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(IWineD3DStateBlockImpl *stateblock)
 179 {
 180     return stateblock->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(
 484     IWineD3DDevice* device,
 485     char usePixelShader,
 486     char useVertexShader) {
 487
 488     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl *) device;
 489     const struct shader_arb_priv* const priv = (const struct shader_arb_priv *) deviceImpl->shader_priv;
 490     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 491     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 492
 493     if (!usePixelShader) {
 494         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 495         return;
 496     }
 497
 498     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 499         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 500         UINT i;
 501         WORD active = fixup->super.active;
 502         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 503
 504         for (i = 0; active; active >>= 1, ++i) {
 505             const unsigned char idx = fixup->super.idx[i];
 506             const IWineD3DTextureImpl* const tex = (const IWineD3DTextureImpl*) stateBlock->textures[i];
 507             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 508
 509             if (!(active & 1)) continue;
 510
 511             if (!tex) {
 512                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 513                 continue;
 514             }
 515
 516             if (idx % 2) {
 517                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
 518             } else {
 519                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
 520             }
 521         }
 522
 523         for (i = 0; i < fixup->super.num_consts; ++i) {
 524             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 525                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 526         }
 527     }
 528 }
 529
 530 /* GL locking is done by the caller. */
 531 static inline void shader_arb_ps_local_constants(IWineD3DDeviceImpl* deviceImpl)
 532 {
 533     const struct wined3d_context *context = context_get_current();
 534     IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
 535     const struct wined3d_gl_info *gl_info = context->gl_info;
 536     unsigned char i;
 537     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 538     const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 539
 540     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 541     {
 542         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 543
 544         /* The state manager takes care that this function is always called if the bump env matrix changes */
 545         const float *data = (const float *)&stateBlock->state.texture_states[texunit][WINED3DTSS_BUMPENVMAT00];
 546         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 547                 gl_shader->bumpenvmatconst[i].const_num, data));
 548
 549         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 550         {
 551             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 552              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 553              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 554              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 555             */
 556             const float *scale = (const float *)&stateBlock->state.texture_states[texunit][WINED3DTSS_BUMPENVLSCALE];
 557             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 558                     gl_shader->luminanceconst[i].const_num, scale));
 559         }
 560     }
 561     checkGLcall("Load bumpmap consts");
 562
 563     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 564     {
 565         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 566         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 567         * ycorrection.z: 1.0
 568         * ycorrection.w: 0.0
 569         */
 570         float val[4];
 571         val[0] = context->render_offscreen ? 0.0f
 572                 : deviceImpl->render_targets[0]->currentDesc.Height;
 573         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 574         val[2] = 1.0f;
 575         val[3] = 0.0f;
 576         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 577         checkGLcall("y correction loading");
 578     }
 579
 580     if (!gl_shader->num_int_consts) return;
 581
 582     for(i = 0; i < MAX_CONST_I; i++)
 583     {
 584         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 585         {
 586             float val[4];
 587             val[0] = (float) stateBlock->pixelShaderConstantI[4 * i];
 588             val[1] = (float) stateBlock->pixelShaderConstantI[4 * i + 1];
 589             val[2] = (float) stateBlock->pixelShaderConstantI[4 * i + 2];
 590             val[3] = -1.0f;
 591
 592             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 593         }
 594     }
 595     checkGLcall("Load ps int consts");
 596 }
 597
 598 /* GL locking is done by the caller. */
 599 static inline void shader_arb_vs_local_constants(IWineD3DDeviceImpl* deviceImpl)
 600 {
 601     IWineD3DStateBlockImpl* stateBlock;
 602     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
 603     unsigned char i;
 604     struct shader_arb_priv *priv = deviceImpl->shader_priv;
 605     const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 606
 607     /* Upload the position fixup */
 608     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, deviceImpl->posFixup));
 609
 610     if (!gl_shader->num_int_consts) return;
 611
 612     stateBlock = deviceImpl->stateBlock;
 613
 614     for(i = 0; i < MAX_CONST_I; i++)
 615     {
 616         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 617         {
 618             float val[4];
 619             val[0] = (float) stateBlock->vertexShaderConstantI[4 * i];
 620             val[1] = (float) stateBlock->vertexShaderConstantI[4 * i + 1];
 621             val[2] = (float) stateBlock->vertexShaderConstantI[4 * i + 2];
 622             val[3] = -1.0f;
 623
 624             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 625         }
 626     }
 627     checkGLcall("Load vs int consts");
 628 }
 629
 630 /**
 631  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 632  *
 633  * We only support float constants in ARB at the moment, so don't
 634  * worry about the Integers or Booleans
 635  */
 636 /* GL locking is done by the caller (state handler) */
 637 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 638 {
 639     IWineD3DDeviceImpl *device = context->swapchain->device;
 640     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
 641     const struct wined3d_gl_info *gl_info = context->gl_info;
 642
 643     if (useVertexShader) {
 644         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
 645
 646         /* Load DirectX 9 float constants for vertex shader */
 647         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 648                 device->highest_dirty_vs_const, stateBlock->vertexShaderConstantF, context->vshader_const_dirty);
 649         shader_arb_vs_local_constants(device);
 650     }
 651
 652     if (usePixelShader) {
 653         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
 654
 655         /* Load DirectX 9 float constants for pixel shader */
 656         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 657                 device->highest_dirty_ps_const, stateBlock->pixelShaderConstantF, context->pshader_const_dirty);
 658         shader_arb_ps_local_constants(device);
 659     }
 660 }
 661
 662 static void shader_arb_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
 663 {
 664     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 665     struct wined3d_context *context = context_get_current();
 666
 667     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 668      * context. On a context switch the old context will be fully dirtified */
 669     if (!context || context->swapchain->device != This) return;
 670
 671     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
 672     This->highest_dirty_vs_const = max(This->highest_dirty_vs_const, start + count);
 673 }
 674
 675 static void shader_arb_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
 676 {
 677     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
 678     struct wined3d_context *context = context_get_current();
 679
 680     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 681      * context. On a context switch the old context will be fully dirtified */
 682     if (!context || context->swapchain->device != This) return;
 683
 684     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
 685     This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, start + count);
 686 }
 687
 688 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
 689 {
 690     DWORD *ret;
 691     DWORD idx = 0;
 692     const local_constant *lconst;
 693
 694     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
 695
 696     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
 697     if(!ret) {
 698         ERR("Out of memory\n");
 699         return NULL;
 700     }
 701
 702     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 703         ret[lconst->idx] = idx++;
 704     }
 705     return ret;
 706 }
 707
 708 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 709 static DWORD shader_generate_arb_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
 710         struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
 711         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
 712 {
 713     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
 714     DWORD i, next_local = 0;
 715     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 716     unsigned max_constantsF;
 717     const local_constant *lconst;
 718     DWORD map;
 719
 720     /* In pixel shaders, all private constants are program local, we don't need anything
 721      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 722      * If we need a private constant the GL implementation will squeeze it in somewhere
 723      *
 724      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 725      * immediate values. The posFixup is loaded using program.env for now, so always
 726      * subtract one from the number of constants. If the shader uses indirect addressing,
 727      * account for the helper const too because we have to declare all availabke d3d constants
 728      * and don't know which are actually used.
 729      */
 730     if (pshader)
 731     {
 732         max_constantsF = gl_info->limits.arb_ps_native_constants;
 733         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 734         if (max_constantsF < 24)
 735             max_constantsF = gl_info->limits.arb_ps_float_constants;
 736     }
 737     else
 738     {
 739         max_constantsF = gl_info->limits.arb_vs_native_constants;
 740         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 741          * Also prevents max_constantsF from becoming less than 0 and
 742          * wrapping . */
 743         if (max_constantsF < 96)
 744             max_constantsF = gl_info->limits.arb_vs_float_constants;
 745
 746         if(This->baseShader.reg_maps.usesrelconstF) {
 747             DWORD highest_constf = 0, clip_limit;
 748
 749             max_constantsF -= reserved_vs_const(This, gl_info);
 750             max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
 751
 752             for(i = 0; i < This->baseShader.limits.constant_float; i++)
 753             {
 754                 DWORD idx = i >> 5;
 755                 DWORD shift = i & 0x1f;
 756                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 757             }
 758
 759             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 760             {
 761                 if(ctx->cur_vs_args->super.clip_enabled)
 762                     clip_limit = gl_info->limits.clipplanes;
 763                 else
 764                     clip_limit = 0;
 765             }
 766             else
 767             {
 768                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 769                 clip_limit = min(count_bits(mask), 4);
 770             }
 771             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 772             max_constantsF -= *num_clipplanes;
 773             if(*num_clipplanes < clip_limit)
 774             {
 775                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 776             }
 777         }
 778         else
 779         {
 780             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 781             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 782         }
 783     }
 784
 785     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 786     {
 787         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 788     }
 789
 790     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 791     {
 792         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 793     }
 794
 795     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 796     {
 797         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 798         {
 799             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 800         }
 801     }
 802
 803     /* Load local constants using the program-local space,
 804      * this avoids reloading them each time the shader is used
 805      */
 806     if(lconst_map) {
 807         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
 808             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 809                            lconst_map[lconst->idx]);
 810             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 811         }
 812     }
 813
 814     /* After subtracting privately used constants from the hardware limit(they are loaded as
 815      * local constants), make sure the shader doesn't violate the env constant limit
 816      */
 817     if(pshader)
 818     {
 819         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 820     }
 821     else
 822     {
 823         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 824     }
 825
 826     /* Avoid declaring more constants than needed */
 827     max_constantsF = min(max_constantsF, This->baseShader.limits.constant_float);
 828
 829     /* we use the array-based constants array if the local constants are marked for loading,
 830      * because then we use indirect addressing, or when the local constant list is empty,
 831      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 832      * local constants do not declare the loaded constants as an array because ARB compilers usually
 833      * do not optimize unused constants away
 834      */
 835     if(This->baseShader.reg_maps.usesrelconstF) {
 836         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 837         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 838                     max_constantsF, max_constantsF - 1);
 839     } else {
 840         for(i = 0; i < max_constantsF; i++) {
 841             DWORD idx, mask;
 842             idx = i >> 5;
 843             mask = 1 << (i & 0x1f);
 844             if(!shader_constant_is_local(This, i) && (This->baseShader.reg_maps.constf[idx] & mask)) {
 845                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 846             }
 847         }
 848     }
 849
 850     return next_local;
 851 }
 852
 853 static const char * const shift_tab[] = {
 854     "dummy",     /*  0 (none) */
 855     "coefmul.x", /*  1 (x2)   */
 856     "coefmul.y", /*  2 (x4)   */
 857     "coefmul.z", /*  3 (x8)   */
 858     "coefmul.w", /*  4 (x16)  */
 859     "dummy",     /*  5 (x32)  */
 860     "dummy",     /*  6 (x64)  */
 861     "dummy",     /*  7 (x128) */
 862     "dummy",     /*  8 (d256) */
 863     "dummy",     /*  9 (d128) */
 864     "dummy",     /* 10 (d64)  */
 865     "dummy",     /* 11 (d32)  */
 866     "coefdiv.w", /* 12 (d16)  */
 867     "coefdiv.z", /* 13 (d8)   */
 868     "coefdiv.y", /* 14 (d4)   */
 869     "coefdiv.x"  /* 15 (d2)   */
 870 };
 871
 872 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 873         const struct wined3d_shader_dst_param *dst, char *write_mask)
 874 {
 875     char *ptr = write_mask;
 876
 877     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 878     {
 879         *ptr++ = '.';
 880         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 881         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 882         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 883         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 884     }
 885
 886     *ptr = '\0';
 887 }
 888
 889 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 890 {
 891     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 892      * but addressed as "rgba". To fix this we need to swap the register's x
 893      * and z components. */
 894     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 895     char *ptr = swizzle_str;
 896
 897     /* swizzle bits fields: wwzzyyxx */
 898     DWORD swizzle = param->swizzle;
 899     DWORD swizzle_x = swizzle & 0x03;
 900     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 901     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 902     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 903
 904     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 905      * generate a swizzle string. Unless we need to our own swizzling. */
 906     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 907     {
 908         *ptr++ = '.';
 909         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 910             *ptr++ = swizzle_chars[swizzle_x];
 911         } else {
 912             *ptr++ = swizzle_chars[swizzle_x];
 913             *ptr++ = swizzle_chars[swizzle_y];
 914             *ptr++ = swizzle_chars[swizzle_z];
 915             *ptr++ = swizzle_chars[swizzle_w];
 916         }
 917     }
 918
 919     *ptr = '\0';
 920 }
 921
 922 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 923 {
 924     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 925     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 926
 927     if (!strcmp(priv->addr_reg, src)) return;
 928
 929     strcpy(priv->addr_reg, src);
 930     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 931 }
 932
 933 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 934         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 935
 936 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 937         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 938 {
 939     /* oPos, oFog and oPts in D3D */
 940     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 941     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
 942     BOOL pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
 943     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 944
 945     *is_color = FALSE;
 946
 947     switch (reg->type)
 948     {
 949         case WINED3DSPR_TEMP:
 950             sprintf(register_name, "R%u", reg->idx);
 951             break;
 952
 953         case WINED3DSPR_INPUT:
 954             if (pshader)
 955             {
 956                 if(This->baseShader.reg_maps.shader_version.major < 3)
 957                 {
 958                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 959                     else strcpy(register_name, "fragment.color.secondary");
 960                 }
 961                 else
 962                 {
 963                     if(reg->rel_addr)
 964                     {
 965                         char rel_reg[50];
 966                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 967
 968                         if (!strcmp(rel_reg, "**aL_emul**"))
 969                         {
 970                             DWORD idx = ctx->aL + reg->idx;
 971                             if(idx < MAX_REG_INPUT)
 972                             {
 973                                 strcpy(register_name, ctx->ps_input[idx]);
 974                             }
 975                             else
 976                             {
 977                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
 978                                 sprintf(register_name, "out_of_bounds_%u", idx);
 979                             }
 980                         }
 981                         else if(This->baseShader.reg_maps.input_registers & 0x0300)
 982                         {
 983                             /* There are two ways basically:
 984                              *
 985                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
 986                              *    That means trouble if the loop also contains a breakc or if the control values
 987                              *    aren't local constants.
 988                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
 989                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
 990                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
 991                              *    ADAC to load the condition code register and pop it again afterwards
 992                              */
 993                             FIXME("Relative input register addressing with more than 8 registers\n");
 994
 995                             /* This is better than nothing for now */
 996                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
 997                         }
 998                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
 999                         {
1000                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1001                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1002                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1003                              * proper varyings, or loop unrolling
1004                              *
1005                              * For now use the texcoords and hope for the best
1006                              */
1007                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1008                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1009                         }
1010                         else
1011                         {
1012                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1013                              * pulls GL_NV_fragment_program2 in
1014                              */
1015                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1016                         }
1017                     }
1018                     else
1019                     {
1020                         if(reg->idx < MAX_REG_INPUT)
1021                         {
1022                             strcpy(register_name, ctx->ps_input[reg->idx]);
1023                         }
1024                         else
1025                         {
1026                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1027                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1028                         }
1029                     }
1030                 }
1031             }
1032             else
1033             {
1034                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1035                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1036             }
1037             break;
1038
1039         case WINED3DSPR_CONST:
1040             if (!pshader && reg->rel_addr)
1041             {
1042                 BOOL aL = FALSE;
1043                 char rel_reg[50];
1044                 UINT rel_offset = ((IWineD3DVertexShaderImpl *)This)->rel_offset;
1045                 if(This->baseShader.reg_maps.shader_version.major < 2) {
1046                     sprintf(rel_reg, "A0.x");
1047                 } else {
1048                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1049                     if(ctx->target_version == ARB) {
1050                         if (!strcmp(rel_reg, "**aL_emul**"))
1051                         {
1052                             aL = TRUE;
1053                         } else {
1054                             shader_arb_request_a0(ins, rel_reg);
1055                             sprintf(rel_reg, "A0.x");
1056                         }
1057                     }
1058                 }
1059                 if(aL)
1060                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1061                 else if (reg->idx >= rel_offset)
1062                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1063                 else
1064                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1065             }
1066             else
1067             {
1068                 if (This->baseShader.reg_maps.usesrelconstF)
1069                     sprintf(register_name, "C[%u]", reg->idx);
1070                 else
1071                     sprintf(register_name, "C%u", reg->idx);
1072             }
1073             break;
1074
1075         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1076             if (pshader) {
1077                 if(This->baseShader.reg_maps.shader_version.major == 1 &&
1078                    This->baseShader.reg_maps.shader_version.minor <= 3) {
1079                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1080                      * and as source to most instructions. For some instructions it is the texcoord
1081                      * input. Those instructions know about the special use
1082                      */
1083                     sprintf(register_name, "T%u", reg->idx);
1084                 } else {
1085                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1086                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1087                 }
1088             }
1089             else
1090             {
1091                 if(This->baseShader.reg_maps.shader_version.major == 1 || ctx->target_version >= NV2)
1092                 {
1093                     sprintf(register_name, "A%u", reg->idx);
1094                 }
1095                 else
1096                 {
1097                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1098                 }
1099             }
1100             break;
1101
1102         case WINED3DSPR_COLOROUT:
1103             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1104             {
1105                 strcpy(register_name, "TMP_COLOR");
1106             }
1107             else
1108             {
1109                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1110                 if(This->baseShader.reg_maps.highest_render_target > 0)
1111                 {
1112                     sprintf(register_name, "result.color[%u]", reg->idx);
1113                 }
1114                 else
1115                 {
1116                     strcpy(register_name, "result.color");
1117                 }
1118             }
1119             break;
1120
1121         case WINED3DSPR_RASTOUT:
1122             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1123             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1124             break;
1125
1126         case WINED3DSPR_DEPTHOUT:
1127             strcpy(register_name, "result.depth");
1128             break;
1129
1130         case WINED3DSPR_ATTROUT:
1131         /* case WINED3DSPR_OUTPUT: */
1132             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1133             else strcpy(register_name, ctx->color_output[reg->idx]);
1134             break;
1135
1136         case WINED3DSPR_TEXCRDOUT:
1137             if (pshader)
1138             {
1139                 sprintf(register_name, "oT[%u]", reg->idx);
1140             }
1141             else
1142             {
1143                 if(This->baseShader.reg_maps.shader_version.major < 3)
1144                 {
1145                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1146                 }
1147                 else
1148                 {
1149                     strcpy(register_name, ctx->vs_output[reg->idx]);
1150                 }
1151             }
1152             break;
1153
1154         case WINED3DSPR_LOOP:
1155             if(ctx->target_version >= NV2)
1156             {
1157                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1158                 if(pshader) sprintf(register_name, "A0.x");
1159                 else sprintf(register_name, "aL.y");
1160             }
1161             else
1162             {
1163                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1164                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1165                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1166                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1167                  * indexing
1168                  */
1169                 sprintf(register_name, "**aL_emul**");
1170             }
1171
1172             break;
1173
1174         case WINED3DSPR_CONSTINT:
1175             sprintf(register_name, "I%u", reg->idx);
1176             break;
1177
1178         case WINED3DSPR_MISCTYPE:
1179             if (!reg->idx)
1180             {
1181                 sprintf(register_name, "vpos");
1182             }
1183             else if(reg->idx == 1)
1184             {
1185                 sprintf(register_name, "fragment.facing.x");
1186             }
1187             else
1188             {
1189                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1190             }
1191             break;
1192
1193         default:
1194             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1195             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1196             break;
1197     }
1198 }
1199
1200 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1201         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1202 {
1203     char register_name[255];
1204     char write_mask[6];
1205     BOOL is_color;
1206
1207     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1208     strcpy(str, register_name);
1209
1210     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1211     strcat(str, write_mask);
1212 }
1213
1214 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1215 {
1216     switch(channel_source)
1217     {
1218         case CHANNEL_SOURCE_ZERO: return "0";
1219         case CHANNEL_SOURCE_ONE: return "1";
1220         case CHANNEL_SOURCE_X: return "x";
1221         case CHANNEL_SOURCE_Y: return "y";
1222         case CHANNEL_SOURCE_Z: return "z";
1223         case CHANNEL_SOURCE_W: return "w";
1224         default:
1225             FIXME("Unhandled channel source %#x\n", channel_source);
1226             return "undefined";
1227     }
1228 }
1229
1230 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1231         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1232 {
1233     DWORD mask;
1234
1235     if (is_complex_fixup(fixup))
1236     {
1237         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1238         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1239         return;
1240     }
1241
1242     mask = 0;
1243     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1244     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1245     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1246     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1247     mask &= dst_mask;
1248
1249     if (mask)
1250     {
1251         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1252                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1253                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1254     }
1255
1256     mask = 0;
1257     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1258     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1259     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1260     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1261     mask &= dst_mask;
1262
1263     if (mask)
1264     {
1265         char reg_mask[6];
1266         char *ptr = reg_mask;
1267
1268         if (mask != WINED3DSP_WRITEMASK_ALL)
1269         {
1270             *ptr++ = '.';
1271             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1272             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1273             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1274             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1275         }
1276         *ptr = '\0';
1277
1278         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1279     }
1280 }
1281
