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

   1 /*
   2  * Pixel and vertex shaders implementation using ARB_vertex_program
   3  * and ARB_fragment_program GL extensions.
   4  *
   5  * Copyright 2002-2003 Jason Edmeades
   6  * Copyright 2002-2003 Raphael Junqueira
   7  * Copyright 2004 Christian Costa
   8  * Copyright 2005 Oliver Stieber
   9  * Copyright 2006 Ivan Gyurdiev
  10  * Copyright 2006 Jason Green
  11  * Copyright 2006 Henri Verbeet
  12  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  13  * Copyright 2009 Henri Verbeet for CodeWeavers
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  28  */
  29
  30 #include "config.h"
  31
  32 #include <math.h>
  33 #include <stdio.h>
  34
  35 #include "wined3d_private.h"
  36
  37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
  39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
  40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
  41
  42 /* Extract a line. Note that this modifies the source string. */
  43 static char *get_line(char **ptr)
  44 {
  45     char *p, *q;
  46
  47     p = *ptr;
  48     if (!(q = strstr(p, "\n")))
  49     {
  50         if (!*p) return NULL;
  51         *ptr += strlen(p);
  52         return p;
  53     }
  54     *q = '\0';
  55     *ptr = q + 1;
  56
  57     return p;
  58 }
  59
  60 static void shader_arb_dump_program_source(const char *source)
  61 {
  62     ULONG source_size;
  63     char *ptr, *line, *tmp;
  64
  65     source_size = strlen(source) + 1;
  66     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
  67     if (!tmp)
  68     {
  69         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
  70         return;
  71     }
  72     memcpy(tmp, source, source_size);
  73
  74     ptr = tmp;
  75     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
  76     FIXME("\n");
  77
  78     HeapFree(GetProcessHeap(), 0, tmp);
  79 }
  80
  81 enum arb_helper_value
  82 {
  83     ARB_ZERO,
  84     ARB_ONE,
  85     ARB_TWO,
  86     ARB_0001,
  87     ARB_EPS,
  88
  89     ARB_VS_REL_OFFSET
  90 };
  91
  92 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
  93 {
  94     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
  95     {
  96         ERR("Geometry shaders are unsupported\n");
  97         return "bad";
  98     }
  99
 100     if (shader == WINED3D_SHADER_TYPE_PIXEL)
 101     {
 102         switch (value)
 103         {
 104             case ARB_ZERO: return "ps_helper_const.x";
 105             case ARB_ONE: return "ps_helper_const.y";
 106             case ARB_TWO: return "coefmul.x";
 107             case ARB_0001: return "ps_helper_const.xxxy";
 108             case ARB_EPS: return "ps_helper_const.z";
 109             default: break;
 110         }
 111     }
 112     else
 113     {
 114         switch (value)
 115         {
 116             case ARB_ZERO: return "helper_const.x";
 117             case ARB_ONE: return "helper_const.y";
 118             case ARB_TWO: return "helper_const.z";
 119             case ARB_EPS: return "helper_const.w";
 120             case ARB_0001: return "helper_const.xxxy";
 121             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
 122         }
 123     }
 124     FIXME("Unmanaged %s shader helper constant requested: %u\n",
 125           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
 126     switch (value)
 127     {
 128         case ARB_ZERO: return "0.0";
 129         case ARB_ONE: return "1.0";
 130         case ARB_TWO: return "2.0";
 131         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
 132         case ARB_EPS: return "1e-8";
 133         default: return "bad";
 134     }
 135 }
 136
 137 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
 138 {
 139     return state->lowest_disabled_stage < 7;
 140 }
 141
 142 /* ARB_program_shader private data */
 143
 144 struct control_frame
 145 {
 146     struct                          list entry;
 147     enum
 148     {
 149         IF,
 150         IFC,
 151         LOOP,
 152         REP
 153     } type;
 154     BOOL                            muting;
 155     BOOL                            outer_loop;
 156     union
 157     {
 158         unsigned int                loop;
 159         unsigned int                ifc;
 160     } no;
 161     struct wined3d_shader_loop_control loop_control;
 162     BOOL                            had_else;
 163 };
 164
 165 struct arb_ps_np2fixup_info
 166 {
 167     struct ps_np2fixup_info         super;
 168     /* For ARB we need a offset value:
 169      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
 170      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
 171      * array we need an offset to the index inside the program local parameter array. */
 172     UINT                            offset;
 173 };
 174
 175 struct arb_ps_compile_args
 176 {
 177     struct ps_compile_args          super;
 178     WORD                            bools;
 179     WORD                            clip;  /* only a boolean, use a WORD for alignment */
 180     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 181 };
 182
 183 struct stb_const_desc
 184 {
 185     unsigned char           texunit;
 186     UINT                    const_num;
 187 };
 188
 189 struct arb_ps_compiled_shader
 190 {
 191     struct arb_ps_compile_args      args;
 192     struct arb_ps_np2fixup_info     np2fixup_info;
 193     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
 194     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
 195     UINT                            int_consts[MAX_CONST_I];
 196     GLuint                          prgId;
 197     UINT                            ycorrection;
 198     unsigned char                   numbumpenvmatconsts;
 199     char                            num_int_consts;
 200 };
 201
 202 struct arb_vs_compile_args
 203 {
 204     struct vs_compile_args          super;
 205     union
 206     {
 207         struct
 208         {
 209             WORD                    bools;
 210             unsigned char           clip_texcoord;
 211             unsigned char           clipplane_mask;
 212         }                           boolclip;
 213         DWORD                       boolclip_compare;
 214     } clip;
 215     DWORD                           ps_signature;
 216     union
 217     {
 218         unsigned char               samplers[4];
 219         DWORD                       samplers_compare;
 220     } vertex;
 221     unsigned char                   loop_ctrl[MAX_CONST_I][3];
 222 };
 223
 224 struct arb_vs_compiled_shader
 225 {
 226     struct arb_vs_compile_args      args;
 227     GLuint                          prgId;
 228     UINT                            int_consts[MAX_CONST_I];
 229     char                            num_int_consts;
 230     char                            need_color_unclamp;
 231     UINT                            pos_fixup;
 232 };
 233
 234 struct recorded_instruction
 235 {
 236     struct wined3d_shader_instruction ins;
 237     struct list entry;
 238 };
 239
 240 struct shader_arb_ctx_priv
 241 {
 242     char addr_reg[20];
 243     enum
 244     {
 245         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
 246         ARB,
 247         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
 248         NV2,
 249         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
 250         NV3
 251     } target_version;
 252
 253     const struct arb_vs_compile_args    *cur_vs_args;
 254     const struct arb_ps_compile_args    *cur_ps_args;
 255     const struct arb_ps_compiled_shader *compiled_fprog;
 256     const struct arb_vs_compiled_shader *compiled_vprog;
 257     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
 258     struct list                         control_frames;
 259     struct list                         record;
 260     BOOL                                recording;
 261     BOOL                                muted;
 262     unsigned int                        num_loops, loop_depth, num_ifcs;
 263     int                                 aL;
 264
 265     unsigned int                        vs_clipplanes;
 266     BOOL                                footer_written;
 267     BOOL                                in_main_func;
 268
 269     /* For 3.0 vertex shaders */
 270     const char                          *vs_output[MAX_REG_OUTPUT];
 271     /* For 2.x and earlier vertex shaders */
 272     const char                          *texcrd_output[8], *color_output[2], *fog_output;
 273
 274     /* 3.0 pshader input for compatibility with fixed function */
 275     const char                          *ps_input[MAX_REG_INPUT];
 276 };
 277
 278 struct ps_signature
 279 {
 280     struct wined3d_shader_signature_element *sig;
 281     DWORD                               idx;
 282     struct wine_rb_entry                entry;
 283 };
 284
 285 struct arb_pshader_private {
 286     struct arb_ps_compiled_shader   *gl_shaders;
 287     UINT                            num_gl_shaders, shader_array_size;
 288     DWORD                           input_signature_idx;
 289     DWORD                           clipplane_emulation;
 290     BOOL                            clamp_consts;
 291 };
 292
 293 struct arb_vshader_private {
 294     struct arb_vs_compiled_shader   *gl_shaders;
 295     UINT                            num_gl_shaders, shader_array_size;
 296     UINT rel_offset;
 297 };
 298
 299 struct shader_arb_priv
 300 {
 301     GLuint                  current_vprogram_id;
 302     GLuint                  current_fprogram_id;
 303     const struct arb_ps_compiled_shader *compiled_fprog;
 304     const struct arb_vs_compiled_shader *compiled_vprog;
 305     GLuint                  depth_blt_vprogram_id;
 306     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
 307     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
 308     BOOL                    use_arbfp_fixed_func;
 309     struct wine_rb_tree     fragment_shaders;
 310     BOOL                    last_ps_const_clamped;
 311     BOOL                    last_vs_color_unclamp;
 312
 313     struct wine_rb_tree     signature_tree;
 314     DWORD ps_sig_number;
 315
 316     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
 317     char *vshader_const_dirty, *pshader_const_dirty;
 318     const struct wined3d_context *last_context;
 319 };
 320
 321 /* GL locking for state handlers is done by the caller. */
 322 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
 323         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 324 {
 325     if (shader_data->rel_offset) return TRUE;
 326     if (!reg_maps->usesmova) return FALSE;
 327     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
 328 }
 329
 330 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
 331 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
 332 {
 333     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
 334             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
 335 }
 336
 337 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
 338         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 339 {
 340     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
 341     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
 342     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
 343     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
 344     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
 345     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
 346     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
 347     return FALSE;
 348 }
 349
 350 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 351         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 352 {
 353     unsigned int ret = 1;
