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 #include "wine/port.h"
  32
  33 #include <math.h>
  34 #include <stdio.h>
  35
  36 #include "wined3d_private.h"
  37
  38 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  41 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  42
  43 /* Extract a line. Note that this modifies the source string. */
  44 static char *get_line(char **ptr)
  45 {
  46     char *p, *q;
  47
  48     p = *ptr;
  49     if (!(q = strstr(p, "\n")))
  50     {
  51         if (!*p) return NULL;
  52         *ptr += strlen(p);
  53         return p;
  54     }
  55     *q = '\0';
  56     *ptr = q + 1;
  57
  58     return p;
  59 }
  60
  61 static void shader_arb_dump_program_source(const char *source)
  62 {
  63     ULONG source_size;
  64     char *ptr, *line, *tmp;
  65
  66     source_size = strlen(source) + 1;
  67     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
  68     if (!tmp)
  69     {
  70         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
  71         return;
  72     }
  73     memcpy(tmp, source, source_size);
  74
  75     ptr = tmp;
  76     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
  77     FIXME("\n");
  78
  79     HeapFree(GetProcessHeap(), 0, tmp);
  80 }
  81
  82 enum arb_helper_value
  83 {
  84     ARB_ZERO,
  85     ARB_ONE,
  86     ARB_TWO,
  87     ARB_0001,
  88     ARB_EPS,
  89
  90     ARB_VS_REL_OFFSET
  91 };
  92
  93 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
  94 {
  95     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
  96     {
  97         ERR("Geometry shaders are unsupported\n");
  98         return "bad";
  99     }
 100
 101     if (shader == WINED3D_SHADER_TYPE_PIXEL)
 102     {
 103         switch (value)
 104         {
 105             case ARB_ZERO: return "ps_helper_const.x";
 106             case ARB_ONE: return "ps_helper_const.y";
 107             case ARB_TWO: return "coefmul.x";
 108             case ARB_0001: return "ps_helper_const.xxxy";
 109             case ARB_EPS: return "ps_helper_const.z";
 110             default: break;
 111         }
 112     }
 113     else
 114     {
 115         switch (value)
 116         {
 117             case ARB_ZERO: return "helper_const.x";
 118             case ARB_ONE: return "helper_const.y";
 119             case ARB_TWO: return "helper_const.z";
 120             case ARB_EPS: return "helper_const.w";
 121             case ARB_0001: return "helper_const.xxxy";
 122             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
 123         }
 124     }
 125     FIXME("Unmanaged %s shader helper constant requested: %u\n",
 126           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
 127     switch (value)
 128     {
 129         case ARB_ZERO: return "0.0";
 130         case ARB_ONE: return "1.0";
 131         case ARB_TWO: return "2.0";
 132         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
 133         case ARB_EPS: return "1e-8";
 134         default: return "bad";
 135     }
 136 }
 137
 138 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
 139 {
 140     return state->lowest_disabled_stage < 7;
 141 }
 142
 143 /* ARB_program_shader private data */
 144
 145 struct control_frame
 146 {
 147     struct                          list entry;
 148     enum
 149     {
 150         IF,
 151         IFC,
 152         LOOP,
 153         REP
 154     } type;
 155     BOOL                            muting;
 156     BOOL                            outer_loop;
 157     union
 158     {
 159         unsigned int                loop;
 160         unsigned int                ifc;
 161     } no;
 162     struct wined3d_shader_loop_control loop_control;
 163     BOOL                            had_else;
 164 };
 165
 166 struct arb_ps_np2fixup_info
 167 {
 168     struct ps_np2fixup_info         super;
 169     /* For ARB we need a offset value:
 170      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 171      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 172      * array we need an offset to the index inside the program local parameter array. */
 173     UINT                            offset;
 174 };
 175
 176 struct arb_ps_compile_args
 177 {
 178     struct ps_compile_args          super;
 179     WORD                            bools;
 180     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 181     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 182 };
 183
 184 struct stb_const_desc
 185 {
 186     unsigned char           texunit;
 187     UINT                    const_num;
 188 };
 189
 190 struct arb_ps_compiled_shader
 191 {
 192     struct arb_ps_compile_args      args;
 193     struct arb_ps_np2fixup_info     np2fixup_info;
 194     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 195     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 196     UINT                            int_consts[MAX_CONST_I];
 197     GLuint                          prgId;
 198     UINT                            ycorrection;
 199     unsigned char                   numbumpenvmatconsts;
 200     char                            num_int_consts;
 201 };
 202
 203 struct arb_vs_compile_args
 204 {
 205     struct vs_compile_args          super;
 206     union
 207     {
 208         struct
 209         {
 210             WORD                    bools;
 211             unsigned char           clip_texcoord;
 212             unsigned char           clipplane_mask;
 213         }                           boolclip;
 214         DWORD                       boolclip_compare;
 215     } clip;
 216     DWORD                           ps_signature;
 217     union
 218     {
 219         unsigned char               samplers[4];
 220         DWORD                       samplers_compare;
 221     } vertex;
 222     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 223 };
 224
 225 struct arb_vs_compiled_shader
 226 {
 227     struct arb_vs_compile_args      args;
 228     GLuint                          prgId;
 229     UINT                            int_consts[MAX_CONST_I];
 230     char                            num_int_consts;
 231     char                            need_color_unclamp;
 232     UINT                            pos_fixup;
 233 };
 234
 235 struct recorded_instruction
 236 {
 237     struct wined3d_shader_instruction ins;
 238     struct list entry;
 239 };
 240
 241 struct shader_arb_ctx_priv
 242 {
 243     char addr_reg[20];
 244     enum
 245     {
 246         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 247         ARB,
 248         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 249         NV2,
 250         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 251         NV3
 252     } target_version;
 253
 254     const struct arb_vs_compile_args    *cur_vs_args;
 255     const struct arb_ps_compile_args    *cur_ps_args;
 256     const struct arb_ps_compiled_shader *compiled_fprog;
 257     const struct arb_vs_compiled_shader *compiled_vprog;
 258     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 259     struct list                         control_frames;
 260     struct list                         record;
 261     BOOL                                recording;
 262     BOOL                                muted;
 263     unsigned int                        num_loops, loop_depth, num_ifcs;
 264     int                                 aL;
 265
 266     unsigned int                        vs_clipplanes;
 267     BOOL                                footer_written;
 268     BOOL                                in_main_func;
 269
 270     /* For 3.0 vertex shaders */
 271     const char                          *vs_output[MAX_REG_OUTPUT];
 272     /* For 2.x and earlier vertex shaders */
 273     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 274
 275     /* 3.0 pshader input for compatibility with fixed function */
 276     const char                          *ps_input[MAX_REG_INPUT];
 277 };
 278
 279 struct ps_signature
 280 {
 281     struct wined3d_shader_signature_element *sig;
 282     DWORD                               idx;
 283     struct wine_rb_entry                entry;
 284 };
 285
 286 struct arb_pshader_private {
 287     struct arb_ps_compiled_shader   *gl_shaders;
 288     UINT                            num_gl_shaders, shader_array_size;
 289     DWORD                           input_signature_idx;
 290     DWORD                           clipplane_emulation;
 291     BOOL                            clamp_consts;
 292 };
 293
 294 struct arb_vshader_private {
 295     struct arb_vs_compiled_shader   *gl_shaders;
 296     UINT                            num_gl_shaders, shader_array_size;
 297     UINT rel_offset;
 298 };
 299
 300 struct shader_arb_priv
 301 {
 302     GLuint                  current_vprogram_id;
 303     GLuint                  current_fprogram_id;
 304     const struct arb_ps_compiled_shader *compiled_fprog;
 305     const struct arb_vs_compiled_shader *compiled_vprog;
 306     GLuint                  depth_blt_vprogram_id;
 307     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 308     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 309     BOOL                    use_arbfp_fixed_func;
 310     struct wine_rb_tree     fragment_shaders;
 311     BOOL                    last_ps_const_clamped;
 312     BOOL                    last_vs_color_unclamp;
 313
 314     struct wine_rb_tree     signature_tree;
 315     DWORD ps_sig_number;
 316
 317     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
 318     char *vshader_const_dirty, *pshader_const_dirty;
 319     const struct wined3d_context *last_context;
 320 };
 321
 322 /* GL locking for state handlers is done by the caller. */
 323 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 324         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 325 {
 326     if (shader_data->rel_offset) return TRUE;
 327     if (!reg_maps->usesmova) return FALSE;
 328     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 329 }
 330
 331 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 332 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 333 {
 334     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 335             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 336 }
 337
 338 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 339         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 340 {
 341     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 342     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 343     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 344     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 345     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 346     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 347     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 348     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 349     return FALSE;
 350 }
 351
 352 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 353         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 354 {
 355     unsigned int ret = 1;
 356     /* We use one PARAM for the pos fixup, and in some cases one to load
 357      * some immediate values into the shader. */
 358     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 359     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 360     return ret;
 361 }
 362
 363 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 364  * When constant_list == NULL, it will load all the constants.
 365  *
 366  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 367  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 368  */
 369 /* GL locking is done by the caller */
 370 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 371         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 372         const float *constants, char *dirty_consts)
 373 {
 374     struct wined3d_shader_lconst *lconst;
 375     DWORD i, j;
 376     unsigned int ret;
 377
 378     if (TRACE_ON(d3d_constants))
 379     {
 380         for(i = 0; i < max_constants; i++) {
 381             if(!dirty_consts[i]) continue;
 382             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 383                         constants[i * 4 + 0], constants[i * 4 + 1],
 384                         constants[i * 4 + 2], constants[i * 4 + 3]);
 385         }
 386     }
 387
 388     i = 0;
 389
 390     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 391     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 392     {
 393         float lcl_const[4];
 394         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 395          * shaders, the first 8 constants are marked dirty for reload
 396          */
 397         for(; i < min(8, max_constants); i++) {
 398             if(!dirty_consts[i]) continue;
 399             dirty_consts[i] = 0;
 400
 401             j = 4 * i;
 402             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 403             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 404             else lcl_const[0] = constants[j + 0];
 405
 406             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 407             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 408             else lcl_const[1] = constants[j + 1];
 409
 410             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 411             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 412             else lcl_const[2] = constants[j + 2];
 413
 414             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 415             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 416             else lcl_const[3] = constants[j + 3];
 417
 418             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 419         }
 420
 421         /* If further constants are dirty, reload them without clamping.
 422          *
 423          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 424          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 425          * above would always re-check the first 8 constants since max_constant remains at the init
 426          * value
 427          */
 428     }
 429
 430     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 431     {
 432         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 433          * or just reloading *all* constants at once
 434          *
 435         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 436          */
 437         for(; i < max_constants; i++) {
 438             if(!dirty_consts[i]) continue;
 439
 440             /* Find the next block of dirty constants */
 441             dirty_consts[i] = 0;
 442             j = i;
 443             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 444                 dirty_consts[i] = 0;
 445             }
 446
 447             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 448         }
 449     } else {
 450         for(; i < max_constants; i++) {
 451             if(dirty_consts[i]) {
 452                 dirty_consts[i] = 0;
 453                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 454             }
 455         }
 456     }
 457     checkGLcall("glProgramEnvParameter4fvARB()");
 458
 459     /* Load immediate constants */
 460     if (shader->load_local_constsF)
 461     {
 462         if (TRACE_ON(d3d_shader))
 463         {
 464             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 465             {
 466                 GLfloat* values = (GLfloat*)lconst->value;
 467                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 468                         values[0], values[1], values[2], values[3]);
 469             }
 470         }
 471         /* Immediate constants are clamped for 1.X shaders at loading times */
 472         ret = 0;
 473         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 474         {
 475             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 476             ret = max(ret, lconst->idx + 1);
 477             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 478         }
 479         checkGLcall("glProgramEnvParameter4fvARB()");
 480         return ret; /* The loaded immediate constants need reloading for the next shader */
 481     } else {
 482         return 0; /* No constants are dirty now */
 483     }
 484 }
 485
 486 /**
 487  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 488  */
 489 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 490         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 491 {
 492     const struct shader_arb_priv * priv = shader_priv;
 493
 494     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 495     if (!use_ps(state)) return;
 496
 497     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 498         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 499         UINT i;
 500         WORD active = fixup->super.active;
 501         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 502
 503         for (i = 0; active; active >>= 1, ++i)
 504         {
 505             const struct wined3d_texture *tex = state->textures[i];
 506             const unsigned char idx = fixup->super.idx[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             {
 518                 tex_dim[2] = tex->pow2_matrix[0];
 519                 tex_dim[3] = tex->pow2_matrix[5];
 520             }
 521             else
 522             {
 523                 tex_dim[0] = tex->pow2_matrix[0];
 524                 tex_dim[1] = tex->pow2_matrix[5];
 525             }
 526         }
 527
 528         for (i = 0; i < fixup->super.num_consts; ++i) {
 529             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 530                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 531         }
 532     }
 533 }
 534
 535 /* GL locking is done by the caller. */
 536 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 537         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 538 {
 539     const struct wined3d_gl_info *gl_info = context->gl_info;
 540     unsigned char i;
 541
 542     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 543     {
 544         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 545
 546         /* The state manager takes care that this function is always called if the bump env matrix changes */
 547         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 548         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 549                 gl_shader->bumpenvmatconst[i].const_num, data));
 550
 551         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 552         {
 553             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 554              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 555              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 556              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 557             */
 558             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 559             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 560                     gl_shader->luminanceconst[i].const_num, scale));
 561         }
 562     }
 563     checkGLcall("Load bumpmap consts");
 564
 565     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 566     {
 567         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 568         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 569         * ycorrection.z: 1.0
 570         * ycorrection.w: 0.0
 571         */
 572         float val[4];
 573         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 574         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 575         val[2] = 1.0f;
 576         val[3] = 0.0f;
 577         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 578         checkGLcall("y correction loading");
 579     }
 580
 581     if (!gl_shader->num_int_consts) return;
 582
 583     for(i = 0; i < MAX_CONST_I; i++)
 584     {
 585         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 586         {
 587             float val[4];
 588             val[0] = (float)state->ps_consts_i[4 * i];
 589             val[1] = (float)state->ps_consts_i[4 * i + 1];
 590             val[2] = (float)state->ps_consts_i[4 * i + 2];
 591             val[3] = -1.0f;
 592
 593             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 594         }
 595     }
 596     checkGLcall("Load ps int consts");
 597 }
 598
 599 /* GL locking is done by the caller. */
 600 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 601         const struct wined3d_context *context, const struct wined3d_state *state)
 602 {
 603     const struct wined3d_gl_info *gl_info = context->gl_info;
 604     float position_fixup[4];
 605     unsigned char i;
 606
 607     /* Upload the position fixup */
 608     shader_get_position_fixup(context, state, position_fixup);
 609     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 610
 611     if (!gl_shader->num_int_consts) return;
 612
 613     for(i = 0; i < MAX_CONST_I; i++)
 614     {
 615         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 616         {
 617             float val[4];
 618             val[0] = (float)state->vs_consts_i[4 * i];
 619             val[1] = (float)state->vs_consts_i[4 * i + 1];
 620             val[2] = (float)state->vs_consts_i[4 * i + 2];
 621             val[3] = -1.0f;
 622
 623             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 624         }
 625     }
 626     checkGLcall("Load vs int consts");
 627 }
 628
 629 /**
 630  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 631  *
 632  * We only support float constants in ARB at the moment, so don't
 633  * worry about the Integers or Booleans
 634  */
 635 /* GL locking is done by the caller (state handler) */
 636 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 637 {
 638     struct wined3d_device *device = context->swapchain->device;
 639     const struct wined3d_state *state = &device->stateBlock->state;
 640     const struct wined3d_gl_info *gl_info = context->gl_info;
 641     struct shader_arb_priv *priv = device->shader_priv;
 642
 643     if (context != priv->last_context)
 644     {
 645         memset(priv->vshader_const_dirty, 1,
 646                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 647         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 648
 649         memset(priv->pshader_const_dirty, 1,
 650                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 651         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 652
 653         priv->last_context = context;
 654     }
 655
 656     if (useVertexShader)
 657     {
 658         struct wined3d_shader *vshader = state->vertex_shader;
 659         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 660
 661         /* Load DirectX 9 float constants for vertex shader */
 662         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 663                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 664         shader_arb_vs_local_constants(gl_shader, context, state);
 665     }
 666
 667     if (usePixelShader)
 668     {
 669         struct wined3d_shader *pshader = state->pixel_shader;
 670         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 671         UINT rt_height = state->fb->render_targets[0]->resource.height;
 672
 673         /* Load DirectX 9 float constants for pixel shader */
 674         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 675                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 676         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 677     }
 678 }
 679
 680 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 681 {
 682     struct wined3d_context *context = context_get_current();
 683     struct shader_arb_priv *priv = device->shader_priv;
 684
 685     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 686      * context. On a context switch the old context will be fully dirtified */
 687     if (!context || context->swapchain->device != device) return;
 688
 689     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 690     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 691 }
 692
 693 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 694 {
 695     struct wined3d_context *context = context_get_current();
 696     struct shader_arb_priv *priv = device->shader_priv;
 697
 698     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 699      * context. On a context switch the old context will be fully dirtified */
 700     if (!context || context->swapchain->device != device) return;
 701
 702     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 703     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 704 }
 705
 706 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 707 {
 708     const struct wined3d_shader_lconst *lconst;
 709     DWORD *ret;
 710     DWORD idx = 0;
 711
 712     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 713         return NULL;
 714
 715     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 716     if (!ret)
 717     {
 718         ERR("Out of memory\n");
 719         return NULL;
 720     }
 721
 722     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 723     {
 724         ret[lconst->idx] = idx++;
 725     }
 726     return ret;
 727 }
 728
 729 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 730 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 731         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 732         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 733         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 734 {
 735     DWORD i, next_local = 0;
 736     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 737     const struct wined3d_shader_lconst *lconst;
 738     unsigned max_constantsF;
 739     DWORD map;
 740
 741     /* In pixel shaders, all private constants are program local, we don't need anything
 742      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 743      * If we need a private constant the GL implementation will squeeze it in somewhere
 744      *
 745      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 746      * immediate values. The posFixup is loaded using program.env for now, so always
 747      * subtract one from the number of constants. If the shader uses indirect addressing,
 748      * account for the helper const too because we have to declare all available d3d constants
 749      * and don't know which are actually used.