1282 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1283 {
1284     DWORD mod;
1285     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1286     if (!ins->dst_count) return "";
1287
1288     mod = ins->dst[0].modifiers;
1289
1290     /* Silently ignore PARTIALPRECISION if its not supported */
1291     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1292
1293     if(mod & WINED3DSPDM_MSAMPCENTROID)
1294     {
1295         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1296         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1297     }
1298
1299     switch(mod)
1300     {
1301         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1302             return "H_SAT";
1303
1304         case WINED3DSPDM_SATURATE:
1305             return "_SAT";
1306
1307         case WINED3DSPDM_PARTIALPRECISION:
1308             return "H";
1309
1310         case 0:
1311             return "";
1312
1313         default:
1314             FIXME("Unknown modifiers 0x%08x\n", mod);
1315             return "";
1316     }
1317 }
1318
1319 #define TEX_PROJ        0x1
1320 #define TEX_BIAS        0x2
1321 #define TEX_LOD         0x4
1322 #define TEX_DERIV       0x10
1323
1324 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1325         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1326 {
1327     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1328     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1329     const char *tex_type;
1330     BOOL np2_fixup = FALSE;
1331     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1332     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1333     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1334     const char *mod;
1335     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1336
1337     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1338     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1339
1340     switch(sampler_type) {
1341         case WINED3DSTT_1D:
1342             tex_type = "1D";
1343             break;
1344
1345         case WINED3DSTT_2D:
1346             if(device->stateBlock->textures[sampler_idx] &&
1347                IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
1348                 tex_type = "RECT";
1349             } else {
1350                 tex_type = "2D";
1351             }
1352             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1353             {
1354                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1355                 {
1356                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1357                     else np2_fixup = TRUE;
1358                 }
1359             }
1360             break;
1361
1362         case WINED3DSTT_VOLUME:
1363             tex_type = "3D";
1364             break;
1365
1366         case WINED3DSTT_CUBE:
1367             tex_type = "CUBE";
1368             break;
1369
1370         default:
1371             ERR("Unexpected texture type %d\n", sampler_type);
1372             tex_type = "";
1373     }
1374
1375     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1376      * so don't use shader_arb_get_modifier
1377      */
1378     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1379     else mod = "";
1380
1381     /* Fragment samplers always have indentity mapping */
1382     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1383     {
1384         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1385     }
1386
1387     if (flags & TEX_DERIV)
1388     {
1389         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1390         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1391         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1392                        dsx, dsy,sampler_idx, tex_type);
1393     }
1394     else if(flags & TEX_LOD)
1395     {
1396         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1397         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1398         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1399                        sampler_idx, tex_type);
1400     }
1401     else if (flags & TEX_BIAS)
1402     {
1403         /* Shouldn't be possible, but let's check for it */
1404         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1405         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1406         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1407     }
1408     else if (flags & TEX_PROJ)
1409     {
1410         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1411     }
1412     else
1413     {
1414         if (np2_fixup)
1415         {
1416             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1417             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1418                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1419
1420             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1421         }
1422         else
1423             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1424     }
1425
1426     if (pshader)
1427     {
1428         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1429                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1430                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1431                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1432     }
1433 }
1434
1435 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1436         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1437 {
1438     /* Generate a line that does the input modifier computation and return the input register to use */
1439     BOOL is_color = FALSE;
1440     char regstr[256];
1441     char swzstr[20];
1442     int insert_line;
1443     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1444     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1445     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1446     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1447
1448     /* Assume a new line will be added */
1449     insert_line = 1;
1450
1451     /* Get register name */
1452     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1453     shader_arb_get_swizzle(src, is_color, swzstr);
1454
1455     switch (src->modifiers)
1456     {
1457     case WINED3DSPSM_NONE:
1458         sprintf(outregstr, "%s%s", regstr, swzstr);
1459         insert_line = 0;
1460         break;
1461     case WINED3DSPSM_NEG:
1462         sprintf(outregstr, "-%s%s", regstr, swzstr);
1463         insert_line = 0;
1464         break;
1465     case WINED3DSPSM_BIAS:
1466         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1467         break;
1468     case WINED3DSPSM_BIASNEG:
1469         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1470         break;
1471     case WINED3DSPSM_SIGN:
1472         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1473         break;
1474     case WINED3DSPSM_SIGNNEG:
1475         shader_addline(buffer, "MAD T%c, %s, %s, %s;\n", 'A' + tmpreg, regstr, two, one);
1476         break;
1477     case WINED3DSPSM_COMP:
1478         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1479         break;
1480     case WINED3DSPSM_X2:
1481         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1482         break;
1483     case WINED3DSPSM_X2NEG:
1484         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1485         break;
1486     case WINED3DSPSM_DZ:
1487         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1488         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1489         break;
1490     case WINED3DSPSM_DW:
1491         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1492         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1493         break;
1494     case WINED3DSPSM_ABS:
1495         if(ctx->target_version >= NV2) {
1496             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1497             insert_line = 0;
1498         } else {
1499             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1500         }
1501         break;
1502     case WINED3DSPSM_ABSNEG:
1503         if(ctx->target_version >= NV2) {
1504             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1505         } else {
1506             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1507             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1508         }
1509         insert_line = 0;
1510         break;
1511     default:
1512         sprintf(outregstr, "%s%s", regstr, swzstr);
1513         insert_line = 0;
1514     }
1515
1516     /* Return modified or original register, with swizzle */
1517     if (insert_line)
1518         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1519 }
1520
1521 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1522 {
1523     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1524     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1525     char dst_name[50];
1526     char src_name[2][50];
1527     DWORD sampler_code = dst->reg.idx;
1528
1529     shader_arb_get_dst_param(ins, dst, dst_name);
1530
1531     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1532      *
1533      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1534      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1535      * temps is done.
1536      */
1537     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1538     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1539     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1540     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1541     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1542
1543     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1544     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1545 }
1546
1547 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1548 {
1549     *extra_char = ' ';
1550     switch(mod)
1551     {
1552         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1553         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1554         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1555         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1556         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1557         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1558         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1559         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1560         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1561         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1562         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1563         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1564         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1565     }
1566     FIXME("Unknown modifier %u\n", mod);
1567     return mod;
1568 }
1569
1570 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1571 {
1572     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1573     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1574     char dst_name[50];
1575     char src_name[3][50];
1576     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1577             ins->ctx->reg_maps->shader_version.minor);
1578     BOOL is_color;
1579
1580     shader_arb_get_dst_param(ins, dst, dst_name);
1581     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1582
1583     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1584     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1585     {
1586         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1587     } else {
1588         struct wined3d_shader_src_param src0_copy = ins->src[0];
1589         char extra_neg;
1590
1591         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1592         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1593
1594         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1595         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1596         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1597         /* No modifiers supported on CMP */
1598         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1599
1600         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1601         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1602         {
1603             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1604             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1605         }
1606     }
1607 }
1608
1609 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1610 {
1611     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1612     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1613     char dst_name[50];
1614     char src_name[3][50];
1615     BOOL is_color;
1616
1617     shader_arb_get_dst_param(ins, dst, dst_name);
1618
1619     /* Generate input register names (with modifiers) */
1620     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1621     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1622     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1623
1624     /* No modifiers are supported on CMP */
1625     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1626                    src_name[0], src_name[2], src_name[1]);
1627
1628     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1629     {
1630         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1631         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1632     }
1633 }
1634
1635 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1636  * dst = dot2(src0, src1) + src2 */
1637 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1638 {
1639     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1640     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1641     char dst_name[50];
1642     char src_name[3][50];
1643     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1644
1645     shader_arb_get_dst_param(ins, dst, dst_name);
1646     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1647     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1648
1649     if(ctx->target_version >= NV3)
1650     {
1651         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1652         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1653         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1654                        dst_name, src_name[0], src_name[1], src_name[2]);
1655     }
1656     else if(ctx->target_version >= NV2)
1657     {
1658         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1659          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1660          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1661          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1662          *
1663          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1664          *
1665          * .xyxy and other swizzles that we could get with this are not valid in
1666          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1667          */
1668         struct wined3d_shader_src_param tmp_param = ins->src[1];
1669         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1670         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1671
1672         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1673
1674         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1675                        dst_name, src_name[2], src_name[0], src_name[1]);
1676     }
1677     else
1678     {
1679         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1680         /* Emulate a DP2 with a DP3 and 0.0. Don't use the dest as temp register, it could be src[1] or src[2]
1681         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1682         */
1683         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1684         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1685         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1686         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1687     }
1688 }
1689
1690 /* Map the opcode 1-to-1 to the GL code */
1691 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1692 {
1693     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1694     const char *instruction;
1695     char arguments[256], dst_str[50];
1696     unsigned int i;
1697     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1698
1699     switch (ins->handler_idx)
1700     {
1701         case WINED3DSIH_ABS: instruction = "ABS"; break;
1702         case WINED3DSIH_ADD: instruction = "ADD"; break;
1703         case WINED3DSIH_CRS: instruction = "XPD"; break;
1704         case WINED3DSIH_DP3: instruction = "DP3"; break;
1705         case WINED3DSIH_DP4: instruction = "DP4"; break;
1706         case WINED3DSIH_DST: instruction = "DST"; break;
1707         case WINED3DSIH_FRC: instruction = "FRC"; break;
1708         case WINED3DSIH_LIT: instruction = "LIT"; break;
1709         case WINED3DSIH_LRP: instruction = "LRP"; break;
1710         case WINED3DSIH_MAD: instruction = "MAD"; break;
1711         case WINED3DSIH_MAX: instruction = "MAX"; break;
1712         case WINED3DSIH_MIN: instruction = "MIN"; break;
1713         case WINED3DSIH_MOV: instruction = "MOV"; break;
1714         case WINED3DSIH_MUL: instruction = "MUL"; break;
1715         case WINED3DSIH_SGE: instruction = "SGE"; break;
1716         case WINED3DSIH_SLT: instruction = "SLT"; break;
1717         case WINED3DSIH_SUB: instruction = "SUB"; break;
1718         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1719         case WINED3DSIH_DSX: instruction = "DDX"; break;
1720         default: instruction = "";
1721             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1722             break;
1723     }
1724
1725     /* Note that shader_arb_add_dst_param() adds spaces. */
1726     arguments[0] = '\0';
1727     shader_arb_get_dst_param(ins, dst, dst_str);
1728     for (i = 0; i < ins->src_count; ++i)
1729     {
1730         char operand[100];
1731         strcat(arguments, ", ");
1732         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1733         strcat(arguments, operand);
1734     }
1735     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1736 }
1737
1738 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1739 {
1740     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1741     shader_addline(buffer, "NOP;\n");
1742 }
1743
1744 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1745 {
1746     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1747     BOOL pshader = shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type);
1748     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1749     const char *zero = arb_get_helper_value(shader->baseShader.reg_maps.shader_version.type, ARB_ZERO);
1750     const char *one = arb_get_helper_value(shader->baseShader.reg_maps.shader_version.type, ARB_ONE);
1751     const char *two = arb_get_helper_value(shader->baseShader.reg_maps.shader_version.type, ARB_TWO);
1752
1753     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1754     char src0_param[256];
1755
1756     if(ins->handler_idx == WINED3DSIH_MOVA) {
1757         char write_mask[6];
1758         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1759
1760         if(ctx->target_version >= NV2) {
1761             shader_hw_map2gl(ins);
1762             return;
1763         }
1764         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1765         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1766
1767         /* This implements the mova formula used in GLSL. The first two instructions
1768          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1769          * in this case:
1770          * mova A0.x, 0.0
1771          *
1772          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1773          *
1774          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1775          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1776          */
1777         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1778         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1779
1780         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1781         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1782         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1783         if (((IWineD3DVertexShaderImpl *)shader)->rel_offset)
1784         {
1785             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1786         }
1787         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1788
1789         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1790     } else if (ins->ctx->reg_maps->shader_version.major == 1
1791           && !shader_is_pshader_version(ins->ctx->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     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1884     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
1885     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1886     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1887             ins->ctx->reg_maps->shader_version.minor);
1888     struct wined3d_shader_src_param src;
1889
1890     char reg_dest[40];
1891     char reg_coord[40];
1892     DWORD reg_sampler_code;
1893     WORD myflags = 0;
1894
1895     /* All versions have a destination register */
1896     shader_arb_get_dst_param(ins, dst, reg_dest);
1897
1898     /* 1.0-1.4: Use destination register number as texture code.
1899        2.0+: Use provided sampler number as texure code. */
1900     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1901         reg_sampler_code = dst->reg.idx;
1902     else
1903         reg_sampler_code = ins->src[1].reg.idx;
1904
1905     /* 1.0-1.3: Use the texcoord varying.
1906        1.4+: Use provided coordinate source register. */
1907     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1908         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1909     else {
1910         /* TEX is the only instruction that can handle DW and DZ natively */
1911         src = ins->src[0];
1912         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1913         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1914         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1915     }
1916
1917     /* projection flag:
1918      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1919      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1920      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1921      */
1922     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1923     {
1924         DWORD flags = 0;
1925         if (reg_sampler_code < MAX_TEXTURES)
1926             flags = deviceImpl->stateBlock->state.texture_states[reg_sampler_code][WINED3DTSS_TEXTURETRANSFORMFLAGS];
1927         if (flags & WINED3DTTFF_PROJECTED)
1928             myflags |= TEX_PROJ;
1929     }
1930     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1931     {
1932         DWORD src_mod = ins->src[0].modifiers;
1933         if (src_mod == WINED3DSPSM_DZ) {
1934             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1935              * varying register, so we need a temp reg
1936              */
1937             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1938             strcpy(reg_coord, "TA");
1939             myflags |= TEX_PROJ;
1940         } else if(src_mod == WINED3DSPSM_DW) {
1941             myflags |= TEX_PROJ;
1942         }
1943     } else {
1944         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1945         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1946     }
1947     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1948 }
1949
1950 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1951 {
1952     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1953     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1954     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1955             ins->ctx->reg_maps->shader_version.minor);
1956     char dst_str[50];
1957
1958     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1959     {
1960         DWORD reg = dst->reg.idx;
1961
1962         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1963         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1964     } else {
1965         char reg_src[40];
1966
1967         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1968         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1969         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1970    }
1971 }
1972
1973 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1974 {
1975      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1976      IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1977      IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
1978      DWORD flags;
1979
1980      DWORD reg1 = ins->dst[0].reg.idx;
1981      char dst_str[50];
1982      char src_str[50];
1983
1984      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1985      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1986      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1987      /* Move .x first in case src_str is "TA" */
1988      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1989      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1990      flags = reg1 < MAX_TEXTURES ? deviceImpl->stateBlock->state.texture_states[reg1][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
1991      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
1992 }
1993
1994 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
1995 {
1996      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1997
1998      DWORD reg1 = ins->dst[0].reg.idx;
1999      char dst_str[50];
2000      char src_str[50];
2001
2002      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2003      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2004      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2005      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2006      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2007      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2008 }
2009
2010 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2011 {
2012     DWORD reg1 = ins->dst[0].reg.idx;
2013     char dst_str[50];
2014     char src_str[50];
2015
2016     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2017     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2018     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2019     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2020 }
2021
2022 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2023 {
2024     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2025     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)shader->baseShader.device;
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 (device->stateBlock->state.texture_states[reg_dest_code][WINED3DTSS_TEXTURETRANSFORMFLAGS]
2058             & WINED3DTTFF_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     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2098     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2099     DWORD flags;
2100     DWORD reg = ins->dst[0].reg.idx;
2101     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2102     char dst_str[50];
2103     char src0_name[50];
2104     char dst_reg[50];
2105     BOOL is_color;
2106
2107     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2108     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2109
2110     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2111     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2112     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2113     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->state.texture_states[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2114     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2115 }
2116
2117 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2118 {
2119     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2120     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2121     DWORD reg = ins->dst[0].reg.idx;
2122     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2123     char src0_name[50], dst_name[50];
2124     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2125     BOOL is_color;
2126
2127     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2128      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2129      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2130      */
2131     tmp_reg.idx = reg + 2 - current_state->current_row;
2132     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2133
2134     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2135     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2136                    dst_name, 'x' + current_state->current_row, reg, src0_name);
2137     current_state->texcoord_w[current_state->current_row++] = reg;
2138 }
2139
2140 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2141 {
2142     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2143     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2144     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2145     DWORD flags;
2146     DWORD reg = ins->dst[0].reg.idx;
2147     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2148     char dst_str[50];
2149     char src0_name[50], dst_name[50];
2150     BOOL is_color;
2151
2152     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2153     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2154     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2155
2156     /* Sample the texture using the calculated coordinates */
2157     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2158     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->state.texture_states[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2159     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2160     current_state->current_row = 0;
2161 }
2162
2163 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2164 {
2165     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2166     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2167     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2168     DWORD flags;
2169     DWORD reg = ins->dst[0].reg.idx;
2170     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2171     char dst_str[50];
2172     char src0_name[50];
2173     char dst_reg[50];
2174     BOOL is_color;
2175
2176     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2177      * components for temporary data storage
2178      */
2179     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2180     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2181     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2182
2183     /* Construct the eye-ray vector from w coordinates */
2184     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", current_state->texcoord_w[0]);
2185     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", current_state->texcoord_w[1]);
2186     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2187
2188     /* Calculate reflection vector
2189      */
2190     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2191     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2192     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2193     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2194     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2195     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2196     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2197
2198     /* Sample the texture using the calculated coordinates */
2199     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2200     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->state.texture_states[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2201     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2202     current_state->current_row = 0;
2203 }
2204
2205 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2206 {
2207     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2208     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
2209     SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
2210     DWORD flags;
2211     DWORD reg = ins->dst[0].reg.idx;
2212     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2213     char dst_str[50];
2214     char src0_name[50];
2215     char src1_name[50];
2216     char dst_reg[50];
2217     BOOL is_color;
2218
2219     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2220     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2221     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2222     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2223     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2224
2225     /* Calculate reflection vector.
