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