 750      */
 751     if (pshader)
 752     {
 753         max_constantsF = gl_info->limits.arb_ps_native_constants;
 754         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 755         if (max_constantsF < 24)
 756             max_constantsF = gl_info->limits.arb_ps_float_constants;
 757     }
 758     else
 759     {
 760         const struct arb_vshader_private *shader_data = shader->backend_data;
 761         max_constantsF = gl_info->limits.arb_vs_native_constants;
 762         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 763          * Also prevents max_constantsF from becoming less than 0 and
 764          * wrapping . */
 765         if (max_constantsF < 96)
 766             max_constantsF = gl_info->limits.arb_vs_float_constants;
 767
 768         if (reg_maps->usesrelconstF)
 769         {
 770             DWORD highest_constf = 0, clip_limit;
 771
 772             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 773             max_constantsF -= count_bits(reg_maps->integer_constants);
 774
 775             for (i = 0; i < shader->limits.constant_float; ++i)
 776             {
 777                 DWORD idx = i >> 5;
 778                 DWORD shift = i & 0x1f;
 779                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 780             }
 781
 782             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 783             {
 784                 if(ctx->cur_vs_args->super.clip_enabled)
 785                     clip_limit = gl_info->limits.clipplanes;
 786                 else
 787                     clip_limit = 0;
 788             }
 789             else
 790             {
 791                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 792                 clip_limit = min(count_bits(mask), 4);
 793             }
 794             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 795             max_constantsF -= *num_clipplanes;
 796             if(*num_clipplanes < clip_limit)
 797             {
 798                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 799             }
 800         }
 801         else
 802         {
 803             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 804             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 805         }
 806     }
 807
 808     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 809     {
 810         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 811     }
 812
 813     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 814     {
 815         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 816     }
 817
 818     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 819     {
 820         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 821         {
 822             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 823         }
 824     }
 825
 826     /* Load local constants using the program-local space,
 827      * this avoids reloading them each time the shader is used
 828      */
 829     if (lconst_map)
 830     {
 831         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 832         {
 833             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 834                            lconst_map[lconst->idx]);
 835             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 836         }
 837     }
 838
 839     /* After subtracting privately used constants from the hardware limit(they are loaded as
 840      * local constants), make sure the shader doesn't violate the env constant limit
 841      */
 842     if(pshader)
 843     {
 844         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 845     }
 846     else
 847     {
 848         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 849     }
 850
 851     /* Avoid declaring more constants than needed */
 852     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 853
 854     /* we use the array-based constants array if the local constants are marked for loading,
 855      * because then we use indirect addressing, or when the local constant list is empty,
 856      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 857      * local constants do not declare the loaded constants as an array because ARB compilers usually
 858      * do not optimize unused constants away
 859      */
 860     if (reg_maps->usesrelconstF)
 861     {
 862         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 863         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 864                     max_constantsF, max_constantsF - 1);
 865     } else {
 866         for(i = 0; i < max_constantsF; i++) {
 867             DWORD idx, mask;
 868             idx = i >> 5;
 869             mask = 1 << (i & 0x1f);
 870             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 871             {
 872                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 873             }
 874         }
 875     }
 876
 877     return next_local;
 878 }
 879
 880 static const char * const shift_tab[] = {
 881     "dummy",     /*  0 (none) */
 882     "coefmul.x", /*  1 (x2)   */
 883     "coefmul.y", /*  2 (x4)   */
 884     "coefmul.z", /*  3 (x8)   */
 885     "coefmul.w", /*  4 (x16)  */
 886     "dummy",     /*  5 (x32)  */
 887     "dummy",     /*  6 (x64)  */
 888     "dummy",     /*  7 (x128) */
 889     "dummy",     /*  8 (d256) */
 890     "dummy",     /*  9 (d128) */
 891     "dummy",     /* 10 (d64)  */
 892     "dummy",     /* 11 (d32)  */
 893     "coefdiv.w", /* 12 (d16)  */
 894     "coefdiv.z", /* 13 (d8)   */
 895     "coefdiv.y", /* 14 (d4)   */
 896     "coefdiv.x"  /* 15 (d2)   */
 897 };
 898
 899 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 900         const struct wined3d_shader_dst_param *dst, char *write_mask)
 901 {
 902     char *ptr = write_mask;
 903
 904     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 905     {
 906         *ptr++ = '.';
 907         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 908         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 909         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 910         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 911     }
 912
 913     *ptr = '\0';
 914 }
 915
 916 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 917 {
 918     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 919      * but addressed as "rgba". To fix this we need to swap the register's x
 920      * and z components. */
 921     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 922     char *ptr = swizzle_str;
 923
 924     /* swizzle bits fields: wwzzyyxx */
 925     DWORD swizzle = param->swizzle;
 926     DWORD swizzle_x = swizzle & 0x03;
 927     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 928     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 929     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 930
 931     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 932      * generate a swizzle string. Unless we need to our own swizzling. */
 933     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 934     {
 935         *ptr++ = '.';
 936         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 937             *ptr++ = swizzle_chars[swizzle_x];
 938         } else {
 939             *ptr++ = swizzle_chars[swizzle_x];
 940             *ptr++ = swizzle_chars[swizzle_y];
 941             *ptr++ = swizzle_chars[swizzle_z];
 942             *ptr++ = swizzle_chars[swizzle_w];
 943         }
 944     }
 945
 946     *ptr = '\0';
 947 }
 948
 949 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 950 {
 951     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 952     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 953
 954     if (!strcmp(priv->addr_reg, src)) return;
 955
 956     strcpy(priv->addr_reg, src);
 957     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 958 }
 959
 960 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 961         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 962
 963 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 964         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 965 {
 966     /* oPos, oFog and oPts in D3D */
 967     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
 968     const struct wined3d_shader *shader = ins->ctx->shader;
 969     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 970     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 971     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 972
 973     *is_color = FALSE;
 974
 975     switch (reg->type)
 976     {
 977         case WINED3DSPR_TEMP:
 978             sprintf(register_name, "R%u", reg->idx);
 979             break;
 980
 981         case WINED3DSPR_INPUT:
 982             if (pshader)
 983             {
 984                 if (reg_maps->shader_version.major < 3)
 985                 {
 986                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 987                     else strcpy(register_name, "fragment.color.secondary");
 988                 }
 989                 else
 990                 {
 991                     if(reg->rel_addr)
 992                     {
 993                         char rel_reg[50];
 994                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 995
 996                         if (!strcmp(rel_reg, "**aL_emul**"))
 997                         {
 998                             DWORD idx = ctx->aL + reg->idx;
 999                             if(idx < MAX_REG_INPUT)
1000                             {
1001                                 strcpy(register_name, ctx->ps_input[idx]);
1002                             }
1003                             else
1004                             {
1005                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1006                                 sprintf(register_name, "out_of_bounds_%u", idx);
1007                             }
1008                         }
1009                         else if (reg_maps->input_registers & 0x0300)
1010                         {
1011                             /* There are two ways basically:
1012                              *
1013                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1014                              *    That means trouble if the loop also contains a breakc or if the control values
1015                              *    aren't local constants.
1016                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1017                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1018                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1019                              *    ADAC to load the condition code register and pop it again afterwards
1020                              */
1021                             FIXME("Relative input register addressing with more than 8 registers\n");
1022
1023                             /* This is better than nothing for now */
1024                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1025                         }
1026                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1027                         {
1028                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1029                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1030                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1031                              * proper varyings, or loop unrolling
1032                              *
1033                              * For now use the texcoords and hope for the best
1034                              */
1035                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1036                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1037                         }
1038                         else
1039                         {
1040                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1041                              * pulls GL_NV_fragment_program2 in
1042                              */
1043                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1044                         }
1045                     }
1046                     else
1047                     {
1048                         if(reg->idx < MAX_REG_INPUT)
1049                         {
1050                             strcpy(register_name, ctx->ps_input[reg->idx]);
1051                         }
1052                         else
1053                         {
1054                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1055                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1056                         }
1057                     }
1058                 }
1059             }
1060             else
1061             {
1062                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1063                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1064             }
1065             break;
1066
1067         case WINED3DSPR_CONST:
1068             if (!pshader && reg->rel_addr)
1069             {
1070                 const struct arb_vshader_private *shader_data = shader->backend_data;
1071                 UINT rel_offset = shader_data->rel_offset;
1072                 BOOL aL = FALSE;
1073                 char rel_reg[50];
1074                 if (reg_maps->shader_version.major < 2)
1075                 {
1076                     sprintf(rel_reg, "A0.x");
1077                 } else {
1078                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1079                     if(ctx->target_version == ARB) {
1080                         if (!strcmp(rel_reg, "**aL_emul**"))
1081                         {
1082                             aL = TRUE;
1083                         } else {
1084                             shader_arb_request_a0(ins, rel_reg);
1085                             sprintf(rel_reg, "A0.x");
1086                         }
1087                     }
1088                 }
1089                 if(aL)
1090                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1091                 else if (reg->idx >= rel_offset)
1092                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1093                 else
1094                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1095             }
1096             else
1097             {
1098                 if (reg_maps->usesrelconstF)
1099                     sprintf(register_name, "C[%u]", reg->idx);
1100                 else
1101                     sprintf(register_name, "C%u", reg->idx);
1102             }
1103             break;
1104
1105         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1106             if (pshader)
1107             {
1108                 if (reg_maps->shader_version.major == 1
1109                         && reg_maps->shader_version.minor <= 3)
1110                 {
1111                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1112                      * and as source to most instructions. For some instructions it is the texcoord
1113                      * input. Those instructions know about the special use
1114                      */
1115                     sprintf(register_name, "T%u", reg->idx);
1116                 } else {
1117                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1118                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1119                 }
1120             }
1121             else
1122             {
1123                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1124                 {
1125                     sprintf(register_name, "A%u", reg->idx);
1126                 }
1127                 else
1128                 {
1129                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1130                 }
1131             }
1132             break;
1133
1134         case WINED3DSPR_COLOROUT:
1135             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1136             {
1137                 strcpy(register_name, "TMP_COLOR");
1138             }
1139             else
1140             {
1141                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1142                 if (reg_maps->rt_mask > 1)
1143                 {
1144                     sprintf(register_name, "result.color[%u]", reg->idx);
1145                 }
1146                 else
1147                 {
1148                     strcpy(register_name, "result.color");
1149                 }
1150             }
1151             break;
1152
1153         case WINED3DSPR_RASTOUT:
1154             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1155             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1156             break;
1157
1158         case WINED3DSPR_DEPTHOUT:
1159             strcpy(register_name, "result.depth");
1160             break;
1161
1162         case WINED3DSPR_ATTROUT:
1163         /* case WINED3DSPR_OUTPUT: */
1164             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1165             else strcpy(register_name, ctx->color_output[reg->idx]);
1166             break;
1167
1168         case WINED3DSPR_TEXCRDOUT:
1169             if (pshader)
1170             {
1171                 sprintf(register_name, "oT[%u]", reg->idx);
1172             }
1173             else
1174             {
1175                 if (reg_maps->shader_version.major < 3)
1176                 {
1177                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1178                 }
1179                 else
1180                 {
1181                     strcpy(register_name, ctx->vs_output[reg->idx]);
1182                 }
1183             }
1184             break;
1185
1186         case WINED3DSPR_LOOP:
1187             if(ctx->target_version >= NV2)
1188             {
1189                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1190                 if(pshader) sprintf(register_name, "A0.x");
1191                 else sprintf(register_name, "aL.y");
1192             }
1193             else
1194             {
1195                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1196                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1197                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1198                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1199                  * indexing
1200                  */
1201                 sprintf(register_name, "**aL_emul**");
1202             }
1203
1204             break;
1205
1206         case WINED3DSPR_CONSTINT:
1207             sprintf(register_name, "I%u", reg->idx);
1208             break;
1209
1210         case WINED3DSPR_MISCTYPE:
1211             if (!reg->idx)
1212             {
1213                 sprintf(register_name, "vpos");
1214             }
1215             else if(reg->idx == 1)
1216             {
1217                 sprintf(register_name, "fragment.facing.x");
1218             }
1219             else
1220             {
1221                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1222             }
1223             break;
1224
1225         default:
1226             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1227             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1228             break;
1229     }
1230 }
1231
1232 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1233         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1234 {
1235     char register_name[255];
1236     char write_mask[6];
1237     BOOL is_color;
1238
1239     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1240     strcpy(str, register_name);
1241
1242     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1243     strcat(str, write_mask);
1244 }
1245
1246 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1247 {
1248     switch(channel_source)
1249     {
1250         case CHANNEL_SOURCE_ZERO: return "0";
1251         case CHANNEL_SOURCE_ONE: return "1";
1252         case CHANNEL_SOURCE_X: return "x";
1253         case CHANNEL_SOURCE_Y: return "y";
1254         case CHANNEL_SOURCE_Z: return "z";
1255         case CHANNEL_SOURCE_W: return "w";
1256         default:
1257             FIXME("Unhandled channel source %#x\n", channel_source);
1258             return "undefined";
1259     }
1260 }
1261
1262 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1263         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1264 {
1265     DWORD mask;
1266
1267     if (is_complex_fixup(fixup))
1268     {
1269         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1270         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1271         return;
1272     }
1273
1274     mask = 0;
1275     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1276     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1277     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1278     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1279     mask &= dst_mask;
1280
1281     if (mask)
1282     {
1283         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1284                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1285                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1286     }
1287
1288     mask = 0;
1289     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1290     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1291     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1292     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1293     mask &= dst_mask;
1294
1295     if (mask)
1296     {
1297         char reg_mask[6];
1298         char *ptr = reg_mask;
1299
1300         if (mask != WINED3DSP_WRITEMASK_ALL)
1301         {
1302             *ptr++ = '.';
1303             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1304             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1305             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1306             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1307         }
1308         *ptr = '\0';
1309
1310         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1311     }
1312 }
1313
1314 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1315 {
1316     DWORD mod;
1317     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1318     if (!ins->dst_count) return "";
1319
1320     mod = ins->dst[0].modifiers;
1321
1322     /* Silently ignore PARTIALPRECISION if its not supported */
1323     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1324
1325     if(mod & WINED3DSPDM_MSAMPCENTROID)
1326     {
1327         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1328         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1329     }
1330
1331     switch(mod)
1332     {
1333         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1334             return "H_SAT";
1335
1336         case WINED3DSPDM_SATURATE:
1337             return "_SAT";
1338
1339         case WINED3DSPDM_PARTIALPRECISION:
1340             return "H";
1341
1342         case 0:
1343             return "";
1344
1345         default:
1346             FIXME("Unknown modifiers 0x%08x\n", mod);
1347             return "";
1348     }
1349 }
1350
1351 #define TEX_PROJ        0x1
1352 #define TEX_BIAS        0x2
1353 #define TEX_LOD         0x4
1354 #define TEX_DERIV       0x10
1355
1356 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1357         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1358 {
1359     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1360     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1361     const struct wined3d_shader *shader = ins->ctx->shader;
1362     const struct wined3d_texture *texture;
1363     const char *tex_type;
1364     BOOL np2_fixup = FALSE;
1365     struct wined3d_device *device = shader->device;
1366     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1367     const char *mod;
1368     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1369
1370     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1371     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1372
1373     switch(sampler_type) {
1374         case WINED3DSTT_1D:
1375             tex_type = "1D";
1376             break;
1377
1378         case WINED3DSTT_2D:
1379             texture = device->stateBlock->state.textures[sampler_idx];
1380             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1381             {
1382                 tex_type = "RECT";
1383             } else {
1384                 tex_type = "2D";
1385             }
1386             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1387             {
1388                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1389                 {
1390                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1391                     else np2_fixup = TRUE;
1392                 }
1393             }
1394             break;
1395
1396         case WINED3DSTT_VOLUME:
1397             tex_type = "3D";
1398             break;
1399
1400         case WINED3DSTT_CUBE:
1401             tex_type = "CUBE";
1402             break;
1403
1404         default:
1405             ERR("Unexpected texture type %d\n", sampler_type);
1406             tex_type = "";
1407     }
1408
1409     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1410      * so don't use shader_arb_get_modifier
1411      */
1412     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1413     else mod = "";
1414
1415     /* Fragment samplers always have indentity mapping */
1416     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1417     {
1418         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1419     }
1420
1421     if (flags & TEX_DERIV)
1422     {
1423         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1424         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1425         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1426                        dsx, dsy,sampler_idx, tex_type);
1427     }
1428     else if(flags & TEX_LOD)
1429     {
1430         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1431         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1432         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1433                        sampler_idx, tex_type);
1434     }
1435     else if (flags & TEX_BIAS)
1436     {
1437         /* Shouldn't be possible, but let's check for it */
1438         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1439         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1440         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1441     }
1442     else if (flags & TEX_PROJ)
1443     {
1444         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1445     }
1446     else
1447     {
1448         if (np2_fixup)
1449         {
1450             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1451             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1452                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1453
1454             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1455         }
1456         else
1457             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1458     }
1459
1460     if (pshader)
1461     {
1462         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1463                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1464                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1465                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1466     }
1467 }
1468
1469 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1470         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1471 {
1472     /* Generate a line that does the input modifier computation and return the input register to use */
1473     BOOL is_color = FALSE;
1474     char regstr[256];
1475     char swzstr[20];
1476     int insert_line;
1477     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1478     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1479     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1480     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1481
1482     /* Assume a new line will be added */
1483     insert_line = 1;
1484
1485     /* Get register name */
1486     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1487     shader_arb_get_swizzle(src, is_color, swzstr);
1488
1489     switch (src->modifiers)
1490     {
1491     case WINED3DSPSM_NONE:
1492         sprintf(outregstr, "%s%s", regstr, swzstr);
1493         insert_line = 0;
1494         break;
1495     case WINED3DSPSM_NEG:
1496         sprintf(outregstr, "-%s%s", regstr, swzstr);
1497         insert_line = 0;
1498         break;
1499     case WINED3DSPSM_BIAS:
1500         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1501         break;
1502     case WINED3DSPSM_BIASNEG:
1503         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1504         break;
1505     case WINED3DSPSM_SIGN:
1506         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1507         break;
1508     case WINED3DSPSM_SIGNNEG:
1509         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1510         break;
1511     case WINED3DSPSM_COMP:
1512         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1513         break;
1514     case WINED3DSPSM_X2:
1515         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1516         break;
1517     case WINED3DSPSM_X2NEG:
1518         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1519         break;
1520     case WINED3DSPSM_DZ:
1521         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1522         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1523         break;
1524     case WINED3DSPSM_DW:
1525         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1526         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1527         break;
1528     case WINED3DSPSM_ABS:
1529         if(ctx->target_version >= NV2) {
1530             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1531             insert_line = 0;
1532         } else {
1533             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1534         }
1535         break;
1536     case WINED3DSPSM_ABSNEG:
1537         if(ctx->target_version >= NV2) {
1538             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1539         } else {
1540             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1541             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1542         }
1543         insert_line = 0;
1544         break;
1545     default:
1546         sprintf(outregstr, "%s%s", regstr, swzstr);
1547         insert_line = 0;
1548     }
1549
1550     /* Return modified or original register, with swizzle */
1551     if (insert_line)
1552         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1553 }
1554
1555 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1556 {
1557     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1558     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1559     char dst_name[50];
1560     char src_name[2][50];
1561     DWORD sampler_code = dst->reg.idx;
1562
1563     shader_arb_get_dst_param(ins, dst, dst_name);
1564
1565     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1566      *
1567      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1568      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1569      * temps is done.
1570      */
1571     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1572     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1573     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1574     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1575     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1576
1577     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1578     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1579 }
1580
1581 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1582 {
1583     *extra_char = ' ';
1584     switch(mod)
1585     {
1586         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1587         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1588         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1589         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1590         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1591         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1592         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1593         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1594         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1595         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1596         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1597         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1598         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1599     }
1600     FIXME("Unknown modifier %u\n", mod);
1601     return mod;
1602 }
1603
1604 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1605 {
1606     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1607     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1608     char dst_name[50];
1609     char src_name[3][50];
1610     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1611             ins->ctx->reg_maps->shader_version.minor);
1612
1613     shader_arb_get_dst_param(ins, dst, dst_name);
1614     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1615
1616     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1617     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1618     {
1619         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1620     }
1621     else
1622     {
1623         struct wined3d_shader_src_param src0_copy = ins->src[0];
1624         char extra_neg;
1625
1626         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1627         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1628
1629         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1630         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1631         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1632         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1633                 dst_name, src_name[1], src_name[2]);
1634     }
1635 }
1636
1637 static void pshader_hw_cmp(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
1644     shader_arb_get_dst_param(ins, dst, dst_name);
1645
1646     /* Generate input register names (with modifiers) */
1647     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1648     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1649     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1650
1651     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1652             dst_name, src_name[0], src_name[2], src_name[1]);
1653 }
1654
1655 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1656  * dst = dot2(src0, src1) + src2 */
1657 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1658 {
1659     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1660     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1661     char dst_name[50];
1662     char src_name[3][50];
1663     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1664
1665     shader_arb_get_dst_param(ins, dst, dst_name);
1666     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1667     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1668
1669     if(ctx->target_version >= NV3)
1670     {
1671         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1672         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1673         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1674                        dst_name, src_name[0], src_name[1], src_name[2]);
1675     }
1676     else if(ctx->target_version >= NV2)
1677     {
1678         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1679          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1680          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1681          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1682          *
1683          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1684          *
1685          * .xyxy and other swizzles that we could get with this are not valid in
1686          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1687          */
1688         struct wined3d_shader_src_param tmp_param = ins->src[1];
1689         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1690         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1691
1692         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1693
1694         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1695                        dst_name, src_name[2], src_name[0], src_name[1]);
1696     }
1697     else
1698     {
1699         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1700         /* 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]
1701         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1702         */
1703         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1704         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1705         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1706         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1707     }
1708 }
1709
1710 /* Map the opcode 1-to-1 to the GL code */
1711 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1712 {
1713     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1714     const char *instruction;
1715     char arguments[256], dst_str[50];
1716     unsigned int i;
1717     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1718
1719     switch (ins->handler_idx)
1720     {
1721         case WINED3DSIH_ABS: instruction = "ABS"; break;
1722         case WINED3DSIH_ADD: instruction = "ADD"; break;
1723         case WINED3DSIH_CRS: instruction = "XPD"; break;
1724         case WINED3DSIH_DP3: instruction = "DP3"; break;
1725         case WINED3DSIH_DP4: instruction = "DP4"; break;
1726         case WINED3DSIH_DST: instruction = "DST"; break;
1727         case WINED3DSIH_FRC: instruction = "FRC"; break;
1728         case WINED3DSIH_LIT: instruction = "LIT"; break;
1729         case WINED3DSIH_LRP: instruction = "LRP"; break;
1730         case WINED3DSIH_MAD: instruction = "MAD"; break;
1731         case WINED3DSIH_MAX: instruction = "MAX"; break;
1732         case WINED3DSIH_MIN: instruction = "MIN"; break;
1733         case WINED3DSIH_MOV: instruction = "MOV"; break;
1734         case WINED3DSIH_MUL: instruction = "MUL"; break;
1735         case WINED3DSIH_SGE: instruction = "SGE"; break;
1736         case WINED3DSIH_SLT: instruction = "SLT"; break;
1737         case WINED3DSIH_SUB: instruction = "SUB"; break;
1738         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1739         case WINED3DSIH_DSX: instruction = "DDX"; break;
1740         default: instruction = "";
1741             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1742             break;
1743     }
1744
1745     /* Note that shader_arb_add_dst_param() adds spaces. */
1746     arguments[0] = '\0';
1747     shader_arb_get_dst_param(ins, dst, dst_str);
1748     for (i = 0; i < ins->src_count; ++i)
1749     {
1750         char operand[100];
1751         strcat(arguments, ", ");
1752         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1753         strcat(arguments, operand);
1754     }
1755     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1756 }
1757
1758 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1759 {
1760     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1761     shader_addline(buffer, "NOP;\n");
1762 }
1763
1764 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1765 {
1766     const struct wined3d_shader *shader = ins->ctx->shader;
1767     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1768     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1769     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1770     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1771     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1772     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1773
1774     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1775     char src0_param[256];
1776
1777     if (ins->handler_idx == WINED3DSIH_MOVA)
1778     {
1779         const struct arb_vshader_private *shader_data = shader->backend_data;
1780         char write_mask[6];
1781         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1782
1783         if(ctx->target_version >= NV2) {
1784             shader_hw_map2gl(ins);
1785             return;
1786         }
1787         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1788         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1789
1790         /* This implements the mova formula used in GLSL. The first two instructions
1791          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1792          * in this case:
1793          * mova A0.x, 0.0
1794          *
1795          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1796          *
1797          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1798          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1799          */
1800         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1801         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1802
1803         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1804         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1805         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1806         if (shader_data->rel_offset)
1807         {
1808             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1809         }
1810         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1811
1812         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1813     }
1814     else if (reg_maps->shader_version.major == 1
1815           && !shader_is_pshader_version(reg_maps->shader_version.type)
1816           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1817     {
1818         const struct arb_vshader_private *shader_data = shader->backend_data;
1819         src0_param[0] = '\0';
1820
1821         if (shader_data->rel_offset)
1822         {
1823             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1824             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1825             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1826             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1827         }
1828         else
1829         {
1830             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1831              * with more than one component. Thus replicate the first source argument over all
1832              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1833             struct wined3d_shader_src_param tmp_src = ins->src[0];
1834             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1835             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1836             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1837         }
1838     }
1839     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1840     {
1841         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1842         {
1843             shader_addline(buffer, "#mov handled in srgb write code\n");
1844             return;
1845         }
1846         shader_hw_map2gl(ins);
1847     }
1848     else
1849     {
1850         shader_hw_map2gl(ins);
1851     }
1852 }
1853
1854 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1855 {
1856     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1857     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1858     char reg_dest[40];
1859
1860     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1861      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1862      */
1863     shader_arb_get_dst_param(ins, dst, reg_dest);
1864
1865     if (ins->ctx->reg_maps->shader_version.major >= 2)
1866     {
1867         const char *kilsrc = "TA";
1868         BOOL is_color;
1869
1870         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1871         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1872         {
1873             kilsrc = reg_dest;
1874         }
1875         else
1876         {
1877             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1878              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1879              * masked out components to 0(won't kill)
1880              */
1881             char x = '0', y = '0', z = '0', w = '0';
1882             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1883             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1884             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1885             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1886             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1887         }
1888         shader_addline(buffer, "KIL %s;\n", kilsrc);
1889     } else {
1890         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1891          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1892          *
1893          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1894          * or pass in any temporary register(in shader phase 2)
1895          */
1896         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1897             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1898         } else {
1899             shader_arb_get_dst_param(ins, dst, reg_dest);
1900         }
1901         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1902         shader_addline(buffer, "KIL TA;\n");
1903     }
1904 }
1905
1906 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1907 {
1908     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1909     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1910     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1911             ins->ctx->reg_maps->shader_version.minor);
1912     struct wined3d_shader_src_param src;
1913
1914     char reg_dest[40];
1915     char reg_coord[40];
1916     DWORD reg_sampler_code;
1917     WORD myflags = 0;
1918     BOOL swizzle_coord = FALSE;
1919
1920     /* All versions have a destination register */
1921     shader_arb_get_dst_param(ins, dst, reg_dest);
1922
1923     /* 1.0-1.4: Use destination register number as texture code.