 354     /* We use one PARAM for the pos fixup, and in some cases one to load
 355      * some immediate values into the shader. */
 356     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 357     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 358     return ret;
 359 }
 360
 361 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 362  * When constant_list == NULL, it will load all the constants.
 363  *
 364  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 365  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 366  */
 367 /* GL locking is done by the caller */
 368 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 369         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 370         const float *constants, char *dirty_consts)
 371 {
 372     struct wined3d_shader_lconst *lconst;
 373     DWORD i, j;
 374     unsigned int ret;
 375
 376     if (TRACE_ON(d3d_constants))
 377     {
 378         for(i = 0; i < max_constants; i++) {
 379             if(!dirty_consts[i]) continue;
 380             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 381                         constants[i * 4 + 0], constants[i * 4 + 1],
 382                         constants[i * 4 + 2], constants[i * 4 + 3]);
 383         }
 384     }
 385
 386     i = 0;
 387
 388     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 389     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 390     {
 391         float lcl_const[4];
 392         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 393          * shaders, the first 8 constants are marked dirty for reload
 394          */
 395         for(; i < min(8, max_constants); i++) {
 396             if(!dirty_consts[i]) continue;
 397             dirty_consts[i] = 0;
 398
 399             j = 4 * i;
 400             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 401             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 402             else lcl_const[0] = constants[j + 0];
 403
 404             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 405             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 406             else lcl_const[1] = constants[j + 1];
 407
 408             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 409             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 410             else lcl_const[2] = constants[j + 2];
 411
 412             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 413             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 414             else lcl_const[3] = constants[j + 3];
 415
 416             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 417         }
 418
 419         /* If further constants are dirty, reload them without clamping.
 420          *
 421          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 422          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 423          * above would always re-check the first 8 constants since max_constant remains at the init
 424          * value
 425          */
 426     }
 427
 428     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 429     {
 430         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 431          * or just reloading *all* constants at once
 432          *
 433         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 434          */
 435         for(; i < max_constants; i++) {
 436             if(!dirty_consts[i]) continue;
 437
 438             /* Find the next block of dirty constants */
 439             dirty_consts[i] = 0;
 440             j = i;
 441             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 442                 dirty_consts[i] = 0;
 443             }
 444
 445             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 446         }
 447     } else {
 448         for(; i < max_constants; i++) {
 449             if(dirty_consts[i]) {
 450                 dirty_consts[i] = 0;
 451                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 452             }
 453         }
 454     }
 455     checkGLcall("glProgramEnvParameter4fvARB()");
 456
 457     /* Load immediate constants */
 458     if (shader->load_local_constsF)
 459     {
 460         if (TRACE_ON(d3d_shader))
 461         {
 462             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 463             {
 464                 GLfloat* values = (GLfloat*)lconst->value;
 465                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 466                         values[0], values[1], values[2], values[3]);
 467             }
 468         }
 469         /* Immediate constants are clamped for 1.X shaders at loading times */
 470         ret = 0;
 471         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 472         {
 473             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 474             ret = max(ret, lconst->idx + 1);
 475             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 476         }
 477         checkGLcall("glProgramEnvParameter4fvARB()");
 478         return ret; /* The loaded immediate constants need reloading for the next shader */
 479     } else {
 480         return 0; /* No constants are dirty now */
 481     }
 482 }
 483
 484 /**
 485  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 486  */
 487 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 488         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 489 {
 490     const struct shader_arb_priv * priv = shader_priv;
 491
 492     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 493     if (!use_ps(state)) return;
 494
 495     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 496         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 497         UINT i;
 498         WORD active = fixup->super.active;
 499         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 500
 501         for (i = 0; active; active >>= 1, ++i)
 502         {
 503             const struct wined3d_texture *tex = state->textures[i];
 504             const unsigned char idx = fixup->super.idx[i];
 505             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 506
 507             if (!(active & 1)) continue;
 508
 509             if (!tex) {
 510                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 511                 continue;
 512             }
 513
 514             if (idx % 2)
 515             {
 516                 tex_dim[2] = tex->pow2_matrix[0];
 517                 tex_dim[3] = tex->pow2_matrix[5];
 518             }
 519             else
 520             {
 521                 tex_dim[0] = tex->pow2_matrix[0];
 522                 tex_dim[1] = tex->pow2_matrix[5];
 523             }
 524         }
 525
 526         for (i = 0; i < fixup->super.num_consts; ++i) {
 527             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 528                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 529         }
 530     }
 531 }
 532
 533 /* GL locking is done by the caller. */
 534 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 535         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 536 {
 537     const struct wined3d_gl_info *gl_info = context->gl_info;
 538     unsigned char i;
 539
 540     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 541     {
 542         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 543
 544         /* The state manager takes care that this function is always called if the bump env matrix changes */
 545         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 546         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 547                 gl_shader->bumpenvmatconst[i].const_num, data));
 548
 549         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 550         {
 551             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 552              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 553              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 554              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 555             */
 556             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 557             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 558                     gl_shader->luminanceconst[i].const_num, scale));
 559         }
 560     }
 561     checkGLcall("Load bumpmap consts");
 562
 563     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 564     {
 565         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 566         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 567         * ycorrection.z: 1.0
 568         * ycorrection.w: 0.0
 569         */
 570         float val[4];
 571         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 572         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 573         val[2] = 1.0f;
 574         val[3] = 0.0f;
 575         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 576         checkGLcall("y correction loading");
 577     }
 578
 579     if (!gl_shader->num_int_consts) return;
 580
 581     for(i = 0; i < MAX_CONST_I; i++)
 582     {
 583         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 584         {
 585             float val[4];
 586             val[0] = (float)state->ps_consts_i[4 * i];
 587             val[1] = (float)state->ps_consts_i[4 * i + 1];
 588             val[2] = (float)state->ps_consts_i[4 * i + 2];
 589             val[3] = -1.0f;
 590
 591             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 592         }
 593     }
 594     checkGLcall("Load ps int consts");
 595 }
 596
 597 /* GL locking is done by the caller. */
 598 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 599         const struct wined3d_context *context, const struct wined3d_state *state)
 600 {
 601     const struct wined3d_gl_info *gl_info = context->gl_info;
 602     float position_fixup[4];
 603     unsigned char i;
 604
 605     /* Upload the position fixup */
 606     shader_get_position_fixup(context, state, position_fixup);
 607     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 608
 609     if (!gl_shader->num_int_consts) return;
 610
 611     for(i = 0; i < MAX_CONST_I; i++)
 612     {
 613         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 614         {
 615             float val[4];
 616             val[0] = (float)state->vs_consts_i[4 * i];
 617             val[1] = (float)state->vs_consts_i[4 * i + 1];
 618             val[2] = (float)state->vs_consts_i[4 * i + 2];
 619             val[3] = -1.0f;
 620
 621             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 622         }
 623     }
 624     checkGLcall("Load vs int consts");
 625 }
 626
 627 /**
 628  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 629  *
 630  * We only support float constants in ARB at the moment, so don't
 631  * worry about the Integers or Booleans
 632  */
 633 /* GL locking is done by the caller (state handler) */
 634 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 635 {
 636     struct wined3d_device *device = context->swapchain->device;
 637     const struct wined3d_state *state = &device->stateBlock->state;
 638     const struct wined3d_gl_info *gl_info = context->gl_info;
 639     struct shader_arb_priv *priv = device->shader_priv;
 640
 641     if (context != priv->last_context)
 642     {
 643         memset(priv->vshader_const_dirty, 1,
 644                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 645         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 646
 647         memset(priv->pshader_const_dirty, 1,
 648                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 649         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 650
 651         priv->last_context = context;
