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(IWineD3DDeviceImpl *device, UINT start, UINT count)
 657 {
 658     struct wined3d_context *context = context_get_current();
 659
 660     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 661      * context. On a context switch the old context will be fully dirtified */
 662     if (!context || context->swapchain->device != device) return;
 663
 664     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 665     device->highest_dirty_vs_const = max(device->highest_dirty_vs_const, start + count);
 666 }
 667
 668 static void shader_arb_update_float_pixel_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
 669 {
 670     struct wined3d_context *context = context_get_current();
 671
 672     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 673      * context. On a context switch the old context will be fully dirtified */
 674     if (!context || context->swapchain->device != device) return;
 675
 676     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 677     device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, start + count);
 678 }
 679
 680 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 681 {
 682     DWORD *ret;
 683     DWORD idx = 0;
 684     const local_constant *lconst;
 685
 686     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 687
 688     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 689     if(!ret) {
 690         ERR("Out of memory\n");
 691         return NULL;
 692     }
 693
 694     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 695         ret[lconst->idx] = idx++;
 696     }
 697     return ret;
 698 }
 699
 700 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 701 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 702         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 703         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 704 {
 705     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 706     DWORD i, next_local = 0;
 707     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 708     unsigned max_constantsF;
 709     const local_constant *lconst;
 710     DWORD map;
 711
 712     /* In pixel shaders, all private constants are program local, we don't need anything
 713      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 714      * If we need a private constant the GL implementation will squeeze it in somewhere
 715      *
 716      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 717      * immediate values. The posFixup is loaded using program.env for now, so always
 718      * subtract one from the number of constants. If the shader uses indirect addressing,
 719      * account for the helper const too because we have to declare all availabke d3d constants
 720      * and don't know which are actually used.
 721      */
 722     if (pshader)
 723     {
 724         max_constantsF = gl_info->limits.arb_ps_native_constants;
 725         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 726         if (max_constantsF < 24)
 727             max_constantsF = gl_info->limits.arb_ps_float_constants;
 728     }
 729     else
 730     {
 731         max_constantsF = gl_info->limits.arb_vs_native_constants;
 732         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 733          * Also prevents max_constantsF from becoming less than 0 and
 734          * wrapping . */
 735         if (max_constantsF < 96)
 736             max_constantsF = gl_info->limits.arb_vs_float_constants;
 737
 738         if(This->baseShader.reg_maps.usesrelconstF) {
 739             DWORD highest_constf = 0, clip_limit;
 740
 741             max_constantsF -= reserved_vs_const(This, gl_info);
 742             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 743
 744             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 745             {
 746                 DWORD idx = i >> 5;
 747                 DWORD shift = i & 0x1f;
 748                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 749             }
 750
 751             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 752             {
 753                 if(ctx->cur_vs_args->super.clip_enabled)
 754                     clip_limit = gl_info->limits.clipplanes;
 755                 else
 756                     clip_limit = 0;
 757             }
 758             else
 759             {
 760                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 761                 clip_limit = min(count_bits(mask), 4);
 762             }
 763             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 764             max_constantsF -= *num_clipplanes;
 765             if(*num_clipplanes < clip_limit)
 766             {
 767                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 768             }
 769         }
 770         else
 771         {
 772             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 773             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 774         }
 775     }
 776
 777     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 778     {
 779         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 780     }
 781
 782     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 783     {
 784         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 785     }
 786
 787     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 788     {
 789         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 790         {
 791             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 792         }
 793     }
 794
 795     /* Load local constants using the program-local space,
 796      * this avoids reloading them each time the shader is used
 797      */
 798     if(lconst_map) {
 799         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 800             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 801                            lconst_map[lconst->idx]);
 802             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 803         }
 804     }
 805
 806     /* After subtracting privately used constants from the hardware limit(they are loaded as
 807      * local constants), make sure the shader doesn't violate the env constant limit
 808      */
 809     if(pshader)
 810     {
 811         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 812     }
 813     else
 814     {
 815         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 816     }
 817
 818     /* Avoid declaring more constants than needed */
 819     max_constantsF = min(max_constantsF, This->baseShader.limits.constant_float);
 820
 821     /* we use the array-based constants array if the local constants are marked for loading,
 822      * because then we use indirect addressing, or when the local constant list is empty,
 823      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 824      * local constants do not declare the loaded constants as an array because ARB compilers usually
 825      * do not optimize unused constants away
 826      */
 827     if(This->baseShader.reg_maps.usesrelconstF) {
 828         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 829         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 830                     max_constantsF, max_constantsF - 1);
 831     } else {
 832         for(i = 0; i < max_constantsF; i++) {
 833             DWORD idx, mask;
 834             idx = i >> 5;
 835             mask = 1 << (i & 0x1f);
 836             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 837                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 838             }
 839         }
 840     }
 841
 842     return next_local;
 843 }
 844
 845 static const char * const shift_tab[] = {
 846     "dummy",     /*  0 (none) */
 847     "coefmul.x", /*  1 (x2)   */
 848     "coefmul.y", /*  2 (x4)   */
 849     "coefmul.z", /*  3 (x8)   */
 850     "coefmul.w", /*  4 (x16)  */
 851     "dummy",     /*  5 (x32)  */
 852     "dummy",     /*  6 (x64)  */
 853     "dummy",     /*  7 (x128) */
 854     "dummy",     /*  8 (d256) */
 855     "dummy",     /*  9 (d128) */
 856     "dummy",     /* 10 (d64)  */
 857     "dummy",     /* 11 (d32)  */
 858     "coefdiv.w", /* 12 (d16)  */
 859     "coefdiv.z", /* 13 (d8)   */
 860     "coefdiv.y", /* 14 (d4)   */
 861     "coefdiv.x"  /* 15 (d2)   */
 862 };
 863
 864 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 865         const struct wined3d_shader_dst_param *dst, char *write_mask)
 866 {
 867     char *ptr = write_mask;
 868
 869     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 870     {
 871         *ptr++ = '.';
 872         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 873         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 874         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 875         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 876     }
 877
 878     *ptr = '\0';
 879 }
 880
 881 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 882 {
 883     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 884      * but addressed as "rgba". To fix this we need to swap the register's x
 885      * and z components. */
 886     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 887     char *ptr = swizzle_str;
 888
 889     /* swizzle bits fields: wwzzyyxx */
 890     DWORD swizzle = param->swizzle;
 891     DWORD swizzle_x = swizzle & 0x03;
 892     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 893     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 894     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 895
 896     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 897      * generate a swizzle string. Unless we need to our own swizzling. */
 898     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 899     {
 900         *ptr++ = '.';
 901         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 902             *ptr++ = swizzle_chars[swizzle_x];
 903         } else {
 904             *ptr++ = swizzle_chars[swizzle_x];
 905             *ptr++ = swizzle_chars[swizzle_y];
 906             *ptr++ = swizzle_chars[swizzle_z];
 907             *ptr++ = swizzle_chars[swizzle_w];
 908         }
 909     }
 910
 911     *ptr = '\0';
 912 }
 913
 914 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 915 {
 916     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 917     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 918
 919     if (!strcmp(priv->addr_reg, src)) return;
 920
 921     strcpy(priv->addr_reg, src);
 922     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 923 }
 924
 925 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 926         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 927
 928 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 929         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 930 {
 931     /* oPos, oFog and oPts in D3D */
 932     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 933     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 934     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 935     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 936     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 937
 938     *is_color = FALSE;
 939
 940     switch (reg->type)
 941     {
 942         case WINED3DSPR_TEMP:
 943             sprintf(register_name, "R%u", reg->idx);
 944             break;
 945
 946         case WINED3DSPR_INPUT:
 947             if (pshader)
 948             {
 949                 if (reg_maps->shader_version.major < 3)
 950                 {
 951                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 952                     else strcpy(register_name, "fragment.color.secondary");
 953                 }
 954                 else
 955                 {
 956                     if(reg->rel_addr)
 957                     {
 958                         char rel_reg[50];
 959                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 960
 961                         if (!strcmp(rel_reg, "**aL_emul**"))
 962                         {
 963                             DWORD idx = ctx->aL + reg->idx;
 964                             if(idx < MAX_REG_INPUT)
 965                             {
 966                                 strcpy(register_name, ctx->ps_input[idx]);
 967                             }
 968                             else
 969                             {
 970                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 971                                 sprintf(register_name, "out_of_bounds_%u", idx);
 972                             }
 973                         }
 974                         else if (reg_maps->input_registers & 0x0300)
 975                         {
 976                             /* There are two ways basically:
 977                              *
 978                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 979                              *    That means trouble if the loop also contains a breakc or if the control values
 980                              *    aren't local constants.
 981                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 982                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 983                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 984                              *    ADAC to load the condition code register and pop it again afterwards
 985                              */
 986                             FIXME("Relative input register addressing with more than 8 registers\n");
 987
 988                             /* This is better than nothing for now */
 989                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 990                         }
 991                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 992                         {
 993                             /* This is problematic because we'd have to consult the ctx->ps_input strings
 994                              * for where to find the varying. Some may be "0.0", others can be texcoords or
 995                              * colors. This needs either a pipeline replacement to make the vertex shader feed
 996                              * proper varyings, or loop unrolling
 997                              *
 998                              * For now use the texcoords and hope for the best
 999                              */
1000                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1001                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1002                         }
1003                         else
1004                         {
1005                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1006                              * pulls GL_NV_fragment_program2 in
1007                              */
1008                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1009                         }
1010                     }
1011                     else
1012                     {
1013                         if(reg->idx < MAX_REG_INPUT)
1014                         {
1015                             strcpy(register_name, ctx->ps_input[reg->idx]);
1016                         }
1017                         else
1018                         {
1019                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1020                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1021                         }
1022                     }
1023                 }
1024             }
1025             else
1026             {
1027                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1028                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1029             }
1030             break;
1031
1032         case WINED3DSPR_CONST:
1033             if (!pshader && reg->rel_addr)
1034             {
1035                 BOOL aL = FALSE;
1036                 char rel_reg[50];
1037                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
1038                 if (reg_maps->shader_version.major < 2)
1039                 {
1040                     sprintf(rel_reg, "A0.x");
1041                 } else {
1042                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1043                     if(ctx->target_version == ARB) {
1044                         if (!strcmp(rel_reg, "**aL_emul**"))
1045                         {
1046                             aL = TRUE;
1047                         } else {
1048                             shader_arb_request_a0(ins, rel_reg);
1049                             sprintf(rel_reg, "A0.x");
1050                         }
1051                     }
1052                 }
1053                 if(aL)
1054                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1055                 else if (reg->idx >= rel_offset)
1056                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1057                 else
1058                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1059             }
1060             else
1061             {
1062                 if (reg_maps->usesrelconstF)
1063                     sprintf(register_name, "C[%u]", reg->idx);
1064                 else
1065                     sprintf(register_name, "C%u", reg->idx);
1066             }
1067             break;
1068
1069         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1070             if (pshader)
1071             {
1072                 if (reg_maps->shader_version.major == 1
1073                         && reg_maps->shader_version.minor <= 3)
1074                 {
1075                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1076                      * and as source to most instructions. For some instructions it is the texcoord
1077                      * input. Those instructions know about the special use
1078                      */
1079                     sprintf(register_name, "T%u", reg->idx);
1080                 } else {
1081                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1082                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1083                 }
1084             }
1085             else
1086             {
1087                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1088                 {
1089                     sprintf(register_name, "A%u", reg->idx);
1090                 }
1091                 else
1092                 {
1093                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1094                 }
1095             }
1096             break;
1097
1098         case WINED3DSPR_COLOROUT:
1099             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1100             {
1101                 strcpy(register_name, "TMP_COLOR");
1102             }
1103             else
1104             {
1105                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1106                 if (reg_maps->highest_render_target > 0)
1107                 {
1108                     sprintf(register_name, "result.color[%u]", reg->idx);
1109                 }
1110                 else
1111                 {
1112                     strcpy(register_name, "result.color");
1113                 }
1114             }
1115             break;
1116
1117         case WINED3DSPR_RASTOUT:
1118             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1119             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1120             break;
1121
1122         case WINED3DSPR_DEPTHOUT:
1123             strcpy(register_name, "result.depth");
1124             break;
1125
1126         case WINED3DSPR_ATTROUT:
1127         /* case WINED3DSPR_OUTPUT: */
1128             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1129             else strcpy(register_name, ctx->color_output[reg->idx]);
1130             break;
1131
1132         case WINED3DSPR_TEXCRDOUT:
1133             if (pshader)
1134             {
1135                 sprintf(register_name, "oT[%u]", reg->idx);
1136             }
1137             else
1138             {
1139                 if (reg_maps->shader_version.major < 3)
1140                 {
1141                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1142                 }
1143                 else
1144                 {
1145                     strcpy(register_name, ctx->vs_output[reg->idx]);
1146                 }
1147             }
1148             break;
1149
1150         case WINED3DSPR_LOOP:
1151             if(ctx->target_version >= NV2)
1152             {
1153                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1154                 if(pshader) sprintf(register_name, "A0.x");
1155                 else sprintf(register_name, "aL.y");
1156             }
1157             else
1158             {
1159                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1160                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1161                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1162                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1163                  * indexing
1164                  */
1165                 sprintf(register_name, "**aL_emul**");
1166             }
1167
1168             break;
1169
1170         case WINED3DSPR_CONSTINT:
1171             sprintf(register_name, "I%u", reg->idx);
1172             break;
1173
1174         case WINED3DSPR_MISCTYPE:
1175             if (!reg->idx)
1176             {
1177                 sprintf(register_name, "vpos");
1178             }
1179             else if(reg->idx == 1)
1180             {
1181                 sprintf(register_name, "fragment.facing.x");
1182             }
1183             else
1184             {
1185                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1186             }
1187             break;
1188
1189         default:
1190             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1191             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1192             break;
1193     }
1194 }
1195
1196 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1197         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1198 {
1199     char register_name[255];
1200     char write_mask[6];
1201     BOOL is_color;
1202
1203     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1204     strcpy(str, register_name);
1205
1206     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1207     strcat(str, write_mask);
1208 }
1209
1210 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1211 {
1212     switch(channel_source)
1213     {
1214         case CHANNEL_SOURCE_ZERO: return "0";
1215         case CHANNEL_SOURCE_ONE: return "1";
1216         case CHANNEL_SOURCE_X: return "x";
1217         case CHANNEL_SOURCE_Y: return "y";
1218         case CHANNEL_SOURCE_Z: return "z";
1219         case CHANNEL_SOURCE_W: return "w";
1220         default:
1221             FIXME("Unhandled channel source %#x\n", channel_source);
1222             return "undefined";
1223     }
1224 }
1225
1226 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1227         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1228 {
1229     DWORD mask;
1230
1231     if (is_complex_fixup(fixup))
1232     {
1233         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1234         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1235         return;
1236     }
1237
1238     mask = 0;
1239     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1240     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1241     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1242     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1243     mask &= dst_mask;
1244
1245     if (mask)
1246     {
1247         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1248                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1249                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1250     }
1251
1252     mask = 0;
1253     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1254     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1255     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1256     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1257     mask &= dst_mask;
1258
1259     if (mask)
1260     {
1261         char reg_mask[6];
1262         char *ptr = reg_mask;
1263
1264         if (mask != WINED3DSP_WRITEMASK_ALL)
1265         {
1266             *ptr++ = '.';
1267             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1268             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1269             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1270             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1271         }
1272         *ptr = '\0';
1273
1274         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1275     }
1276 }
1277
1278 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1279 {
1280     DWORD mod;
1281     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1282     if (!ins->dst_count) return "";
1283
1284     mod = ins->dst[0].modifiers;
1285
1286     /* Silently ignore PARTIALPRECISION if its not supported */
1287     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1288
1289     if(mod & WINED3DSPDM_MSAMPCENTROID)
1290     {
1291         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1292         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1293     }
1294
1295     switch(mod)
1296     {
1297         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1298             return "H_SAT";
1299
1300         case WINED3DSPDM_SATURATE:
1301             return "_SAT";
1302
1303         case WINED3DSPDM_PARTIALPRECISION:
1304             return "H";
1305
1306         case 0:
1307             return "";
1308
1309         default:
1310             FIXME("Unknown modifiers 0x%08x\n", mod);
1311             return "";
1312     }
1313 }
1314
1315 #define TEX_PROJ        0x1
1316 #define TEX_BIAS        0x2
1317 #define TEX_LOD         0x4
1318 #define TEX_DERIV       0x10
1319
1320 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1321         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1322 {
1323     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1324     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1325     IWineD3DBaseTextureImpl *texture;
1326     const char *tex_type;
1327     BOOL np2_fixup = FALSE;
1328     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1329     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1330     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1331     const char *mod;
1332     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1333
1334     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1335     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1336
1337     switch(sampler_type) {
1338         case WINED3DSTT_1D:
1339             tex_type = "1D";
1340             break;
1341
1342         case WINED3DSTT_2D:
1343             texture = device->stateBlock->state.textures[sampler_idx];
1344             if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1345             {
1346                 tex_type = "RECT";
1347             } else {
1348                 tex_type = "2D";
1349             }
1350             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1351             {
1352                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1353                 {
1354                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1355                     else np2_fixup = TRUE;
1356                 }
1357             }
1358             break;
1359
1360         case WINED3DSTT_VOLUME:
1361             tex_type = "3D";
1362             break;
1363
1364         case WINED3DSTT_CUBE:
1365             tex_type = "CUBE";
1366             break;
1367
1368         default:
1369             ERR("Unexpected texture type %d\n", sampler_type);
1370             tex_type = "";
1371     }
1372
1373     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1374      * so don't use shader_arb_get_modifier
1375      */
1376     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1377     else mod = "";
1378
1379     /* Fragment samplers always have indentity mapping */
1380     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1381     {
1382         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1383     }
1384
1385     if (flags & TEX_DERIV)
1386     {
1387         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1388         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1389         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1390                        dsx, dsy,sampler_idx, tex_type);
1391     }
1392     else if(flags & TEX_LOD)
1393     {
1394         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1395         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1396         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1397                        sampler_idx, tex_type);
1398     }
1399     else if (flags & TEX_BIAS)
1400     {
1401         /* Shouldn't be possible, but let's check for it */
1402         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1403         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1404         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1405     }
1406     else if (flags & TEX_PROJ)
1407     {
1408         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1409     }
1410     else
1411     {
1412         if (np2_fixup)
1413         {
1414             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1415             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1416                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1417
1418             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1419         }
1420         else
1421             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1422     }
1423
1424     if (pshader)
1425     {
1426         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1427                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1428                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1429                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1430     }
1431 }
1432
1433 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1434         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1435 {
1436     /* Generate a line that does the input modifier computation and return the input register to use */
1437     BOOL is_color = FALSE;
1438     char regstr[256];
1439     char swzstr[20];
1440     int insert_line;
1441     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1442     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1443     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1444     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1445
1446     /* Assume a new line will be added */
1447     insert_line = 1;
1448
1449     /* Get register name */
1450     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1451     shader_arb_get_swizzle(src, is_color, swzstr);
1452
1453     switch (src->modifiers)
1454     {
1455     case WINED3DSPSM_NONE:
1456         sprintf(outregstr, "%s%s", regstr, swzstr);
1457         insert_line = 0;
1458         break;
1459     case WINED3DSPSM_NEG:
1460         sprintf(outregstr, "-%s%s", regstr, swzstr);
1461         insert_line = 0;
1462         break;
1463     case WINED3DSPSM_BIAS:
1464         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1465         break;
1466     case WINED3DSPSM_BIASNEG:
1467         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1468         break;
1469     case WINED3DSPSM_SIGN:
1470         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1471         break;
1472     case WINED3DSPSM_SIGNNEG:
1473         shader_addline(buffer, "MAD T%c, %s, %s, %s;\n", 'A' + tmpreg, regstr, two, one);
1474         break;
1475     case WINED3DSPSM_COMP:
1476         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1477         break;
1478     case WINED3DSPSM_X2:
1479         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1480         break;
1481     case WINED3DSPSM_X2NEG:
1482         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1483         break;
1484     case WINED3DSPSM_DZ:
1485         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1486         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1487         break;
1488     case WINED3DSPSM_DW:
1489         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1490         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1491         break;
1492     case WINED3DSPSM_ABS:
1493         if(ctx->target_version >= NV2) {
1494             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1495             insert_line = 0;
1496         } else {
1497             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1498         }
1499         break;
1500     case WINED3DSPSM_ABSNEG:
1501         if(ctx->target_version >= NV2) {
1502             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1503         } else {
1504             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1505             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1506         }
1507         insert_line = 0;
1508         break;
1509     default:
1510         sprintf(outregstr, "%s%s", regstr, swzstr);
1511         insert_line = 0;
1512     }
1513
1514     /* Return modified or original register, with swizzle */
1515     if (insert_line)
1516         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1517 }
1518
1519 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1520 {
1521     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1522     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1523     char dst_name[50];
1524     char src_name[2][50];
1525     DWORD sampler_code = dst->reg.idx;
1526
1527     shader_arb_get_dst_param(ins, dst, dst_name);
1528
1529     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1530      *
1531      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1532      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1533      * temps is done.
