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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31
  32 #include <math.h>
  33 #include <stdio.h>
  34
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 /* Extract a line. Note that this modifies the source string. */
  43 static char *get_line(char **ptr)
  44 {
  45     char *p, *q;
  46
  47     p = *ptr;
  48     if (!(q = strstr(p, "\n")))
  49     {
  50         if (!*p) return NULL;
  51         *ptr += strlen(p);
  52         return p;
  53     }
  54     *q = '\0';
  55     *ptr = q + 1;
  56
  57     return p;
  58 }
  59
  60 static void shader_arb_dump_program_source(const char *source)
  61 {
  62     ULONG source_size;
  63     char *ptr, *line, *tmp;
  64
  65     source_size = strlen(source) + 1;
  66     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
  67     if (!tmp)
  68     {
  69         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
  70         return;
  71     }
  72     memcpy(tmp, source, source_size);
  73
  74     ptr = tmp;
  75     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
  76     FIXME("\n");
  77
  78     HeapFree(GetProcessHeap(), 0, tmp);
  79 }
  80
  81 /* GL locking for state handlers is done by the caller. */
  82 static BOOL need_rel_addr_const(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info)
  83 {
  84     if (shader->baseShader.reg_maps.shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
  85     {
  86         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset) return TRUE;
  87     }
  88     if (!shader->baseShader.reg_maps.usesmova) return FALSE;
  89     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
  90 }
  91
  92 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
  93 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
  94 {
  95     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
  96             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
  97 }
  98
  99 static BOOL need_helper_const(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info)
 100 {
 101     if (need_rel_addr_const(shader, gl_info)) return TRUE;
 102     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 103     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 104     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 105     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 106     if (shader->baseShader.reg_maps.usesnrm) return TRUE; /* 0.0 */
 107     if (shader->baseShader.reg_maps.usesrcp) return TRUE; /* EPS */
 108     return FALSE;
 109 }
 110
 111 static unsigned int reserved_vs_const(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info)
 112 {
 113     unsigned int ret = 1;
 114     /* We use one PARAM for the pos fixup, and in some cases one to load
 115      * some immediate values into the shader
 116      */
 117     if(need_helper_const(shader, gl_info)) ret++;
 118     if(need_rel_addr_const(shader, gl_info)) ret++;
 119     return ret;
 120 }
 121
 122 enum arb_helper_value
 123 {
 124     ARB_ZERO,
 125     ARB_ONE,
 126     ARB_TWO,
 127     ARB_0001,
 128     ARB_EPS,
 129
 130     ARB_VS_REL_OFFSET
 131 };
 132
 133 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
 134 {
 135     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
 136     {
 137         ERR("Geometry shaders are unsupported\n");
 138         return "bad";
 139     }
 140
 141     if (shader == WINED3D_SHADER_TYPE_PIXEL)
 142     {
 143         switch (value)
 144         {
 145             case ARB_ZERO: return "ps_helper_const.x";
 146             case ARB_ONE: return "ps_helper_const.y";
 147             case ARB_TWO: return "coefmul.x";
 148             case ARB_0001: return "helper_const.xxxy";
 149             case ARB_EPS: return "ps_helper_const.z";
 150             default: break;
 151         }
 152     }
 153     else
 154     {
 155         switch (value)
 156         {
 157             case ARB_ZERO: return "helper_const.x";
 158             case ARB_ONE: return "helper_const.y";
 159             case ARB_TWO: return "helper_const.z";
 160             case ARB_EPS: return "helper_const.w";
 161             case ARB_0001: return "helper_const.xxxy";
 162             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
 163         }
 164     }
 165     FIXME("Unmanaged %s shader helper constant requested: %u\n",
 166           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
 167     switch (value)
 168     {
 169         case ARB_ZERO: return "0.0";
 170         case ARB_ONE: return "1.0";
 171         case ARB_TWO: return "2.0";
 172         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
 173         case ARB_EPS: return "1e-8";
 174         default: return "bad";
 175     }
 176 }
 177
 178 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
 179 {
 180     return state->lowest_disabled_stage < 7;
 181 }
 182
 183 /* ARB_program_shader private data */
 184
 185 struct control_frame
 186 {
 187     struct                          list entry;
 188     enum
 189     {
 190         IF,
 191         IFC,
 192         LOOP,
 193         REP
 194     } type;
 195     BOOL                            muting;
 196     BOOL                            outer_loop;
 197     union
 198     {
 199         unsigned int                loop;
 200         unsigned int                ifc;
 201     } no;
 202     struct wined3d_shader_loop_control loop_control;
 203     BOOL                            had_else;
 204 };
 205
 206 struct arb_ps_np2fixup_info
 207 {
 208     struct ps_np2fixup_info         super;
 209     /* For ARB we need a offset value:
 210      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 211      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 212      * array we need an offset to the index inside the program local parameter array. */
 213     UINT                            offset;
 214 };
 215
 216 struct arb_ps_compile_args
 217 {
 218     struct ps_compile_args          super;
 219     WORD                            bools;
 220     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 221     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 222 };
 223
 224 struct stb_const_desc
 225 {
 226     unsigned char           texunit;
 227     UINT                    const_num;
 228 };
 229
 230 struct arb_ps_compiled_shader
 231 {
 232     struct arb_ps_compile_args      args;
 233     struct arb_ps_np2fixup_info     np2fixup_info;
 234     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 235     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 236     UINT                            int_consts[MAX_CONST_I];
 237     GLuint                          prgId;
 238     UINT                            ycorrection;
 239     unsigned char                   numbumpenvmatconsts;
 240     char                            num_int_consts;
 241 };
 242
 243 struct arb_vs_compile_args
 244 {
 245     struct vs_compile_args          super;
 246     union
 247     {
 248         struct
 249         {
 250             WORD                    bools;
 251             unsigned char           clip_texcoord;
 252             unsigned char           clipplane_mask;
 253         }                           boolclip;
 254         DWORD                       boolclip_compare;
 255     } clip;
 256     DWORD                           ps_signature;
 257     union
 258     {
 259         unsigned char               samplers[4];
 260         DWORD                       samplers_compare;
 261     } vertex;
 262     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 263 };
 264
 265 struct arb_vs_compiled_shader
 266 {
 267     struct arb_vs_compile_args      args;
 268     GLuint                          prgId;
 269     UINT                            int_consts[MAX_CONST_I];
 270     char                            num_int_consts;
 271     char                            need_color_unclamp;
 272     UINT                            pos_fixup;
 273 };
 274
 275 struct recorded_instruction
 276 {
 277     struct wined3d_shader_instruction ins;
 278     struct list entry;
 279 };
 280
 281 struct shader_arb_ctx_priv
 282 {
 283     char addr_reg[20];
 284     enum
 285     {
 286         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 287         ARB,
 288         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 289         NV2,
 290         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 291         NV3
 292     } target_version;
 293
 294     const struct arb_vs_compile_args    *cur_vs_args;
 295     const struct arb_ps_compile_args    *cur_ps_args;
 296     const struct arb_ps_compiled_shader *compiled_fprog;
 297     const struct arb_vs_compiled_shader *compiled_vprog;
 298     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 299     struct list                         control_frames;
 300     struct list                         record;
 301     BOOL                                recording;
 302     BOOL                                muted;
 303     unsigned int                        num_loops, loop_depth, num_ifcs;
 304     int                                 aL;
 305
 306     unsigned int                        vs_clipplanes;
 307     BOOL                                footer_written;
 308     BOOL                                in_main_func;
 309
 310     /* For 3.0 vertex shaders */
 311     const char                          *vs_output[MAX_REG_OUTPUT];
 312     /* For 2.x and earlier vertex shaders */
 313     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 314
 315     /* 3.0 pshader input for compatibility with fixed function */
 316     const char                          *ps_input[MAX_REG_INPUT];
 317 };
 318
 319 struct ps_signature
 320 {
 321     struct wined3d_shader_signature_element *sig;
 322     DWORD                               idx;
 323     struct wine_rb_entry                entry;
 324 };
 325
 326 struct arb_pshader_private {
 327     struct arb_ps_compiled_shader   *gl_shaders;
 328     UINT                            num_gl_shaders, shader_array_size;
 329     BOOL                            has_signature_idx;
 330     DWORD                           input_signature_idx;
 331     DWORD                           clipplane_emulation;
 332     BOOL                            clamp_consts;
 333 };
 334
 335 struct arb_vshader_private {
 336     struct arb_vs_compiled_shader   *gl_shaders;
 337     UINT                            num_gl_shaders, shader_array_size;
 338 };
 339
 340 struct shader_arb_priv
 341 {
 342     GLuint                  current_vprogram_id;
 343     GLuint                  current_fprogram_id;
 344     const struct arb_ps_compiled_shader *compiled_fprog;
 345     const struct arb_vs_compiled_shader *compiled_vprog;
 346     GLuint                  depth_blt_vprogram_id;
 347     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 348     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 349     BOOL                    use_arbfp_fixed_func;
 350     struct wine_rb_tree     fragment_shaders;
 351     BOOL                    last_ps_const_clamped;
 352     BOOL                    last_vs_color_unclamp;
 353
 354     struct wine_rb_tree     signature_tree;
 355     DWORD ps_sig_number;
 356 };
 357
 358 /********************************************************
 359  * ARB_[vertex/fragment]_program helper functions follow
 360  ********************************************************/
 361
 362 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 363  * When constant_list == NULL, it will load all the constants.
 364  *
 365  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 366  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 367  */
 368 /* GL locking is done by the caller */
 369 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
 370         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
 371 {
 372     local_constant* lconst;
 373     DWORD i, j;
 374     unsigned int ret;
 375
 376     if (TRACE_ON(d3d_constants))
 377     {
 378         for(i = 0; i < max_constants; i++) {
 379             if(!dirty_consts[i]) continue;
 380             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 381                         constants[i * 4 + 0], constants[i * 4 + 1],
 382                         constants[i * 4 + 2], constants[i * 4 + 3]);
 383         }
 384     }
 385
 386     i = 0;
 387
 388     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 389     if (target_type == GL_FRAGMENT_PROGRAM_ARB && This->baseShader.reg_maps.shader_version.major == 1)
 390     {
 391         float lcl_const[4];
 392         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 393          * shaders, the first 8 constants are marked dirty for reload
 394          */
 395         for(; i < min(8, max_constants); i++) {
 396             if(!dirty_consts[i]) continue;
 397             dirty_consts[i] = 0;
 398
 399             j = 4 * i;
 400             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 401             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 402             else lcl_const[0] = constants[j + 0];
 403
 404             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 405             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 406             else lcl_const[1] = constants[j + 1];
 407
 408             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 409             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 410             else lcl_const[2] = constants[j + 2];
 411
 412             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 413             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 414             else lcl_const[3] = constants[j + 3];
 415
 416             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 417         }
 418
 419         /* If further constants are dirty, reload them without clamping.
 420          *
 421          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 422          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 423          * above would always re-check the first 8 constants since max_constant remains at the init
 424          * value
 425          */
 426     }
 427
 428     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 429     {
 430         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 431          * or just reloading *all* constants at once
 432          *
 433         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 434          */
 435         for(; i < max_constants; i++) {
 436             if(!dirty_consts[i]) continue;
 437
 438             /* Find the next block of dirty constants */
 439             dirty_consts[i] = 0;
 440             j = i;
 441             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 442                 dirty_consts[i] = 0;
 443             }
 444
 445             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 446         }
 447     } else {
 448         for(; i < max_constants; i++) {
 449             if(dirty_consts[i]) {
 450                 dirty_consts[i] = 0;
 451                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 452             }
 453         }
 454     }
 455     checkGLcall("glProgramEnvParameter4fvARB()");
 456
 457     /* Load immediate constants */
 458     if(This->baseShader.load_local_constsF) {
 459         if (TRACE_ON(d3d_shader)) {
 460             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 461                 GLfloat* values = (GLfloat*)lconst->value;
 462                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 463                         values[0], values[1], values[2], values[3]);
 464             }
 465         }
 466         /* Immediate constants are clamped for 1.X shaders at loading times */
 467         ret = 0;
 468         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 469             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 470             ret = max(ret, lconst->idx + 1);
 471             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 472         }
 473         checkGLcall("glProgramEnvParameter4fvARB()");
 474         return ret; /* The loaded immediate constants need reloading for the next shader */
 475     } else {
 476         return 0; /* No constants are dirty now */
 477     }
 478 }
 479
 480 /**
 481  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 482  */
 483 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 484         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 485 {
 486     const struct shader_arb_priv * priv = shader_priv;
 487
 488     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 489     if (!use_ps(state)) return;
 490
 491     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 492         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 493         UINT i;
 494         WORD active = fixup->super.active;
 495         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 496
 497         for (i = 0; active; active >>= 1, ++i)
 498         {
 499             const IWineD3DBaseTextureImpl *tex = state->textures[i];
 500             const unsigned char idx = fixup->super.idx[i];
 501             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 502
 503             if (!(active & 1)) continue;
 504
 505             if (!tex) {
 506                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 507                 continue;
 508             }
 509
 510             if (idx % 2) {
 511                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 512             } else {
 513                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 514             }
 515         }
 516
 517         for (i = 0; i < fixup->super.num_consts; ++i) {
 518             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 519                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 520         }
 521     }
 522 }
 523
 524 /* GL locking is done by the caller. */
 525 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 526         const struct wined3d_context *context, const struct wined3d_state *state, float rt_height)
 527 {
 528     const struct wined3d_gl_info *gl_info = context->gl_info;
 529     unsigned char i;
 530
 531     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 532     {
 533         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 534
 535         /* The state manager takes care that this function is always called if the bump env matrix changes */
 536         const float *data = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVMAT00];
 537         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 538                 gl_shader->bumpenvmatconst[i].const_num, data));
 539
 540         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 541         {
 542             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 543              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 544              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 545              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 546             */
 547             const float *scale = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVLSCALE];
 548             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 549                     gl_shader->luminanceconst[i].const_num, scale));
 550         }
 551     }
 552     checkGLcall("Load bumpmap consts");
 553
 554     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 555     {
 556         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 557         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 558         * ycorrection.z: 1.0
 559         * ycorrection.w: 0.0
 560         */
 561         float val[4];
 562         val[0] = context->render_offscreen ? 0.0f : rt_height;
 563         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 564         val[2] = 1.0f;
 565         val[3] = 0.0f;
 566         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 567         checkGLcall("y correction loading");
 568     }
 569
 570     if (!gl_shader->num_int_consts) return;
 571
 572     for(i = 0; i < MAX_CONST_I; i++)
 573     {
 574         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 575         {
 576             float val[4];
 577             val[0] = (float)state->ps_consts_i[4 * i];
 578             val[1] = (float)state->ps_consts_i[4 * i + 1];
 579             val[2] = (float)state->ps_consts_i[4 * i + 2];
 580             val[3] = -1.0f;
 581
 582             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 583         }
 584     }
 585     checkGLcall("Load ps int consts");
 586 }
 587
 588 /* GL locking is done by the caller. */
 589 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 590         const struct wined3d_context *context, const struct wined3d_state *state)
 591 {
 592     const struct wined3d_gl_info *gl_info = context->gl_info;
 593     float position_fixup[4];
 594     unsigned char i;
 595
 596     /* Upload the position fixup */
 597     shader_get_position_fixup(context, state, position_fixup);
 598     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 599
 600     if (!gl_shader->num_int_consts) return;
 601
 602     for(i = 0; i < MAX_CONST_I; i++)
 603     {
 604         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 605         {
 606             float val[4];
 607             val[0] = (float)state->vs_consts_i[4 * i];
 608             val[1] = (float)state->vs_consts_i[4 * i + 1];
 609             val[2] = (float)state->vs_consts_i[4 * i + 2];
 610             val[3] = -1.0f;
 611
 612             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 613         }
 614     }
 615     checkGLcall("Load vs int consts");
 616 }
 617
 618 /**
 619  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 620  *
 621  * We only support float constants in ARB at the moment, so don't
 622  * worry about the Integers or Booleans
 623  */
 624 /* GL locking is done by the caller (state handler) */
 625 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 626 {
 627     IWineD3DDeviceImpl *device = context->swapchain->device;
 628     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 629     const struct wined3d_gl_info *gl_info = context->gl_info;
 630     struct shader_arb_priv *priv = device->shader_priv;
 631
 632     if (useVertexShader)
 633     {
 634         IWineD3DBaseShaderImpl *vshader = (IWineD3DBaseShaderImpl *)stateBlock->state.vertex_shader;
 635         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 636
 637         /* Load DirectX 9 float constants for vertex shader */
 638         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 639                 device->highest_dirty_vs_const, stateBlock->state.vs_consts_f, context->vshader_const_dirty);
 640         shader_arb_vs_local_constants(gl_shader, context, &stateBlock->state);
 641     }
 642
 643     if (usePixelShader)
 644     {
 645         IWineD3DBaseShaderImpl *pshader = (IWineD3DBaseShaderImpl *)stateBlock->state.pixel_shader;
 646         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 647         float rt_height = device->render_targets[0]->currentDesc.Height;
 648
 649         /* Load DirectX 9 float constants for pixel shader */
 650         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 651                 device->highest_dirty_ps_const, stateBlock->state.ps_consts_f, context->pshader_const_dirty);
 652         shader_arb_ps_local_constants(gl_shader, context, &stateBlock->state, rt_height);
 653     }
 654 }
 655
 656 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 657 {
 658     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 659     struct wined3d_context *context = context_get_current();
 660
 661     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 662      * context. On a context switch the old context will be fully dirtified */
 663     if (!context || context->swapchain->device != This) return;
 664
 665     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 666     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 667 }
 668
 669 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 670 {
 671     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 672     struct wined3d_context *context = context_get_current();
 673
 674     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 675      * context. On a context switch the old context will be fully dirtified */
 676     if (!context || context->swapchain->device != This) return;
 677
 678     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 679     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 680 }
 681
 682 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 683 {
 684     DWORD *ret;
 685     DWORD idx = 0;
 686     const local_constant *lconst;
 687
 688     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 689
 690     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 691     if(!ret) {
 692         ERR("Out of memory\n");
 693         return NULL;
 694     }
 695
 696     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 697         ret[lconst->idx] = idx++;
 698     }
 699     return ret;
 700 }
 701
 702 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 703 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 704         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 705         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 706 {
 707     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 708     DWORD i, next_local = 0;
 709     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 710     unsigned max_constantsF;
 711     const local_constant *lconst;
 712     DWORD map;
 713
 714     /* In pixel shaders, all private constants are program local, we don't need anything
 715      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 716      * If we need a private constant the GL implementation will squeeze it in somewhere
 717      *
 718      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 719      * immediate values. The posFixup is loaded using program.env for now, so always
 720      * subtract one from the number of constants. If the shader uses indirect addressing,
 721      * account for the helper const too because we have to declare all availabke d3d constants
 722      * and don't know which are actually used.
