         return FALSE;
7332     }
7333
7334     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7335         return FALSE;
7336
7337     src_fixup = get_complex_fixup(src_format->color_fixup);
7338     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7339     {
7340         TRACE("Checking support for fixup:\n");
7341         dump_color_fixup_desc(src_format->color_fixup);
7342     }
7343
7344     if (!is_identity_fixup(dst_format->color_fixup))
7345     {
7346         TRACE("Destination fixups are not supported\n");
7347         return FALSE;
7348     }
7349
7350     if (is_identity_fixup(src_format->color_fixup))
7351     {
7352         TRACE("[OK]\n");
7353         return TRUE;
7354     }
7355
7356      /* We only support YUV conversions. */
7357     if (!is_complex_fixup(src_format->color_fixup))
7358     {
7359         TRACE("[FAILED]\n");
7360         return FALSE;
7361     }
7362
7363     switch(src_fixup)
7364     {
7365         case COMPLEX_FIXUP_YUY2:
7366         case COMPLEX_FIXUP_UYVY:
7367         case COMPLEX_FIXUP_YV12:
7368         case COMPLEX_FIXUP_P8:
7369             TRACE("[OK]\n");
7370             return TRUE;
7371
7372         default:
7373             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7374             TRACE("[FAILED]\n");
7375             return FALSE;
7376     }
7377 }
7378
7379 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7380         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7381         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7382 {
7383     struct wined3d_context *context;
7384     RECT src_rect = *src_rect_in;
7385     RECT dst_rect = *dst_rect_in;
7386
7387     /* Now load the surface */
7388     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7389             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7390     {
7391         /* Without FBO blits transferring from the drawable to the texture is
7392          * expensive, because we have to flip the data in sysmem. Since we can
7393          * flip in the blitter, we don't actually need that flip anyway. So we
7394          * use the surface's texture as scratch texture, and flip the source
7395          * rectangle instead. */
7396         surface_load_fb_texture(src_surface, FALSE);
7397
7398         src_rect.top = src_surface->resource.height - src_rect.top;
7399         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7400     }
7401     else
7402         surface_internal_preload(src_surface, SRGB_RGB);
7403
7404     /* Activate the destination context, set it up for blitting */
7405     context = context_acquire(device, dst_surface);
7406     context_apply_blit_state(context, device);
7407
7408     if (!surface_is_offscreen(dst_surface))
7409         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7410
7411     arbfp_blit_set(device->blit_priv, context, src_surface);
7412
7413     /* Draw a textured quad */
7414     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7415
7416     /* Leave the opengl state valid for blitting */
7417     arbfp_blit_unset(context->gl_info);
7418
7419     if (wined3d_settings.strict_draw_ordering
7420             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7421             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7422         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7423
7424     context_release(context);
7425
7426     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7427     return WINED3D_OK;
7428 }
7429
7430 /* Do not call while under the GL lock. */
7431 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7432         const RECT *dst_rect, const struct wined3d_color *color)
7433 {
7434     FIXME("Color filling not implemented by arbfp_blit\n");
7435     return WINED3DERR_INVALIDCALL;
7436 }
7437
7438 /* Do not call while under the GL lock. */
7439 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7440         struct wined3d_surface *surface, const RECT *rect, float depth)
7441 {
7442     FIXME("Depth filling not implemented by arbfp_blit.\n");
7443     return WINED3DERR_INVALIDCALL;
7444 }
7445
7446 const struct blit_shader arbfp_blit = {
7447     arbfp_blit_alloc,
7448     arbfp_blit_free,
7449     arbfp_blit_set,
7450     arbfp_blit_unset,
7451     arbfp_blit_supported,
7452     arbfp_blit_color_fill,
7453     arbfp_blit_depth_fill,
7454 };