1534      */
1535     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1536     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1537     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1538     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1539     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1540
1541     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1542     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1543 }
1544
1545 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1546 {
1547     *extra_char = ' ';
1548     switch(mod)
1549     {
1550         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1551         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1552         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1553         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1554         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1555         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1556         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1557         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1558         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1559         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1560         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1561         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1562         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1563     }
1564     FIXME("Unknown modifier %u\n", mod);
1565     return mod;
1566 }
1567
1568 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1569 {
1570     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1571     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1572     char dst_name[50];
1573     char src_name[3][50];
1574     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1575             ins->ctx->reg_maps->shader_version.minor);
1576     BOOL is_color;
1577
1578     shader_arb_get_dst_param(ins, dst, dst_name);
1579     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1580
1581     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1582     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1583     {
1584         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1585     } else {
1586         struct wined3d_shader_src_param src0_copy = ins->src[0];
1587         char extra_neg;
1588
1589         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1590         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1591
1592         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1593         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1594         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1595         /* No modifiers supported on CMP */
1596         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1597
1598         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1599         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1600         {
1601             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1602             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1603         }
1604     }
1605 }
1606
1607 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1608 {
1609     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1610     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1611     char dst_name[50];
1612     char src_name[3][50];
1613     BOOL is_color;
1614
1615     shader_arb_get_dst_param(ins, dst, dst_name);
1616
1617     /* Generate input register names (with modifiers) */
1618     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1619     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1620     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1621
1622     /* No modifiers are supported on CMP */
1623     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1624                    src_name[0], src_name[2], src_name[1]);
1625
1626     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1627     {
1628         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1629         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1630     }
1631 }
1632
1633 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1634  * dst = dot2(src0, src1) + src2 */
1635 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1636 {
1637     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1638     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1639     char dst_name[50];
1640     char src_name[3][50];
1641     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1642
1643     shader_arb_get_dst_param(ins, dst, dst_name);
1644     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1645     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1646
1647     if(ctx->target_version >= NV3)
1648     {
1649         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1650         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1651         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1652                        dst_name, src_name[0], src_name[1], src_name[2]);
1653     }
1654     else if(ctx->target_version >= NV2)
1655     {
1656         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1657          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1658          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1659          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1660          *
1661          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1662          *
1663          * .xyxy and other swizzles that we could get with this are not valid in
1664          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1665          */
1666         struct wined3d_shader_src_param tmp_param = ins->src[1];
1667         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1668         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1669
1670         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1671
1672         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1673                        dst_name, src_name[2], src_name[0], src_name[1]);
1674     }
1675     else
1676     {
1677         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1678         /* 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]
1679         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1680         */
1681         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1682         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1683         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1684         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1685     }
1686 }
1687
1688 /* Map the opcode 1-to-1 to the GL code */
1689 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1690 {
1691     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1692     const char *instruction;
1693     char arguments[256], dst_str[50];
1694     unsigned int i;
1695     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1696
1697     switch (ins->handler_idx)
1698     {
1699         case WINED3DSIH_ABS: instruction = "ABS"; break;
1700         case WINED3DSIH_ADD: instruction = "ADD"; break;
1701         case WINED3DSIH_CRS: instruction = "XPD"; break;
1702         case WINED3DSIH_DP3: instruction = "DP3"; break;
1703         case WINED3DSIH_DP4: instruction = "DP4"; break;
1704         case WINED3DSIH_DST: instruction = "DST"; break;
1705         case WINED3DSIH_FRC: instruction = "FRC"; break;
1706         case WINED3DSIH_LIT: instruction = "LIT"; break;
1707         case WINED3DSIH_LRP: instruction = "LRP"; break;
1708         case WINED3DSIH_MAD: instruction = "MAD"; break;
1709         case WINED3DSIH_MAX: instruction = "MAX"; break;
1710         case WINED3DSIH_MIN: instruction = "MIN"; break;
1711         case WINED3DSIH_MOV: instruction = "MOV"; break;
1712         case WINED3DSIH_MUL: instruction = "MUL"; break;
1713         case WINED3DSIH_SGE: instruction = "SGE"; break;
1714         case WINED3DSIH_SLT: instruction = "SLT"; break;
1715         case WINED3DSIH_SUB: instruction = "SUB"; break;
1716         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1717         case WINED3DSIH_DSX: instruction = "DDX"; break;
1718         default: instruction = "";
1719             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1720             break;
1721     }
1722
1723     /* Note that shader_arb_add_dst_param() adds spaces. */
1724     arguments[0] = '\0';
1725     shader_arb_get_dst_param(ins, dst, dst_str);
1726     for (i = 0; i < ins->src_count; ++i)
1727     {
1728         char operand[100];
1729         strcat(arguments, ", ");
1730         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1731         strcat(arguments, operand);
1732     }
1733     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1734 }
1735
1736 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1737 {
1738     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1739     shader_addline(buffer, "NOP;\n");
1740 }
1741
1742 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1743 {
1744     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1745     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1746     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1747     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1748     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1749     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1750     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1751
1752     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1753     char src0_param[256];
1754
1755     if(ins->handler_idx == WINED3DSIH_MOVA) {
1756         char write_mask[6];
1757         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1758
1759         if(ctx->target_version >= NV2) {
1760             shader_hw_map2gl(ins);
1761             return;
1762         }
1763         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1764         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1765
1766         /* This implements the mova formula used in GLSL. The first two instructions
1767          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1768          * in this case:
1769          * mova A0.x, 0.0
1770          *
1771          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1772          *
1773          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1774          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1775          */
1776         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1777         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1778
1779         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1780         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1781         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1782         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1783         {
1784             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1785         }
1786         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1787
1788         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1789     }
1790     else if (reg_maps->shader_version.major == 1
1791           && !shader_is_pshader_version(reg_maps->shader_version.type)
1792           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1793     {
1794         src0_param[0] = '\0';
1795         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1796         {
1797             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1798             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1799             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1800             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1801         }
1802         else
1803         {
1804             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1805              * with more than one component. Thus replicate the first source argument over all
1806              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1807             struct wined3d_shader_src_param tmp_src = ins->src[0];
1808             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1809             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1810             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1811         }
1812     }
1813     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1814     {
1815         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1816         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1817         {
1818             shader_addline(buffer, "#mov handled in srgb write code\n");
1819             return;
1820         }
1821         shader_hw_map2gl(ins);
1822     }
1823     else
1824     {
1825         shader_hw_map2gl(ins);
1826     }
1827 }
1828
1829 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1830 {
1831     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1832     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1833     char reg_dest[40];
1834
1835     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1836      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1837      */
1838     shader_arb_get_dst_param(ins, dst, reg_dest);
1839
1840     if (ins->ctx->reg_maps->shader_version.major >= 2)
1841     {
1842         const char *kilsrc = "TA";
1843         BOOL is_color;
1844
1845         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1846         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1847         {
1848             kilsrc = reg_dest;
1849         }
1850         else
1851         {
1852             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1853              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1854              * masked out components to 0(won't kill)
1855              */
1856             char x = '0', y = '0', z = '0', w = '0';
1857             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1858             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1859             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1860             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1861             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1862         }
1863         shader_addline(buffer, "KIL %s;\n", kilsrc);
1864     } else {
1865         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1866          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1867          *
1868          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1869          * or pass in any temporary register(in shader phase 2)
1870          */
1871         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1872             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1873         } else {
1874             shader_arb_get_dst_param(ins, dst, reg_dest);
1875         }
1876         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1877         shader_addline(buffer, "KIL TA;\n");
1878     }
1879 }
1880
1881 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1882 {
1883     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1884     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1885     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1886             ins->ctx->reg_maps->shader_version.minor);
1887     struct wined3d_shader_src_param src;
1888
1889     char reg_dest[40];
1890     char reg_coord[40];
1891     DWORD reg_sampler_code;
1892     WORD myflags = 0;
1893
1894     /* All versions have a destination register */
1895     shader_arb_get_dst_param(ins, dst, reg_dest);
1896
1897     /* 1.0-1.4: Use destination register number as texture code.
1898        2.0+: Use provided sampler number as texure code. */
1899     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1900         reg_sampler_code = dst->reg.idx;
1901     else
1902         reg_sampler_code = ins->src[1].reg.idx;
1903
1904     /* 1.0-1.3: Use the texcoord varying.
1905        1.4+: Use provided coordinate source register. */
1906     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1907         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1908     else {
1909         /* TEX is the only instruction that can handle DW and DZ natively */
1910         src = ins->src[0];
1911         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1912         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1913         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1914     }
1915
1916     /* projection flag:
1917      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1918      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1919      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1920      */
1921     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1922     {
1923         DWORD flags = 0;
1924         if (reg_sampler_code < MAX_TEXTURES)
1925             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1926         if (flags & WINED3D_PSARGS_PROJECTED)
1927             myflags |= TEX_PROJ;
1928     }
1929     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1930     {
1931         DWORD src_mod = ins->src[0].modifiers;
1932         if (src_mod == WINED3DSPSM_DZ) {
1933             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1934              * varying register, so we need a temp reg
1935              */
1936             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1937             strcpy(reg_coord, "TA");
1938             myflags |= TEX_PROJ;
1939         } else if(src_mod == WINED3DSPSM_DW) {
1940             myflags |= TEX_PROJ;
1941         }
1942     } else {
1943         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1944         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1945     }
1946     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1947 }
1948
1949 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1950 {
1951     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1952     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1953     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1954             ins->ctx->reg_maps->shader_version.minor);
1955     char dst_str[50];
1956
1957     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1958     {
1959         DWORD reg = dst->reg.idx;
1960
1961         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1962         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1963     } else {
1964         char reg_src[40];
1965
1966         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1967         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1968         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1969    }
1970 }
1971
1972 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1973 {
1974      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1975      DWORD flags = 0;
1976
1977      DWORD reg1 = ins->dst[0].reg.idx;
1978      char dst_str[50];
1979      char src_str[50];
1980
1981      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1982      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1983      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1984      /* Move .x first in case src_str is "TA" */
1985      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1986      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1987      if (reg1 < MAX_TEXTURES)
1988      {
1989          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1990          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1991      }
1992      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1993 }
1994
1995 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1996 {
1997      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1998
1999      DWORD reg1 = ins->dst[0].reg.idx;
2000      char dst_str[50];
2001      char src_str[50];
2002
2003      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2004      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2005      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2006      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2007      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2008      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2009 }
2010
2011 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2012 {
2013     DWORD reg1 = ins->dst[0].reg.idx;
2014     char dst_str[50];
2015     char src_str[50];
2016
2017     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2018     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2019     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2020     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2021 }
2022
2023 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2024 {
2025     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2026     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2027     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2028     char reg_coord[40], dst_reg[50], src_reg[50];
2029     DWORD reg_dest_code;
2030
2031     /* All versions have a destination register. The Tx where the texture coordinates come
2032      * from is the varying incarnation of the texture register
2033      */
2034     reg_dest_code = dst->reg.idx;
2035     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2036     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2037     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2038
2039     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2040      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2041      *
2042      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2043      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2044      *
2045      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2046      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2047      * extension.
2048      */
2049     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2050     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2051     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2052     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2053
2054     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2055      * so we can't let the GL handle this.
2056      */
2057     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2058             & WINED3D_PSARGS_PROJECTED)
2059     {
2060         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2061         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2062         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2063     } else {
2064         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2065     }
2066
2067     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2068
2069     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2070     {
2071         /* No src swizzles are allowed, so this is ok */
2072         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2073                        src_reg, reg_dest_code, reg_dest_code);
2074         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2075     }
2076 }
2077
2078 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2079 {
2080     DWORD reg = ins->dst[0].reg.idx;
2081     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2082     char src0_name[50], dst_name[50];
2083     BOOL is_color;
2084     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2085
2086     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2087     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2088      * T<reg+1> register. Use this register to store the calculated vector
2089      */
2090     tmp_reg.idx = reg + 1;
2091     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2092     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2093 }
2094
2095 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2096 {
2097     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2098     DWORD flags;
2099     DWORD reg = ins->dst[0].reg.idx;
2100     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2101     char dst_str[50];
2102     char src0_name[50];
2103     char dst_reg[50];
2104     BOOL is_color;
2105
2106     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2107     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2108
2109     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2110     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2111     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2112     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2113     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2114 }
2115
2116 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2117 {
2118     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2119     DWORD reg = ins->dst[0].reg.idx;
2120     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2121     char src0_name[50], dst_name[50];
2122     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2123     BOOL is_color;
2124
2125     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2126      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2127      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2128      */
2129     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2130     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2131
2132     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2133     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2134                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2135     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2136 }
2137
2138 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2139 {
2140     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2141     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2142     DWORD flags;
2143     DWORD reg = ins->dst[0].reg.idx;
2144     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2145     char dst_str[50];
2146     char src0_name[50], dst_name[50];
2147     BOOL is_color;
2148
2149     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2150     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2151     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2152
2153     /* Sample the texture using the calculated coordinates */
2154     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2155     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2156     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2157     tex_mx->current_row = 0;
2158 }
2159
2160 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2161 {
2162     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2163     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2164     DWORD flags;
2165     DWORD reg = ins->dst[0].reg.idx;
2166     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2167     char dst_str[50];
2168     char src0_name[50];
2169     char dst_reg[50];
2170     BOOL is_color;
2171
2172     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2173      * components for temporary data storage
2174      */
2175     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2176     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2177     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2178
2179     /* Construct the eye-ray vector from w coordinates */
2180     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2181     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2182     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2183
2184     /* Calculate reflection vector
2185      */
2186     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2187     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2188     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2189     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2190     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2191     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2192     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2193
2194     /* Sample the texture using the calculated coordinates */
2195     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2196     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2197     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2198     tex_mx->current_row = 0;
2199 }
2200
2201 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2202 {
2203     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2204     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2205     DWORD flags;
2206     DWORD reg = ins->dst[0].reg.idx;
2207     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2208     char dst_str[50];
2209     char src0_name[50];
2210     char src1_name[50];
2211     char dst_reg[50];
2212     BOOL is_color;
2213
2214     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2215     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2216     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2217     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2218     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2219
2220     /* Calculate reflection vector.