2226      *
2227      *                   dot(N, E)
2228      * dst_reg.xyz = 2 * --------- * N - E
2229      *                   dot(N, N)
2230      *
2231      * Which normalizes the normal vector
2232      */
2233     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2234     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2235     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2236     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2237     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2238     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2239
2240     /* Sample the texture using the calculated coordinates */
2241     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2242     flags = reg < MAX_TEXTURES ? deviceImpl->stateBlock->state.texture_states[reg][WINED3DTSS_TEXTURETRANSFORMFLAGS] : 0;
2243     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2244     current_state->current_row = 0;
2245 }
2246
2247 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2248 {
2249     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2250     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2251     char dst_name[50];
2252     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2253     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2254
2255     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2256      * which is essentially an input, is the destination register because it is the first
2257      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2258      * here(writemasks/swizzles are not valid on texdepth)
2259      */
2260     shader_arb_get_dst_param(ins, dst, dst_name);
2261
2262     /* According to the msdn, the source register(must be r5) is unusable after
2263      * the texdepth instruction, so we're free to modify it
2264      */
2265     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2266
2267     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2268      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2269      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2270      */
2271     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2272     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2273     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2274     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2275 }
2276
2277 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2278  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2279  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2280 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2281 {
2282     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2283     DWORD sampler_idx = ins->dst[0].reg.idx;
2284     char src0[50];
2285     char dst_str[50];
2286
2287     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2288     shader_addline(buffer, "MOV TB, 0.0;\n");
2289     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2290
2291     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2292     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2293 }
2294
2295 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2296  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2297 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2298 {
2299     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2300     char src0[50];
2301     char dst_str[50];
2302     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2303
2304     /* Handle output register */
2305     shader_arb_get_dst_param(ins, dst, dst_str);
2306     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2307     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2308 }
2309
2310 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2311  * Perform the 3rd row of a 3x3 matrix multiply */
2312 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2313 {
2314     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2315     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2316     char dst_str[50], dst_name[50];
2317     char src0[50];
2318     BOOL is_color;
2319
2320     shader_arb_get_dst_param(ins, dst, dst_str);
2321     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2322     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2323     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2324     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2325 }
2326
2327 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2328  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2329  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2330  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2331  */
2332 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2333 {
2334     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2335     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2336     char src0[50], dst_name[50];
2337     BOOL is_color;
2338     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2339     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2340
2341     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2342     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2343     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2344
2345     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2346      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2347      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2348      */
2349     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2350     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2351     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2352     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2353 }
2354
2355 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2356     Vertex/Pixel shaders to ARB_vertex_program codes */
2357 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2358 {
2359     int i;
2360     int nComponents = 0;
2361     struct wined3d_shader_dst_param tmp_dst = {{0}};
2362     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2363     struct wined3d_shader_instruction tmp_ins;
2364
2365     memset(&tmp_ins, 0, sizeof(tmp_ins));
2366
2367     /* Set constants for the temporary argument */
2368     tmp_ins.ctx = ins->ctx;
2369     tmp_ins.dst_count = 1;
2370     tmp_ins.dst = &tmp_dst;
2371     tmp_ins.src_count = 2;
2372     tmp_ins.src = tmp_src;
2373
2374     switch(ins->handler_idx)
2375     {
2376         case WINED3DSIH_M4x4:
2377             nComponents = 4;
2378             tmp_ins.handler_idx = WINED3DSIH_DP4;
2379             break;
2380         case WINED3DSIH_M4x3:
2381             nComponents = 3;
2382             tmp_ins.handler_idx = WINED3DSIH_DP4;
2383             break;
2384         case WINED3DSIH_M3x4:
2385             nComponents = 4;
2386             tmp_ins.handler_idx = WINED3DSIH_DP3;
2387             break;
2388         case WINED3DSIH_M3x3:
2389             nComponents = 3;
2390             tmp_ins.handler_idx = WINED3DSIH_DP3;
2391             break;
2392         case WINED3DSIH_M3x2:
2393             nComponents = 2;
2394             tmp_ins.handler_idx = WINED3DSIH_DP3;
2395             break;
2396         default:
2397             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2398             break;
2399     }
2400
2401     tmp_dst = ins->dst[0];
2402     tmp_src[0] = ins->src[0];
2403     tmp_src[1] = ins->src[1];
2404     for (i = 0; i < nComponents; i++) {
2405         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2406         shader_hw_map2gl(&tmp_ins);
2407         ++tmp_src[1].reg.idx;
2408     }
2409 }
2410
2411 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2412 {
2413     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2414     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2415     const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2416
2417     char dst[50];
2418     char src[50];
2419
2420     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2421     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2422     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2423     {
2424         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2425          * .w is used
2426          */
2427         strcat(src, ".w");
2428     }
2429
2430     /* TODO: If the destination is readable, and not the same as the source, the destination
2431      * can be used instead of TA
2432      */
2433     if (priv->target_version >= NV2)
2434     {
2435         shader_addline(buffer, "MOVC TA.x, %s;\n", src);
2436         shader_addline(buffer, "MOV TA.x (EQ.x), %s;\n", flt_eps);
2437         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2438     }
2439     else
2440     {
2441         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2442         shader_addline(buffer, "ABS TA.x, %s;\n", src);
2443         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2444         shader_addline(buffer, "MAD TA.x, TA.y, %s, %s;\n", flt_eps, src);
2445         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2446     }
2447 }
2448
2449 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2450 {
2451     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2452     const char *instruction;
2453
2454     char dst[50];
2455     char src[50];
2456
2457     switch(ins->handler_idx)
2458     {
2459         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2460         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2461         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2462         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2463         default: instruction = "";
2464             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2465             break;
2466     }
2467
2468     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2469     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2470     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2471     {
2472         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2473          * .w is used
2474          */
2475         strcat(src, ".w");
2476     }
2477
2478     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2479 }
2480
2481 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2482 {
2483     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2484     char dst_name[50];
2485     char src_name[50];
2486     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2487     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2488     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2489
2490     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2491     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2492
2493     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2494      * otherwise NRM or RSQ would return NaN */
2495     if(pshader && priv->target_version >= NV3)
2496     {
2497         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2498          *
2499          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2500          */
2501         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2502         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2503         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2504     }
2505     else if(priv->target_version >= NV2)
2506     {
2507         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2508         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2509         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2510                        src_name);
2511     }
2512     else
2513     {
2514         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2515
2516         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2517         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2518          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2519          */
2520         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2521         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2522
2523         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2524         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2525         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2526                     src_name);
2527     }
2528 }
2529
2530 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2531 {
2532     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2533     char dst_name[50];
2534     char src_name[3][50];
2535
2536     /* ARB_fragment_program has a convenient LRP instruction */
2537     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2538         shader_hw_map2gl(ins);
2539         return;
2540     }
2541
2542     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2543     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2544     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2545     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2546
2547     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2548     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2549                    dst_name, src_name[0], src_name[2]);
2550 }
2551
2552 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2553 {
2554     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2555      * must contain fixed constants. So we need a separate function to filter those constants and
2556      * can't use map2gl
2557      */
2558     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2559     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2560     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2561     char dst_name[50];
2562     char src_name0[50], src_name1[50], src_name2[50];
2563     BOOL is_color;
2564
2565     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2566     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2567         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2568         /* No modifiers are supported on SCS */
2569         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2570
2571         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2572         {
2573             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2574             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2575         }
2576     } else if(priv->target_version >= NV2) {
2577         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2578
2579         /* Sincos writemask must be .x, .y or .xy */
2580         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2581             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2582         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2583             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2584     } else {
2585         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2586          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2587          *
2588          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2589          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2590          *
2591          * The constants we get are:
2592          *
2593          *  +1   +1,     -1     -1     +1      +1      -1       -1
2594          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2595          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2596          *
2597          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2598          *
2599          * (x/2)^2 = x^2 / 4
2600          * (x/2)^3 = x^3 / 8
2601          * (x/2)^4 = x^4 / 16
2602          * (x/2)^5 = x^5 / 32
2603          * etc
2604          *
2605          * To get the final result:
2606          * sin(x) = 2 * sin(x/2) * cos(x/2)
2607          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2608          * (from sin(x+y) and cos(x+y) rules)
2609          *
2610          * As per MSDN, dst.z is undefined after the operation, and so is
2611          * dst.x and dst.y if they're masked out by the writemask. Ie
2612          * sincos dst.y, src1, c0, c1
2613          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2614          * vsa.exe also stops with an error if the dest register is the same register as the source
2615          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2616          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2617          */
2618         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2619         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2620         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2621
2622         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2623         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2624         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2625         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2626         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2627         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2628
2629         /* sin(x/2)
2630          *
2631          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2632          * properly merge that with MULs in the code above?
2633          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2634          * we can merge the sine and cosine MAD rows to calculate them together.
2635          */
2636         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2637         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2638         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2639         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2640
2641         /* cos(x/2) */
2642         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2643         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2644         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2645
2646         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2647             /* cos x */
2648             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2649             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2650         }
2651         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2652             /* sin x */
2653             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2654             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2655         }
2656     }
2657 }
2658
2659 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2660 {
2661     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2662     char dst_name[50];
2663     char src_name[50];
2664     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2665
2666     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2667     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2668
2669     /* SGN is only valid in vertex shaders */
2670     if(ctx->target_version >= NV2) {
2671         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2672         return;
2673     }
2674
2675     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2676      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2677      */
2678     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2679         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2680     } else {
2681         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2682          * Then use TA, and calculate the final result
2683          *
2684          * Not reading from TA? Store the first result in TA to avoid overwriting the
2685          * destination if src reg = dst reg
2686          */
2687         if(strstr(src_name, "TA"))
2688         {
2689             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2690             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2691             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2692         }
2693         else
2694         {
2695             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2696             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2697             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2698         }
2699     }
2700 }
2701
2702 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2703 {
2704     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2705     char src[50];
2706     char dst[50];
2707     char dst_name[50];
2708     BOOL is_color;
2709
2710     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2711     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2712     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2713
2714     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2715     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2716 }
2717
2718 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2719 {
2720     *need_abs = FALSE;
2721
2722     switch(mod)
2723     {
2724         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2725         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2726         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2727         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2728         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2729         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2730         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2731         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2732         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2733         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2734         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2735         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2736         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2737     }
2738     FIXME("Unknown modifier %u\n", mod);
2739     return mod;
2740 }
2741
2742 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2743 {
2744     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2745     char src0[50], src1[50], dst[50];
2746     struct wined3d_shader_src_param src0_copy = ins->src[0];
2747     BOOL need_abs = FALSE;
2748     const char *instr;
2749     BOOL arg2 = FALSE;
2750
2751     switch(ins->handler_idx)
2752     {
2753         case WINED3DSIH_LOG:  instr = "LG2"; break;
2754         case WINED3DSIH_LOGP: instr = "LOG"; break;
2755         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2756         default:
2757             ERR("Unexpected instruction %d\n", ins->handler_idx);
2758             return;
2759     }
2760
2761     /* LOG, LOGP and POW operate on the absolute value of the input */
2762     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2763
2764     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2765     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2766     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2767
2768     if(need_abs)
2769     {
2770         shader_addline(buffer, "ABS TA, %s;\n", src0);
2771         if(arg2)
2772         {
2773             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2774         }
2775         else
2776         {
2777             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2778         }
2779     }
2780     else if(arg2)
2781     {
2782         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2783     }
2784     else
2785     {
2786         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2787     }
2788 }
2789
2790 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2791 {
2792     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2793     char src_name[50];
2794     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2795
2796     /* src0 is aL */
2797     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2798
2799     if(vshader)
2800     {
2801         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2802         struct list *e = list_head(&priv->control_frames);
2803         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2804
2805         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2806         /* The constant loader makes sure to load -1 into iX.w */
2807         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2808         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2809         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2810     }
2811     else
2812     {
2813         shader_addline(buffer, "LOOP %s;\n", src_name);
2814     }
2815 }
2816
2817 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2818 {
2819     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2820     char src_name[50];
2821     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2822
2823     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2824
2825     /* The constant loader makes sure to load -1 into iX.w */
2826     if(vshader)
2827     {
2828         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2829         struct list *e = list_head(&priv->control_frames);
2830         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2831
2832         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2833
2834         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2835         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2836         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2837     }
2838     else
2839     {
2840         shader_addline(buffer, "REP %s;\n", src_name);
2841     }
2842 }
2843
2844 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2845 {
2846     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2847     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2848
2849     if(vshader)
2850     {
2851         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2852         struct list *e = list_head(&priv->control_frames);
2853         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2854
2855         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2856         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2857         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2858
2859         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2860     }
2861     else
2862     {
2863         shader_addline(buffer, "ENDLOOP;\n");
2864     }
2865 }
2866
2867 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2868 {
2869     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2870     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2871
2872     if(vshader)
2873     {
2874         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2875         struct list *e = list_head(&priv->control_frames);
2876         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2877
2878         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2879         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2880         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2881
2882         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2883     }
2884     else
2885     {
2886         shader_addline(buffer, "ENDREP;\n");
2887     }
2888 }
2889
2890 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2891 {
2892     struct control_frame *control_frame;
2893
2894     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2895     {
2896         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2897     }
2898     ERR("Could not find loop for break\n");
2899     return NULL;
2900 }
2901
2902 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2903 {
2904     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2905     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2906     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2907
2908     if(vshader)
2909     {
2910         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2911     }
2912     else
2913     {
2914         shader_addline(buffer, "BRK;\n");
2915     }
2916 }
2917
2918 static const char *get_compare(COMPARISON_TYPE flags)
2919 {
2920     switch (flags)
2921     {
2922         case COMPARISON_GT: return "GT";
2923         case COMPARISON_EQ: return "EQ";
2924         case COMPARISON_GE: return "GE";
2925         case COMPARISON_LT: return "LT";
2926         case COMPARISON_NE: return "NE";
2927         case COMPARISON_LE: return "LE";
2928         default:
2929             FIXME("Unrecognized comparison value: %u\n", flags);
2930             return "(\?\?)";
2931     }
2932 }
2933
2934 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2935 {
2936     switch (flags)
2937     {
2938         case COMPARISON_GT: return COMPARISON_LE;
2939         case COMPARISON_EQ: return COMPARISON_NE;
2940         case COMPARISON_GE: return COMPARISON_LT;
2941         case COMPARISON_LT: return COMPARISON_GE;
2942         case COMPARISON_NE: return COMPARISON_EQ;
2943         case COMPARISON_LE: return COMPARISON_GT;
2944         default:
2945             FIXME("Unrecognized comparison value: %u\n", flags);
2946             return -1;
2947     }
2948 }
2949
2950 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2951 {
2952     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2953     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2954     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2955     char src_name0[50];
2956     char src_name1[50];
2957     const char *comp = get_compare(ins->flags);
2958
2959     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2960     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2961
2962     if(vshader)
2963     {
2964         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2965          * away the subtraction result
2966          */
2967         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2968         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2969     }
2970     else
2971     {
2972         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2973         shader_addline(buffer, "BRK (%s.x);\n", comp);
2974     }
2975 }
2976
2977 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2978 {
2979     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2980     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2981     struct list *e = list_head(&priv->control_frames);
2982     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2983     const char *comp;
2984     char src_name0[50];
2985     char src_name1[50];
2986     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2987
2988     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2989     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2990
2991     if(vshader)
2992     {
2993         /* Invert the flag. We jump to the else label if the condition is NOT true */
2994         comp = get_compare(invert_compare(ins->flags));
2995         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2996         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
2997     }
2998     else
2999     {
3000         comp = get_compare(ins->flags);
3001         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3002         shader_addline(buffer, "IF %s.x;\n", comp);
3003     }
3004 }
3005
3006 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3007 {
3008     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3009     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3010     struct list *e = list_head(&priv->control_frames);
3011     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3012     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3013
3014     if(vshader)
3015     {
3016         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3017         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3018         control_frame->had_else = TRUE;
3019     }
3020     else
3021     {
3022         shader_addline(buffer, "ELSE;\n");
3023     }
3024 }
3025
3026 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3027 {
3028     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3029     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3030     struct list *e = list_head(&priv->control_frames);
3031     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3032     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3033
3034     if(vshader)
3035     {
3036         if(control_frame->had_else)
3037         {
3038             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3039         }
3040         else
3041         {
3042             shader_addline(buffer, "#No else branch. else is endif\n");
3043             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3044         }
3045     }
3046     else
3047     {
3048         shader_addline(buffer, "ENDIF;\n");
3049     }
3050 }
3051
3052 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3053 {
3054     DWORD sampler_idx = ins->src[1].reg.idx;
3055     char reg_dest[40];
3056     char reg_src[3][40];
3057     WORD flags = TEX_DERIV;
3058
3059     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3060     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3061     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3062     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3063
3064     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3065     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3066
3067     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3068 }
3069
3070 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3071 {
3072     DWORD sampler_idx = ins->src[1].reg.idx;
3073     char reg_dest[40];
3074     char reg_coord[40];
3075     WORD flags = TEX_LOD;
3076
3077     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3078     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3079
3080     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3081     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3082
3083     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3084 }
3085
3086 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3087 {
3088     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3089     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3090
3091     priv->in_main_func = FALSE;
3092     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3093      * subroutine, don't generate a label that will make GL complain
3094      */
3095     if(priv->target_version == ARB) return;
3096
3097     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3098 }
3099
3100 static void vshader_add_footer(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
3101         const struct arb_vs_compile_args *args, struct shader_arb_ctx_priv *priv_ctx)
3102 {
3103     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3104     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3105     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
3106     unsigned int i;
3107
3108     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3109      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3110      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3111      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3112      */
3113     if(args->super.fog_src == VS_FOG_Z) {
3114         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3115     } else if (!reg_maps->fog) {
3116         /* posFixup.x is always 1.0, so we can savely use it */
3117         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3118     }
3119
3120     /* Write the final position.
3121      *
3122      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3123      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3124      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3125      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3126      */
3127     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3128     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3129     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3130
3131     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3132     {
3133         if(args->super.clip_enabled)
3134         {
3135             for(i = 0; i < priv_ctx->vs_clipplanes; i++)
3136             {
3137                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3138             }
3139         }
3140     }
3141     else if(args->clip.boolclip.clip_texcoord)
3142     {
3143         unsigned int cur_clip = 0;
3144         char component[4] = {'x', 'y', 'z', 'w'};
3145         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3146
3147         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3148         {
3149             if(args->clip.boolclip.clipplane_mask & (1 << i))
3150             {
3151                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3152                                component[cur_clip++], i);
3153             }
3154         }
3155         switch(cur_clip)
3156         {
3157             case 0:
3158                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3159                 break;
3160             case 1:
3161                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3162                 break;
3163             case 2:
3164                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3165                 break;
3166             case 3:
3167                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3168                 break;
3169         }
3170         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3171                        args->clip.boolclip.clip_texcoord - 1);
3172     }
3173
3174     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3175      * and the glsl equivalent
3176      */
3177     if(need_helper_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
3178         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3179         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3180     } else {
3181         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3182         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3183     }
3184
3185     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3186
3187     priv_ctx->footer_written = TRUE;
3188 }
3189
3190 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3191 {
3192     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3193     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3194     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *) ins->ctx->shader;
3195     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3196
3197     if(priv->target_version == ARB) return;
3198
3199     if(vshader)
3200     {
3201         if(priv->in_main_func) vshader_add_footer((IWineD3DVertexShaderImpl *) shader, buffer, priv->cur_vs_args, priv);
3202     }
3203
3204     shader_addline(buffer, "RET;\n");
3205 }
3206
3207 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3208 {
3209     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3210     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3211 }
3212
3213 /* GL locking is done by the caller */
3214 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3215 {
3216     GLuint program_id = 0;
3217     GLint pos;
3218
3219     const char *blt_vprogram =
3220         "!!ARBvp1.0\n"
3221         "PARAM c[1] = { { 1, 0.5 } };\n"
3222         "MOV result.position, vertex.position;\n"
3223         "MOV result.color, c[0].x;\n"
3224         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3225         "END\n";
3226
3227     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3228     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3229     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3230             strlen(blt_vprogram), blt_vprogram));
3231     checkGLcall("glProgramStringARB()");
3232
3233     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3234     if (pos != -1)
3235     {
3236         FIXME("Vertex program error at position %d: %s\n\n", pos,
3237             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3238         shader_arb_dump_program_source(blt_vprogram);
3239     }
3240     else
3241     {
3242         GLint native;
3243
3244         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3245         checkGLcall("glGetProgramivARB()");
3246         if (!native) WARN("Program exceeds native resource limits.\n");
3247     }
3248
3249     return program_id;
3250 }
3251
3252 /* GL locking is done by the caller */
3253 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3254         enum tex_types tex_type, BOOL masked)
3255 {
3256     GLuint program_id = 0;
3257     const char *fprogram;
3258     GLint pos;
3259
3260     static const char * const blt_fprograms_full[tex_type_count] =
3261     {
3262         /* tex_1d */
3263         NULL,
3264         /* tex_2d */
3265         "!!ARBfp1.0\n"
3266         "TEMP R0;\n"
3267         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3268         "MOV result.depth.z, R0.x;\n"
3269         "END\n",
3270         /* tex_3d */
3271         NULL,
3272         /* tex_cube */
3273         "!!ARBfp1.0\n"
3274         "TEMP R0;\n"
3275         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3276         "MOV result.depth.z, R0.x;\n"
3277         "END\n",
3278         /* tex_rect */
3279         "!!ARBfp1.0\n"
3280         "TEMP R0;\n"
3281         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3282         "MOV result.depth.z, R0.x;\n"
3283         "END\n",
3284     };
3285
3286     static const char * const blt_fprograms_masked[tex_type_count] =
3287     {
3288         /* tex_1d */
3289         NULL,
3290         /* tex_2d */
3291         "!!ARBfp1.0\n"
3292         "PARAM mask = program.local[0];\n"
3293         "TEMP R0;\n"
3294         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3295         "MUL R0.x, R0.x, R0.y;\n"
3296         "KIL -R0.x;\n"
3297         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3298         "MOV result.depth.z, R0.x;\n"
3299         "END\n",
3300         /* tex_3d */
3301         NULL,
3302         /* tex_cube */
3303         "!!ARBfp1.0\n"
3304         "PARAM mask = program.local[0];\n"
3305         "TEMP R0;\n"
3306         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3307         "MUL R0.x, R0.x, R0.y;\n"
3308         "KIL -R0.x;\n"
3309         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3310         "MOV result.depth.z, R0.x;\n"
3311         "END\n",
3312         /* tex_rect */
3313         "!!ARBfp1.0\n"
3314         "PARAM mask = program.local[0];\n"
3315         "TEMP R0;\n"
3316         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3317         "MUL R0.x, R0.x, R0.y;\n"
3318         "KIL -R0.x;\n"
3319         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3320         "MOV result.depth.z, R0.x;\n"
3321         "END\n",
3322     };
3323
3324     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3325     if (!fprogram)
3326     {
3327         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3328         tex_type = tex_2d;
3329         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3330     }
3331
3332     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3333     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3334     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3335     checkGLcall("glProgramStringARB()");
3336
3337     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3338     if (pos != -1)
3339     {
3340         FIXME("Fragment program error at position %d: %s\n\n", pos,
3341             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3342         shader_arb_dump_program_source(fprogram);
3343     }
3344     else
3345     {
3346         GLint native;
3347
3348         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3349         checkGLcall("glGetProgramivARB()");
3350         if (!native) WARN("Program exceeds native resource limits.\n");
3351     }
3352
3353     return program_id;
3354 }
3355
3356 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3357         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3358 {
3359     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3360
3361     if(condcode)
3362     {
3363         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3364         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3365         /* Calculate the > 0.0031308 case */
3366         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3367         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3368         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3369         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3370         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3371         /* Calculate the < case */
3372         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3373     }
3374     else
3375     {
3376         /* Calculate the > 0.0031308 case */
3377         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3378         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3379         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3380         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3381         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3382         /* Calculate the < case */
3383         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3384         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3385         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3386         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3387         /* Store the components > 0.0031308 in the destination */
3388         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3389         /* Add the components that are < 0.0031308 */
3390         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3391         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3392         * result.color writes(.rgb first, then .a), or handle overwriting already written
3393         * components. The assembler uses a temporary register in this case, which is usually
3394         * not allocated from one of our registers that were used earlier.