 652     }
 653
 654     if (useVertexShader)
 655     {
 656         struct wined3d_shader *vshader = state->vertex_shader;
 657         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 658
 659         /* Load DirectX 9 float constants for vertex shader */
 660         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 661                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 662         shader_arb_vs_local_constants(gl_shader, context, state);
 663     }
 664
 665     if (usePixelShader)
 666     {
 667         struct wined3d_shader *pshader = state->pixel_shader;
 668         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 669         UINT rt_height = state->fb->render_targets[0]->resource.height;
 670
 671         /* Load DirectX 9 float constants for pixel shader */
 672         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 673                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 674         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 675     }
 676 }
 677
 678 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 679 {
 680     struct wined3d_context *context = context_get_current();
 681     struct shader_arb_priv *priv = device->shader_priv;
 682
 683     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 684      * context. On a context switch the old context will be fully dirtified */
 685     if (!context || context->swapchain->device != device) return;
 686
 687     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 688     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 689 }
 690
 691 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 692 {
 693     struct wined3d_context *context = context_get_current();
 694     struct shader_arb_priv *priv = device->shader_priv;
 695
 696     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 697      * context. On a context switch the old context will be fully dirtified */
 698     if (!context || context->swapchain->device != device) return;
 699
 700     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 701     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 702 }
 703
 704 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 705 {
 706     const struct wined3d_shader_lconst *lconst;
 707     DWORD *ret;
 708     DWORD idx = 0;
 709
 710     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 711         return NULL;
 712
 713     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 714     if (!ret)
 715     {
 716         ERR("Out of memory\n");
 717         return NULL;
 718     }
 719
 720     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 721     {
 722         ret[lconst->idx] = idx++;
 723     }
 724     return ret;
 725 }
 726
 727 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 728 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 729         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 730         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 731         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 732 {
 733     DWORD i, next_local = 0;
 734     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 735     const struct wined3d_shader_lconst *lconst;
 736     unsigned max_constantsF;
 737     DWORD map;
 738
 739     /* In pixel shaders, all private constants are program local, we don't need anything
 740      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 741      * If we need a private constant the GL implementation will squeeze it in somewhere
 742      *
 743      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 744      * immediate values. The posFixup is loaded using program.env for now, so always
 745      * subtract one from the number of constants. If the shader uses indirect addressing,
 746      * account for the helper const too because we have to declare all available d3d constants
 747      * and don't know which are actually used.
 748      */
 749     if (pshader)
 750     {
 751         max_constantsF = gl_info->limits.arb_ps_native_constants;
 752         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 753         if (max_constantsF < 24)
 754             max_constantsF = gl_info->limits.arb_ps_float_constants;
 755     }
 756     else
 757     {
 758         const struct arb_vshader_private *shader_data = shader->backend_data;
 759         max_constantsF = gl_info->limits.arb_vs_native_constants;
 760         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 761          * Also prevents max_constantsF from becoming less than 0 and
 762          * wrapping . */
 763         if (max_constantsF < 96)
 764             max_constantsF = gl_info->limits.arb_vs_float_constants;
 765
 766         if (reg_maps->usesrelconstF)
 767         {
 768             DWORD highest_constf = 0, clip_limit;
 769
 770             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 771             max_constantsF -= count_bits(reg_maps->integer_constants);
 772
 773             for (i = 0; i < shader->limits.constant_float; ++i)
 774             {
 775                 DWORD idx = i >> 5;
 776                 DWORD shift = i & 0x1f;
 777                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 778             }
 779
 780             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 781             {
 782                 if(ctx->cur_vs_args->super.clip_enabled)
 783                     clip_limit = gl_info->limits.clipplanes;
 784                 else
 785                     clip_limit = 0;
 786             }
 787             else
 788             {
 789                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 790                 clip_limit = min(count_bits(mask), 4);
 791             }
 792             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 793             max_constantsF -= *num_clipplanes;
 794             if(*num_clipplanes < clip_limit)
 795             {
 796                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 797             }
 798         }
 799         else
 800         {
 801             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 802             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 803         }
 804     }
 805
 806     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 807     {
 808         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 809     }
 810
 811     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 812     {
 813         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 814     }
 815
 816     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 817     {
 818         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 819         {
 820             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 821         }
 822     }
 823
 824     /* Load local constants using the program-local space,
 825      * this avoids reloading them each time the shader is used
 826      */
 827     if (lconst_map)
 828     {
 829         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 830         {
 831             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 832                            lconst_map[lconst->idx]);
 833             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 834         }
 835     }
 836
 837     /* After subtracting privately used constants from the hardware limit(they are loaded as
 838      * local constants), make sure the shader doesn't violate the env constant limit
 839      */
 840     if(pshader)
 841     {
 842         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 843     }
 844     else
 845     {
 846         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 847     }
 848
 849     /* Avoid declaring more constants than needed */
 850     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 851
 852     /* we use the array-based constants array if the local constants are marked for loading,
 853      * because then we use indirect addressing, or when the local constant list is empty,
 854      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 855      * local constants do not declare the loaded constants as an array because ARB compilers usually
 856      * do not optimize unused constants away
 857      */
 858     if (reg_maps->usesrelconstF)
 859     {
 860         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 861         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 862                     max_constantsF, max_constantsF - 1);
 863     } else {
 864         for(i = 0; i < max_constantsF; i++) {
 865             DWORD idx, mask;
 866             idx = i >> 5;
 867             mask = 1 << (i & 0x1f);
 868             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 869             {
 870                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 871             }
 872         }
 873     }
 874
 875     return next_local;
 876 }
 877
 878 static const char * const shift_tab[] = {
 879     "dummy",     /*  0 (none) */
 880     "coefmul.x", /*  1 (x2)   */
 881     "coefmul.y", /*  2 (x4)   */
 882     "coefmul.z", /*  3 (x8)   */
 883     "coefmul.w", /*  4 (x16)  */
 884     "dummy",     /*  5 (x32)  */
 885     "dummy",     /*  6 (x64)  */
 886     "dummy",     /*  7 (x128) */
 887     "dummy",     /*  8 (d256) */
 888     "dummy",     /*  9 (d128) */
 889     "dummy",     /* 10 (d64)  */
 890     "dummy",     /* 11 (d32)  */
 891     "coefdiv.w", /* 12 (d16)  */
 892     "coefdiv.z", /* 13 (d8)   */
 893     "coefdiv.y", /* 14 (d4)   */
 894     "coefdiv.x"  /* 15 (d2)   */
 895 };
 896
 897 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 898         const struct wined3d_shader_dst_param *dst, char *write_mask)
 899 {
 900     char *ptr = write_mask;
 901
 902     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 903     {
 904         *ptr++ = '.';
 905         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 906         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 907         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 908         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 909     }
 910
 911     *ptr = '\0';
 912 }
 913
 914 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 915 {
 916     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 917      * but addressed as "rgba". To fix this we need to swap the register's x
 918      * and z components. */
 919     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 920     char *ptr = swizzle_str;
 921
 922     /* swizzle bits fields: wwzzyyxx */
 923     DWORD swizzle = param->swizzle;
 924     DWORD swizzle_x = swizzle & 0x03;
 925     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 926     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 927     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 928
 929     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 930      * generate a swizzle string. Unless we need to our own swizzling. */
 931     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 932     {
 933         *ptr++ = '.';
 934         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 935             *ptr++ = swizzle_chars[swizzle_x];
 936         } else {
 937             *ptr++ = swizzle_chars[swizzle_x];
 938             *ptr++ = swizzle_chars[swizzle_y];
 939             *ptr++ = swizzle_chars[swizzle_z];
 940             *ptr++ = swizzle_chars[swizzle_w];
 941         }
 942     }
 943
 944     *ptr = '\0';
 945 }
 946
 947 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 948 {
 949     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 950     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 951
 952     if (!strcmp(priv->addr_reg, src)) return;
 953
 954     strcpy(priv->addr_reg, src);
 955     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 956 }