2221      *
2222      *                   dot(N, E)
2223      * dst_reg.xyz = 2 * --------- * N - E
2224      *                   dot(N, N)
2225      *
2226      * Which normalizes the normal vector
2227      */
2228     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2229     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2230     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2231     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2232     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2233     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2234
2235     /* Sample the texture using the calculated coordinates */
2236     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2237     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2238     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2239     tex_mx->current_row = 0;
2240 }
2241
2242 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2243 {
2244     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2245     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2246     char dst_name[50];
2247     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2248     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2249
2250     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2251      * which is essentially an input, is the destination register because it is the first
2252      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2253      * here(writemasks/swizzles are not valid on texdepth)
2254      */
2255     shader_arb_get_dst_param(ins, dst, dst_name);
2256
2257     /* According to the msdn, the source register(must be r5) is unusable after
2258      * the texdepth instruction, so we're free to modify it
2259      */
2260     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2261
2262     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2263      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2264      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2265      */
2266     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2267     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2268     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2269     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2270 }
2271
2272 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2273  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2274  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2275 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2276 {
2277     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2278     DWORD sampler_idx = ins->dst[0].reg.idx;
2279     char src0[50];
2280     char dst_str[50];
2281
2282     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2283     shader_addline(buffer, "MOV TB, 0.0;\n");
2284     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2285
2286     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2287     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2288 }
2289
2290 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2291  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2292 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2293 {
2294     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2295     char src0[50];
2296     char dst_str[50];
2297     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2298
2299     /* Handle output register */
2300     shader_arb_get_dst_param(ins, dst, dst_str);
2301     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2302     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2303 }
2304
2305 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2306  * Perform the 3rd row of a 3x3 matrix multiply */
2307 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2308 {
2309     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2310     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2311     char dst_str[50], dst_name[50];
2312     char src0[50];
2313     BOOL is_color;
2314
2315     shader_arb_get_dst_param(ins, dst, dst_str);
2316     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2317     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2318     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2319     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2320 }
2321
2322 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2323  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2324  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2325  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2326  */
2327 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2328 {
2329     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2330     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2331     char src0[50], dst_name[50];
2332     BOOL is_color;
2333     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2334     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2335
2336     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2337     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2338     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2339
2340     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2341      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2342      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2343      */
2344     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2345     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2346     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2347     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2348 }
2349
2350 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2351     Vertex/Pixel shaders to ARB_vertex_program codes */
2352 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2353 {
2354     int i;
2355     int nComponents = 0;
2356     struct wined3d_shader_dst_param tmp_dst = {{0}};
2357     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2358     struct wined3d_shader_instruction tmp_ins;
2359
2360     memset(&tmp_ins, 0, sizeof(tmp_ins));
2361
2362     /* Set constants for the temporary argument */
2363     tmp_ins.ctx = ins->ctx;
2364     tmp_ins.dst_count = 1;
2365     tmp_ins.dst = &tmp_dst;
2366     tmp_ins.src_count = 2;
2367     tmp_ins.src = tmp_src;
2368
2369     switch(ins->handler_idx)
2370     {
2371         case WINED3DSIH_M4x4:
2372             nComponents = 4;
2373             tmp_ins.handler_idx = WINED3DSIH_DP4;
2374             break;
2375         case WINED3DSIH_M4x3:
2376             nComponents = 3;
2377             tmp_ins.handler_idx = WINED3DSIH_DP4;
2378             break;
2379         case WINED3DSIH_M3x4:
2380             nComponents = 4;
2381             tmp_ins.handler_idx = WINED3DSIH_DP3;
2382             break;
2383         case WINED3DSIH_M3x3:
2384             nComponents = 3;
2385             tmp_ins.handler_idx = WINED3DSIH_DP3;
2386             break;
2387         case WINED3DSIH_M3x2:
2388             nComponents = 2;
2389             tmp_ins.handler_idx = WINED3DSIH_DP3;
2390             break;
2391         default:
2392             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2393             break;
2394     }
2395
2396     tmp_dst = ins->dst[0];
2397     tmp_src[0] = ins->src[0];
2398     tmp_src[1] = ins->src[1];
2399     for (i = 0; i < nComponents; i++) {
2400         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2401         shader_hw_map2gl(&tmp_ins);
2402         ++tmp_src[1].reg.idx;
2403     }
2404 }
2405
2406 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2407 {
2408     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2409     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2410     const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2411
2412     char dst[50];
2413     char src[50];
2414
2415     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2416     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2417     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2418     {
2419         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2420          * .w is used
2421          */
2422         strcat(src, ".w");
2423     }
2424
2425     /* TODO: If the destination is readable, and not the same as the source, the destination
2426      * can be used instead of TA
2427      */
2428     if (priv->target_version >= NV2)
2429     {
2430         shader_addline(buffer, "MOVC TA.x, %s;\n", src);
2431         shader_addline(buffer, "MOV TA.x (EQ.x), %s;\n", flt_eps);
2432         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2433     }
2434     else
2435     {
2436         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2437         shader_addline(buffer, "ABS TA.x, %s;\n", src);
2438         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2439         shader_addline(buffer, "MAD TA.x, TA.y, %s, %s;\n", flt_eps, src);
2440         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2441     }
2442 }
2443
2444 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2445 {
2446     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2447     const char *instruction;
2448
2449     char dst[50];
2450     char src[50];
2451
2452     switch(ins->handler_idx)
2453     {
2454         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2455         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2456         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2457         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2458         default: instruction = "";
2459             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2460             break;
2461     }
2462
2463     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2464     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2465     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2466     {
2467         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2468          * .w is used
2469          */
2470         strcat(src, ".w");
2471     }
2472
2473     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2474 }
2475
2476 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2477 {
2478     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2479     char dst_name[50];
2480     char src_name[50];
2481     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2482     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2483     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2484
2485     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2486     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2487
2488     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2489      * otherwise NRM or RSQ would return NaN */
2490     if(pshader && priv->target_version >= NV3)
2491     {
2492         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2493          *
2494          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2495          */
2496         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2497         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2498         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2499     }
2500     else if(priv->target_version >= NV2)
2501     {
2502         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2503         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2504         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2505                        src_name);
2506     }
2507     else
2508     {
2509         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2510
2511         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2512         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2513          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2514          */
2515         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2516         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2517
2518         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2519         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2520         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2521                     src_name);
2522     }
2523 }
2524
2525 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2526 {
2527     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2528     char dst_name[50];
2529     char src_name[3][50];
2530
2531     /* ARB_fragment_program has a convenient LRP instruction */
2532     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2533         shader_hw_map2gl(ins);
2534         return;
2535     }
2536
2537     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2538     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2539     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2540     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2541
2542     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2543     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2544                    dst_name, src_name[0], src_name[2]);
2545 }
2546
2547 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2548 {
2549     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2550      * must contain fixed constants. So we need a separate function to filter those constants and
2551      * can't use map2gl
2552      */
2553     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2554     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2555     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2556     char dst_name[50];
2557     char src_name0[50], src_name1[50], src_name2[50];
2558     BOOL is_color;
2559
2560     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2561     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2562         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2563         /* No modifiers are supported on SCS */
2564         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2565
2566         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2567         {
2568             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2569             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2570         }
2571     } else if(priv->target_version >= NV2) {
2572         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2573
2574         /* Sincos writemask must be .x, .y or .xy */
2575         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2576             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2577         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2578             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2579     } else {
2580         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2581          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2582          *
2583          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2584          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2585          *
2586          * The constants we get are:
2587          *
2588          *  +1   +1,     -1     -1     +1      +1      -1       -1
2589          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2590          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2591          *
2592          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2593          *
2594          * (x/2)^2 = x^2 / 4
2595          * (x/2)^3 = x^3 / 8
2596          * (x/2)^4 = x^4 / 16
2597          * (x/2)^5 = x^5 / 32
2598          * etc
2599          *
2600          * To get the final result:
2601          * sin(x) = 2 * sin(x/2) * cos(x/2)
2602          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2603          * (from sin(x+y) and cos(x+y) rules)
2604          *
2605          * As per MSDN, dst.z is undefined after the operation, and so is
2606          * dst.x and dst.y if they're masked out by the writemask. Ie
2607          * sincos dst.y, src1, c0, c1
2608          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2609          * vsa.exe also stops with an error if the dest register is the same register as the source
2610          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2611          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2612          */
2613         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2614         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2615         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2616
2617         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2618         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2619         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2620         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2621         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2622         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2623
2624         /* sin(x/2)
2625          *
2626          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2627          * properly merge that with MULs in the code above?
2628          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2629          * we can merge the sine and cosine MAD rows to calculate them together.
2630          */
2631         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2632         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2633         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2634         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2635
2636         /* cos(x/2) */
2637         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2638         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2639         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2640
2641         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2642             /* cos x */
2643             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2644             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2645         }
2646         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2647             /* sin x */
2648             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2649             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2650         }
2651     }
2652 }
2653
2654 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2655 {
2656     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2657     char dst_name[50];
2658     char src_name[50];
2659     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2660
2661     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2662     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2663
2664     /* SGN is only valid in vertex shaders */
2665     if(ctx->target_version >= NV2) {
2666         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2667         return;
2668     }
2669
2670     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2671      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2672      */
2673     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2674         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2675     } else {
2676         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2677          * Then use TA, and calculate the final result
2678          *
2679          * Not reading from TA? Store the first result in TA to avoid overwriting the
2680          * destination if src reg = dst reg
2681          */
2682         if(strstr(src_name, "TA"))
2683         {
2684             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2685             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2686             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2687         }
2688         else
2689         {
2690             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2691             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2692             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2693         }
2694     }
2695 }
2696
2697 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2698 {
2699     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2700     char src[50];
2701     char dst[50];
2702     char dst_name[50];
2703     BOOL is_color;
2704
2705     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2706     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2707     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2708
2709     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2710     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2711 }
2712
2713 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2714 {
2715     *need_abs = FALSE;
2716
2717     switch(mod)
2718     {
2719         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2720         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2721         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2722         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2723         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2724         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2725         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2726         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2727         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2728         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2729         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2730         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2731         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2732     }
2733     FIXME("Unknown modifier %u\n", mod);
2734     return mod;
2735 }
2736
2737 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2738 {
2739     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2740     char src0[50], src1[50], dst[50];
2741     struct wined3d_shader_src_param src0_copy = ins->src[0];
2742     BOOL need_abs = FALSE;
2743     const char *instr;
2744     BOOL arg2 = FALSE;
2745
2746     switch(ins->handler_idx)
2747     {
2748         case WINED3DSIH_LOG:  instr = "LG2"; break;
2749         case WINED3DSIH_LOGP: instr = "LOG"; break;
2750         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2751         default:
2752             ERR("Unexpected instruction %d\n", ins->handler_idx);
2753             return;
2754     }
2755
2756     /* LOG, LOGP and POW operate on the absolute value of the input */
2757     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2758
2759     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2760     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2761     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2762
2763     if(need_abs)
2764     {
2765         shader_addline(buffer, "ABS TA, %s;\n", src0);
2766         if(arg2)
2767         {
2768             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2769         }
2770         else
2771         {
2772             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2773         }
2774     }
2775     else if(arg2)
2776     {
2777         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2778     }
2779     else
2780     {
2781         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2782     }
2783 }
2784
2785 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2786 {
2787     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2788     char src_name[50];
2789     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2790
2791     /* src0 is aL */
2792     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2793
2794     if(vshader)
2795     {
2796         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2797         struct list *e = list_head(&priv->control_frames);
2798         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2799
2800         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2801         /* The constant loader makes sure to load -1 into iX.w */
2802         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2803         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2804         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2805     }
2806     else
2807     {
2808         shader_addline(buffer, "LOOP %s;\n", src_name);
2809     }
2810 }
2811
2812 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2813 {
2814     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2815     char src_name[50];
2816     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2817
2818     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2819
2820     /* The constant loader makes sure to load -1 into iX.w */
2821     if(vshader)
2822     {
2823         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2824         struct list *e = list_head(&priv->control_frames);
2825         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2826
2827         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2828
2829         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2830         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2831         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2832     }
2833     else
2834     {
2835         shader_addline(buffer, "REP %s;\n", src_name);
2836     }
2837 }
2838
2839 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2840 {
2841     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2842     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2843
2844     if(vshader)
2845     {
2846         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2847         struct list *e = list_head(&priv->control_frames);
2848         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2849
2850         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2851         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2852         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2853
2854         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2855     }
2856     else
2857     {
2858         shader_addline(buffer, "ENDLOOP;\n");
2859     }
2860 }
2861
2862 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2863 {
2864     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2865     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2866
2867     if(vshader)
2868     {
2869         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2870         struct list *e = list_head(&priv->control_frames);
2871         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2872
2873         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2874         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2875         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2876
2877         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2878     }
2879     else
2880     {
2881         shader_addline(buffer, "ENDREP;\n");
2882     }
2883 }
2884
2885 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2886 {
2887     struct control_frame *control_frame;
2888
2889     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2890     {
2891         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2892     }
2893     ERR("Could not find loop for break\n");
2894     return NULL;
2895 }
2896
2897 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2898 {
2899     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2900     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2901     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2902
2903     if(vshader)
2904     {
2905         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2906     }
2907     else
2908     {
2909         shader_addline(buffer, "BRK;\n");
2910     }
2911 }
2912
2913 static const char *get_compare(COMPARISON_TYPE flags)
2914 {
2915     switch (flags)
2916     {
2917         case COMPARISON_GT: return "GT";
2918         case COMPARISON_EQ: return "EQ";
2919         case COMPARISON_GE: return "GE";
2920         case COMPARISON_LT: return "LT";
2921         case COMPARISON_NE: return "NE";
2922         case COMPARISON_LE: return "LE";
2923         default:
2924             FIXME("Unrecognized comparison value: %u\n", flags);
2925             return "(\?\?)";
2926     }
2927 }
2928
2929 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2930 {
2931     switch (flags)
2932     {
2933         case COMPARISON_GT: return COMPARISON_LE;
2934         case COMPARISON_EQ: return COMPARISON_NE;
2935         case COMPARISON_GE: return COMPARISON_LT;
2936         case COMPARISON_LT: return COMPARISON_GE;
2937         case COMPARISON_NE: return COMPARISON_EQ;
2938         case COMPARISON_LE: return COMPARISON_GT;
2939         default:
2940             FIXME("Unrecognized comparison value: %u\n", flags);
2941             return -1;
2942     }
2943 }
2944
2945 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2946 {
2947     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2948     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2949     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2950     char src_name0[50];
2951     char src_name1[50];
2952     const char *comp = get_compare(ins->flags);
2953
2954     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2955     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2956
2957     if(vshader)
2958     {
2959         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2960          * away the subtraction result
2961          */
2962         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2963         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2964     }
2965     else
2966     {
2967         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2968         shader_addline(buffer, "BRK (%s.x);\n", comp);
2969     }
2970 }
2971
2972 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2973 {
2974     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2975     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2976     struct list *e = list_head(&priv->control_frames);
2977     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2978     const char *comp;
2979     char src_name0[50];
2980     char src_name1[50];
2981     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2982
2983     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2984     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2985
2986     if(vshader)
2987     {
2988         /* Invert the flag. We jump to the else label if the condition is NOT true */
2989         comp = get_compare(invert_compare(ins->flags));
2990         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2991         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
2992     }
2993     else
2994     {
2995         comp = get_compare(ins->flags);
2996         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2997         shader_addline(buffer, "IF %s.x;\n", comp);
2998     }
2999 }
3000
3001 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3002 {
3003     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3004     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3005     struct list *e = list_head(&priv->control_frames);
3006     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3007     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3008
3009     if(vshader)
3010     {
3011         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3012         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3013         control_frame->had_else = TRUE;
3014     }
3015     else
3016     {
3017         shader_addline(buffer, "ELSE;\n");
3018     }
3019 }
3020
3021 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3022 {
3023     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3024     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3025     struct list *e = list_head(&priv->control_frames);
3026     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3027     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3028
3029     if(vshader)
3030     {
3031         if(control_frame->had_else)
3032         {
3033             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3034         }
3035         else
3036         {
3037             shader_addline(buffer, "#No else branch. else is endif\n");
3038             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3039         }
3040     }
3041     else
3042     {
3043         shader_addline(buffer, "ENDIF;\n");
3044     }
3045 }
3046
3047 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3048 {
3049     DWORD sampler_idx = ins->src[1].reg.idx;
3050     char reg_dest[40];
3051     char reg_src[3][40];
3052     WORD flags = TEX_DERIV;
3053
3054     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3055     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3056     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3057     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3058
3059     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3060     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3061
3062     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3063 }
3064
3065 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3066 {
3067     DWORD sampler_idx = ins->src[1].reg.idx;
3068     char reg_dest[40];
3069     char reg_coord[40];
3070     WORD flags = TEX_LOD;
3071
3072     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3073     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3074
3075     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3076     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3077
3078     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3079 }
3080
3081 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3082 {
3083     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3084     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3085
3086     priv->in_main_func = FALSE;
3087     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3088      * subroutine, don't generate a label that will make GL complain
3089      */
3090     if(priv->target_version == ARB) return;
3091
3092     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3093 }
3094
3095 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3096         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
3097 {
3098     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3099     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3100     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3101     unsigned int i;
3102
3103     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3104      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3105      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3106      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3107      */
3108     if(args->super.fog_src == VS_FOG_Z) {
3109         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3110     } else if (!reg_maps->fog) {
3111         /* posFixup.x is always 1.0, so we can savely use it */
3112         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3113     }
3114
3115     /* Write the final position.
3116      *
3117      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3118      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3119      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3120      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3121      */
3122     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3123     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3124     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3125
3126     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3127     {
3128         if(args->super.clip_enabled)
3129         {
3130             for(i = 0; i < priv_ctx->vs_clipplanes; i++)
3131             {
3132                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3133             }
3134         }
3135     }
3136     else if(args->clip.boolclip.clip_texcoord)
3137     {
3138         unsigned int cur_clip = 0;
3139         char component[4] = {'x', 'y', 'z', 'w'};
3140         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3141
3142         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3143         {
3144             if(args->clip.boolclip.clipplane_mask & (1 << i))
3145             {
3146                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3147                                component[cur_clip++], i);
3148             }
3149         }
3150         switch(cur_clip)
3151         {
3152             case 0:
3153                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3154                 break;
3155             case 1:
3156                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3157                 break;
3158             case 2:
3159                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3160                 break;
3161             case 3:
3162                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3163                 break;
3164         }
3165         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3166                        args->clip.boolclip.clip_texcoord - 1);
3167     }
3168
3169     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3170      * and the glsl equivalent
3171      */
3172     if(need_helper_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
3173         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3174         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3175     } else {
3176         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3177         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3178     }
3179
3180     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3181
3182     priv_ctx->footer_written = TRUE;
3183 }
3184
3185 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3186 {
3187     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3188     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3189     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
3190     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3191
3192     if(priv->target_version == ARB) return;
3193
3194     if(vshader)
3195     {
3196         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
3197     }
3198
3199     shader_addline(buffer, "RET;\n");
3200 }
3201
3202 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3203 {
3204     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3205     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3206 }
3207
3208 /* GL locking is done by the caller */
3209 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3210 {
3211     GLuint program_id = 0;
3212     GLint pos;
3213
3214     const char *blt_vprogram =
3215         "!!ARBvp1.0\n"
3216         "PARAM c[1] = { { 1, 0.5 } };\n"
3217         "MOV result.position, vertex.position;\n"
3218         "MOV result.color, c[0].x;\n"
3219         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3220         "END\n";
3221
3222     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3223     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3224     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3225             strlen(blt_vprogram), blt_vprogram));
3226     checkGLcall("glProgramStringARB()");
3227
3228     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3229     if (pos != -1)
3230     {
3231         FIXME("Vertex program error at position %d: %s\n\n", pos,
3232             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3233         shader_arb_dump_program_source(blt_vprogram);
3234     }
3235     else
3236     {
3237         GLint native;
3238
3239         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3240         checkGLcall("glGetProgramivARB()");
3241         if (!native) WARN("Program exceeds native resource limits.\n");
3242     }
3243
3244     return program_id;
3245 }
3246
3247 /* GL locking is done by the caller */
3248 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3249         enum tex_types tex_type, BOOL masked)
3250 {
3251     GLuint program_id = 0;
3252     const char *fprogram;
3253     GLint pos;
3254
3255     static const char * const blt_fprograms_full[tex_type_count] =
3256     {
3257         /* tex_1d */
3258         NULL,
3259         /* tex_2d */
3260         "!!ARBfp1.0\n"
3261         "TEMP R0;\n"
3262         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3263         "MOV result.depth.z, R0.x;\n"
3264         "END\n",
3265         /* tex_3d */
3266         NULL,
3267         /* tex_cube */
3268         "!!ARBfp1.0\n"
3269         "TEMP R0;\n"
3270         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3271         "MOV result.depth.z, R0.x;\n"
3272         "END\n",
3273         /* tex_rect */
3274         "!!ARBfp1.0\n"
3275         "TEMP R0;\n"
3276         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3277         "MOV result.depth.z, R0.x;\n"
3278         "END\n",
3279     };
3280
3281     static const char * const blt_fprograms_masked[tex_type_count] =
3282     {
3283         /* tex_1d */
3284         NULL,
3285         /* tex_2d */
3286         "!!ARBfp1.0\n"
3287         "PARAM mask = program.local[0];\n"
3288         "TEMP R0;\n"
3289         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3290         "MUL R0.x, R0.x, R0.y;\n"
3291         "KIL -R0.x;\n"
3292         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3293         "MOV result.depth.z, R0.x;\n"
3294         "END\n",
3295         /* tex_3d */
3296         NULL,
3297         /* tex_cube */
3298         "!!ARBfp1.0\n"
3299         "PARAM mask = program.local[0];\n"
3300         "TEMP R0;\n"
3301         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3302         "MUL R0.x, R0.x, R0.y;\n"
3303         "KIL -R0.x;\n"
3304         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3305         "MOV result.depth.z, R0.x;\n"
3306         "END\n",
3307         /* tex_rect */
3308         "!!ARBfp1.0\n"
3309         "PARAM mask = program.local[0];\n"
3310         "TEMP R0;\n"
3311         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3312         "MUL R0.x, R0.x, R0.y;\n"
3313         "KIL -R0.x;\n"
3314         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3315         "MOV result.depth.z, R0.x;\n"
3316         "END\n",
3317     };
3318
3319     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3320     if (!fprogram)
3321     {
3322         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3323         tex_type = tex_2d;
3324         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3325     }
3326
3327     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3328     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3329     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3330     checkGLcall("glProgramStringARB()");
3331
3332     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3333     if (pos != -1)
3334     {
3335         FIXME("Fragment program error at position %d: %s\n\n", pos,
3336             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3337         shader_arb_dump_program_source(fprogram);
3338     }
3339     else
3340     {
3341         GLint native;
3342
3343         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3344         checkGLcall("glGetProgramivARB()");
3345         if (!native) WARN("Program exceeds native resource limits.\n");
3346     }
3347
3348     return program_id;
3349 }
3350
3351 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3352         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3353 {
3354     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3355
3356     if(condcode)
3357     {
3358         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3359         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3360         /* Calculate the > 0.0031308 case */
3361         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3362         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3363         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3364         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3365         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3366         /* Calculate the < case */
3367         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3368     }
3369     else
3370     {
3371         /* Calculate the > 0.0031308 case */
3372         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3373         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3374         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3375         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3376         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3377         /* Calculate the < case */
3378         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3379         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3380         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3381         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3382         /* Store the components > 0.0031308 in the destination */
3383         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3384         /* Add the components that are < 0.0031308 */
3385         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3386         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3387         * result.color writes(.rgb first, then .a), or handle overwriting already written
3388         * components. The assembler uses a temporary register in this case, which is usually
3389         * not allocated from one of our registers that were used earlier.