 723      */
 724     if (pshader)
 725     {
 726         max_constantsF = gl_info->limits.arb_ps_native_constants;
 727         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 728         if (max_constantsF < 24)
 729             max_constantsF = gl_info->limits.arb_ps_float_constants;
 730     }
 731     else
 732     {
 733         max_constantsF = gl_info->limits.arb_vs_native_constants;
 734         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 735          * Also prevents max_constantsF from becoming less than 0 and
 736          * wrapping . */
 737         if (max_constantsF < 96)
 738             max_constantsF = gl_info->limits.arb_vs_float_constants;
 739
 740         if(This->baseShader.reg_maps.usesrelconstF) {
 741             DWORD highest_constf = 0, clip_limit;
 742
 743             max_constantsF -= reserved_vs_const(This, gl_info);
 744             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 745
 746             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 747             {
 748                 DWORD idx = i >> 5;
 749                 DWORD shift = i & 0x1f;
 750                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 751             }
 752
 753             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 754             {
 755                 if(ctx->cur_vs_args->super.clip_enabled)
 756                     clip_limit = gl_info->limits.clipplanes;
 757                 else
 758                     clip_limit = 0;
 759             }
 760             else
 761             {
 762                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 763                 clip_limit = min(count_bits(mask), 4);
 764             }
 765             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 766             max_constantsF -= *num_clipplanes;
 767             if(*num_clipplanes < clip_limit)
 768             {
 769                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 770             }
 771         }
 772         else
 773         {
 774             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 775             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 776         }
 777     }
 778
 779     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 780     {
 781         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 782     }
 783
 784     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 785     {
 786         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 787     }
 788
 789     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 790     {
 791         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 792         {
 793             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 794         }
 795     }
 796
 797     /* Load local constants using the program-local space,
 798      * this avoids reloading them each time the shader is used
 799      */
 800     if(lconst_map) {
 801         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 802             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 803                            lconst_map[lconst->idx]);
 804             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 805         }
 806     }
 807
 808     /* After subtracting privately used constants from the hardware limit(they are loaded as
 809      * local constants), make sure the shader doesn't violate the env constant limit
 810      */
 811     if(pshader)
 812     {
 813         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 814     }
 815     else
 816     {
 817         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 818     }
 819
 820     /* Avoid declaring more constants than needed */
 821     max_constantsF = min(max_constantsF, This->baseShader.limits.constant_float);
 822
 823     /* we use the array-based constants array if the local constants are marked for loading,
 824      * because then we use indirect addressing, or when the local constant list is empty,
 825      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 826      * local constants do not declare the loaded constants as an array because ARB compilers usually
 827      * do not optimize unused constants away
 828      */
 829     if(This->baseShader.reg_maps.usesrelconstF) {
 830         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 831         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 832                     max_constantsF, max_constantsF - 1);
 833     } else {
 834         for(i = 0; i < max_constantsF; i++) {
 835             DWORD idx, mask;
 836             idx = i >> 5;
 837             mask = 1 << (i & 0x1f);
 838             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 839                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 840             }
 841         }
 842     }
 843
 844     return next_local;
 845 }
 846
 847 static const char * const shift_tab[] = {
 848     "dummy",     /*  0 (none) */
 849     "coefmul.x", /*  1 (x2)   */
 850     "coefmul.y", /*  2 (x4)   */
 851     "coefmul.z", /*  3 (x8)   */
 852     "coefmul.w", /*  4 (x16)  */
 853     "dummy",     /*  5 (x32)  */
 854     "dummy",     /*  6 (x64)  */
 855     "dummy",     /*  7 (x128) */
 856     "dummy",     /*  8 (d256) */
 857     "dummy",     /*  9 (d128) */
 858     "dummy",     /* 10 (d64)  */
 859     "dummy",     /* 11 (d32)  */
 860     "coefdiv.w", /* 12 (d16)  */
 861     "coefdiv.z", /* 13 (d8)   */
 862     "coefdiv.y", /* 14 (d4)   */
 863     "coefdiv.x"  /* 15 (d2)   */
 864 };
 865
 866 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 867         const struct wined3d_shader_dst_param *dst, char *write_mask)
 868 {
 869     char *ptr = write_mask;
 870
 871     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 872     {
 873         *ptr++ = '.';
 874         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 875         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 876         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 877         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 878     }
 879
 880     *ptr = '\0';
 881 }
 882
 883 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 884 {
 885     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 886      * but addressed as "rgba". To fix this we need to swap the register's x
 887      * and z components. */
 888     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 889     char *ptr = swizzle_str;
 890
 891     /* swizzle bits fields: wwzzyyxx */
 892     DWORD swizzle = param->swizzle;
 893     DWORD swizzle_x = swizzle & 0x03;
 894     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 895     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 896     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 897
 898     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 899      * generate a swizzle string. Unless we need to our own swizzling. */
 900     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 901     {
 902         *ptr++ = '.';
 903         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 904             *ptr++ = swizzle_chars[swizzle_x];
 905         } else {
 906             *ptr++ = swizzle_chars[swizzle_x];
 907             *ptr++ = swizzle_chars[swizzle_y];
 908             *ptr++ = swizzle_chars[swizzle_z];
 909             *ptr++ = swizzle_chars[swizzle_w];
 910         }
 911     }
 912
 913     *ptr = '\0';
 914 }
 915
 916 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 917 {
 918     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 919     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 920
 921     if (!strcmp(priv->addr_reg, src)) return;
 922
 923     strcpy(priv->addr_reg, src);
 924     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 925 }
 926
 927 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 928         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 929
 930 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 931         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 932 {
 933     /* oPos, oFog and oPts in D3D */
 934     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 935     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 936     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 937     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 938     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 939
 940     *is_color = FALSE;
 941
 942     switch (reg->type)
 943     {
 944         case WINED3DSPR_TEMP:
 945             sprintf(register_name, "R%u", reg->idx);
 946             break;
 947
 948         case WINED3DSPR_INPUT:
 949             if (pshader)
 950             {
 951                 if (reg_maps->shader_version.major < 3)
 952                 {
 953                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 954                     else strcpy(register_name, "fragment.color.secondary");
 955                 }
 956                 else
 957                 {
 958                     if(reg->rel_addr)
 959                     {
 960                         char rel_reg[50];
 961                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 962
 963                         if (!strcmp(rel_reg, "**aL_emul**"))
 964                         {
 965                             DWORD idx = ctx->aL + reg->idx;
 966                             if(idx < MAX_REG_INPUT)
 967                             {
 968                                 strcpy(register_name, ctx->ps_input[idx]);
 969                             }
 970                             else
 971                             {
 972                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 973                                 sprintf(register_name, "out_of_bounds_%u", idx);
 974                             }
 975                         }
 976                         else if (reg_maps->input_registers & 0x0300)
 977                         {
 978                             /* There are two ways basically:
 979                              *
 980                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 981                              *    That means trouble if the loop also contains a breakc or if the control values
 982                              *    aren't local constants.
 983                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 984                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 985                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 986                              *    ADAC to load the condition code register and pop it again afterwards
 987                              */
 988                             FIXME("Relative input register addressing with more than 8 registers\n");
 989
 990                             /* This is better than nothing for now */
 991                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 992                         }
 993                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 994                         {
 995                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 996                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 997                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 998                              * proper varyings, or loop unrolling
 999                              *
1000                              * For now use the texcoords and hope for the best
1001                              */
1002                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1003                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1004                         }
1005                         else
1006                         {
1007                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1008                              * pulls GL_NV_fragment_program2 in
1009                              */
1010                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1011                         }
1012                     }
1013                     else
1014                     {
1015                         if(reg->idx < MAX_REG_INPUT)
1016                         {
1017                             strcpy(register_name, ctx->ps_input[reg->idx]);
1018                         }
1019                         else
1020                         {
1021                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1022                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1023                         }
1024                     }
1025                 }
1026             }
1027             else
1028             {
1029                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1030                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1031             }
1032             break;
1033
1034         case WINED3DSPR_CONST:
1035             if (!pshader && reg->rel_addr)
1036             {
1037                 BOOL aL = FALSE;
1038                 char rel_reg[50];
1039                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
1040                 if (reg_maps->shader_version.major < 2)
1041                 {
1042                     sprintf(rel_reg, "A0.x");
1043                 } else {
1044                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1045                     if(ctx->target_version == ARB) {
1046                         if (!strcmp(rel_reg, "**aL_emul**"))
1047                         {
1048                             aL = TRUE;
1049                         } else {
1050                             shader_arb_request_a0(ins, rel_reg);
1051                             sprintf(rel_reg, "A0.x");
1052                         }
1053                     }
1054                 }
1055                 if(aL)
1056                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1057                 else if (reg->idx >= rel_offset)
1058                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1059                 else
1060                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1061             }
1062             else
1063             {
1064                 if (reg_maps->usesrelconstF)
1065                     sprintf(register_name, "C[%u]", reg->idx);
1066                 else
1067                     sprintf(register_name, "C%u", reg->idx);
1068             }
1069             break;
1070
1071         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1072             if (pshader)
1073             {
1074                 if (reg_maps->shader_version.major == 1
1075                         && reg_maps->shader_version.minor <= 3)
1076                 {
1077                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1078                      * and as source to most instructions. For some instructions it is the texcoord
1079                      * input. Those instructions know about the special use
1080                      */
1081                     sprintf(register_name, "T%u", reg->idx);
1082                 } else {
1083                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1084                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1085                 }
1086             }
1087             else
1088             {
1089                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1090                 {
1091                     sprintf(register_name, "A%u", reg->idx);
1092                 }
1093                 else
1094                 {
1095                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1096                 }
1097             }
1098             break;
1099
1100         case WINED3DSPR_COLOROUT:
1101             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1102             {
1103                 strcpy(register_name, "TMP_COLOR");
1104             }
1105             else
1106             {
1107                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1108                 if (reg_maps->highest_render_target > 0)
1109                 {
1110                     sprintf(register_name, "result.color[%u]", reg->idx);
1111                 }
1112                 else
1113                 {
1114                     strcpy(register_name, "result.color");
1115                 }
1116             }
1117             break;
1118
1119         case WINED3DSPR_RASTOUT:
1120             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1121             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1122             break;
1123
1124         case WINED3DSPR_DEPTHOUT:
1125             strcpy(register_name, "result.depth");
1126             break;
1127
1128         case WINED3DSPR_ATTROUT:
1129         /* case WINED3DSPR_OUTPUT: */
1130             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1131             else strcpy(register_name, ctx->color_output[reg->idx]);
1132             break;
1133
1134         case WINED3DSPR_TEXCRDOUT:
1135             if (pshader)
1136             {
1137                 sprintf(register_name, "oT[%u]", reg->idx);
1138             }
1139             else
1140             {
1141                 if (reg_maps->shader_version.major < 3)
1142                 {
1143                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1144                 }
1145                 else
1146                 {
1147                     strcpy(register_name, ctx->vs_output[reg->idx]);
1148                 }
1149             }
1150             break;
1151
1152         case WINED3DSPR_LOOP:
1153             if(ctx->target_version >= NV2)
1154             {
1155                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1156                 if(pshader) sprintf(register_name, "A0.x");
1157                 else sprintf(register_name, "aL.y");
1158             }
1159             else
1160             {
1161                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1162                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1163                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1164                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1165                  * indexing
1166                  */
1167                 sprintf(register_name, "**aL_emul**");
1168             }
1169
1170             break;
1171
1172         case WINED3DSPR_CONSTINT:
1173             sprintf(register_name, "I%u", reg->idx);
1174             break;
1175
1176         case WINED3DSPR_MISCTYPE:
1177             if (!reg->idx)
1178             {
1179                 sprintf(register_name, "vpos");
1180             }
1181             else if(reg->idx == 1)
1182             {
1183                 sprintf(register_name, "fragment.facing.x");
1184             }
1185             else
1186             {
1187                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1188             }
1189             break;
1190
1191         default:
1192             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1193             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1194             break;
1195     }
1196 }
1197
1198 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1199         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1200 {
1201     char register_name[255];
1202     char write_mask[6];
1203     BOOL is_color;
1204
1205     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1206     strcpy(str, register_name);
1207
1208     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1209     strcat(str, write_mask);
1210 }
1211
1212 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1213 {
1214     switch(channel_source)
1215     {
1216         case CHANNEL_SOURCE_ZERO: return "0";
1217         case CHANNEL_SOURCE_ONE: return "1";
1218         case CHANNEL_SOURCE_X: return "x";
1219         case CHANNEL_SOURCE_Y: return "y";
1220         case CHANNEL_SOURCE_Z: return "z";
1221         case CHANNEL_SOURCE_W: return "w";
1222         default:
1223             FIXME("Unhandled channel source %#x\n", channel_source);
1224             return "undefined";
1225     }
1226 }
1227
1228 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1229         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1230 {
1231     DWORD mask;
1232
1233     if (is_complex_fixup(fixup))
1234     {
1235         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1236         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1237         return;
1238     }
1239
1240     mask = 0;
1241     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1242     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1243     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1244     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1245     mask &= dst_mask;
1246
1247     if (mask)
1248     {
1249         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1250                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1251                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1252     }
1253
1254     mask = 0;
1255     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1256     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1257     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1258     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1259     mask &= dst_mask;
1260
1261     if (mask)
1262     {
1263         char reg_mask[6];
1264         char *ptr = reg_mask;
1265
1266         if (mask != WINED3DSP_WRITEMASK_ALL)
1267         {
1268             *ptr++ = '.';
1269             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1270             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1271             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1272             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1273         }
1274         *ptr = '\0';
1275
1276         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1277     }
1278 }
1279
1280 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1281 {
1282     DWORD mod;
1283     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1284     if (!ins->dst_count) return "";
1285
1286     mod = ins->dst[0].modifiers;
1287
1288     /* Silently ignore PARTIALPRECISION if its not supported */
1289     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1290
1291     if(mod & WINED3DSPDM_MSAMPCENTROID)
1292     {
1293         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1294         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1295     }
1296
1297     switch(mod)
1298     {
1299         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1300             return "H_SAT";
1301
1302         case WINED3DSPDM_SATURATE:
1303             return "_SAT";
1304
1305         case WINED3DSPDM_PARTIALPRECISION:
1306             return "H";
1307
1308         case 0:
1309             return "";
1310
1311         default:
1312             FIXME("Unknown modifiers 0x%08x\n", mod);
1313             return "";
1314     }
1315 }
1316
1317 #define TEX_PROJ        0x1
1318 #define TEX_BIAS        0x2
1319 #define TEX_LOD         0x4
1320 #define TEX_DERIV       0x10
1321
1322 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1323         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1324 {
1325     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1326     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1327     IWineD3DBaseTextureImpl *texture;
1328     const char *tex_type;
1329     BOOL np2_fixup = FALSE;
1330     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1331     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1332     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1333     const char *mod;
1334     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1335
1336     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1337     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1338
1339     switch(sampler_type) {
1340         case WINED3DSTT_1D:
1341             tex_type = "1D";
1342             break;
1343
1344         case WINED3DSTT_2D:
1345             texture = device->stateBlock->state.textures[sampler_idx];
1346             if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1347             {
1348                 tex_type = "RECT";
1349             } else {
1350                 tex_type = "2D";
1351             }
1352             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1353             {
1354                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1355                 {
1356                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1357                     else np2_fixup = TRUE;
1358                 }
1359             }
1360             break;
1361
1362         case WINED3DSTT_VOLUME:
1363             tex_type = "3D";
1364             break;
1365
1366         case WINED3DSTT_CUBE:
1367             tex_type = "CUBE";
1368             break;
1369
1370         default:
1371             ERR("Unexpected texture type %d\n", sampler_type);
1372             tex_type = "";
1373     }
1374
1375     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1376      * so don't use shader_arb_get_modifier
1377      */
1378     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1379     else mod = "";
1380
1381     /* Fragment samplers always have indentity mapping */
1382     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1383     {
1384         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1385     }
1386
1387     if (flags & TEX_DERIV)
1388     {
1389         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1390         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1391         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1392                        dsx, dsy,sampler_idx, tex_type);
1393     }
1394     else if(flags & TEX_LOD)
1395     {
1396         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1397         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1398         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1399                        sampler_idx, tex_type);
1400     }
1401     else if (flags & TEX_BIAS)
1402     {
1403         /* Shouldn't be possible, but let's check for it */
1404         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1405         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1406         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1407     }
1408     else if (flags & TEX_PROJ)
1409     {
1410         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1411     }
1412     else
1413     {
1414         if (np2_fixup)
1415         {
1416             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1417             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1418                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1419
1420             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1421         }
1422         else
1423             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1424     }
1425
1426     if (pshader)
1427     {
1428         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1429                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1430                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1431                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1432     }
1433 }
1434
1435 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1436         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1437 {
1438     /* Generate a line that does the input modifier computation and return the input register to use */
1439     BOOL is_color = FALSE;
1440     char regstr[256];
1441     char swzstr[20];
1442     int insert_line;
1443     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1444     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1445     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1446     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1447
1448     /* Assume a new line will be added */
1449     insert_line = 1;
1450
1451     /* Get register name */
1452     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1453     shader_arb_get_swizzle(src, is_color, swzstr);
1454
1455     switch (src->modifiers)
1456     {
1457     case WINED3DSPSM_NONE:
1458         sprintf(outregstr, "%s%s", regstr, swzstr);
1459         insert_line = 0;
1460         break;
1461     case WINED3DSPSM_NEG:
1462         sprintf(outregstr, "-%s%s", regstr, swzstr);
1463         insert_line = 0;
1464         break;
1465     case WINED3DSPSM_BIAS:
1466         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1467         break;
1468     case WINED3DSPSM_BIASNEG:
1469         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1470         break;
1471     case WINED3DSPSM_SIGN:
1472         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1473         break;
1474     case WINED3DSPSM_SIGNNEG:
1475         shader_addline(buffer, "MAD T%c, %s, %s, %s;\n", 'A' + tmpreg, regstr, two, one);
1476         break;
1477     case WINED3DSPSM_COMP:
1478         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1479         break;
1480     case WINED3DSPSM_X2:
1481         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1482         break;
1483     case WINED3DSPSM_X2NEG:
1484         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1485         break;
1486     case WINED3DSPSM_DZ:
1487         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1488         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1489         break;
1490     case WINED3DSPSM_DW:
1491         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1492         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1493         break;
1494     case WINED3DSPSM_ABS:
1495         if(ctx->target_version >= NV2) {
1496             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1497             insert_line = 0;
1498         } else {
1499             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1500         }
1501         break;
1502     case WINED3DSPSM_ABSNEG:
1503         if(ctx->target_version >= NV2) {
1504             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1505         } else {
1506             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1507             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1508         }
1509         insert_line = 0;
1510         break;
1511     default:
1512         sprintf(outregstr, "%s%s", regstr, swzstr);
1513         insert_line = 0;
1514     }
1515
1516     /* Return modified or original register, with swizzle */
1517     if (insert_line)
1518         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1519 }
1520
1521 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1522 {
1523     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1524     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1525     char dst_name[50];
1526     char src_name[2][50];
1527     DWORD sampler_code = dst->reg.idx;
1528
1529     shader_arb_get_dst_param(ins, dst, dst_name);
1530
1531     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1532      *
1533      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1534      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1535      * temps is done.
1536      */
1537     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1538     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1539     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1540     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1541     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1542
1543     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1544     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1545 }
1546
1547 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1548 {
1549     *extra_char = ' ';
1550     switch(mod)
1551     {
1552         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1553         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1554         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1555         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1556         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1557         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1558         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1559         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1560         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1561         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1562         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1563         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1564         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1565     }
1566     FIXME("Unknown modifier %u\n", mod);
1567     return mod;
1568 }
1569
1570 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1571 {
1572     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1573     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1574     char dst_name[50];
1575     char src_name[3][50];
1576     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1577             ins->ctx->reg_maps->shader_version.minor);
1578     BOOL is_color;
1579
1580     shader_arb_get_dst_param(ins, dst, dst_name);
1581     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1582
1583     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1584     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1585     {
1586         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1587     } else {
1588         struct wined3d_shader_src_param src0_copy = ins->src[0];
1589         char extra_neg;
1590
1591         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1592         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1593
1594         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1595         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1596         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1597         /* No modifiers supported on CMP */
1598         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1599
1600         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1601         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1602         {
1603             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1604             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1605         }
1606     }
1607 }
1608
1609 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1610 {
1611     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1612     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1613     char dst_name[50];
1614     char src_name[3][50];
1615     BOOL is_color;
1616
1617     shader_arb_get_dst_param(ins, dst, dst_name);
1618
1619     /* Generate input register names (with modifiers) */
1620     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1621     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1622     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1623
1624     /* No modifiers are supported on CMP */
1625     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1626                    src_name[0], src_name[2], src_name[1]);
1627
1628     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1629     {
1630         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1631         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1632     }
1633 }
1634
1635 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1636  * dst = dot2(src0, src1) + src2 */
1637 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1638 {
1639     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1640     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1641     char dst_name[50];
1642     char src_name[3][50];
1643     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1644
1645     shader_arb_get_dst_param(ins, dst, dst_name);
1646     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1647     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1648
1649     if(ctx->target_version >= NV3)
1650     {
1651         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1652         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1653         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1654                        dst_name, src_name[0], src_name[1], src_name[2]);
1655     }
1656     else if(ctx->target_version >= NV2)
1657     {
1658         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1659          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1660          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1661          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1662          *
1663          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1664          *
1665          * .xyxy and other swizzles that we could get with this are not valid in
1666          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1667          */
1668         struct wined3d_shader_src_param tmp_param = ins->src[1];
1669         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1670         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1671
1672         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1673
1674         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1675                        dst_name, src_name[2], src_name[0], src_name[1]);
1676     }
1677     else
1678     {
1679         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1680         /* 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]
1681         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1682         */
1683         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1684         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1685         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1686         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1687     }
1688 }
1689
1690 /* Map the opcode 1-to-1 to the GL code */
1691 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1692 {
1693     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1694     const char *instruction;
1695     char arguments[256], dst_str[50];
1696     unsigned int i;
1697     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1698
1699     switch (ins->handler_idx)
1700     {
1701         case WINED3DSIH_ABS: instruction = "ABS"; break;
1702         case WINED3DSIH_ADD: instruction = "ADD"; break;
1703         case WINED3DSIH_CRS: instruction = "XPD"; break;
1704         case WINED3DSIH_DP3: instruction = "DP3"; break;
1705         case WINED3DSIH_DP4: instruction = "DP4"; break;
1706         case WINED3DSIH_DST: instruction = "DST"; break;
1707         case WINED3DSIH_FRC: instruction = "FRC"; break;
1708         case WINED3DSIH_LIT: instruction = "LIT"; break;
1709         case WINED3DSIH_LRP: instruction = "LRP"; break;
1710         case WINED3DSIH_MAD: instruction = "MAD"; break;
1711         case WINED3DSIH_MAX: instruction = "MAX"; break;
1712         case WINED3DSIH_MIN: instruction = "MIN"; break;
1713         case WINED3DSIH_MOV: instruction = "MOV"; break;
1714         case WINED3DSIH_MUL: instruction = "MUL"; break;
1715         case WINED3DSIH_SGE: instruction = "SGE"; break;
1716         case WINED3DSIH_SLT: instruction = "SLT"; break;
1717         case WINED3DSIH_SUB: instruction = "SUB"; break;
1718         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1719         case WINED3DSIH_DSX: instruction = "DDX"; break;
1720         default: instruction = "";
1721             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1722             break;
1723     }
1724
1725     /* Note that shader_arb_add_dst_param() adds spaces. */
1726     arguments[0] = '\0';
1727     shader_arb_get_dst_param(ins, dst, dst_str);
1728     for (i = 0; i < ins->src_count; ++i)
1729     {
1730         char operand[100];
1731         strcat(arguments, ", ");
1732         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1733         strcat(arguments, operand);
1734     }
1735     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1736 }
1737
1738 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1739 {
1740     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1741     shader_addline(buffer, "NOP;\n");
1742 }
1743
1744 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1745 {
1746     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1747     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1748     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1749     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1750     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1751     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1752     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1753
1754     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1755     char src0_param[256];
1756
1757     if(ins->handler_idx == WINED3DSIH_MOVA) {
1758         char write_mask[6];
1759         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1760
1761         if(ctx->target_version >= NV2) {
1762             shader_hw_map2gl(ins);
1763             return;
1764         }
1765         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1766         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1767
1768         /* This implements the mova formula used in GLSL. The first two instructions
1769          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1770          * in this case:
1771          * mova A0.x, 0.0
1772          *
1773          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1774          *
1775          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1776          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1777          */
1778         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1779         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1780
1781         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1782         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1783         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1784         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1785         {
1786             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1787         }
1788         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1789
1790         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1791     }
1792     else if (reg_maps->shader_version.major == 1
1793           && !shader_is_pshader_version(reg_maps->shader_version.type)
1794           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1795     {
1796         src0_param[0] = '\0';
1797         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1798         {
1799             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1800             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1801             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1802             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1803         }
1804         else
1805         {
1806             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1807              * with more than one component. Thus replicate the first source argument over all
1808              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1809             struct wined3d_shader_src_param tmp_src = ins->src[0];
1810             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1811             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1812             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1813         }
1814     }
1815     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1816     {
1817         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1818         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1819         {
1820             shader_addline(buffer, "#mov handled in srgb write code\n");
1821             return;
1822         }
1823         shader_hw_map2gl(ins);
1824     }
1825     else
1826     {
1827         shader_hw_map2gl(ins);
1828     }
1829 }
1830
1831 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1832 {
1833     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1834     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1835     char reg_dest[40];
1836
1837     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1838      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1839      */
1840     shader_arb_get_dst_param(ins, dst, reg_dest);
1841
1842     if (ins->ctx->reg_maps->shader_version.major >= 2)
1843     {
1844         const char *kilsrc = "TA";
1845         BOOL is_color;
1846
1847         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1848         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1849         {
1850             kilsrc = reg_dest;
1851         }
1852         else
1853         {
1854             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1855              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1856              * masked out components to 0(won't kill)
1857              */
1858             char x = '0', y = '0', z = '0', w = '0';
1859             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1860             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1861             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1862             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1863             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1864         }
1865         shader_addline(buffer, "KIL %s;\n", kilsrc);
1866     } else {
1867         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1868          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1869          *
1870          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1871          * or pass in any temporary register(in shader phase 2)
1872          */
1873         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1874             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1875         } else {
1876             shader_arb_get_dst_param(ins, dst, reg_dest);
1877         }
1878         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1879         shader_addline(buffer, "KIL TA;\n");
1880     }
1881 }
1882
1883 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1884 {
1885     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1886     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1887     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1888             ins->ctx->reg_maps->shader_version.minor);
1889     struct wined3d_shader_src_param src;
1890
1891     char reg_dest[40];
1892     char reg_coord[40];
1893     DWORD reg_sampler_code;
1894     WORD myflags = 0;
1895
1896     /* All versions have a destination register */
1897     shader_arb_get_dst_param(ins, dst, reg_dest);
1898
1899     /* 1.0-1.4: Use destination register number as texture code.