 957
 958 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 959         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 960
 961 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 962         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 963 {
 964     /* oPos, oFog and oPts in D3D */
 965     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
 966     const struct wined3d_shader *shader = ins->ctx->shader;
 967     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 968     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 969     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 970
 971     *is_color = FALSE;
 972
 973     switch (reg->type)
 974     {
 975         case WINED3DSPR_TEMP:
 976             sprintf(register_name, "R%u", reg->idx);
 977             break;
 978
 979         case WINED3DSPR_INPUT:
 980             if (pshader)
 981             {
 982                 if (reg_maps->shader_version.major < 3)
 983                 {
 984                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 985                     else strcpy(register_name, "fragment.color.secondary");
 986                 }
 987                 else
 988                 {
 989                     if(reg->rel_addr)
 990                     {
 991                         char rel_reg[50];
 992                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 993
 994                         if (!strcmp(rel_reg, "**aL_emul**"))
 995                         {
 996                             DWORD idx = ctx->aL + reg->idx;
 997                             if(idx < MAX_REG_INPUT)
 998                             {
 999                                 strcpy(register_name, ctx->ps_input[idx]);
1000                             }
1001                             else
1002                             {
1003                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1004                                 sprintf(register_name, "out_of_bounds_%u", idx);
1005                             }
1006                         }
1007                         else if (reg_maps->input_registers & 0x0300)
1008                         {
1009                             /* There are two ways basically:
1010                              *
1011                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1012                              *    That means trouble if the loop also contains a breakc or if the control values
1013                              *    aren't local constants.
1014                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1015                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1016                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1017                              *    ADAC to load the condition code register and pop it again afterwards
1018                              */
1019                             FIXME("Relative input register addressing with more than 8 registers\n");
1020
1021                             /* This is better than nothing for now */
1022                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1023                         }
1024                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1025                         {
1026                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1027                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1028                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1029                              * proper varyings, or loop unrolling
1030                              *
1031                              * For now use the texcoords and hope for the best
1032                              */
1033                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1034                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1035                         }
1036                         else
1037                         {
1038                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1039                              * pulls GL_NV_fragment_program2 in
1040                              */
1041                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1042                         }
1043                     }
1044                     else
1045                     {
1046                         if(reg->idx < MAX_REG_INPUT)
1047                         {
1048                             strcpy(register_name, ctx->ps_input[reg->idx]);
1049                         }
1050                         else
1051                         {
1052                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1053                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1054                         }
1055                     }
1056                 }
1057             }
1058             else
1059             {
1060                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1061                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1062             }
1063             break;
1064
1065         case WINED3DSPR_CONST:
1066             if (!pshader && reg->rel_addr)
1067             {
1068                 const struct arb_vshader_private *shader_data = shader->backend_data;
1069                 UINT rel_offset = shader_data->rel_offset;
1070                 BOOL aL = FALSE;
1071                 char rel_reg[50];
1072                 if (reg_maps->shader_version.major < 2)
1073                 {
1074                     sprintf(rel_reg, "A0.x");
1075                 } else {
1076                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1077                     if(ctx->target_version == ARB) {
1078                         if (!strcmp(rel_reg, "**aL_emul**"))
1079                         {
1080                             aL = TRUE;
1081                         } else {
1082                             shader_arb_request_a0(ins, rel_reg);
1083                             sprintf(rel_reg, "A0.x");
1084                         }
1085                     }
1086                 }
1087                 if(aL)
1088                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1089                 else if (reg->idx >= rel_offset)
1090                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1091                 else
1092                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1093             }
1094             else
1095             {
1096                 if (reg_maps->usesrelconstF)
1097                     sprintf(register_name, "C[%u]", reg->idx);
1098                 else
1099                     sprintf(register_name, "C%u", reg->idx);
1100             }
1101             break;
1102
1103         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1104             if (pshader)
1105             {
1106                 if (reg_maps->shader_version.major == 1
1107                         && reg_maps->shader_version.minor <= 3)
1108                 {
1109                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1110                      * and as source to most instructions. For some instructions it is the texcoord
1111                      * input. Those instructions know about the special use
1112                      */
1113                     sprintf(register_name, "T%u", reg->idx);
1114                 } else {
1115                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1116                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1117                 }
1118             }
1119             else
1120             {
1121                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1122                 {
1123                     sprintf(register_name, "A%u", reg->idx);
1124                 }
1125                 else
1126                 {
1127                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1128                 }
1129             }
1130             break;
1131
1132         case WINED3DSPR_COLOROUT:
1133             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1134             {
1135                 strcpy(register_name, "TMP_COLOR");
1136             }
1137             else
1138             {
1139                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1140                 if (reg_maps->rt_mask > 1)
1141                 {
1142                     sprintf(register_name, "result.color[%u]", reg->idx);
1143                 }
1144                 else
1145                 {
1146                     strcpy(register_name, "result.color");
1147                 }
1148             }
1149             break;
1150
1151         case WINED3DSPR_RASTOUT:
1152             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1153             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1154             break;
1155
1156         case WINED3DSPR_DEPTHOUT:
1157             strcpy(register_name, "result.depth");
1158             break;
1159
1160         case WINED3DSPR_ATTROUT:
1161         /* case WINED3DSPR_OUTPUT: */
1162             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1163             else strcpy(register_name, ctx->color_output[reg->idx]);
1164             break;
1165
1166         case WINED3DSPR_TEXCRDOUT:
1167             if (pshader)
1168             {
1169                 sprintf(register_name, "oT[%u]", reg->idx);
1170             }
1171             else
1172             {
1173                 if (reg_maps->shader_version.major < 3)
1174                 {
1175                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1176                 }
1177                 else
1178                 {
1179                     strcpy(register_name, ctx->vs_output[reg->idx]);
1180                 }
1181             }
1182             break;
1183
1184         case WINED3DSPR_LOOP:
1185             if(ctx->target_version >= NV2)
1186             {
1187                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1188                 if(pshader) sprintf(register_name, "A0.x");
1189                 else sprintf(register_name, "aL.y");
1190             }
1191             else
1192             {
1193                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1194                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1195                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1196                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1197                  * indexing
1198                  */
1199                 sprintf(register_name, "**aL_emul**");
1200             }
1201
1202             break;
1203
1204         case WINED3DSPR_CONSTINT:
1205             sprintf(register_name, "I%u", reg->idx);
1206             break;
1207
1208         case WINED3DSPR_MISCTYPE:
1209             if (!reg->idx)
1210             {
1211                 sprintf(register_name, "vpos");
1212             }
1213             else if(reg->idx == 1)
1214             {
1215                 sprintf(register_name, "fragment.facing.x");
1216             }
1217             else
1218             {
1219                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1220             }
1221             break;
1222
1223         default:
1224             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1225             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1226             break;
1227     }
1228 }
1229
1230 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1231         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1232 {
1233     char register_name[255];
1234     char write_mask[6];
1235     BOOL is_color;
1236
1237     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1238     strcpy(str, register_name);
1239