1924        2.0+: Use provided sampler number as texure code. */
1925     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1926         reg_sampler_code = dst->reg.idx;
1927     else
1928         reg_sampler_code = ins->src[1].reg.idx;
1929
1930     /* 1.0-1.3: Use the texcoord varying.
1931        1.4+: Use provided coordinate source register. */
1932     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1933         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1934     else {
1935         /* TEX is the only instruction that can handle DW and DZ natively */
1936         src = ins->src[0];
1937         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1938         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1939         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1940     }
1941
1942     /* projection flag:
1943      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1944      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1945      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1946      */
1947     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1948     {
1949         DWORD flags = 0;
1950         if (reg_sampler_code < MAX_TEXTURES)
1951             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1952         if (flags & WINED3D_PSARGS_PROJECTED)
1953         {
1954             myflags |= TEX_PROJ;
1955             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1956                 swizzle_coord = TRUE;
1957         }
1958     }
1959     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1960     {
1961         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1962         if (src_mod == WINED3DSPSM_DZ)
1963         {
1964             swizzle_coord = TRUE;
1965             myflags |= TEX_PROJ;
1966         } else if(src_mod == WINED3DSPSM_DW) {
1967             myflags |= TEX_PROJ;
1968         }
1969     } else {
1970         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1971         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1972     }
1973
1974     if (swizzle_coord)
1975     {
1976         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
1977          * reg_coord is a read-only varying register, so we need a temp reg */
1978         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1979         strcpy(reg_coord, "TA");
1980     }
1981
1982     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1983 }
1984
1985 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1986 {
1987     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1988     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1989     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1990             ins->ctx->reg_maps->shader_version.minor);
1991     char dst_str[50];
1992
1993     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1994     {
1995         DWORD reg = dst->reg.idx;
1996
1997         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1998         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1999     } else {
2000         char reg_src[40];
2001
2002         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2003         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2004         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2005    }
2006 }
2007
2008 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2009 {
2010      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2011      DWORD flags = 0;
2012
2013      DWORD reg1 = ins->dst[0].reg.idx;
2014      char dst_str[50];
2015      char src_str[50];
2016
2017      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2018      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2019      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2020      /* Move .x first in case src_str is "TA" */
2021      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2022      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2023      if (reg1 < MAX_TEXTURES)
2024      {
2025          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2026          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2027      }
2028      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2029 }
2030
2031 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2032 {
2033      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2034
2035      DWORD reg1 = ins->dst[0].reg.idx;
2036      char dst_str[50];
2037      char src_str[50];
2038
2039      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2040      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2041      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2042      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2043      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2044      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2045 }
2046
2047 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2048 {
2049     DWORD reg1 = ins->dst[0].reg.idx;
2050     char dst_str[50];
2051     char src_str[50];
2052
2053     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2054     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2055     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2056     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2057 }
2058
2059 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2060 {
2061     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2062     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2063     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2064     char reg_coord[40], dst_reg[50], src_reg[50];
2065     DWORD reg_dest_code;
2066
2067     /* All versions have a destination register. The Tx where the texture coordinates come
2068      * from is the varying incarnation of the texture register
2069      */
2070     reg_dest_code = dst->reg.idx;
2071     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2072     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2073     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2074
2075     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2076      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2077      *
2078      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2079      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2080      *
2081      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2082      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2083      * extension.
2084      */
2085     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2086     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2087     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2088     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2089
2090     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2091      * so we can't let the GL handle this.
2092      */
2093     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2094             & WINED3D_PSARGS_PROJECTED)
2095     {
2096         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2097         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2098         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2099     } else {
2100         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2101     }
2102
2103     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2104
2105     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2106     {
2107         /* No src swizzles are allowed, so this is ok */
2108         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2109                        src_reg, reg_dest_code, reg_dest_code);
2110         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2111     }
2112 }
2113
2114 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2115 {
2116     DWORD reg = ins->dst[0].reg.idx;
2117     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2118     char src0_name[50], dst_name[50];
2119     BOOL is_color;
2120     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2121
2122     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2123     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2124      * T<reg+1> register. Use this register to store the calculated vector
2125      */
2126     tmp_reg.idx = reg + 1;
2127     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2128     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2129 }
2130
2131 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2132 {
2133     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2134     DWORD flags;
2135     DWORD reg = ins->dst[0].reg.idx;
2136     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2137     char dst_str[50];
2138     char src0_name[50];
2139     char dst_reg[50];
2140     BOOL is_color;
2141
2142     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2143     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2144
2145     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2146     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2147     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2148     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2149     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2150 }
2151
2152 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2153 {
2154     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2155     DWORD reg = ins->dst[0].reg.idx;
2156     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2157     char src0_name[50], dst_name[50];
2158     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2159     BOOL is_color;
2160
2161     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2162      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2163      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2164      */
2165     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2166     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2167
2168     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2169     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2170                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2171     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2172 }
2173
2174 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2175 {
2176     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2177     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2178     DWORD flags;
2179     DWORD reg = ins->dst[0].reg.idx;
2180     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2181     char dst_str[50];
2182     char src0_name[50], dst_name[50];
2183     BOOL is_color;
2184
2185     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2186     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2187     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2188
2189     /* Sample the texture using the calculated coordinates */
2190     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2191     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2192     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2193     tex_mx->current_row = 0;
2194 }
2195
2196 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2197 {
2198     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2199     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2200     DWORD flags;
2201     DWORD reg = ins->dst[0].reg.idx;
2202     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2203     char dst_str[50];
2204     char src0_name[50];
2205     char dst_reg[50];
2206     BOOL is_color;
2207
2208     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2209      * components for temporary data storage
2210      */
2211     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2212     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2213     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2214
2215     /* Construct the eye-ray vector from w coordinates */
2216     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2217     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2218     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2219
2220     /* Calculate reflection vector
2221      */
2222     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2223     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2224     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2225     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2226     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2227     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2228     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2229
2230     /* Sample the texture using the calculated coordinates */
2231     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2232     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2233     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2234     tex_mx->current_row = 0;
2235 }
2236
2237 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2238 {
2239     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2240     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2241     DWORD flags;
2242     DWORD reg = ins->dst[0].reg.idx;
2243     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2244     char dst_str[50];
2245     char src0_name[50];
2246     char src1_name[50];
2247     char dst_reg[50];
2248     BOOL is_color;
2249
2250     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2251     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2252     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2253     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2254     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2255
2256     /* Calculate reflection vector.
2257      *
2258      *                   dot(N, E)
2259      * dst_reg.xyz = 2 * --------- * N - E
2260      *                   dot(N, N)
2261      *
2262      * Which normalizes the normal vector
2263      */
2264     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2265     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2266     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2267     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2268     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2269     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2270
2271     /* Sample the texture using the calculated coordinates */
2272     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2273     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2274     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2275     tex_mx->current_row = 0;
2276 }
2277
2278 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2279 {
2280     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2281     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2282     char dst_name[50];
2283     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2284     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2285
2286     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2287      * which is essentially an input, is the destination register because it is the first
2288      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2289      * here(writemasks/swizzles are not valid on texdepth)
2290      */
2291     shader_arb_get_dst_param(ins, dst, dst_name);
2292
2293     /* According to the msdn, the source register(must be r5) is unusable after
2294      * the texdepth instruction, so we're free to modify it
2295      */
2296     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2297
2298     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2299      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2300      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2301      */
2302     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2303     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2304     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2305     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2306 }
2307
2308 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2309  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2310  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2311 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2312 {
2313     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2314     DWORD sampler_idx = ins->dst[0].reg.idx;
2315     char src0[50];
2316     char dst_str[50];
2317
2318     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2319     shader_addline(buffer, "MOV TB, 0.0;\n");
2320     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2321
2322     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2323     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2324 }
2325
2326 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2327  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2328 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2329 {
2330     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2331     char src0[50];
2332     char dst_str[50];
2333     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2334
2335     /* Handle output register */
2336     shader_arb_get_dst_param(ins, dst, dst_str);
2337     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2338     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2339 }
2340
2341 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2342  * Perform the 3rd row of a 3x3 matrix multiply */
2343 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2344 {
2345     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2346     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2347     char dst_str[50], dst_name[50];
2348     char src0[50];
2349     BOOL is_color;
2350
2351     shader_arb_get_dst_param(ins, dst, dst_str);
2352     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2353     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2354     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2355     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2356 }
2357
2358 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2359  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2360  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2361  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2362  */
2363 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2364 {
2365     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2366     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2367     char src0[50], dst_name[50];
2368     BOOL is_color;
2369     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2370     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2371
2372     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2373     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2374     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2375
2376     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2377      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2378      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2379      */
2380     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2381     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2382     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2383     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2384 }
2385
2386 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2387     Vertex/Pixel shaders to ARB_vertex_program codes */
2388 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2389 {
2390     int i;
2391     int nComponents = 0;
2392     struct wined3d_shader_dst_param tmp_dst = {{0}};
2393     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2394     struct wined3d_shader_instruction tmp_ins;
2395
2396     memset(&tmp_ins, 0, sizeof(tmp_ins));
2397
2398     /* Set constants for the temporary argument */
2399     tmp_ins.ctx = ins->ctx;
2400     tmp_ins.dst_count = 1;
2401     tmp_ins.dst = &tmp_dst;
2402     tmp_ins.src_count = 2;
2403     tmp_ins.src = tmp_src;
2404
2405     switch(ins->handler_idx)
2406     {
2407         case WINED3DSIH_M4x4:
2408             nComponents = 4;
2409             tmp_ins.handler_idx = WINED3DSIH_DP4;
2410             break;
2411         case WINED3DSIH_M4x3:
2412             nComponents = 3;
2413             tmp_ins.handler_idx = WINED3DSIH_DP4;
2414             break;
2415         case WINED3DSIH_M3x4:
2416             nComponents = 4;
2417             tmp_ins.handler_idx = WINED3DSIH_DP3;
2418             break;
2419         case WINED3DSIH_M3x3:
2420             nComponents = 3;
2421             tmp_ins.handler_idx = WINED3DSIH_DP3;
2422             break;
2423         case WINED3DSIH_M3x2:
2424             nComponents = 2;
2425             tmp_ins.handler_idx = WINED3DSIH_DP3;
2426             break;
2427         default:
2428             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2429             break;
2430     }
2431
2432     tmp_dst = ins->dst[0];
2433     tmp_src[0] = ins->src[0];
2434     tmp_src[1] = ins->src[1];
2435     for (i = 0; i < nComponents; i++) {
2436         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2437         shader_hw_map2gl(&tmp_ins);
2438         ++tmp_src[1].reg.idx;
2439     }
2440 }
2441
2442 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2443 {
2444     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2445
2446     char dst[50];
2447     char src[50];
2448
2449     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2450     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2451     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2452     {
2453         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2454          * .w is used
2455          */
2456         strcat(src, ".w");
2457     }
2458
2459     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2460 }
2461
2462 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2463 {
2464     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2465     const char *instruction;
2466
2467     char dst[50];
2468     char src[50];
2469
2470     switch(ins->handler_idx)
2471     {
2472         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2473         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2474         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2475         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2476         default: instruction = "";
2477             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2478             break;
2479     }
2480
2481     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2482     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2483     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2484     {
2485         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2486          * .w is used
2487          */
2488         strcat(src, ".w");
2489     }
2490
2491     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2492 }
2493
2494 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2495 {
2496     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2497     char dst_name[50];
2498     char src_name[50];
2499     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2500     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2501     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2502
2503     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2504     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2505
2506     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2507      * otherwise NRM or RSQ would return NaN */
2508     if(pshader && priv->target_version >= NV3)
2509     {
2510         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2511          *
2512          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2513          */
2514         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2515         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2516         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2517     }
2518     else if(priv->target_version >= NV2)
2519     {
2520         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2521         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2522         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2523                        src_name);
2524     }
2525     else
2526     {
2527         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2528
2529         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2530         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2531          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2532          */
2533         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2534         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2535
2536         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2537         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2538         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2539                     src_name);
2540     }
2541 }
2542
2543 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2544 {
2545     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2546     char dst_name[50];
2547     char src_name[3][50];
2548
2549     /* ARB_fragment_program has a convenient LRP instruction */
2550     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2551         shader_hw_map2gl(ins);
2552         return;
2553     }
2554
2555     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2556     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2557     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2558     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2559
2560     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2561     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2562                    dst_name, src_name[0], src_name[2]);
2563 }
2564
2565 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2566 {
2567     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2568      * must contain fixed constants. So we need a separate function to filter those constants and
2569      * can't use map2gl
2570      */
2571     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2572     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2573     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2574     char dst_name[50];
2575     char src_name0[50], src_name1[50], src_name2[50];
2576     BOOL is_color;
2577
2578     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2579     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2580         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2581         /* No modifiers are supported on SCS */
2582         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2583
2584         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2585         {
2586             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2587             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2588         }
2589     } else if(priv->target_version >= NV2) {
2590         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2591
2592         /* Sincos writemask must be .x, .y or .xy */
2593         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2594             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2595         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2596             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2597     } else {
2598         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2599          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2600          *
2601          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2602          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2603          *
2604          * The constants we get are:
2605          *
2606          *  +1   +1,     -1     -1     +1      +1      -1       -1
2607          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2608          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2609          *
2610          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2611          *
2612          * (x/2)^2 = x^2 / 4
2613          * (x/2)^3 = x^3 / 8
2614          * (x/2)^4 = x^4 / 16
2615          * (x/2)^5 = x^5 / 32
2616          * etc
2617          *
2618          * To get the final result:
2619          * sin(x) = 2 * sin(x/2) * cos(x/2)
2620          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2621          * (from sin(x+y) and cos(x+y) rules)
2622          *
2623          * As per MSDN, dst.z is undefined after the operation, and so is
2624          * dst.x and dst.y if they're masked out by the writemask. Ie
2625          * sincos dst.y, src1, c0, c1
2626          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2627          * vsa.exe also stops with an error if the dest register is the same register as the source
2628          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2629          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2630          */
2631         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2632         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2633         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2634
2635         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2636         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2637         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2638         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2639         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2640         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2641
2642         /* sin(x/2)
2643          *
2644          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2645          * properly merge that with MULs in the code above?
2646          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2647          * we can merge the sine and cosine MAD rows to calculate them together.
