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