3395         */
3396     }
3397     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3398 }
3399
3400 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3401 {
3402     const local_constant *constant;
3403
3404     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3405     {
3406         if (constant->idx == idx)
3407         {
3408             return constant->value;
3409         }
3410     }
3411     return NULL;
3412 }
3413
3414 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3415                           struct shader_arb_ctx_priv *priv)
3416 {
3417     static const char * const texcoords[8] =
3418     {
3419         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3420         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3421     };
3422     unsigned int i;
3423     const struct wined3d_shader_signature_element *sig = This->baseShader.input_signature;
3424     const char *semantic_name;
3425     DWORD semantic_idx;
3426
3427     switch(args->super.vp_mode)
3428     {
3429         case pretransformed:
3430         case fixedfunction:
3431             /* The pixelshader has to collect the varyings on its own. In any case properly load
3432              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3433              * other attribs to 0.0.
3434              *
3435              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3436              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3437              * load the texcoord attrib pointers to match the pixel shader signature
3438              */
3439             for(i = 0; i < MAX_REG_INPUT; i++)
3440             {
3441                 semantic_name = sig[i].semantic_name;
3442                 semantic_idx = sig[i].semantic_idx;
3443                 if (!semantic_name) continue;
3444
3445                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3446                 {
3447                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3448                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3449                     else priv->ps_input[i] = "0.0";
3450                 }
3451                 else if(args->super.vp_mode == fixedfunction)
3452                 {
3453                     priv->ps_input[i] = "0.0";
3454                 }
3455                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3456                 {
3457                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3458                     else priv->ps_input[i] = "0.0";
3459                 }
3460                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3461                 {
3462                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3463                     else priv->ps_input[i] = "0.0";
3464                 }
3465                 else
3466                 {
3467                     priv->ps_input[i] = "0.0";
3468                 }
3469
3470                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3471             }
3472             break;
3473
3474         case vertexshader:
3475             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3476              * fragment.color
3477              */
3478             for(i = 0; i < 8; i++)
3479             {
3480                 priv->ps_input[i] = texcoords[i];
3481             }
3482             priv->ps_input[8] = "fragment.color.primary";
3483             priv->ps_input[9] = "fragment.color.secondary";
3484             break;
3485     }
3486 }
3487
3488 /* GL locking is done by the caller */
3489 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This, struct wined3d_shader_buffer *buffer,
3490         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3491 {
3492     const shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3493     CONST DWORD *function = This->baseShader.function;
3494     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3495     const local_constant *lconst;
3496     GLuint retval;
3497     char fragcolor[16];
3498     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3499     struct shader_arb_ctx_priv priv_ctx;
3500     BOOL dcl_td = FALSE;
3501     BOOL want_nv_prog = FALSE;
3502     struct arb_pshader_private *shader_priv = This->baseShader.backend_data;
3503     GLint errPos;
3504     DWORD map;
3505
3506     char srgbtmp[4][4];
3507     unsigned int i, found = 0;
3508
3509     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3510     {
3511         if (!(map & 1)
3512                 || (This->color0_mov && i == This->color0_reg)
3513                 || (reg_maps->shader_version.major < 2 && !i))
3514             continue;
3515
3516         sprintf(srgbtmp[found], "R%u", i);
3517         ++found;
3518         if (found == 4) break;
3519     }
3520
3521     switch(found) {
3522         case 0:
3523             sprintf(srgbtmp[0], "TA");
3524             sprintf(srgbtmp[1], "TB");
3525             sprintf(srgbtmp[2], "TC");
3526             sprintf(srgbtmp[3], "TD");
3527             dcl_td = TRUE;
3528             break;
3529         case 1:
3530             sprintf(srgbtmp[1], "TA");
3531             sprintf(srgbtmp[2], "TB");
3532             sprintf(srgbtmp[3], "TC");
3533             break;
3534         case 2:
3535             sprintf(srgbtmp[2], "TA");
3536             sprintf(srgbtmp[3], "TB");
3537             break;
3538         case 3:
3539             sprintf(srgbtmp[3], "TA");
3540             break;
3541         case 4:
3542             break;
3543     }
3544
3545     /*  Create the hw ARB shader */
3546     memset(&priv_ctx, 0, sizeof(priv_ctx));
3547     priv_ctx.cur_ps_args = args;
3548     priv_ctx.compiled_fprog = compiled;
3549     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3550     init_ps_input(This, args, &priv_ctx);
3551     list_init(&priv_ctx.control_frames);
3552
3553     /* Avoid enabling NV_fragment_program* if we do not need it.
3554      *
3555      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3556      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3557      * is faster than what we gain from using higher native instructions. There are some things though
3558      * that cannot be emulated. In that case enable the extensions.
3559      * If the extension is enabled, instruction handlers that support both ways will use it.
3560      *
3561      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3562      * So enable the best we can get.
3563      */
3564     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3565        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3566     {
3567         want_nv_prog = TRUE;
3568     }
3569
3570     shader_addline(buffer, "!!ARBfp1.0\n");
3571     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3572     {
3573         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3574         priv_ctx.target_version = NV3;
3575     }
3576     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3577     {
3578         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3579         priv_ctx.target_version = NV2;
3580     } else {
3581         if(want_nv_prog)
3582         {
3583             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3584              * limits properly
3585              */
3586             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3587             ERR("Try GLSL\n");
3588         }
3589         priv_ctx.target_version = ARB;
3590     }
3591
3592     if(This->baseShader.reg_maps.highest_render_target > 0)
3593     {
3594         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3595     }
3596
3597     if (reg_maps->shader_version.major < 3)
3598     {
3599         switch(args->super.fog) {
3600             case FOG_OFF:
3601                 break;
3602             case FOG_LINEAR:
3603                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3604                 break;
3605             case FOG_EXP:
3606                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3607                 break;
3608             case FOG_EXP2:
3609                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3610                 break;
3611         }
3612     }
3613
3614     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3615      * unused temps away(but occupies them for the whole shader if they're used once). Always
3616      * declaring them avoids tricky bookkeeping work
3617      */
3618     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3619     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3620     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3621     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3622     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3623     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3624     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3625
3626     if (reg_maps->shader_version.major < 2)
3627     {
3628         strcpy(fragcolor, "R0");
3629     } else {
3630         if(args->super.srgb_correction) {
3631             if(This->color0_mov) {
3632                 sprintf(fragcolor, "R%u", This->color0_reg);
3633             } else {
3634                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3635                 strcpy(fragcolor, "TMP_COLOR");
3636             }
3637         } else {
3638             strcpy(fragcolor, "result.color");
3639         }
3640     }
3641
3642     if(args->super.srgb_correction) {
3643         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3644                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3645         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3646                        srgb_sub_high, 0.0, 0.0, 0.0);
3647     }
3648
3649     /* Base Declarations */
3650     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
3651             reg_maps, buffer, gl_info, lconst_map, NULL, &priv_ctx);
3652
3653     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3654     {
3655         if (!(map & 1)) continue;
3656
3657         cur = compiled->numbumpenvmatconsts;
3658         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3659         compiled->bumpenvmatconst[cur].texunit = i;
3660         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3661         compiled->luminanceconst[cur].texunit = i;
3662
3663         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3664          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3665          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3666          * textures due to conditional NP2 restrictions)
3667          *
3668          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3669          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3670          * their location is shader dependent anyway and they cannot be loaded globally.
3671          */
3672         compiled->bumpenvmatconst[cur].const_num = next_local++;
3673         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3674                        i, compiled->bumpenvmatconst[cur].const_num);
3675         compiled->numbumpenvmatconsts = cur + 1;
3676
3677         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3678
3679         compiled->luminanceconst[cur].const_num = next_local++;
3680         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3681                        i, compiled->luminanceconst[cur].const_num);
3682     }
3683
3684     for(i = 0; i < MAX_CONST_I; i++)
3685     {
3686         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3687         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3688         {
3689             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3690
3691             if(control_values)
3692             {
3693                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3694                                 control_values[0], control_values[1], control_values[2]);
3695             }
3696             else
3697             {
3698                 compiled->int_consts[i] = next_local;
3699                 compiled->num_int_consts++;
3700                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3701             }
3702         }
3703     }
3704
3705     if(reg_maps->vpos || reg_maps->usesdsy)
3706     {
3707         compiled->ycorrection = next_local;
3708         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3709
3710         if(reg_maps->vpos)
3711         {
3712             shader_addline(buffer, "TEMP vpos;\n");
3713             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3714              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3715              * ycorrection.z: 1.0
3716              * ycorrection.w: 0.0
3717              */
3718             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3719             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3720         }
3721     }
3722     else
3723     {
3724         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3725     }
3726
3727     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3728      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3729      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3730      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3731      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3732      * shader compilation errors and the subsequent errors when drawing with this shader. */
3733     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3734
3735         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3736         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3737         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3738
3739         fixup->offset = next_local;
3740         fixup->super.active = 0;
3741
3742         cur = 0;
3743         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3744             if (!(map & (1 << i))) continue;
3745
3746             if (fixup->offset + (cur >> 1) < max_lconsts) {
3747                 fixup->super.active |= (1 << i);
3748                 fixup->super.idx[i] = cur++;
3749             } else {
3750                 FIXME("No free constant found to load NP2 fixup data into shader. "
3751                       "Sampling from this texture will probably look wrong.\n");
3752                 break;
3753             }
3754         }
3755
3756         fixup->super.num_consts = (cur + 1) >> 1;
3757         if (fixup->super.num_consts) {
3758             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3759                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3760         }
3761
3762         next_local += fixup->super.num_consts;
3763     }
3764
3765     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3766     {
3767         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3768     }
3769
3770     /* Base Shader Body */
3771     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3772
3773     if(args->super.srgb_correction) {
3774         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3775                                   priv_ctx.target_version >= NV2);
3776     }
3777
3778     if(strcmp(fragcolor, "result.color")) {
3779         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3780     }
3781     shader_addline(buffer, "END\n");
3782
3783     /* TODO: change to resource.glObjectHandle or something like that */
3784     GL_EXTCALL(glGenProgramsARB(1, &retval));
3785
3786     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3787     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3788
3789     TRACE("Created hw pixel shader, prg=%d\n", retval);
3790     /* Create the program and check for errors */
3791     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3792                buffer->bsize, buffer->buffer));
3793     checkGLcall("glProgramStringARB()");
3794
3795     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3796     if (errPos != -1)
3797     {
3798         FIXME("HW PixelShader Error at position %d: %s\n\n",
3799               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3800         shader_arb_dump_program_source(buffer->buffer);
3801         retval = 0;
3802     }
3803     else
3804     {
3805         GLint native;
3806
3807         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3808         checkGLcall("glGetProgramivARB()");
3809         if (!native) WARN("Program exceeds native resource limits.\n");
3810     }
3811
3812     /* Load immediate constants */
3813     if(lconst_map) {
3814         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3815             const float *value = (const float *)lconst->value;
3816             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3817             checkGLcall("glProgramLocalParameter4fvARB");
3818         }
3819         HeapFree(GetProcessHeap(), 0, lconst_map);
3820     }
3821
3822     return retval;
3823 }
3824
3825 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3826 {
3827     unsigned int i;
3828     int ret;
3829
3830     for(i = 0; i < MAX_REG_INPUT; i++)
3831     {
3832         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3833         {
3834             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3835             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3836             continue;
3837         }
3838
3839         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3840         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3841         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3842         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3843         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3844         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3845     }
3846     return 0;
3847 }
3848
3849 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3850 {
3851     struct wined3d_shader_signature_element *new;
3852     int i;
3853     char *name;
3854
3855     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3856     for(i = 0; i < MAX_REG_INPUT; i++)
3857     {
3858         if (!sig[i].semantic_name) continue;
3859
3860         new[i] = sig[i];
3861         /* Clone the semantic string */
3862         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3863         strcpy(name, sig[i].semantic_name);
3864         new[i].semantic_name = name;
3865     }
3866     return new;
3867 }
3868
3869 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3870 {
3871     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3872     struct ps_signature *found_sig;
3873
3874     if (entry)
3875     {
3876         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3877         TRACE("Found existing signature %u\n", found_sig->idx);
3878         return found_sig->idx;
3879     }
3880     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3881     found_sig->sig = clone_sig(sig);
3882     found_sig->idx = priv->ps_sig_number++;
3883     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3884     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3885     {
3886         ERR("Failed to insert program entry.\n");
3887     }
3888     return found_sig->idx;
3889 }
3890
3891 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3892                                   struct arb_vs_compiled_shader *compiled)
3893 {
3894     unsigned int i, j;
3895     static const char * const texcoords[8] =
3896     {
3897         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3898         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3899     };
3900     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
3901     IWineD3DBaseShaderClass *baseshader = &shader->baseShader;
3902     const struct wined3d_shader_signature_element *sig;
3903     const char *semantic_name;
3904     DWORD semantic_idx, reg_idx;
3905
3906     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3907      * and varying 9 to result.color.secondary
3908      */
3909     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3910     {
3911         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3912         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3913         "result.color.primary", "result.color.secondary"
3914     };
3915
3916     if(sig_num == ~0)
3917     {
3918         TRACE("Pixel shader uses builtin varyings\n");
3919         /* Map builtins to builtins */
3920         for(i = 0; i < 8; i++)
3921         {
3922             priv_ctx->texcrd_output[i] = texcoords[i];
3923         }
3924         priv_ctx->color_output[0] = "result.color.primary";
3925         priv_ctx->color_output[1] = "result.color.secondary";
3926         priv_ctx->fog_output = "result.fogcoord";
3927
3928         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3929         for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
3930         {
3931             semantic_name = baseshader->output_signature[i].semantic_name;
3932             if (!semantic_name) continue;
3933
3934             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3935             {
3936                 TRACE("o%u is TMP_OUT\n", i);
3937                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3938                 else priv_ctx->vs_output[i] = "TA";
3939             }
3940             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3941             {
3942                 TRACE("o%u is result.pointsize\n", i);
3943                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3944                 else priv_ctx->vs_output[i] = "TA";
3945             }
3946             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3947             {
3948                 TRACE("o%u is result.color.?, idx %u\n", i, baseshader->output_signature[i].semantic_idx);
3949                 if (!baseshader->output_signature[i].semantic_idx)
3950                     priv_ctx->vs_output[i] = "result.color.primary";
3951                 else if (baseshader->output_signature[i].semantic_idx == 1)
3952                     priv_ctx->vs_output[i] = "result.color.secondary";
3953                 else priv_ctx->vs_output[i] = "TA";
3954             }
3955             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3956             {
3957                 TRACE("o%u is %s\n", i, texcoords[baseshader->output_signature[i].semantic_idx]);
3958                 if (baseshader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3959                 else priv_ctx->vs_output[i] = texcoords[baseshader->output_signature[i].semantic_idx];
3960             }
3961             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3962             {
3963                 TRACE("o%u is result.fogcoord\n", i);
3964                 if (baseshader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3965                 else priv_ctx->vs_output[i] = "result.fogcoord";
3966             }
3967             else
3968             {
3969                 priv_ctx->vs_output[i] = "TA";
3970             }
3971         }
3972         return;
3973     }
3974
3975     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3976      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3977      */
3978     sig = ((IWineD3DPixelShaderImpl *)device->stateBlock->pixelShader)->baseShader.input_signature;
3979     TRACE("Pixel shader uses declared varyings\n");
3980
3981     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3982     for(i = 0; i < 8; i++)
3983     {
3984         priv_ctx->texcrd_output[i] = "TA";
3985     }
3986     priv_ctx->color_output[0] = "TA";
3987     priv_ctx->color_output[1] = "TA";
3988     priv_ctx->fog_output = "TA";
3989
3990     for(i = 0; i < MAX_REG_INPUT; i++)
3991     {
3992         semantic_name = sig[i].semantic_name;
3993         semantic_idx = sig[i].semantic_idx;
3994         reg_idx = sig[i].register_idx;
3995         if (!semantic_name) continue;
3996
3997         /* If a declared input register is not written by builtin arguments, don't write to it.
3998          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
3999          *
4000          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4001          * to TMP_OUT in any case
4002          */
4003         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4004         {
4005             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4006         }
4007         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4008         {
4009             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4010         }
4011         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4012         {
4013             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4014         }
4015         else
4016         {
4017             continue;
4018         }
4019
4020         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4021                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4022         {
4023             compiled->need_color_unclamp = TRUE;
4024         }
4025     }
4026
4027     /* Map declared to declared */
4028     for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
4029     {
4030         /* Write unread output to TA to throw them away */
4031         priv_ctx->vs_output[i] = "TA";
4032         semantic_name = baseshader->output_signature[i].semantic_name;
4033         if (!semantic_name) continue;
4034
4035         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4036                 && !baseshader->output_signature[i].semantic_idx)
4037         {
4038             priv_ctx->vs_output[i] = "TMP_OUT";
4039             continue;
4040         }
4041         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4042                 && !baseshader->output_signature[i].semantic_idx)
4043         {
4044             priv_ctx->vs_output[i] = "result.pointsize";
4045             continue;
4046         }
4047
4048         for(j = 0; j < MAX_REG_INPUT; j++)
4049         {
4050             if (!sig[j].semantic_name) continue;
4051
4052             if (!strcmp(sig[j].semantic_name, semantic_name)
4053                     && sig[j].semantic_idx == baseshader->output_signature[i].semantic_idx)
4054             {
4055                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4056
4057                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4058                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4059                 {
4060                     compiled->need_color_unclamp = TRUE;
4061                 }
4062             }
4063         }
4064     }
4065 }
4066
4067 /* GL locking is done by the caller */
4068 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This, struct wined3d_shader_buffer *buffer,
4069         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4070 {
4071     const shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4072     CONST DWORD *function = This->baseShader.function;
4073     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4074     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4075     const local_constant *lconst;
4076     GLuint ret;
4077     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
4078     struct shader_arb_ctx_priv priv_ctx;
4079     unsigned int i;
4080     GLint errPos;
4081
4082     memset(&priv_ctx, 0, sizeof(priv_ctx));
4083     priv_ctx.cur_vs_args = args;
4084     list_init(&priv_ctx.control_frames);
4085     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
4086
4087     /*  Create the hw ARB shader */
4088     shader_addline(buffer, "!!ARBvp1.0\n");
4089
4090     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4091      * mesurable performance penalty, and we can always make use of it for clipplanes.