2648          */
2649         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2650         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2651         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2652         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2653
2654         /* cos(x/2) */
2655         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2656         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2657         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2658
2659         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2660             /* cos x */
2661             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2662             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2663         }
2664         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2665             /* sin x */
2666             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2667             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2668         }
2669     }
2670 }
2671
2672 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2673 {
2674     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2675     char dst_name[50];
2676     char src_name[50];
2677     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2678
2679     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2680     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2681
2682     /* SGN is only valid in vertex shaders */
2683     if(ctx->target_version >= NV2) {
2684         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2685         return;
2686     }
2687
2688     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2689      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2690      */
2691     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2692         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2693     } else {
2694         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2695          * Then use TA, and calculate the final result
2696          *
2697          * Not reading from TA? Store the first result in TA to avoid overwriting the
2698          * destination if src reg = dst reg
2699          */
2700         if(strstr(src_name, "TA"))
2701         {
2702             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2703             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2704             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2705         }
2706         else
2707         {
2708             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2709             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2710             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2711         }
2712     }
2713 }
2714
2715 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2716 {
2717     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2718     char src[50];
2719     char dst[50];
2720     char dst_name[50];
2721     BOOL is_color;
2722
2723     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2724     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2725     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2726
2727     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2728     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2729 }
2730
2731 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2732 {
2733     *need_abs = FALSE;
2734
2735     switch(mod)
2736     {
2737         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2738         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2739         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2740         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2741         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2742         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2743         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2744         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2745         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2746         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2747         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2748         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2749         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2750     }
2751     FIXME("Unknown modifier %u\n", mod);
2752     return mod;
2753 }
2754
2755 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2756 {
2757     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2758     char src0[50], dst[50];
2759     struct wined3d_shader_src_param src0_copy = ins->src[0];
2760     BOOL need_abs = FALSE;
2761     const char *instr;
2762
2763     switch(ins->handler_idx)
2764     {
2765         case WINED3DSIH_LOG:  instr = "LG2"; break;
2766         case WINED3DSIH_LOGP: instr = "LOG"; break;
2767         default:
2768             ERR("Unexpected instruction %d\n", ins->handler_idx);
2769             return;
2770     }
2771
2772     /* LOG and LOGP operate on the absolute value of the input */
2773     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2774
2775     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2776     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2777
2778     if(need_abs)
2779     {
2780         shader_addline(buffer, "ABS TA, %s;\n", src0);
2781         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2782     }
2783     else
2784     {
2785         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2786     }
2787 }
2788
2789 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2790 {
2791     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2792     char src0[50], src1[50], dst[50];
2793     struct wined3d_shader_src_param src0_copy = ins->src[0];
2794     BOOL need_abs = FALSE;
2795     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2796     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2797
2798     /* POW operates on the absolute value of the input */
2799     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2800
2801     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2802     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2803     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2804
2805     if (need_abs)
2806         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2807     else
2808         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2809
2810     if (priv->target_version >= NV2)
2811     {
2812         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2813         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2814         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2815     }
2816     else
2817     {
2818         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2819         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2820
2821         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2822         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2823         /* Possibly add flt_eps to avoid getting float special values */
2824         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2825         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2826         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2827         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2828     }
2829 }
2830
2831 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2832 {
2833     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2834     char src_name[50];
2835     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2836
2837     /* src0 is aL */
2838     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2839
2840     if(vshader)
2841     {
2842         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2843         struct list *e = list_head(&priv->control_frames);
2844         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2845
2846         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2847         /* The constant loader makes sure to load -1 into iX.w */
2848         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2849         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2850         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2851     }
2852     else
2853     {
2854         shader_addline(buffer, "LOOP %s;\n", src_name);
2855     }
2856 }
2857
2858 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2859 {
2860     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2861     char src_name[50];
2862     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2863
2864     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2865
2866     /* The constant loader makes sure to load -1 into iX.w */
2867     if(vshader)
2868     {
2869         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2870         struct list *e = list_head(&priv->control_frames);
2871         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2872
2873         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2874
2875         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2876         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2877         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2878     }
2879     else
2880     {
2881         shader_addline(buffer, "REP %s;\n", src_name);
2882     }
2883 }
2884
2885 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2886 {
2887     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2888     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2889
2890     if(vshader)
2891     {
2892         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2893         struct list *e = list_head(&priv->control_frames);
2894         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2895
2896         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2897         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2898         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2899
2900         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2901     }
2902     else
2903     {
2904         shader_addline(buffer, "ENDLOOP;\n");
2905     }
2906 }
2907
2908 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2909 {
2910     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2911     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2912
2913     if(vshader)
2914     {
2915         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2916         struct list *e = list_head(&priv->control_frames);
2917         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2918
2919         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2920         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2921         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2922
2923         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2924     }
2925     else
2926     {
2927         shader_addline(buffer, "ENDREP;\n");
2928     }
2929 }
2930
2931 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2932 {
2933     struct control_frame *control_frame;
2934
2935     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2936     {
2937         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2938     }
2939     ERR("Could not find loop for break\n");
2940     return NULL;
2941 }
2942
2943 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2944 {
2945     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2946     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2947     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2948
2949     if(vshader)
2950     {
2951         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2952     }
2953     else
2954     {
2955         shader_addline(buffer, "BRK;\n");
2956     }
2957 }
2958
2959 static const char *get_compare(enum wined3d_shader_rel_op op)
2960 {
2961     switch (op)
2962     {
2963         case WINED3D_SHADER_REL_OP_GT: return "GT";
2964         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2965         case WINED3D_SHADER_REL_OP_GE: return "GE";
2966         case WINED3D_SHADER_REL_OP_LT: return "LT";
2967         case WINED3D_SHADER_REL_OP_NE: return "NE";
2968         case WINED3D_SHADER_REL_OP_LE: return "LE";
2969         default:
2970             FIXME("Unrecognized operator %#x.\n", op);
2971             return "(\?\?)";
2972     }
2973 }
2974
2975 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2976 {
2977     switch (op)
2978     {
2979         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2980         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2981         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2982         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2983         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2984         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2985         default:
2986             FIXME("Unrecognized operator %#x.\n", op);
2987             return -1;
2988     }
2989 }
2990
2991 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2992 {
2993     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2994     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2995     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2996     char src_name0[50];
2997     char src_name1[50];
2998     const char *comp = get_compare(ins->flags);
2999
3000     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3001     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3002
3003     if(vshader)
3004     {
3005         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3006          * away the subtraction result
3007          */
3008         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3009         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3010     }
3011     else
3012     {
3013         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3014         shader_addline(buffer, "BRK (%s.x);\n", comp);
3015     }
3016 }
3017
3018 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3019 {
3020     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3021     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3022     struct list *e = list_head(&priv->control_frames);
3023     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3024     const char *comp;
3025     char src_name0[50];
3026     char src_name1[50];
3027     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3028
3029     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3030     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3031
3032     if(vshader)
3033     {
3034         /* Invert the flag. We jump to the else label if the condition is NOT true */
3035         comp = get_compare(invert_compare(ins->flags));
3036         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3037         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3038     }
3039     else
3040     {
3041         comp = get_compare(ins->flags);
3042         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3043         shader_addline(buffer, "IF %s.x;\n", comp);
3044     }
3045 }
3046
3047 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3048 {
3049     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3050     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3051     struct list *e = list_head(&priv->control_frames);
3052     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3053     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3054
3055     if(vshader)
3056     {
3057         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3058         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3059         control_frame->had_else = TRUE;
3060     }
3061     else
3062     {
3063         shader_addline(buffer, "ELSE;\n");
3064     }
3065 }
3066
3067 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3068 {
3069     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3070     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3071     struct list *e = list_head(&priv->control_frames);
3072     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3073     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3074
3075     if(vshader)
3076     {
3077         if(control_frame->had_else)
3078         {
3079             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3080         }
3081         else
3082         {
3083             shader_addline(buffer, "#No else branch. else is endif\n");
3084             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3085         }
3086     }
3087     else
3088     {
3089         shader_addline(buffer, "ENDIF;\n");
3090     }
3091 }
3092
3093 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3094 {
3095     DWORD sampler_idx = ins->src[1].reg.idx;
3096     char reg_dest[40];
3097     char reg_src[3][40];
3098     WORD flags = TEX_DERIV;
3099
3100     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3101     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3102     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3103     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3104
3105     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3106     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3107
3108     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3109 }
3110
3111 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3112 {
3113     DWORD sampler_idx = ins->src[1].reg.idx;
3114     char reg_dest[40];
3115     char reg_coord[40];
3116     WORD flags = TEX_LOD;
3117
3118     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3119     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3120
3121     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3122     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3123
3124     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3125 }
3126
3127 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3128 {
3129     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3130     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3131
3132     priv->in_main_func = FALSE;
3133     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3134      * subroutine, don't generate a label that will make GL complain
3135      */
3136     if(priv->target_version == ARB) return;
3137
3138     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3139 }
3140
3141 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3142         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3143         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3144         struct wined3d_shader_buffer *buffer)
3145 {
3146     unsigned int i;
3147
3148     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3149      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3150      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3151      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3152      */
3153     if (args->super.fog_src == VS_FOG_Z)
3154     {
3155         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3156     }
3157     else
3158     {
3159         if (!reg_maps->fog)
3160         {
3161             /* posFixup.x is always 1.0, so we can safely use it */
3162             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3163         }
3164         else
3165         {
3166             /* Clamp fogcoord */
3167             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3168             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3169
3170             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3171             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3172         }
3173     }
3174
3175     /* Clipplanes are always stored without y inversion */
3176     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3177     {
3178         if (args->super.clip_enabled)
3179         {
3180             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3181             {
3182                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3183             }
3184         }
3185     }
3186     else if (args->clip.boolclip.clip_texcoord)
3187     {
3188         unsigned int cur_clip = 0;
3189         char component[4] = {'x', 'y', 'z', 'w'};
3190         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3191
3192         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3193         {
3194             if (args->clip.boolclip.clipplane_mask & (1 << i))
3195             {
3196                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3197                                component[cur_clip++], i);
3198             }
3199         }
3200         switch (cur_clip)
3201         {
3202             case 0:
3203                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3204                 break;
3205             case 1:
3206                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3207                 break;
3208             case 2:
3209                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3210                 break;
3211             case 3:
3212                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3213                 break;
3214         }
3215         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3216                        args->clip.boolclip.clip_texcoord - 1);
3217     }
3218
3219     /* Write the final position.
3220      *
3221      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3222      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3223      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3224      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3225      */
3226     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3227     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3228     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3229
3230     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3231      * and the glsl equivalent
3232      */
3233     if (need_helper_const(shader_data, reg_maps, gl_info))
3234     {
3235         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3236         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3237     }
3238     else
3239     {
3240         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3241         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3242     }
3243
3244     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3245
3246     priv_ctx->footer_written = TRUE;
3247 }
3248
3249 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3250 {
3251     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3252     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3253     const struct wined3d_shader *shader = ins->ctx->shader;
3254     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3255
3256     if(priv->target_version == ARB) return;
3257
3258     if(vshader)
3259     {
3260         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3261                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3262     }
3263
3264     shader_addline(buffer, "RET;\n");
3265 }
3266
3267 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3268 {
3269     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3270     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3271 }
3272
3273 /* GL locking is done by the caller */
3274 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3275 {
3276     GLuint program_id = 0;
3277     GLint pos;
3278
3279     const char *blt_vprogram =
3280         "!!ARBvp1.0\n"
3281         "PARAM c[1] = { { 1, 0.5 } };\n"
3282         "MOV result.position, vertex.position;\n"
3283         "MOV result.color, c[0].x;\n"
3284         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3285         "END\n";
3286
3287     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3288     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3289     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3290             strlen(blt_vprogram), blt_vprogram));
3291     checkGLcall("glProgramStringARB()");
3292
3293     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3294     if (pos != -1)
3295     {
3296         FIXME("Vertex program error at position %d: %s\n\n", pos,
3297             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3298         shader_arb_dump_program_source(blt_vprogram);
3299     }
3300     else
3301     {
3302         GLint native;
3303
3304         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3305         checkGLcall("glGetProgramivARB()");
3306         if (!native) WARN("Program exceeds native resource limits.\n");
3307     }
3308
3309     return program_id;
3310 }
3311
3312 /* GL locking is done by the caller */
3313 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3314         enum tex_types tex_type, BOOL masked)
3315 {
3316     GLuint program_id = 0;
3317     const char *fprogram;
3318     GLint pos;
3319
3320     static const char * const blt_fprograms_full[tex_type_count] =
3321     {
3322         /* tex_1d */
3323         NULL,
3324         /* tex_2d */
3325         "!!ARBfp1.0\n"
3326         "TEMP R0;\n"
3327         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3328         "MOV result.depth.z, R0.x;\n"
3329         "END\n",
3330         /* tex_3d */
3331         NULL,
3332         /* tex_cube */
3333         "!!ARBfp1.0\n"
3334         "TEMP R0;\n"
3335         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3336         "MOV result.depth.z, R0.x;\n"
3337         "END\n",
3338         /* tex_rect */
3339         "!!ARBfp1.0\n"
3340         "TEMP R0;\n"
3341         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3342         "MOV result.depth.z, R0.x;\n"
3343         "END\n",
3344     };
3345
3346     static const char * const blt_fprograms_masked[tex_type_count] =
3347     {
3348         /* tex_1d */
3349         NULL,
3350         /* tex_2d */
3351         "!!ARBfp1.0\n"
3352         "PARAM mask = program.local[0];\n"
3353         "TEMP R0;\n"
3354         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3355         "MUL R0.x, R0.x, R0.y;\n"
3356         "KIL -R0.x;\n"
3357         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3358         "MOV result.depth.z, R0.x;\n"
3359         "END\n",
3360         /* tex_3d */
3361         NULL,
3362         /* tex_cube */
3363         "!!ARBfp1.0\n"
3364         "PARAM mask = program.local[0];\n"
3365         "TEMP R0;\n"
3366         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3367         "MUL R0.x, R0.x, R0.y;\n"
3368         "KIL -R0.x;\n"
3369         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3370         "MOV result.depth.z, R0.x;\n"
3371         "END\n",
3372         /* tex_rect */
3373         "!!ARBfp1.0\n"
3374         "PARAM mask = program.local[0];\n"
3375         "TEMP R0;\n"
3376         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3377         "MUL R0.x, R0.x, R0.y;\n"
3378         "KIL -R0.x;\n"
3379         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3380         "MOV result.depth.z, R0.x;\n"
3381         "END\n",
3382     };
3383
3384     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3385     if (!fprogram)
3386     {
3387         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3388         tex_type = tex_2d;
3389         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3390     }
3391
3392     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3393     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3394     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3395     checkGLcall("glProgramStringARB()");
3396
3397     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3398     if (pos != -1)
3399     {
3400         FIXME("Fragment program error at position %d: %s\n\n", pos,
3401             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3402         shader_arb_dump_program_source(fprogram);
3403     }
3404     else
3405     {
3406         GLint native;
3407
3408         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3409         checkGLcall("glGetProgramivARB()");
3410         if (!native) WARN("Program exceeds native resource limits.\n");
3411     }
3412
3413     return program_id;
3414 }
3415
3416 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3417         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3418 {
3419     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3420
3421     if(condcode)
3422     {
3423         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3424         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3425         /* Calculate the > 0.0031308 case */
3426         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3427         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3428         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3429         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3430         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3431         /* Calculate the < case */
3432         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3433     }
3434     else
3435     {
3436         /* Calculate the > 0.0031308 case */
3437         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3438         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3439         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3440         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3441         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3442         /* Calculate the < case */
3443         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3444         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3445         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3446         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3447         /* Store the components > 0.0031308 in the destination */
3448         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3449         /* Add the components that are < 0.0031308 */
3450         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3451         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3452         * result.color writes(.rgb first, then .a), or handle overwriting already written
3453         * components. The assembler uses a temporary register in this case, which is usually
3454         * not allocated from one of our registers that were used earlier.
3455         */
3456     }
3457     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3458 }
3459
3460 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3461 {
3462     const struct wined3d_shader_lconst *constant;
3463
3464     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3465     {
3466         if (constant->idx == idx)
3467         {
3468             return constant->value;
3469         }
3470     }
3471     return NULL;
3472 }
3473
3474 static void init_ps_input(const struct wined3d_shader *shader,
3475         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3476 {
3477     static const char * const texcoords[8] =
3478     {
3479         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3480         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3481     };
3482     unsigned int i;
3483     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3484     const char *semantic_name;
3485     DWORD semantic_idx;
3486
3487     switch(args->super.vp_mode)
3488     {
3489         case pretransformed:
3490         case fixedfunction:
3491             /* The pixelshader has to collect the varyings on its own. In any case properly load
3492              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3493              * other attribs to 0.0.
3494              *
3495              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3496              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3497              * load the texcoord attrib pointers to match the pixel shader signature
3498              */
3499             for(i = 0; i < MAX_REG_INPUT; i++)
3500             {
3501                 semantic_name = sig[i].semantic_name;
3502                 semantic_idx = sig[i].semantic_idx;
3503                 if (!semantic_name) continue;
3504
3505                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3506                 {
3507                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3508                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3509                     else priv->ps_input[i] = "0.0";
3510                 }
3511                 else if(args->super.vp_mode == fixedfunction)
3512                 {
3513                     priv->ps_input[i] = "0.0";
3514                 }
3515                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3516                 {
3517                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3518                     else priv->ps_input[i] = "0.0";
3519                 }
3520                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3521                 {
3522                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3523                     else priv->ps_input[i] = "0.0";
3524                 }
3525                 else
3526                 {
3527                     priv->ps_input[i] = "0.0";
3528                 }
3529
3530                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3531             }
3532             break;
3533
3534         case vertexshader:
3535             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3536              * fragment.color
3537              */
3538             for(i = 0; i < 8; i++)
3539             {
3540                 priv->ps_input[i] = texcoords[i];
3541             }
3542             priv->ps_input[8] = "fragment.color.primary";
3543             priv->ps_input[9] = "fragment.color.secondary";
3544             break;
3545     }
3546 }
3547
3548 /* GL locking is done by the caller */
3549 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3550         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3551         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3552 {
3553     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3554     const struct wined3d_shader_lconst *lconst;
3555     const DWORD *function = shader->function;
3556     GLuint retval;
3557     char fragcolor[16];
3558     DWORD *lconst_map = local_const_mapping(shader), next_local;
3559     struct shader_arb_ctx_priv priv_ctx;
3560     BOOL dcl_td = FALSE;
3561     BOOL want_nv_prog = FALSE;
3562     struct arb_pshader_private *shader_priv = shader->backend_data;
3563     GLint errPos;
3564     DWORD map;
3565
3566     char srgbtmp[4][4];
3567     unsigned int i, found = 0;
3568
3569     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3570     {
3571         if (!(map & 1)
3572                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3573                 || (reg_maps->shader_version.major < 2 && !i))
3574             continue;
3575
3576         sprintf(srgbtmp[found], "R%u", i);
3577         ++found;
3578         if (found == 4) break;
3579     }
3580
3581     switch(found) {
3582         case 0:
3583             sprintf(srgbtmp[0], "TA");
3584             sprintf(srgbtmp[1], "TB");
3585             sprintf(srgbtmp[2], "TC");
3586             sprintf(srgbtmp[3], "TD");
3587             dcl_td = TRUE;
3588             break;
3589         case 1:
3590             sprintf(srgbtmp[1], "TA");
3591             sprintf(srgbtmp[2], "TB");
3592             sprintf(srgbtmp[3], "TC");
3593             break;
3594         case 2:
3595             sprintf(srgbtmp[2], "TA");
3596             sprintf(srgbtmp[3], "TB");
3597             break;
3598         case 3:
3599             sprintf(srgbtmp[3], "TA");
3600             break;
3601         case 4:
3602             break;
3603     }
3604
3605     /*  Create the hw ARB shader */
3606     memset(&priv_ctx, 0, sizeof(priv_ctx));
3607     priv_ctx.cur_ps_args = args;
3608     priv_ctx.compiled_fprog = compiled;
3609     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3610     init_ps_input(shader, args, &priv_ctx);
3611     list_init(&priv_ctx.control_frames);
3612
3613     /* Avoid enabling NV_fragment_program* if we do not need it.
3614      *
3615      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3616      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3617      * is faster than what we gain from using higher native instructions. There are some things though
3618      * that cannot be emulated. In that case enable the extensions.
3619      * If the extension is enabled, instruction handlers that support both ways will use it.
3620      *
3621      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3622      * So enable the best we can get.
3623      */
3624     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3625        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3626     {
3627         want_nv_prog = TRUE;
3628     }
3629
3630     shader_addline(buffer, "!!ARBfp1.0\n");
3631     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3632     {
3633         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3634         priv_ctx.target_version = NV3;
3635     }
3636     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3637     {
3638         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3639         priv_ctx.target_version = NV2;
3640     } else {
3641         if(want_nv_prog)
3642         {
3643             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3644              * limits properly
3645              */
3646             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3647             ERR("Try GLSL\n");
3648         }
3649         priv_ctx.target_version = ARB;
3650     }
3651
3652     if (reg_maps->rt_mask > 1)
3653     {
3654         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3655     }
3656
3657     if (reg_maps->shader_version.major < 3)
3658     {
3659         switch(args->super.fog) {
3660             case FOG_OFF:
3661                 break;
3662             case FOG_LINEAR:
3663                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3664                 break;
3665             case FOG_EXP:
3666                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3667                 break;
3668             case FOG_EXP2:
3669                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3670                 break;
3671         }
3672     }
3673
3674     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3675      * unused temps away(but occupies them for the whole shader if they're used once). Always
3676      * declaring them avoids tricky bookkeeping work
3677      */
3678     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3679     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3680     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3681     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3682     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3683     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3684     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3685
3686     if (reg_maps->shader_version.major < 2)
3687     {
3688         strcpy(fragcolor, "R0");
3689     }
3690     else
3691     {
3692         if (args->super.srgb_correction)
3693         {
3694             if (shader->u.ps.color0_mov)
3695             {
3696                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3697             }
3698             else
3699             {
3700                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3701                 strcpy(fragcolor, "TMP_COLOR");
3702             }
3703         } else {
3704             strcpy(fragcolor, "result.color");
3705         }
3706     }
3707
3708     if(args->super.srgb_correction) {
3709         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3710                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3711         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3712                        srgb_sub_high, 0.0, 0.0, 0.0);
3713     }
3714
3715     /* Base Declarations */
3716     next_local = shader_generate_arb_declarations(shader, reg_maps,
3717             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3718
3719     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3720     {
3721         unsigned char bump_const;
3722
3723         if (!(map & 1)) continue;
3724
3725         bump_const = compiled->numbumpenvmatconsts;
3726         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3727         compiled->bumpenvmatconst[bump_const].texunit = i;
3728         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3729         compiled->luminanceconst[bump_const].texunit = i;
3730
3731         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3732          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3733          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3734          * textures due to conditional NP2 restrictions)
3735          *
3736          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3737          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3738          * their location is shader dependent anyway and they cannot be loaded globally.