1240     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1241     strcat(str, write_mask);
1242 }
1243
1244 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1245 {
1246     switch(channel_source)
1247     {
1248         case CHANNEL_SOURCE_ZERO: return "0";
1249         case CHANNEL_SOURCE_ONE: return "1";
1250         case CHANNEL_SOURCE_X: return "x";
1251         case CHANNEL_SOURCE_Y: return "y";
1252         case CHANNEL_SOURCE_Z: return "z";
1253         case CHANNEL_SOURCE_W: return "w";
1254         default:
1255             FIXME("Unhandled channel source %#x\n", channel_source);
1256             return "undefined";
1257     }
1258 }
1259
1260 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1261         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1262 {
1263     DWORD mask;
1264
1265     if (is_complex_fixup(fixup))
1266     {
1267         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1268         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1269         return;
1270     }
1271
1272     mask = 0;
1273     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1274     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1275     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1276     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1277     mask &= dst_mask;
1278
1279     if (mask)
1280     {
1281         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1282                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1283                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1284     }
1285
1286     mask = 0;
1287     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1288     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1289     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1290     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1291     mask &= dst_mask;
1292
1293     if (mask)
1294     {
1295         char reg_mask[6];
1296         char *ptr = reg_mask;
1297
1298         if (mask != WINED3DSP_WRITEMASK_ALL)
1299         {
1300             *ptr++ = '.';
1301             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1302             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1303             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1304             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1305         }
1306         *ptr = '\0';
1307
1308         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1309     }
1310 }
1311
1312 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1313 {
1314     DWORD mod;
1315     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1316     if (!ins->dst_count) return "";
1317
1318     mod = ins->dst[0].modifiers;
1319
1320     /* Silently ignore PARTIALPRECISION if its not supported */
1321     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1322
1323     if(mod & WINED3DSPDM_MSAMPCENTROID)
1324     {
1325         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1326         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1327     }
1328
1329     switch(mod)
1330     {
1331         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1332             return "H_SAT";
1333
1334         case WINED3DSPDM_SATURATE:
1335             return "_SAT";
1336
1337         case WINED3DSPDM_PARTIALPRECISION:
1338             return "H";
1339
1340         case 0:
1341             return "";
1342
1343         default:
1344             FIXME("Unknown modifiers 0x%08x\n", mod);
1345             return "";
1346     }
1347 }
1348
1349 #define TEX_PROJ        0x1
1350 #define TEX_BIAS        0x2
1351 #define TEX_LOD         0x4
1352 #define TEX_DERIV       0x10
1353
1354 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1355         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1356 {
1357     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1358     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1359     const struct wined3d_shader *shader = ins->ctx->shader;
1360     const struct wined3d_texture *texture;
1361     const char *tex_type;
1362     BOOL np2_fixup = FALSE;
1363     struct wined3d_device *device = shader->device;
1364     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1365     const char *mod;
1366     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1367
1368     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1369     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1370
1371     switch(sampler_type) {
1372         case WINED3DSTT_1D:
1373             tex_type = "1D";
1374             break;
1375
1376         case WINED3DSTT_2D:
1377             texture = device->stateBlock->state.textures[sampler_idx];
1378             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1379             {
1380                 tex_type = "RECT";
1381             } else {
1382                 tex_type = "2D";
1383             }
1384             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1385             {
1386                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1387                 {
1388                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1389                     else np2_fixup = TRUE;
1390                 }
1391             }
1392             break;
1393
1394         case WINED3DSTT_VOLUME:
1395             tex_type = "3D";
1396             break;
1397
1398         case WINED3DSTT_CUBE:
1399             tex_type = "CUBE";
1400             break;
1401
1402         default:
1403             ERR("Unexpected texture type %d\n", sampler_type);
1404             tex_type = "";
1405     }
1406
1407     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1408      * so don't use shader_arb_get_modifier
1409      */
1410     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1411     else mod = "";
1412
1413     /* Fragment samplers always have indentity mapping */
1414     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1415     {
1416         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1417     }
1418
1419     if (flags & TEX_DERIV)
1420     {
1421         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1422         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1423         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1424                        dsx, dsy,sampler_idx, tex_type);
1425     }
1426     else if(flags & TEX_LOD)
1427     {
1428         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1429         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1430         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1431                        sampler_idx, tex_type);
1432     }
1433     else if (flags & TEX_BIAS)
1434     {
1435         /* Shouldn't be possible, but let's check for it */
1436         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1437         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1438         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1439     }
1440     else if (flags & TEX_PROJ)
1441     {
1442         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1443     }
1444     else
1445     {
1446         if (np2_fixup)
1447         {
1448             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1449             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1450                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1451
1452             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1453         }
1454         else
1455             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1456     }
1457
1458     if (pshader)
1459     {
1460         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1461                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1462                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1463                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1464     }
1465 }
1466
1467 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1468         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1469 {
1470     /* Generate a line that does the input modifier computation and return the input register to use */
1471     BOOL is_color = FALSE;
1472     char regstr[256];
1473     char swzstr[20];
1474     int insert_line;
1475     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1476     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1477     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1478     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1479
1480     /* Assume a new line will be added */
1481     insert_line = 1;
1482
1483     /* Get register name */
1484     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1485     shader_arb_get_swizzle(src, is_color, swzstr);
1486
1487     switch (src->modifiers)
1488     {
1489     case WINED3DSPSM_NONE:
1490         sprintf(outregstr, "%s%s", regstr, swzstr);
1491         insert_line = 0;
1492         break;
1493     case WINED3DSPSM_NEG:
1494         sprintf(outregstr, "-%s%s", regstr, swzstr);
1495         insert_line = 0;
1496         break;
1497     case WINED3DSPSM_BIAS:
1498         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1499         break;
1500     case WINED3DSPSM_BIASNEG:
1501         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1502         break;
1503     case WINED3DSPSM_SIGN:
1504         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1505         break;
1506     case WINED3DSPSM_SIGNNEG:
1507         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1508         break;
1509     case WINED3DSPSM_COMP:
1510         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1511         break;
1512     case WINED3DSPSM_X2:
1513         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1514         break;
1515     case WINED3DSPSM_X2NEG:
1516         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1517         break;
1518     case WINED3DSPSM_DZ:
1519         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1520         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1521         break;
1522     case WINED3DSPSM_DW:
1523         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1524         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1525         break;
1526     case WINED3DSPSM_ABS:
1527         if(ctx->target_version >= NV2) {
1528             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1529             insert_line = 0;
1530         } else {
1531             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1532         }
1533         break;
1534     case WINED3DSPSM_ABSNEG:
1535         if(ctx->target_version >= NV2) {
1536             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1537         } else {
1538             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1539             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1540         }
1541         insert_line = 0;
1542         break;
1543     default:
1544         sprintf(outregstr, "%s%s", regstr, swzstr);
1545         insert_line = 0;
1546     }
1547
1548     /* Return modified or original register, with swizzle */
1549     if (insert_line)
1550         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1551 }
1552
1553 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1554 {
1555     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1556     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1557     char dst_name[50];
1558     char src_name[2][50];
1559     DWORD sampler_code = dst->reg.idx;
1560
1561     shader_arb_get_dst_param(ins, dst, dst_name);
1562
1563     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1564      *
1565      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1566      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1567      * temps is done.