4092      */
4093     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4094     {
4095         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4096         priv_ctx.target_version = NV3;
4097         shader_addline(buffer, "ADDRESS aL;\n");
4098     }
4099     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4100     {
4101         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4102         priv_ctx.target_version = NV2;
4103         shader_addline(buffer, "ADDRESS aL;\n");
4104     } else {
4105         priv_ctx.target_version = ARB;
4106     }
4107
4108     shader_addline(buffer, "TEMP TMP_OUT;\n");
4109     if(need_helper_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
4110         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4111     }
4112     if(need_rel_addr_const((IWineD3DBaseShaderImpl *) This, gl_info)) {
4113         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", This->rel_offset);
4114         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4115     }
4116
4117     shader_addline(buffer, "TEMP TA;\n");
4118
4119     /* Base Declarations */
4120     next_local = shader_generate_arb_declarations((IWineD3DBaseShader *)This,
4121             reg_maps, buffer, gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4122
4123     for(i = 0; i < MAX_CONST_I; i++)
4124     {
4125         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4126         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4127         {
4128             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
4129
4130             if(control_values)
4131             {
4132                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4133                                 control_values[0], control_values[1], control_values[2]);
4134             }
4135             else
4136             {
4137                 compiled->int_consts[i] = next_local;
4138                 compiled->num_int_consts++;
4139                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4140             }
4141         }
4142     }
4143
4144     /* We need a constant to fixup the final position */
4145     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4146     compiled->pos_fixup = next_local++;
4147
4148     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4149      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4150      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4151      * a replacement shader depend on the texcoord.w being set properly.
4152      *
4153      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4154      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4155      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4156      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4157      * this can eat a number of instructions, so skip it unless this cap is set as well
4158      */
4159     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4160     {
4161         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4162         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4163
4164         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4165         {
4166             int i;
4167             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4168             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4169             {
4170                 if (This->baseShader.reg_maps.texcoord_mask[i]
4171                         && This->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4172                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4173             }
4174         }
4175     }
4176
4177     /* The shader starts with the main function */
4178     priv_ctx.in_main_func = TRUE;
4179     /* Base Shader Body */
4180     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4181
4182     if(!priv_ctx.footer_written) vshader_add_footer(This, buffer, args, &priv_ctx);
4183
4184     shader_addline(buffer, "END\n");
4185
4186     /* TODO: change to resource.glObjectHandle or something like that */
4187     GL_EXTCALL(glGenProgramsARB(1, &ret));
4188
4189     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4190     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4191
4192     TRACE("Created hw vertex shader, prg=%d\n", ret);
4193     /* Create the program and check for errors */
4194     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4195                buffer->bsize, buffer->buffer));
4196     checkGLcall("glProgramStringARB()");
4197
4198     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4199     if (errPos != -1)
4200     {
4201         FIXME("HW VertexShader Error at position %d: %s\n\n",
4202               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4203         shader_arb_dump_program_source(buffer->buffer);
4204         ret = -1;
4205     }
4206     else
4207     {
4208         GLint native;
4209
4210         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4211         checkGLcall("glGetProgramivARB()");
4212         if (!native) WARN("Program exceeds native resource limits.\n");
4213
4214         /* Load immediate constants */
4215         if(lconst_map) {
4216             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
4217                 const float *value = (const float *)lconst->value;
4218                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4219             }
4220         }
4221     }
4222     HeapFree(GetProcessHeap(), 0, lconst_map);
4223
4224     return ret;
4225 }
4226
4227 /* GL locking is done by the caller */
4228 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
4229 {
4230     UINT i;
4231     DWORD new_size;
4232     struct arb_ps_compiled_shader *new_array;
4233     struct wined3d_shader_buffer buffer;
4234     struct arb_pshader_private *shader_data;
4235     GLuint ret;
4236
4237     if (!shader->baseShader.backend_data)
4238     {
4239         IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) shader->baseShader.device;
4240         const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4241         struct shader_arb_priv *priv = device->shader_priv;
4242
4243         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4244         shader_data = shader->baseShader.backend_data;
4245         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
4246
4247         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
4248         else shader_data->input_signature_idx = find_input_signature(priv, shader->baseShader.input_signature);
4249
4250         shader_data->has_signature_idx = TRUE;
4251         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4252
4253         if (!device->vs_clipping)
4254             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
4255                     gl_info->limits.texture_stages - 1);
4256         else
4257             shader_data->clipplane_emulation = ~0U;
4258     }
4259     shader_data = shader->baseShader.backend_data;
4260
4261     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4262      * so a linear search is more performant than a hashmap or a binary search
4263      * (cache coherency etc)
4264      */
4265     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4266     {
4267         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4268             return &shader_data->gl_shaders[i];
4269     }
4270
4271     TRACE("No matching GL shader found, compiling a new shader\n");
4272     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4273         if (shader_data->num_gl_shaders)
4274         {
4275             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4276             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4277                                     new_size * sizeof(*shader_data->gl_shaders));
4278         } else {
4279             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4280             new_size = 1;
4281         }
4282
4283         if(!new_array) {
4284             ERR("Out of memory\n");
4285             return 0;
4286         }
4287         shader_data->gl_shaders = new_array;
4288         shader_data->shader_array_size = new_size;
4289     }
4290
4291     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4292
4293     pixelshader_update_samplers(&shader->baseShader.reg_maps,
4294             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4295
4296     if (!shader_buffer_init(&buffer))
4297     {
4298         ERR("Failed to initialize shader buffer.\n");
4299         return 0;
4300     }
4301
4302     ret = shader_arb_generate_pshader(shader, &buffer, args,
4303                                       &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4304     shader_buffer_free(&buffer);
4305     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4306
4307     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4308 }
4309
4310 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4311                                  const DWORD use_map, BOOL skip_int) {
4312     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4313     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4314     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4315     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4316     if(stored->ps_signature != new->ps_signature) return FALSE;
4317     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4318     if(skip_int) return TRUE;
4319
4320     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4321 }
4322
4323 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4324 {
4325     UINT i;
4326     DWORD new_size;
4327     struct arb_vs_compiled_shader *new_array;
4328     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4329     struct wined3d_shader_buffer buffer;
4330     struct arb_vshader_private *shader_data;
4331     GLuint ret;
4332     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4333
4334     if (!shader->baseShader.backend_data)
4335     {
4336         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4337     }
4338     shader_data = shader->baseShader.backend_data;
4339
4340     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4341      * so a linear search is more performant than a hashmap or a binary search
4342      * (cache coherency etc)
4343      */
4344     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4345         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4346                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4347         {
4348             return &shader_data->gl_shaders[i];
4349         }
4350     }
4351
4352     TRACE("No matching GL shader found, compiling a new shader\n");
4353
4354     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4355         if (shader_data->num_gl_shaders)
4356         {
4357             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4358             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4359                                     new_size * sizeof(*shader_data->gl_shaders));
4360         } else {
4361             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4362             new_size = 1;
4363         }
4364
4365         if(!new_array) {
4366             ERR("Out of memory\n");
4367             return 0;
4368         }
4369         shader_data->gl_shaders = new_array;
4370         shader_data->shader_array_size = new_size;
4371     }
4372
4373     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4374
4375     if (!shader_buffer_init(&buffer))
4376     {
4377         ERR("Failed to initialize shader buffer.\n");
4378         return 0;
4379     }
4380
4381     ret = shader_arb_generate_vshader(shader, &buffer, args,
4382             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4383     shader_buffer_free(&buffer);
4384     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4385
4386     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4387 }
4388
4389 static inline void find_arb_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4390         struct arb_ps_compile_args *args)
4391 {
4392     int i;
4393     WORD int_skip;
4394     const struct wined3d_gl_info *gl_info = &((IWineD3DDeviceImpl *)shader->baseShader.device)->adapter->gl_info;
4395     find_ps_compile_args(shader, stateblock, &args->super);
4396
4397     /* This forces all local boolean constants to 1 to make them stateblock independent */
4398     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4399
4400     for(i = 0; i < MAX_CONST_B; i++)
4401     {
4402         if(stateblock->pixelShaderConstantB[i]) args->bools |= ( 1 << i);
4403     }
4404
4405     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4406      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4407      * duplicate the shader than have a no-op KIL instruction in every shader
4408      */
4409     if ((!((IWineD3DDeviceImpl *)shader->baseShader.device)->vs_clipping) && use_vs(stateblock)
4410             && stateblock->state.render_states[WINED3DRS_CLIPPING]
4411             && stateblock->state.render_states[WINED3DRS_CLIPPLANEENABLE])
4412         args->clip = 1;
4413     else
4414         args->clip = 0;
4415
4416     /* Skip if unused or local, or supported natively */
4417     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4418     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4419     {
4420         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4421         return;
4422     }
4423
4424     for(i = 0; i < MAX_CONST_I; i++)
4425     {
4426         if(int_skip & (1 << i))
4427         {
4428             args->loop_ctrl[i][0] = 0;
4429             args->loop_ctrl[i][1] = 0;
4430             args->loop_ctrl[i][2] = 0;
4431         }
4432         else
4433         {
4434             args->loop_ctrl[i][0] = stateblock->pixelShaderConstantI[i * 4];
4435             args->loop_ctrl[i][1] = stateblock->pixelShaderConstantI[i * 4 + 1];
4436             args->loop_ctrl[i][2] = stateblock->pixelShaderConstantI[i * 4 + 2];
4437         }
4438     }
4439 }
4440
4441 static inline void find_arb_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock,
4442         struct arb_vs_compile_args *args)
4443 {
4444     int i;
4445     WORD int_skip;
4446     IWineD3DDeviceImpl *dev = (IWineD3DDeviceImpl *)shader->baseShader.device;
4447     const struct wined3d_gl_info *gl_info = &dev->adapter->gl_info;
4448     find_vs_compile_args(shader, stateblock, &args->super);
4449
4450     args->clip.boolclip_compare = 0;
4451     if(use_ps(stateblock))
4452     {
4453         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) stateblock->pixelShader;
4454         struct arb_pshader_private *shader_priv = ps->baseShader.backend_data;
4455         args->ps_signature = shader_priv->input_signature_idx;
4456
4457         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4458     }
4459     else
4460     {
4461         args->ps_signature = ~0;
4462         if(!dev->vs_clipping)
4463         {
4464             args->clip.boolclip.clip_texcoord = ffp_clip_emul(stateblock) ? gl_info->limits.texture_stages : 0;
4465         }
4466         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4467     }
4468
4469     if (args->clip.boolclip.clip_texcoord)
4470     {
4471         if (stateblock->state.render_states[WINED3DRS_CLIPPING])
4472             args->clip.boolclip.clipplane_mask = (unsigned char)stateblock->state.render_states[WINED3DRS_CLIPPLANEENABLE];
4473         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4474     }
4475
4476     /* This forces all local boolean constants to 1 to make them stateblock independent */
4477     args->clip.boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4478     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4479     for(i = 0; i < MAX_CONST_B; i++)
4480     {
4481         if(stateblock->vertexShaderConstantB[i]) args->clip.boolclip.bools |= ( 1 << i);
4482     }
4483
4484     args->vertex.samplers[0] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4485     args->vertex.samplers[1] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4486     args->vertex.samplers[2] = dev->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4487     args->vertex.samplers[3] = 0;
4488
4489     /* Skip if unused or local */
4490     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4491     /* This is about flow control, not clipping. */
4492     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4493     {
4494         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4495         return;
4496     }
4497
4498     for(i = 0; i < MAX_CONST_I; i++)
4499     {
4500         if(int_skip & (1 << i))
4501         {
4502             args->loop_ctrl[i][0] = 0;
4503             args->loop_ctrl[i][1] = 0;
4504             args->loop_ctrl[i][2] = 0;
4505         }
4506         else
4507         {
4508             args->loop_ctrl[i][0] = stateblock->vertexShaderConstantI[i * 4];
4509             args->loop_ctrl[i][1] = stateblock->vertexShaderConstantI[i * 4 + 1];
4510             args->loop_ctrl[i][2] = stateblock->vertexShaderConstantI[i * 4 + 2];
4511         }
4512     }
4513 }
4514
4515 /* GL locking is done by the caller */
4516 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4517 {
4518     IWineD3DDeviceImpl *This = context->swapchain->device;
4519     struct shader_arb_priv *priv = This->shader_priv;
4520     const struct wined3d_gl_info *gl_info = context->gl_info;
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         struct arb_ps_compile_args compile_args;
4526         struct arb_ps_compiled_shader *compiled;
4527         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) This->stateBlock->pixelShader;
4528
4529         TRACE("Using pixel shader %p\n", This->stateBlock->pixelShader);
4530         find_arb_ps_compile_args(ps, This->stateBlock, &compile_args);
4531         compiled = find_arb_pshader(ps, &compile_args);
4532         priv->current_fprogram_id = compiled->prgId;
4533         priv->compiled_fprog = compiled;
4534
4535         /* Bind the fragment program */
4536         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4537         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4538
4539         if(!priv->use_arbfp_fixed_func) {
4540             /* Enable OpenGL fragment programs */
4541             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4542             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4543         }
4544         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4545
4546         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4547          * a 1.x and newer shader, reload the first 8 constants
4548          */
4549         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts)
4550         {
4551             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts;
4552             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4553             for(i = 0; i < 8; i++)
4554             {
4555                 context->pshader_const_dirty[i] = 1;
4556             }
4557             /* Also takes care of loading local constants */
4558             shader_arb_load_constants(context, TRUE, FALSE);
4559         }
4560         else
4561         {
4562             shader_arb_ps_local_constants(This);
4563         }
4564
4565         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4566         if (compiled->np2fixup_info.super.active)
4567             shader_arb_load_np2fixup_constants((IWineD3DDevice *)This, usePS, useVS);
4568     }
4569     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4570     {
4571         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4572         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4573         * replacement shader
4574         */
4575         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4576         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4577         priv->current_fprogram_id = 0;
4578     }
4579
4580     if (useVS) {
4581         struct arb_vs_compile_args compile_args;
4582         struct arb_vs_compiled_shader *compiled;
4583         IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) This->stateBlock->vertexShader;
4584
4585         TRACE("Using vertex shader %p\n", This->stateBlock->vertexShader);
4586         find_arb_vs_compile_args(vs, This->stateBlock, &compile_args);
4587         compiled = find_arb_vshader(vs, &compile_args);
4588         priv->current_vprogram_id = compiled->prgId;
4589         priv->compiled_vprog = compiled;
4590
4591         /* Bind the vertex program */
4592         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4593         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4594
4595         /* Enable OpenGL vertex programs */
4596         glEnable(GL_VERTEX_PROGRAM_ARB);
4597         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4598         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4599         shader_arb_vs_local_constants(This);
4600
4601         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4602             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4603
4604             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4605             {
4606                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4607                 checkGLcall("glClampColorARB");
4608             } else {
4609                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4610             }
4611         }
4612     }
4613     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4614     {
4615         priv->current_vprogram_id = 0;
4616         glDisable(GL_VERTEX_PROGRAM_ARB);
4617         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4618     }
4619 }
4620
4621 /* GL locking is done by the caller */
4622 static void shader_arb_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type, const SIZE *ds_mask_size)
4623 {
4624     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4625     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4626     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4627     struct shader_arb_priv *priv = This->shader_priv;
4628     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4629     GLuint *blt_fprogram;
4630
4631     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4632     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4633     glEnable(GL_VERTEX_PROGRAM_ARB);
4634
4635     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4636     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4637     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4638     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4639     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4640 }
4641
4642 /* GL locking is done by the caller */
4643 static void shader_arb_deselect_depth_blt(IWineD3DDevice *iface) {
4644     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4645     struct shader_arb_priv *priv = This->shader_priv;
4646     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4647
4648     if (priv->current_vprogram_id) {
4649         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4650         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4651
4652         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4653     } else {
4654         glDisable(GL_VERTEX_PROGRAM_ARB);
4655         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4656     }
4657
4658     if (priv->current_fprogram_id) {
4659         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4660         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4661
4662         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4663     } else if(!priv->use_arbfp_fixed_func) {
4664         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4665         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4666     }
4667 }
4668
4669 static void shader_arb_destroy(IWineD3DBaseShader *iface) {
4670     IWineD3DBaseShaderImpl *baseShader = (IWineD3DBaseShaderImpl *) iface;
4671     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)baseShader->baseShader.device;
4672     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4673
4674     if (shader_is_pshader_version(baseShader->baseShader.reg_maps.shader_version.type))
4675     {
4676         struct arb_pshader_private *shader_data = baseShader->baseShader.backend_data;
4677         UINT i;
4678
4679         if(!shader_data) return; /* This can happen if a shader was never compiled */
4680
4681         if (shader_data->num_gl_shaders)
4682         {
4683             struct wined3d_context *context = context_acquire(device, NULL);
4684
4685             ENTER_GL();
4686             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4687             {
4688                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4689                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4690             }
4691             LEAVE_GL();
4692
4693             context_release(context);
4694         }
4695
4696         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4697         HeapFree(GetProcessHeap(), 0, shader_data);
4698         baseShader->baseShader.backend_data = NULL;
4699     }
4700     else
4701     {
4702         struct arb_vshader_private *shader_data = baseShader->baseShader.backend_data;
4703         UINT i;
4704
4705         if(!shader_data) return; /* This can happen if a shader was never compiled */
4706
4707         if (shader_data->num_gl_shaders)
4708         {
4709             struct wined3d_context *context = context_acquire(device, NULL);
4710
4711             ENTER_GL();
4712             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4713             {
4714                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4715                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4716             }
4717             LEAVE_GL();
4718
4719             context_release(context);
4720         }
4721
4722         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4723         HeapFree(GetProcessHeap(), 0, shader_data);
4724         baseShader->baseShader.backend_data = NULL;
4725     }
4726 }
4727
4728 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4729 {
4730     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4731     return compare_sig(key, e->sig);
4732 }
4733
4734 static const struct wine_rb_functions sig_tree_functions =
4735 {
4736     wined3d_rb_alloc,
4737     wined3d_rb_realloc,
4738     wined3d_rb_free,
4739     sig_tree_compare
4740 };
4741
4742 static HRESULT shader_arb_alloc(IWineD3DDevice *iface) {
4743     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4744     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4745     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4746     {
4747         ERR("RB tree init failed\n");
4748         HeapFree(GetProcessHeap(), 0, priv);
4749         return E_OUTOFMEMORY;
4750     }
4751     This->shader_priv = priv;
4752     return WINED3D_OK;
4753 }
4754
4755 static void release_signature(struct wine_rb_entry *entry, void *context)
4756 {
4757     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4758     int i;
4759     for(i = 0; i < MAX_REG_INPUT; i++)
4760     {
4761         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4762     }
4763     HeapFree(GetProcessHeap(), 0, sig->sig);
4764     HeapFree(GetProcessHeap(), 0, sig);
4765 }
4766
4767 /* Context activation is done by the caller. */
4768 static void shader_arb_free(IWineD3DDevice *iface) {
4769     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4770     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4771     struct shader_arb_priv *priv = This->shader_priv;
4772     int i;
4773
4774     ENTER_GL();
4775     if(priv->depth_blt_vprogram_id) {
4776         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4777     }
4778     for (i = 0; i < tex_type_count; ++i)
4779     {
4780         if (priv->depth_blt_fprogram_id_full[i])
4781         {
4782             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4783         }
4784         if (priv->depth_blt_fprogram_id_masked[i])
4785         {
4786             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4787         }
4788     }
4789     LEAVE_GL();
4790
4791     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4792     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4793 }
4794
4795 static BOOL shader_arb_dirty_const(IWineD3DDevice *iface) {
4796     return TRUE;
4797 }
4798
4799 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4800 {
4801     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4802     {
4803         DWORD vs_consts;
4804
4805         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4806          * for vertex programs. If the native limit is less than that it's
4807          * not very useful, and e.g. Mesa swrast returns 0, probably to
4808          * indicate it's a software implementation. */
4809         if (gl_info->limits.arb_vs_native_constants < 96)
4810             vs_consts = gl_info->limits.arb_vs_float_constants;
4811         else
4812             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4813
4814         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4815         {
4816             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4817             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4818         }
4819         else if (vs_consts >= 256)
4820         {
4821             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4822             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4823             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4824         }
4825         else
4826         {
4827             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4828             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4829         }
4830         pCaps->MaxVertexShaderConst = vs_consts;
4831     }
4832     else
4833     {
4834         pCaps->VertexShaderVersion = 0;
4835         pCaps->MaxVertexShaderConst = 0;
4836     }
4837
4838     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4839     {
4840         DWORD ps_consts;
4841
4842         /* Similar as above for vertex programs, but the minimum for fragment
4843          * programs is 24. */
4844         if (gl_info->limits.arb_ps_native_constants < 24)
4845             ps_consts = gl_info->limits.arb_ps_float_constants;
4846         else
4847             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4848
4849         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4850         {
4851             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4852             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4853         }
4854         else if (ps_consts >= 32)
4855         {
4856             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4857             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4858             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4859         }
4860         else
4861         {
4862             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4863             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4864         }
4865         pCaps->PixelShader1xMaxValue = 8.0f;
4866         pCaps->MaxPixelShaderConst = ps_consts;
4867     }
4868     else
4869     {
4870         pCaps->PixelShaderVersion = 0;
4871         pCaps->PixelShader1xMaxValue = 0.0f;
4872         pCaps->MaxPixelShaderConst = 0;
4873     }
4874
4875     pCaps->VSClipping = use_nv_clip(gl_info);
4876 }
4877
4878 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4879 {
4880     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4881     {
4882         TRACE("Checking support for color_fixup:\n");
4883         dump_color_fixup_desc(fixup);
4884     }
4885
4886     /* We support everything except complex conversions. */
4887     if (!is_complex_fixup(fixup))
4888     {
4889         TRACE("[OK]\n");
4890         return TRUE;
4891     }
4892
4893     TRACE("[FAILED]\n");
4894     return FALSE;
4895 }
4896
4897 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4898     DWORD shift;
4899     char write_mask[20], regstr[50];
4900     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4901     BOOL is_color = FALSE;
4902     const struct wined3d_shader_dst_param *dst;
4903
4904     if (!ins->dst_count) return;
4905
4906     dst = &ins->dst[0];
4907     shift = dst->shift;
4908     if (!shift) return; /* Saturate alone is handled by the instructions */
4909
4910     shader_arb_get_write_mask(ins, dst, write_mask);
4911     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4912
4913     /* Generate a line that does the output modifier computation
4914      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4915      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4916      */
4917     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4918                    regstr, write_mask, regstr, shift_tab[shift]);
4919 }
4920
4921 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4922 {
4923     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4924     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4925     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4926     /* WINED3DSIH_BREAK         */ shader_hw_break,
4927     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4928     /* WINED3DSIH_BREAKP        */ NULL,
4929     /* WINED3DSIH_CALL          */ shader_hw_call,
4930     /* WINED3DSIH_CALLNZ        */ NULL,
4931     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4932     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4933     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4934     /* WINED3DSIH_CUT           */ NULL,
4935     /* WINED3DSIH_DCL           */ NULL,
4936     /* WINED3DSIH_DEF           */ NULL,
4937     /* WINED3DSIH_DEFB          */ NULL,
4938     /* WINED3DSIH_DEFI          */ NULL,
4939     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4940     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4941     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4942     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4943     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4944     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4945     /* WINED3DSIH_ELSE          */ shader_hw_else,
4946     /* WINED3DSIH_EMIT          */ NULL,