3390         */
3391     }
3392     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3393 }
3394
3395 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3396 {
3397     const local_constant *constant;
3398
3399     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3400     {
3401         if (constant->idx == idx)
3402         {
3403             return constant->value;
3404         }
3405     }
3406     return NULL;
3407 }
3408
3409 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3410                           struct shader_arb_ctx_priv *priv)
3411 {
3412     static const char * const texcoords[8] =
3413     {
3414         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3415         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3416     };
3417     unsigned int i;
3418     const struct wined3d_shader_signature_element *sig = This->baseShader.input_signature;
3419     const char *semantic_name;
3420     DWORD semantic_idx;
3421
3422     switch(args->super.vp_mode)
3423     {
3424         case pretransformed:
3425         case fixedfunction:
3426             /* The pixelshader has to collect the varyings on its own. In any case properly load
3427              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3428              * other attribs to 0.0.
3429              *
3430              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3431              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3432              * load the texcoord attrib pointers to match the pixel shader signature
3433              */
3434             for(i = 0; i < MAX_REG_INPUT; i++)
3435             {
3436                 semantic_name = sig[i].semantic_name;
3437                 semantic_idx = sig[i].semantic_idx;
3438                 if (!semantic_name) continue;
3439
3440                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3441                 {
3442                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3443                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3444                     else priv->ps_input[i] = "0.0";
3445                 }
3446                 else if(args->super.vp_mode == fixedfunction)
3447                 {
3448                     priv->ps_input[i] = "0.0";
3449                 }
3450                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3451                 {
3452                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3453                     else priv->ps_input[i] = "0.0";
3454                 }
3455                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3456                 {
3457                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3458                     else priv->ps_input[i] = "0.0";
3459                 }
3460                 else
3461                 {
3462                     priv->ps_input[i] = "0.0";
3463                 }
3464
3465                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3466             }
3467             break;
3468
3469         case vertexshader:
3470             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3471              * fragment.color
3472              */
3473             for(i = 0; i < 8; i++)
3474             {
3475                 priv->ps_input[i] = texcoords[i];
3476             }
3477             priv->ps_input[8] = "fragment.color.primary";
3478             priv->ps_input[9] = "fragment.color.secondary";
3479             break;
3480     }
3481 }
3482
3483 /* GL locking is done by the caller */
3484 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3485         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3486 {
3487     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3488     CONST DWORD *function = This->baseShader.function;
3489     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3490     const local_constant *lconst;
3491     GLuint retval;
3492     char fragcolor[16];
3493     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3494     struct shader_arb_ctx_priv priv_ctx;
3495     BOOL dcl_td = FALSE;
3496     BOOL want_nv_prog = FALSE;
3497     struct arb_pshader_private *shader_priv = This->baseShader.backend_data;
3498     GLint errPos;
3499     DWORD map;
3500
3501     char srgbtmp[4][4];
3502     unsigned int i, found = 0;
3503
3504     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3505     {
3506         if (!(map & 1)
3507                 || (This->color0_mov && i == This->color0_reg)
3508                 || (reg_maps->shader_version.major < 2 && !i))
3509             continue;
3510
3511         sprintf(srgbtmp[found], "R%u", i);
3512         ++found;
3513         if (found == 4) break;
3514     }
3515
3516     switch(found) {
3517         case 0:
3518             sprintf(srgbtmp[0], "TA");
3519             sprintf(srgbtmp[1], "TB");
3520             sprintf(srgbtmp[2], "TC");
3521             sprintf(srgbtmp[3], "TD");
3522             dcl_td = TRUE;
3523             break;
3524         case 1:
3525             sprintf(srgbtmp[1], "TA");
3526             sprintf(srgbtmp[2], "TB");
3527             sprintf(srgbtmp[3], "TC");
3528             break;
3529         case 2:
3530             sprintf(srgbtmp[2], "TA");
3531             sprintf(srgbtmp[3], "TB");
3532             break;
3533         case 3:
3534             sprintf(srgbtmp[3], "TA");
3535             break;
3536         case 4:
3537             break;
3538     }
3539
3540     /*  Create the hw ARB shader */
3541     memset(&priv_ctx, 0, sizeof(priv_ctx));
3542     priv_ctx.cur_ps_args = args;
3543     priv_ctx.compiled_fprog = compiled;
3544     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3545     init_ps_input(This, args, &priv_ctx);
3546     list_init(&priv_ctx.control_frames);
3547
3548     /* Avoid enabling NV_fragment_program* if we do not need it.
3549      *
3550      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3551      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3552      * is faster than what we gain from using higher native instructions. There are some things though
3553      * that cannot be emulated. In that case enable the extensions.
3554      * If the extension is enabled, instruction handlers that support both ways will use it.
3555      *
3556      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3557      * So enable the best we can get.
3558      */
3559     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3560        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3561     {
3562         want_nv_prog = TRUE;
3563     }
3564
3565     shader_addline(buffer, "!!ARBfp1.0\n");
3566     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3567     {
3568         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3569         priv_ctx.target_version = NV3;
3570     }
3571     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3572     {
3573         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3574         priv_ctx.target_version = NV2;
3575     } else {
3576         if(want_nv_prog)
3577         {
3578             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3579              * limits properly
3580              */
3581             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3582             ERR("Try GLSL\n");
3583         }
3584         priv_ctx.target_version = ARB;
3585     }
3586
3587     if(This->baseShader.reg_maps.highest_render_target > 0)
3588     {
3589         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3590     }
3591
3592     if (reg_maps->shader_version.major < 3)
3593     {
3594         switch(args->super.fog) {
3595             case FOG_OFF:
3596                 break;
3597             case FOG_LINEAR:
3598                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3599                 break;
3600             case FOG_EXP:
3601                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3602                 break;
3603             case FOG_EXP2:
3604                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3605                 break;
3606         }
3607     }
3608
3609     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3610      * unused temps away(but occupies them for the whole shader if they're used once). Always
3611      * declaring them avoids tricky bookkeeping work
3612      */
3613     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3614     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3615     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3616     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3617     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3618     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3619     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3620
3621     if (reg_maps->shader_version.major < 2)
3622     {
3623         strcpy(fragcolor, "R0");
3624     } else {
3625         if(args->super.srgb_correction) {
3626             if(This->color0_mov) {
3627                 sprintf(fragcolor, "R%u", This->color0_reg);
3628             } else {
3629                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3630                 strcpy(fragcolor, "TMP_COLOR");
3631             }
3632         } else {
3633             strcpy(fragcolor, "result.color");
3634         }
3635     }
3636
3637     if(args->super.srgb_correction) {
3638         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3639                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3640         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3641                        srgb_sub_high, 0.0, 0.0, 0.0);
3642     }
3643
3644     /* Base Declarations */
3645     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
3646             reg_maps, buffer, gl_info, lconst_map, NULL, &priv_ctx);
3647
3648     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3649     {
3650         if (!(map & 1)) continue;
3651
3652         cur = compiled->numbumpenvmatconsts;
3653         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3654         compiled->bumpenvmatconst[cur].texunit = i;
3655         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3656         compiled->luminanceconst[cur].texunit = i;
3657
3658         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3659          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3660          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3661          * textures due to conditional NP2 restrictions)
3662          *
3663          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3664          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3665          * their location is shader dependent anyway and they cannot be loaded globally.
3666          */
3667         compiled->bumpenvmatconst[cur].const_num = next_local++;
3668         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3669                        i, compiled->bumpenvmatconst[cur].const_num);
3670         compiled->numbumpenvmatconsts = cur + 1;
3671
3672         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3673
3674         compiled->luminanceconst[cur].const_num = next_local++;
3675         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3676                        i, compiled->luminanceconst[cur].const_num);
3677     }
3678
3679     for(i = 0; i < MAX_CONST_I; i++)
3680     {
3681         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3682         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3683         {
3684             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3685
3686             if(control_values)
3687             {
3688                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3689                                 control_values[0], control_values[1], control_values[2]);
3690             }
3691             else
3692             {
3693                 compiled->int_consts[i] = next_local;
3694                 compiled->num_int_consts++;
3695                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3696             }
3697         }
3698     }
3699
3700     if(reg_maps->vpos || reg_maps->usesdsy)
3701     {
3702         compiled->ycorrection = next_local;
3703         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3704
3705         if(reg_maps->vpos)
3706         {
3707             shader_addline(buffer, "TEMP vpos;\n");
3708             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3709              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3710              * ycorrection.z: 1.0
3711              * ycorrection.w: 0.0
3712              */
3713             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3714             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3715         }
3716     }
3717     else
3718     {
3719         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3720     }
3721
3722     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3723      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3724      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3725      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3726      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3727      * shader compilation errors and the subsequent errors when drawing with this shader. */
3728     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3729
3730         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3731         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3732         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3733
3734         fixup->offset = next_local;
3735         fixup->super.active = 0;
3736
3737         cur = 0;
3738         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3739             if (!(map & (1 << i))) continue;
3740
3741             if (fixup->offset + (cur >> 1) < max_lconsts) {
3742                 fixup->super.active |= (1 << i);
3743                 fixup->super.idx[i] = cur++;
3744             } else {
3745                 FIXME("No free constant found to load NP2 fixup data into shader. "
3746                       "Sampling from this texture will probably look wrong.\n");
3747                 break;
3748             }
3749         }
3750
3751         fixup->super.num_consts = (cur + 1) >> 1;
3752         if (fixup->super.num_consts) {
3753             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3754                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3755         }
3756
3757         next_local += fixup->super.num_consts;
3758     }
3759
3760     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3761     {
3762         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3763     }
3764
3765     /* Base Shader Body */
3766     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3767
3768     if(args->super.srgb_correction) {
3769         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3770                                   priv_ctx.target_version >= NV2);
3771     }
3772
3773     if(strcmp(fragcolor, "result.color")) {
3774         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3775     }
3776     shader_addline(buffer, "END\n");
3777
3778     /* TODO: change to resource.glObjectHandle or something like that */
3779     GL_EXTCALL(glGenProgramsARB(1, &retval));
3780
3781     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3782     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3783
3784     TRACE("Created hw pixel shader, prg=%d\n", retval);
3785     /* Create the program and check for errors */
3786     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3787                buffer->bsize, buffer->buffer));
3788     checkGLcall("glProgramStringARB()");
3789
3790     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3791     if (errPos != -1)
3792     {
3793         FIXME("HW PixelShader Error at position %d: %s\n\n",
3794               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3795         shader_arb_dump_program_source(buffer->buffer);
3796         retval = 0;
3797     }
3798     else
3799     {
3800         GLint native;
3801
3802         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3803         checkGLcall("glGetProgramivARB()");
3804         if (!native) WARN("Program exceeds native resource limits.\n");
3805     }
3806
3807     /* Load immediate constants */
3808     if(lconst_map) {
3809         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3810             const float *value = (const float *)lconst->value;
3811             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3812             checkGLcall("glProgramLocalParameter4fvARB");
3813         }
3814         HeapFree(GetProcessHeap(), 0, lconst_map);
3815     }
3816
3817     return retval;
3818 }
3819
3820 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3821 {
3822     unsigned int i;
3823     int ret;
3824
3825     for(i = 0; i < MAX_REG_INPUT; i++)
3826     {
3827         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3828         {
3829             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3830             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3831             continue;
3832         }
3833
3834         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3835         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3836         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3837         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3838         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3839         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3840     }
3841     return 0;
3842 }
3843
3844 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3845 {
3846     struct wined3d_shader_signature_element *new;
3847     int i;
3848     char *name;
3849
3850     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3851     for(i = 0; i < MAX_REG_INPUT; i++)
3852     {
3853         if (!sig[i].semantic_name) continue;
3854
3855         new[i] = sig[i];
3856         /* Clone the semantic string */
3857         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3858         strcpy(name, sig[i].semantic_name);
3859         new[i].semantic_name = name;
3860     }
3861     return new;
3862 }
3863
3864 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3865 {
3866     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3867     struct ps_signature *found_sig;
3868
3869     if (entry)
3870     {
3871         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3872         TRACE("Found existing signature %u\n", found_sig->idx);
3873         return found_sig->idx;
3874     }
3875     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3876     found_sig->sig = clone_sig(sig);
3877     found_sig->idx = priv->ps_sig_number++;
3878     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3879     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3880     {
3881         ERR("Failed to insert program entry.\n");
3882     }
3883     return found_sig->idx;
3884 }
3885
3886 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3887                                   struct arb_vs_compiled_shader *compiled)
3888 {
3889     unsigned int i, j;
3890     static const char * const texcoords[8] =
3891     {
3892         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3893         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3894     };
3895     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3896     IWineD3DBaseShaderClass *baseshader = &shader->baseShader;
3897     const struct wined3d_shader_signature_element *sig;
3898     const char *semantic_name;
3899     DWORD semantic_idx, reg_idx;
3900
3901     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3902      * and varying 9 to result.color.secondary
3903      */
3904     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3905     {
3906         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3907         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3908         "result.color.primary", "result.color.secondary"
3909     };
3910
3911     if(sig_num == ~0)
3912     {
3913         TRACE("Pixel shader uses builtin varyings\n");
3914         /* Map builtins to builtins */
3915         for(i = 0; i < 8; i++)
3916         {
3917             priv_ctx->texcrd_output[i] = texcoords[i];
3918         }
3919         priv_ctx->color_output[0] = "result.color.primary";
3920         priv_ctx->color_output[1] = "result.color.secondary";
3921         priv_ctx->fog_output = "result.fogcoord";
3922
3923         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3924         for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
3925         {
3926             semantic_name = baseshader->output_signature[i].semantic_name;
3927             if (!semantic_name) continue;
3928
3929             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3930             {
3931                 TRACE("o%u is TMP_OUT\n", i);
3932                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3933                 else priv_ctx->vs_output[i] = "TA";
3934             }
3935             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3936             {
3937                 TRACE("o%u is result.pointsize\n", i);
3938                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3939                 else priv_ctx->vs_output[i] = "TA";
3940             }
3941             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3942             {
3943                 TRACE("o%u is result.color.?, idx %u\n", i, baseshader->output_signature[i].semantic_idx);
3944                 if (!baseshader->output_signature[i].semantic_idx)
3945                     priv_ctx->vs_output[i] = "result.color.primary";
3946                 else if (baseshader->output_signature[i].semantic_idx == 1)
3947                     priv_ctx->vs_output[i] = "result.color.secondary";
3948                 else priv_ctx->vs_output[i] = "TA";
3949             }
3950             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3951             {
3952                 TRACE("o%u is %s\n", i, texcoords[baseshader->output_signature[i].semantic_idx]);
3953                 if (baseshader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3954                 else priv_ctx->vs_output[i] = texcoords[baseshader->output_signature[i].semantic_idx];
3955             }
3956             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3957             {
3958                 TRACE("o%u is result.fogcoord\n", i);
3959                 if (baseshader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3960                 else priv_ctx->vs_output[i] = "result.fogcoord";
3961             }
3962             else
3963             {
3964                 priv_ctx->vs_output[i] = "TA";
3965             }
3966         }
3967         return;
3968     }
3969
3970     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3971      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3972      */
3973     sig = device->stateBlock->state.pixel_shader->baseShader.input_signature;
3974     TRACE("Pixel shader uses declared varyings\n");
3975
3976     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3977     for(i = 0; i < 8; i++)
3978     {
3979         priv_ctx->texcrd_output[i] = "TA";
3980     }
3981     priv_ctx->color_output[0] = "TA";
3982     priv_ctx->color_output[1] = "TA";
3983     priv_ctx->fog_output = "TA";
3984
3985     for(i = 0; i < MAX_REG_INPUT; i++)
3986     {
3987         semantic_name = sig[i].semantic_name;
3988         semantic_idx = sig[i].semantic_idx;
3989         reg_idx = sig[i].register_idx;
3990         if (!semantic_name) continue;
3991
3992         /* If a declared input register is not written by builtin arguments, don't write to it.