3739          */
3740         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3741         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3742                        i, compiled->bumpenvmatconst[bump_const].const_num);
3743         compiled->numbumpenvmatconsts = bump_const + 1;
3744
3745         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3746
3747         compiled->luminanceconst[bump_const].const_num = next_local++;
3748         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3749                        i, compiled->luminanceconst[bump_const].const_num);
3750     }
3751
3752     for(i = 0; i < MAX_CONST_I; i++)
3753     {
3754         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3755         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3756         {
3757             const DWORD *control_values = find_loop_control_values(shader, i);
3758
3759             if(control_values)
3760             {
3761                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3762                                 control_values[0], control_values[1], control_values[2]);
3763             }
3764             else
3765             {
3766                 compiled->int_consts[i] = next_local;
3767                 compiled->num_int_consts++;
3768                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3769             }
3770         }
3771     }
3772
3773     if(reg_maps->vpos || reg_maps->usesdsy)
3774     {
3775         compiled->ycorrection = next_local;
3776         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3777
3778         if(reg_maps->vpos)
3779         {
3780             shader_addline(buffer, "TEMP vpos;\n");
3781             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3782              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3783              * ycorrection.z: 1.0
3784              * ycorrection.w: 0.0
3785              */
3786             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3787             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3788         }
3789     }
3790     else
3791     {
3792         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3793     }
3794
3795     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3796      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3797      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3798      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3799      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3800      * shader compilation errors and the subsequent errors when drawing with this shader. */
3801     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3802         unsigned char cur_fixup_sampler = 0;
3803
3804         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3805         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3806         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3807
3808         fixup->offset = next_local;
3809         fixup->super.active = 0;
3810
3811         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3812             if (!(map & (1 << i))) continue;
3813
3814             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3815                 fixup->super.active |= (1 << i);
3816                 fixup->super.idx[i] = cur_fixup_sampler++;
3817             } else {
3818                 FIXME("No free constant found to load NP2 fixup data into shader. "
3819                       "Sampling from this texture will probably look wrong.\n");
3820                 break;
3821             }
3822         }
3823
3824         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3825         if (fixup->super.num_consts) {
3826             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3827                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3828         }
3829     }
3830
3831     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3832     {
3833         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3834     }
3835
3836     /* Base Shader Body */
3837     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3838
3839     if(args->super.srgb_correction) {
3840         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3841                                   priv_ctx.target_version >= NV2);
3842     }
3843
3844     if(strcmp(fragcolor, "result.color")) {
3845         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3846     }
3847     shader_addline(buffer, "END\n");
3848
3849     /* TODO: change to resource.glObjectHandle or something like that */
3850     GL_EXTCALL(glGenProgramsARB(1, &retval));
3851
3852     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3853     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3854
3855     TRACE("Created hw pixel shader, prg=%d\n", retval);
3856     /* Create the program and check for errors */
3857     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3858                buffer->bsize, buffer->buffer));
3859     checkGLcall("glProgramStringARB()");
3860
3861     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3862     if (errPos != -1)
3863     {
3864         FIXME("HW PixelShader Error at position %d: %s\n\n",
3865               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3866         shader_arb_dump_program_source(buffer->buffer);
3867         retval = 0;
3868     }
3869     else
3870     {
3871         GLint native;
3872
3873         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3874         checkGLcall("glGetProgramivARB()");
3875         if (!native) WARN("Program exceeds native resource limits.\n");
3876     }
3877
3878     /* Load immediate constants */
3879     if (lconst_map)
3880     {
3881         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3882         {
3883             const float *value = (const float *)lconst->value;
3884             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3885             checkGLcall("glProgramLocalParameter4fvARB");
3886         }
3887         HeapFree(GetProcessHeap(), 0, lconst_map);
3888     }
3889
3890     return retval;
3891 }
3892
3893 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3894 {
3895     unsigned int i;
3896     int ret;
3897
3898     for(i = 0; i < MAX_REG_INPUT; i++)
3899     {
3900         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3901         {
3902             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3903             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3904             continue;
3905         }
3906
3907         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3908         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3909         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3910         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3911         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3912         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3913     }
3914     return 0;
3915 }
3916
3917 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3918 {
3919     struct wined3d_shader_signature_element *new;
3920     int i;
3921     char *name;
3922
3923     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3924     for(i = 0; i < MAX_REG_INPUT; i++)
3925     {
3926         if (!sig[i].semantic_name) continue;
3927
3928         new[i] = sig[i];
3929         /* Clone the semantic string */
3930         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3931         strcpy(name, sig[i].semantic_name);
3932         new[i].semantic_name = name;
3933     }
3934     return new;
3935 }
3936
3937 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3938 {
3939     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3940     struct ps_signature *found_sig;
3941
3942     if (entry)
3943     {
3944         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3945         TRACE("Found existing signature %u\n", found_sig->idx);
3946         return found_sig->idx;
3947     }
3948     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3949     found_sig->sig = clone_sig(sig);
3950     found_sig->idx = priv->ps_sig_number++;
3951     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3952     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3953     {
3954         ERR("Failed to insert program entry.\n");
3955     }
3956     return found_sig->idx;
3957 }
3958
3959 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3960         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3961 {
3962     unsigned int i, j;
3963     static const char * const texcoords[8] =
3964     {
3965         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3966         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3967     };
3968     struct wined3d_device *device = shader->device;
3969     const struct wined3d_shader_signature_element *sig;
3970     const char *semantic_name;
3971     DWORD semantic_idx, reg_idx;
3972
3973     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3974      * and varying 9 to result.color.secondary
3975      */
3976     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3977     {
3978         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3979         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3980         "result.color.primary", "result.color.secondary"
3981     };
3982
3983     if(sig_num == ~0)
3984     {
3985         TRACE("Pixel shader uses builtin varyings\n");
3986         /* Map builtins to builtins */
3987         for(i = 0; i < 8; i++)
3988         {
3989             priv_ctx->texcrd_output[i] = texcoords[i];
3990         }
3991         priv_ctx->color_output[0] = "result.color.primary";
3992         priv_ctx->color_output[1] = "result.color.secondary";
3993         priv_ctx->fog_output = "TMP_FOGCOORD";
3994
3995         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3996         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3997         {
3998             semantic_name = shader->output_signature[i].semantic_name;
3999             if (!semantic_name) continue;
4000
4001             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4002             {
4003                 TRACE("o%u is TMP_OUT\n", i);
4004                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4005                 else priv_ctx->vs_output[i] = "TA";
4006             }
4007             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4008             {
4009                 TRACE("o%u is result.pointsize\n", i);
4010                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4011                 else priv_ctx->vs_output[i] = "TA";
4012             }
4013             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4014             {
4015                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4016                 if (!shader->output_signature[i].semantic_idx)
4017                     priv_ctx->vs_output[i] = "result.color.primary";
4018                 else if (shader->output_signature[i].semantic_idx == 1)
4019                     priv_ctx->vs_output[i] = "result.color.secondary";
4020                 else priv_ctx->vs_output[i] = "TA";
4021             }
4022             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4023             {
4024                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4025                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4026                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4027             }
4028             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4029             {
4030                 TRACE("o%u is result.fogcoord\n", i);
4031                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4032                 else priv_ctx->vs_output[i] = "result.fogcoord";
4033             }
4034             else
4035             {
4036                 priv_ctx->vs_output[i] = "TA";
4037             }
4038         }
4039         return;
4040     }
4041
4042     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4043      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4044      */
4045     sig = device->stateBlock->state.pixel_shader->input_signature;
4046     TRACE("Pixel shader uses declared varyings\n");
4047
4048     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4049     for(i = 0; i < 8; i++)
4050     {
4051         priv_ctx->texcrd_output[i] = "TA";
4052     }
4053     priv_ctx->color_output[0] = "TA";
4054     priv_ctx->color_output[1] = "TA";
4055     priv_ctx->fog_output = "TA";
4056
4057     for(i = 0; i < MAX_REG_INPUT; i++)
4058     {
4059         semantic_name = sig[i].semantic_name;
4060         semantic_idx = sig[i].semantic_idx;
4061         reg_idx = sig[i].register_idx;
4062         if (!semantic_name) continue;
4063
4064         /* If a declared input register is not written by builtin arguments, don't write to it.
4065          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4066          *
4067          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4068          * to TMP_OUT in any case
4069          */
4070         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4071         {
4072             if (semantic_idx < 8)
4073                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4074         }
4075         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4076         {
4077             if (semantic_idx < 2)
4078                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4079         }
4080         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4081         {
4082             if (!semantic_idx)
4083                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4084         }
4085         else
4086         {
4087             continue;
4088         }
4089
4090         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4091                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4092         {
4093             compiled->need_color_unclamp = TRUE;
4094         }
4095     }
4096
4097     /* Map declared to declared */
4098     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4099     {
4100         /* Write unread output to TA to throw them away */
4101         priv_ctx->vs_output[i] = "TA";
4102         semantic_name = shader->output_signature[i].semantic_name;
4103         if (!semantic_name) continue;
4104
4105         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4106                 && !shader->output_signature[i].semantic_idx)
4107         {
4108             priv_ctx->vs_output[i] = "TMP_OUT";
4109             continue;
4110         }
4111         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4112                 && !shader->output_signature[i].semantic_idx)
4113         {
4114             priv_ctx->vs_output[i] = "result.pointsize";
4115             continue;
4116         }
4117
4118         for(j = 0; j < MAX_REG_INPUT; j++)
4119         {
4120             if (!sig[j].semantic_name) continue;
4121
4122             if (!strcmp(sig[j].semantic_name, semantic_name)
4123                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4124             {
4125                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4126
4127                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4128                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4129                 {
4130                     compiled->need_color_unclamp = TRUE;
4131                 }
4132             }
4133         }
4134     }
4135 }
4136
4137 /* GL locking is done by the caller */
4138 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4139         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4140         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4141 {
4142     const struct arb_vshader_private *shader_data = shader->backend_data;
4143     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4144     const struct wined3d_shader_lconst *lconst;
4145     const DWORD *function = shader->function;
4146     GLuint ret;
4147     DWORD next_local, *lconst_map = local_const_mapping(shader);
4148     struct shader_arb_ctx_priv priv_ctx;
4149     unsigned int i;
4150     GLint errPos;
4151
4152     memset(&priv_ctx, 0, sizeof(priv_ctx));
4153     priv_ctx.cur_vs_args = args;
4154     list_init(&priv_ctx.control_frames);
4155     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4156
4157     /*  Create the hw ARB shader */
4158     shader_addline(buffer, "!!ARBvp1.0\n");
4159
4160     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4161      * mesurable performance penalty, and we can always make use of it for clipplanes.
4162      */
4163     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4164     {
4165         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4166         priv_ctx.target_version = NV3;
4167         shader_addline(buffer, "ADDRESS aL;\n");
4168     }
4169     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4170     {
4171         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4172         priv_ctx.target_version = NV2;
4173         shader_addline(buffer, "ADDRESS aL;\n");
4174     } else {
4175         priv_ctx.target_version = ARB;
4176     }
4177
4178     shader_addline(buffer, "TEMP TMP_OUT;\n");
4179     if (reg_maps->fog)
4180         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4181     if (need_helper_const(shader_data, reg_maps, gl_info))
4182     {
4183         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4184     }
4185     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4186     {
4187         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4188         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4189     }
4190
4191     shader_addline(buffer, "TEMP TA;\n");
4192     shader_addline(buffer, "TEMP TB;\n");
4193
4194     /* Base Declarations */
4195     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4196             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4197
4198     for(i = 0; i < MAX_CONST_I; i++)
4199     {
4200         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4201         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4202         {
4203             const DWORD *control_values = find_loop_control_values(shader, i);
4204
4205             if(control_values)
4206             {
4207                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4208                                 control_values[0], control_values[1], control_values[2]);
4209             }
4210             else
4211             {
4212                 compiled->int_consts[i] = next_local;
4213                 compiled->num_int_consts++;
4214                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4215             }
4216         }
4217     }
4218
4219     /* We need a constant to fixup the final position */
4220     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4221     compiled->pos_fixup = next_local++;
4222
4223     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4224      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4225      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4226      * a replacement shader depend on the texcoord.w being set properly.
4227      *
4228      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4229      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4230      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4231      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4232      * this can eat a number of instructions, so skip it unless this cap is set as well
4233      */
4234     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4235     {
4236         struct wined3d_device *device = shader->device;
4237         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4238         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4239
4240         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4241         {
4242             int i;
4243             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4244             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4245             {
4246                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4247                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4248             }
4249         }
4250     }
4251
4252     /* The shader starts with the main function */
4253     priv_ctx.in_main_func = TRUE;
4254     /* Base Shader Body */
4255     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4256
4257     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4258             shader_data, args, reg_maps, gl_info, buffer);
4259
4260     shader_addline(buffer, "END\n");
4261
4262     /* TODO: change to resource.glObjectHandle or something like that */
4263     GL_EXTCALL(glGenProgramsARB(1, &ret));
4264
4265     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4266     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4267
4268     TRACE("Created hw vertex shader, prg=%d\n", ret);
4269     /* Create the program and check for errors */
4270     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4271                buffer->bsize, buffer->buffer));
4272     checkGLcall("glProgramStringARB()");
4273
4274     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4275     if (errPos != -1)
4276     {
4277         FIXME("HW VertexShader Error at position %d: %s\n\n",
4278               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4279         shader_arb_dump_program_source(buffer->buffer);
4280         ret = -1;
4281     }
4282     else
4283     {
4284         GLint native;
4285
4286         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4287         checkGLcall("glGetProgramivARB()");
4288         if (!native) WARN("Program exceeds native resource limits.\n");
4289
4290         /* Load immediate constants */
4291         if (lconst_map)
4292         {
4293             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4294             {
4295                 const float *value = (const float *)lconst->value;
4296                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4297             }
4298         }
4299     }
4300     HeapFree(GetProcessHeap(), 0, lconst_map);
4301
4302     return ret;
4303 }
4304
4305 /* GL locking is done by the caller */
4306 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4307         const struct arb_ps_compile_args *args)
4308 {
4309     struct wined3d_device *device = shader->device;
4310     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4311     UINT i;
4312     DWORD new_size;
4313     struct arb_ps_compiled_shader *new_array;
4314     struct wined3d_shader_buffer buffer;
4315     struct arb_pshader_private *shader_data;
4316     GLuint ret;
4317
4318     if (!shader->backend_data)
4319     {
4320         struct shader_arb_priv *priv = device->shader_priv;
4321
4322         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4323         shader_data = shader->backend_data;
4324         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4325
4326         if (shader->reg_maps.shader_version.major < 3)
4327             shader_data->input_signature_idx = ~0;
4328         else
4329             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4330
4331         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4332
4333         if (!device->vs_clipping)
4334             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4335                     gl_info->limits.texture_stages - 1);
4336         else
4337             shader_data->clipplane_emulation = ~0U;
4338     }
4339     shader_data = shader->backend_data;
4340
4341     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4342      * so a linear search is more performant than a hashmap or a binary search
4343      * (cache coherency etc)
4344      */
4345     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4346     {
4347         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4348             return &shader_data->gl_shaders[i];
4349     }
4350
4351     TRACE("No matching GL shader found, compiling a new shader\n");
4352     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4353         if (shader_data->num_gl_shaders)
4354         {
4355             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4356             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4357                                     new_size * sizeof(*shader_data->gl_shaders));
4358         } else {
4359             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4360             new_size = 1;
4361         }
4362
4363         if(!new_array) {
4364             ERR("Out of memory\n");
4365             return 0;
4366         }
4367         shader_data->gl_shaders = new_array;
4368         shader_data->shader_array_size = new_size;
4369     }
4370
4371     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4372
4373     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
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_pshader(shader, gl_info, &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 BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4390                                  const DWORD use_map, BOOL skip_int) {
4391     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4392     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4393     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4394     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4395     if(stored->ps_signature != new->ps_signature) return FALSE;
4396     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4397     if(skip_int) return TRUE;
4398
4399     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4400 }
4401
4402 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4403         const struct arb_vs_compile_args *args)
4404 {
4405     struct wined3d_device *device = shader->device;
4406     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4407     DWORD use_map = device->strided_streams.use_map;
4408     UINT i;
4409     DWORD new_size;
4410     struct arb_vs_compiled_shader *new_array;
4411     struct wined3d_shader_buffer buffer;
4412     struct arb_vshader_private *shader_data;
4413     GLuint ret;
4414
4415     if (!shader->backend_data)
4416     {
4417         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4418
4419         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4420         shader_data = shader->backend_data;
4421
4422         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4423                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4424         {
4425             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4426             {
4427                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4428                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4429                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4430             }
4431             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4432                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4433             else if (reg_maps->max_rel_offset > 63)
4434                 shader_data->rel_offset = reg_maps->min_rel_offset;
4435         }
4436     }
4437     shader_data = shader->backend_data;
4438
4439     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4440      * so a linear search is more performant than a hashmap or a binary search
4441      * (cache coherency etc)
4442      */
4443     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4444         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4445                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4446         {
4447             return &shader_data->gl_shaders[i];
4448         }
4449     }
4450
4451     TRACE("No matching GL shader found, compiling a new shader\n");
4452
4453     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4454         if (shader_data->num_gl_shaders)
4455         {
4456             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4457             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4458                                     new_size * sizeof(*shader_data->gl_shaders));
4459         } else {
4460             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4461             new_size = 1;
4462         }
4463
4464         if(!new_array) {
4465             ERR("Out of memory\n");
4466             return 0;
4467         }
4468         shader_data->gl_shaders = new_array;
4469         shader_data->shader_array_size = new_size;
4470     }
4471
4472     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4473
4474     if (!shader_buffer_init(&buffer))
4475     {
4476         ERR("Failed to initialize shader buffer.\n");
4477         return 0;
4478     }
4479
4480     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4481             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4482     shader_buffer_free(&buffer);
4483     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4484
4485     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4486 }
4487
4488 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4489         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4490 {
4491     struct wined3d_device *device = shader->device;
4492     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4493     int i;
4494     WORD int_skip;
4495
4496     find_ps_compile_args(state, shader, &args->super);
4497
4498     /* This forces all local boolean constants to 1 to make them stateblock independent */
4499     args->bools = shader->reg_maps.local_bool_consts;
4500
4501     for(i = 0; i < MAX_CONST_B; i++)
4502     {
4503         if (state->ps_consts_b[i])
4504             args->bools |= ( 1 << i);
4505     }
4506
4507     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4508      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4509      * duplicate the shader than have a no-op KIL instruction in every shader
4510      */
4511     if (!device->vs_clipping && use_vs(state)
4512             && state->render_states[WINED3D_RS_CLIPPING]
4513             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4514         args->clip = 1;
4515     else
4516         args->clip = 0;
4517
4518     /* Skip if unused or local, or supported natively */
4519     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4520     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4521     {
4522         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4523         return;
4524     }
4525
4526     for(i = 0; i < MAX_CONST_I; i++)
4527     {
4528         if(int_skip & (1 << i))
4529         {
4530             args->loop_ctrl[i][0] = 0;
4531             args->loop_ctrl[i][1] = 0;
4532             args->loop_ctrl[i][2] = 0;
4533         }
4534         else
4535         {
4536             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4537             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4538             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4539         }
4540     }
4541 }
4542
4543 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4544         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4545 {
4546     struct wined3d_device *device = shader->device;
4547     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4548     int i;
4549     WORD int_skip;
4550
4551     find_vs_compile_args(state, shader, &args->super);
4552
4553     args->clip.boolclip_compare = 0;
4554     if (use_ps(state))
4555     {
4556         const struct wined3d_shader *ps = state->pixel_shader;
4557         const struct arb_pshader_private *shader_priv = ps->backend_data;
4558         args->ps_signature = shader_priv->input_signature_idx;
4559
4560         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4561     }
4562     else
4563     {
4564         args->ps_signature = ~0;
4565         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4566         {
4567             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4568         }
4569         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4570     }
4571
4572     if (args->clip.boolclip.clip_texcoord)
4573     {
4574         if (state->render_states[WINED3D_RS_CLIPPING])
4575             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4576         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4577     }
4578
4579     /* This forces all local boolean constants to 1 to make them stateblock independent */
4580     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4581     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4582     for(i = 0; i < MAX_CONST_B; i++)
4583     {
4584         if (state->vs_consts_b[i])
4585             args->clip.boolclip.bools |= ( 1 << i);
4586     }
4587
4588     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4589     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4590     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4591     args->vertex.samplers[3] = 0;
4592
4593     /* Skip if unused or local */
4594     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4595     /* This is about flow control, not clipping. */
4596     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4597     {
4598         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4599         return;
4600     }
4601
4602     for(i = 0; i < MAX_CONST_I; i++)
4603     {
4604         if(int_skip & (1 << i))
4605         {
4606             args->loop_ctrl[i][0] = 0;
4607             args->loop_ctrl[i][1] = 0;
4608             args->loop_ctrl[i][2] = 0;
4609         }
4610         else
4611         {
4612             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4613             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4614             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4615         }
4616     }
4617 }
4618
4619 /* GL locking is done by the caller */
4620 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4621 {
4622     struct wined3d_device *device = context->swapchain->device;
4623     struct shader_arb_priv *priv = device->shader_priv;
4624     const struct wined3d_gl_info *gl_info = context->gl_info;
4625     const struct wined3d_state *state = &device->stateBlock->state;
4626     int i;
4627
4628     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4629     if (usePS)
4630     {
4631         struct wined3d_shader *ps = state->pixel_shader;
4632         struct arb_ps_compile_args compile_args;
4633         struct arb_ps_compiled_shader *compiled;
4634
4635         TRACE("Using pixel shader %p.\n", ps);
4636         find_arb_ps_compile_args(state, ps, &compile_args);
4637         compiled = find_arb_pshader(ps, &compile_args);
4638         priv->current_fprogram_id = compiled->prgId;
4639         priv->compiled_fprog = compiled;
4640
4641         /* Bind the fragment program */
4642         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4643         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4644
4645         if(!priv->use_arbfp_fixed_func) {
4646             /* Enable OpenGL fragment programs */
4647             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4648             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4649         }
4650         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4651                 device, priv->current_fprogram_id);
4652
4653         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4654          * a 1.x and newer shader, reload the first 8 constants
4655          */
4656         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4657         {
4658             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4659             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4660             for(i = 0; i < 8; i++)