1568      */
1569     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1570     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1571     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1572     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1573     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1574
1575     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1576     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1577 }
1578
1579 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1580 {
1581     *extra_char = ' ';
1582     switch(mod)
1583     {
1584         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1585         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1586         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1587         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1588         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1589         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1590         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1591         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1592         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1593         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1594         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1595         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1596         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1597     }
1598     FIXME("Unknown modifier %u\n", mod);
1599     return mod;
1600 }
1601
1602 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1603 {
1604     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1605     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1606     char dst_name[50];
1607     char src_name[3][50];
1608     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1609             ins->ctx->reg_maps->shader_version.minor);
1610
1611     shader_arb_get_dst_param(ins, dst, dst_name);
1612     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1613
1614     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1615     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1616     {
1617         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1618     }
1619     else
1620     {
1621         struct wined3d_shader_src_param src0_copy = ins->src[0];
1622         char extra_neg;
1623
1624         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1625         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1626
1627         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1628         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1629         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1630         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1631                 dst_name, src_name[1], src_name[2]);
1632     }
1633 }
1634
1635 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1636 {
1637     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1638     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1639     char dst_name[50];
1640     char src_name[3][50];
1641
1642     shader_arb_get_dst_param(ins, dst, dst_name);
1643
1644     /* Generate input register names (with modifiers) */
1645     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1646     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1647     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1648
1649     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1650             dst_name, src_name[0], src_name[2], src_name[1]);
1651 }
1652
1653 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1654  * dst = dot2(src0, src1) + src2 */
1655 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1656 {
1657     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1658     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1659     char dst_name[50];
1660     char src_name[3][50];
1661     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1662
1663     shader_arb_get_dst_param(ins, dst, dst_name);
1664     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1665     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1666
1667     if(ctx->target_version >= NV3)
1668     {
1669         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1670         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1671         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1672                        dst_name, src_name[0], src_name[1], src_name[2]);
1673     }
1674     else if(ctx->target_version >= NV2)
1675     {
1676         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1677          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1678          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1679          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1680          *
1681          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1682          *
1683          * .xyxy and other swizzles that we could get with this are not valid in
1684          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1685          */
1686         struct wined3d_shader_src_param tmp_param = ins->src[1];
1687         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1688         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1689
1690         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1691
1692         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1693                        dst_name, src_name[2], src_name[0], src_name[1]);
1694     }
1695     else
1696     {
1697         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1698         /* 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]
1699         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1700         */
1701         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1702         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1703         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1704         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1705     }
1706 }
1707
1708 /* Map the opcode 1-to-1 to the GL code */
1709 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1710 {
1711     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1712     const char *instruction;
1713     char arguments[256], dst_str[50];
1714     unsigned int i;
1715     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1716
1717     switch (ins->handler_idx)
1718     {
1719         case WINED3DSIH_ABS: instruction = "ABS"; break;
1720         case WINED3DSIH_ADD: instruction = "ADD"; break;
1721         case WINED3DSIH_CRS: instruction = "XPD"; break;
1722         case WINED3DSIH_DP3: instruction = "DP3"; break;
1723         case WINED3DSIH_DP4: instruction = "DP4"; break;
1724         case WINED3DSIH_DST: instruction = "DST"; break;
1725         case WINED3DSIH_FRC: instruction = "FRC"; break;
1726         case WINED3DSIH_LIT: instruction = "LIT"; break;
1727         case WINED3DSIH_LRP: instruction = "LRP"; break;
1728         case WINED3DSIH_MAD: instruction = "MAD"; break;
1729         case WINED3DSIH_MAX: instruction = "MAX"; break;
1730         case WINED3DSIH_MIN: instruction = "MIN"; break;
1731         case WINED3DSIH_MOV: instruction = "MOV"; break;
1732         case WINED3DSIH_MUL: instruction = "MUL"; break;
1733         case WINED3DSIH_SGE: instruction = "SGE"; break;
1734         case WINED3DSIH_SLT: instruction = "SLT"; break;
1735         case WINED3DSIH_SUB: instruction = "SUB"; break;
1736         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1737         case WINED3DSIH_DSX: instruction = "DDX"; break;
1738         default: instruction = "";
1739             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1740             break;
1741     }
1742
1743     /* Note that shader_arb_add_dst_param() adds spaces. */
1744     arguments[0] = '\0';
1745     shader_arb_get_dst_param(ins, dst, dst_str);
1746     for (i = 0; i < ins->src_count; ++i)
1747     {
1748         char operand[100];
1749         strcat(arguments, ", ");
1750         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1751         strcat(arguments, operand);
1752     }
1753     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1754 }
1755
1756 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1757 {
1758     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1759     shader_addline(buffer, "NOP;\n");
1760 }
1761
1762 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1763 {
1764     const struct wined3d_shader *shader = ins->ctx->shader;
1765     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1766     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1767     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1768     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1769     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1770     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1771
1772     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1773     char src0_param[256];
1774
1775     if (ins->handler_idx == WINED3DSIH_MOVA)
1776     {
1777         const struct arb_vshader_private *shader_data = shader->backend_data;
1778         char write_mask[6];
1779         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1780
1781         if(ctx->target_version >= NV2) {
1782             shader_hw_map2gl(ins);
1783             return;
1784         }
1785         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1786         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1787
1788         /* This implements the mova formula used in GLSL. The first two instructions
1789          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1790          * in this case:
1791          * mova A0.x, 0.0
1792          *
1793          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1794          *
1795          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1796          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1797          */
1798         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1799         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1800
1801         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1802         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1803         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1804         if (shader_data->rel_offset)
1805         {
1806             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1807         }
1808         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1809
1810         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1811     }
1812     else if (reg_maps->shader_version.major == 1
1813           && !shader_is_pshader_version(reg_maps->shader_version.type)
1814           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1815     {
1816         const struct arb_vshader_private *shader_data = shader->backend_data;
1817         src0_param[0] = '\0';
1818
1819         if (shader_data->rel_offset)