3993          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3994          *
3995          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
3996          * to TMP_OUT in any case
3997          */
3998         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3999         {
4000             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4001         }
4002         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4003         {
4004             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4005         }
4006         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4007         {
4008             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4009         }
4010         else
4011         {
4012             continue;
4013         }
4014
4015         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4016                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4017         {
4018             compiled->need_color_unclamp = TRUE;
4019         }
4020     }
4021
4022     /* Map declared to declared */
4023     for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
4024     {
4025         /* Write unread output to TA to throw them away */
4026         priv_ctx->vs_output[i] = "TA";
4027         semantic_name = baseshader->output_signature[i].semantic_name;
4028         if (!semantic_name) continue;
4029
4030         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4031                 && !baseshader->output_signature[i].semantic_idx)
4032         {
4033             priv_ctx->vs_output[i] = "TMP_OUT";
4034             continue;
4035         }
4036         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4037                 && !baseshader->output_signature[i].semantic_idx)
4038         {
4039             priv_ctx->vs_output[i] = "result.pointsize";
4040             continue;
4041         }
4042
4043         for(j = 0; j < MAX_REG_INPUT; j++)
4044         {
4045             if (!sig[j].semantic_name) continue;
4046
4047             if (!strcmp(sig[j].semantic_name, semantic_name)
4048                     && sig[j].semantic_idx == baseshader->output_signature[i].semantic_idx)
4049             {
4050                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4051
4052                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4053                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4054                 {
4055                     compiled->need_color_unclamp = TRUE;
4056                 }
4057             }
4058         }
4059     }
4060 }
4061
4062 /* GL locking is done by the caller */
4063 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
4064         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4065 {
4066     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4067     CONST DWORD *function = This->baseShader.function;
4068     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4069     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4070     const local_constant *lconst;
4071     GLuint ret;
4072     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
4073     struct shader_arb_ctx_priv priv_ctx;
4074     unsigned int i;
4075     GLint errPos;
4076
4077     memset(&priv_ctx, 0, sizeof(priv_ctx));
4078     priv_ctx.cur_vs_args = args;
4079     list_init(&priv_ctx.control_frames);
4080     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
4081
4082     /*  Create the hw ARB shader */
4083     shader_addline(buffer, "!!ARBvp1.0\n");
4084
4085     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4086      * mesurable performance penalty, and we can always make use of it for clipplanes.
4087      */
4088     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4089     {
4090         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4091         priv_ctx.target_version = NV3;
4092         shader_addline(buffer, "ADDRESS aL;\n");
4093     }
4094     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4095     {
4096         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4097         priv_ctx.target_version = NV2;
4098         shader_addline(buffer, "ADDRESS aL;\n");
4099     } else {
4100         priv_ctx.target_version = ARB;
4101     }
4102
4103     shader_addline(buffer, "TEMP TMP_OUT;\n");
4104     if(need_helper_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
4105         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4106     }
4107     if(need_rel_addr_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
4108         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", This->rel_offset);
4109         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4110     }
4111
4112     shader_addline(buffer, "TEMP TA;\n");
4113
4114     /* Base Declarations */
4115     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
4116             reg_maps, buffer, gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4117
4118     for(i = 0; i < MAX_CONST_I; i++)
4119     {
4120         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4121         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4122         {
4123             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
4124
4125             if(control_values)
4126             {
4127                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4128                                 control_values[0], control_values[1], control_values[2]);
4129             }
4130             else
4131             {
4132                 compiled->int_consts[i] = next_local;
4133                 compiled->num_int_consts++;
4134                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4135             }
4136         }
4137     }
4138
4139     /* We need a constant to fixup the final position */
4140     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4141     compiled->pos_fixup = next_local++;
4142
4143     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4144      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4145      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4146      * a replacement shader depend on the texcoord.w being set properly.
4147      *
4148      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4149      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4150      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4151      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4152      * this can eat a number of instructions, so skip it unless this cap is set as well
4153      */
4154     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4155     {
4156         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4157         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4158
4159         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4160         {
4161             int i;
4162             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4163             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4164             {
4165                 if (This->baseShader.reg_maps.texcoord_mask[i]
4166                         && This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4167                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4168             }
4169         }
4170     }
4171
4172     /* The shader starts with the main function */
4173     priv_ctx.in_main_func = TRUE;
4174     /* Base Shader Body */
4175     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4176
4177     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
4178
4179     shader_addline(buffer, "END\n");
4180
4181     /* TODO: change to resource.glObjectHandle or something like that */
4182     GL_EXTCALL(glGenProgramsARB(1, &ret));
4183
4184     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4185     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4186
4187     TRACE("Created hw vertex shader, prg=%d\n", ret);
4188     /* Create the program and check for errors */
4189     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4190                buffer->bsize, buffer->buffer));
4191     checkGLcall("glProgramStringARB()");
4192
4193     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4194     if (errPos != -1)
4195     {
4196         FIXME("HW VertexShader Error at position %d: %s\n\n",
4197               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4198         shader_arb_dump_program_source(buffer->buffer);
4199         ret = -1;
4200     }
4201     else
4202     {
4203         GLint native;
4204
4205         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4206         checkGLcall("glGetProgramivARB()");
4207         if (!native) WARN("Program exceeds native resource limits.\n");
4208
4209         /* Load immediate constants */
4210         if(lconst_map) {
4211             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
4212                 const float *value = (const float *)lconst->value;
4213                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4214             }
4215         }
4216     }
4217     HeapFree(GetProcessHeap(), 0, lconst_map);
4218
4219     return ret;
4220 }
4221
4222 /* GL locking is done by the caller */
4223 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
4224 {
4225     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)shader->baseShader.device;
4226     UINT i;
4227     DWORD new_size;
4228     struct arb_ps_compiled_shader *new_array;
4229     struct wined3d_shader_buffer buffer;
4230     struct arb_pshader_private *shader_data;
4231     GLuint ret;
4232
4233     if (!shader->baseShader.backend_data)
4234     {
4235         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4236         struct shader_arb_priv *priv = device->shader_priv;
4237
4238         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4239         shader_data = shader->baseShader.backend_data;
4240         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
4241
4242         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
4243         else shader_data->input_signature_idx = find_input_signature(priv, shader->baseShader.input_signature);
4244
4245         shader_data->has_signature_idx = TRUE;
4246         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4247
4248         if (!device->vs_clipping)
4249             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
4250                     gl_info->limits.texture_stages - 1);
4251         else
4252             shader_data->clipplane_emulation = ~0U;
4253     }
4254     shader_data = shader->baseShader.backend_data;
4255
4256     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4257      * so a linear search is more performant than a hashmap or a binary search
4258      * (cache coherency etc)
4259      */
4260     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4261     {
4262         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4263             return &shader_data->gl_shaders[i];
4264     }
4265
4266     TRACE("No matching GL shader found, compiling a new shader\n");
4267     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4268         if (shader_data->num_gl_shaders)
4269         {
4270             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4271             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4272                                     new_size * sizeof(*shader_data->gl_shaders));
4273         } else {
4274             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4275             new_size = 1;
4276         }
4277
4278         if(!new_array) {
4279             ERR("Out of memory\n");
4280             return 0;
4281         }
4282         shader_data->gl_shaders = new_array;
4283         shader_data->shader_array_size = new_size;
4284     }
4285
4286     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4287
4288     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4289             (IWineD3DBaseTexture **)device->stateBlock->state.textures);
4290
4291     if (!shader_buffer_init(&buffer))
4292     {
4293         ERR("Failed to initialize shader buffer.\n");
4294         return 0;
4295     }
4296
4297     ret = shader_arb_generate_pshader(shader, &buffer, args,
4298                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4299     shader_buffer_free(&buffer);
4300     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4301
4302     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4303 }
4304
4305 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4306                                  const DWORD use_map, BOOL skip_int) {
4307     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4308     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4309     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4310     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4311     if(stored->ps_signature != new->ps_signature) return FALSE;
4312     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4313     if(skip_int) return TRUE;
4314
4315     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4316 }
4317
4318 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4319 {
4320     UINT i;
4321     DWORD new_size;
4322     struct arb_vs_compiled_shader *new_array;
4323     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4324     struct wined3d_shader_buffer buffer;
4325     struct arb_vshader_private *shader_data;
4326     GLuint ret;
4327     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4328
4329     if (!shader->baseShader.backend_data)
4330     {
4331         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4332     }
4333     shader_data = shader->baseShader.backend_data;
4334
4335     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4336      * so a linear search is more performant than a hashmap or a binary search
4337      * (cache coherency etc)
4338      */
4339     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4340         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4341                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4342         {
4343             return &shader_data->gl_shaders[i];
4344         }
4345     }
4346
4347     TRACE("No matching GL shader found, compiling a new shader\n");
4348
4349     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4350         if (shader_data->num_gl_shaders)
4351         {
4352             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4353             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4354                                     new_size * sizeof(*shader_data->gl_shaders));
4355         } else {
4356             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4357             new_size = 1;
4358         }
4359
4360         if(!new_array) {
4361             ERR("Out of memory\n");
4362             return 0;
4363         }
4364         shader_data->gl_shaders = new_array;
4365         shader_data->shader_array_size = new_size;
4366     }
4367
4368     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4369
4370     if (!shader_buffer_init(&buffer))
4371     {
4372         ERR("Failed to initialize shader buffer.\n");
4373         return 0;
4374     }
4375
4376     ret = shader_arb_generate_vshader(shader, &buffer, args,
4377             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4378     shader_buffer_free(&buffer);
4379     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4380
4381     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4382 }
4383
4384 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4385         IWineD3DPixelShaderImpl *shader, struct arb_ps_compile_args *args)
4386 {
4387     int i;
4388     WORD int_skip;
4389     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4390
4391     find_ps_compile_args(state, shader, &args->super);
4392
4393     /* This forces all local boolean constants to 1 to make them stateblock independent */
4394     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4395
4396     for(i = 0; i < MAX_CONST_B; i++)
4397     {
4398         if (state->ps_consts_b[i])
4399             args->bools |= ( 1 << i);
4400     }
4401
4402     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4403      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4404      * duplicate the shader than have a no-op KIL instruction in every shader
4405      */
4406     if ((!((IWineD3DDeviceImpl *)shader->baseShader.device)->vs_clipping) && use_vs(state)
4407             && state->render_states[WINED3DRS_CLIPPING]
4408             && state->render_states[WINED3DRS_CLIPPLANEENABLE])
4409         args->clip = 1;
4410     else
4411         args->clip = 0;
4412
4413     /* Skip if unused or local, or supported natively */
4414     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4415     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4416     {
4417         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4418         return;
4419     }
4420
4421     for(i = 0; i < MAX_CONST_I; i++)
4422     {
4423         if(int_skip & (1 << i))
4424         {
4425             args->loop_ctrl[i][0] = 0;
4426             args->loop_ctrl[i][1] = 0;
4427             args->loop_ctrl[i][2] = 0;
4428         }
4429         else
4430         {
4431             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4432             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4433             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4434         }
4435     }
4436 }
4437
4438 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4439         IWineD3DVertexShaderImpl *shader, struct arb_vs_compile_args *args)
4440 {
4441     int i;
4442     WORD int_skip;
4443     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4444     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4445
4446     find_vs_compile_args(state, shader, &args->super);
4447
4448     args->clip.boolclip_compare = 0;
4449     if (use_ps(state))
4450     {
4451         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
4452         struct arb_pshader_private *shader_priv = ps->baseShader.backend_data;
4453         args->ps_signature = shader_priv->input_signature_idx;
4454
4455         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4456     }
4457     else
4458     {
4459         args->ps_signature = ~0;
4460         if(!dev->vs_clipping)
4461         {
4462             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4463         }
4464         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4465     }
4466
4467     if (args->clip.boolclip.clip_texcoord)
4468     {
4469         if (state->render_states[WINED3DRS_CLIPPING])
4470             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3DRS_CLIPPLANEENABLE];
4471         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4472     }
4473
4474     /* This forces all local boolean constants to 1 to make them stateblock independent */
4475     args->clip.boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4476     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4477     for(i = 0; i < MAX_CONST_B; i++)
4478     {
4479         if (state->vs_consts_b[i])
4480             args->clip.boolclip.bools |= ( 1 << i);
4481     }
4482
4483     args->vertex.samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4484     args->vertex.samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4485     args->vertex.samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4486     args->vertex.samplers[3] = 0;
4487
4488     /* Skip if unused or local */
4489     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4490     /* This is about flow control, not clipping. */
4491     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4492     {
4493         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4494         return;
4495     }
4496
4497     for(i = 0; i < MAX_CONST_I; i++)
4498     {
4499         if(int_skip & (1 << i))
4500         {
4501             args->loop_ctrl[i][0] = 0;
4502             args->loop_ctrl[i][1] = 0;
4503             args->loop_ctrl[i][2] = 0;
4504         }
4505         else
4506         {
4507             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4508             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4509             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4510         }
4511     }
4512 }
4513
4514 /* GL locking is done by the caller */
4515 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4516 {
4517     IWineD3DDeviceImpl *This = context->swapchain->device;
4518     struct shader_arb_priv *priv = This->shader_priv;
4519     const struct wined3d_gl_info *gl_info = context->gl_info;
4520     const struct wined3d_state *state = &This->stateBlock->state;
4521     int i;
4522
4523     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4524     if (usePS)
4525     {
4526         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
4527         struct arb_ps_compile_args compile_args;
4528         struct arb_ps_compiled_shader *compiled;
4529
4530         TRACE("Using pixel shader %p.\n", ps);
4531         find_arb_ps_compile_args(state, ps, &compile_args);
4532         compiled = find_arb_pshader(ps, &compile_args);
4533         priv->current_fprogram_id = compiled->prgId;
4534         priv->compiled_fprog = compiled;
4535
4536         /* Bind the fragment program */
4537         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4538         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4539
4540         if(!priv->use_arbfp_fixed_func) {
4541             /* Enable OpenGL fragment programs */
4542             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4543             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4544         }
4545         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4546
4547         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4548          * a 1.x and newer shader, reload the first 8 constants
4549          */
4550         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts)
4551         {
4552             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts;
4553             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4554             for(i = 0; i < 8; i++)
4555             {
4556                 context->pshader_const_dirty[i] = 1;
4557             }
4558             /* Also takes care of loading local constants */
4559             shader_arb_load_constants(context, TRUE, FALSE);
4560         }
4561         else
4562         {
4563             float rt_height = This->render_targets[0]->currentDesc.Height;
4564             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4565         }
4566
4567         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4568         if (compiled->np2fixup_info.super.active)
4569             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4570     }
4571     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4572     {
4573         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4574         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4575         * replacement shader
4576         */
4577         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4578         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4579         priv->current_fprogram_id = 0;
4580     }
4581
4582     if (useVS)
4583     {
4584         IWineD3DVertexShaderImpl *vs = state->vertex_shader;
4585         struct arb_vs_compile_args compile_args;
4586         struct arb_vs_compiled_shader *compiled;
4587
4588         TRACE("Using vertex shader %p\n", vs);
4589         find_arb_vs_compile_args(state, vs, &compile_args);
4590         compiled = find_arb_vshader(vs, &compile_args);
4591         priv->current_vprogram_id = compiled->prgId;
4592         priv->compiled_vprog = compiled;
4593
4594         /* Bind the vertex program */
4595         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4596         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4597
4598         /* Enable OpenGL vertex programs */
4599         glEnable(GL_VERTEX_PROGRAM_ARB);
4600         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4601         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4602         shader_arb_vs_local_constants(compiled, context, state);
4603
4604         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4605             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4606
4607             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4608             {
4609                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4610                 checkGLcall("glClampColorARB");
4611             } else {
4612                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4613             }
4614         }
4615     }
4616     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4617     {
4618         priv->current_vprogram_id = 0;
4619         glDisable(GL_VERTEX_PROGRAM_ARB);
4620         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4621     }
4622 }
4623
4624 /* GL locking is done by the caller */
4625 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4626         enum tex_types tex_type, const SIZE *ds_mask_size)
4627 {
4628     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4629     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4630     struct shader_arb_priv *priv = shader_priv;
4631     GLuint *blt_fprogram;
4632
4633     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4634     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4635     glEnable(GL_VERTEX_PROGRAM_ARB);
4636
4637     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4638     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4639     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4640     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4641     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4642 }
4643
4644 /* GL locking is done by the caller */
4645 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4646 {
4647     struct shader_arb_priv *priv = shader_priv;
4648
4649     if (priv->current_vprogram_id) {
4650         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4651         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4652
4653         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4654     }
4655     else
4656     {
4657         glDisable(GL_VERTEX_PROGRAM_ARB);
4658         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4659     }
4660
4661     if (priv->current_fprogram_id) {
4662         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4663         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4664
4665         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4666     }
4667     else if(!priv->use_arbfp_fixed_func)
4668     {
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(IWineD3DDeviceImpl *device)
4748 {
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     device->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(IWineD3DDeviceImpl *device)
4774 {
4775     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4776     struct shader_arb_priv *priv = device->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, device->shader_priv);
4798 }
4799
4800 static BOOL shader_arb_dirty_const(void)
4801 {
4802     return TRUE;
4803 }
4804
4805 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4806 {
4807     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4808     {
4809         DWORD vs_consts;
4810
4811         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4812          * for vertex programs. If the native limit is less than that it's
4813          * not very useful, and e.g. Mesa swrast returns 0, probably to
4814          * indicate it's a software implementation. */
4815         if (gl_info->limits.arb_vs_native_constants < 96)
4816             vs_consts = gl_info->limits.arb_vs_float_constants;
4817         else
4818             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4819
4820         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4821         {
4822             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4823             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4824         }
4825         else if (vs_consts >= 256)
4826         {
4827             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4828             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4829             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4830         }
4831         else
4832         {
4833             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4834             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4835         }
4836         pCaps->MaxVertexShaderConst = vs_consts;
4837     }
4838     else
4839     {
4840         pCaps->VertexShaderVersion = 0;
4841         pCaps->MaxVertexShaderConst = 0;
4842     }
4843
4844     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4845     {
4846         DWORD ps_consts;
4847
4848         /* Similar as above for vertex programs, but the minimum for fragment
4849          * programs is 24. */
4850         if (gl_info->limits.arb_ps_native_constants < 24)
4851             ps_consts = gl_info->limits.arb_ps_float_constants;
4852         else
4853             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4854
4855         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4856         {
4857             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4858             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4859         }
4860         else if (ps_consts >= 32)
4861         {
4862             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4863             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4864             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4865         }
4866         else
4867         {
4868             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4869             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4870         }
4871         pCaps->PixelShader1xMaxValue = 8.0f;
4872         pCaps->MaxPixelShaderConst = ps_consts;
4873     }
4874     else
4875     {
4876         pCaps->PixelShaderVersion = 0;
4877         pCaps->PixelShader1xMaxValue = 0.0f;
4878         pCaps->MaxPixelShaderConst = 0;
4879     }
4880
4881     pCaps->VSClipping = use_nv_clip(gl_info);
4882 }
4883
4884 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4885 {
4886     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4887     {
4888         TRACE("Checking support for color_fixup:\n");
4889         dump_color_fixup_desc(fixup);
4890     }
4891
4892     /* We support everything except complex conversions. */
4893     if (!is_complex_fixup(fixup))
4894     {
4895         TRACE("[OK]\n");
4896         return TRUE;
4897     }
4898
4899     TRACE("[FAILED]\n");
4900     return FALSE;
4901 }
4902
4903 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4904     DWORD shift;
4905     char write_mask[20], regstr[50];
4906     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4907     BOOL is_color = FALSE;
4908     const struct wined3d_shader_dst_param *dst;
4909
4910     if (!ins->dst_count) return;
4911
4912     dst = &ins->dst[0];
4913     shift = dst->shift;
4914     if (!shift) return; /* Saturate alone is handled by the instructions */
4915
4916     shader_arb_get_write_mask(ins, dst, write_mask);
4917     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4918
4919     /* Generate a line that does the output modifier computation
4920      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4921      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4922      */
4923     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4924                    regstr, write_mask, regstr, shift_tab[shift]);
4925 }
4926
4927 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4928 {
4929     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4930     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4931     /* WINED3DSIH_AND           */ NULL,
4932     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4933     /* WINED3DSIH_BREAK         */ shader_hw_break,
4934     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4935     /* WINED3DSIH_BREAKP        */ NULL,
4936     /* WINED3DSIH_CALL          */ shader_hw_call,
4937     /* WINED3DSIH_CALLNZ        */ NULL,
4938     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4939     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4940     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4941     /* WINED3DSIH_CUT           */ NULL,
4942     /* WINED3DSIH_DCL           */ NULL,
4943     /* WINED3DSIH_DEF           */ NULL,
4944     /* WINED3DSIH_DEFB          */ NULL,
4945     /* WINED3DSIH_DEFI          */ NULL,
4946     /* WINED3DSIH_DIV           */ NULL,
4947     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4948     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4949     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4950     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4951     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4952     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4953     /* WINED3DSIH_ELSE          */ shader_hw_else,
4954     /* WINED3DSIH_EMIT          */ NULL,
4955     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4956     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4957     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4958     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4959     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4960     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4961     /* WINED3DSIH_FTOI          */ NULL,
4962     /* WINED3DSIH_IADD          */ NULL,
4963     /* WINED3DSIH_IEQ           */ NULL,
4964     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4965     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4966     /* WINED3DSIH_IGE           */ NULL,