4661             {
4662                 priv->pshader_const_dirty[i] = 1;
4663             }
4664             /* Also takes care of loading local constants */
4665             shader_arb_load_constants(context, TRUE, FALSE);
4666         }
4667         else
4668         {
4669             UINT rt_height = state->fb->render_targets[0]->resource.height;
4670             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4671         }
4672
4673         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4674         if (compiled->np2fixup_info.super.active)
4675             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4676     }
4677     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4678     {
4679         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4680         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4681         * replacement shader
4682         */
4683         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4684         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4685         priv->current_fprogram_id = 0;
4686     }
4687
4688     if (useVS)
4689     {
4690         struct wined3d_shader *vs = state->vertex_shader;
4691         struct arb_vs_compile_args compile_args;
4692         struct arb_vs_compiled_shader *compiled;
4693
4694         TRACE("Using vertex shader %p\n", vs);
4695         find_arb_vs_compile_args(state, vs, &compile_args);
4696         compiled = find_arb_vshader(vs, &compile_args);
4697         priv->current_vprogram_id = compiled->prgId;
4698         priv->compiled_vprog = compiled;
4699
4700         /* Bind the vertex program */
4701         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4702         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4703
4704         /* Enable OpenGL vertex programs */
4705         glEnable(GL_VERTEX_PROGRAM_ARB);
4706         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4707         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4708         shader_arb_vs_local_constants(compiled, context, state);
4709
4710         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4711             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4712
4713             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4714             {
4715                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4716                 checkGLcall("glClampColorARB");
4717             } else {
4718                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4719             }
4720         }
4721     }
4722     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4723     {
4724         priv->current_vprogram_id = 0;
4725         glDisable(GL_VERTEX_PROGRAM_ARB);
4726         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4727     }
4728 }
4729
4730 /* GL locking is done by the caller */
4731 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4732         enum tex_types tex_type, const SIZE *ds_mask_size)
4733 {
4734     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4735     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4736     struct shader_arb_priv *priv = shader_priv;
4737     GLuint *blt_fprogram;
4738
4739     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4740     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4741     glEnable(GL_VERTEX_PROGRAM_ARB);
4742
4743     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4744     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4745     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4746     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4747     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4748 }
4749
4750 /* GL locking is done by the caller */
4751 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4752 {
4753     struct shader_arb_priv *priv = shader_priv;
4754
4755     if (priv->current_vprogram_id) {
4756         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4757         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4758
4759         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4760     }
4761     else
4762     {
4763         glDisable(GL_VERTEX_PROGRAM_ARB);
4764         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4765     }
4766
4767     if (priv->current_fprogram_id) {
4768         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4769         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4770
4771         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4772     }
4773     else if(!priv->use_arbfp_fixed_func)
4774     {
4775         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4776         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4777     }
4778 }
4779
4780 static void shader_arb_destroy(struct wined3d_shader *shader)
4781 {
4782     struct wined3d_device *device = shader->device;
4783     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4784
4785     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4786     {
4787         struct arb_pshader_private *shader_data = shader->backend_data;
4788         UINT i;
4789
4790         if(!shader_data) return; /* This can happen if a shader was never compiled */
4791
4792         if (shader_data->num_gl_shaders)
4793         {
4794             struct wined3d_context *context = context_acquire(device, NULL);
4795
4796             ENTER_GL();
4797             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4798             {
4799                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4800                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4801             }
4802             LEAVE_GL();
4803
4804             context_release(context);
4805         }
4806
4807         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4808         HeapFree(GetProcessHeap(), 0, shader_data);
4809         shader->backend_data = NULL;
4810     }
4811     else
4812     {
4813         struct arb_vshader_private *shader_data = shader->backend_data;
4814         UINT i;
4815
4816         if(!shader_data) return; /* This can happen if a shader was never compiled */
4817
4818         if (shader_data->num_gl_shaders)
4819         {
4820             struct wined3d_context *context = context_acquire(device, NULL);
4821
4822             ENTER_GL();
4823             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4824             {
4825                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4826                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4827             }
4828             LEAVE_GL();
4829
4830             context_release(context);
4831         }
4832
4833         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4834         HeapFree(GetProcessHeap(), 0, shader_data);
4835         shader->backend_data = NULL;
4836     }
4837 }
4838
4839 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4840 {
4841     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4842     return compare_sig(key, e->sig);
4843 }
4844
4845 static const struct wine_rb_functions sig_tree_functions =
4846 {
4847     wined3d_rb_alloc,
4848     wined3d_rb_realloc,
4849     wined3d_rb_free,
4850     sig_tree_compare
4851 };
4852
4853 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4854 {
4855     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4856
4857     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4858             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4859     if (!priv->vshader_const_dirty)
4860         goto fail;
4861     memset(priv->vshader_const_dirty, 1,
4862            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4863
4864     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4865             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4866     if (!priv->pshader_const_dirty)
4867         goto fail;
4868     memset(priv->pshader_const_dirty, 1,
4869             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4870
4871     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4872     {
4873         ERR("RB tree init failed\n");
4874         goto fail;
4875     }
4876     device->shader_priv = priv;
4877     return WINED3D_OK;
4878
4879 fail:
4880     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4881     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4882     HeapFree(GetProcessHeap(), 0, priv);
4883     return E_OUTOFMEMORY;
4884 }
4885
4886 static void release_signature(struct wine_rb_entry *entry, void *context)
4887 {
4888     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4889     int i;
4890     for(i = 0; i < MAX_REG_INPUT; i++)
4891     {
4892         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4893     }
4894     HeapFree(GetProcessHeap(), 0, sig->sig);
4895     HeapFree(GetProcessHeap(), 0, sig);
4896 }
4897
4898 /* Context activation is done by the caller. */
4899 static void shader_arb_free(struct wined3d_device *device)
4900 {
4901     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4902     struct shader_arb_priv *priv = device->shader_priv;
4903     int i;
4904
4905     ENTER_GL();
4906     if(priv->depth_blt_vprogram_id) {
4907         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4908     }
4909     for (i = 0; i < tex_type_count; ++i)
4910     {
4911         if (priv->depth_blt_fprogram_id_full[i])
4912         {
4913             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4914         }
4915         if (priv->depth_blt_fprogram_id_masked[i])
4916         {
4917             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4918         }
4919     }
4920     LEAVE_GL();
4921
4922     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4923     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4924     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4925     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4926 }
4927
4928 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4929 {
4930     struct shader_arb_priv *priv = shader_priv;
4931
4932     if (priv->last_context == context)
4933         priv->last_context = NULL;
4934 }
4935
4936 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4937 {
4938     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4939     {
4940         DWORD vs_consts;
4941
4942         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4943          * for vertex programs. If the native limit is less than that it's
4944          * not very useful, and e.g. Mesa swrast returns 0, probably to
4945          * indicate it's a software implementation. */
4946         if (gl_info->limits.arb_vs_native_constants < 96)
4947             vs_consts = gl_info->limits.arb_vs_float_constants;
4948         else
4949             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4950
4951         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4952         {
4953             caps->VertexShaderVersion = 3;
4954             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4955         }
4956         else if (vs_consts >= 256)
4957         {
4958             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4959             caps->VertexShaderVersion = 2;
4960             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4961         }
4962         else
4963         {
4964             caps->VertexShaderVersion = 1;
4965             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4966         }
4967         caps->MaxVertexShaderConst = vs_consts;
4968     }
4969     else
4970     {
4971         caps->VertexShaderVersion = 0;
4972         caps->MaxVertexShaderConst = 0;
4973     }
4974
4975     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4976     {
4977         DWORD ps_consts;
4978
4979         /* Similar as above for vertex programs, but the minimum for fragment
4980          * programs is 24. */
4981         if (gl_info->limits.arb_ps_native_constants < 24)
4982             ps_consts = gl_info->limits.arb_ps_float_constants;
4983         else
4984             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4985
4986         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4987         {
4988             caps->PixelShaderVersion = 3;
4989             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4990         }
4991         else if (ps_consts >= 32)
4992         {
4993             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4994             caps->PixelShaderVersion = 2;
4995             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4996         }
4997         else
4998         {
4999             caps->PixelShaderVersion = 1;
5000             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5001         }
5002         caps->PixelShader1xMaxValue = 8.0f;
5003         caps->MaxPixelShaderConst = ps_consts;
5004     }
5005     else
5006     {
5007         caps->PixelShaderVersion = 0;
5008         caps->PixelShader1xMaxValue = 0.0f;
5009         caps->MaxPixelShaderConst = 0;
5010     }
5011
5012     caps->VSClipping = use_nv_clip(gl_info);
5013 }
5014
5015 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5016 {
5017     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5018     {
5019         TRACE("Checking support for color_fixup:\n");
5020         dump_color_fixup_desc(fixup);
5021     }
5022
5023     /* We support everything except complex conversions. */
5024     if (!is_complex_fixup(fixup))
5025     {
5026         TRACE("[OK]\n");
5027         return TRUE;
5028     }
5029
5030     TRACE("[FAILED]\n");
5031     return FALSE;
5032 }
5033
5034 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5035     DWORD shift;
5036     char write_mask[20], regstr[50];
5037     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5038     BOOL is_color = FALSE;
5039     const struct wined3d_shader_dst_param *dst;
5040
5041     if (!ins->dst_count) return;
5042
5043     dst = &ins->dst[0];
5044     shift = dst->shift;
5045     if (!shift) return; /* Saturate alone is handled by the instructions */
5046
5047     shader_arb_get_write_mask(ins, dst, write_mask);
5048     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5049
5050     /* Generate a line that does the output modifier computation
5051      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5052      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5053      */
5054     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5055                    regstr, write_mask, regstr, shift_tab[shift]);
5056 }
5057
5058 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5059 {
5060     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
5061     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
5062     /* WINED3DSIH_AND           */ NULL,
5063     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5064     /* WINED3DSIH_BREAK         */ shader_hw_break,
5065     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5066     /* WINED3DSIH_BREAKP        */ NULL,
5067     /* WINED3DSIH_CALL          */ shader_hw_call,
5068     /* WINED3DSIH_CALLNZ        */ NULL,
5069     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5070     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5071     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5072     /* WINED3DSIH_CUT           */ NULL,
5073     /* WINED3DSIH_DCL           */ NULL,
5074     /* WINED3DSIH_DEF           */ NULL,
5075     /* WINED3DSIH_DEFB          */ NULL,
5076     /* WINED3DSIH_DEFI          */ NULL,
5077     /* WINED3DSIH_DIV           */ NULL,
5078     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5079     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5080     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5081     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5082     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5083     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5084     /* WINED3DSIH_ELSE          */ shader_hw_else,
5085     /* WINED3DSIH_EMIT          */ NULL,
5086     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5087     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5088     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5089     /* WINED3DSIH_EQ            */ NULL,
5090     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5091     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5092     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5093     /* WINED3DSIH_FTOI          */ NULL,
5094     /* WINED3DSIH_GE            */ NULL,
5095     /* WINED3DSIH_IADD          */ NULL,
5096     /* WINED3DSIH_IEQ           */ NULL,
5097     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5098     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5099     /* WINED3DSIH_IGE           */ NULL,
5100     /* WINED3DSIH_IMUL          */ NULL,
5101     /* WINED3DSIH_ITOF          */ NULL,
5102     /* WINED3DSIH_LABEL         */ shader_hw_label,
5103     /* WINED3DSIH_LD            */ NULL,
5104     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5105     /* WINED3DSIH_LOG           */ shader_hw_log,
5106     /* WINED3DSIH_LOGP          */ shader_hw_log,
5107     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5108     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5109     /* WINED3DSIH_LT            */ NULL,
5110     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5111     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5112     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5113     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5114     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5115     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5116     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5117     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5118     /* WINED3DSIH_MOV           */ shader_hw_mov,
5119     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5120     /* WINED3DSIH_MOVC          */ NULL,
5121     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5122     /* WINED3DSIH_NOP           */ shader_hw_nop,
5123     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5124     /* WINED3DSIH_PHASE         */ NULL,
5125     /* WINED3DSIH_POW           */ shader_hw_pow,
5126     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5127     /* WINED3DSIH_REP           */ shader_hw_rep,
5128     /* WINED3DSIH_RET           */ shader_hw_ret,
5129     /* WINED3DSIH_ROUND_NI      */ NULL,
5130     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5131     /* WINED3DSIH_SAMPLE        */ NULL,
5132     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5133     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5134     /* WINED3DSIH_SETP          */ NULL,
5135     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5136     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5137     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5138     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5139     /* WINED3DSIH_SQRT          */ NULL,
5140     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5141     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5142     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5143     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5144     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5145     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5146     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5147     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5148     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5149     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5150     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5151     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5152     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5153     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5154     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5155     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5156     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5157     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5158     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5159     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5160     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5161     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5162     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5163     /* WINED3DSIH_UDIV          */ NULL,
5164     /* WINED3DSIH_USHR          */ NULL,
5165     /* WINED3DSIH_UTOF          */ NULL,
5166     /* WINED3DSIH_XOR           */ NULL,
5167 };
5168
5169 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5170         const struct wined3d_shader *shader, DWORD idx)
5171 {
5172     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5173     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5174     const struct wined3d_shader_lconst *constant;
5175     WORD bools = 0;
5176     WORD flag = (1 << idx);
5177     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5178
5179     if (reg_maps->local_bool_consts & flag)
5180     {
5181         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5182         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5183         {
5184             if (constant->idx == idx)
5185             {
5186                 return constant->value[0];
5187             }
5188         }
5189         ERR("Local constant not found\n");
5190         return FALSE;
5191     }
5192     else
5193     {
5194         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5195         else bools = priv->cur_ps_args->bools;
5196         return bools & flag;
5197     }
5198 }
5199
5200 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5201         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5202 {
5203     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5204     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5205
5206     /* Integer constants can either be a local constant, or they can be stored in the shader
5207      * type specific compile args. */
5208     if (reg_maps->local_int_consts & (1 << idx))
5209     {
5210         const struct wined3d_shader_lconst *constant;
5211
5212         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5213         {
5214             if (constant->idx == idx)
5215             {
5216                 loop_control->count = constant->value[0];
5217                 loop_control->start = constant->value[1];
5218                 /* Step is signed. */
5219                 loop_control->step = (int)constant->value[2];
5220                 return;
5221             }
5222         }
5223         /* If this happens the flag was set incorrectly */
5224         ERR("Local constant not found\n");
5225         loop_control->count = 0;
5226         loop_control->start = 0;
5227         loop_control->step = 0;
5228         return;
5229     }
5230
5231     switch (reg_maps->shader_version.type)
5232     {
5233         case WINED3D_SHADER_TYPE_VERTEX:
5234             /* Count and aL start value are unsigned */
5235             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5236             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5237             /* Step is signed. */
5238             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5239             break;
5240
5241         case WINED3D_SHADER_TYPE_PIXEL:
5242             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5243             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5244             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5245             break;
5246
5247         default:
5248             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5249             break;
5250     }
5251 }
5252
5253 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5254 {
5255     unsigned int i;
5256     struct wined3d_shader_dst_param *dst_param = NULL;
5257     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5258     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5259     if(!rec)
5260     {
5261         ERR("Out of memory\n");
5262         return;
5263     }
5264
5265     rec->ins = *ins;
5266     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5267     if(!dst_param) goto free;
5268     *dst_param = *ins->dst;
5269     if(ins->dst->reg.rel_addr)
5270     {
5271         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5272         if(!rel_addr) goto free;
5273         *rel_addr = *ins->dst->reg.rel_addr;
5274         dst_param->reg.rel_addr = rel_addr;
5275     }
5276     rec->ins.dst = dst_param;
5277
5278     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5279     if(!src_param) goto free;
5280     for(i = 0; i < ins->src_count; i++)
5281     {
5282         src_param[i] = ins->src[i];
5283         if(ins->src[i].reg.rel_addr)
5284         {
5285             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5286             if(!rel_addr) goto free;
5287             *rel_addr = *ins->src[i].reg.rel_addr;
5288             src_param[i].reg.rel_addr = rel_addr;
5289         }
5290     }
5291     rec->ins.src = src_param;
5292     list_add_tail(list, &rec->entry);
5293     return;
5294
5295 free:
5296     ERR("Out of memory\n");
5297     if(dst_param)
5298     {
5299         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5300         HeapFree(GetProcessHeap(), 0, dst_param);
5301     }
5302     if(src_param)
5303     {
5304         for(i = 0; i < ins->src_count; i++)
5305         {
5306             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5307         }
5308         HeapFree(GetProcessHeap(), 0, src_param);
5309     }
5310     HeapFree(GetProcessHeap(), 0, rec);
5311 }
5312
5313 static void free_recorded_instruction(struct list *list)
5314 {
5315     struct recorded_instruction *rec_ins, *entry2;
5316     unsigned int i;
5317
5318     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5319     {
5320         list_remove(&rec_ins->entry);
5321         if(rec_ins->ins.dst)
5322         {
5323             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5324             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5325         }
5326         if(rec_ins->ins.src)
5327         {
5328             for(i = 0; i < rec_ins->ins.src_count; i++)
5329             {
5330                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5331             }
5332             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5333         }
5334         HeapFree(GetProcessHeap(), 0, rec_ins);
5335     }
5336 }
5337
5338 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5339     SHADER_HANDLER hw_fct;
5340     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5341     const struct wined3d_shader *shader = ins->ctx->shader;
5342     struct control_frame *control_frame;
5343     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5344     BOOL bool_const;
5345
5346     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5347     {
5348         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5349         list_add_head(&priv->control_frames, &control_frame->entry);
5350
5351         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5352         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5353
5354         if(priv->target_version >= NV2)
5355         {
5356             control_frame->no.loop = priv->num_loops++;
5357             priv->loop_depth++;
5358         }
5359         else
5360         {
5361             /* Don't bother recording when we're in a not used if branch */
5362             if(priv->muted)
5363             {
5364                 return;
5365             }
5366
5367             if(!priv->recording)
5368             {
5369                 list_init(&priv->record);
5370                 priv->recording = TRUE;
5371                 control_frame->outer_loop = TRUE;
5372                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5373                 return; /* Instruction is handled */
5374             }
5375             /* Record this loop in the outer loop's recording */
5376         }
5377     }
5378     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5379     {
5380         if(priv->target_version >= NV2)
5381         {
5382             /* Nothing to do. The control frame is popped after the HW instr handler */
5383         }
5384         else
5385         {
5386             struct list *e = list_head(&priv->control_frames);
5387             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5388             list_remove(&control_frame->entry);
5389
5390             if(control_frame->outer_loop)
5391             {
5392                 unsigned int iteration;
5393                 int aL = 0;
5394                 struct list copy;
5395
5396                 /* Turn off recording before playback */
5397                 priv->recording = FALSE;
5398
5399                 /* Move the recorded instructions to a separate list and get them out of the private data
5400                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5401                  * be recorded again, thus priv->record might be overwritten
5402                  */
5403                 list_init(&copy);
5404                 list_move_tail(&copy, &priv->record);
5405                 list_init(&priv->record);
5406
5407                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5408                 {
5409                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5410                                    control_frame->loop_control.count, control_frame->loop_control.start,
5411                                    control_frame->loop_control.step);
5412                     aL = control_frame->loop_control.start;
5413                 }
5414                 else
5415                 {
5416                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5417                 }
5418
5419                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5420                 {
5421                     struct recorded_instruction *rec_ins;
5422                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5423                     {
5424                         priv->aL = aL;
5425                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5426                     }
5427                     else
5428                     {
5429                         shader_addline(buffer, "#Iteration %u\n", iteration);
5430                     }
5431
5432                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5433                     {
5434                         shader_arb_handle_instruction(&rec_ins->ins);
5435                     }
5436
5437                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5438                     {
5439                         aL += control_frame->loop_control.step;
5440                     }
5441                 }
5442                 shader_addline(buffer, "#end loop/rep\n");
5443
5444                 free_recorded_instruction(&copy);
5445                 HeapFree(GetProcessHeap(), 0, control_frame);
5446                 return; /* Instruction is handled */
5447             }
5448             else
5449             {
5450                 /* This is a nested loop. Proceed to the normal recording function */
5451                 HeapFree(GetProcessHeap(), 0, control_frame);
5452             }
5453         }
5454     }
5455
5456     if(priv->recording)
5457     {
5458         record_instruction(&priv->record, ins);
5459         return;
5460     }
5461
5462     /* boolean if */
5463     if(ins->handler_idx == WINED3DSIH_IF)
5464     {
5465         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5466         list_add_head(&priv->control_frames, &control_frame->entry);
5467         control_frame->type = IF;
5468
5469         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5470         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5471         if (!priv->muted && !bool_const)
5472         {
5473             shader_addline(buffer, "#if(FALSE){\n");
5474             priv->muted = TRUE;
5475             control_frame->muting = TRUE;
5476         }
5477         else shader_addline(buffer, "#if(TRUE) {\n");
5478
5479         return; /* Instruction is handled */
5480     }
5481     else if(ins->handler_idx == WINED3DSIH_IFC)
5482     {
5483         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5484         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5485         control_frame->type = IFC;
5486         control_frame->no.ifc = priv->num_ifcs++;
5487         list_add_head(&priv->control_frames, &control_frame->entry);
5488     }
5489     else if(ins->handler_idx == WINED3DSIH_ELSE)
5490     {
5491         struct list *e = list_head(&priv->control_frames);
5492         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5493
5494         if(control_frame->type == IF)
5495         {
5496             shader_addline(buffer, "#} else {\n");
5497             if(!priv->muted && !control_frame->muting)
5498             {
5499                 priv->muted = TRUE;
5500                 control_frame->muting = TRUE;
5501             }
5502             else if(control_frame->muting) priv->muted = FALSE;
5503             return; /* Instruction is handled. */
5504         }
5505         /* In case of an ifc, generate a HW shader instruction */
5506     }
5507     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5508     {
5509         struct list *e = list_head(&priv->control_frames);
5510         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5511
5512         if(control_frame->type == IF)
5513         {
5514             shader_addline(buffer, "#} endif\n");
5515             if(control_frame->muting) priv->muted = FALSE;
5516             list_remove(&control_frame->entry);
5517             HeapFree(GetProcessHeap(), 0, control_frame);
5518             return; /* Instruction is handled */
5519         }
5520     }
5521
5522     if(priv->muted) return;
5523
5524     /* Select handler */
5525     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5526
5527     /* Unhandled opcode */
5528     if (!hw_fct)
5529     {
5530         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5531         return;
5532     }
5533     hw_fct(ins);
5534
5535     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5536     {
5537         struct list *e = list_head(&priv->control_frames);
5538         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5539         list_remove(&control_frame->entry);
5540         HeapFree(GetProcessHeap(), 0, control_frame);
5541         priv->loop_depth--;
5542     }
5543     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5544     {
5545         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5546         struct list *e = list_head(&priv->control_frames);
5547         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5548         list_remove(&control_frame->entry);
5549         HeapFree(GetProcessHeap(), 0, control_frame);
5550     }
5551
5552
5553     shader_arb_add_instruction_modifiers(ins);
5554 }
5555
5556 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5557 {
5558     shader_arb_handle_instruction,
5559     shader_arb_select,
5560     shader_arb_select_depth_blt,
5561     shader_arb_deselect_depth_blt,
5562     shader_arb_update_float_vertex_constants,
5563     shader_arb_update_float_pixel_constants,
5564     shader_arb_load_constants,
5565     shader_arb_load_np2fixup_constants,
5566     shader_arb_destroy,
5567     shader_arb_alloc,
5568     shader_arb_free,
5569     shader_arb_context_destroyed,
5570     shader_arb_get_caps,
5571     shader_arb_color_fixup_supported,
5572 };
5573
5574 /* ARB_fragment_program fixed function pipeline replacement definitions */
5575 #define ARB_FFP_CONST_TFACTOR           0
