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

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