4967     /* WINED3DSIH_IMUL          */ NULL,
4968     /* WINED3DSIH_ITOF          */ NULL,
4969     /* WINED3DSIH_LABEL         */ shader_hw_label,
4970     /* WINED3DSIH_LD            */ NULL,
4971     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4972     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4973     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4974     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4975     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4976     /* WINED3DSIH_LT            */ NULL,
4977     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4978     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4979     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4980     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4981     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4982     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4983     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4984     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4985     /* WINED3DSIH_MOV           */ shader_hw_mov,
4986     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4987     /* WINED3DSIH_MOVC          */ NULL,
4988     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4989     /* WINED3DSIH_NOP           */ shader_hw_nop,
4990     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4991     /* WINED3DSIH_PHASE         */ NULL,
4992     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4993     /* WINED3DSIH_RCP           */ shader_hw_rcp,
4994     /* WINED3DSIH_REP           */ shader_hw_rep,
4995     /* WINED3DSIH_RET           */ shader_hw_ret,
4996     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4997     /* WINED3DSIH_SAMPLE        */ NULL,
4998     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
4999     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5000     /* WINED3DSIH_SETP          */ NULL,
5001     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5002     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5003     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5004     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5005     /* WINED3DSIH_SQRT          */ NULL,
5006     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5007     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5008     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5009     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5010     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5011     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5012     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5013     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5014     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5015     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5016     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5017     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5018     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5019     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5020     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5021     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5022     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5023     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5024     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5025     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5026     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5027     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5028     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5029     /* WINED3DSIH_UTOF          */ NULL,
5030 };
5031
5032 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
5033 {
5034     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5035     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5036     WORD bools = 0;
5037     WORD flag = (1 << idx);
5038     const local_constant *constant;
5039     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5040
5041     if (reg_maps->local_bool_consts & flag)
5042     {
5043         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5044         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
5045         {
5046             if (constant->idx == idx)
5047             {
5048                 return constant->value[0];
5049             }
5050         }
5051         ERR("Local constant not found\n");
5052         return FALSE;
5053     }
5054     else
5055     {
5056         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5057         else bools = priv->cur_ps_args->bools;
5058         return bools & flag;
5059     }
5060 }
5061
5062 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5063         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
5064 {
5065     const struct shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5066     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5067
5068     /* Integer constants can either be a local constant, or they can be stored in the shader
5069      * type specific compile args. */
5070     if (reg_maps->local_int_consts & (1 << idx))
5071     {
5072         const local_constant *constant;
5073
5074         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
5075         {
5076             if (constant->idx == idx)
5077             {
5078                 loop_control->count = constant->value[0];
5079                 loop_control->start = constant->value[1];
5080                 /* Step is signed. */
5081                 loop_control->step = (int)constant->value[2];
5082                 return;
5083             }
5084         }
5085         /* If this happens the flag was set incorrectly */
5086         ERR("Local constant not found\n");
5087         loop_control->count = 0;
5088         loop_control->start = 0;
5089         loop_control->step = 0;
5090         return;
5091     }
5092
5093     switch (reg_maps->shader_version.type)
5094     {
5095         case WINED3D_SHADER_TYPE_VERTEX:
5096             /* Count and aL start value are unsigned */
5097             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5098             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5099             /* Step is signed. */
5100             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5101             break;
5102
5103         case WINED3D_SHADER_TYPE_PIXEL:
5104             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5105             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5106             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5107             break;
5108
5109         default:
5110             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5111             break;
5112     }
5113 }
5114
5115 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5116 {
5117     unsigned int i;
5118     struct wined3d_shader_dst_param *dst_param = NULL;
5119     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5120     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5121     if(!rec)
5122     {
5123         ERR("Out of memory\n");
5124         return;
5125     }
5126
5127     rec->ins = *ins;
5128     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5129     if(!dst_param) goto free;
5130     *dst_param = *ins->dst;
5131     if(ins->dst->reg.rel_addr)
5132     {
5133         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5134         if(!rel_addr) goto free;
5135         *rel_addr = *ins->dst->reg.rel_addr;
5136         dst_param->reg.rel_addr = rel_addr;
5137     }
5138     rec->ins.dst = dst_param;
5139
5140     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5141     if(!src_param) goto free;
5142     for(i = 0; i < ins->src_count; i++)
5143     {
5144         src_param[i] = ins->src[i];
5145         if(ins->src[i].reg.rel_addr)
5146         {
5147             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5148             if(!rel_addr) goto free;
5149             *rel_addr = *ins->src[i].reg.rel_addr;
5150             src_param[i].reg.rel_addr = rel_addr;
5151         }
5152     }
5153     rec->ins.src = src_param;
5154     list_add_tail(list, &rec->entry);
5155     return;
5156
5157 free:
5158     ERR("Out of memory\n");
5159     if(dst_param)
5160     {
5161         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5162         HeapFree(GetProcessHeap(), 0, dst_param);
5163     }
5164     if(src_param)
5165     {
5166         for(i = 0; i < ins->src_count; i++)
5167         {
5168             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5169         }
5170         HeapFree(GetProcessHeap(), 0, src_param);
5171     }
5172     HeapFree(GetProcessHeap(), 0, rec);
5173 }
5174
5175 static void free_recorded_instruction(struct list *list)
5176 {
5177     struct recorded_instruction *rec_ins, *entry2;
5178     unsigned int i;
5179
5180     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5181     {
5182         list_remove(&rec_ins->entry);
5183         if(rec_ins->ins.dst)
5184         {
5185             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5186             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5187         }
5188         if(rec_ins->ins.src)
5189         {
5190             for(i = 0; i < rec_ins->ins.src_count; i++)
5191             {
5192                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5193             }
5194             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5195         }
5196         HeapFree(GetProcessHeap(), 0, rec_ins);
5197     }
5198 }
5199
5200 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5201     SHADER_HANDLER hw_fct;
5202     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5203     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
5204     struct control_frame *control_frame;
5205     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5206     BOOL bool_const;
5207
5208     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5209     {
5210         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5211         list_add_head(&priv->control_frames, &control_frame->entry);
5212
5213         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5214         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5215
5216         if(priv->target_version >= NV2)
5217         {
5218             control_frame->no.loop = priv->num_loops++;
5219             priv->loop_depth++;
5220         }
5221         else
5222         {
5223             /* Don't bother recording when we're in a not used if branch */
5224             if(priv->muted)
5225             {
5226                 return;
5227             }
5228
5229             if(!priv->recording)
5230             {
5231                 list_init(&priv->record);
5232                 priv->recording = TRUE;
5233                 control_frame->outer_loop = TRUE;
5234                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
5235                 return; /* Instruction is handled */
5236             }
5237             /* Record this loop in the outer loop's recording */
5238         }
5239     }
5240     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5241     {
5242         if(priv->target_version >= NV2)
5243         {
5244             /* Nothing to do. The control frame is popped after the HW instr handler */
5245         }
5246         else
5247         {
5248             struct list *e = list_head(&priv->control_frames);
5249             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5250             list_remove(&control_frame->entry);
5251
5252             if(control_frame->outer_loop)
5253             {
5254                 unsigned int iteration;
5255                 int aL = 0;
5256                 struct list copy;
5257
5258                 /* Turn off recording before playback */
5259                 priv->recording = FALSE;
5260
5261                 /* Move the recorded instructions to a separate list and get them out of the private data
5262                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5263                  * be recorded again, thus priv->record might be overwritten
5264                  */
5265                 list_init(&copy);
5266                 list_move_tail(&copy, &priv->record);
5267                 list_init(&priv->record);
5268
5269                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5270                 {
5271                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5272                                    control_frame->loop_control.count, control_frame->loop_control.start,
5273                                    control_frame->loop_control.step);
5274                     aL = control_frame->loop_control.start;
5275                 }
5276                 else
5277                 {
5278                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5279                 }
5280
5281                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5282                 {
5283                     struct recorded_instruction *rec_ins;
5284                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5285                     {
5286                         priv->aL = aL;
5287                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5288                     }
5289                     else
5290                     {
5291                         shader_addline(buffer, "#Iteration %u\n", iteration);
5292                     }
5293
5294                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5295                     {
5296                         shader_arb_handle_instruction(&rec_ins->ins);
5297                     }
5298
5299                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5300                     {
5301                         aL += control_frame->loop_control.step;
5302                     }
5303                 }
5304                 shader_addline(buffer, "#end loop/rep\n");
5305
5306                 free_recorded_instruction(&copy);
5307                 HeapFree(GetProcessHeap(), 0, control_frame);
5308                 return; /* Instruction is handled */
5309             }
5310             else
5311             {
5312                 /* This is a nested loop. Proceed to the normal recording function */
5313                 HeapFree(GetProcessHeap(), 0, control_frame);
5314             }
5315         }
5316     }
5317
5318     if(priv->recording)
5319     {
5320         record_instruction(&priv->record, ins);
5321         return;
5322     }
5323
5324     /* boolean if */
5325     if(ins->handler_idx == WINED3DSIH_IF)
5326     {
5327         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5328         list_add_head(&priv->control_frames, &control_frame->entry);
5329         control_frame->type = IF;
5330
5331         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
5332         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5333         if (!priv->muted && !bool_const)
5334         {
5335             shader_addline(buffer, "#if(FALSE){\n");
5336             priv->muted = TRUE;
5337             control_frame->muting = TRUE;
5338         }
5339         else shader_addline(buffer, "#if(TRUE) {\n");
5340
5341         return; /* Instruction is handled */
5342     }
5343     else if(ins->handler_idx == WINED3DSIH_IFC)
5344     {
5345         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5346         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5347         control_frame->type = IFC;
5348         control_frame->no.ifc = priv->num_ifcs++;
5349         list_add_head(&priv->control_frames, &control_frame->entry);
5350     }
5351     else if(ins->handler_idx == WINED3DSIH_ELSE)
5352     {
5353         struct list *e = list_head(&priv->control_frames);
5354         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5355
5356         if(control_frame->type == IF)
5357         {
5358             shader_addline(buffer, "#} else {\n");
5359             if(!priv->muted && !control_frame->muting)
5360             {
5361                 priv->muted = TRUE;
5362                 control_frame->muting = TRUE;
5363             }
5364             else if(control_frame->muting) priv->muted = FALSE;
5365             return; /* Instruction is handled. */
5366         }
5367         /* In case of an ifc, generate a HW shader instruction */
5368     }
5369     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5370     {
5371         struct list *e = list_head(&priv->control_frames);
5372         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5373
5374         if(control_frame->type == IF)
5375         {
5376             shader_addline(buffer, "#} endif\n");
5377             if(control_frame->muting) priv->muted = FALSE;
5378             list_remove(&control_frame->entry);
5379             HeapFree(GetProcessHeap(), 0, control_frame);
5380             return; /* Instruction is handled */
5381         }
5382     }
5383
5384     if(priv->muted) return;
5385
5386     /* Select handler */
5387     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5388
5389     /* Unhandled opcode */
5390     if (!hw_fct)
5391     {
5392         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5393         return;
5394     }
5395     hw_fct(ins);
5396
5397     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5398     {
5399         struct list *e = list_head(&priv->control_frames);
5400         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5401         list_remove(&control_frame->entry);
5402         HeapFree(GetProcessHeap(), 0, control_frame);
5403         priv->loop_depth--;
5404     }
5405     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5406     {
5407         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5408         struct list *e = list_head(&priv->control_frames);
5409         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5410         list_remove(&control_frame->entry);
5411         HeapFree(GetProcessHeap(), 0, control_frame);
5412     }
5413
5414
5415     shader_arb_add_instruction_modifiers(ins);
5416 }
5417
5418 const shader_backend_t arb_program_shader_backend = {
5419     shader_arb_handle_instruction,
5420     shader_arb_select,
5421     shader_arb_select_depth_blt,
5422     shader_arb_deselect_depth_blt,
5423     shader_arb_update_float_vertex_constants,
5424     shader_arb_update_float_pixel_constants,
5425     shader_arb_load_constants,
5426     shader_arb_load_np2fixup_constants,
5427     shader_arb_destroy,
5428     shader_arb_alloc,
5429     shader_arb_free,
5430     shader_arb_dirty_const,
5431     shader_arb_get_caps,
5432     shader_arb_color_fixup_supported,
5433 };
5434
5435 /* ARB_fragment_program fixed function pipeline replacement definitions */
5436 #define ARB_FFP_CONST_TFACTOR           0
5437 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5438 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5439 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5440 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5441
5442 struct arbfp_ffp_desc
5443 {
5444     struct ffp_frag_desc parent;
5445     GLuint shader;
5446     unsigned int num_textures_used;
5447 };
5448
5449 /* Context activation is done by the caller. */
5450 static void arbfp_enable(BOOL enable)
5451 {
5452     ENTER_GL();
5453     if(enable) {
5454         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5455         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5456     } else {
5457         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5458         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5459     }
5460     LEAVE_GL();
5461 }
5462
5463 static HRESULT arbfp_alloc(IWineD3DDeviceImpl *device)
5464 {
5465     struct shader_arb_priv *priv;
5466     /* Share private data between the shader backend and the pipeline replacement, if both
5467      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5468      * if no pixel shader is bound or not
5469      */
5470     if (device->shader_backend == &arb_program_shader_backend)
5471     {
5472         device->fragment_priv = device->shader_priv;
5473     }
5474     else
5475     {
5476         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5477         if (!device->fragment_priv) return E_OUTOFMEMORY;
5478     }
5479     priv = device->fragment_priv;
5480     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5481     {
5482         ERR("Failed to initialize rbtree.\n");
5483         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5484         return E_OUTOFMEMORY;
5485     }
5486     priv->use_arbfp_fixed_func = TRUE;
5487     return WINED3D_OK;
5488 }
5489
5490 /* Context activation is done by the caller. */
5491 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5492 {
5493     const struct wined3d_gl_info *gl_info = context;
5494     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5495
5496     ENTER_GL();
5497     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5498     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5499     HeapFree(GetProcessHeap(), 0, entry_arb);
5500     LEAVE_GL();
5501 }
5502
5503 /* Context activation is done by the caller. */
5504 static void arbfp_free(IWineD3DDeviceImpl *device)
5505 {
5506     struct shader_arb_priv *priv = device->fragment_priv;
5507
5508     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5509     priv->use_arbfp_fixed_func = FALSE;
5510
5511     if (device->shader_backend != &arb_program_shader_backend)
5512     {
5513         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5514     }
5515 }
5516
5517 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5518 {
5519     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5520     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5521                            WINED3DTEXOPCAPS_SELECTARG1                  |
5522                            WINED3DTEXOPCAPS_SELECTARG2                  |
5523                            WINED3DTEXOPCAPS_MODULATE4X                  |
5524                            WINED3DTEXOPCAPS_MODULATE2X                  |
5525                            WINED3DTEXOPCAPS_MODULATE                    |
5526                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5527                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5528                            WINED3DTEXOPCAPS_ADD                         |
5529                            WINED3DTEXOPCAPS_SUBTRACT                    |
5530                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5531                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5532                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5533                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5534                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5535                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5536                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5537                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5538                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5539                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5540                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5541                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5542                            WINED3DTEXOPCAPS_LERP                        |
5543                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5544                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5545
5546     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5547
5548     caps->MaxTextureBlendStages   = 8;
5549     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5550 }
5551
5552 static void state_texfactor_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5553 {
5554     const struct wined3d_gl_info *gl_info = context->gl_info;
5555     const struct wined3d_state *state = &stateblock->state;
5556     IWineD3DDeviceImpl *device = stateblock->device;
5557     float col[4];
5558
5559     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5560      * application provided constants
5561      */
5562     if (device->shader_backend == &arb_program_shader_backend)
5563     {
5564         if (use_ps(state)) return;
5565
5566         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5567         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5568     }
5569
5570     D3DCOLORTOGLFLOAT4(state->render_states[WINED3DRS_TEXTUREFACTOR], col);
5571     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5572     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5573
5574 }
5575
5576 static void state_arb_specularenable(DWORD state_id,
5577         IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5578 {
5579     const struct wined3d_gl_info *gl_info = context->gl_info;
5580     const struct wined3d_state *state = &stateblock->state;
5581     IWineD3DDeviceImpl *device = stateblock->device;
5582     float col[4];
5583
5584     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5585      * application provided constants
5586      */
5587     if (device->shader_backend == &arb_program_shader_backend)
5588     {
5589         if (use_ps(state)) return;
5590
5591         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5592         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5593     }
5594
5595     if (state->render_states[WINED3DRS_SPECULARENABLE])
5596     {
5597         /* The specular color has no alpha */
5598         col[0] = 1.0f; col[1] = 1.0f;
5599         col[2] = 1.0f; col[3] = 0.0f;
5600     } else {
5601         col[0] = 0.0f; col[1] = 0.0f;
5602         col[2] = 0.0f; col[3] = 0.0f;
5603     }
5604     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5605     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5606 }
5607
5608 static void set_bumpmat_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5609 {
5610     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5611     const struct wined3d_gl_info *gl_info = context->gl_info;
5612     const struct wined3d_state *state = &stateblock->state;
5613     IWineD3DDeviceImpl *device = stateblock->device;
5614     float mat[2][2];
5615
5616     if (use_ps(state))
5617     {
5618         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
5619         if (stage && (ps->baseShader.reg_maps.bumpmat & (1 << stage)))
5620         {
5621             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5622              * anyway
5623              */
5624             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5625                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5626         }
5627
5628         if(device->shader_backend == &arb_program_shader_backend) {
5629             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5630             return;
5631         }
5632     } else if(device->shader_backend == &arb_program_shader_backend) {
5633         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5634         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5635     }
5636
5637     mat[0][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5638     mat[0][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5639     mat[1][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5640     mat[1][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5641
5642     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5643     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5644 }
5645
5646 static void tex_bumpenvlum_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5647 {
5648     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5649     const struct wined3d_gl_info *gl_info = context->gl_info;
5650     const struct wined3d_state *state = &stateblock->state;
5651     IWineD3DDeviceImpl *device = stateblock->device;
5652     float param[4];
5653
5654     if (use_ps(state))
5655     {
5656         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
5657         if (stage && (ps->baseShader.reg_maps.luminanceparams & (1 << stage)))
5658         {
5659             /* The pixel shader has to know the luminance offset. Do a constants update if it
5660              * isn't scheduled anyway
5661              */
5662             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5663                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5664         }
5665
5666         if(device->shader_backend == &arb_program_shader_backend) {
5667             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5668             return;
5669         }
5670     } else if(device->shader_backend == &arb_program_shader_backend) {
5671         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5672         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5673     }
5674
5675     param[0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5676     param[1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5677     param[2] = 0.0f;
5678     param[3] = 0.0f;
5679
5680     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5681     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5682 }
5683
5684 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5685 {
5686     const char *ret;
5687
5688     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5689
5690     switch(arg & WINED3DTA_SELECTMASK) {
5691         case WINED3DTA_DIFFUSE:
5692             ret = "fragment.color.primary"; break;
5693
5694         case WINED3DTA_CURRENT:
5695             if (!stage) ret = "fragment.color.primary";
5696             else ret = "ret";
5697             break;
5698
5699         case WINED3DTA_TEXTURE:
5700             switch(stage) {
5701                 case 0: ret = "tex0"; break;
5702                 case 1: ret = "tex1"; break;
5703                 case 2: ret = "tex2"; break;
5704                 case 3: ret = "tex3"; break;
5705                 case 4: ret = "tex4"; break;
5706                 case 5: ret = "tex5"; break;
5707                 case 6: ret = "tex6"; break;
5708                 case 7: ret = "tex7"; break;
5709                 default: ret = "unknown texture";
5710             }
5711             break;
5712
5713         case WINED3DTA_TFACTOR:
5714             ret = "tfactor"; break;
5715
5716         case WINED3DTA_SPECULAR:
5717             ret = "fragment.color.secondary"; break;
5718
5719         case WINED3DTA_TEMP:
5720             ret = "tempreg"; break;
5721
5722         case WINED3DTA_CONSTANT:
5723             FIXME("Implement perstage constants\n");
5724             switch(stage) {
5725                 case 0: ret = "const0"; break;
5726                 case 1: ret = "const1"; break;
5727                 case 2: ret = "const2"; break;
5728                 case 3: ret = "const3"; break;
5729                 case 4: ret = "const4"; break;
5730                 case 5: ret = "const5"; break;
5731                 case 6: ret = "const6"; break;
5732                 case 7: ret = "const7"; break;
5733                 default: ret = "unknown constant";
5734             }
5735             break;
5736
5737         default:
5738             return "unknown";
5739     }
5740
5741     if(arg & WINED3DTA_COMPLEMENT) {
5742         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5743         if(argnum == 0) ret = "arg0";
5744         if(argnum == 1) ret = "arg1";
5745         if(argnum == 2) ret = "arg2";
5746     }
5747     if(arg & WINED3DTA_ALPHAREPLICATE) {
5748         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5749         if(argnum == 0) ret = "arg0";
5750         if(argnum == 1) ret = "arg1";
5751         if(argnum == 2) ret = "arg2";
5752     }
5753     return ret;
5754 }
5755
5756 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5757         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5758 {
5759     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5760     unsigned int mul = 1;
5761     BOOL mul_final_dest = FALSE;
5762
5763     if(color && alpha) dstmask = "";
5764     else if(color) dstmask = ".xyz";
5765     else dstmask = ".w";
5766
5767     if(dst == tempreg) dstreg = "tempreg";
5768     else dstreg = "ret";
5769
5770     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5771     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5772     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5773
5774     switch(op) {
5775         case WINED3DTOP_DISABLE:
5776             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5777             break;
5778
5779         case WINED3DTOP_SELECTARG2:
5780             arg1 = arg2;
5781         case WINED3DTOP_SELECTARG1:
5782             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5783             break;
5784
5785         case WINED3DTOP_MODULATE4X:
5786             mul = 2;
5787         case WINED3DTOP_MODULATE2X:
5788             mul *= 2;
5789             if (!strcmp(dstreg, "result.color"))
5790             {
5791                 dstreg = "ret";
5792                 mul_final_dest = TRUE;
5793             }
5794         case WINED3DTOP_MODULATE:
5795             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5796             break;
5797
5798         case WINED3DTOP_ADDSIGNED2X:
5799             mul = 2;
5800             if (!strcmp(dstreg, "result.color"))
5801             {
5802                 dstreg = "ret";
5803                 mul_final_dest = TRUE;
5804             }
5805         case WINED3DTOP_ADDSIGNED:
5806             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5807             arg2 = "arg2";
5808         case WINED3DTOP_ADD:
5809             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5810             break;
5811
5812         case WINED3DTOP_SUBTRACT:
5813             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5814             break;
5815
5816         case WINED3DTOP_ADDSMOOTH:
5817             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5818             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5819             break;
5820
5821         case WINED3DTOP_BLENDCURRENTALPHA:
5822             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5823             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5824             break;
5825         case WINED3DTOP_BLENDFACTORALPHA:
5826             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5827             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5828             break;
5829         case WINED3DTOP_BLENDTEXTUREALPHA:
5830             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5831             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5832             break;
5833         case WINED3DTOP_BLENDDIFFUSEALPHA:
5834             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5835             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5836             break;
5837
5838         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5839             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5840             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5841             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5842             break;
5843
5844         /* D3DTOP_PREMODULATE ???? */
5845
5846         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5847             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5848             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5849             break;
5850         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5851             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5852             break;
5853         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5854             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5855             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5856             break;
5857         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5858             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5859             break;
5860
5861         case WINED3DTOP_DOTPRODUCT3:
5862             mul = 4;
5863             if (!strcmp(dstreg, "result.color"))
5864             {
5865                 dstreg = "ret";
5866                 mul_final_dest = TRUE;
5867             }
5868             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5869             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5870             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5871             break;
5872
5873         case WINED3DTOP_MULTIPLYADD:
5874             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5875             break;
5876
5877         case WINED3DTOP_LERP:
5878             /* The msdn is not quite right here */
5879             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5880             break;
5881
5882         case WINED3DTOP_BUMPENVMAP:
5883         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5884             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5885             break;
5886
5887         default:
5888             FIXME("Unhandled texture op %08x\n", op);
5889     }
5890
5891     if(mul == 2) {
5892         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5893     } else if(mul == 4) {
5894         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5895     }
5896 }
5897
5898 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5899 {
5900     const struct wined3d_gl_info *gl_info = &stateblock->device->adapter->gl_info;
5901     unsigned int stage;
5902     struct wined3d_shader_buffer buffer;
5903     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5904     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5905     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5906     const char *textype;
5907     const char *instr, *sat;
5908     char colorcor_dst[8];
5909     GLuint ret;
5910     DWORD arg0, arg1, arg2;
5911     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5912     BOOL op_equal;
5913     const char *final_combiner_src = "ret";
5914     GLint pos;
5915
5916     /* Find out which textures are read */
5917     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5918         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5919         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5920         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5921         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5922         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5923         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5924         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5925
5926         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5927         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5928         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5929             bump_used[stage] = TRUE;
5930             tex_read[stage] = TRUE;
5931         }
5932         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5933             bump_used[stage] = TRUE;
5934             tex_read[stage] = TRUE;
5935             luminance_used[stage] = TRUE;
5936         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5937             tfactor_used = TRUE;
5938         }
5939
5940         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5941             tfactor_used = TRUE;
5942         }
5943
5944         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5945         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5946             tempreg_used = TRUE;
5947         }
5948
5949         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5950         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5951         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5952         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5953         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5954         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5955         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5956
5957         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5958             tempreg_used = TRUE;
5959         }
5960         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5961             tfactor_used = TRUE;
5962         }
5963     }
5964
5965     /* Shader header */
5966     if (!shader_buffer_init(&buffer))
5967     {
5968         ERR("Failed to initialize shader buffer.\n");
5969         return 0;
5970     }
5971
5972     shader_addline(&buffer, "!!ARBfp1.0\n");
5973
5974     switch(settings->fog) {
5975         case FOG_OFF:                                                         break;
5976         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5977         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5978         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5979         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5980     }
5981
5982     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5983     shader_addline(&buffer, "TEMP TMP;\n");
5984     shader_addline(&buffer, "TEMP ret;\n");
5985     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5986     shader_addline(&buffer, "TEMP arg0;\n");
5987     shader_addline(&buffer, "TEMP arg1;\n");
5988     shader_addline(&buffer, "TEMP arg2;\n");
5989     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5990         if(!tex_read[stage]) continue;
5991         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5992         if(!bump_used[stage]) continue;
5993         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5994         if(!luminance_used[stage]) continue;
5995         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5996     }
5997     if(tfactor_used) {
5998         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5999     }
6000         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6001
6002     if(settings->sRGB_write) {
6003         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6004                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6005         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6006                        srgb_sub_high, 0.0, 0.0, 0.0);
6007     }
6008
6009     if (ffp_clip_emul(&stateblock->state) && settings->emul_clipplanes)
6010         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6011
6012     /* Generate texture sampling instructions) */
6013     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
6014         if(!tex_read[stage]) continue;
6015
6016         switch(settings->op[stage].tex_type) {
6017             case tex_1d:                    textype = "1D";     break;
6018             case tex_2d:                    textype = "2D";     break;
6019             case tex_3d:                    textype = "3D";     break;
6020             case tex_cube:                  textype = "CUBE";   break;
6021             case tex_rect:                  textype = "RECT";   break;
6022             default: textype = "unexpected_textype";   break;
6023         }
6024
6025         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
6026            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6027             sat = "";
6028         } else {
6029             sat = "_SAT";
6030         }
6031
6032         if(settings->op[stage].projected == proj_none) {
6033             instr = "TEX";
6034         } else if(settings->op[stage].projected == proj_count4 ||
6035                   settings->op[stage].projected == proj_count3) {
6036             instr = "TXP";
6037         } else {
6038             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6039             instr = "TXP";
6040         }
6041
6042         if(stage > 0 &&
6043            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6044             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6045             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6046             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6047             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6048             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6049
6050             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6051              * so multiply the displacement with the dividing parameter before passing it to TXP
6052              */
6053             if (settings->op[stage].projected != proj_none) {
6054                 if(settings->op[stage].projected == proj_count4) {
6055                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6056                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6057                 } else {
6058                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6059                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6060                 }
6061             } else {
6062                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6063             }
6064
6065             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6066                            instr, sat, stage, stage, textype);
6067             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6068                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6069                                stage - 1, stage - 1, stage - 1);
6070                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6071             }
6072         } else if(settings->op[stage].projected == proj_count3) {
6073             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6074             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6075             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6076                             instr, sat, stage, stage, textype);
6077         } else {
6078             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6079                             instr, sat, stage, stage, stage, textype);
6080         }
6081
6082         sprintf(colorcor_dst, "tex%u", stage);
6083         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6084                 settings->op[stage].color_fixup);
6085     }
6086
6087     /* Generate the main shader */
6088     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6089     {
6090         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6091         {
6092             if (!stage) final_combiner_src = "fragment.color.primary";
6093             break;
6094         }
6095
6096         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6097            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6098             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6099         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6100                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6101             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6102         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6103                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6104             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6105         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6106                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6107             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6108         } else {
6109             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6110                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6111                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6112                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6113         }
6114
6115         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6116             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6117                           settings->op[stage].cop, settings->op[stage].carg0,
6118                           settings->op[stage].carg1, settings->op[stage].carg2);
6119             if (!stage)
6120                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6121         }
6122         else if (op_equal)
6123         {
6124             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6125                           settings->op[stage].cop, settings->op[stage].carg0,
6126                           settings->op[stage].carg1, settings->op[stage].carg2);
6127         } else {
6128             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6129                           settings->op[stage].cop, settings->op[stage].carg0,
6130                           settings->op[stage].carg1, settings->op[stage].carg2);
6131             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6132                           settings->op[stage].aop, settings->op[stage].aarg0,
6133                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6134         }
6135     }
6136
6137     if(settings->sRGB_write) {
6138         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6139         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6140         shader_addline(&buffer, "MOV result.color, ret;\n");
6141     } else {
6142         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6143     }
6144
6145     /* Footer */
6146     shader_addline(&buffer, "END\n");
6147
6148     /* Generate the shader */
6149     GL_EXTCALL(glGenProgramsARB(1, &ret));
6150     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6151     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6152             strlen(buffer.buffer), buffer.buffer));
6153     checkGLcall("glProgramStringARB()");
6154
6155     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6156     if (pos != -1)
6157     {
6158         FIXME("Fragment program error at position %d: %s\n\n", pos,
6159               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6160         shader_arb_dump_program_source(buffer.buffer);
6161     }
6162     else
6163     {
6164         GLint native;
6165
6166         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6167         checkGLcall("glGetProgramivARB()");
6168         if (!native) WARN("Program exceeds native resource limits.\n");
6169     }
6170
6171     shader_buffer_free(&buffer);
6172     return ret;
6173 }
6174
6175 static void fragment_prog_arbfp(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6176 {
6177     const struct wined3d_gl_info *gl_info = context->gl_info;
6178     const struct wined3d_state *state = &stateblock->state;
6179     IWineD3DDeviceImpl *device = stateblock->device;
6180     struct shader_arb_priv *priv = device->fragment_priv;
6181     BOOL use_vshader = use_vs(state);
6182     BOOL use_pshader = use_ps(state);
6183     struct ffp_frag_settings settings;
6184     const struct arbfp_ffp_desc *desc;
6185     unsigned int i;
6186
6187     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6188
6189     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
6190         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6191             /* Reload fixed function constants since they collide with the pixel shader constants */
6192             for(i = 0; i < MAX_TEXTURES; i++) {
6193                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6194             }
6195             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6196             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6197         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6198             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6199         }
6200         return;
6201     }
6202
6203     if(!use_pshader) {
6204         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
6205         gen_ffp_frag_op(stateblock, &settings, FALSE);
6206         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6207         if(!desc) {
6208             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6209             if (!new_desc)
6210             {
6211                 ERR("Out of memory\n");
6212                 return;
6213             }
6214             new_desc->num_textures_used = 0;
6215             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6216             {
6217                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6218                 new_desc->num_textures_used = i;
6219             }
6220
6221             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6222             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
6223             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6224             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6225             desc = new_desc;
6226         }
6227
6228         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6229          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6230          * deactivate it.
6231          */
6232         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6233         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6234         priv->current_fprogram_id = desc->shader;
6235
6236         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6237             /* Reload fixed function constants since they collide with the pixel shader constants */
6238             for(i = 0; i < MAX_TEXTURES; i++) {
6239                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6240             }
6241             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6242             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6243         }
6244         context->last_was_pshader = FALSE;
6245     } else {
6246         context->last_was_pshader = TRUE;
6247     }
6248
6249     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6250      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6251      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6252      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6253      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6254      *
6255      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6256      * shader handler
6257      */
6258     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6259         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6260
6261         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6262             stateblock_apply_state(STATE_VERTEXSHADERCONSTANT, stateblock, context);
6263     }
6264     if (use_pshader) stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
6265 }
6266
6267 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
6268  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
6269  * state table, so we need to handle that with a forwarding function. The other invisible side effect
6270  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
6271  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
6272  */
6273 static void state_arbfp_fog(DWORD state_id, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6274 {
6275     const struct wined3d_state *state = &stateblock->state;
6276     enum fogsource new_source;
6277
6278     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6279
6280     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6281         fragment_prog_arbfp(state_id, stateblock, context);
6282     }
6283
6284     if (!state->render_states[WINED3DRS_FOGENABLE]) return;
6285
6286     if (state->render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6287     {
6288         if (use_vs(state))
6289         {
6290             new_source = FOGSOURCE_VS;
6291         }
6292         else
6293         {
6294             if (state->render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6295                 new_source = FOGSOURCE_COORD;
6296             else
6297                 new_source = FOGSOURCE_FFP;
6298         }
6299     } else {
6300         new_source = FOGSOURCE_FFP;
6301     }
6302     if(new_source != context->fog_source) {
6303         context->fog_source = new_source;
6304         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6305     }
6306 }
6307
6308 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6309 {
6310     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6311         fragment_prog_arbfp(state, stateblock, context);
6312     }
6313 }
6314
6315 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6316     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6317     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6318     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6319     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6320     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6321     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6322     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6323     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6324     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6325     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6326     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6327     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6328     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6329     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6330     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6331     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6332     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6333     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6334     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6335     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6336     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6337     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6338     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6339     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6340     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6341     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6342     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6343     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6344     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6345     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6346     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6347     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6348     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6349     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6350     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6351     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6352     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6353     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6354     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6355     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6356     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6357     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6358     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6359     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6360     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6361     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6362     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6363     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6364     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6365     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6366     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6367     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6368     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6369     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6370     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6371     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6372     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6373     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6374     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6375     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6376     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6377     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6378     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6379     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6380     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6381     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6382     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6383     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6384     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6385     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6386     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6387     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6388     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6389     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6390     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6391     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6392     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6393     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6394     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6395     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6396     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6397     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6398     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6399     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6400     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6401     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6402     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6403     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6404     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6405     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6406     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6407     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6408     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6409     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6410     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6411     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6412     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6413     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6414     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6415     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6416     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6417     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6418     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6419     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6420     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6421     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6422     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6423     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6424     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6425     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6426     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6427     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6428     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6429     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6430     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6431     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6432     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6433     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6434     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6435     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6436     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6437     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6438     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6439     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6440     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6441     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6442     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6443     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6444     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6445     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6446     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6447     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6448     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6449     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6450     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6451     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6452     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6453     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6454     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6455     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6456 };
6457
6458 const struct fragment_pipeline arbfp_fragment_pipeline = {
6459     arbfp_enable,
6460     arbfp_get_caps,
6461     arbfp_alloc,
6462     arbfp_free,
6463     shader_arb_color_fixup_supported,
6464     arbfp_fragmentstate_template,
6465     TRUE /* We can disable projected textures */
6466 };
6467
6468 struct arbfp_blit_priv {
6469     GLenum yuy2_rect_shader, yuy2_2d_shader;
6470     GLenum uyvy_rect_shader, uyvy_2d_shader;
6471     GLenum yv12_rect_shader, yv12_2d_shader;
6472     GLenum p8_rect_shader, p8_2d_shader;
6473     GLuint palette_texture;
6474 };
6475
6476 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6477     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6478     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6479     if(!device->blit_priv) {
6480         ERR("Out of memory\n");
6481         return E_OUTOFMEMORY;
6482     }
6483     return WINED3D_OK;
6484 }
6485
6486 /* Context activation is done by the caller. */
6487 static void arbfp_blit_free(IWineD3DDevice *iface) {
6488     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6489     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6490     struct arbfp_blit_priv *priv = device->blit_priv;
6491
6492     ENTER_GL();
6493     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6494     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6495     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6496     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6497     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6498     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6499     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6500     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6501     checkGLcall("Delete yuv and p8 programs");
6502
6503     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6504     LEAVE_GL();
6505
6506     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6507     device->blit_priv = NULL;
6508 }
6509
6510 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6511         GLenum textype, char *luminance)
6512 {
6513     char chroma;
6514     const char *tex, *texinstr;
6515
6516     if (fixup == COMPLEX_FIXUP_UYVY) {
6517         chroma = 'x';
6518         *luminance = 'w';
6519     } else {
6520         chroma = 'w';
6521         *luminance = 'x';
6522     }
6523     switch(textype) {
6524         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6525         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6526         default:
6527             /* This is more tricky than just replacing the texture type - we have to navigate
6528              * properly in the texture to find the correct chroma values
6529              */
6530             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6531             return FALSE;
6532     }
6533
6534     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6535      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6536      * filtering when we sample the texture.