1900        2.0+: Use provided sampler number as texure code. */
1901     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1902         reg_sampler_code = dst->reg.idx;
1903     else
1904         reg_sampler_code = ins->src[1].reg.idx;
1905
1906     /* 1.0-1.3: Use the texcoord varying.
1907        1.4+: Use provided coordinate source register. */
1908     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1909         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1910     else {
1911         /* TEX is the only instruction that can handle DW and DZ natively */
1912         src = ins->src[0];
1913         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1914         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1915         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1916     }
1917
1918     /* projection flag:
1919      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1920      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1921      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1922      */
1923     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1924     {
1925         DWORD flags = 0;
1926         if (reg_sampler_code < MAX_TEXTURES)
1927             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1928         if (flags & WINED3D_PSARGS_PROJECTED)
1929             myflags |= TEX_PROJ;
1930     }
1931     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1932     {
1933         DWORD src_mod = ins->src[0].modifiers;
1934         if (src_mod == WINED3DSPSM_DZ) {
1935             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1936              * varying register, so we need a temp reg
1937              */
1938             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1939             strcpy(reg_coord, "TA");
1940             myflags |= TEX_PROJ;
1941         } else if(src_mod == WINED3DSPSM_DW) {
1942             myflags |= TEX_PROJ;
1943         }
1944     } else {
1945         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1946         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1947     }
1948     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1949 }
1950
1951 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1952 {
1953     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1954     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1955     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1956             ins->ctx->reg_maps->shader_version.minor);
1957     char dst_str[50];
1958
1959     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1960     {
1961         DWORD reg = dst->reg.idx;
1962
1963         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1964         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1965     } else {
1966         char reg_src[40];
1967
1968         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1969         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1970         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1971    }
1972 }
1973
1974 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1975 {
1976      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1977      DWORD flags = 0;
1978
1979      DWORD reg1 = ins->dst[0].reg.idx;
1980      char dst_str[50];
1981      char src_str[50];
1982
1983      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1984      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1985      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1986      /* Move .x first in case src_str is "TA" */
1987      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1988      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1989      if (reg1 < MAX_TEXTURES)
1990      {
1991          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1992          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1993      }
1994      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1995 }
1996
1997 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1998 {
1999      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2000
2001      DWORD reg1 = ins->dst[0].reg.idx;
2002      char dst_str[50];
2003      char src_str[50];
2004
2005      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2006      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2007      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2008      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2009      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2010      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2011 }
2012
2013 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2014 {
2015     DWORD reg1 = ins->dst[0].reg.idx;
2016     char dst_str[50];
2017     char src_str[50];
2018
2019     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2020     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2021     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2022     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2023 }
2024
2025 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2026 {
2027     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2028     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2029     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2030     char reg_coord[40], dst_reg[50], src_reg[50];
2031     DWORD reg_dest_code;
2032
2033     /* All versions have a destination register. The Tx where the texture coordinates come
2034      * from is the varying incarnation of the texture register
2035      */
2036     reg_dest_code = dst->reg.idx;
2037     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2038     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2039     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2040
2041     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2042      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2043      *
2044      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2045      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2046      *
2047      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2048      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2049      * extension.
2050      */
2051     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2052     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2053     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2054     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2055
2056     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2057      * so we can't let the GL handle this.
2058      */
2059     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2060             & WINED3D_PSARGS_PROJECTED)
2061     {
2062         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2063         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2064         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2065     } else {
2066         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2067     }
2068
2069     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2070
2071     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2072     {
2073         /* No src swizzles are allowed, so this is ok */
2074         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2075                        src_reg, reg_dest_code, reg_dest_code);
2076         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2077     }
2078 }
2079
2080 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2081 {
2082     DWORD reg = ins->dst[0].reg.idx;
2083     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2084     char src0_name[50], dst_name[50];
2085     BOOL is_color;
2086     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2087
2088     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2089     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2090      * T<reg+1> register. Use this register to store the calculated vector
2091      */
2092     tmp_reg.idx = reg + 1;
2093     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2094     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2095 }
2096
2097 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2098 {
2099     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2100     DWORD flags;
2101     DWORD reg = ins->dst[0].reg.idx;
2102     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2103     char dst_str[50];
2104     char src0_name[50];
2105     char dst_reg[50];
2106     BOOL is_color;
2107
2108     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2109     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2110
2111     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2112     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2113     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2114     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2115     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2116 }
2117
2118 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2119 {
2120     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2121     DWORD reg = ins->dst[0].reg.idx;
2122     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2123     char src0_name[50], dst_name[50];
2124     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2125     BOOL is_color;
2126
2127     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2128      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2129      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2130      */
2131     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2132     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2133
2134     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2135     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2136                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2137     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2138 }
2139
2140 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2141 {
2142     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2143     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2144     DWORD flags;
2145     DWORD reg = ins->dst[0].reg.idx;
2146     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2147     char dst_str[50];
2148     char src0_name[50], dst_name[50];
2149     BOOL is_color;
2150
2151     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2152     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2153     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2154
2155     /* Sample the texture using the calculated coordinates */
2156     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2157     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2158     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2159     tex_mx->current_row = 0;
2160 }
2161
2162 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2163 {
2164     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2165     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2166     DWORD flags;
2167     DWORD reg = ins->dst[0].reg.idx;
2168     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2169     char dst_str[50];
2170     char src0_name[50];
2171     char dst_reg[50];
2172     BOOL is_color;
2173
2174     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2175      * components for temporary data storage
2176      */
2177     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2178     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2179     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2180
2181     /* Construct the eye-ray vector from w coordinates */
2182     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2183     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2184     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2185
2186     /* Calculate reflection vector
2187      */
2188     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2189     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2190     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2191     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2192     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2193     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2194     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
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_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2200     tex_mx->current_row = 0;
2201 }
2202
2203 static void pshader_hw_texm3x3spec(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;
2209     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2210     char dst_str[50];
2211     char src0_name[50];
2212     char src1_name[50];
2213     char dst_reg[50];
2214     BOOL is_color;
2215
2216     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2217     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2218     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2219     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2220     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2221
2222     /* Calculate reflection vector.
2223      *
2224      *                   dot(N, E)
2225      * dst_reg.xyz = 2 * --------- * N - E
2226      *                   dot(N, N)
2227      *
2228      * Which normalizes the normal vector
2229      */
2230     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2231     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2232     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2233     shader_addline(buffer, "MUL %s.w, %s.w, TC.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, -%s;\n", dst_reg, dst_reg, src1_name);
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 & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2241     tex_mx->current_row = 0;
2242 }
2243
2244 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2245 {
2246     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2247     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2248     char dst_name[50];
2249     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2250     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2251
2252     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2253      * which is essentially an input, is the destination register because it is the first
2254      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2255      * here(writemasks/swizzles are not valid on texdepth)
2256      */
2257     shader_arb_get_dst_param(ins, dst, dst_name);
2258
2259     /* According to the msdn, the source register(must be r5) is unusable after
2260      * the texdepth instruction, so we're free to modify it
2261      */
2262     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2263
2264     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2265      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2266      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2267      */
2268     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2269     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2270     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2271     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2272 }
2273
2274 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2275  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2276  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2277 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2278 {
2279     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2280     DWORD sampler_idx = ins->dst[0].reg.idx;
2281     char src0[50];
2282     char dst_str[50];
2283
2284     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2285     shader_addline(buffer, "MOV TB, 0.0;\n");
2286     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2287
2288     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2289     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2290 }
2291
2292 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2293  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2294 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2295 {
2296     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2297     char src0[50];
2298     char dst_str[50];
2299     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2300
2301     /* Handle output register */
2302     shader_arb_get_dst_param(ins, dst, dst_str);
2303     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2304     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2305 }
2306
2307 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2308  * Perform the 3rd row of a 3x3 matrix multiply */
2309 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2310 {
2311     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2312     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2313     char dst_str[50], dst_name[50];
2314     char src0[50];
2315     BOOL is_color;
2316
2317     shader_arb_get_dst_param(ins, dst, dst_str);
2318     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2319     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2320     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2321     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2322 }
2323
2324 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2325  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2326  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2327  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2328  */
2329 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2330 {
2331     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2332     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2333     char src0[50], dst_name[50];
2334     BOOL is_color;
2335     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2336     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2337
2338     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2339     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2340     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2341
2342     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2343      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2344      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2345      */
2346     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2347     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2348     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2349     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2350 }
2351
2352 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2353     Vertex/Pixel shaders to ARB_vertex_program codes */
2354 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2355 {
2356     int i;
2357     int nComponents = 0;
2358     struct wined3d_shader_dst_param tmp_dst = {{0}};
2359     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2360     struct wined3d_shader_instruction tmp_ins;
2361
2362     memset(&tmp_ins, 0, sizeof(tmp_ins));
2363
2364     /* Set constants for the temporary argument */
2365     tmp_ins.ctx = ins->ctx;
2366     tmp_ins.dst_count = 1;
2367     tmp_ins.dst = &tmp_dst;
2368     tmp_ins.src_count = 2;
2369     tmp_ins.src = tmp_src;
2370
2371     switch(ins->handler_idx)
2372     {
2373         case WINED3DSIH_M4x4:
2374             nComponents = 4;
2375             tmp_ins.handler_idx = WINED3DSIH_DP4;
2376             break;
2377         case WINED3DSIH_M4x3:
2378             nComponents = 3;
2379             tmp_ins.handler_idx = WINED3DSIH_DP4;
2380             break;
2381         case WINED3DSIH_M3x4:
2382             nComponents = 4;
2383             tmp_ins.handler_idx = WINED3DSIH_DP3;
2384             break;
2385         case WINED3DSIH_M3x3:
2386             nComponents = 3;
2387             tmp_ins.handler_idx = WINED3DSIH_DP3;
2388             break;
2389         case WINED3DSIH_M3x2:
2390             nComponents = 2;
2391             tmp_ins.handler_idx = WINED3DSIH_DP3;
2392             break;
2393         default:
2394             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2395             break;
2396     }
2397
2398     tmp_dst = ins->dst[0];
2399     tmp_src[0] = ins->src[0];
2400     tmp_src[1] = ins->src[1];
2401     for (i = 0; i < nComponents; i++) {
2402         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2403         shader_hw_map2gl(&tmp_ins);
2404         ++tmp_src[1].reg.idx;
2405     }
2406 }
2407
2408 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2409 {
2410     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2411     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2412     const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2413
2414     char dst[50];
2415     char src[50];
2416
2417     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2418     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2419     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2420     {
2421         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2422          * .w is used
2423          */
2424         strcat(src, ".w");
2425     }
2426
2427     /* TODO: If the destination is readable, and not the same as the source, the destination
2428      * can be used instead of TA
2429      */
2430     if (priv->target_version >= NV2)
2431     {
2432         shader_addline(buffer, "MOVC TA.x, %s;\n", src);
2433         shader_addline(buffer, "MOV TA.x (EQ.x), %s;\n", flt_eps);
2434         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2435     }
2436     else
2437     {
2438         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2439         shader_addline(buffer, "ABS TA.x, %s;\n", src);
2440         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2441         shader_addline(buffer, "MAD TA.x, TA.y, %s, %s;\n", flt_eps, src);
2442         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2443     }
2444 }
2445
2446 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2447 {
2448     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2449     const char *instruction;
2450
2451     char dst[50];
2452     char src[50];
2453
2454     switch(ins->handler_idx)
2455     {
2456         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2457         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2458         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2459         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2460         default: instruction = "";
2461             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2462             break;
2463     }
2464
2465     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2466     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2467     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2468     {
2469         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2470          * .w is used
2471          */
2472         strcat(src, ".w");
2473     }
2474
2475     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2476 }
2477
2478 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2479 {
2480     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2481     char dst_name[50];
2482     char src_name[50];
2483     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2484     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2485     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2486
2487     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2488     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2489
2490     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2491      * otherwise NRM or RSQ would return NaN */
2492     if(pshader && priv->target_version >= NV3)
2493     {
2494         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2495          *
2496          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2497          */
2498         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2499         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2500         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2501     }
2502     else if(priv->target_version >= NV2)
2503     {
2504         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2505         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2506         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2507                        src_name);
2508     }
2509     else
2510     {
2511         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2512
2513         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2514         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2515          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2516          */
2517         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2518         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2519
2520         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2521         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2522         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2523                     src_name);
2524     }
2525 }
2526
2527 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2528 {
2529     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2530     char dst_name[50];
2531     char src_name[3][50];
2532
2533     /* ARB_fragment_program has a convenient LRP instruction */
2534     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2535         shader_hw_map2gl(ins);
2536         return;
2537     }
2538
2539     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2540     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2541     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2542     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2543
2544     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2545     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2546                    dst_name, src_name[0], src_name[2]);
2547 }
2548
2549 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2550 {
2551     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2552      * must contain fixed constants. So we need a separate function to filter those constants and
2553      * can't use map2gl
2554      */
2555     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2556     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2557     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2558     char dst_name[50];
2559     char src_name0[50], src_name1[50], src_name2[50];
2560     BOOL is_color;
2561
2562     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2563     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2564         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2565         /* No modifiers are supported on SCS */
2566         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2567
2568         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2569         {
2570             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2571             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2572         }
2573     } else if(priv->target_version >= NV2) {
2574         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2575
2576         /* Sincos writemask must be .x, .y or .xy */
2577         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2578             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2579         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2580             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2581     } else {
2582         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2583          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2584          *
2585          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2586          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2587          *
2588          * The constants we get are:
2589          *
2590          *  +1   +1,     -1     -1     +1      +1      -1       -1
2591          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2592          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2593          *
2594          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2595          *
2596          * (x/2)^2 = x^2 / 4
2597          * (x/2)^3 = x^3 / 8
2598          * (x/2)^4 = x^4 / 16
2599          * (x/2)^5 = x^5 / 32
2600          * etc
2601          *
2602          * To get the final result:
2603          * sin(x) = 2 * sin(x/2) * cos(x/2)
2604          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2605          * (from sin(x+y) and cos(x+y) rules)
2606          *
2607          * As per MSDN, dst.z is undefined after the operation, and so is
2608          * dst.x and dst.y if they're masked out by the writemask. Ie
2609          * sincos dst.y, src1, c0, c1
2610          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2611          * vsa.exe also stops with an error if the dest register is the same register as the source
2612          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2613          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2614          */
2615         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2616         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2617         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2618
2619         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2620         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2621         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2622         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2623         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2624         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2625
2626         /* sin(x/2)
2627          *
2628          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2629          * properly merge that with MULs in the code above?
2630          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2631          * we can merge the sine and cosine MAD rows to calculate them together.
