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