4947     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4948     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4949     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4950     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4951     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4952     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4953     /* WINED3DSIH_IADD          */ NULL,
4954     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4955     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4956     /* WINED3DSIH_IGE           */ NULL,
4957     /* WINED3DSIH_LABEL         */ shader_hw_label,
4958     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
4959     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
4960     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
4961     /* WINED3DSIH_LOOP          */ shader_hw_loop,
4962     /* WINED3DSIH_LRP           */ shader_hw_lrp,
4963     /* WINED3DSIH_LT            */ NULL,
4964     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
4965     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
4966     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
4967     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
4968     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
4969     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
4970     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
4971     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
4972     /* WINED3DSIH_MOV           */ shader_hw_mov,
4973     /* WINED3DSIH_MOVA          */ shader_hw_mov,
4974     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
4975     /* WINED3DSIH_NOP           */ shader_hw_nop,
4976     /* WINED3DSIH_NRM           */ shader_hw_nrm,
4977     /* WINED3DSIH_PHASE         */ NULL,
4978     /* WINED3DSIH_POW           */ shader_hw_log_pow,
4979     /* WINED3DSIH_RCP           */ shader_hw_rcp,
4980     /* WINED3DSIH_REP           */ shader_hw_rep,
4981     /* WINED3DSIH_RET           */ shader_hw_ret,
4982     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
4983     /* WINED3DSIH_SETP          */ NULL,
4984     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
4985     /* WINED3DSIH_SGN           */ shader_hw_sgn,
4986     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
4987     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
4988     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
4989     /* WINED3DSIH_TEX           */ pshader_hw_tex,
4990     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
4991     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
4992     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
4993     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
4994     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
4995     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
4996     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
4997     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
4998     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
4999     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5000     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5001     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5002     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5003     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5004     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5005     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5006     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5007     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5008     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5009     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5010     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5011 };
5012
5013 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
5014 {
5015     BOOL vshader = shader_is_vshader_version(This->baseShader.reg_maps.shader_version.type);
5016     WORD bools = 0;
5017     WORD flag = (1 << idx);
5018     const local_constant *constant;
5019     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5020
5021     if(This->baseShader.reg_maps.local_bool_consts & flag)
5022     {
5023         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5024         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
5025         {
5026             if (constant->idx == idx)
5027             {
5028                 return constant->value[0];
5029             }
5030         }
5031         ERR("Local constant not found\n");
5032         return FALSE;
5033     }
5034     else
5035     {
5036         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5037         else bools = priv->cur_ps_args->bools;
5038         return bools & flag;
5039     }
5040 }
5041
5042 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5043         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
5044 {
5045     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5046
5047     /* Integer constants can either be a local constant, or they can be stored in the shader
5048      * type specific compile args. */
5049     if (This->baseShader.reg_maps.local_int_consts & (1 << idx))
5050     {
5051         const local_constant *constant;
5052
5053         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
5054         {
5055             if (constant->idx == idx)
5056             {
5057                 loop_control->count = constant->value[0];
5058                 loop_control->start = constant->value[1];
5059                 /* Step is signed. */
5060                 loop_control->step = (int)constant->value[2];
5061                 return;
5062             }
5063         }
5064         /* If this happens the flag was set incorrectly */
5065         ERR("Local constant not found\n");
5066         loop_control->count = 0;
5067         loop_control->start = 0;
5068         loop_control->step = 0;
5069         return;
5070     }
5071
5072     switch (This->baseShader.reg_maps.shader_version.type)
5073     {
5074         case WINED3D_SHADER_TYPE_VERTEX:
5075             /* Count and aL start value are unsigned */
5076             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5077             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5078             /* Step is signed. */
5079             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5080             break;
5081
5082         case WINED3D_SHADER_TYPE_PIXEL:
5083             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5084             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5085             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5086             break;
5087
5088         default:
5089             FIXME("Unhandled shader type %#x.\n", This->baseShader.reg_maps.shader_version.type);
5090             break;
5091     }
5092 }
5093
5094 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5095 {
5096     unsigned int i;
5097     struct wined3d_shader_dst_param *dst_param = NULL;
5098     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5099     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5100     if(!rec)
5101     {
5102         ERR("Out of memory\n");
5103         return;
5104     }
5105
5106     rec->ins = *ins;
5107     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5108     if(!dst_param) goto free;
5109     *dst_param = *ins->dst;
5110     if(ins->dst->reg.rel_addr)
5111     {
5112         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5113         if(!rel_addr) goto free;
5114         *rel_addr = *ins->dst->reg.rel_addr;
5115         dst_param->reg.rel_addr = rel_addr;
5116     }
5117     rec->ins.dst = dst_param;
5118
5119     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5120     if(!src_param) goto free;
5121     for(i = 0; i < ins->src_count; i++)
5122     {
5123         src_param[i] = ins->src[i];
5124         if(ins->src[i].reg.rel_addr)
5125         {
5126             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5127             if(!rel_addr) goto free;
5128             *rel_addr = *ins->src[i].reg.rel_addr;
5129             src_param[i].reg.rel_addr = rel_addr;
5130         }
5131     }
5132     rec->ins.src = src_param;
5133     list_add_tail(list, &rec->entry);
5134     return;
5135
5136 free:
5137     ERR("Out of memory\n");
5138     if(dst_param)
5139     {
5140         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5141         HeapFree(GetProcessHeap(), 0, dst_param);
5142     }
5143     if(src_param)
5144     {
5145         for(i = 0; i < ins->src_count; i++)
5146         {
5147             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5148         }
5149         HeapFree(GetProcessHeap(), 0, src_param);
5150     }
5151     HeapFree(GetProcessHeap(), 0, rec);
5152 }
5153
5154 static void free_recorded_instruction(struct list *list)
5155 {
5156     struct recorded_instruction *rec_ins, *entry2;
5157     unsigned int i;
5158
5159     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5160     {
5161         list_remove(&rec_ins->entry);
5162         if(rec_ins->ins.dst)
5163         {
5164             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5165             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5166         }
5167         if(rec_ins->ins.src)
5168         {
5169             for(i = 0; i < rec_ins->ins.src_count; i++)
5170             {
5171                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5172             }
5173             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5174         }
5175         HeapFree(GetProcessHeap(), 0, rec_ins);
5176     }
5177 }
5178
5179 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5180     SHADER_HANDLER hw_fct;
5181     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5182     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
5183     struct control_frame *control_frame;
5184     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5185     BOOL bool_const;
5186
5187     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5188     {
5189         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5190         list_add_head(&priv->control_frames, &control_frame->entry);
5191
5192         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5193         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5194
5195         if(priv->target_version >= NV2)
5196         {
5197             control_frame->no.loop = priv->num_loops++;
5198             priv->loop_depth++;
5199         }
5200         else
5201         {
5202             /* Don't bother recording when we're in a not used if branch */
5203             if(priv->muted)
5204             {
5205                 return;
5206             }
5207
5208             if(!priv->recording)
5209             {
5210                 list_init(&priv->record);
5211                 priv->recording = TRUE;
5212                 control_frame->outer_loop = TRUE;
5213                 get_loop_control_const(ins, This, ins->src[0].reg.idx, &control_frame->loop_control);
5214                 return; /* Instruction is handled */
5215             }
5216             /* Record this loop in the outer loop's recording */
5217         }
5218     }
5219     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5220     {
5221         if(priv->target_version >= NV2)
5222         {
5223             /* Nothing to do. The control frame is popped after the HW instr handler */
5224         }
5225         else
5226         {
5227             struct list *e = list_head(&priv->control_frames);
5228             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5229             list_remove(&control_frame->entry);
5230
5231             if(control_frame->outer_loop)
5232             {
5233                 unsigned int iteration;
5234                 int aL = 0;
5235                 struct list copy;
5236
5237                 /* Turn off recording before playback */
5238                 priv->recording = FALSE;
5239
5240                 /* Move the recorded instructions to a separate list and get them out of the private data
5241                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5242                  * be recorded again, thus priv->record might be overwritten
5243                  */
5244                 list_init(&copy);
5245                 list_move_tail(&copy, &priv->record);
5246                 list_init(&priv->record);
5247
5248                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5249                 {
5250                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5251                                    control_frame->loop_control.count, control_frame->loop_control.start,
5252                                    control_frame->loop_control.step);
5253                     aL = control_frame->loop_control.start;
5254                 }
5255                 else
5256                 {
5257                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5258                 }
5259
5260                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5261                 {
5262                     struct recorded_instruction *rec_ins;
5263                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5264                     {
5265                         priv->aL = aL;
5266                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5267                     }
5268                     else
5269                     {
5270                         shader_addline(buffer, "#Iteration %u\n", iteration);
5271                     }
5272
5273                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5274                     {
5275                         shader_arb_handle_instruction(&rec_ins->ins);
5276                     }
5277
5278                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5279                     {
5280                         aL += control_frame->loop_control.step;
5281                     }
5282                 }
5283                 shader_addline(buffer, "#end loop/rep\n");
5284
5285                 free_recorded_instruction(&copy);
5286                 HeapFree(GetProcessHeap(), 0, control_frame);
5287                 return; /* Instruction is handled */
5288             }
5289             else
5290             {
5291                 /* This is a nested loop. Proceed to the normal recording function */
5292                 HeapFree(GetProcessHeap(), 0, control_frame);
5293             }
5294         }
5295     }
5296
5297     if(priv->recording)
5298     {
5299         record_instruction(&priv->record, ins);
5300         return;
5301     }
5302
5303     /* boolean if */
5304     if(ins->handler_idx == WINED3DSIH_IF)
5305     {
5306         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5307         list_add_head(&priv->control_frames, &control_frame->entry);
5308         control_frame->type = IF;
5309
5310         bool_const = get_bool_const(ins, This, ins->src[0].reg.idx);
5311         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5312         if (!priv->muted && !bool_const)
5313         {
5314             shader_addline(buffer, "#if(FALSE){\n");
5315             priv->muted = TRUE;
5316             control_frame->muting = TRUE;
5317         }
5318         else shader_addline(buffer, "#if(TRUE) {\n");
5319
5320         return; /* Instruction is handled */
5321     }
5322     else if(ins->handler_idx == WINED3DSIH_IFC)
5323     {
5324         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5325         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5326         control_frame->type = IFC;
5327         control_frame->no.ifc = priv->num_ifcs++;
5328         list_add_head(&priv->control_frames, &control_frame->entry);
5329     }
5330     else if(ins->handler_idx == WINED3DSIH_ELSE)
5331     {
5332         struct list *e = list_head(&priv->control_frames);
5333         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5334
5335         if(control_frame->type == IF)
5336         {
5337             shader_addline(buffer, "#} else {\n");
5338             if(!priv->muted && !control_frame->muting)
5339             {
5340                 priv->muted = TRUE;
5341                 control_frame->muting = TRUE;
5342             }
5343             else if(control_frame->muting) priv->muted = FALSE;
5344             return; /* Instruction is handled. */
5345         }
5346         /* In case of an ifc, generate a HW shader instruction */
5347     }
5348     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5349     {
5350         struct list *e = list_head(&priv->control_frames);
5351         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5352
5353         if(control_frame->type == IF)
5354         {
5355             shader_addline(buffer, "#} endif\n");
5356             if(control_frame->muting) priv->muted = FALSE;
5357             list_remove(&control_frame->entry);
5358             HeapFree(GetProcessHeap(), 0, control_frame);
5359             return; /* Instruction is handled */
5360         }
5361     }
5362
5363     if(priv->muted) return;
5364
5365     /* Select handler */
5366     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5367
5368     /* Unhandled opcode */
5369     if (!hw_fct)
5370     {
5371         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5372         return;
5373     }
5374     hw_fct(ins);
5375
5376     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5377     {
5378         struct list *e = list_head(&priv->control_frames);
5379         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5380         list_remove(&control_frame->entry);
5381         HeapFree(GetProcessHeap(), 0, control_frame);
5382         priv->loop_depth--;
5383     }
5384     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5385     {
5386         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5387         struct list *e = list_head(&priv->control_frames);
5388         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5389         list_remove(&control_frame->entry);
5390         HeapFree(GetProcessHeap(), 0, control_frame);
5391     }
5392
5393
5394     shader_arb_add_instruction_modifiers(ins);
5395 }
5396
5397 const shader_backend_t arb_program_shader_backend = {
5398     shader_arb_handle_instruction,
5399     shader_arb_select,
5400     shader_arb_select_depth_blt,
5401     shader_arb_deselect_depth_blt,
5402     shader_arb_update_float_vertex_constants,
5403     shader_arb_update_float_pixel_constants,
5404     shader_arb_load_constants,
5405     shader_arb_load_np2fixup_constants,
5406     shader_arb_destroy,
5407     shader_arb_alloc,
5408     shader_arb_free,
5409     shader_arb_dirty_const,
5410     shader_arb_get_caps,
5411     shader_arb_color_fixup_supported,
5412 };
5413
5414 /* ARB_fragment_program fixed function pipeline replacement definitions */
5415 #define ARB_FFP_CONST_TFACTOR           0
5416 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5417 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5418 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5419 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5420
5421 struct arbfp_ffp_desc
5422 {
5423     struct ffp_frag_desc parent;
5424     GLuint shader;
5425     unsigned int num_textures_used;
5426 };
5427
5428 /* Context activation is done by the caller. */
5429 static void arbfp_enable(IWineD3DDevice *iface, BOOL enable) {
5430     ENTER_GL();
5431     if(enable) {
5432         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5433         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5434     } else {
5435         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5436         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5437     }
5438     LEAVE_GL();
5439 }
5440
5441 static HRESULT arbfp_alloc(IWineD3DDevice *iface) {
5442     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5443     struct shader_arb_priv *priv;
5444     /* Share private data between the shader backend and the pipeline replacement, if both
5445      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5446      * if no pixel shader is bound or not
5447      */
5448     if(This->shader_backend == &arb_program_shader_backend) {
5449         This->fragment_priv = This->shader_priv;
5450     } else {
5451         This->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5452         if(!This->fragment_priv) return E_OUTOFMEMORY;
5453     }
5454     priv = This->fragment_priv;
5455     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5456     {
5457         ERR("Failed to initialize rbtree.\n");
5458         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5459         return E_OUTOFMEMORY;
5460     }
5461     priv->use_arbfp_fixed_func = TRUE;
5462     return WINED3D_OK;
5463 }
5464
5465 /* Context activation is done by the caller. */
5466 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5467 {
5468     const struct wined3d_gl_info *gl_info = context;
5469     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5470
5471     ENTER_GL();
5472     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5473     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5474     HeapFree(GetProcessHeap(), 0, entry_arb);
5475     LEAVE_GL();
5476 }
5477
5478 /* Context activation is done by the caller. */
5479 static void arbfp_free(IWineD3DDevice *iface) {
5480     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *) iface;
5481     struct shader_arb_priv *priv = This->fragment_priv;
5482
5483     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &This->adapter->gl_info);
5484     priv->use_arbfp_fixed_func = FALSE;
5485
5486     if(This->shader_backend != &arb_program_shader_backend) {
5487         HeapFree(GetProcessHeap(), 0, This->fragment_priv);
5488     }
5489 }
5490
5491 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5492 {
5493     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5494     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5495                            WINED3DTEXOPCAPS_SELECTARG1                  |
5496                            WINED3DTEXOPCAPS_SELECTARG2                  |
5497                            WINED3DTEXOPCAPS_MODULATE4X                  |
5498                            WINED3DTEXOPCAPS_MODULATE2X                  |
5499                            WINED3DTEXOPCAPS_MODULATE                    |
5500                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5501                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5502                            WINED3DTEXOPCAPS_ADD                         |
5503                            WINED3DTEXOPCAPS_SUBTRACT                    |
5504                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5505                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5506                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5507                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5508                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5509                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5510                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5511                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5512                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5513                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5514                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5515                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5516                            WINED3DTEXOPCAPS_LERP                        |
5517                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5518                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5519
5520     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5521
5522     caps->MaxTextureBlendStages   = 8;
5523     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5524 }
5525
5526 static void state_texfactor_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5527 {
5528     const struct wined3d_gl_info *gl_info = context->gl_info;
5529     IWineD3DDeviceImpl *device = stateblock->device;
5530     float col[4];
5531
5532     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5533      * application provided constants
5534      */
5535     if(device->shader_backend == &arb_program_shader_backend) {
5536         if (use_ps(stateblock)) return;
5537
5538         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5539         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5540     }
5541
5542     D3DCOLORTOGLFLOAT4(stateblock->state.render_states[WINED3DRS_TEXTUREFACTOR], col);
5543     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5544     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5545
5546 }
5547
5548 static void state_arb_specularenable(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5549 {
5550     const struct wined3d_gl_info *gl_info = context->gl_info;
5551     IWineD3DDeviceImpl *device = stateblock->device;
5552     float col[4];
5553
5554     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5555      * application provided constants
5556      */
5557     if(device->shader_backend == &arb_program_shader_backend) {
5558         if (use_ps(stateblock)) return;
5559
5560         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5561         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5562     }
5563
5564     if (stateblock->state.render_states[WINED3DRS_SPECULARENABLE])
5565     {
5566         /* The specular color has no alpha */
5567         col[0] = 1.0f; col[1] = 1.0f;
5568         col[2] = 1.0f; col[3] = 0.0f;
5569     } else {
5570         col[0] = 0.0f; col[1] = 0.0f;
5571         col[2] = 0.0f; col[3] = 0.0f;
5572     }
5573     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5574     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5575 }
5576
5577 static void set_bumpmat_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5578 {
5579     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5580     const struct wined3d_gl_info *gl_info = context->gl_info;
5581     IWineD3DDeviceImpl *device = stateblock->device;
5582     float mat[2][2];
5583
5584     if (use_ps(stateblock))
5585     {
5586         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *)stateblock->pixelShader;
5587         if (stage && (ps->baseShader.reg_maps.bumpmat & (1 << stage)))
5588         {
5589             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5590              * anyway
5591              */
5592             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5593                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5594         }
5595
5596         if(device->shader_backend == &arb_program_shader_backend) {
5597             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5598             return;
5599         }
5600     } else if(device->shader_backend == &arb_program_shader_backend) {
5601         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5602         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5603     }
5604
5605     mat[0][0] = *((float *)&stateblock->state.texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5606     mat[0][1] = *((float *)&stateblock->state.texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5607     mat[1][0] = *((float *)&stateblock->state.texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5608     mat[1][1] = *((float *)&stateblock->state.texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5609
5610     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5611     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5612 }
5613
5614 static void tex_bumpenvlum_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
5615 {
5616     DWORD stage = (state - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5617     const struct wined3d_gl_info *gl_info = context->gl_info;
5618     IWineD3DDeviceImpl *device = stateblock->device;
5619     float param[4];
5620
5621     if (use_ps(stateblock))
5622     {
5623         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *)stateblock->pixelShader;
5624         if (stage && (ps->baseShader.reg_maps.luminanceparams & (1 << stage)))
5625         {
5626             /* The pixel shader has to know the luminance offset. Do a constants update if it
5627              * isn't scheduled anyway
5628              */
5629             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5630                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5631         }
5632
5633         if(device->shader_backend == &arb_program_shader_backend) {
5634             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5635             return;
5636         }
5637     } else if(device->shader_backend == &arb_program_shader_backend) {
5638         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5639         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5640     }
5641
5642     param[0] = *((float *)&stateblock->state.texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5643     param[1] = *((float *)&stateblock->state.texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5644     param[2] = 0.0f;
5645     param[3] = 0.0f;
5646
5647     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5648     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5649 }
5650
5651 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5652 {
5653     const char *ret;
5654
5655     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5656
5657     switch(arg & WINED3DTA_SELECTMASK) {
5658         case WINED3DTA_DIFFUSE:
5659             ret = "fragment.color.primary"; break;
5660
5661         case WINED3DTA_CURRENT:
5662             if (!stage) ret = "fragment.color.primary";
5663             else ret = "ret";
5664             break;
5665
5666         case WINED3DTA_TEXTURE:
5667             switch(stage) {
5668                 case 0: ret = "tex0"; break;
5669                 case 1: ret = "tex1"; break;
5670                 case 2: ret = "tex2"; break;
5671                 case 3: ret = "tex3"; break;
5672                 case 4: ret = "tex4"; break;
5673                 case 5: ret = "tex5"; break;
5674                 case 6: ret = "tex6"; break;
5675                 case 7: ret = "tex7"; break;
5676                 default: ret = "unknown texture";
5677             }
5678             break;
5679
5680         case WINED3DTA_TFACTOR:
5681             ret = "tfactor"; break;
5682
5683         case WINED3DTA_SPECULAR:
5684             ret = "fragment.color.secondary"; break;
5685
5686         case WINED3DTA_TEMP:
5687             ret = "tempreg"; break;
5688
5689         case WINED3DTA_CONSTANT:
5690             FIXME("Implement perstage constants\n");
5691             switch(stage) {
5692                 case 0: ret = "const0"; break;
5693                 case 1: ret = "const1"; break;
5694                 case 2: ret = "const2"; break;
5695                 case 3: ret = "const3"; break;
5696                 case 4: ret = "const4"; break;
5697                 case 5: ret = "const5"; break;
5698                 case 6: ret = "const6"; break;
5699                 case 7: ret = "const7"; break;
5700                 default: ret = "unknown constant";
5701             }
5702             break;
5703
5704         default:
5705             return "unknown";
5706     }
5707
5708     if(arg & WINED3DTA_COMPLEMENT) {
5709         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5710         if(argnum == 0) ret = "arg0";
5711         if(argnum == 1) ret = "arg1";
5712         if(argnum == 2) ret = "arg2";
5713     }
5714     if(arg & WINED3DTA_ALPHAREPLICATE) {
5715         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5716         if(argnum == 0) ret = "arg0";
5717         if(argnum == 1) ret = "arg1";
5718         if(argnum == 2) ret = "arg2";
5719     }
5720     return ret;
5721 }
5722
5723 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5724         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5725 {
5726     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5727     unsigned int mul = 1;
5728     BOOL mul_final_dest = FALSE;
5729
5730     if(color && alpha) dstmask = "";
5731     else if(color) dstmask = ".xyz";
5732     else dstmask = ".w";
5733
5734     if(dst == tempreg) dstreg = "tempreg";
5735     else dstreg = "ret";
5736
5737     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5738     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5739     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5740
5741     switch(op) {
5742         case WINED3DTOP_DISABLE:
5743             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5744             break;
5745
5746         case WINED3DTOP_SELECTARG2:
5747             arg1 = arg2;
5748         case WINED3DTOP_SELECTARG1:
5749             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5750             break;
5751
5752         case WINED3DTOP_MODULATE4X:
5753             mul = 2;
5754         case WINED3DTOP_MODULATE2X:
5755             mul *= 2;
5756             if (!strcmp(dstreg, "result.color"))
5757             {
5758                 dstreg = "ret";
5759                 mul_final_dest = TRUE;
5760             }
5761         case WINED3DTOP_MODULATE:
5762             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5763             break;
5764
5765         case WINED3DTOP_ADDSIGNED2X:
5766             mul = 2;
5767             if (!strcmp(dstreg, "result.color"))
5768             {
5769                 dstreg = "ret";
5770                 mul_final_dest = TRUE;
5771             }
5772         case WINED3DTOP_ADDSIGNED:
5773             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5774             arg2 = "arg2";
5775         case WINED3DTOP_ADD:
5776             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5777             break;
5778
5779         case WINED3DTOP_SUBTRACT:
5780             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5781             break;
5782
5783         case WINED3DTOP_ADDSMOOTH:
5784             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5785             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5786             break;
5787
5788         case WINED3DTOP_BLENDCURRENTALPHA:
5789             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5790             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5791             break;
5792         case WINED3DTOP_BLENDFACTORALPHA:
5793             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5794             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5795             break;
5796         case WINED3DTOP_BLENDTEXTUREALPHA:
5797             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5798             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5799             break;
5800         case WINED3DTOP_BLENDDIFFUSEALPHA:
5801             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5802             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5803             break;
5804
5805         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5806             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5807             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5808             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5809             break;
5810
5811         /* D3DTOP_PREMODULATE ???? */
5812
5813         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5814             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5815             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5816             break;
5817         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5818             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5819             break;
5820         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5821             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5822             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5823             break;
5824         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5825             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5826             break;
5827
5828         case WINED3DTOP_DOTPRODUCT3:
5829             mul = 4;
5830             if (!strcmp(dstreg, "result.color"))
5831             {
5832                 dstreg = "ret";
5833                 mul_final_dest = TRUE;
5834             }
5835             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5836             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5837             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5838             break;
5839
5840         case WINED3DTOP_MULTIPLYADD:
5841             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5842             break;
5843
5844         case WINED3DTOP_LERP:
5845             /* The msdn is not quite right here */
5846             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5847             break;
5848
5849         case WINED3DTOP_BUMPENVMAP:
5850         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5851             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5852             break;
5853
5854         default:
5855             FIXME("Unhandled texture op %08x\n", op);
5856     }
5857
5858     if(mul == 2) {
5859         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5860     } else if(mul == 4) {
5861         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5862     }
5863 }
5864
5865 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, IWineD3DStateBlockImpl *stateblock)
5866 {
5867     const struct wined3d_gl_info *gl_info = &stateblock->device->adapter->gl_info;
5868     unsigned int stage;
5869     struct wined3d_shader_buffer buffer;
5870     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5871     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5872     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5873     const char *textype;
5874     const char *instr, *sat;
5875     char colorcor_dst[8];
5876     GLuint ret;
5877     DWORD arg0, arg1, arg2;
5878     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5879     BOOL op_equal;
5880     const char *final_combiner_src = "ret";
5881     GLint pos;
5882
5883     /* Find out which textures are read */
5884     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5885         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5886         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5887         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5888         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5889         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5890         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5891         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5892
5893         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5894         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5895         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5896             bump_used[stage] = TRUE;
5897             tex_read[stage] = TRUE;
5898         }
5899         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5900             bump_used[stage] = TRUE;
5901             tex_read[stage] = TRUE;
5902             luminance_used[stage] = TRUE;
5903         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5904             tfactor_used = TRUE;
5905         }
5906
5907         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5908             tfactor_used = TRUE;
5909         }
5910
5911         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5912         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5913             tempreg_used = TRUE;
5914         }
5915
5916         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5917         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5918         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5919         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5920         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5921         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5922         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5923
5924         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5925             tempreg_used = TRUE;
5926         }
5927         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5928             tfactor_used = TRUE;
5929         }
5930     }
5931
5932     /* Shader header */
5933     if (!shader_buffer_init(&buffer))
5934     {
5935         ERR("Failed to initialize shader buffer.\n");
5936         return 0;
5937     }
5938
5939     shader_addline(&buffer, "!!ARBfp1.0\n");
5940
5941     switch(settings->fog) {
5942         case FOG_OFF:                                                         break;
5943         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
5944         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
5945         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
5946         default: FIXME("Unexpected fog setting %d\n", settings->fog);
5947     }
5948
5949     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
5950     shader_addline(&buffer, "TEMP TMP;\n");
5951     shader_addline(&buffer, "TEMP ret;\n");
5952     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
5953     shader_addline(&buffer, "TEMP arg0;\n");
5954     shader_addline(&buffer, "TEMP arg1;\n");
5955     shader_addline(&buffer, "TEMP arg2;\n");
5956     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5957         if(!tex_read[stage]) continue;
5958         shader_addline(&buffer, "TEMP tex%u;\n", stage);
5959         if(!bump_used[stage]) continue;
5960         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
5961         if(!luminance_used[stage]) continue;
5962         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
5963     }
5964     if(tfactor_used) {
5965         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
5966     }
5967         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
5968
5969     if(settings->sRGB_write) {
5970         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
5971                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
5972         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
5973                        srgb_sub_high, 0.0, 0.0, 0.0);
5974     }
5975
5976     if(ffp_clip_emul(stateblock) && settings->emul_clipplanes) shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
5977
5978     /* Generate texture sampling instructions) */
5979     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
5980         if(!tex_read[stage]) continue;
5981
5982         switch(settings->op[stage].tex_type) {
5983             case tex_1d:                    textype = "1D";     break;
5984             case tex_2d:                    textype = "2D";     break;
5985             case tex_3d:                    textype = "3D";     break;
5986             case tex_cube:                  textype = "CUBE";   break;
5987             case tex_rect:                  textype = "RECT";   break;
5988             default: textype = "unexpected_textype";   break;
5989         }
5990
5991         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
5992            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5993             sat = "";
5994         } else {
5995             sat = "_SAT";
5996         }
5997
5998         if(settings->op[stage].projected == proj_none) {
5999             instr = "TEX";
6000         } else if(settings->op[stage].projected == proj_count4 ||
6001                   settings->op[stage].projected == proj_count3) {
6002             instr = "TXP";
6003         } else {
6004             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6005             instr = "TXP";
6006         }
6007
6008         if(stage > 0 &&
6009            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6010             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6011             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6012             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6013             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6014             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6015
6016             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6017              * so multiply the displacement with the dividing parameter before passing it to TXP
6018              */
6019             if (settings->op[stage].projected != proj_none) {
6020                 if(settings->op[stage].projected == proj_count4) {
6021                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6022                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6023                 } else {
6024                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6025                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6026                 }
6027             } else {
6028                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6029             }
6030
6031             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6032                            instr, sat, stage, stage, textype);
6033             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6034                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6035                                stage - 1, stage - 1, stage - 1);
6036                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6037             }
6038         } else if(settings->op[stage].projected == proj_count3) {
6039             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6040             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6041             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6042                             instr, sat, stage, stage, textype);
6043         } else {
6044             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6045                             instr, sat, stage, stage, stage, textype);
6046         }
6047
6048         sprintf(colorcor_dst, "tex%u", stage);
6049         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6050                 settings->op[stage].color_fixup);
6051     }
6052
6053     /* Generate the main shader */
6054     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6055     {
6056         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6057         {
6058             if (!stage) final_combiner_src = "fragment.color.primary";
6059             break;
6060         }
6061
6062         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6063            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6064             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6065         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6066                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6067             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6068         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6069                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6070             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6071         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6072                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6073             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6074         } else {
6075             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6076                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6077                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6078                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6079         }
6080
6081         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6082             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6083                           settings->op[stage].cop, settings->op[stage].carg0,
6084                           settings->op[stage].carg1, settings->op[stage].carg2);
6085             if (!stage)
6086                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6087         }
6088         else if (op_equal)
6089         {
6090             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6091                           settings->op[stage].cop, settings->op[stage].carg0,
6092                           settings->op[stage].carg1, settings->op[stage].carg2);
6093         } else {
6094             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6095                           settings->op[stage].cop, settings->op[stage].carg0,
6096                           settings->op[stage].carg1, settings->op[stage].carg2);
6097             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6098                           settings->op[stage].aop, settings->op[stage].aarg0,
6099                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6100         }
6101     }
6102
6103     if(settings->sRGB_write) {
6104         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6105         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6106         shader_addline(&buffer, "MOV result.color, ret;\n");
6107     } else {
6108         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6109     }
6110
6111     /* Footer */
6112     shader_addline(&buffer, "END\n");
6113
6114     /* Generate the shader */
6115     GL_EXTCALL(glGenProgramsARB(1, &ret));
6116     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6117     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6118             strlen(buffer.buffer), buffer.buffer));
6119     checkGLcall("glProgramStringARB()");
6120
6121     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6122     if (pos != -1)
6123     {
6124         FIXME("Fragment program error at position %d: %s\n\n", pos,
6125               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6126         shader_arb_dump_program_source(buffer.buffer);
6127     }
6128     else
6129     {
6130         GLint native;
6131
6132         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6133         checkGLcall("glGetProgramivARB()");
6134         if (!native) WARN("Program exceeds native resource limits.\n");
6135     }
6136
6137     shader_buffer_free(&buffer);
6138     return ret;
6139 }
6140
6141 static void fragment_prog_arbfp(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6142 {
6143     const struct wined3d_gl_info *gl_info = context->gl_info;
6144     IWineD3DDeviceImpl *device = stateblock->device;
6145     struct shader_arb_priv *priv = device->fragment_priv;
6146     BOOL use_pshader = use_ps(stateblock);
6147     BOOL use_vshader = use_vs(stateblock);
6148     struct ffp_frag_settings settings;
6149     const struct arbfp_ffp_desc *desc;
6150     unsigned int i;
6151
6152     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
6153
6154     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
6155         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6156             /* Reload fixed function constants since they collide with the pixel shader constants */
6157             for(i = 0; i < MAX_TEXTURES; i++) {
6158                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6159             }
6160             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6161             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6162         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6163             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6164         }
6165         return;
6166     }
6167
6168     if(!use_pshader) {
6169         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
6170         gen_ffp_frag_op(stateblock, &settings, FALSE);
6171         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6172         if(!desc) {
6173             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6174             if (!new_desc)
6175             {
6176                 ERR("Out of memory\n");
6177                 return;
6178             }
6179             new_desc->num_textures_used = 0;
6180             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6181             {
6182                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6183                 new_desc->num_textures_used = i;
6184             }
6185
6186             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6187             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
6188             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6189             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6190             desc = new_desc;
6191         }
6192
6193         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6194          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6195          * deactivate it.
6196          */
6197         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6198         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6199         priv->current_fprogram_id = desc->shader;
6200
6201         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6202             /* Reload fixed function constants since they collide with the pixel shader constants */
6203             for(i = 0; i < MAX_TEXTURES; i++) {
6204                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6205             }
6206             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6207             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6208         }
6209         context->last_was_pshader = FALSE;
6210     } else {
6211         context->last_was_pshader = TRUE;
6212     }
6213
6214     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6215      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6216      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6217      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6218      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6219      *
6220      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6221      * shader handler
6222      */
6223     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6224         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6225
6226         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6227             stateblock_apply_state(STATE_VERTEXSHADERCONSTANT, stateblock, context);
6228     }
6229     if (use_pshader) stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
6230 }
6231
6232 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
6233  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
6234  * state table, so we need to handle that with a forwarding function. The other invisible side effect
6235  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
6236  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
6237  */
6238 static void state_arbfp_fog(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6239 {
6240     enum fogsource new_source;
6241
6242     TRACE("state %#x, stateblock %p, context %p\n", state, stateblock, context);
6243
6244     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6245         fragment_prog_arbfp(state, stateblock, context);
6246     }
6247
6248     if (!stateblock->state.render_states[WINED3DRS_FOGENABLE]) return;
6249
6250     if (stateblock->state.render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6251     {
6252         if(use_vs(stateblock)) {
6253             new_source = FOGSOURCE_VS;
6254         }
6255         else
6256         {
6257             if (stateblock->state.render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6258                 new_source = FOGSOURCE_COORD;
6259             else
6260                 new_source = FOGSOURCE_FFP;
6261         }
6262     } else {
6263         new_source = FOGSOURCE_FFP;
6264     }
6265     if(new_source != context->fog_source) {
6266         context->fog_source = new_source;
6267         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6268     }
6269 }
6270
6271 static void textransform(DWORD state, IWineD3DStateBlockImpl *stateblock, struct wined3d_context *context)
6272 {
6273     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6274         fragment_prog_arbfp(state, stateblock, context);
6275     }
6276 }
6277
6278 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6279     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6280     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6281     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6282     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6283     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6284     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6285     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6286     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6287     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6288     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6289     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6290     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6291     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6292     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6293     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6294     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6295     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6296     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6297     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6298     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6299     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6300     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6301     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6302     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6303     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6304     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6305     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6306     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6307     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6308     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6309     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6310     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6311     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6312     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6313     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6314     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6315     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6316     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6317     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6318     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6319     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6320     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6321     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6322     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6323     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6324     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6325     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6326     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6327     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6328     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6329     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6330     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6331     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6332     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6333     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6334     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6335     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6336     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6337     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6338     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6339     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6340     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6341     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6342     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6343     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6344     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6345     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6346     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6347     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6348     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6349     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6350     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6351     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6352     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6353     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6354     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6355     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6356     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6357     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6358     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6359     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6360     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6361     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6362     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6363     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6364     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6365     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6366     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6367     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6368     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6369     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6370     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6371     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6372     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6373     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6374     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6375     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6376     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6377     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6378     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6379     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6380     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6381     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6382     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6383     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6384     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6385     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6386     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6387     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6388     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6389     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6390     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6391     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6392     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6393     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6394     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6395     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6396     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6397     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6398     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6399     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6400     {STATE_SAMPLER(0),                                    { STATE_SAMPLER(0),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6401     {STATE_SAMPLER(1),                                    { STATE_SAMPLER(1),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6402     {STATE_SAMPLER(2),                                    { STATE_SAMPLER(2),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6403     {STATE_SAMPLER(3),                                    { STATE_SAMPLER(3),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6404     {STATE_SAMPLER(4),                                    { STATE_SAMPLER(4),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6405     {STATE_SAMPLER(5),                                    { STATE_SAMPLER(5),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6406     {STATE_SAMPLER(6),                                    { STATE_SAMPLER(6),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6407     {STATE_SAMPLER(7),                                    { STATE_SAMPLER(7),                                   sampler_texdim          }, WINED3D_GL_EXT_NONE             },
6408     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6409     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6410     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6411     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6412     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6413     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6414     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6415     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6416     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6417     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6418     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6419     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6420     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6421     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6422     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6423     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6424     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6425     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6426     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6427 };
6428
6429 const struct fragment_pipeline arbfp_fragment_pipeline = {
6430     arbfp_enable,
6431     arbfp_get_caps,
6432     arbfp_alloc,
6433     arbfp_free,
6434     shader_arb_color_fixup_supported,
6435     arbfp_fragmentstate_template,
6436     TRUE /* We can disable projected textures */
6437 };
6438
6439 struct arbfp_blit_priv {
6440     GLenum yuy2_rect_shader, yuy2_2d_shader;
6441     GLenum uyvy_rect_shader, uyvy_2d_shader;
6442     GLenum yv12_rect_shader, yv12_2d_shader;
6443     GLenum p8_rect_shader, p8_2d_shader;
6444     GLuint palette_texture;
6445 };
6446
6447 static HRESULT arbfp_blit_alloc(IWineD3DDevice *iface) {
6448     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6449     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6450     if(!device->blit_priv) {
6451         ERR("Out of memory\n");
6452         return E_OUTOFMEMORY;
6453     }
6454     return WINED3D_OK;
6455 }
6456
6457 /* Context activation is done by the caller. */
6458 static void arbfp_blit_free(IWineD3DDevice *iface) {
6459     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6460     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6461     struct arbfp_blit_priv *priv = device->blit_priv;
6462
6463     ENTER_GL();
6464     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6465     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6466     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6467     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6468     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6469     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6470     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6471     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6472     checkGLcall("Delete yuv and p8 programs");
6473
6474     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6475     LEAVE_GL();
6476
6477     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6478     device->blit_priv = NULL;
6479 }
6480
6481 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6482         GLenum textype, char *luminance)
6483 {
6484     char chroma;
6485     const char *tex, *texinstr;
6486
6487     if (fixup == COMPLEX_FIXUP_UYVY) {
6488         chroma = 'x';
6489         *luminance = 'w';
6490     } else {
6491         chroma = 'w';
6492         *luminance = 'x';
6493     }
6494     switch(textype) {
6495         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6496         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6497         default:
6498             /* This is more tricky than just replacing the texture type - we have to navigate
6499              * properly in the texture to find the correct chroma values
6500              */
6501             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6502             return FALSE;
6503     }
6504
6505     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6506      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6507      * filtering when we sample the texture.