2632          */
2633         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2634         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2635         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2636         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2637
2638         /* cos(x/2) */
2639         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2640         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2641         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2642
2643         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2644             /* cos x */
2645             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2646             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2647         }
2648         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2649             /* sin x */
2650             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2651             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2652         }
2653     }
2654 }
2655
2656 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2657 {
2658     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2659     char dst_name[50];
2660     char src_name[50];
2661     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2662
2663     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2664     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2665
2666     /* SGN is only valid in vertex shaders */
2667     if(ctx->target_version >= NV2) {
2668         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2669         return;
2670     }
2671
2672     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2673      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2674      */
2675     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2676         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2677     } else {
2678         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2679          * Then use TA, and calculate the final result
2680          *
2681          * Not reading from TA? Store the first result in TA to avoid overwriting the
2682          * destination if src reg = dst reg
2683          */
2684         if(strstr(src_name, "TA"))
2685         {
2686             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2687             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2688             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2689         }
2690         else
2691         {
2692             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2693             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2694             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2695         }
2696     }
2697 }
2698
2699 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2700 {
2701     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2702     char src[50];
2703     char dst[50];
2704     char dst_name[50];
2705     BOOL is_color;
2706
2707     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2708     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2709     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2710
2711     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2712     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2713 }
2714
2715 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2716 {
2717     *need_abs = FALSE;
2718
2719     switch(mod)
2720     {
2721         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2722         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2723         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2724         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2725         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2726         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2727         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2728         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2729         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2730         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2731         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2732         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2733         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2734     }
2735     FIXME("Unknown modifier %u\n", mod);
2736     return mod;
2737 }
2738
2739 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2740 {
2741     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2742     char src0[50], src1[50], dst[50];
2743     struct wined3d_shader_src_param src0_copy = ins->src[0];
2744     BOOL need_abs = FALSE;
2745     const char *instr;
2746     BOOL arg2 = FALSE;
2747
2748     switch(ins->handler_idx)
2749     {
2750         case WINED3DSIH_LOG:  instr = "LG2"; break;
2751         case WINED3DSIH_LOGP: instr = "LOG"; break;
2752         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2753         default:
2754             ERR("Unexpected instruction %d\n", ins->handler_idx);
2755             return;
2756     }
2757
2758     /* LOG, LOGP and POW operate on the absolute value of the input */
2759     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2760
2761     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2762     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2763     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2764
2765     if(need_abs)
2766     {
2767         shader_addline(buffer, "ABS TA, %s;\n", src0);
2768         if(arg2)
2769         {
2770             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2771         }
2772         else
2773         {
2774             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2775         }
2776     }
2777     else if(arg2)
2778     {
2779         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2780     }
2781     else
2782     {
2783         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2784     }
2785 }
2786
2787 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2788 {
2789     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2790     char src_name[50];
2791     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2792
2793     /* src0 is aL */
2794     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2795
2796     if(vshader)
2797     {
2798         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2799         struct list *e = list_head(&priv->control_frames);
2800         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2801
2802         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2803         /* The constant loader makes sure to load -1 into iX.w */
2804         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2805         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2806         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2807     }
2808     else
2809     {
2810         shader_addline(buffer, "LOOP %s;\n", src_name);
2811     }
2812 }
2813
2814 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2815 {
2816     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2817     char src_name[50];
2818     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2819
2820     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2821
2822     /* The constant loader makes sure to load -1 into iX.w */
2823     if(vshader)
2824     {
2825         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2826         struct list *e = list_head(&priv->control_frames);
2827         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2828
2829         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2830
2831         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2832         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2833         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2834     }
2835     else
2836     {
2837         shader_addline(buffer, "REP %s;\n", src_name);
2838     }
2839 }
2840
2841 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2842 {
2843     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2844     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2845
2846     if(vshader)
2847     {
2848         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2849         struct list *e = list_head(&priv->control_frames);
2850         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2851
2852         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2853         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2854         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2855
2856         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2857     }
2858     else
2859     {
2860         shader_addline(buffer, "ENDLOOP;\n");
2861     }
2862 }
2863
2864 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2865 {
2866     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2867     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2868
2869     if(vshader)
2870     {
2871         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2872         struct list *e = list_head(&priv->control_frames);
2873         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2874
2875         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2876         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2877         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2878
2879         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2880     }
2881     else
2882     {
2883         shader_addline(buffer, "ENDREP;\n");
2884     }
2885 }
2886
2887 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2888 {
2889     struct control_frame *control_frame;
2890
2891     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2892     {
2893         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2894     }
2895     ERR("Could not find loop for break\n");
2896     return NULL;
2897 }
2898
2899 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2900 {
2901     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2902     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2903     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2904
2905     if(vshader)
2906     {
2907         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2908     }
2909     else
2910     {
2911         shader_addline(buffer, "BRK;\n");
2912     }
2913 }
2914
2915 static const char *get_compare(COMPARISON_TYPE flags)
2916 {
2917     switch (flags)
2918     {
2919         case COMPARISON_GT: return "GT";
2920         case COMPARISON_EQ: return "EQ";
2921         case COMPARISON_GE: return "GE";
2922         case COMPARISON_LT: return "LT";
2923         case COMPARISON_NE: return "NE";
2924         case COMPARISON_LE: return "LE";
2925         default:
2926             FIXME("Unrecognized comparison value: %u\n", flags);
2927             return "(\?\?)";
2928     }
2929 }
2930
2931 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2932 {
2933     switch (flags)
2934     {
2935         case COMPARISON_GT: return COMPARISON_LE;
2936         case COMPARISON_EQ: return COMPARISON_NE;
2937         case COMPARISON_GE: return COMPARISON_LT;
2938         case COMPARISON_LT: return COMPARISON_GE;
2939         case COMPARISON_NE: return COMPARISON_EQ;
2940         case COMPARISON_LE: return COMPARISON_GT;
2941         default:
2942             FIXME("Unrecognized comparison value: %u\n", flags);
2943             return -1;
2944     }
2945 }
2946
2947 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2948 {
2949     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2950     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2951     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2952     char src_name0[50];
2953     char src_name1[50];
2954     const char *comp = get_compare(ins->flags);
2955
2956     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2957     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2958
2959     if(vshader)
2960     {
2961         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2962          * away the subtraction result
2963          */
2964         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2965         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2966     }
2967     else
2968     {
2969         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2970         shader_addline(buffer, "BRK (%s.x);\n", comp);
2971     }
2972 }
2973
2974 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2975 {
2976     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2977     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2978     struct list *e = list_head(&priv->control_frames);
2979     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2980     const char *comp;
2981     char src_name0[50];
2982     char src_name1[50];
2983     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2984
2985     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2986     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2987
2988     if(vshader)
2989     {
2990         /* Invert the flag. We jump to the else label if the condition is NOT true */
2991         comp = get_compare(invert_compare(ins->flags));
2992         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2993         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
2994     }
2995     else
2996     {
2997         comp = get_compare(ins->flags);
2998         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2999         shader_addline(buffer, "IF %s.x;\n", comp);
3000     }
3001 }
3002
3003 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3004 {
3005     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3006     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3007     struct list *e = list_head(&priv->control_frames);
3008     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3009     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3010
3011     if(vshader)
3012     {
3013         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3014         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3015         control_frame->had_else = TRUE;
3016     }
3017     else
3018     {
3019         shader_addline(buffer, "ELSE;\n");
3020     }
3021 }
3022
3023 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3024 {
3025     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3026     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3027     struct list *e = list_head(&priv->control_frames);
3028     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3029     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3030
3031     if(vshader)
3032     {
3033         if(control_frame->had_else)
3034         {
3035             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3036         }
3037         else
3038         {
3039             shader_addline(buffer, "#No else branch. else is endif\n");
3040             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3041         }
3042     }
3043     else
3044     {
3045         shader_addline(buffer, "ENDIF;\n");
3046     }
3047 }
3048
3049 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3050 {
3051     DWORD sampler_idx = ins->src[1].reg.idx;
3052     char reg_dest[40];
3053     char reg_src[3][40];
3054     WORD flags = TEX_DERIV;
3055
3056     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3057     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3058     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3059     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3060
3061     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3062     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3063
3064     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3065 }
3066
3067 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3068 {
3069     DWORD sampler_idx = ins->src[1].reg.idx;
3070     char reg_dest[40];
3071     char reg_coord[40];
3072     WORD flags = TEX_LOD;
3073
3074     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3075     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3076
3077     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3078     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3079
3080     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3081 }
3082
3083 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3084 {
3085     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3086     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3087
3088     priv->in_main_func = FALSE;
3089     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3090      * subroutine, don't generate a label that will make GL complain
3091      */
3092     if(priv->target_version == ARB) return;
3093
3094     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3095 }
3096
3097 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3098         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
3099 {
3100     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3101     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3102     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3103     unsigned int i;
3104
3105     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3106      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3107      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3108      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3109      */
3110     if(args->super.fog_src == VS_FOG_Z) {
3111         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3112     } else if (!reg_maps->fog) {
3113         /* posFixup.x is always 1.0, so we can savely use it */
3114         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3115     }
3116
3117     /* Write the final position.
3118      *
3119      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3120      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3121      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3122      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3123      */
3124     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3125     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3126     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3127
3128     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3129     {
3130         if(args->super.clip_enabled)
3131         {
3132             for(i = 0; i < priv_ctx->vs_clipplanes; i++)
3133             {
3134                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3135             }
3136         }
3137     }
3138     else if(args->clip.boolclip.clip_texcoord)
3139     {
3140         unsigned int cur_clip = 0;
3141         char component[4] = {'x', 'y', 'z', 'w'};
3142         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3143
3144         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3145         {
3146             if(args->clip.boolclip.clipplane_mask & (1 << i))
3147             {
3148                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3149                                component[cur_clip++], i);
3150             }
3151         }
3152         switch(cur_clip)
3153         {
3154             case 0:
3155                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3156                 break;
3157             case 1:
3158                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3159                 break;
3160             case 2:
3161                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3162                 break;
3163             case 3:
3164                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3165                 break;
3166         }
3167         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3168                        args->clip.boolclip.clip_texcoord - 1);
3169     }
3170
3171     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3172      * and the glsl equivalent
3173      */
3174     if(need_helper_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
3175         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3176         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3177     } else {
3178         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3179         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3180     }
3181
3182     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3183
3184     priv_ctx->footer_written = TRUE;
3185 }
3186
3187 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3188 {
3189     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3190     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3191     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
3192     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3193
3194     if(priv->target_version == ARB) return;
3195
3196     if(vshader)
3197     {
3198         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
3199     }
3200
3201     shader_addline(buffer, "RET;\n");
3202 }
3203
3204 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3205 {
3206     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3207     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3208 }
3209
3210 /* GL locking is done by the caller */
3211 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3212 {
3213     GLuint program_id = 0;
3214     GLint pos;
3215
3216     const char *blt_vprogram =
3217         "!!ARBvp1.0\n"
3218         "PARAM c[1] = { { 1, 0.5 } };\n"
3219         "MOV result.position, vertex.position;\n"
3220         "MOV result.color, c[0].x;\n"
3221         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3222         "END\n";
3223
3224     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3225     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3226     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3227             strlen(blt_vprogram), blt_vprogram));
3228     checkGLcall("glProgramStringARB()");
3229
3230     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3231     if (pos != -1)
3232     {
3233         FIXME("Vertex program error at position %d: %s\n\n", pos,
3234             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3235         shader_arb_dump_program_source(blt_vprogram);
3236     }
3237     else
3238     {
3239         GLint native;
3240
3241         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3242         checkGLcall("glGetProgramivARB()");
3243         if (!native) WARN("Program exceeds native resource limits.\n");
3244     }
3245
3246     return program_id;
3247 }
3248
3249 /* GL locking is done by the caller */
3250 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3251         enum tex_types tex_type, BOOL masked)
3252 {
3253     GLuint program_id = 0;
3254     const char *fprogram;
3255     GLint pos;
3256
3257     static const char * const blt_fprograms_full[tex_type_count] =
3258     {
3259         /* tex_1d */
3260         NULL,
3261         /* tex_2d */
3262         "!!ARBfp1.0\n"
3263         "TEMP R0;\n"
3264         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3265         "MOV result.depth.z, R0.x;\n"
3266         "END\n",
3267         /* tex_3d */
3268         NULL,
3269         /* tex_cube */
3270         "!!ARBfp1.0\n"
3271         "TEMP R0;\n"
3272         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3273         "MOV result.depth.z, R0.x;\n"
3274         "END\n",
3275         /* tex_rect */
3276         "!!ARBfp1.0\n"
3277         "TEMP R0;\n"
3278         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3279         "MOV result.depth.z, R0.x;\n"
3280         "END\n",
3281     };
3282
3283     static const char * const blt_fprograms_masked[tex_type_count] =
3284     {
3285         /* tex_1d */
3286         NULL,
3287         /* tex_2d */
3288         "!!ARBfp1.0\n"
3289         "PARAM mask = program.local[0];\n"
3290         "TEMP R0;\n"
3291         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3292         "MUL R0.x, R0.x, R0.y;\n"
3293         "KIL -R0.x;\n"
3294         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3295         "MOV result.depth.z, R0.x;\n"
3296         "END\n",
3297         /* tex_3d */
3298         NULL,
3299         /* tex_cube */
3300         "!!ARBfp1.0\n"
3301         "PARAM mask = program.local[0];\n"
3302         "TEMP R0;\n"
3303         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3304         "MUL R0.x, R0.x, R0.y;\n"
3305         "KIL -R0.x;\n"
3306         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3307         "MOV result.depth.z, R0.x;\n"
3308         "END\n",
3309         /* tex_rect */
3310         "!!ARBfp1.0\n"
3311         "PARAM mask = program.local[0];\n"
3312         "TEMP R0;\n"
3313         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3314         "MUL R0.x, R0.x, R0.y;\n"
3315         "KIL -R0.x;\n"
3316         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3317         "MOV result.depth.z, R0.x;\n"
3318         "END\n",
3319     };
3320
3321     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3322     if (!fprogram)
3323     {
3324         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3325         tex_type = tex_2d;
3326         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3327     }
3328
3329     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3330     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3331     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3332     checkGLcall("glProgramStringARB()");
3333
3334     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3335     if (pos != -1)
3336     {
3337         FIXME("Fragment program error at position %d: %s\n\n", pos,
3338             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3339         shader_arb_dump_program_source(fprogram);
3340     }
3341     else
3342     {
3343         GLint native;
3344
3345         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3346         checkGLcall("glGetProgramivARB()");
3347         if (!native) WARN("Program exceeds native resource limits.\n");
3348     }
3349
3350     return program_id;
3351 }
3352
3353 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3354         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3355 {
3356     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3357
3358     if(condcode)
3359     {
3360         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3361         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3362         /* Calculate the > 0.0031308 case */
3363         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3364         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3365         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3366         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3367         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3368         /* Calculate the < case */
3369         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3370     }
3371     else
3372     {
3373         /* Calculate the > 0.0031308 case */
3374         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3375         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3376         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3377         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3378         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3379         /* Calculate the < case */
3380         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3381         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3382         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3383         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3384         /* Store the components > 0.0031308 in the destination */
3385         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3386         /* Add the components that are < 0.0031308 */
3387         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3388         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3389         * result.color writes(.rgb first, then .a), or handle overwriting already written
3390         * components. The assembler uses a temporary register in this case, which is usually
3391         * not allocated from one of our registers that were used earlier.
3392         */
3393     }
3394     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3395 }
3396
3397 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3398 {
3399     const local_constant *constant;
3400
3401     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3402     {
3403         if (constant->idx == idx)
3404         {
3405             return constant->value;
3406         }
3407     }
3408     return NULL;
3409 }
3410
3411 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3412                           struct shader_arb_ctx_priv *priv)
3413 {
3414     static const char * const texcoords[8] =
3415     {
3416         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3417         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3418     };
3419     unsigned int i;
3420     const struct wined3d_shader_signature_element *sig = This->baseShader.input_signature;
3421     const char *semantic_name;
3422     DWORD semantic_idx;
3423
3424     switch(args->super.vp_mode)
3425     {
3426         case pretransformed:
3427         case fixedfunction:
3428             /* The pixelshader has to collect the varyings on its own. In any case properly load
3429              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3430              * other attribs to 0.0.
3431              *
3432              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3433              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3434              * load the texcoord attrib pointers to match the pixel shader signature
3435              */
3436             for(i = 0; i < MAX_REG_INPUT; i++)
3437             {
3438                 semantic_name = sig[i].semantic_name;
3439                 semantic_idx = sig[i].semantic_idx;
3440                 if (!semantic_name) continue;
3441
3442                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3443                 {
3444                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3445                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3446                     else priv->ps_input[i] = "0.0";
3447                 }
3448                 else if(args->super.vp_mode == fixedfunction)
3449                 {
3450                     priv->ps_input[i] = "0.0";
3451                 }
3452                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3453                 {
3454                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3455                     else priv->ps_input[i] = "0.0";
3456                 }
3457                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3458                 {
3459                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3460                     else priv->ps_input[i] = "0.0";
3461                 }
3462                 else
3463                 {
3464                     priv->ps_input[i] = "0.0";
3465                 }
3466
3467                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3468             }
3469             break;
3470
3471         case vertexshader:
3472             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3473              * fragment.color
3474              */
3475             for(i = 0; i < 8; i++)
3476             {
3477                 priv->ps_input[i] = texcoords[i];
3478             }
3479             priv->ps_input[8] = "fragment.color.primary";
3480             priv->ps_input[9] = "fragment.color.secondary";
3481             break;
3482     }
3483 }
3484
3485 /* GL locking is done by the caller */
3486 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3487         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3488 {
3489     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3490     CONST DWORD *function = This->baseShader.function;
3491     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3492     const local_constant *lconst;
3493     GLuint retval;
3494     char fragcolor[16];
3495     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3496     struct shader_arb_ctx_priv priv_ctx;
3497     BOOL dcl_td = FALSE;
3498     BOOL want_nv_prog = FALSE;
3499     struct arb_pshader_private *shader_priv = This->baseShader.backend_data;
3500     GLint errPos;
3501     DWORD map;
3502
3503     char srgbtmp[4][4];
3504     unsigned int i, found = 0;
3505
3506     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3507     {
3508         if (!(map & 1)
3509                 || (This->color0_mov && i == This->color0_reg)
3510                 || (reg_maps->shader_version.major < 2 && !i))
3511             continue;
3512
3513         sprintf(srgbtmp[found], "R%u", i);
3514         ++found;
3515         if (found == 4) break;
3516     }
3517
3518     switch(found) {
3519         case 0:
3520             sprintf(srgbtmp[0], "TA");
3521             sprintf(srgbtmp[1], "TB");
3522             sprintf(srgbtmp[2], "TC");
3523             sprintf(srgbtmp[3], "TD");
3524             dcl_td = TRUE;
3525             break;
3526         case 1:
3527             sprintf(srgbtmp[1], "TA");
3528             sprintf(srgbtmp[2], "TB");
3529             sprintf(srgbtmp[3], "TC");
3530             break;
3531         case 2:
3532             sprintf(srgbtmp[2], "TA");
3533             sprintf(srgbtmp[3], "TB");
3534             break;
3535         case 3:
3536             sprintf(srgbtmp[3], "TA");
3537             break;
3538         case 4:
3539             break;
3540     }
3541
3542     /*  Create the hw ARB shader */
3543     memset(&priv_ctx, 0, sizeof(priv_ctx));
3544     priv_ctx.cur_ps_args = args;
3545     priv_ctx.compiled_fprog = compiled;
3546     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3547     init_ps_input(This, args, &priv_ctx);
3548     list_init(&priv_ctx.control_frames);
3549
3550     /* Avoid enabling NV_fragment_program* if we do not need it.
3551      *
3552      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3553      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3554      * is faster than what we gain from using higher native instructions. There are some things though
3555      * that cannot be emulated. In that case enable the extensions.
3556      * If the extension is enabled, instruction handlers that support both ways will use it.
3557      *
3558      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3559      * So enable the best we can get.
3560      */
3561     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3562        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3563     {
3564         want_nv_prog = TRUE;
3565     }
3566
3567     shader_addline(buffer, "!!ARBfp1.0\n");
3568     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3569     {
3570         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3571         priv_ctx.target_version = NV3;
3572     }
3573     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3574     {
3575         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3576         priv_ctx.target_version = NV2;
3577     } else {
3578         if(want_nv_prog)
3579         {
3580             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3581              * limits properly
3582              */
3583             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3584             ERR("Try GLSL\n");
3585         }
3586         priv_ctx.target_version = ARB;
3587     }
3588
3589     if(This->baseShader.reg_maps.highest_render_target > 0)
3590     {
3591         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3592     }
3593
3594     if (reg_maps->shader_version.major < 3)
3595     {
3596         switch(args->super.fog) {
3597             case FOG_OFF:
3598                 break;
3599             case FOG_LINEAR:
3600                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3601                 break;
3602             case FOG_EXP:
3603                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3604                 break;
3605             case FOG_EXP2:
3606                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3607                 break;
3608         }
3609     }
3610
3611     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3612      * unused temps away(but occupies them for the whole shader if they're used once). Always
3613      * declaring them avoids tricky bookkeeping work
3614      */
3615     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3616     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3617     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3618     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3619     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3620     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3621     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3622
3623     if (reg_maps->shader_version.major < 2)
3624     {
3625         strcpy(fragcolor, "R0");
3626     } else {
3627         if(args->super.srgb_correction) {
3628             if(This->color0_mov) {
3629                 sprintf(fragcolor, "R%u", This->color0_reg);
3630             } else {
3631                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3632                 strcpy(fragcolor, "TMP_COLOR");
3633             }
3634         } else {
3635             strcpy(fragcolor, "result.color");
3636         }
3637     }
3638
3639     if(args->super.srgb_correction) {
3640         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3641                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3642         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3643                        srgb_sub_high, 0.0, 0.0, 0.0);
3644     }
3645
3646     /* Base Declarations */
3647     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
3648             reg_maps, buffer, gl_info, lconst_map, NULL, &priv_ctx);
3649
3650     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3651     {
3652         if (!(map & 1)) continue;
3653
3654         cur = compiled->numbumpenvmatconsts;
3655         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3656         compiled->bumpenvmatconst[cur].texunit = i;
3657         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3658         compiled->luminanceconst[cur].texunit = i;
3659
3660         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3661          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3662          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3663          * textures due to conditional NP2 restrictions)
3664          *
3665          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3666          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3667          * their location is shader dependent anyway and they cannot be loaded globally.