5576 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5577 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5578 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5579 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5580
5581 struct arbfp_ffp_desc
5582 {
5583     struct ffp_frag_desc parent;
5584     GLuint shader;
5585     unsigned int num_textures_used;
5586 };
5587
5588 /* Context activation and GL locking are done by the caller. */
5589 static void arbfp_enable(BOOL enable)
5590 {
5591     if(enable) {
5592         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5593         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5594     } else {
5595         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5596         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5597     }
5598 }
5599
5600 static HRESULT arbfp_alloc(struct wined3d_device *device)
5601 {
5602     struct shader_arb_priv *priv;
5603     /* Share private data between the shader backend and the pipeline replacement, if both
5604      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5605      * if no pixel shader is bound or not
5606      */
5607     if (device->shader_backend == &arb_program_shader_backend)
5608     {
5609         device->fragment_priv = device->shader_priv;
5610     }
5611     else
5612     {
5613         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5614         if (!device->fragment_priv) return E_OUTOFMEMORY;
5615     }
5616     priv = device->fragment_priv;
5617     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5618     {
5619         ERR("Failed to initialize rbtree.\n");
5620         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5621         return E_OUTOFMEMORY;
5622     }
5623     priv->use_arbfp_fixed_func = TRUE;
5624     return WINED3D_OK;
5625 }
5626
5627 /* Context activation is done by the caller. */
5628 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5629 {
5630     const struct wined3d_gl_info *gl_info = context;
5631     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5632
5633     ENTER_GL();
5634     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5635     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5636     HeapFree(GetProcessHeap(), 0, entry_arb);
5637     LEAVE_GL();
5638 }
5639
5640 /* Context activation is done by the caller. */
5641 static void arbfp_free(struct wined3d_device *device)
5642 {
5643     struct shader_arb_priv *priv = device->fragment_priv;
5644
5645     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5646     priv->use_arbfp_fixed_func = FALSE;
5647
5648     if (device->shader_backend != &arb_program_shader_backend)
5649     {
5650         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5651     }
5652 }
5653
5654 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5655 {
5656     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5657     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5658                            WINED3DTEXOPCAPS_SELECTARG1                  |
5659                            WINED3DTEXOPCAPS_SELECTARG2                  |
5660                            WINED3DTEXOPCAPS_MODULATE4X                  |
5661                            WINED3DTEXOPCAPS_MODULATE2X                  |
5662                            WINED3DTEXOPCAPS_MODULATE                    |
5663                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5664                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5665                            WINED3DTEXOPCAPS_ADD                         |
5666                            WINED3DTEXOPCAPS_SUBTRACT                    |
5667                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5668                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5669                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5670                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5671                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5672                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5673                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5674                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5675                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5676                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5677                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5678                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5679                            WINED3DTEXOPCAPS_LERP                        |
5680                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5681                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5682
5683     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5684
5685     caps->MaxTextureBlendStages   = 8;
5686     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5687 }
5688
5689 static void state_texfactor_arbfp(struct wined3d_context *context,
5690         const struct wined3d_state *state, DWORD state_id)
5691 {
5692     struct wined3d_device *device = context->swapchain->device;
5693     const struct wined3d_gl_info *gl_info = context->gl_info;
5694     float col[4];
5695
5696     /* Don't load the parameter if we're using an arbfp pixel shader,
5697      * otherwise we'll overwrite application provided constants. */
5698     if (device->shader_backend == &arb_program_shader_backend)
5699     {
5700         struct shader_arb_priv *priv;
5701
5702         if (use_ps(state)) return;
5703
5704         priv = device->shader_priv;
5705         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5706         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5707     }
5708
5709     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5710     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5711     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5712 }
5713
5714 static void state_arb_specularenable(struct wined3d_context *context,
5715         const struct wined3d_state *state, DWORD state_id)
5716 {
5717     struct wined3d_device *device = context->swapchain->device;
5718     const struct wined3d_gl_info *gl_info = context->gl_info;
5719     float col[4];
5720
5721     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5722      * application provided constants
5723      */
5724     if (device->shader_backend == &arb_program_shader_backend)
5725     {
5726         struct shader_arb_priv *priv;
5727
5728         if (use_ps(state)) return;
5729
5730         priv = device->shader_priv;
5731         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5732         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5733     }
5734
5735     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5736     {
5737         /* The specular color has no alpha */
5738         col[0] = 1.0f; col[1] = 1.0f;
5739         col[2] = 1.0f; col[3] = 0.0f;
5740     } else {
5741         col[0] = 0.0f; col[1] = 0.0f;
5742         col[2] = 0.0f; col[3] = 0.0f;
5743     }
5744     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5745     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5746 }
5747
5748 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5749 {
5750     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5751     struct wined3d_device *device = context->swapchain->device;
5752     const struct wined3d_gl_info *gl_info = context->gl_info;
5753     float mat[2][2];
5754
5755     if (use_ps(state))
5756     {
5757         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5758         {
5759             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5760              * anyway
5761              */
5762             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5763                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5764         }
5765
5766         if(device->shader_backend == &arb_program_shader_backend) {
5767             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5768             return;
5769         }
5770     }
5771     else if (device->shader_backend == &arb_program_shader_backend)
5772     {
5773         struct shader_arb_priv *priv = device->shader_priv;
5774         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5775         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5776     }
5777
5778     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5779     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5780     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5781     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5782
5783     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5784     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5785 }
5786
5787 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5788         const struct wined3d_state *state, DWORD state_id)
5789 {
5790     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5791     struct wined3d_device *device = context->swapchain->device;
5792     const struct wined3d_gl_info *gl_info = context->gl_info;
5793     float param[4];
5794
5795     if (use_ps(state))
5796     {
5797         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5798         {
5799             /* The pixel shader has to know the luminance offset. Do a constants update if it
5800              * isn't scheduled anyway
5801              */
5802             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5803                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5804         }
5805
5806         if(device->shader_backend == &arb_program_shader_backend) {
5807             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5808             return;
5809         }
5810     }
5811     else if (device->shader_backend == &arb_program_shader_backend)
5812     {
5813         struct shader_arb_priv *priv = device->shader_priv;
5814         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5815         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5816     }
5817
5818     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5819     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5820     param[2] = 0.0f;
5821     param[3] = 0.0f;
5822
5823     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5824     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5825 }
5826
5827 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5828 {
5829     const char *ret;
5830
5831     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5832
5833     switch(arg & WINED3DTA_SELECTMASK) {
5834         case WINED3DTA_DIFFUSE:
5835             ret = "fragment.color.primary"; break;
5836
5837         case WINED3DTA_CURRENT:
5838             if (!stage) ret = "fragment.color.primary";
5839             else ret = "ret";
5840             break;
5841
5842         case WINED3DTA_TEXTURE:
5843             switch(stage) {
5844                 case 0: ret = "tex0"; break;
5845                 case 1: ret = "tex1"; break;
5846                 case 2: ret = "tex2"; break;
5847                 case 3: ret = "tex3"; break;
5848                 case 4: ret = "tex4"; break;
5849                 case 5: ret = "tex5"; break;
5850                 case 6: ret = "tex6"; break;
5851                 case 7: ret = "tex7"; break;
5852                 default: ret = "unknown texture";
5853             }
5854             break;
5855
5856         case WINED3DTA_TFACTOR:
5857             ret = "tfactor"; break;
5858
5859         case WINED3DTA_SPECULAR:
5860             ret = "fragment.color.secondary"; break;
5861
5862         case WINED3DTA_TEMP:
5863             ret = "tempreg"; break;
5864
5865         case WINED3DTA_CONSTANT:
5866             FIXME("Implement perstage constants\n");
5867             switch(stage) {
5868                 case 0: ret = "const0"; break;
5869                 case 1: ret = "const1"; break;
5870                 case 2: ret = "const2"; break;
5871                 case 3: ret = "const3"; break;
5872                 case 4: ret = "const4"; break;
5873                 case 5: ret = "const5"; break;
5874                 case 6: ret = "const6"; break;
5875                 case 7: ret = "const7"; break;
5876                 default: ret = "unknown constant";
5877             }
5878             break;
5879
5880         default:
5881             return "unknown";
5882     }
5883
5884     if(arg & WINED3DTA_COMPLEMENT) {
5885         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5886         if(argnum == 0) ret = "arg0";
5887         if(argnum == 1) ret = "arg1";
5888         if(argnum == 2) ret = "arg2";
5889     }
5890     if(arg & WINED3DTA_ALPHAREPLICATE) {
5891         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5892         if(argnum == 0) ret = "arg0";
5893         if(argnum == 1) ret = "arg1";
5894         if(argnum == 2) ret = "arg2";
5895     }
5896     return ret;
5897 }
5898
5899 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5900         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5901 {
5902     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5903     unsigned int mul = 1;
5904     BOOL mul_final_dest = FALSE;
5905
5906     if(color && alpha) dstmask = "";
5907     else if(color) dstmask = ".xyz";
5908     else dstmask = ".w";
5909
5910     if(dst == tempreg) dstreg = "tempreg";
5911     else dstreg = "ret";
5912
5913     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5914     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5915     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5916
5917     switch (op)
5918     {
5919         case WINED3D_TOP_DISABLE:
5920             if (!stage)
5921                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5922             break;
5923
5924         case WINED3D_TOP_SELECT_ARG2:
5925             arg1 = arg2;
5926             /* FALLTHROUGH */
5927         case WINED3D_TOP_SELECT_ARG1:
5928             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5929             break;
5930
5931         case WINED3D_TOP_MODULATE_4X:
5932             mul = 2;
5933             /* FALLTHROUGH */
5934         case WINED3D_TOP_MODULATE_2X:
5935             mul *= 2;
5936             if (!strcmp(dstreg, "result.color"))
5937             {
5938                 dstreg = "ret";
5939                 mul_final_dest = TRUE;
5940             }
5941             /* FALLTHROUGH */
5942         case WINED3D_TOP_MODULATE:
5943             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5944             break;
5945
5946         case WINED3D_TOP_ADD_SIGNED_2X:
5947             mul = 2;
5948             if (!strcmp(dstreg, "result.color"))
5949             {
5950                 dstreg = "ret";
5951                 mul_final_dest = TRUE;
5952             }
5953             /* FALLTHROUGH */
5954         case WINED3D_TOP_ADD_SIGNED:
5955             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5956             arg2 = "arg2";
5957             /* FALLTHROUGH */
5958         case WINED3D_TOP_ADD:
5959             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5960             break;
5961
5962         case WINED3D_TOP_SUBTRACT:
5963             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5964             break;
5965
5966         case WINED3D_TOP_ADD_SMOOTH:
5967             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5968             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5969             break;
5970
5971         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
5972             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5973             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5974             break;
5975         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
5976             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5977             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5978             break;
5979         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
5980             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5981             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5982             break;
5983         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
5984             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5985             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5986             break;
5987
5988         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
5989             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5990             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5991             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5992             break;
5993
5994         /* D3DTOP_PREMODULATE ???? */
5995
5996         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
5997             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5998             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5999             break;
6000         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6001             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6002             break;
6003         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6004             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6005             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6006             break;
6007         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6008             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6009             break;
6010
6011         case WINED3D_TOP_DOTPRODUCT3:
6012             mul = 4;
6013             if (!strcmp(dstreg, "result.color"))
6014             {
6015                 dstreg = "ret";
6016                 mul_final_dest = TRUE;
6017             }
6018             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6019             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6020             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6021             break;
6022
6023         case WINED3D_TOP_MULTIPLY_ADD:
6024             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6025             break;
6026
6027         case WINED3D_TOP_LERP:
6028             /* The msdn is not quite right here */
6029             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6030             break;
6031
6032         case WINED3D_TOP_BUMPENVMAP:
6033         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6034             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6035             break;
6036
6037         default:
6038             FIXME("Unhandled texture op %08x\n", op);
6039     }
6040
6041     if(mul == 2) {
6042         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6043     } else if(mul == 4) {
6044         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6045     }
6046 }
6047
6048 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6049 {
6050     unsigned int stage;
6051     struct wined3d_shader_buffer buffer;
6052     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6053     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6054     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6055     UINT lowest_disabled_stage;
6056     const char *textype;
6057     const char *instr, *sat;
6058     char colorcor_dst[8];
6059     GLuint ret;
6060     DWORD arg0, arg1, arg2;
6061     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6062     BOOL op_equal;
6063     const char *final_combiner_src = "ret";
6064     GLint pos;
6065
6066     /* Find out which textures are read */
6067     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6068     {
6069         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6070             break;
6071         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6072         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6073         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6074         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6075         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6076         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6077
6078         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6079             tex_read[stage] = TRUE;
6080         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6081             tex_read[stage] = TRUE;
6082         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6083         {
6084             bump_used[stage] = TRUE;
6085             tex_read[stage] = TRUE;
6086         }
6087         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6088         {
6089             bump_used[stage] = TRUE;
6090             tex_read[stage] = TRUE;
6091             luminance_used[stage] = TRUE;
6092         }
6093         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6094         {
6095             tfactor_used = TRUE;
6096         }
6097
6098         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6099             tfactor_used = TRUE;
6100         }
6101
6102         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6103         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6104             tempreg_used = TRUE;
6105         }
6106
6107         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6108             continue;
6109         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6110         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6111         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6112         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6113         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6114         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6115
6116         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6117             tempreg_used = TRUE;
6118         }
6119         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6120             tfactor_used = TRUE;
6121         }
6122     }
6123     lowest_disabled_stage = stage;
6124
6125     /* Shader header */
6126     if (!shader_buffer_init(&buffer))
6127     {
6128         ERR("Failed to initialize shader buffer.\n");
6129         return 0;
6130     }
6131
6132     shader_addline(&buffer, "!!ARBfp1.0\n");
6133
6134     switch(settings->fog) {
6135         case FOG_OFF:                                                         break;
6136         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6137         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6138         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6139         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6140     }
6141
6142     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6143     shader_addline(&buffer, "TEMP TMP;\n");
6144     shader_addline(&buffer, "TEMP ret;\n");
6145     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6146     shader_addline(&buffer, "TEMP arg0;\n");
6147     shader_addline(&buffer, "TEMP arg1;\n");
6148     shader_addline(&buffer, "TEMP arg2;\n");
6149     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6150         if(!tex_read[stage]) continue;
6151         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6152         if(!bump_used[stage]) continue;
6153         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6154         if(!luminance_used[stage]) continue;
6155         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6156     }
6157     if(tfactor_used) {
6158         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6159     }
6160         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6161
6162     if(settings->sRGB_write) {
6163         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6164                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6165         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6166                        srgb_sub_high, 0.0, 0.0, 0.0);
6167     }
6168
6169     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6170         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6171
6172     /* Generate texture sampling instructions) */
6173     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6174     {
6175         if (!tex_read[stage])
6176             continue;
6177
6178         switch(settings->op[stage].tex_type) {
6179             case tex_1d:                    textype = "1D";     break;
6180             case tex_2d:                    textype = "2D";     break;
6181             case tex_3d:                    textype = "3D";     break;
6182             case tex_cube:                  textype = "CUBE";   break;
6183             case tex_rect:                  textype = "RECT";   break;
6184             default: textype = "unexpected_textype";   break;
6185         }
6186
6187         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6188                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6189             sat = "";
6190         else
6191             sat = "_SAT";
6192
6193         if(settings->op[stage].projected == proj_none) {
6194             instr = "TEX";
6195         } else if(settings->op[stage].projected == proj_count4 ||
6196                   settings->op[stage].projected == proj_count3) {
6197             instr = "TXP";
6198         } else {
6199             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6200             instr = "TXP";
6201         }
6202
6203         if (stage > 0
6204                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6205                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6206         {
6207             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6208             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6209             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6210             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6211
6212             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6213              * so multiply the displacement with the dividing parameter before passing it to TXP
6214              */
6215             if (settings->op[stage].projected != proj_none) {
6216                 if(settings->op[stage].projected == proj_count4) {
6217                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6218                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6219                 } else {
6220                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6221                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6222                 }
6223             } else {
6224                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6225             }
6226
6227             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6228                     instr, sat, stage, stage, textype);
6229             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6230             {
6231                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6232                                stage - 1, stage - 1, stage - 1);
6233                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6234             }
6235         } else if(settings->op[stage].projected == proj_count3) {
6236             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6237             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6238             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6239                             instr, sat, stage, stage, textype);
6240         } else {
6241             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6242                             instr, sat, stage, stage, stage, textype);
6243         }
6244
6245         sprintf(colorcor_dst, "tex%u", stage);
6246         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6247                 settings->op[stage].color_fixup);
6248     }
6249
6250     /* Generate the main shader */
6251     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6252     {
6253         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6254         {
6255             if (!stage)
6256                 final_combiner_src = "fragment.color.primary";
6257             break;
6258         }
6259
6260         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6261                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6262             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6263         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6264                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6265             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6266         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6267                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6268             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6269         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6270                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6271             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6272         else
6273             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6274                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6275                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6276                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6277
6278         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6279         {
6280             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6281                           settings->op[stage].cop, settings->op[stage].carg0,
6282                           settings->op[stage].carg1, settings->op[stage].carg2);
6283             if (!stage)
6284                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6285         }
6286         else if (op_equal)
6287         {
6288             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6289                           settings->op[stage].cop, settings->op[stage].carg0,
6290                           settings->op[stage].carg1, settings->op[stage].carg2);
6291         } else {
6292             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6293                           settings->op[stage].cop, settings->op[stage].carg0,
6294                           settings->op[stage].carg1, settings->op[stage].carg2);
6295             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6296                           settings->op[stage].aop, settings->op[stage].aarg0,
6297                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6298         }
6299     }
6300
6301     if(settings->sRGB_write) {
6302         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6303         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6304         shader_addline(&buffer, "MOV result.color, ret;\n");
6305     } else {
6306         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6307     }
6308
6309     /* Footer */
6310     shader_addline(&buffer, "END\n");
6311
6312     /* Generate the shader */
6313     GL_EXTCALL(glGenProgramsARB(1, &ret));
6314     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6315     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6316             strlen(buffer.buffer), buffer.buffer));
6317     checkGLcall("glProgramStringARB()");
6318
6319     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6320     if (pos != -1)
6321     {
6322         FIXME("Fragment program error at position %d: %s\n\n", pos,
6323               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6324         shader_arb_dump_program_source(buffer.buffer);
6325     }
6326     else
6327     {
6328         GLint native;
6329
6330         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6331         checkGLcall("glGetProgramivARB()");
6332         if (!native) WARN("Program exceeds native resource limits.\n");
6333     }
6334
6335     shader_buffer_free(&buffer);
6336     return ret;
6337 }
6338
6339 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6340 {
6341     const struct wined3d_device *device = context->swapchain->device;
6342     const struct wined3d_gl_info *gl_info = context->gl_info;
6343     struct shader_arb_priv *priv = device->fragment_priv;
6344     BOOL use_vshader = use_vs(state);
6345     BOOL use_pshader = use_ps(state);
6346     struct ffp_frag_settings settings;
6347     const struct arbfp_ffp_desc *desc;
6348     unsigned int i;
6349
6350     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6351
6352     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6353     {
6354         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6355         {
6356             /* Reload fixed function constants since they collide with the
6357              * pixel shader constants. */
6358             for (i = 0; i < MAX_TEXTURES; ++i)
6359             {
6360                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6361             }
6362             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6363             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6364         }
6365         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6366         {
6367             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6368         }
6369         return;
6370     }
6371
6372     if (!use_pshader)
6373     {
6374         /* Find or create a shader implementing the fixed function pipeline
6375          * settings, then activate it. */
6376         gen_ffp_frag_op(device, state, &settings, FALSE);
6377         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6378         if(!desc) {
6379             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6380             if (!new_desc)
6381             {
6382                 ERR("Out of memory\n");
6383                 return;
6384             }
6385             new_desc->num_textures_used = 0;
6386             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6387             {
6388                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6389                     break;
6390                 new_desc->num_textures_used = i;
6391             }
6392
6393             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6394             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6395             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6396             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6397             desc = new_desc;
6398         }
6399
6400         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6401          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6402          * deactivate it.
6403          */
6404         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6405         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6406         priv->current_fprogram_id = desc->shader;
6407
6408         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6409         {
6410             /* Reload fixed function constants since they collide with the
6411              * pixel shader constants. */
6412             for (i = 0; i < MAX_TEXTURES; ++i)
6413             {
6414                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6415             }
6416             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6417             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6418         }
6419         context->last_was_pshader = FALSE;
6420     } else {
6421         context->last_was_pshader = TRUE;
6422     }
6423
6424     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6425      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function
6426      * replacement shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6427      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6428      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6429      *
6430      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6431      * shader handler.