1820         {
1821             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1822             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1823             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1824             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1825         }
1826         else
1827         {
1828             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1829              * with more than one component. Thus replicate the first source argument over all
1830              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1831             struct wined3d_shader_src_param tmp_src = ins->src[0];
1832             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1833             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1834             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1835         }
1836     }
1837     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1838     {
1839         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1840         {
1841             shader_addline(buffer, "#mov handled in srgb write code\n");
1842             return;
1843         }
1844         shader_hw_map2gl(ins);
1845     }
1846     else
1847     {
1848         shader_hw_map2gl(ins);
1849     }
1850 }
1851
1852 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1853 {
1854     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1855     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1856     char reg_dest[40];
1857
1858     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1859      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1860      */
1861     shader_arb_get_dst_param(ins, dst, reg_dest);
1862
1863     if (ins->ctx->reg_maps->shader_version.major >= 2)
1864     {
1865         const char *kilsrc = "TA";
1866         BOOL is_color;
1867
1868         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1869         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1870         {
1871             kilsrc = reg_dest;
1872         }
1873         else
1874         {
1875             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1876              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1877              * masked out components to 0(won't kill)
1878              */
1879             char x = '0', y = '0', z = '0', w = '0';
1880             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1881             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1882             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1883             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1884             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1885         }
1886         shader_addline(buffer, "KIL %s;\n", kilsrc);
1887     } else {
1888         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1889          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1890          *
1891          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1892          * or pass in any temporary register(in shader phase 2)
1893          */
1894         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1895             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1896         } else {
1897             shader_arb_get_dst_param(ins, dst, reg_dest);
1898         }
1899         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1900         shader_addline(buffer, "KIL TA;\n");
1901     }
1902 }
1903
1904 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1905 {
1906     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1907     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1908     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1909             ins->ctx->reg_maps->shader_version.minor);
1910     struct wined3d_shader_src_param src;
1911
1912     char reg_dest[40];
1913     char reg_coord[40];
1914     DWORD reg_sampler_code;
1915     WORD myflags = 0;
1916     BOOL swizzle_coord = FALSE;
1917
1918     /* All versions have a destination register */
1919     shader_arb_get_dst_param(ins, dst, reg_dest);
1920
1921     /* 1.0-1.4: Use destination register number as texture code.
1922        2.0+: Use provided sampler number as texure code. */
1923     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1924         reg_sampler_code = dst->reg.idx;
1925     else
1926         reg_sampler_code = ins->src[1].reg.idx;
1927
1928     /* 1.0-1.3: Use the texcoord varying.
1929        1.4+: Use provided coordinate source register. */
1930     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1931         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1932     else {
1933         /* TEX is the only instruction that can handle DW and DZ natively */
1934         src = ins->src[0];
1935         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1936         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1937         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1938     }
1939
1940     /* projection flag:
1941      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1942      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1943      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1944      */
1945     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1946     {
1947         DWORD flags = 0;
1948         if (reg_sampler_code < MAX_TEXTURES)
1949             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1950         if (flags & WINED3D_PSARGS_PROJECTED)
1951         {
1952             myflags |= TEX_PROJ;
1953             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1954                 swizzle_coord = TRUE;
1955         }
1956     }
1957     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1958     {
1959         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1960         if (src_mod == WINED3DSPSM_DZ)
1961         {
1962             swizzle_coord = TRUE;
1963             myflags |= TEX_PROJ;
1964         } else if(src_mod == WINED3DSPSM_DW) {
1965             myflags |= TEX_PROJ;
1966         }
1967     } else {
1968         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1969         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1970     }
1971
1972     if (swizzle_coord)
1973     {
1974         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
1975          * reg_coord is a read-only varying register, so we need a temp reg */
1976         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1977         strcpy(reg_coord, "TA");
1978     }
1979
1980     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1981 }
1982
1983 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1984 {
1985     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1986     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1987     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1988             ins->ctx->reg_maps->shader_version.minor);
1989     char dst_str[50];
1990
1991     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1992     {
1993         DWORD reg = dst->reg.idx;
1994
1995         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1996         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1997     } else {
1998         char reg_src[40];
1999
2000         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2001         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2002         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2003    }
2004 }
2005
2006 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2007 {
2008      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2009      DWORD flags = 0;
2010
2011      DWORD reg1 = ins->dst[0].reg.idx;
2012      char dst_str[50];
2013      char src_str[50];
2014
2015      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2016      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2017      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2018      /* Move .x first in case src_str is "TA" */
2019      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2020      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2021      if (reg1 < MAX_TEXTURES)
2022      {
2023          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2024          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2025      }
2026      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2027 }
2028
2029 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2030 {
2031      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2032
2033      DWORD reg1 = ins->dst[0].reg.idx;
2034      char dst_str[50];
2035      char src_str[50];
2036
2037      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2038      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2039      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2040      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2041      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2042      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2043 }
2044
2045 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2046 {
2047     DWORD reg1 = ins->dst[0].reg.idx;
2048     char dst_str[50];
2049     char src_str[50];
2050
2051     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2052     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2053     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2054     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2055 }
2056
2057 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2058 {
2059     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2060     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2061     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2062     char reg_coord[40], dst_reg[50], src_reg[50];
2063     DWORD reg_dest_code;
2064
2065     /* All versions have a destination register. The Tx where the texture coordinates come
2066      * from is the varying incarnation of the texture register
2067      */
2068     reg_dest_code = dst->reg.idx;
2069     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2070     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2071     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2072
2073     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2074      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2075      *
2076      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2077      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2078      *
2079      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2080      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2081      * extension.
2082      */
2083     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2084     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2085     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2086     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2087
2088     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2089      * so we can't let the GL handle this.