3668          */
3669         compiled->bumpenvmatconst[cur].const_num = next_local++;
3670         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3671                        i, compiled->bumpenvmatconst[cur].const_num);
3672         compiled->numbumpenvmatconsts = cur + 1;
3673
3674         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3675
3676         compiled->luminanceconst[cur].const_num = next_local++;
3677         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3678                        i, compiled->luminanceconst[cur].const_num);
3679     }
3680
3681     for(i = 0; i < MAX_CONST_I; i++)
3682     {
3683         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3684         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3685         {
3686             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3687
3688             if(control_values)
3689             {
3690                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3691                                 control_values[0], control_values[1], control_values[2]);
3692             }
3693             else
3694             {
3695                 compiled->int_consts[i] = next_local;
3696                 compiled->num_int_consts++;
3697                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3698             }
3699         }
3700     }
3701
3702     if(reg_maps->vpos || reg_maps->usesdsy)
3703     {
3704         compiled->ycorrection = next_local;
3705         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3706
3707         if(reg_maps->vpos)
3708         {
3709             shader_addline(buffer, "TEMP vpos;\n");
3710             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3711              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3712              * ycorrection.z: 1.0
3713              * ycorrection.w: 0.0
3714              */
3715             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3716             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3717         }
3718     }
3719     else
3720     {
3721         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3722     }
3723
3724     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3725      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3726      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3727      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3728      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3729      * shader compilation errors and the subsequent errors when drawing with this shader. */
3730     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3731
3732         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3733         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3734         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3735
3736         fixup->offset = next_local;
3737         fixup->super.active = 0;
3738
3739         cur = 0;
3740         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3741             if (!(map & (1 << i))) continue;
3742
3743             if (fixup->offset + (cur >> 1) < max_lconsts) {
3744                 fixup->super.active |= (1 << i);
3745                 fixup->super.idx[i] = cur++;
3746             } else {
3747                 FIXME("No free constant found to load NP2 fixup data into shader. "
3748                       "Sampling from this texture will probably look wrong.\n");
3749                 break;
3750             }
3751         }
3752
3753         fixup->super.num_consts = (cur + 1) >> 1;
3754         if (fixup->super.num_consts) {
3755             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3756                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3757         }
3758
3759         next_local += fixup->super.num_consts;
3760     }
3761
3762     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3763     {
3764         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3765     }
3766
3767     /* Base Shader Body */
3768     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3769
3770     if(args->super.srgb_correction) {
3771         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3772                                   priv_ctx.target_version >= NV2);
3773     }
3774
3775     if(strcmp(fragcolor, "result.color")) {
3776         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3777     }
3778     shader_addline(buffer, "END\n");
3779
3780     /* TODO: change to resource.glObjectHandle or something like that */
3781     GL_EXTCALL(glGenProgramsARB(1, &retval));
3782
3783     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3784     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3785
3786     TRACE("Created hw pixel shader, prg=%d\n", retval);
3787     /* Create the program and check for errors */
3788     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3789                buffer->bsize, buffer->buffer));
3790     checkGLcall("glProgramStringARB()");
3791
3792     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3793     if (errPos != -1)
3794     {
3795         FIXME("HW PixelShader Error at position %d: %s\n\n",
3796               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3797         shader_arb_dump_program_source(buffer->buffer);
3798         retval = 0;
3799     }
3800     else
3801     {
3802         GLint native;
3803
3804         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3805         checkGLcall("glGetProgramivARB()");
3806         if (!native) WARN("Program exceeds native resource limits.\n");
3807     }
3808
3809     /* Load immediate constants */
3810     if(lconst_map) {
3811         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3812             const float *value = (const float *)lconst->value;
3813             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3814             checkGLcall("glProgramLocalParameter4fvARB");
3815         }
3816         HeapFree(GetProcessHeap(), 0, lconst_map);
3817     }
3818
3819     return retval;
3820 }
3821
3822 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3823 {
3824     unsigned int i;
3825     int ret;
3826
3827     for(i = 0; i < MAX_REG_INPUT; i++)
3828     {
3829         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3830         {
3831             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3832             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3833             continue;
3834         }
3835
3836         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3837         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3838         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3839         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3840         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3841         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3842     }
3843     return 0;
3844 }
3845
3846 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3847 {
3848     struct wined3d_shader_signature_element *new;
3849     int i;
3850     char *name;
3851
3852     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3853     for(i = 0; i < MAX_REG_INPUT; i++)
3854     {
3855         if (!sig[i].semantic_name) continue;
3856
3857         new[i] = sig[i];
3858         /* Clone the semantic string */
3859         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3860         strcpy(name, sig[i].semantic_name);
3861         new[i].semantic_name = name;
3862     }
3863     return new;
3864 }
3865
3866 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3867 {
3868     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3869     struct ps_signature *found_sig;
3870
3871     if (entry)
3872     {
3873         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3874         TRACE("Found existing signature %u\n", found_sig->idx);
3875         return found_sig->idx;
3876     }
3877     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3878     found_sig->sig = clone_sig(sig);
3879     found_sig->idx = priv->ps_sig_number++;
3880     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3881     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3882     {
3883         ERR("Failed to insert program entry.\n");
3884     }
3885     return found_sig->idx;
3886 }
3887
3888 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3889                                   struct arb_vs_compiled_shader *compiled)
3890 {
3891     unsigned int i, j;
3892     static const char * const texcoords[8] =
3893     {
3894         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3895         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3896     };
3897     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3898     IWineD3DBaseShaderClass *baseshader = &shader->baseShader;
3899     const struct wined3d_shader_signature_element *sig;
3900     const char *semantic_name;
3901     DWORD semantic_idx, reg_idx;
3902
3903     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3904      * and varying 9 to result.color.secondary
3905      */
3906     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3907     {
3908         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3909         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3910         "result.color.primary", "result.color.secondary"
3911     };
3912
3913     if(sig_num == ~0)
3914     {
3915         TRACE("Pixel shader uses builtin varyings\n");
3916         /* Map builtins to builtins */
3917         for(i = 0; i < 8; i++)
3918         {
3919             priv_ctx->texcrd_output[i] = texcoords[i];
3920         }
3921         priv_ctx->color_output[0] = "result.color.primary";
3922         priv_ctx->color_output[1] = "result.color.secondary";
3923         priv_ctx->fog_output = "result.fogcoord";
3924
3925         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3926         for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
3927         {
3928             semantic_name = baseshader->output_signature[i].semantic_name;
3929             if (!semantic_name) continue;
3930
3931             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3932             {
3933                 TRACE("o%u is TMP_OUT\n", i);
3934                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3935                 else priv_ctx->vs_output[i] = "TA";
3936             }
3937             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3938             {
3939                 TRACE("o%u is result.pointsize\n", i);
3940                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3941                 else priv_ctx->vs_output[i] = "TA";
3942             }
3943             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3944             {
3945                 TRACE("o%u is result.color.?, idx %u\n", i, baseshader->output_signature[i].semantic_idx);
3946                 if (!baseshader->output_signature[i].semantic_idx)
3947                     priv_ctx->vs_output[i] = "result.color.primary";
3948                 else if (baseshader->output_signature[i].semantic_idx == 1)
3949                     priv_ctx->vs_output[i] = "result.color.secondary";
3950                 else priv_ctx->vs_output[i] = "TA";
3951             }
3952             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3953             {
3954                 TRACE("o%u is %s\n", i, texcoords[baseshader->output_signature[i].semantic_idx]);
3955                 if (baseshader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3956                 else priv_ctx->vs_output[i] = texcoords[baseshader->output_signature[i].semantic_idx];
3957             }
3958             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3959             {
3960                 TRACE("o%u is result.fogcoord\n", i);
3961                 if (baseshader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3962                 else priv_ctx->vs_output[i] = "result.fogcoord";
3963             }
3964             else
3965             {
3966                 priv_ctx->vs_output[i] = "TA";
3967             }
3968         }
3969         return;
3970     }
3971
3972     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3973      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3974      */
3975     sig = device->stateBlock->state.pixel_shader->baseShader.input_signature;
3976     TRACE("Pixel shader uses declared varyings\n");
3977
3978     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3979     for(i = 0; i < 8; i++)
3980     {
3981         priv_ctx->texcrd_output[i] = "TA";
3982     }
3983     priv_ctx->color_output[0] = "TA";
3984     priv_ctx->color_output[1] = "TA";
3985     priv_ctx->fog_output = "TA";
3986
3987     for(i = 0; i < MAX_REG_INPUT; i++)
3988     {
3989         semantic_name = sig[i].semantic_name;
3990         semantic_idx = sig[i].semantic_idx;
3991         reg_idx = sig[i].register_idx;
3992         if (!semantic_name) continue;
3993
3994         /* If a declared input register is not written by builtin arguments, don't write to it.
3995          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3996          *
3997          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3998          * to TMP_OUT in any case
3999          */
4000         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4001         {
4002             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4003         }
4004         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4005         {
4006             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4007         }
4008         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4009         {
4010             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4011         }
4012         else
4013         {
4014             continue;
4015         }
4016
4017         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4018                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4019         {
4020             compiled->need_color_unclamp = TRUE;
4021         }
4022     }
4023
4024     /* Map declared to declared */
4025     for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
4026     {
4027         /* Write unread output to TA to throw them away */
4028         priv_ctx->vs_output[i] = "TA";
4029         semantic_name = baseshader->output_signature[i].semantic_name;
4030         if (!semantic_name) continue;
4031
4032         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4033                 && !baseshader->output_signature[i].semantic_idx)
4034         {
4035             priv_ctx->vs_output[i] = "TMP_OUT";
4036             continue;
4037         }
4038         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4039                 && !baseshader->output_signature[i].semantic_idx)
4040         {
4041             priv_ctx->vs_output[i] = "result.pointsize";
4042             continue;
4043         }
4044
4045         for(j = 0; j < MAX_REG_INPUT; j++)
4046         {
4047             if (!sig[j].semantic_name) continue;
4048
4049             if (!strcmp(sig[j].semantic_name, semantic_name)
4050                     && sig[j].semantic_idx == baseshader->output_signature[i].semantic_idx)
4051             {
4052                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4053
4054                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4055                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4056                 {
4057                     compiled->need_color_unclamp = TRUE;
4058                 }
4059             }
4060         }
4061     }
4062 }
4063
4064 /* GL locking is done by the caller */
4065 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
4066         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4067 {
4068     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4069     CONST DWORD *function = This->baseShader.function;
4070     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4071     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4072     const local_constant *lconst;
4073     GLuint ret;
4074     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
4075     struct shader_arb_ctx_priv priv_ctx;
4076     unsigned int i;
4077     GLint errPos;
4078
4079     memset(&priv_ctx, 0, sizeof(priv_ctx));
4080     priv_ctx.cur_vs_args = args;
4081     list_init(&priv_ctx.control_frames);
4082     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
4083
4084     /*  Create the hw ARB shader */
4085     shader_addline(buffer, "!!ARBvp1.0\n");
4086
4087     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4088      * mesurable performance penalty, and we can always make use of it for clipplanes.
4089      */
4090     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4091     {
4092         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4093         priv_ctx.target_version = NV3;
4094         shader_addline(buffer, "ADDRESS aL;\n");
4095     }
4096     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4097     {
4098         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4099         priv_ctx.target_version = NV2;
4100         shader_addline(buffer, "ADDRESS aL;\n");
4101     } else {
4102         priv_ctx.target_version = ARB;
4103     }
4104
4105     shader_addline(buffer, "TEMP TMP_OUT;\n");
4106     if(need_helper_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
4107         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4108     }
4109     if(need_rel_addr_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
4110         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", This->rel_offset);
4111         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4112     }
4113
4114     shader_addline(buffer, "TEMP TA;\n");
4115
4116     /* Base Declarations */
4117     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
4118             reg_maps, buffer, gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4119
4120     for(i = 0; i < MAX_CONST_I; i++)
4121     {
4122         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4123         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4124         {
4125             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
4126
4127             if(control_values)
4128             {
4129                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4130                                 control_values[0], control_values[1], control_values[2]);
4131             }
4132             else
4133             {
4134                 compiled->int_consts[i] = next_local;
4135                 compiled->num_int_consts++;
4136                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4137             }
4138         }
4139     }
4140
4141     /* We need a constant to fixup the final position */
4142     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4143     compiled->pos_fixup = next_local++;
4144
4145     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4146      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4147      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4148      * a replacement shader depend on the texcoord.w being set properly.
4149      *
4150      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4151      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4152      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4153      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4154      * this can eat a number of instructions, so skip it unless this cap is set as well
4155      */
4156     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4157     {
4158         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4159         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4160
4161         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4162         {
4163             int i;
4164             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4165             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4166             {
4167                 if (This->baseShader.reg_maps.texcoord_mask[i]
4168                         && This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4169                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4170             }
4171         }
4172     }
4173
4174     /* The shader starts with the main function */
4175     priv_ctx.in_main_func = TRUE;
4176     /* Base Shader Body */
4177     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4178
4179     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
4180
4181     shader_addline(buffer, "END\n");
4182
4183     /* TODO: change to resource.glObjectHandle or something like that */
4184     GL_EXTCALL(glGenProgramsARB(1, &ret));
4185
4186     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4187     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4188
4189     TRACE("Created hw vertex shader, prg=%d\n", ret);
4190     /* Create the program and check for errors */
4191     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4192                buffer->bsize, buffer->buffer));
4193     checkGLcall("glProgramStringARB()");
4194
4195     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4196     if (errPos != -1)
4197     {
4198         FIXME("HW VertexShader Error at position %d: %s\n\n",
4199               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4200         shader_arb_dump_program_source(buffer->buffer);
4201         ret = -1;
4202     }
4203     else
4204     {
4205         GLint native;
4206
4207         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4208         checkGLcall("glGetProgramivARB()");
4209         if (!native) WARN("Program exceeds native resource limits.\n");
4210
4211         /* Load immediate constants */
4212         if(lconst_map) {
4213             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
4214                 const float *value = (const float *)lconst->value;
4215                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4216             }
4217         }
4218     }
4219     HeapFree(GetProcessHeap(), 0, lconst_map);
4220
4221     return ret;
4222 }
4223
4224 /* GL locking is done by the caller */
4225 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
4226 {
4227     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)shader->baseShader.device;
4228     UINT i;
4229     DWORD new_size;
4230     struct arb_ps_compiled_shader *new_array;
4231     struct wined3d_shader_buffer buffer;
4232     struct arb_pshader_private *shader_data;
4233     GLuint ret;
4234
4235     if (!shader->baseShader.backend_data)
4236     {
4237         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4238         struct shader_arb_priv *priv = device->shader_priv;
4239
4240         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4241         shader_data = shader->baseShader.backend_data;
4242         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
4243
4244         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
4245         else shader_data->input_signature_idx = find_input_signature(priv, shader->baseShader.input_signature);
4246
4247         shader_data->has_signature_idx = TRUE;
4248         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4249
4250         if (!device->vs_clipping)
4251             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
4252                     gl_info->limits.texture_stages - 1);
4253         else
4254             shader_data->clipplane_emulation = ~0U;
4255     }
4256     shader_data = shader->baseShader.backend_data;
4257
4258     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4259      * so a linear search is more performant than a hashmap or a binary search
4260      * (cache coherency etc)
4261      */
4262     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4263     {
4264         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4265             return &shader_data->gl_shaders[i];
4266     }
4267
4268     TRACE("No matching GL shader found, compiling a new shader\n");
4269     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4270         if (shader_data->num_gl_shaders)
4271         {
4272             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4273             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4274                                     new_size * sizeof(*shader_data->gl_shaders));
4275         } else {
4276             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4277             new_size = 1;
4278         }
4279
4280         if(!new_array) {
4281             ERR("Out of memory\n");
4282             return 0;
4283         }
4284         shader_data->gl_shaders = new_array;
4285         shader_data->shader_array_size = new_size;
4286     }
4287
4288     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4289
4290     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4291             (IWineD3DBaseTexture **)device->stateBlock->state.textures);
4292
4293     if (!shader_buffer_init(&buffer))
4294     {
4295         ERR("Failed to initialize shader buffer.\n");
4296         return 0;
4297     }
4298
4299     ret = shader_arb_generate_pshader(shader, &buffer, args,
4300                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4301     shader_buffer_free(&buffer);
4302     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4303
4304     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4305 }
4306
4307 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4308                                  const DWORD use_map, BOOL skip_int) {
4309     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4310     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4311     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4312     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4313     if(stored->ps_signature != new->ps_signature) return FALSE;
4314     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4315     if(skip_int) return TRUE;
4316
4317     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4318 }
4319
4320 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4321 {
4322     UINT i;
4323     DWORD new_size;
4324     struct arb_vs_compiled_shader *new_array;
4325     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4326     struct wined3d_shader_buffer buffer;
4327     struct arb_vshader_private *shader_data;
4328     GLuint ret;
4329     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4330
4331     if (!shader->baseShader.backend_data)
4332     {
4333         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4334     }
4335     shader_data = shader->baseShader.backend_data;
4336
4337     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4338      * so a linear search is more performant than a hashmap or a binary search
4339      * (cache coherency etc)
4340      */
4341     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4342         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4343                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4344         {
4345             return &shader_data->gl_shaders[i];
4346         }
4347     }
4348
4349     TRACE("No matching GL shader found, compiling a new shader\n");
4350
4351     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4352         if (shader_data->num_gl_shaders)
4353         {
4354             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4355             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4356                                     new_size * sizeof(*shader_data->gl_shaders));
4357         } else {
4358             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4359             new_size = 1;
4360         }
4361
4362         if(!new_array) {
4363             ERR("Out of memory\n");
4364             return 0;
4365         }
4366         shader_data->gl_shaders = new_array;
4367         shader_data->shader_array_size = new_size;
4368     }
4369
4370     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4371
4372     if (!shader_buffer_init(&buffer))
4373     {
4374         ERR("Failed to initialize shader buffer.\n");
4375         return 0;
4376     }
4377
4378     ret = shader_arb_generate_vshader(shader, &buffer, args,
4379             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4380     shader_buffer_free(&buffer);
4381     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4382
4383     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4384 }
4385
4386 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4387         IWineD3DPixelShaderImpl *shader, struct arb_ps_compile_args *args)
4388 {
4389     int i;
4390     WORD int_skip;
4391     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4392
4393     find_ps_compile_args(state, shader, &args->super);
4394
4395     /* This forces all local boolean constants to 1 to make them stateblock independent */
4396     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4397
4398     for(i = 0; i < MAX_CONST_B; i++)
4399     {
4400         if (state->ps_consts_b[i])
4401             args->bools |= ( 1 << i);
4402     }
4403
4404     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4405      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4406      * duplicate the shader than have a no-op KIL instruction in every shader
4407      */
4408     if ((!((IWineD3DDeviceImpl *)shader->baseShader.device)->vs_clipping) && use_vs(state)
4409             && state->render_states[WINED3DRS_CLIPPING]
4410             && state->render_states[WINED3DRS_CLIPPLANEENABLE])
4411         args->clip = 1;
4412     else
4413         args->clip = 0;
4414
4415     /* Skip if unused or local, or supported natively */
4416     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4417     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4418     {
4419         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4420         return;
4421     }
4422
4423     for(i = 0; i < MAX_CONST_I; i++)
4424     {
4425         if(int_skip & (1 << i))
4426         {
4427             args->loop_ctrl[i][0] = 0;
4428             args->loop_ctrl[i][1] = 0;
4429             args->loop_ctrl[i][2] = 0;
4430         }
4431         else
4432         {
4433             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4434             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4435             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4436         }
4437     }
4438 }
4439
4440 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4441         IWineD3DVertexShaderImpl *shader, struct arb_vs_compile_args *args)
4442 {
4443     int i;
4444     WORD int_skip;
4445     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4446     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4447
4448     find_vs_compile_args(state, shader, &args->super);
4449
4450     args->clip.boolclip_compare = 0;
4451     if (use_ps(state))
4452     {
4453         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
4454         struct arb_pshader_private *shader_priv = ps->baseShader.backend_data;
4455         args->ps_signature = shader_priv->input_signature_idx;
4456
4457         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4458     }
4459     else
4460     {
4461         args->ps_signature = ~0;
4462         if(!dev->vs_clipping)
4463         {
4464             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4465         }
4466         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4467     }
4468
4469     if (args->clip.boolclip.clip_texcoord)
4470     {
4471         if (state->render_states[WINED3DRS_CLIPPING])
4472             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3DRS_CLIPPLANEENABLE];
4473         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4474     }
4475
4476     /* This forces all local boolean constants to 1 to make them stateblock independent */
4477     args->clip.boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4478     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4479     for(i = 0; i < MAX_CONST_B; i++)
4480     {
4481         if (state->vs_consts_b[i])
4482             args->clip.boolclip.bools |= ( 1 << i);
4483     }
4484
4485     args->vertex.samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4486     args->vertex.samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4487     args->vertex.samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4488     args->vertex.samplers[3] = 0;
4489
4490     /* Skip if unused or local */
4491     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4492     /* This is about flow control, not clipping. */
4493     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4494     {
4495         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4496         return;
4497     }
4498
4499     for(i = 0; i < MAX_CONST_I; i++)
4500     {
4501         if(int_skip & (1 << i))
4502         {
4503             args->loop_ctrl[i][0] = 0;
4504             args->loop_ctrl[i][1] = 0;
4505             args->loop_ctrl[i][2] = 0;
4506         }
4507         else
4508         {
4509             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4510             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4511             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4512         }
4513     }
4514 }
4515
4516 /* GL locking is done by the caller */
4517 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4518 {
4519     IWineD3DDeviceImpl *This = context->swapchain->device;
4520     struct shader_arb_priv *priv = This->shader_priv;
4521     const struct wined3d_gl_info *gl_info = context->gl_info;
4522     const struct wined3d_state *state = &This->stateBlock->state;
4523     int i;
4524
4525     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4526     if (usePS)
4527     {
4528         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
4529         struct arb_ps_compile_args compile_args;
4530         struct arb_ps_compiled_shader *compiled;
4531
4532         TRACE("Using pixel shader %p.\n", ps);
4533         find_arb_ps_compile_args(state, ps, &compile_args);
4534         compiled = find_arb_pshader(ps, &compile_args);
4535         priv->current_fprogram_id = compiled->prgId;
4536         priv->compiled_fprog = compiled;
4537
4538         /* Bind the fragment program */
4539         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4540         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4541
4542         if(!priv->use_arbfp_fixed_func) {
4543             /* Enable OpenGL fragment programs */
4544             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4545             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4546         }