6537      *
6538      * These are the rules for reading the chroma:
6539      *
6540      * Even pixel: Cr
6541      * Even pixel: U
6542      * Odd pixel: V
6543      *
6544      * So we have to get the sampling x position in non-normalized coordinates in integers
6545      */
6546     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6547         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6548         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6549     } else {
6550         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6551     }
6552     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6553      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6554      * 0.5, so add 0.5.
6555      */
6556     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6557     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6558
6559     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6560      * even and odd pixels respectively
6561      */
6562     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6563     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6564
6565     /* Sample Pixel 1 */
6566     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6567
6568     /* Put the value into either of the chroma values */
6569     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6570     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6571     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6572     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6573
6574     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6575      * the pixel right to the current one. Otherwise, sample the left pixel.
6576      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6577      */
6578     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6579     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6580     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6581
6582     /* Put the value into the other chroma */
6583     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6584     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6585     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6586     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6587
6588     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6589      * the current one and lerp the two U and V values
6590      */
6591
6592     /* This gives the correctly filtered luminance value */
6593     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6594
6595     return TRUE;
6596 }
6597
6598 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6599 {
6600     const char *tex;
6601
6602     switch(textype) {
6603         case GL_TEXTURE_2D:             tex = "2D";     break;
6604         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6605         default:
6606             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6607             return FALSE;
6608     }
6609
6610     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6611      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6612      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6613      * pitch of the luminance plane, the packing into the gl texture is a bit
6614      * unfortunate. If the whole texture is interpreted as luminance data it looks
6615      * approximately like this:
6616      *
6617      *        +----------------------------------+----
6618      *        |                                  |
6619      *        |                                  |
6620      *        |                                  |
6621      *        |                                  |
6622      *        |                                  |   2
6623      *        |            LUMINANCE             |   -
6624      *        |                                  |   3
6625      *        |                                  |
6626      *        |                                  |
6627      *        |                                  |
6628      *        |                                  |
6629      *        +----------------+-----------------+----
6630      *        |                |                 |
6631      *        |  U even rows   |  U odd rows     |
6632      *        |                |                 |   1
6633      *        +----------------+------------------   -
6634      *        |                |                 |   3
6635      *        |  V even rows   |  V odd rows     |
6636      *        |                |                 |
6637      *        +----------------+-----------------+----
6638      *        |                |                 |
6639      *        |     0.5        |       0.5       |
6640      *
6641      * So it appears as if there are 4 chroma images, but in fact the odd rows
6642      * in the chroma images are in the same row as the even ones. So its is
6643      * kinda tricky to read
6644      *
6645      * When reading from rectangle textures, keep in mind that the input y coordinates
6646      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6647      */
6648     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6649             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6650
6651     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6652     /* the chroma planes have only half the width */
6653     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6654
6655     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6656      * the coordinate. Also read the right side of the image when reading odd lines
6657      *
6658      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6659      * bleeding
6660      */
6661     if(textype == GL_TEXTURE_2D) {
6662
6663         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6664
6665         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6666
6667         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6668         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6669
6670         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6671         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6672         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6673         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6674         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6675
6676         /* clamp, keep the half pixel origin in mind */
6677         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6678         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6679         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6680         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6681     } else {
6682         /* Read from [size - size+size/4] */
6683         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6684         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6685
6686         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6687         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6688         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6689         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6690         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6691         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6692
6693         /* Make sure to read exactly from the pixel center */
6694         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6695         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6696
6697         /* Clamp */
6698         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6699         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6700         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6701         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6702         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6703     }
6704     /* Read the texture, put the result into the output register */
6705     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6706     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6707
6708     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6709      * No need to clamp because we're just reusing the already clamped value from above
6710      */
6711     if(textype == GL_TEXTURE_2D) {
6712         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6713     } else {
6714         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6715     }
6716     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6717     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6718
6719     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6720      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6721      * values due to filtering
6722      */
6723     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6724     if(textype == GL_TEXTURE_2D) {
6725         /* Multiply the y coordinate by 2/3 and clamp it */
6726         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6727         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6728         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6729         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6730     } else {
6731         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6732          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6733          * is bigger
6734          */
6735         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6736         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6737         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6738     }
6739     *luminance = 'a';
6740
6741     return TRUE;
6742 }
6743
6744 static GLuint gen_p8_shader(IWineD3DDeviceImpl *device, GLenum textype)
6745 {
6746     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6747     GLenum shader;
6748     struct wined3d_shader_buffer buffer;
6749     struct arbfp_blit_priv *priv = device->blit_priv;
6750     GLint pos;
6751
6752     /* Shader header */
6753     if (!shader_buffer_init(&buffer))
6754     {
6755         ERR("Failed to initialize shader buffer.\n");
6756         return 0;
6757     }
6758
6759     ENTER_GL();
6760     GL_EXTCALL(glGenProgramsARB(1, &shader));
6761     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6762     LEAVE_GL();
6763     if(!shader) {
6764         shader_buffer_free(&buffer);
6765         return 0;
6766     }
6767
6768     shader_addline(&buffer, "!!ARBfp1.0\n");
6769     shader_addline(&buffer, "TEMP index;\n");
6770
6771     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6772     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6773
6774     /* The alpha-component contains the palette index */
6775     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6776         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6777     else
6778         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6779
6780     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6781     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6782
6783     /* Use the alpha-component as an index in the palette to get the final color */
6784     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6785     shader_addline(&buffer, "END\n");
6786
6787     ENTER_GL();
6788     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6789             strlen(buffer.buffer), buffer.buffer));
6790     checkGLcall("glProgramStringARB()");
6791
6792     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6793     if (pos != -1)
6794     {
6795         FIXME("Fragment program error at position %d: %s\n\n", pos,
6796               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6797         shader_arb_dump_program_source(buffer.buffer);
6798     }
6799
6800     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6801         priv->p8_rect_shader = shader;
6802     else
6803         priv->p8_2d_shader = shader;
6804
6805     shader_buffer_free(&buffer);
6806     LEAVE_GL();
6807
6808     return shader;
6809 }
6810
6811 /* Context activation is done by the caller. */
6812 static void upload_palette(IWineD3DSurfaceImpl *surface)
6813 {
6814     BYTE table[256][4];
6815     IWineD3DDeviceImpl *device = surface->resource.device;
6816     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6817     struct arbfp_blit_priv *priv = device->blit_priv;
6818     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6819
6820     d3dfmt_p8_init_palette(surface, table, colorkey);
6821
6822     ENTER_GL();
6823     if (!priv->palette_texture)
6824         glGenTextures(1, &priv->palette_texture);
6825
6826     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6827     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6828
6829     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6830
6831     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6832     /* Make sure we have discrete color levels. */
6833     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6834     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6835     /* Upload the palette */
6836     /* TODO: avoid unneeed uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6837     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6838
6839     /* Switch back to unit 0 in which the 2D texture will be stored. */
6840     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE0));
6841     LEAVE_GL();
6842 }
6843
6844 /* Context activation is done by the caller. */
6845 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum complex_fixup yuv_fixup, GLenum textype)
6846 {
6847     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6848     GLenum shader;
6849     struct wined3d_shader_buffer buffer;
6850     char luminance_component;
6851     struct arbfp_blit_priv *priv = device->blit_priv;
6852     GLint pos;
6853
6854     /* Shader header */
6855     if (!shader_buffer_init(&buffer))
6856     {
6857         ERR("Failed to initialize shader buffer.\n");
6858         return 0;
6859     }
6860
6861     ENTER_GL();
6862     GL_EXTCALL(glGenProgramsARB(1, &shader));
6863     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6864     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6865     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6866     LEAVE_GL();
6867     if(!shader) {
6868         shader_buffer_free(&buffer);
6869         return 0;
6870     }
6871
6872     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6873      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6874      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6875      * each single pixel it contains, and one U and one V value shared between both
6876      * pixels.
6877      *
6878      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6879      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6880      * take the format into account when generating the read swizzles
6881      *
6882      * Reading the Y value is straightforward - just sample the texture. The hardware
6883      * takes care of filtering in the horizontal and vertical direction.
6884      *
6885      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6886      * because that would mix the U and V values of one pixel or two adjacent pixels.
6887      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6888      * regardless of the filtering setting. Vertical filtering works automatically
6889      * though - the U and V values of two rows are mixed nicely.
6890      *
6891      * Appart of avoiding filtering issues, the code has to know which value it just
6892      * read, and where it can find the other one. To determine this, it checks if
6893      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6894      *
6895      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6896      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6897      *
6898      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6899      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6900      * in an unfiltered situation. Finding the luminance on the other hand requires
6901      * finding out if it is an odd or even pixel. The real drawback of this approach
6902      * is filtering. This would have to be emulated completely in the shader, reading
6903      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6904      * vertically. Beyond that it would require adjustments to the texture handling
6905      * code to deal with the width scaling
6906      */
6907     shader_addline(&buffer, "!!ARBfp1.0\n");
6908     shader_addline(&buffer, "TEMP luminance;\n");
6909     shader_addline(&buffer, "TEMP temp;\n");
6910     shader_addline(&buffer, "TEMP chroma;\n");
6911     shader_addline(&buffer, "TEMP texcrd;\n");
6912     shader_addline(&buffer, "TEMP texcrd2;\n");
6913     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6914     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6915     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6916
6917     switch (yuv_fixup)
6918     {
6919         case COMPLEX_FIXUP_UYVY:
6920         case COMPLEX_FIXUP_YUY2:
6921             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6922             {
6923                 shader_buffer_free(&buffer);
6924                 return 0;
6925             }
6926             break;
6927
6928         case COMPLEX_FIXUP_YV12:
6929             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6930             {
6931                 shader_buffer_free(&buffer);
6932                 return 0;
6933             }
6934             break;
6935
6936         default:
6937             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6938             shader_buffer_free(&buffer);
6939             return 0;
6940     }
6941
6942     /* Calculate the final result. Formula is taken from
6943      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6944      * ranges from -0.5 to 0.5
6945      */
6946     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6947
6948     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6949     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6950     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6951     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6952     shader_addline(&buffer, "END\n");
6953
6954     ENTER_GL();
6955     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6956             strlen(buffer.buffer), buffer.buffer));
6957     checkGLcall("glProgramStringARB()");
6958
6959     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6960     if (pos != -1)
6961     {
6962         FIXME("Fragment program error at position %d: %s\n\n", pos,
6963               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6964         shader_arb_dump_program_source(buffer.buffer);
6965     }
6966     else
6967     {
6968         GLint native;
6969
6970         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6971         checkGLcall("glGetProgramivARB()");
6972         if (!native) WARN("Program exceeds native resource limits.\n");
6973     }
6974
6975     shader_buffer_free(&buffer);
6976     LEAVE_GL();
6977
6978     switch (yuv_fixup)
6979     {
6980         case COMPLEX_FIXUP_YUY2:
6981             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6982             else priv->yuy2_2d_shader = shader;
6983             break;
6984
6985         case COMPLEX_FIXUP_UYVY:
6986             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6987             else priv->uyvy_2d_shader = shader;
6988             break;
6989
6990         case COMPLEX_FIXUP_YV12:
6991             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6992             else priv->yv12_2d_shader = shader;
6993             break;
6994         default:
6995             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
6996     }
6997
6998     return shader;
6999 }
7000
7001 /* Context activation is done by the caller. */
7002 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, IWineD3DSurfaceImpl *surface)
7003 {
7004     GLenum shader;
7005     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
7006     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7007     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7008     struct arbfp_blit_priv *priv = device->blit_priv;
7009     enum complex_fixup fixup;
7010     GLenum textype = surface->texture_target;
7011
7012     if (!is_complex_fixup(surface->resource.format->color_fixup))
7013     {
7014         TRACE("Fixup:\n");
7015         dump_color_fixup_desc(surface->resource.format->color_fixup);
7016         /* Don't bother setting up a shader for unconverted formats */
7017         ENTER_GL();
7018         glEnable(textype);
7019         checkGLcall("glEnable(textype)");
7020         LEAVE_GL();
7021         return WINED3D_OK;
7022     }
7023
7024     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7025
7026     switch(fixup)
7027     {
7028         case COMPLEX_FIXUP_YUY2:
7029             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7030             break;
7031
7032         case COMPLEX_FIXUP_UYVY:
7033             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7034             break;
7035
7036         case COMPLEX_FIXUP_YV12:
7037             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7038             break;
7039
7040         case COMPLEX_FIXUP_P8:
7041             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7042             if (!shader) shader = gen_p8_shader(device, textype);
7043
7044             upload_palette(surface);
7045             break;
7046
7047         default:
7048             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7049             ENTER_GL();
7050             glEnable(textype);
7051             checkGLcall("glEnable(textype)");
7052             LEAVE_GL();
7053             return E_NOTIMPL;
7054     }
7055
7056     if (!shader) shader = gen_yuv_shader(device, fixup, textype);
7057
7058     ENTER_GL();
7059     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7060     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7061     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7062     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7063     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7064     checkGLcall("glProgramLocalParameter4fvARB");
7065     LEAVE_GL();
7066
7067     return WINED3D_OK;
7068 }
7069
7070 /* Context activation is done by the caller. */
7071 static void arbfp_blit_unset(IWineD3DDevice *iface) {
7072     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
7073     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7074
7075     ENTER_GL();
7076     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7077     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7078     glDisable(GL_TEXTURE_2D);
7079     checkGLcall("glDisable(GL_TEXTURE_2D)");
7080     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7081     {
7082         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7083         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7084     }
7085     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7086     {
7087         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7088         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7089     }
7090     LEAVE_GL();
7091 }
7092
7093 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum blit_operation blit_op,
7094         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7095         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7096 {
7097     enum complex_fixup src_fixup;
7098
7099     if (blit_op != BLIT_OP_BLIT)
7100     {
7101         TRACE("Unsupported blit_op=%d\n", blit_op);
7102         return FALSE;
7103     }
7104
7105     src_fixup = get_complex_fixup(src_format->color_fixup);
7106     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7107     {
7108         TRACE("Checking support for fixup:\n");
7109         dump_color_fixup_desc(src_format->color_fixup);
7110     }
7111
7112     if (!is_identity_fixup(dst_format->color_fixup))
7113     {
7114         TRACE("Destination fixups are not supported\n");
7115         return FALSE;
7116     }
7117
7118     if (is_identity_fixup(src_format->color_fixup))
7119     {
7120         TRACE("[OK]\n");
7121         return TRUE;
7122     }
7123
7124      /* We only support YUV conversions. */
7125     if (!is_complex_fixup(src_format->color_fixup))
7126     {
7127         TRACE("[FAILED]\n");
7128         return FALSE;
7129     }
7130
7131     switch(src_fixup)
7132     {
7133         case COMPLEX_FIXUP_YUY2:
7134         case COMPLEX_FIXUP_UYVY:
7135         case COMPLEX_FIXUP_YV12:
7136         case COMPLEX_FIXUP_P8:
7137             TRACE("[OK]\n");
7138             return TRUE;
7139
7140         default:
7141             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7142             TRACE("[FAILED]\n");
7143             return FALSE;
7144     }
7145 }
7146
7147 HRESULT arbfp_blit_surface(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *src_surface, const RECT *src_rect,
7148                            IWineD3DSurfaceImpl *dst_surface, const RECT *dst_rect_in, enum blit_operation blit_op,
7149                            DWORD Filter)
7150 {
7151     IWineD3DSwapChainImpl *dst_swapchain;
7152     struct wined3d_context *context;
7153     RECT dst_rect = *dst_rect_in;
7154
7155     /* Now load the surface */
7156     surface_internal_preload(src_surface, SRGB_RGB);
7157
7158     /* Activate the destination context, set it up for blitting */
7159     context = context_acquire(device, dst_surface);
7160     context_apply_blit_state(context, device);
7161
7162     if (!surface_is_offscreen(dst_surface))
7163         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7164
7165     arbfp_blit_set((IWineD3DDevice *)device, src_surface);
7166
7167     ENTER_GL();
7168
7169     /* Draw a textured quad */
7170     draw_textured_quad(src_surface, src_rect, &dst_rect, Filter);
7171
7172     LEAVE_GL();
7173
7174     /* Leave the opengl state valid for blitting */
7175     arbfp_blit_unset((IWineD3DDevice *)device);
7176
7177     dst_swapchain = dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7178             ? dst_surface->container.u.swapchain : NULL;
7179     if (wined3d_settings.strict_draw_ordering || (dst_swapchain
7180             && (dst_surface == dst_swapchain->front_buffer
7181             || dst_swapchain->num_contexts > 1)))
7182         wglFlush(); /* Flush to ensure ordering across contexts. */
7183
7184     context_release(context);
7185
7186     surface_modify_location(dst_surface, SFLAG_INDRAWABLE, TRUE);
7187     return WINED3D_OK;
7188 }
7189
7190 /* Do not call while under the GL lock. */
7191 static HRESULT arbfp_blit_color_fill(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *dst_surface,
7192         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7193 {
7194     FIXME("Color filling not implemented by arbfp_blit\n");
7195     return WINED3DERR_INVALIDCALL;
7196 }
7197
7198 const struct blit_shader arbfp_blit = {
7199     arbfp_blit_alloc,
7200     arbfp_blit_free,
7201     arbfp_blit_set,
7202     arbfp_blit_unset,
7203     arbfp_blit_supported,
7204     arbfp_blit_color_fill
7205 };