6432      */
6433     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6434     {
6435         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6436
6437         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6438             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6439     }
6440     if (use_pshader)
6441         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6442 }
6443
6444 /* We can't link the fog states to the fragment state directly since the
6445  * vertex pipeline links them to FOGENABLE. A different linking in different
6446  * pipeline parts can't be expressed in the combined state table, so we need
6447  * to handle that with a forwarding function. The other invisible side effect
6448  * is that changing the fog start and fog end (which links to FOGENABLE in
6449  * vertex) results in the fragment_prog_arbfp function being called because
6450  * FOGENABLE is dirty, which calls this function here. */
6451 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6452 {
6453     enum fogsource new_source;
6454
6455     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6456
6457     if (!isStateDirty(context, STATE_PIXELSHADER))
6458         fragment_prog_arbfp(context, state, state_id);
6459
6460     if (!state->render_states[WINED3D_RS_FOGENABLE])
6461         return;
6462
6463     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6464     {
6465         if (use_vs(state))
6466         {
6467             new_source = FOGSOURCE_VS;
6468         }
6469         else
6470         {
6471             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6472                 new_source = FOGSOURCE_COORD;
6473             else
6474                 new_source = FOGSOURCE_FFP;
6475         }
6476     }
6477     else
6478     {
6479         new_source = FOGSOURCE_FFP;
6480     }
6481
6482     if (new_source != context->fog_source)
6483     {
6484         context->fog_source = new_source;
6485         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6486     }
6487 }
6488
6489 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6490 {
6491     if (!isStateDirty(context, STATE_PIXELSHADER))
6492         fragment_prog_arbfp(context, state, state_id);
6493 }
6494
6495 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6496 {
6497     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6498     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6499     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6502     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6503     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6505     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6619     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6620     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6623     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6624     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6626     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6635     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6636     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6637 };
6638
6639 const struct fragment_pipeline arbfp_fragment_pipeline = {
6640     arbfp_enable,
6641     arbfp_get_caps,
6642     arbfp_alloc,
6643     arbfp_free,
6644     shader_arb_color_fixup_supported,
6645     arbfp_fragmentstate_template,
6646     TRUE /* We can disable projected textures */
6647 };
6648
6649 struct arbfp_blit_priv {
6650     GLenum yuy2_rect_shader, yuy2_2d_shader;
6651     GLenum uyvy_rect_shader, uyvy_2d_shader;
6652     GLenum yv12_rect_shader, yv12_2d_shader;
6653     GLenum p8_rect_shader, p8_2d_shader;
6654     GLuint palette_texture;
6655 };
6656
6657 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6658 {
6659     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6660     if(!device->blit_priv) {
6661         ERR("Out of memory\n");
6662         return E_OUTOFMEMORY;
6663     }
6664     return WINED3D_OK;
6665 }
6666
6667 /* Context activation is done by the caller. */
6668 static void arbfp_blit_free(struct wined3d_device *device)
6669 {
6670     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6671     struct arbfp_blit_priv *priv = device->blit_priv;
6672
6673     ENTER_GL();
6674     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6675     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6676     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6677     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6678     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6679     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6680     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6681     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6682     checkGLcall("Delete yuv and p8 programs");
6683
6684     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6685     LEAVE_GL();
6686
6687     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6688     device->blit_priv = NULL;
6689 }
6690
6691 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6692         GLenum textype, char *luminance)
6693 {
6694     char chroma;
6695     const char *tex, *texinstr;
6696
6697     if (fixup == COMPLEX_FIXUP_UYVY) {
6698         chroma = 'x';
6699         *luminance = 'w';
6700     } else {
6701         chroma = 'w';
6702         *luminance = 'x';
6703     }
6704     switch(textype) {
6705         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6706         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6707         default:
6708             /* This is more tricky than just replacing the texture type - we have to navigate
6709              * properly in the texture to find the correct chroma values
6710              */
6711             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6712             return FALSE;
6713     }
6714
6715     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6716      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6717      * filtering when we sample the texture.
6718      *
6719      * These are the rules for reading the chroma:
6720      *
6721      * Even pixel: Cr
6722      * Even pixel: U
6723      * Odd pixel: V
6724      *
6725      * So we have to get the sampling x position in non-normalized coordinates in integers
6726      */
6727     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6728         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6729         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6730     } else {
6731         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6732     }
6733     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6734      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6735      * 0.5, so add 0.5.
6736      */
6737     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6738     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6739
6740     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6741      * even and odd pixels respectively
6742      */
6743     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6744     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6745
6746     /* Sample Pixel 1 */
6747     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6748
6749     /* Put the value into either of the chroma values */
6750     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6751     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6752     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6753     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6754
6755     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6756      * the pixel right to the current one. Otherwise, sample the left pixel.
6757      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6758      */
6759     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6760     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6761     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6762
6763     /* Put the value into the other chroma */
6764     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6765     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6766     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6767     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6768
6769     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6770      * the current one and lerp the two U and V values
6771      */
6772
6773     /* This gives the correctly filtered luminance value */
6774     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6775
6776     return TRUE;
6777 }
6778
6779 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6780 {
6781     const char *tex;
6782
6783     switch(textype) {
6784         case GL_TEXTURE_2D:             tex = "2D";     break;
6785         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6786         default:
6787             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6788             return FALSE;
6789     }
6790
6791     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6792      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6793      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6794      * pitch of the luminance plane, the packing into the gl texture is a bit
6795      * unfortunate. If the whole texture is interpreted as luminance data it looks
6796      * approximately like this:
6797      *
6798      *        +----------------------------------+----
6799      *        |                                  |
6800      *        |                                  |
6801      *        |                                  |
6802      *        |                                  |
6803      *        |                                  |   2
6804      *        |            LUMINANCE             |   -
6805      *        |                                  |   3
6806      *        |                                  |
6807      *        |                                  |
6808      *        |                                  |
6809      *        |                                  |
6810      *        +----------------+-----------------+----
6811      *        |                |                 |
6812      *        |  U even rows   |  U odd rows     |
6813      *        |                |                 |   1
6814      *        +----------------+------------------   -
6815      *        |                |                 |   3
6816      *        |  V even rows   |  V odd rows     |
6817      *        |                |                 |
6818      *        +----------------+-----------------+----
6819      *        |                |                 |
6820      *        |     0.5        |       0.5       |
6821      *
6822      * So it appears as if there are 4 chroma images, but in fact the odd rows
6823      * in the chroma images are in the same row as the even ones. So its is
6824      * kinda tricky to read
6825      *
6826      * When reading from rectangle textures, keep in mind that the input y coordinates
6827      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6828      */
6829     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6830             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6831
6832     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6833     /* the chroma planes have only half the width */
6834     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6835
6836     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6837      * the coordinate. Also read the right side of the image when reading odd lines
6838      *
6839      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6840      * bleeding
6841      */
6842     if(textype == GL_TEXTURE_2D) {
6843
6844         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6845
6846         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6847
6848         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6849         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6850
6851         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6852         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6853         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6854         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6855         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6856
6857         /* clamp, keep the half pixel origin in mind */
6858         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6859         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6860         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6861         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6862     } else {
6863         /* Read from [size - size+size/4] */
6864         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6865         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6866
6867         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6868         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6869         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6870         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6871         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6872         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6873
6874         /* Make sure to read exactly from the pixel center */
6875         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6876         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6877
6878         /* Clamp */
6879         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6880         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6881         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6882         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6883         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6884     }
6885     /* Read the texture, put the result into the output register */
6886     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6887     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6888
6889     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6890      * No need to clamp because we're just reusing the already clamped value from above
6891      */
6892     if(textype == GL_TEXTURE_2D) {
6893         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6894     } else {
6895         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6896     }
6897     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6898     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6899
6900     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6901      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6902      * values due to filtering
6903      */
6904     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6905     if(textype == GL_TEXTURE_2D) {
6906         /* Multiply the y coordinate by 2/3 and clamp it */
6907         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6908         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6909         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6910         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6911     } else {
6912         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6913          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6914          * is bigger
6915          */
6916         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6917         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6918         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6919     }
6920     *luminance = 'a';
6921
6922     return TRUE;
6923 }
6924
6925 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6926         const struct wined3d_gl_info *gl_info, GLenum textype)
6927 {
6928     GLenum shader;
6929     struct wined3d_shader_buffer buffer;
6930     GLint pos;
6931
6932     /* Shader header */
6933     if (!shader_buffer_init(&buffer))
6934     {
6935         ERR("Failed to initialize shader buffer.\n");
6936         return 0;
6937     }
6938
6939     ENTER_GL();
6940     GL_EXTCALL(glGenProgramsARB(1, &shader));
6941     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6942     LEAVE_GL();
6943     if(!shader) {
6944         shader_buffer_free(&buffer);
6945         return 0;
6946     }
6947
6948     shader_addline(&buffer, "!!ARBfp1.0\n");
6949     shader_addline(&buffer, "TEMP index;\n");
6950
6951     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6952     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6953
6954     /* The alpha-component contains the palette index */
6955     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6956         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6957     else
6958         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6959
6960     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6961     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6962
6963     /* Use the alpha-component as an index in the palette to get the final color */
6964     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6965     shader_addline(&buffer, "END\n");
6966
6967     ENTER_GL();
6968     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6969             strlen(buffer.buffer), buffer.buffer));
6970     checkGLcall("glProgramStringARB()");
6971
6972     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6973     if (pos != -1)
6974     {
6975         FIXME("Fragment program error at position %d: %s\n\n", pos,
6976               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6977         shader_arb_dump_program_source(buffer.buffer);
6978     }
6979
6980     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6981         priv->p8_rect_shader = shader;
6982     else
6983         priv->p8_2d_shader = shader;
6984
6985     shader_buffer_free(&buffer);
6986     LEAVE_GL();
6987
6988     return shader;
6989 }
6990
6991 /* Context activation is done by the caller. */
6992 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
6993 {
6994     BYTE table[256][4];
6995     struct wined3d_device *device = surface->resource.device;
6996     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6997     struct arbfp_blit_priv *priv = device->blit_priv;
6998     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6999
7000     d3dfmt_p8_init_palette(surface, table, colorkey);
7001
7002     ENTER_GL();
7003
7004     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7005     {
7006         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
7007         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
7008     }
7009
7010     if (!priv->palette_texture)
7011         glGenTextures(1, &priv->palette_texture);
7012
7013     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7014     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7015
7016     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7017
7018     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7019     /* Make sure we have discrete color levels. */
7020     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7021     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7022     /* Upload the palette */
7023     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7024     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7025
7026     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7027     {
7028         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7029         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7030     }
7031
7032     /* Switch back to unit 0 in which the 2D texture will be stored. */
7033     context_active_texture(context, gl_info, 0);
7034     LEAVE_GL();
7035 }
7036
7037 /* Context activation is done by the caller. */
7038 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7039         enum complex_fixup yuv_fixup, GLenum textype)
7040 {
7041     GLenum shader;
7042     struct wined3d_shader_buffer buffer;
7043     char luminance_component;
7044     GLint pos;
7045
7046     /* Shader header */
7047     if (!shader_buffer_init(&buffer))
7048     {
7049         ERR("Failed to initialize shader buffer.\n");
7050         return 0;
7051     }
7052
7053     ENTER_GL();
7054     GL_EXTCALL(glGenProgramsARB(1, &shader));
7055     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7056     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7057     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7058     LEAVE_GL();
7059     if(!shader) {
7060         shader_buffer_free(&buffer);
7061         return 0;
7062     }
7063
7064     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7065      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7066      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7067      * each single pixel it contains, and one U and one V value shared between both
7068      * pixels.
7069      *
7070      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7071      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7072      * take the format into account when generating the read swizzles
7073      *
7074      * Reading the Y value is straightforward - just sample the texture. The hardware
7075      * takes care of filtering in the horizontal and vertical direction.
7076      *
7077      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7078      * because that would mix the U and V values of one pixel or two adjacent pixels.
7079      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7080      * regardless of the filtering setting. Vertical filtering works automatically
7081      * though - the U and V values of two rows are mixed nicely.
7082      *
7083      * Apart of avoiding filtering issues, the code has to know which value it just
7084      * read, and where it can find the other one. To determine this, it checks if
7085      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7086      *
7087      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7088      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7089      *
7090      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7091      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7092      * in an unfiltered situation. Finding the luminance on the other hand requires
7093      * finding out if it is an odd or even pixel. The real drawback of this approach
7094      * is filtering. This would have to be emulated completely in the shader, reading
7095      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7096      * vertically. Beyond that it would require adjustments to the texture handling
7097      * code to deal with the width scaling
7098      */
7099     shader_addline(&buffer, "!!ARBfp1.0\n");
7100     shader_addline(&buffer, "TEMP luminance;\n");
7101     shader_addline(&buffer, "TEMP temp;\n");
7102     shader_addline(&buffer, "TEMP chroma;\n");
7103     shader_addline(&buffer, "TEMP texcrd;\n");
7104     shader_addline(&buffer, "TEMP texcrd2;\n");
7105     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7106     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7107     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7108
7109     switch (yuv_fixup)
7110     {
7111         case COMPLEX_FIXUP_UYVY:
7112         case COMPLEX_FIXUP_YUY2:
7113             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7114             {
7115                 shader_buffer_free(&buffer);
7116                 return 0;
7117             }
7118             break;
7119
7120         case COMPLEX_FIXUP_YV12:
7121             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7122             {
7123                 shader_buffer_free(&buffer);
7124                 return 0;
7125             }
7126             break;
7127
7128         default:
7129             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7130             shader_buffer_free(&buffer);
7131             return 0;
7132     }
7133
7134     /* Calculate the final result. Formula is taken from
7135      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7136      * ranges from -0.5 to 0.5
7137      */
7138     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7139
7140     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7141     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7142     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7143     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7144     shader_addline(&buffer, "END\n");
7145
7146     ENTER_GL();
7147     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7148             strlen(buffer.buffer), buffer.buffer));
7149     checkGLcall("glProgramStringARB()");
7150
7151     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7152     if (pos != -1)
7153     {
7154         FIXME("Fragment program error at position %d: %s\n\n", pos,
7155               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7156         shader_arb_dump_program_source(buffer.buffer);
7157     }
7158     else
7159     {
7160         GLint native;
7161
7162         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7163         checkGLcall("glGetProgramivARB()");
7164         if (!native) WARN("Program exceeds native resource limits.\n");
7165     }
7166
7167     shader_buffer_free(&buffer);
7168     LEAVE_GL();
7169
7170     switch (yuv_fixup)
7171     {
7172         case COMPLEX_FIXUP_YUY2:
7173             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7174             else priv->yuy2_2d_shader = shader;
7175             break;
7176
7177         case COMPLEX_FIXUP_UYVY:
7178             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7179             else priv->uyvy_2d_shader = shader;
7180             break;
7181
7182         case COMPLEX_FIXUP_YV12:
7183             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7184             else priv->yv12_2d_shader = shader;
7185             break;
7186         default:
7187             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7188     }
7189
7190     return shader;
7191 }
7192
7193 /* Context activation is done by the caller. */
7194 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7195 {
7196     GLenum shader;
7197     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7198     struct arbfp_blit_priv *priv = blit_priv;
7199     enum complex_fixup fixup;
7200     GLenum textype = surface->texture_target;
7201     const struct wined3d_gl_info *gl_info = context->gl_info;
7202
7203     if (surface->flags & SFLAG_CONVERTED)
7204     {
7205         ENTER_GL();
7206         glEnable(textype);
7207         checkGLcall("glEnable(textype)");
7208         LEAVE_GL();
7209         return WINED3D_OK;
7210     }
7211
7212     if (!is_complex_fixup(surface->resource.format->color_fixup))
7213     {
7214         TRACE("Fixup:\n");
7215         dump_color_fixup_desc(surface->resource.format->color_fixup);
7216         /* Don't bother setting up a shader for unconverted formats */
7217         ENTER_GL();
7218         glEnable(textype);
7219         checkGLcall("glEnable(textype)");
7220         LEAVE_GL();
7221         return WINED3D_OK;
7222     }
7223
7224     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7225
7226     switch(fixup)
7227     {
7228         case COMPLEX_FIXUP_YUY2:
7229             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7230             break;
7231
7232         case COMPLEX_FIXUP_UYVY:
7233             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7234             break;
7235
7236         case COMPLEX_FIXUP_YV12:
7237             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7238             break;
7239
7240         case COMPLEX_FIXUP_P8:
7241             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7242             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7243
7244             upload_palette(surface, context);
7245             break;
7246
7247         default:
7248             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7249             ENTER_GL();
7250             glEnable(textype);
7251             checkGLcall("glEnable(textype)");
7252             LEAVE_GL();
7253             return E_NOTIMPL;
7254     }
7255
7256     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7257
7258     ENTER_GL();
7259     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7260     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7261     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7262     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7263     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7264     checkGLcall("glProgramLocalParameter4fvARB");
7265     LEAVE_GL();
7266
7267     return WINED3D_OK;
7268 }
7269
7270 /* Context activation is done by the caller. */
7271 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7272 {
7273     ENTER_GL();
7274     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7275     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7276     glDisable(GL_TEXTURE_2D);
7277     checkGLcall("glDisable(GL_TEXTURE_2D)");
7278     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7279     {
7280         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7281         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7282     }
7283     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7284     {
7285         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7286         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7287     }
7288     LEAVE_GL();
7289 }
7290
7291 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7292         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7293         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7294 {
7295     enum complex_fixup src_fixup;
7296
7297     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7298         return FALSE;
7299
7300     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7301     {
7302         TRACE("Unsupported blit_op=%d\n", blit_op);
7303         return FALSE;
7304     }
7305
7306     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7307         return FALSE;
7308
7309     src_fixup = get_complex_fixup(src_format->color_fixup);
7310     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7311     {
7312         TRACE("Checking support for fixup:\n");
7313         dump_color_fixup_desc(src_format->color_fixup);
7314     }
7315
7316     if (!is_identity_fixup(dst_format->color_fixup))
7317     {
7318         TRACE("Destination fixups are not supported\n");
7319         return FALSE;
7320     }
7321
7322     if (is_identity_fixup(src_format->color_fixup))
7323     {
7324         TRACE("[OK]\n");
7325         return TRUE;
7326     }
7327
7328      /* We only support YUV conversions. */
7329     if (!is_complex_fixup(src_format->color_fixup))
7330     {
7331         TRACE("[FAILED]\n");
7332         return FALSE;
7333     }
7334
7335     switch(src_fixup)
7336     {
7337         case COMPLEX_FIXUP_YUY2:
7338         case COMPLEX_FIXUP_UYVY:
7339         case COMPLEX_FIXUP_YV12:
7340         case COMPLEX_FIXUP_P8:
7341             TRACE("[OK]\n");
7342             return TRUE;
7343
7344         default:
7345             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7346             TRACE("[FAILED]\n");
7347             return FALSE;
7348     }
7349 }
7350
7351 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7352         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7353         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7354 {
7355     struct wined3d_context *context;
7356     RECT src_rect = *src_rect_in;
7357     RECT dst_rect = *dst_rect_in;
7358
7359     /* Now load the surface */
7360     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7361             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7362     {
7363         /* Without FBO blits transferring from the drawable to the texture is
7364          * expensive, because we have to flip the data in sysmem. Since we can
7365          * flip in the blitter, we don't actually need that flip anyway. So we
7366          * use the surface's texture as scratch texture, and flip the source
7367          * rectangle instead. */
7368         surface_load_fb_texture(src_surface, FALSE);
7369
7370         src_rect.top = src_surface->resource.height - src_rect.top;
7371         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7372     }
7373     else
7374         surface_internal_preload(src_surface, SRGB_RGB);
7375
7376     /* Activate the destination context, set it up for blitting */
7377     context = context_acquire(device, dst_surface);
7378     context_apply_blit_state(context, device);
7379
7380     if (!surface_is_offscreen(dst_surface))
7381         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7382
7383     arbfp_blit_set(device->blit_priv, context, src_surface);
7384
7385     ENTER_GL();
7386
7387     /* Draw a textured quad */
7388     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7389
7390     LEAVE_GL();
7391
7392     /* Leave the opengl state valid for blitting */
7393     arbfp_blit_unset(context->gl_info);
7394
7395     if (wined3d_settings.strict_draw_ordering
7396             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7397             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7398         wglFlush(); /* Flush to ensure ordering across contexts. */
7399
7400     context_release(context);
7401
7402     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7403     return WINED3D_OK;
7404 }
7405
7406 /* Do not call while under the GL lock. */
7407 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7408         const RECT *dst_rect, const struct wined3d_color *color)
7409 {
7410     FIXME("Color filling not implemented by arbfp_blit\n");
7411     return WINED3DERR_INVALIDCALL;
7412 }
7413
7414 /* Do not call while under the GL lock. */
7415 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7416         struct wined3d_surface *surface, const RECT *rect, float depth)
7417 {
7418     FIXME("Depth filling not implemented by arbfp_blit.\n");
7419     return WINED3DERR_INVALIDCALL;
7420 }
7421
7422 const struct blit_shader arbfp_blit = {
7423     arbfp_blit_alloc,
7424     arbfp_blit_free,
7425     arbfp_blit_set,
7426     arbfp_blit_unset,
7427     arbfp_blit_supported,
7428     arbfp_blit_color_fill,
7429     arbfp_blit_depth_fill,
7430 };