4547         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4548
4549         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4550          * a 1.x and newer shader, reload the first 8 constants
4551          */
4552         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts)
4553         {
4554             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts;
4555             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4556             for(i = 0; i < 8; i++)
4557             {
4558                 context->pshader_const_dirty[i] = 1;
4559             }
4560             /* Also takes care of loading local constants */
4561             shader_arb_load_constants(context, TRUE, FALSE);
4562         }
4563         else
4564         {
4565             float rt_height = This->render_targets[0]->currentDesc.Height;
4566             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4567         }
4568
4569         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4570         if (compiled->np2fixup_info.super.active)
4571             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4572     }
4573     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4574     {
4575         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4576         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4577         * replacement shader
4578         */
4579         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4580         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4581         priv->current_fprogram_id = 0;
4582     }
4583
4584     if (useVS)
4585     {
4586         IWineD3DVertexShaderImpl *vs = state->vertex_shader;
4587         struct arb_vs_compile_args compile_args;
4588         struct arb_vs_compiled_shader *compiled;
4589
4590         TRACE("Using vertex shader %p\n", vs);
4591         find_arb_vs_compile_args(state, vs, &compile_args);
4592         compiled = find_arb_vshader(vs, &compile_args);
4593         priv->current_vprogram_id = compiled->prgId;
4594         priv->compiled_vprog = compiled;
4595
4596         /* Bind the vertex program */
4597         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4598         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4599
4600         /* Enable OpenGL vertex programs */
4601         glEnable(GL_VERTEX_PROGRAM_ARB);
4602         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4603         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4604         shader_arb_vs_local_constants(compiled, context, state);
4605
4606         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4607             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4608
4609             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4610             {
4611                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4612                 checkGLcall("glClampColorARB");
4613             } else {
4614                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4615             }
4616         }
4617     }
4618     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4619     {
4620         priv->current_vprogram_id = 0;
4621         glDisable(GL_VERTEX_PROGRAM_ARB);
4622         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4623     }
4624 }
4625
4626 /* GL locking is done by the caller */
4627 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type, const SIZE *ds_mask_size)
4628 {
4629     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4630     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4631     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4632     struct shader_arb_priv *priv = This->shader_priv;
4633     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4634     GLuint *blt_fprogram;
4635
4636     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4637     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4638     glEnable(GL_VERTEX_PROGRAM_ARB);
4639
4640     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4641     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4642     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4643     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4644     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4645 }
4646
4647 /* GL locking is done by the caller */
4648 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4649     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4650     struct shader_arb_priv *priv = This->shader_priv;
4651     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4652
4653     if (priv->current_vprogram_id) {
4654         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4655         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4656
4657         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4658     } else {
4659         glDisable(GL_VERTEX_PROGRAM_ARB);
4660         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4661     }
4662
4663     if (priv->current_fprogram_id) {
4664         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4665         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4666
4667         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4668     } else if(!priv->use_arbfp_fixed_func) {
4669         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4670         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4671     }
4672 }
4673
4674 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4675     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4676     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4677     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4678
4679     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4680     {
4681         struct arb_pshader_private *shader_data = baseShader->baseShader.backend_data;
4682         UINT i;
4683
4684         if(!shader_data) return; /* This can happen if a shader was never compiled */
4685
4686         if (shader_data->num_gl_shaders)
4687         {
4688             struct wined3d_context *context = context_acquire(device, NULL);
4689
4690             ENTER_GL();
4691             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4692             {
4693                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4694                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4695             }
4696             LEAVE_GL();
4697
4698             context_release(context);
4699         }
4700
4701         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4702         HeapFree(GetProcessHeap(), 0, shader_data);
4703         baseShader->baseShader.backend_data = NULL;
4704     }
4705     else
4706     {
4707         struct arb_vshader_private *shader_data = baseShader->baseShader.backend_data;
4708         UINT i;
4709
4710         if(!shader_data) return; /* This can happen if a shader was never compiled */
4711
4712         if (shader_data->num_gl_shaders)
4713         {
4714             struct wined3d_context *context = context_acquire(device, NULL);
4715
4716             ENTER_GL();
4717             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4718             {
4719                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4720                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4721             }
4722             LEAVE_GL();
4723
4724             context_release(context);
4725         }
4726
4727         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4728         HeapFree(GetProcessHeap(), 0, shader_data);
4729         baseShader->baseShader.backend_data = NULL;
4730     }
4731 }
4732
4733 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4734 {
4735     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4736     return compare_sig(key, e->sig);
4737 }
4738
4739 static const struct wine_rb_functions sig_tree_functions =
4740 {
4741     wined3d_rb_alloc,
4742     wined3d_rb_realloc,
4743     wined3d_rb_free,
4744     sig_tree_compare
4745 };
4746
4747 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4748     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4749     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4750     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4751     {
4752         ERR("RB tree init failed\n");
4753         HeapFree(GetProcessHeap(), 0, priv);
4754         return E_OUTOFMEMORY;
4755     }
4756     This->shader_priv = priv;
4757     return WINED3D_OK;
4758 }
4759
4760 static void release_signature(struct wine_rb_entry *entry, void *context)
4761 {
4762     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4763     int i;
4764     for(i = 0; i < MAX_REG_INPUT; i++)
4765     {
4766         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4767     }
4768     HeapFree(GetProcessHeap(), 0, sig->sig);
4769     HeapFree(GetProcessHeap(), 0, sig);
4770 }
4771
4772 /* Context activation is done by the caller. */
4773 static void shader_arb_free(IWineD3DDevice *iface) {
4774     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4775     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4776     struct shader_arb_priv *priv = This->shader_priv;
4777     int i;
4778
4779     ENTER_GL();
4780     if(priv->depth_blt_vprogram_id) {
4781         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4782     }
4783     for (i = 0; i < tex_type_count; ++i)
4784     {
4785         if (priv->depth_blt_fprogram_id_full[i])
4786         {
4787             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4788         }
4789         if (priv->depth_blt_fprogram_id_masked[i])
4790         {
4791             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4792         }
4793     }
4794     LEAVE_GL();
4795
4796     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4797     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4798 }
4799
4800 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4801     return TRUE;
4802 }
4803
4804 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4805 {
4806     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4807     {
4808         DWORD vs_consts;
4809
4810         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4811          * for vertex programs. If the native limit is less than that it's
4812          * not very useful, and e.g. Mesa swrast returns 0, probably to
4813          * indicate it's a software implementation. */
4814         if (gl_info->limits.arb_vs_native_constants < 96)
4815             vs_consts = gl_info->limits.arb_vs_float_constants;
4816         else
4817             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4818
4819         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4820         {
4821             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4822             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4823         }
4824         else if (vs_consts >= 256)
4825         {
4826             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4827             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4828             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4829         }
4830         else
4831         {
4832             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4833             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4834         }
4835         pCaps->MaxVertexShaderConst = vs_consts;
4836     }
4837     else
4838     {
4839         pCaps->VertexShaderVersion = 0;
4840         pCaps->MaxVertexShaderConst = 0;
4841     }
4842
4843     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4844     {
4845         DWORD ps_consts;
4846
4847         /* Similar as above for vertex programs, but the minimum for fragment
4848          * programs is 24. */
4849         if (gl_info->limits.arb_ps_native_constants < 24)
4850             ps_consts = gl_info->limits.arb_ps_float_constants;
4851         else
4852             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4853
4854         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4855         {
4856             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4857             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4858         }
4859         else if (ps_consts >= 32)
4860         {
4861             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4862             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4863             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4864         }
4865         else
4866         {
4867             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4868             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4869         }
4870         pCaps->PixelShader1xMaxValue = 8.0f;
4871         pCaps->MaxPixelShaderConst = ps_consts;
4872     }
4873     else
4874     {
4875         pCaps->PixelShaderVersion = 0;
4876         pCaps->PixelShader1xMaxValue = 0.0f;
4877         pCaps->MaxPixelShaderConst = 0;
4878     }
4879
4880     pCaps->VSClipping = use_nv_clip(gl_info);
4881 }
4882
4883 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4884 {
4885     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4886     {
4887         TRACE("Checking support for color_fixup:\n");
4888         dump_color_fixup_desc(fixup);
4889     }
4890
4891     /* We support everything except complex conversions. */
4892     if (!is_complex_fixup(fixup))
4893     {
4894         TRACE("[OK]\n");
4895         return TRUE;
4896     }
4897
4898     TRACE("[FAILED]\n");
4899     return FALSE;
4900 }
4901
4902 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4903     DWORD shift;
4904     char write_mask[20], regstr[50];
4905     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4906     BOOL is_color = FALSE;
4907     const struct wined3d_shader_dst_param *dst;
4908
4909     if (!ins->dst_count) return;
4910
4911     dst = &ins->dst[0];
4912     shift = dst->shift;
4913     if (!shift) return; /* Saturate alone is handled by the instructions */
4914
4915     shader_arb_get_write_mask(ins, dst, write_mask);
4916     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4917
4918     /* Generate a line that does the output modifier computation
4919      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4920      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4921      */
4922     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4923                    regstr, write_mask, regstr, shift_tab[shift]);
4924 }
4925
4926 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4927 {
4928     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4929     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4930     /* WINED3DSIH_AND           */ NULL,
4931     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4932     /* WINED3DSIH_BREAK         */ shader_hw_break,
4933     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4934     /* WINED3DSIH_BREAKP        */ NULL,
4935     /* WINED3DSIH_CALL          */ shader_hw_call,
4936     /* WINED3DSIH_CALLNZ        */ NULL,
4937     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4938     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4939     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4940     /* WINED3DSIH_CUT           */ NULL,
4941     /* WINED3DSIH_DCL           */ NULL,
4942     /* WINED3DSIH_DEF           */ NULL,
4943     /* WINED3DSIH_DEFB          */ NULL,
4944     /* WINED3DSIH_DEFI          */ NULL,
4945     /* WINED3DSIH_DIV           */ NULL,
4946     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4947     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4948     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4949     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4950     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4951     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4952     /* WINED3DSIH_ELSE          */ shader_hw_else,
4953     /* WINED3DSIH_EMIT          */ NULL,
4954     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4955     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4956     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4957     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4958     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4959     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4960     /* WINED3DSIH_FTOI          */ NULL,
4961     /* WINED3DSIH_IADD          */ NULL,
4962     /* WINED3DSIH_IEQ           */ NULL,
4963     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4964     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4965     /* WINED3DSIH_IGE           */ NULL,
4966     /* WINED3DSIH_IMUL          */ NULL,
4967     /* WINED3DSIH_ITOF          */ NULL,
4968     /* WINED3DSIH_LABEL         */ shader_hw_label,
4969     /* WINED3DSIH_LD            */ NULL,
4970     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4971     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4972     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4973     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4974     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4975     /* WINED3DSIH_LT            */ NULL,
4976     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4977     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4978     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4979     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4980     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4981     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4982     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4983     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4984     /* WINED3DSIH_MOV           */ shader_hw_mov,
4985     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4986     /* WINED3DSIH_MOVC          */ NULL,
4987     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4988     /* WINED3DSIH_NOP           */ shader_hw_nop,
4989     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4990     /* WINED3DSIH_PHASE         */ NULL,
4991     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4992     /* WINED3DSIH_RCP           */ shader_hw_rcp,
4993     /* WINED3DSIH_REP           */ shader_hw_rep,
4994     /* WINED3DSIH_RET           */ shader_hw_ret,
4995     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4996     /* WINED3DSIH_SAMPLE        */ NULL,
4997     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
4998     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
4999     /* WINED3DSIH_SETP          */ NULL,
5000     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5001     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5002     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5003     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5004     /* WINED3DSIH_SQRT          */ NULL,
5005     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5006     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5007     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5008     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5009     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5010     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5011     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5012     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5013     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5014     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5015     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5016     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5017     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5018     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5019     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5020     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5021     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5022     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5023     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5024     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5025     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5026     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5027     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5028     /* WINED3DSIH_UTOF          */ NULL,
5029 };
5030
5031 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
5032 {
5033     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5034     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5035     WORD bools = 0;
5036     WORD flag = (1 << idx);
5037     const local_constant *constant;
5038     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5039
5040     if (reg_maps->local_bool_consts & flag)
5041     {
5042         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5043         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
5044         {
5045             if (constant->idx == idx)
5046             {
5047                 return constant->value[0];
5048             }
5049         }
5050         ERR("Local constant not found\n");
5051         return FALSE;
5052     }
5053     else
5054     {
5055         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5056         else bools = priv->cur_ps_args->bools;
5057         return bools & flag;
5058     }
5059 }
5060
5061 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5062         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
5063 {
5064     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5065     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5066
5067     /* Integer constants can either be a local constant, or they can be stored in the shader
5068      * type specific compile args. */
5069     if (reg_maps->local_int_consts & (1 << idx))
5070     {
5071         const local_constant *constant;
5072
5073         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
5074         {
5075             if (constant->idx == idx)
5076             {
5077                 loop_control->count = constant->value[0];
5078                 loop_control->start = constant->value[1];
5079                 /* Step is signed. */
5080                 loop_control->step = (int)constant->value[2];
5081                 return;
5082             }
5083         }
5084         /* If this happens the flag was set incorrectly */
5085         ERR("Local constant not found\n");
5086         loop_control->count = 0;
5087         loop_control->start = 0;
5088         loop_control->step = 0;
5089         return;
5090     }
5091
5092     switch (reg_maps->shader_version.type)
5093     {
5094         case WINED3D_SHADER_TYPE_VERTEX:
5095             /* Count and aL start value are unsigned */
5096             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5097             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5098             /* Step is signed. */
5099             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5100             break;
5101
5102         case WINED3D_SHADER_TYPE_PIXEL:
5103             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5104             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5105             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5106             break;
5107
5108         default:
5109             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5110             break;
5111     }
5112 }
5113
5114 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5115 {
5116     unsigned int i;
5117     struct wined3d_shader_dst_param *dst_param = NULL;
5118     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5119     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5120     if(!rec)
5121     {
5122         ERR("Out of memory\n");
5123         return;
5124     }
5125
5126     rec->ins = *ins;
5127     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5128     if(!dst_param) goto free;
5129     *dst_param = *ins->dst;
5130     if(ins->dst->reg.rel_addr)
5131     {
5132         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5133         if(!rel_addr) goto free;
5134         *rel_addr = *ins->dst->reg.rel_addr;
5135         dst_param->reg.rel_addr = rel_addr;
5136     }
5137     rec->ins.dst = dst_param;
5138
5139     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5140     if(!src_param) goto free;
5141     for(i = 0; i < ins->src_count; i++)
5142     {
5143         src_param[i] = ins->src[i];
5144         if(ins->src[i].reg.rel_addr)
5145         {
5146             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5147             if(!rel_addr) goto free;
5148             *rel_addr = *ins->src[i].reg.rel_addr;
5149             src_param[i].reg.rel_addr = rel_addr;
5150         }
5151     }
5152     rec->ins.src = src_param;
5153     list_add_tail(list, &rec->entry);
5154     return;
5155
5156 free:
5157     ERR("Out of memory\n");
5158     if(dst_param)
5159     {
5160         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5161         HeapFree(GetProcessHeap(), 0, dst_param);
5162     }
5163     if(src_param)
5164     {
5165         for(i = 0; i < ins->src_count; i++)
5166         {
5167             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5168         }
5169         HeapFree(GetProcessHeap(), 0, src_param);
5170     }
5171     HeapFree(GetProcessHeap(), 0, rec);
5172 }
5173
5174 static void free_recorded_instruction(struct list *list)
5175 {
5176     struct recorded_instruction *rec_ins, *entry2;
5177     unsigned int i;
5178
5179     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5180     {
5181         list_remove(&rec_ins->entry);
5182         if(rec_ins->ins.dst)
5183         {
5184             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5185             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5186         }
5187         if(rec_ins->ins.src)
5188         {
5189             for(i = 0; i < rec_ins->ins.src_count; i++)
5190             {
5191                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5192             }
5193             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5194         }
5195         HeapFree(GetProcessHeap(), 0, rec_ins);
5196     }
5197 }
5198
5199 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5200     SHADER_HANDLER hw_fct;
5201     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5202     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
5203     struct control_frame *control_frame;
5204     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5205     BOOL bool_const;
5206
5207     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5208     {
5209         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5210         list_add_head(&priv->control_frames, &control_frame->entry);
5211
5212         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5213         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5214
5215         if(priv->target_version >= NV2)
5216         {
5217             control_frame->no.loop = priv->num_loops++;
5218             priv->loop_depth++;
5219         }
5220         else
5221         {
5222             /* Don't bother recording when we're in a not used if branch */
5223             if(priv->muted)
5224             {
5225                 return;
5226             }
5227
5228             if(!priv->recording)
5229             {
5230                 list_init(&priv->record);
5231                 priv->recording = TRUE;
5232                 control_frame->outer_loop = TRUE;
5233                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
5234                 return; /* Instruction is handled */
5235             }
5236             /* Record this loop in the outer loop's recording */
5237         }
5238     }
5239     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5240     {
5241         if(priv->target_version >= NV2)
5242         {
5243             /* Nothing to do. The control frame is popped after the HW instr handler */
5244         }
5245         else
5246         {
5247             struct list *e = list_head(&priv->control_frames);
5248             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5249             list_remove(&control_frame->entry);
5250
5251             if(control_frame->outer_loop)
5252             {
5253                 unsigned int iteration;
5254                 int aL = 0;
5255                 struct list copy;
5256
5257                 /* Turn off recording before playback */
5258                 priv->recording = FALSE;
5259
5260                 /* Move the recorded instructions to a separate list and get them out of the private data
5261                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5262                  * be recorded again, thus priv->record might be overwritten
5263                  */
5264                 list_init(&copy);
5265                 list_move_tail(&copy, &priv->record);
5266                 list_init(&priv->record);
5267
5268                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5269                 {
5270                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5271                                    control_frame->loop_control.count, control_frame->loop_control.start,
5272                                    control_frame->loop_control.step);
5273                     aL = control_frame->loop_control.start;
5274                 }
5275                 else
5276                 {
5277                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5278                 }
5279
5280                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5281                 {
5282                     struct recorded_instruction *rec_ins;
5283                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5284                     {
5285                         priv->aL = aL;
5286                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5287                     }
5288                     else
5289                     {
5290                         shader_addline(buffer, "#Iteration %u\n", iteration);
5291                     }
5292
5293                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5294                     {
5295                         shader_arb_handle_instruction(&rec_ins->ins);
5296                     }
5297
5298                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5299                     {
5300                         aL += control_frame->loop_control.step;
5301                     }
5302                 }
5303                 shader_addline(buffer, "#end loop/rep\n");
5304
5305                 free_recorded_instruction(&copy);
5306                 HeapFree(GetProcessHeap(), 0, control_frame);
5307                 return; /* Instruction is handled */
5308             }
5309             else
5310             {
5311                 /* This is a nested loop. Proceed to the normal recording function */
5312                 HeapFree(GetProcessHeap(), 0, control_frame);
5313             }
5314         }
5315     }
5316
5317     if(priv->recording)
5318     {
5319         record_instruction(&priv->record, ins);
5320         return;
5321     }
5322
5323     /* boolean if */
5324     if(ins->handler_idx == WINED3DSIH_IF)
5325     {
5326         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5327         list_add_head(&priv->control_frames, &control_frame->entry);
5328         control_frame->type = IF;
5329
5330         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
5331         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5332         if (!priv->muted && !bool_const)
5333         {
5334             shader_addline(buffer, "#if(FALSE){\n");
5335             priv->muted = TRUE;
5336             control_frame->muting = TRUE;
5337         }
5338         else shader_addline(buffer, "#if(TRUE) {\n");
5339
5340         return; /* Instruction is handled */
5341     }
5342     else if(ins->handler_idx == WINED3DSIH_IFC)
5343     {
5344         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5345         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5346         control_frame->type = IFC;
5347         control_frame->no.ifc = priv->num_ifcs++;
5348         list_add_head(&priv->control_frames, &control_frame->entry);
5349     }
5350     else if(ins->handler_idx == WINED3DSIH_ELSE)
5351     {
5352         struct list *e = list_head(&priv->control_frames);
5353         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5354
5355         if(control_frame->type == IF)
5356         {
5357             shader_addline(buffer, "#} else {\n");
5358             if(!priv->muted && !control_frame->muting)
5359             {
5360                 priv->muted = TRUE;
5361                 control_frame->muting = TRUE;
5362             }
5363             else if(control_frame->muting) priv->muted = FALSE;
5364             return; /* Instruction is handled. */
5365         }
5366         /* In case of an ifc, generate a HW shader instruction */
5367     }
5368     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5369     {
5370         struct list *e = list_head(&priv->control_frames);
5371         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5372
5373         if(control_frame->type == IF)
5374         {
5375             shader_addline(buffer, "#} endif\n");
5376             if(control_frame->muting) priv->muted = FALSE;
5377             list_remove(&control_frame->entry);
5378             HeapFree(GetProcessHeap(), 0, control_frame);
5379             return; /* Instruction is handled */
5380         }
5381     }
5382
5383     if(priv->muted) return;
5384
5385     /* Select handler */
5386     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5387
5388     /* Unhandled opcode */
5389     if (!hw_fct)
5390     {
5391         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5392         return;
5393     }
5394     hw_fct(ins);
5395
5396     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5397     {
5398         struct list *e = list_head(&priv->control_frames);
5399         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5400         list_remove(&control_frame->entry);
5401         HeapFree(GetProcessHeap(), 0, control_frame);
5402         priv->loop_depth--;
5403     }
5404     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5405     {
5406         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5407         struct list *e = list_head(&priv->control_frames);
5408         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5409         list_remove(&control_frame->entry);
5410         HeapFree(GetProcessHeap(), 0, control_frame);
5411     }
5412
5413