6508      *
6509      * These are the rules for reading the chroma:
6510      *
6511      * Even pixel: Cr
6512      * Even pixel: U
6513      * Odd pixel: V
6514      *
6515      * So we have to get the sampling x position in non-normalized coordinates in integers
6516      */
6517     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6518         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6519         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6520     } else {
6521         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6522     }
6523     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6524      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6525      * 0.5, so add 0.5.
6526      */
6527     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6528     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6529
6530     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6531      * even and odd pixels respectively
6532      */
6533     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6534     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6535
6536     /* Sample Pixel 1 */
6537     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6538
6539     /* Put the value into either of the chroma values */
6540     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6541     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6542     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6543     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6544
6545     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6546      * the pixel right to the current one. Otherwise, sample the left pixel.
6547      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6548      */
6549     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6550     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6551     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6552
6553     /* Put the value into the other chroma */
6554     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6555     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6556     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6557     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6558
6559     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6560      * the current one and lerp the two U and V values
6561      */
6562
6563     /* This gives the correctly filtered luminance value */
6564     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6565
6566     return TRUE;
6567 }
6568
6569 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6570 {
6571     const char *tex;
6572
6573     switch(textype) {
6574         case GL_TEXTURE_2D:             tex = "2D";     break;
6575         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6576         default:
6577             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6578             return FALSE;
6579     }
6580
6581     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6582      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6583      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6584      * pitch of the luminance plane, the packing into the gl texture is a bit
6585      * unfortunate. If the whole texture is interpreted as luminance data it looks
6586      * approximately like this:
6587      *
6588      *        +----------------------------------+----
6589      *        |                                  |
6590      *        |                                  |
6591      *        |                                  |
6592      *        |                                  |
6593      *        |                                  |   2
6594      *        |            LUMINANCE             |   -
6595      *        |                                  |   3
6596      *        |                                  |
6597      *        |                                  |
6598      *        |                                  |
6599      *        |                                  |
6600      *        +----------------+-----------------+----
6601      *        |                |                 |
6602      *        |  U even rows   |  U odd rows     |
6603      *        |                |                 |   1
6604      *        +----------------+------------------   -
6605      *        |                |                 |   3
6606      *        |  V even rows   |  V odd rows     |
6607      *        |                |                 |
6608      *        +----------------+-----------------+----
6609      *        |                |                 |
6610      *        |     0.5        |       0.5       |
6611      *
6612      * So it appears as if there are 4 chroma images, but in fact the odd rows
6613      * in the chroma images are in the same row as the even ones. So its is
6614      * kinda tricky to read
6615      *
6616      * When reading from rectangle textures, keep in mind that the input y coordinates
6617      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6618      */
6619     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6620             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6621
6622     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6623     /* the chroma planes have only half the width */
6624     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6625
6626     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6627      * the coordinate. Also read the right side of the image when reading odd lines
6628      *
6629      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6630      * bleeding
6631      */
6632     if(textype == GL_TEXTURE_2D) {
6633
6634         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6635
6636         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6637
6638         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6639         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6640
6641         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6642         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6643         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6644         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6645         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6646
6647         /* clamp, keep the half pixel origin in mind */
6648         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6649         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6650         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6651         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6652     } else {
6653         /* Read from [size - size+size/4] */
6654         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6655         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6656
6657         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6658         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6659         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6660         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6661         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6662         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6663
6664         /* Make sure to read exactly from the pixel center */
6665         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6666         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6667
6668         /* Clamp */
6669         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6670         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6671         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6672         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6673         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6674     }
6675     /* Read the texture, put the result into the output register */
6676     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6677     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6678
6679     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6680      * No need to clamp because we're just reusing the already clamped value from above
6681      */
6682     if(textype == GL_TEXTURE_2D) {
6683         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6684     } else {
6685         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6686     }
6687     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6688     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6689
6690     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6691      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6692      * values due to filtering
6693      */
6694     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6695     if(textype == GL_TEXTURE_2D) {
6696         /* Multiply the y coordinate by 2/3 and clamp it */
6697         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6698         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6699         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6700         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6701     } else {
6702         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6703          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6704          * is bigger
6705          */
6706         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6707         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6708         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6709     }
6710     *luminance = 'a';
6711
6712     return TRUE;
6713 }
6714
6715 static GLuint gen_p8_shader(IWineD3DDeviceImpl *device, GLenum textype)
6716 {
6717     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6718     GLenum shader;
6719     struct wined3d_shader_buffer buffer;
6720     struct arbfp_blit_priv *priv = device->blit_priv;
6721     GLint pos;
6722
6723     /* Shader header */
6724     if (!shader_buffer_init(&buffer))
6725     {
6726         ERR("Failed to initialize shader buffer.\n");
6727         return 0;
6728     }
6729
6730     ENTER_GL();
6731     GL_EXTCALL(glGenProgramsARB(1, &shader));
6732     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6733     LEAVE_GL();
6734     if(!shader) {
6735         shader_buffer_free(&buffer);
6736         return 0;
6737     }
6738
6739     shader_addline(&buffer, "!!ARBfp1.0\n");
6740     shader_addline(&buffer, "TEMP index;\n");
6741
6742     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6743     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6744
6745     /* The alpha-component contains the palette index */
6746     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6747         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6748     else
6749         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6750
6751     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6752     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6753
6754     /* Use the alpha-component as an index in the palette to get the final color */
6755     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6756     shader_addline(&buffer, "END\n");
6757
6758     ENTER_GL();
6759     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6760             strlen(buffer.buffer), buffer.buffer));
6761     checkGLcall("glProgramStringARB()");
6762
6763     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6764     if (pos != -1)
6765     {
6766         FIXME("Fragment program error at position %d: %s\n\n", pos,
6767               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6768         shader_arb_dump_program_source(buffer.buffer);
6769     }
6770
6771     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6772         priv->p8_rect_shader = shader;
6773     else
6774         priv->p8_2d_shader = shader;
6775
6776     shader_buffer_free(&buffer);
6777     LEAVE_GL();
6778
6779     return shader;
6780 }
6781
6782 /* Context activation is done by the caller. */
6783 static void upload_palette(IWineD3DSurfaceImpl *surface)
6784 {
6785     BYTE table[256][4];
6786     IWineD3DDeviceImpl *device = surface->resource.device;
6787     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6788     struct arbfp_blit_priv *priv = device->blit_priv;
6789     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6790
6791     d3dfmt_p8_init_palette(surface, table, colorkey);
6792
6793     ENTER_GL();
6794     if (!priv->palette_texture)
6795         glGenTextures(1, &priv->palette_texture);
6796
6797     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6798     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6799
6800     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6801
6802     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6803     /* Make sure we have discrete color levels. */
6804     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6805     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6806     /* Upload the palette */
6807     /* TODO: avoid unneeed uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6808     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6809
6810     /* Switch back to unit 0 in which the 2D texture will be stored. */
6811     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE0));
6812     LEAVE_GL();
6813 }
6814
6815 /* Context activation is done by the caller. */
6816 static GLuint gen_yuv_shader(IWineD3DDeviceImpl *device, enum complex_fixup yuv_fixup, GLenum textype)
6817 {
6818     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6819     GLenum shader;
6820     struct wined3d_shader_buffer buffer;
6821     char luminance_component;
6822     struct arbfp_blit_priv *priv = device->blit_priv;
6823     GLint pos;
6824
6825     /* Shader header */
6826     if (!shader_buffer_init(&buffer))
6827     {
6828         ERR("Failed to initialize shader buffer.\n");
6829         return 0;
6830     }
6831
6832     ENTER_GL();
6833     GL_EXTCALL(glGenProgramsARB(1, &shader));
6834     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6835     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6836     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6837     LEAVE_GL();
6838     if(!shader) {
6839         shader_buffer_free(&buffer);
6840         return 0;
6841     }
6842
6843     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6844      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6845      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6846      * each single pixel it contains, and one U and one V value shared between both
6847      * pixels.
6848      *
6849      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6850      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6851      * take the format into account when generating the read swizzles
6852      *
6853      * Reading the Y value is straightforward - just sample the texture. The hardware
6854      * takes care of filtering in the horizontal and vertical direction.
6855      *
6856      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6857      * because that would mix the U and V values of one pixel or two adjacent pixels.
6858      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6859      * regardless of the filtering setting. Vertical filtering works automatically
6860      * though - the U and V values of two rows are mixed nicely.
6861      *
6862      * Appart of avoiding filtering issues, the code has to know which value it just
6863      * read, and where it can find the other one. To determine this, it checks if
6864      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6865      *
6866      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6867      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6868      *
6869      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6870      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6871      * in an unfiltered situation. Finding the luminance on the other hand requires
6872      * finding out if it is an odd or even pixel. The real drawback of this approach
6873      * is filtering. This would have to be emulated completely in the shader, reading
6874      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6875      * vertically. Beyond that it would require adjustments to the texture handling
6876      * code to deal with the width scaling
6877      */
6878     shader_addline(&buffer, "!!ARBfp1.0\n");
6879     shader_addline(&buffer, "TEMP luminance;\n");
6880     shader_addline(&buffer, "TEMP temp;\n");
6881     shader_addline(&buffer, "TEMP chroma;\n");
6882     shader_addline(&buffer, "TEMP texcrd;\n");
6883     shader_addline(&buffer, "TEMP texcrd2;\n");
6884     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6885     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6886     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6887
6888     switch (yuv_fixup)
6889     {
6890         case COMPLEX_FIXUP_UYVY:
6891         case COMPLEX_FIXUP_YUY2:
6892             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6893             {
6894                 shader_buffer_free(&buffer);
6895                 return 0;
6896             }
6897             break;
6898
6899         case COMPLEX_FIXUP_YV12:
6900             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6901             {
6902                 shader_buffer_free(&buffer);
6903                 return 0;
6904             }
6905             break;
6906
6907         default:
6908             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6909             shader_buffer_free(&buffer);
6910             return 0;
6911     }
6912
6913     /* Calculate the final result. Formula is taken from
6914      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6915      * ranges from -0.5 to 0.5
6916      */
6917     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6918
6919     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6920     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6921     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6922     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6923     shader_addline(&buffer, "END\n");
6924
6925     ENTER_GL();
6926     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6927             strlen(buffer.buffer), buffer.buffer));
6928     checkGLcall("glProgramStringARB()");
6929
6930     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6931     if (pos != -1)
6932     {
6933         FIXME("Fragment program error at position %d: %s\n\n", pos,
6934               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6935         shader_arb_dump_program_source(buffer.buffer);
6936     }
6937     else
6938     {
6939         GLint native;
6940
6941         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6942         checkGLcall("glGetProgramivARB()");
6943         if (!native) WARN("Program exceeds native resource limits.\n");
6944     }
6945
6946     shader_buffer_free(&buffer);
6947     LEAVE_GL();
6948
6949     switch (yuv_fixup)
6950     {
6951         case COMPLEX_FIXUP_YUY2:
6952             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
6953             else priv->yuy2_2d_shader = shader;
6954             break;
6955
6956         case COMPLEX_FIXUP_UYVY:
6957             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
6958             else priv->uyvy_2d_shader = shader;
6959             break;
6960
6961         case COMPLEX_FIXUP_YV12:
6962             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
6963             else priv->yv12_2d_shader = shader;
6964             break;
6965         default:
6966             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
6967     }
6968
6969     return shader;
6970 }
6971
6972 /* Context activation is done by the caller. */
6973 static HRESULT arbfp_blit_set(IWineD3DDevice *iface, IWineD3DSurfaceImpl *surface)
6974 {
6975     GLenum shader;
6976     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
6977     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6978     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
6979     struct arbfp_blit_priv *priv = device->blit_priv;
6980     enum complex_fixup fixup;
6981     GLenum textype = surface->texture_target;
6982
6983     if (!is_complex_fixup(surface->resource.format->color_fixup))
6984     {
6985         TRACE("Fixup:\n");
6986         dump_color_fixup_desc(surface->resource.format->color_fixup);
6987         /* Don't bother setting up a shader for unconverted formats */
6988         ENTER_GL();
6989         glEnable(textype);
6990         checkGLcall("glEnable(textype)");
6991         LEAVE_GL();
6992         return WINED3D_OK;
6993     }
6994
6995     fixup = get_complex_fixup(surface->resource.format->color_fixup);
6996
6997     switch(fixup)
6998     {
6999         case COMPLEX_FIXUP_YUY2:
7000             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7001             break;
7002
7003         case COMPLEX_FIXUP_UYVY:
7004             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7005             break;
7006
7007         case COMPLEX_FIXUP_YV12:
7008             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7009             break;
7010
7011         case COMPLEX_FIXUP_P8:
7012             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7013             if (!shader) shader = gen_p8_shader(device, textype);
7014
7015             upload_palette(surface);
7016             break;
7017
7018         default:
7019             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7020             ENTER_GL();
7021             glEnable(textype);
7022             checkGLcall("glEnable(textype)");
7023             LEAVE_GL();
7024             return E_NOTIMPL;
7025     }
7026
7027     if (!shader) shader = gen_yuv_shader(device, fixup, textype);
7028
7029     ENTER_GL();
7030     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7031     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7032     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7033     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7034     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7035     checkGLcall("glProgramLocalParameter4fvARB");
7036     LEAVE_GL();
7037
7038     return WINED3D_OK;
7039 }
7040
7041 /* Context activation is done by the caller. */
7042 static void arbfp_blit_unset(IWineD3DDevice *iface) {
7043     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) iface;
7044     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7045
7046     ENTER_GL();
7047     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7048     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7049     glDisable(GL_TEXTURE_2D);
7050     checkGLcall("glDisable(GL_TEXTURE_2D)");
7051     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7052     {
7053         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7054         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7055     }
7056     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7057     {
7058         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7059         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7060     }
7061     LEAVE_GL();
7062 }
7063
7064 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum blit_operation blit_op,
7065         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7066         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7067 {
7068     enum complex_fixup src_fixup;
7069
7070     if (blit_op != BLIT_OP_BLIT)
7071     {
7072         TRACE("Unsupported blit_op=%d\n", blit_op);
7073         return FALSE;
7074     }
7075
7076     src_fixup = get_complex_fixup(src_format->color_fixup);
7077     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7078     {
7079         TRACE("Checking support for fixup:\n");
7080         dump_color_fixup_desc(src_format->color_fixup);
7081     }
7082
7083     if (!is_identity_fixup(dst_format->color_fixup))
7084     {
7085         TRACE("Destination fixups are not supported\n");
7086         return FALSE;
7087     }
7088
7089     if (is_identity_fixup(src_format->color_fixup))
7090     {
7091         TRACE("[OK]\n");
7092         return TRUE;
7093     }
7094
7095      /* We only support YUV conversions. */
7096     if (!is_complex_fixup(src_format->color_fixup))
7097     {
7098         TRACE("[FAILED]\n");
7099         return FALSE;
7100     }
7101
7102     switch(src_fixup)
7103     {
7104         case COMPLEX_FIXUP_YUY2:
7105         case COMPLEX_FIXUP_UYVY:
7106         case COMPLEX_FIXUP_YV12:
7107         case COMPLEX_FIXUP_P8:
7108             TRACE("[OK]\n");
7109             return TRUE;
7110
7111         default:
7112             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7113             TRACE("[FAILED]\n");
7114             return FALSE;
7115     }
7116 }
7117
7118 HRESULT arbfp_blit_surface(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *src_surface, const RECT *src_rect,
7119                            IWineD3DSurfaceImpl *dst_surface, const RECT *dst_rect_in, enum blit_operation blit_op,
7120                            DWORD Filter)
7121 {
7122     IWineD3DSwapChainImpl *dst_swapchain;
7123     struct wined3d_context *context;
7124     RECT dst_rect = *dst_rect_in;
7125
7126     /* Now load the surface */
7127     surface_internal_preload(src_surface, SRGB_RGB);
7128
7129     /* Activate the destination context, set it up for blitting */
7130     context = context_acquire(device, dst_surface);
7131     context_apply_blit_state(context, device);
7132
7133     /* The coordinates of the ddraw front buffer are always fullscreen ('screen coordinates',
7134      * while OpenGL coordinates are window relative.
7135      * Also beware of the origin difference(top left vs bottom left).
7136      * Also beware that the front buffer's surface size is screen width x screen height,
7137      * whereas the real gl drawable size is the size of the window. */
7138     dst_swapchain = dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7139             ? dst_surface->container.u.swapchain : NULL;
7140     if (dst_swapchain && dst_surface == dst_swapchain->front_buffer)
7141         surface_translate_frontbuffer_coords(dst_surface, context->win_handle, &dst_rect);
7142
7143     arbfp_blit_set((IWineD3DDevice *)device, src_surface);
7144
7145     ENTER_GL();
7146
7147     /* Draw a textured quad */
7148     draw_textured_quad(src_surface, src_rect, &dst_rect, Filter);
7149
7150     LEAVE_GL();
7151
7152     /* Leave the opengl state valid for blitting */
7153     arbfp_blit_unset((IWineD3DDevice *)device);
7154
7155     if (wined3d_settings.strict_draw_ordering || (dst_swapchain
7156             && (dst_surface == dst_swapchain->front_buffer
7157             || dst_swapchain->num_contexts > 1)))
7158         wglFlush(); /* Flush to ensure ordering across contexts. */
7159
7160     context_release(context);
7161
7162     surface_modify_location(dst_surface, SFLAG_INDRAWABLE, TRUE);
7163     return WINED3D_OK;
7164 }
7165
7166 /* Do not call while under the GL lock. */
7167 static HRESULT arbfp_blit_color_fill(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *dst_surface,
7168         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7169 {
7170     FIXME("Color filling not implemented by arbfp_blit\n");
7171     return WINED3DERR_INVALIDCALL;
7172 }
7173
7174 const struct blit_shader arbfp_blit = {
7175     arbfp_blit_alloc,
7176     arbfp_blit_free,
7177     arbfp_blit_set,
7178     arbfp_blit_unset,
7179     arbfp_blit_supported,
7180     arbfp_blit_color_fill
7181 };