5414     shader_arb_add_instruction_modifiers(ins);
5415 }
5416
5417 const shader_backend_t arb_program_shader_backend = {
5418     shader_arb_handle_instruction,
5419     shader_arb_select,
5420     shader_arb_select_depth_blt,
5421     shader_arb_deselect_depth_blt,
5422     shader_arb_update_float_vertex_constants,
5423     shader_arb_update_float_pixel_constants,
5424     shader_arb_load_constants,
5425     shader_arb_load_np2fixup_constants,
5426     shader_arb_destroy,
5427     shader_arb_alloc,
5428     shader_arb_free,
5429     shader_arb_dirty_const,
5430     shader_arb_get_caps,
5431     shader_arb_color_fixup_supported,
5432 };
5433
5434 /* ARB_fragment_program fixed function pipeline replacement definitions */
5435 #define ARB_FFP_CONST_TFACTOR           0
5436 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5437 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5438 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5439 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5440
5441 struct arbfp_ffp_desc
5442 {
5443     struct ffp_frag_desc parent;
5444     GLuint shader;
5445     unsigned int num_textures_used;
5446 };
5447
5448 /* Context activation is done by the caller. */
5449 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5450     ENTER_GL();
5451     if(enable) {
5452         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5453         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5454     } else {
5455         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5456         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5457     }
5458     LEAVE_GL();
5459 }
5460
5461 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5462     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5463     struct shader_arb_priv *priv;
5464     /* Share private data between the shader backend and the pipeline replacement, if both
5465      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5466      * if no pixel shader is bound or not
5467      */
5468     if(This->shader_backend == &arb_program_shader_backend) {
5469         This->fragment_priv = This->shader_priv;
5470     } else {
5471         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5472         if(!This->fragment_priv) return E_OUTOFMEMORY;
5473     }
5474     priv = This->fragment_priv;
5475     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5476     {
5477         ERR("Failed to initialize rbtree.\n");
5478         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5479         return E_OUTOFMEMORY;
5480     }
5481     priv->use_arbfp_fixed_func = TRUE;
5482     return WINED3D_OK;
5483 }
5484
5485 /* Context activation is done by the caller. */
5486 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5487 {
5488     const struct wined3d_gl_info *gl_info = context;
5489     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5490
5491     ENTER_GL();
5492     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5493     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5494     HeapFree(GetProcessHeap(), 0, entry_arb);
5495     LEAVE_GL();
5496 }
5497
5498 /* Context activation is done by the caller. */
5499 static void arbfp_free(IWineD3DDevice *iface) {
5500     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5501     struct shader_arb_priv *priv = This->fragment_priv;
5502
5503     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5504     priv->use_arbfp_fixed_func = FALSE;
5505
5506     if(This->shader_backend != &arb_program_shader_backend) {
5507         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5508     }
5509 }
5510
5511 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5512 {
5513     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5514     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5515                            WINED3DTEXOPCAPS_SELECTARG1                  |
5516                            WINED3DTEXOPCAPS_SELECTARG2                  |
5517                            WINED3DTEXOPCAPS_MODULATE4X                  |
5518                            WINED3DTEXOPCAPS_MODULATE2X                  |
5519                            WINED3DTEXOPCAPS_MODULATE                    |
5520                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5521                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5522                            WINED3DTEXOPCAPS_ADD                         |
5523                            WINED3DTEXOPCAPS_SUBTRACT                    |
5524                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5525                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5526                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5527                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5528                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5529                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5530                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5531                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5532                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5533                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5534                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5535                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5536                            WINED3DTEXOPCAPS_LERP                        |
5537                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5538                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5539
5540     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5541
5542     caps->MaxTextureBlendStages   = 8;
5543     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5544 }
5545
5546 static void state_texfactor_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5547 {
5548     const struct wined3d_gl_info *gl_info = context->gl_info;
5549     const struct wined3d_state *state = &stateblock->state;
5550     IWineD3DDeviceImpl *device = stateblock->device;
5551     float col[4];
5552
5553     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5554      * application provided constants
5555      */
5556     if (device->shader_backend == &arb_program_shader_backend)
5557     {
5558         if (use_ps(state)) return;
5559
5560         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5561         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5562     }
5563
5564     D3DCOLORTOGLFLOAT4(state->render_states[WINED3DRS_TEXTUREFACTOR], col);
5565     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5566     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5567
5568 }
5569
5570 static void state_arb_specularenable(DWORD state_id,
5571         IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5572 {
5573     const struct wined3d_gl_info *gl_info = context->gl_info;
5574     const struct wined3d_state *state = &stateblock->state;
5575     IWineD3DDeviceImpl *device = stateblock->device;
5576     float col[4];
5577
5578     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5579      * application provided constants
5580      */
5581     if (device->shader_backend == &arb_program_shader_backend)
5582     {
5583         if (use_ps(state)) return;
5584
5585         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5586         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5587     }
5588
5589     if (state->render_states[WINED3DRS_SPECULARENABLE])
5590     {
5591         /* The specular color has no alpha */
5592         col[0] = 1.0f; col[1] = 1.0f;
5593         col[2] = 1.0f; col[3] = 0.0f;
5594     } else {
5595         col[0] = 0.0f; col[1] = 0.0f;
5596         col[2] = 0.0f; col[3] = 0.0f;
5597     }
5598     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5599     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5600 }
5601
5602 static void set_bumpmat_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5603 {
5604     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5605     const struct wined3d_gl_info *gl_info = context->gl_info;
5606     const struct wined3d_state *state = &stateblock->state;
5607     IWineD3DDeviceImpl *device = stateblock->device;
5608     float mat[2][2];
5609
5610     if (use_ps(state))
5611     {
5612         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
5613         if (stage && (ps->baseShader.reg_maps.bumpmat & (1 << stage)))
5614         {
5615             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5616              * anyway
5617              */
5618             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5619                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5620         }
5621
5622         if(device->shader_backend == &arb_program_shader_backend) {
5623             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5624             return;
5625         }
5626     } else if(device->shader_backend == &arb_program_shader_backend) {
5627         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5628         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5629     }
5630
5631     mat[0][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5632     mat[0][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5633     mat[1][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5634     mat[1][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5635
5636     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5637     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5638 }
5639
5640 static void tex_bumpenvlum_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5641 {
5642     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5643     const struct wined3d_gl_info *gl_info = context->gl_info;
5644     const struct wined3d_state *state = &stateblock->state;
5645     IWineD3DDeviceImpl *device = stateblock->device;
5646     float param[4];
5647
5648     if (use_ps(state))
5649     {
5650         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
5651         if (stage && (ps->baseShader.reg_maps.luminanceparams & (1 << stage)))
5652         {
5653             /* The pixel shader has to know the luminance offset. Do a constants update if it
5654              * isn't scheduled anyway
5655              */
5656             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5657                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5658         }
5659
5660         if(device->shader_backend == &arb_program_shader_backend) {
5661             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5662             return;
5663         }
5664     } else if(device->shader_backend == &arb_program_shader_backend) {
5665         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5666         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5667     }
5668
5669     param[0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5670     param[1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5671     param[2] = 0.0f;
5672     param[3] = 0.0f;
5673
5674     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5675     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5676 }
5677
5678 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5679 {
5680     const char *ret;
5681
5682     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5683
5684     switch(arg & WINED3DTA_SELECTMASK) {
5685         case WINED3DTA_DIFFUSE:
5686             ret = "fragment.color.primary"; break;
5687
5688         case WINED3DTA_CURRENT:
5689             if (!stage) ret = "fragment.color.primary";
5690             else ret = "ret";
5691             break;
5692
5693         case WINED3DTA_TEXTURE:
5694             switch(stage) {
5695                 case 0: ret = "tex0"; break;
5696                 case 1: ret = "tex1"; break;
5697                 case 2: ret = "tex2"; break;
5698                 case 3: ret = "tex3"; break;
5699                 case 4: ret = "tex4"; break;
5700                 case 5: ret = "tex5"; break;
5701                 case 6: ret = "tex6"; break;
5702                 case 7: ret = "tex7"; break;
5703                 default: ret = "unknown texture";
5704             }
5705             break;
5706
5707         case WINED3DTA_TFACTOR:
5708             ret = "tfactor"; break;
5709
5710         case WINED3DTA_SPECULAR:
5711             ret = "fragment.color.secondary"; break;
5712
5713         case WINED3DTA_TEMP:
5714             ret = "tempreg"; break;
5715
5716         case WINED3DTA_CONSTANT:
5717             FIXME("Implement perstage constants\n");
5718             switch(stage) {
5719                 case 0: ret = "const0"; break;
5720                 case 1: ret = "const1"; break;
5721                 case 2: ret = "const2"; break;
5722                 case 3: ret = "const3"; break;
5723                 case 4: ret = "const4"; break;
5724                 case 5: ret = "const5"; break;
5725                 case 6: ret = "const6"; break;
5726                 case 7: ret = "const7"; break;
5727                 default: ret = "unknown constant";
5728             }
5729             break;
5730
5731         default:
5732             return "unknown";
5733     }
5734
5735     if(arg & WINED3DTA_COMPLEMENT) {
5736         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5737         if(argnum == 0) ret = "arg0";
5738         if(argnum == 1) ret = "arg1";
5739         if(argnum == 2) ret = "arg2";
5740     }
5741     if(arg & WINED3DTA_ALPHAREPLICATE) {
5742         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5743         if(argnum == 0) ret = "arg0";
5744         if(argnum == 1) ret = "arg1";
5745         if(argnum == 2) ret = "arg2";
5746     }
5747     return ret;
5748 }
5749
5750 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5751         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5752 {
5753     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5754     unsigned int mul = 1;
5755     BOOL mul_final_dest = FALSE;
5756
5757     if(color && alpha) dstmask = "";
5758     else if(color) dstmask = ".xyz";
5759     else dstmask = ".w";
5760
5761     if(dst == tempreg) dstreg = "tempreg";
5762     else dstreg = "ret";
5763
5764     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5765     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5766     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5767
5768     switch(op) {
5769         case WINED3DTOP_DISABLE:
5770             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5771             break;
5772
5773         case WINED3DTOP_SELECTARG2:
5774             arg1 = arg2;
5775         case WINED3DTOP_SELECTARG1:
5776             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5777             break;
5778
5779         case WINED3DTOP_MODULATE4X:
5780             mul = 2;
5781         case WINED3DTOP_MODULATE2X:
5782             mul *= 2;
5783             if (!strcmp(dstreg, "result.color"))
5784             {
5785                 dstreg = "ret";
5786                 mul_final_dest = TRUE;
5787             }
5788         case WINED3DTOP_MODULATE:
5789             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5790             break;
5791
5792         case WINED3DTOP_ADDSIGNED2X:
5793             mul = 2;
5794             if (!strcmp(dstreg, "result.color"))
5795             {
5796                 dstreg = "ret";
5797                 mul_final_dest = TRUE;
5798             }
5799         case WINED3DTOP_ADDSIGNED:
5800             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5801             arg2 = "arg2";
5802         case WINED3DTOP_ADD:
5803             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5804             break;
5805
5806         case WINED3DTOP_SUBTRACT:
5807             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5808             break;
5809
5810         case WINED3DTOP_ADDSMOOTH:
5811             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5812             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5813             break;
5814
5815         case WINED3DTOP_BLENDCURRENTALPHA:
5816             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5817             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5818             break;
5819         case WINED3DTOP_BLENDFACTORALPHA:
5820             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5821             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5822             break;
5823         case WINED3DTOP_BLENDTEXTUREALPHA:
5824             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5825             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5826             break;
5827         case WINED3DTOP_BLENDDIFFUSEALPHA:
5828             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5829             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5830             break;
5831
5832         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5833             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5834             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5835             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5836             break;
5837
5838         /* D3DTOP_PREMODULATE ???? */
5839
5840         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5841             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5842             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5843             break;
5844         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5845             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5846             break;
5847         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5848             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5849             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5850             break;
5851         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5852             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5853             break;
5854
5855         case WINED3DTOP_DOTPRODUCT3:
5856             mul = 4;
5857             if (!strcmp(dstreg, "result.color"))
5858             {
5859                 dstreg = "ret";
5860                 mul_final_dest = TRUE;
5861             }
5862             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5863             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5864             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5865             break;
5866
5867         case WINED3DTOP_MULTIPLYADD:
5868             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5869             break;
5870
5871         case WINED3DTOP_LERP:
5872             /* The msdn is not quite right here */
5873             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5874             break;
5875
5876         case WINED3DTOP_BUMPENVMAP:
5877         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5878             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5879             break;
5880
5881         default:
5882             FIXME("Unhandled texture op %08x\n", op);
5883     }
5884
5885     if(mul == 2) {
5886         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5887     } else if(mul == 4) {
5888         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5889     }
5890 }
5891
5892 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5893 {
5894     const struct wined3d_gl_info *gl_info = &stateblock->device->adapter->gl_info;
5895     unsigned int stage;
5896     struct wined3d_shader_buffer buffer;
5897     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5898     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5899     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5900     const char *textype;
5901     const char *instr, *sat;
5902     char colorcor_dst[8];
5903     GLuint ret;
5904     DWORD arg0, arg1, arg2;
5905     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5906     BOOL op_equal;
5907     const char *final_combiner_src = "ret";
5908     GLint pos;
5909
5910     /* Find out which textures are read */
5911     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5912         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5913         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5914         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5915         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5916         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5917         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5918         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5919
5920         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5921         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5922         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5923             bump_used[stage] = TRUE;
5924             tex_read[stage] = TRUE;
5925         }
5926         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5927             bump_used[stage] = TRUE;
5928             tex_read[stage] = TRUE;
5929             luminance_used[stage] = TRUE;
5930         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5931             tfactor_used = TRUE;
5932         }
5933
5934         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5935             tfactor_used = TRUE;
5936         }
5937
5938         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5939         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5940             tempreg_used = TRUE;
5941         }
5942
5943         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5944         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5945         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5946         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5947         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5948         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5949         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5950
5951         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5952             tempreg_used = TRUE;
5953         }
5954         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5955             tfactor_used = TRUE;
5956         }
5957     }
5958
5959     /* Shader header */
5960     if (!shader_buffer_init(&buffer))
5961     {
5962         ERR("Failed to initialize shader buffer.\n");
5963         return 0;
5964     }
5965
5966     shader_addline(&buffer, "!!ARBfp1.0\n");
5967
5968     switch(settings->fog) {
5969         case FOG_OFF:                                                         break;
5970         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5971         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5972         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5973         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5974     }
5975
5976     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5977     shader_addline(&buffer, "TEMP TMP;\n");
5978     shader_addline(&buffer, "TEMP ret;\n");
5979     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5980     shader_addline(&buffer, "TEMP arg0;\n");
5981     shader_addline(&buffer, "TEMP arg1;\n");
5982     shader_addline(&buffer, "TEMP arg2;\n");
5983     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5984         if(!tex_read[stage]) continue;
5985         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5986         if(!bump_used[stage]) continue;
5987         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5988         if(!luminance_used[stage]) continue;
5989         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5990     }
5991     if(tfactor_used) {
5992         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5993     }
5994         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5995
5996     if(settings->sRGB_write) {
5997         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5998                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5999         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6000                        srgb_sub_high, 0.0, 0.0, 0.0);
6001     }
6002
6003     if (ffp_clip_emul(&stateblock->state) && settings->emul_clipplanes)
6004         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6005
6006     /* Generate texture sampling instructions) */
6007     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
6008         if(!tex_read[stage]) continue;
6009
6010         switch(settings->op[stage].tex_type) {
6011             case tex_1d:                    textype = "1D";     break;
6012             case tex_2d:                    textype = "2D";     break;
6013             case tex_3d:                    textype = "3D";     break;
6014             case tex_cube:                  textype = "CUBE";   break;
6015             case tex_rect:                  textype = "RECT";   break;
6016             default: textype = "unexpected_textype";   break;
6017         }
6018
6019         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
6020            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6021             sat = "";
6022         } else {
6023             sat = "_SAT";
6024         }
6025
6026         if(settings->op[stage].projected == proj_none) {
6027             instr = "TEX";
6028         } else if(settings->op[stage].projected == proj_count4 ||
6029                   settings->op[stage].projected == proj_count3) {
6030             instr = "TXP";
6031         } else {
6032             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6033             instr = "TXP";
6034         }
6035
6036         if(stage > 0 &&
6037            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6038             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6039             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6040             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6041             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6042             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6043
6044             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6045              * so multiply the displacement with the dividing parameter before passing it to TXP
6046              */
6047             if (settings->op[stage].projected != proj_none) {
6048                 if(settings->op[stage].projected == proj_count4) {
6049                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6050                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6051                 } else {
6052                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6053                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6054                 }
6055             } else {
6056                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6057             }
6058
6059             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6060                            instr, sat, stage, stage, textype);
6061             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6062                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6063                                stage - 1, stage - 1, stage - 1);
6064                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6065             }
6066         } else if(settings->op[stage].projected == proj_count3) {
6067             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6068             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6069             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6070                             instr, sat, stage, stage, textype);
6071         } else {
6072             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6073                             instr, sat, stage, stage, stage, textype);
6074         }
6075
6076         sprintf(colorcor_dst, "tex%u", stage);
6077         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6078                 settings->op[stage].color_fixup);
6079     }
6080
6081     /* Generate the main shader */
6082     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6083     {
6084         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6085         {
6086             if (!stage) final_combiner_src = "fragment.color.primary";
6087             break;
6088         }
6089
6090         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6091            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6092             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6093         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6094                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6095             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6096         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6097                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6098             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6099         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6100                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6101             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6102         } else {
6103             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6104                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6105                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6106                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6107         }
6108
6109         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6110             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6111                           settings->op[stage].cop, settings->op[stage].carg0,
6112                           settings->op[stage].carg1, settings->op[stage].carg2);
6113             if (!stage)
6114                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6115         }
6116         else if (op_equal)
6117         {
6118             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6119                           settings->op[stage].cop, settings->op[stage].carg0,
6120                           settings->op[stage].carg1, settings->op[stage].carg2);
6121         } else {
6122             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6123                           settings->op[stage].cop, settings->op[stage].carg0,
6124                           settings->op[stage].carg1, settings->op[stage].carg2);
6125             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6126                           settings->op[stage].aop, settings->op[stage].aarg0,
6127                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6128         }
6129     }
6130
6131     if(settings->sRGB_write) {
6132         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6133         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6134         shader_addline(&buffer, "MOV result.color, ret;\n");
6135     } else {
6136         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6137     }
6138
6139     /* Footer */
6140     shader_addline(&buffer, "END\n");
6141
6142     /* Generate the shader */
6143     GL_EXTCALL(glGenProgramsARB(1, &ret));
6144     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6145     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6146             strlen(buffer.buffer), buffer.buffer));
6147     checkGLcall("glProgramStringARB()");
6148
6149     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6150     if (pos != -1)
6151     {
6152         FIXME("Fragment program error at position %d: %s\n\n", pos,
6153               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6154         shader_arb_dump_program_source(buffer.buffer);
6155     }
6156     else
6157     {
6158         GLint native;
6159
6160         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6161         checkGLcall("glGetProgramivARB()");
6162         if (!native) WARN("Program exceeds native resource limits.\n");
6163     }
6164
6165     shader_buffer_free(&buffer);
6166     return ret;
6167 }
6168
6169 static void fragment_prog_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6170 {
6171     const struct wined3d_gl_info *gl_info = context->gl_info;
6172     const struct wined3d_state *state = &stateblock->state;
6173     IWineD3DDeviceImpl *device = stateblock->device;
6174     struct shader_arb_priv *priv = device->fragment_priv;
6175     BOOL use_vshader = use_vs(state);
6176     BOOL use_pshader = use_ps(state);
6177     struct ffp_frag_settings settings;
6178     const struct arbfp_ffp_desc *desc;
6179     unsigned int i;
6180
6181     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6182
6183     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
6184         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6185             /* Reload fixed function constants since they collide with the pixel shader constants */
6186             for(i = 0; i < MAX_TEXTURES; i++) {
6187                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6188             }
6189             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6190             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6191         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6192             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6193         }
6194         return;
6195     }
6196
6197     if(!use_pshader) {
6198         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
6199         gen_ffp_frag_op(stateblock, &settings, FALSE);
6200         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6201         if(!desc) {
6202             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6203             if (!new_desc)
6204             {
6205                 ERR("Out of memory\n");
6206                 return;
6207             }
6208             new_desc->num_textures_used = 0;
6209             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6210             {
6211                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6212                 new_desc->num_textures_used = i;
6213             }
6214
6215             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6216             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
6217             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6218             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6219             desc = new_desc;
6220         }
6221
6222         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6223          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6224          * deactivate it.
6225          */
6226         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6227         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6228         priv->current_fprogram_id = desc->shader;
6229
6230         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6231             /* Reload fixed function constants since they collide with the pixel shader constants */
6232             for(i = 0; i < MAX_TEXTURES; i++) {
6233                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6234             }
6235             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6236             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6237         }
6238         context->last_was_pshader = FALSE;
6239     } else {
6240         context->last_was_pshader = TRUE;
6241     }
6242
6243     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6244      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6245      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6246      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6247      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6248      *
6249      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6250      * shader handler
6251      */
6252     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6253         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6254
6255         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6256             stateblock_apply_state(STATE_VERTEXSHADERCONSTANT, stateblock, context);
6257     }
6258     if (use_pshader) stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
6259 }
6260
6261 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
6262  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
6263  * state table, so we need to handle that with a forwarding function. The other invisible side effect
6264  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
6265  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
6266  */
6267 static void state_arbfp_fog(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6268 {
6269     const struct wined3d_state *state = &stateblock->state;
6270     enum fogsource new_source;
6271
6272     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6273
6274     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6275         fragment_prog_arbfp(state_id, stateblock, context);
6276     }
6277
6278     if (!state->render_states[WINED3DRS_FOGENABLE]) return;
6279
6280     if (state->render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6281     {
6282         if (use_vs(state))
6283         {
6284             new_source = FOGSOURCE_VS;
6285         }
6286         else
6287         {
6288             if (state->render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6289                 new_source = FOGSOURCE_COORD;
6290             else
6291                 new_source = FOGSOURCE_FFP;
6292         }
6293     } else {
6294         new_source = FOGSOURCE_FFP;
6295     }
6296     if(new_source != context->fog_source) {
6297         context->fog_source = new_source;
6298         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6299     }
6300 }
6301
6302 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6303 {
6304     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6305         fragment_prog_arbfp(state, stateblock, context);
6306     }
6307 }
6308
6309 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6310     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6311     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6312     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6313     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6314     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6315     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6316     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6317     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6318     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6319     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6320     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6321     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6322     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6323     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6324     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6325     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6326     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6327     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6328     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6329     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6330     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6331     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6332     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6333     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6334     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6335     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6336     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6337     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6338     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6339     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6340     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6341     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6342     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6343     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6344     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6345     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6346     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6347     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6348     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6349     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6350     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6351     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6352     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6353     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6354     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6355     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6356     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6357     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6358     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6359     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6360     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6361     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6362     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6363     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6364     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6365     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6366     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6367     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6368     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6369     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6370     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6371     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6372     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6373     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6374     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6375     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6376     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6377     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6378     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6379     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6380     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6381     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6382     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6383     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6384     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6385     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6386     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6387     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6388     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6389     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6390     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6391     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6392     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6393     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6394     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6395     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6396     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6397     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6398     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6399     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6400     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6401     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6402     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6403     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6404     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6405     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6406     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6407     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6408     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6409     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6410     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6411     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6412     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6413     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6414     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6415     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6416     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6417     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6418     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6419     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6420     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6421     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6422     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6423     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6424     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6425     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6426     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6427     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6428     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6429     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6430     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6431     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6432     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6433     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6434     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6435     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6436     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6437     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6438     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6439     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6440     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6441     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6442     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6443     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6444     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6445     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6446     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6447     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6448     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6449     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6450 };
6451
6452 const struct fragment_pipeline arbfp_fragment_pipeline = {
6453     arbfp_enable,
6454     arbfp_get_caps,
6455     arbfp_alloc,
6456     arbfp_free,
6457     shader_arb_color_fixup_supported,
6458     arbfp_fragmentstate_template,
6459     TRUE /* We can disable projected textures */
6460 };
6461
6462 struct arbfp_blit_priv {
6463     GLenum yuy2_rect_shader, yuy2_2d_shader;
6464     GLenum uyvy_rect_shader, uyvy_2d_shader;
6465     GLenum yv12_rect_shader, yv12_2d_shader;
6466     GLenum p8_rect_shader, p8_2d_shader;
6467     GLuint palette_texture;
6468 };
6469
6470 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6471     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6472     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6473     if(!device->blit_priv) {
6474         ERR("Out of memory\n");
6475         return E_OUTOFMEMORY;
6476     }
6477     return WINED3D_OK;
6478 }
6479
6480 /* Context activation is done by the caller. */
6481 static void arbfp_blit_free(IWineD3DDevice *iface) {
6482     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6483     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6484     struct arbfp_blit_priv *priv = device->blit_priv;
6485
6486     ENTER_GL();
6487     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6488     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6489     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6490     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6491     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6492     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6493     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6494     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6495     checkGLcall("Delete yuv and p8 programs");
6496
6497     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6498     LEAVE_GL();
6499
6500     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6501     device->blit_priv = NULL;
6502 }
6503
6504 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6505         GLenum textype, char *luminance)
6506 {
6507     char chroma;
6508     const char *tex, *texinstr;
6509
6510     if (fixup == COMPLEX_FIXUP_UYVY) {
6511         chroma = 'x';
6512         *luminance = 'w';
6513     } else {
6514         chroma = 'w';
6515         *luminance = 'x';
6516     }
6517     switch(textype) {
6518         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6519         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6520         default:
6521             /* This is more tricky than just replacing the texture type - we have to navigate
6522              * properly in the texture to find the correct chroma values
6523              */
6524             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6525             return FALSE;
6526     }
6527
6528     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6529      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6530      * filtering when we sample the texture.
6531      *
6532      * These are the rules for reading the chroma:
6533      *
6534      * Even pixel: Cr
6535      * Even pixel: U
6536      * Odd pixel: V
6537      *
6538      * So we have to get the sampling x position in non-normalized coordinates in integers
6539      */
6540     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6541         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6542         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6543     } else {
6544         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6545     }
6546     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6547      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6548      * 0.5, so add 0.5.
6549      */
6550     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6551     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6552
6553     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6554      * even and odd pixels respectively
6555      */
6556     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6557     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6558
6559     /* Sample Pixel 1 */
6560     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6561
6562     /* Put the value into either of the chroma values */
6563     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6564     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6565     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6566     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6567
6568     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6569      * the pixel right to the current one. Otherwise, sample the left pixel.
6570      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6571      */
6572     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6573     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6574     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6575
6576     /* Put the value into the other chroma */
6577     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6578     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6579     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6580     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6581
6582     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6583      * the current one and lerp the two U and V values
6584      */
6585
6586     /* This gives the correctly filtered luminance value */
6587     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6588
6589     return TRUE;
6590 }
6591
6592 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6593 {
6594     const char *tex;
6595
6596     switch(textype) {
6597         case GL_TEXTURE_2D:             tex = "2D";     break;
6598         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6599         default:
6600             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6601             return FALSE;
6602     }
6603
6604     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6605      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6606      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6607      * pitch of the luminance plane, the packing into the gl texture is a bit
6608      * unfortunate. If the whole texture is interpreted as luminance data it looks
6609      * approximately like this:
6610      *
6611      *        +----------------------------------+----
6612      *        |                                  |
6613      *        |                                  |
6614      *        |                                  |
6615      *        |                                  |
6616      *        |                                  |   2
6617      *        |            LUMINANCE             |   -
6618      *        |                                  |   3
6619      *        |                                  |
6620      *        |                                  |
6621      *        |                                  |
6622      *        |                                  |
6623      *        +----------------+-----------------+----
6624      *        |                |                 |
6625      *        |  U even rows   |  U odd rows     |
6626      *        |                |                 |   1
6627      *        +----------------+------------------   -
6628      *        |                |                 |   3
6629      *        |  V even rows   |  V odd rows     |
6630      *        |                |                 |
6631      *        +----------------+-----------------+----
6632      *        |                |                 |
6633      *        |     0.5        |       0.5       |
6634      *
6635      * So it appears as if there are 4 chroma images, but in fact the odd rows
6636      * in the chroma images are in the same row as the even ones. So its is
6637      * kinda tricky to read
6638      *
6639      * When reading from rectangle textures, keep in mind that the input y coordinates
6640      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6641      */
6642     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6643             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6644
6645     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6646     /* the chroma planes have only half the width */
6647     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6648
6649     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6650      * the coordinate. Also read the right side of the image when reading odd lines
6651      *
6652      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6653      * bleeding
6654      */
6655     if(textype == GL_TEXTURE_2D) {
6656
6657         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6658
6659         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6660
6661         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6662         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6663
6664         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6665         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6666         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6667         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6668         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6669
6670         /* clamp, keep the half pixel origin in mind */
6671         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6672         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6673         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6674         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6675     } else {
6676         /* Read from [size - size+size/4] */
6677         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6678         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6679
6680         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6681         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6682         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6683         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6684         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6685         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6686
6687         /* Make sure to read exactly from the pixel center */
6688         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6689         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6690
6691         /* Clamp */
6692         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6693         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6694         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6695         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6696         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6697     }
6698     /* Read the texture, put the result into the output register */
6699     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6700     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6701
6702     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6703      * No need to clamp because we're just reusing the already clamped value from above
6704      */
6705     if(textype == GL_TEXTURE_2D) {
6706         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6707     } else {
6708         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6709     }
6710     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6711     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6712
6713     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6714      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6715      * values due to filtering
6716      */
6717     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6718     if(textype == GL_TEXTURE_2D) {
6719         /* Multiply the y coordinate by 2/3 and clamp it */
6720         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6721         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6722         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6723         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6724     } else {
6725         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6726          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6727          * is bigger
6728          */
6729         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6730         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6731         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6732     }
6733     *luminance = 'a';
6734
6735     return TRUE;
6736 }
6737
6738 static GLuint gen_p8_shader(IWineD3DDeviceImpl *device, GLenum textype)
6739 {
6740     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6741     GLenum shader;
6742     struct wined3d_shader_buffer buffer;
6743     struct arbfp_blit_priv *priv = device->blit_priv;
6744     GLint pos;
6745
6746     /* Shader header */
6747     if (!shader_buffer_init(&buffer))
6748     {
6749         ERR("Failed to initialize shader buffer.\n");
6750         return 0;
6751     }
6752
6753     ENTER_GL();
6754     GL_EXTCALL(glGenProgramsARB(1, &shader));
6755     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6756     LEAVE_GL();
6757     if(!shader) {
6758         shader_buffer_free(&buffer);
6759         return 0;
6760     }
6761
6762     shader_addline(&buffer, "!!ARBfp1.0\n");
6763     shader_addline(&buffer, "TEMP index;\n");
6764
6765     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6766     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6767
6768     /* The alpha-component contains the palette index */
6769     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6770         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6771     else
6772         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6773
6774     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6775     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6776
6777     /* Use the alpha-component as an index in the palette to get the final color */
6778     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6779     shader_addline(&buffer, "END\n");
6780
6781     ENTER_GL();
6782     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6783             strlen(buffer.buffer), buffer.buffer));
6784     checkGLcall("glProgramStringARB()");
6785
6786     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6787     if (pos != -1)
6788     {
6789         FIXME("Fragment program error at position %d: %s\n\n", pos,
6790               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6791         shader_arb_dump_program_source(buffer.buffer);
6792     }
6793
6794     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6795         priv->p8_rect_shader = shader;
6796     else
6797         priv->p8_2d_shader = shader;
6798
6799     shader_buffer_free(&buffer);
6800     LEAVE_GL();
6801
6802     return shader;
6803 }
6804
6805 /* Context activation is done by the caller. */
6806 static void upload_palette(IWineD3DSurfaceImpl *surface)
6807 {
6808     BYTE table[256][4];
6809     IWineD3DDeviceImpl *device = surface->resource.device;
6810     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6811     struct arbfp_blit_priv *priv = device->blit_priv;
6812     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6813
6814     d3dfmt_p8_init_palette(surface, table, colorkey);
6815
6816     ENTER_GL();
6817     if (!priv->palette_texture)
6818         glGenTextures(1, &priv->palette_texture);
6819
6820     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6821     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6822
6823     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6824
6825     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6826     /* Make sure we have discrete color levels. */
6827     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6828     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6829     /* Upload the palette */
6830     /* TODO: avoid unneeed uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6831     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6832
6833     /* Switch back to unit 0 in which the 2D texture will be stored. */
6834     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE0));
6835     LEAVE_GL();
6836 }
6837
6838 /* Context activation is done by the caller. */
6839 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum complex_fixup yuv_fixup, GLenum textype)
6840 {
6841     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6842     GLenum shader;
6843     struct wined3d_shader_buffer buffer;
6844     char luminance_component;
6845     struct arbfp_blit_priv *priv = device->blit_priv;
6846     GLint pos;
6847
6848     /* Shader header */
6849     if (!shader_buffer_init(&buffer))
6850     {
6851         ERR("Failed to initialize shader buffer.\n");
6852         return 0;
6853     }
6854
6855     ENTER_GL();
6856     GL_EXTCALL(glGenProgramsARB(1, &shader));
6857     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6858     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6859     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6860     LEAVE_GL();
6861     if(!shader) {
6862         shader_buffer_free(&buffer);
6863         return 0;
6864     }
6865
6866     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6867      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6868      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6869      * each single pixel it contains, and one U and one V value shared between both
6870      * pixels.
6871      *
6872      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6873      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6874      * take the format into account when generating the read swizzles
6875      *
6876      * Reading the Y value is straightforward - just sample the texture. The hardware
6877      * takes care of filtering in the horizontal and vertical direction.
6878      *
6879      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6880      * because that would mix the U and V values of one pixel or two adjacent pixels.
6881      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6882      * regardless of the filtering setting. Vertical filtering works automatically
6883      * though - the U and V values of two rows are mixed nicely.
6884      *
6885      * Appart of avoiding filtering issues, the code has to know which value it just
6886      * read, and where it can find the other one. To determine this, it checks if
6887      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6888      *
6889      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6890      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6891      *
6892      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6893      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6894      * in an unfiltered situation. Finding the luminance on the other hand requires
6895      * finding out if it is an odd or even pixel. The real drawback of this approach
6896      * is filtering. This would have to be emulated completely in the shader, reading
6897      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6898      * vertically. Beyond that it would require adjustments to the texture handling
6899      * code to deal with the width scaling
6900      */
6901     shader_addline(&buffer, "!!ARBfp1.0\n");
6902     shader_addline(&buffer, "TEMP luminance;\n");
6903     shader_addline(&buffer, "TEMP temp;\n");
6904     shader_addline(&buffer, "TEMP chroma;\n");
6905     shader_addline(&buffer, "TEMP texcrd;\n");
6906     shader_addline(&buffer, "TEMP texcrd2;\n");
6907     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6908     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6909     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6910
6911     switch (yuv_fixup)
6912     {
6913         case COMPLEX_FIXUP_UYVY:
6914         case COMPLEX_FIXUP_YUY2:
6915             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6916             {
6917                 shader_buffer_free(&buffer);
6918                 return 0;
6919             }
6920             break;
6921
6922         case COMPLEX_FIXUP_YV12:
6923             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6924             {
6925                 shader_buffer_free(&buffer);
6926                 return 0;
6927             }
6928             break;
6929
6930         default:
6931             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6932             shader_buffer_free(&buffer);
6933             return 0;
6934     }
6935
6936     /* Calculate the final result. Formula is taken from
6937      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6938      * ranges from -0.5 to 0.5
6939      */
6940     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6941
6942     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6943     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6944     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6945     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6946     shader_addline(&buffer, "END\n");
6947
6948     ENTER_GL();
6949     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6950             strlen(buffer.buffer), buffer.buffer));
6951     checkGLcall("glProgramStringARB()");
6952
6953     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6954     if (pos != -1)
6955     {
6956         FIXME("Fragment program error at position %d: %s\n\n", pos,
6957               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6958         shader_arb_dump_program_source(buffer.buffer);
6959     }
6960     else
6961     {
6962         GLint native;
6963
6964         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6965         checkGLcall("glGetProgramivARB()");
6966         if (!native) WARN("Program exceeds native resource limits.\n");
6967     }
6968
6969     shader_buffer_free(&buffer);
6970     LEAVE_GL();
6971
6972     switch (yuv_fixup)
6973     {
6974         case COMPLEX_FIXUP_YUY2:
6975             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6976             else priv->yuy2_2d_shader = shader;
6977             break;
6978
6979         case COMPLEX_FIXUP_UYVY:
6980             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6981             else priv->uyvy_2d_shader = shader;
6982             break;
6983
6984         case COMPLEX_FIXUP_YV12:
6985             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6986             else priv->yv12_2d_shader = shader;
6987             break;
6988         default:
6989             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
6990     }
6991
6992     return shader;
6993 }
6994
6995 /* Context activation is done by the caller. */
6996 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, IWineD3DSurfaceImpl *surface)
6997 {
6998     GLenum shader;
6999     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
7000     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7001     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7002     struct arbfp_blit_priv *priv = device->blit_priv;
7003     enum complex_fixup fixup;
7004     GLenum textype = surface->texture_target;
7005
7006     if (!is_complex_fixup(surface->resource.format->color_fixup))
7007     {
7008         TRACE("Fixup:\n");
7009         dump_color_fixup_desc(surface->resource.format->color_fixup);
7010         /* Don't bother setting up a shader for unconverted formats */
7011         ENTER_GL();
7012         glEnable(textype);
7013         checkGLcall("glEnable(textype)");
7014         LEAVE_GL();
7015         return WINED3D_OK;
7016     }
7017
7018     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7019
7020     switch(fixup)
7021     {
7022         case COMPLEX_FIXUP_YUY2:
7023             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7024             break;
7025
7026         case COMPLEX_FIXUP_UYVY:
7027             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7028             break;
7029
7030         case COMPLEX_FIXUP_YV12:
7031             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7032             break;
7033
7034         case COMPLEX_FIXUP_P8:
7035             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7036             if (!shader) shader = gen_p8_shader(device, textype);
7037
7038             upload_palette(surface);
7039             break;
7040
7041         default:
7042             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7043             ENTER_GL();
7044             glEnable(textype);
7045             checkGLcall("glEnable(textype)");
7046             LEAVE_GL();
7047             return E_NOTIMPL;
7048     }
7049
7050     if (!shader) shader = gen_yuv_shader(device, fixup, textype);
7051
7052     ENTER_GL();
7053     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7054     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7055     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7056     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7057     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7058     checkGLcall("glProgramLocalParameter4fvARB");
7059     LEAVE_GL();
7060
7061     return WINED3D_OK;
7062 }
7063
7064 /* Context activation is done by the caller. */
7065 static void arbfp_blit_unset(IWineD3DDevice *iface) {
7066     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
7067     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7068
7069     ENTER_GL();
7070     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7071     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7072     glDisable(GL_TEXTURE_2D);
7073     checkGLcall("glDisable(GL_TEXTURE_2D)");
7074     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7075     {
7076         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7077         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7078     }
7079     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7080     {
7081         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7082         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7083     }
7084     LEAVE_GL();
7085 }
7086
7087 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum blit_operation blit_op,
7088         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7089         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7090 {
7091     enum complex_fixup src_fixup;
7092
7093     if (blit_op != BLIT_OP_BLIT)
7094     {
7095         TRACE("Unsupported blit_op=%d\n", blit_op);
7096         return FALSE;
7097     }
7098
7099     src_fixup = get_complex_fixup(src_format->color_fixup);
7100     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7101     {
7102         TRACE("Checking support for fixup:\n");
7103         dump_color_fixup_desc(src_format->color_fixup);
7104     }
7105
7106     if (!is_identity_fixup(dst_format->color_fixup))
7107     {
7108         TRACE("Destination fixups are not supported\n");
7109         return FALSE;
7110     }
7111
7112     if (is_identity_fixup(src_format->color_fixup))
7113     {
7114         TRACE("[OK]\n");
7115         return TRUE;
7116     }
7117
7118      /* We only support YUV conversions. */
7119     if (!is_complex_fixup(src_format->color_fixup))
7120     {
7121         TRACE("[FAILED]\n");
7122         return FALSE;
7123     }
7124
7125     switch(src_fixup)
7126     {
7127         case COMPLEX_FIXUP_YUY2:
7128         case COMPLEX_FIXUP_UYVY:
7129         case COMPLEX_FIXUP_YV12:
7130         case COMPLEX_FIXUP_P8:
7131             TRACE("[OK]\n");
7132             return TRUE;
7133
7134         default:
7135             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7136             TRACE("[FAILED]\n");
7137             return FALSE;
7138     }
7139 }
7140
7141 HRESULT arbfp_blit_surface(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *src_surface, const RECT *src_rect,
7142                            IWineD3DSurfaceImpl *dst_surface, const RECT *dst_rect_in, enum blit_operation blit_op,
7143                            DWORD Filter)
7144 {
7145     IWineD3DSwapChainImpl *dst_swapchain;
7146     struct wined3d_context *context;
7147     RECT dst_rect = *dst_rect_in;
7148
7149     /* Now load the surface */
7150     surface_internal_preload(src_surface, SRGB_RGB);
7151
7152     /* Activate the destination context, set it up for blitting */
7153     context = context_acquire(device, dst_surface);
7154     context_apply_blit_state(context, device);
7155
7156     if (!surface_is_offscreen(dst_surface))
7157         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7158
7159     arbfp_blit_set((IWineD3DDevice *)device, src_surface);
7160
7161     ENTER_GL();
7162
7163     /* Draw a textured quad */
7164     draw_textured_quad(src_surface, src_rect, &dst_rect, Filter);
7165
7166     LEAVE_GL();
7167
7168     /* Leave the opengl state valid for blitting */
7169     arbfp_blit_unset((IWineD3DDevice *)device);
7170
7171     dst_swapchain = dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7172             ? dst_surface->container.u.swapchain : NULL;
7173     if (wined3d_settings.strict_draw_ordering || (dst_swapchain
7174             && (dst_surface == dst_swapchain->front_buffer
7175             || dst_swapchain->num_contexts > 1)))
7176         wglFlush(); /* Flush to ensure ordering across contexts. */
7177
7178     context_release(context);
7179
7180     surface_modify_location(dst_surface, SFLAG_INDRAWABLE, TRUE);
7181     return WINED3D_OK;
7182 }
7183
7184 /* Do not call while under the GL lock. */
7185 static HRESULT arbfp_blit_color_fill(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *dst_surface,
7186         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7187 {
7188     FIXME("Color filling not implemented by arbfp_blit\n");
7189     return WINED3DERR_INVALIDCALL;
7190 }
7191
7192 const struct blit_shader arbfp_blit = {
7193     arbfp_blit_alloc,
7194     arbfp_blit_free,
7195     arbfp_blit_set,
7196     arbfp_blit_unset,
7197     arbfp_blit_supported,
7198     arbfp_blit_color_fill
7199 };