2090      */
2091     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2092             & WINED3D_PSARGS_PROJECTED)
2093     {
2094         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2095         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2096         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2097     } else {
2098         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2099     }
2100
2101     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2102
2103     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2104     {
2105         /* No src swizzles are allowed, so this is ok */
2106         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2107                        src_reg, reg_dest_code, reg_dest_code);
2108         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2109     }
2110 }
2111
2112 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2113 {
2114     DWORD reg = ins->dst[0].reg.idx;
2115     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2116     char src0_name[50], dst_name[50];
2117     BOOL is_color;
2118     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2119
2120     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2121     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2122      * T<reg+1> register. Use this register to store the calculated vector
2123      */
2124     tmp_reg.idx = reg + 1;
2125     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2126     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2127 }
2128
2129 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2130 {
2131     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2132     DWORD flags;
2133     DWORD reg = ins->dst[0].reg.idx;
2134     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2135     char dst_str[50];
2136     char src0_name[50];
2137     char dst_reg[50];
2138     BOOL is_color;
2139
2140     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2141     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2142
2143     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2144     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2145     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2146     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2147     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2148 }
2149
2150 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2151 {
2152     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2153     DWORD reg = ins->dst[0].reg.idx;
2154     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2155     char src0_name[50], dst_name[50];
2156     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2157     BOOL is_color;
2158
2159     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2160      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2161      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2162      */
2163     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2164     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2165
2166     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2167     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2168                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2169     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2170 }
2171
2172 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2173 {
2174     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2175     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2176     DWORD flags;
2177     DWORD reg = ins->dst[0].reg.idx;
2178     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2179     char dst_str[50];
2180     char src0_name[50], dst_name[50];
2181     BOOL is_color;
2182
2183     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2184     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2185     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2186
2187     /* Sample the texture using the calculated coordinates */
2188     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2189     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2190     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2191     tex_mx->current_row = 0;
2192 }
2193
2194 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2195 {
2196     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2197     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2198     DWORD flags;
2199     DWORD reg = ins->dst[0].reg.idx;
2200     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2201     char dst_str[50];
2202     char src0_name[50];
2203     char dst_reg[50];
2204     BOOL is_color;
2205
2206     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2207      * components for temporary data storage
2208      */
2209     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2210     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2211     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2212
2213     /* Construct the eye-ray vector from w coordinates */
2214     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2215     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2216     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2217
2218     /* Calculate reflection vector
2219      */
2220     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2221     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2222     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2223     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2224     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2225     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2226     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2227
2228     /* Sample the texture using the calculated coordinates */
2229     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2230     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2231     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2232     tex_mx->current_row = 0;
2233 }
2234
2235 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2236 {
2237     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2238     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2239     DWORD flags;
2240     DWORD reg = ins->dst[0].reg.idx;
2241     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2242     char dst_str[50];
2243     char src0_name[50];
2244     char src1_name[50];
2245     char dst_reg[50];
2246     BOOL is_color;
2247
2248     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2249     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2250     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2251     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2252     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2253
2254     /* Calculate reflection vector.
2255      *
2256      *                   dot(N, E)
2257      * dst_reg.xyz = 2 * --------- * N - E
2258      *                   dot(N, N)
2259      *
2260      * Which normalizes the normal vector
2261      */
2262     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2263     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2264     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2265     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2266     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2267     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2268
2269     /* Sample the texture using the calculated coordinates */
2270     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2271     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2272     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2273     tex_mx->current_row = 0;
2274 }
2275
2276 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2277 {
2278     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2279     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2280     char dst_name[50];
2281     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2282     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2283
2284     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2285      * which is essentially an input, is the destination register because it is the first
2286      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2287      * here(writemasks/swizzles are not valid on texdepth)
2288      */
2289     shader_arb_get_dst_param(ins, dst, dst_name);
2290
2291     /* According to the msdn, the source register(must be r5) is unusable after
2292      * the texdepth instruction, so we're free to modify it
2293      */
2294     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2295
2296     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2297      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2298      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2299      */
2300     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2301     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2302     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2303     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2304 }
2305
2306 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2307  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2308  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2309 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2310 {
2311     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2312     DWORD sampler_idx = ins->dst[0].reg.idx;
2313     char src0[50];
2314     char dst_str[50];
2315
2316     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2317     shader_addline(buffer, "MOV TB, 0.0;\n");
2318     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2319
2320     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2321     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2322 }
2323
2324 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2325  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2326 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2327 {
2328     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2329     char src0[50];
2330     char dst_str[50];
2331     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2332
2333     /* Handle output register */
2334     shader_arb_get_dst_param(ins, dst, dst_str);
2335     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2336     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2337 }
2338
2339 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2340  * Perform the 3rd row of a 3x3 matrix multiply */
2341 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2342 {
2343     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2344     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2345     char dst_str[50], dst_name[50];
2346     char src0[50];
2347     BOOL is_color;
2348
2349     shader_arb_get_dst_param(ins, dst, dst_str);
2350     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2351     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2352     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2353     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2354 }
2355
2356 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2357  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2358  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2359  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2360  */
2361 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2362 {
2363     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2364     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2365     char src0[50], dst_name[50];
2366     BOOL is_color;
2367     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2368     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2369
2370     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2371     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2372     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2373
2374     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2375      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2376      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2377      */
2378     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2379     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2380     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2381     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2382 }
2383
2384 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2385     Vertex/Pixel shaders to ARB_vertex_program codes */
2386 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2387 {
2388     int i;
2389     int nComponents = 0;
2390     struct wined3d_shader_dst_param tmp_dst = {{0}};
2391     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2392     struct wined3d_shader_instruction tmp_ins;
2393
2394     memset(&tmp_ins, 0, sizeof(tmp_ins));
2395
2396     /* Set constants for the temporary argument */
2397     tmp_ins.ctx = ins->ctx;
2398     tmp_ins.dst_count = 1;
2399     tmp_ins.dst = &tmp_dst;
2400     tmp_ins.src_count = 2;
2401     tmp_ins.src = tmp_src;
2402
2403     switch(ins->handler_idx)
2404     {
2405         case WINED3DSIH_M4x4:
2406             nComponents = 4;
2407             tmp_ins.handler_idx = WINED3DSIH_DP4;
2408             break;
2409         case WINED3DSIH_M4x3:
2410             nComponents = 3;
2411             tmp_ins.handler_idx = WINED3DSIH_DP4;
2412             break;
2413         case WINED3DSIH_M3x4:
2414             nComponents = 4;
2415             tmp_ins.handler_idx = WINED3DSIH_DP3;
2416             break;
2417         case WINED3DSIH_M3x3:
2418             nComponents = 3;
2419             tmp_ins.handler_idx = WINED3DSIH_DP3;
2420             break;
2421         case WINED3DSIH_M3x2:
2422             nComponents = 2;
2423             tmp_ins.handler_idx = WINED3DSIH_DP3;
2424             break;
2425         default:
2426             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2427             break;
2428     }
2429
2430     tmp_dst = ins->dst[0];
2431     tmp_src[0] = ins->src[0];
2432     tmp_src[1] = ins->src[1];
2433     for (i = 0; i < nComponents; i++) {
2434         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2435         shader_hw_map2gl(&tmp_ins);
2436         ++tmp_src[1].reg.idx;
2437     }
2438 }
2439
2440 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2441 {
2442     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2443
2444     char dst[50];
2445     char src[50];
2446
2447     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2448     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2449     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2450     {
2451         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2452          * .w is used
2453          */
2454         strcat(src, ".w");
2455     }
2456
2457     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2458 }
2459
2460 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2461 {
2462     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2463     const char *instruction;
2464
2465     char dst[50];
2466     char src[50];
2467
2468     switch(ins->handler_idx)
2469     {
2470         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2471         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2472         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2473         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2474         default: instruction = "";
2475             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2476             break;
2477     }
2478
2479     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2480     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2481     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2482     {
2483         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2484          * .w is used
2485          */
2486         strcat(src, ".w");
2487     }
2488
2489     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2490 }
2491
2492 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2493 {
2494     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2495     char dst_name[50];
2496     char src_name[50];
2497     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2498     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2499     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2500
2501     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2502     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2503
2504     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2505      * otherwise NRM or RSQ would return NaN */
2506     if(pshader && priv->target_version >= NV3)
2507     {
2508         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2509          *
2510          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2511          */
2512         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2513         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2514         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2515     }
2516     else if(priv->target_version >= NV2)
2517     {
2518         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2519         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2520         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2521                        src_name);
2522     }
2523     else
2524     {
2525         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2526
2527         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2528         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2529          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2530          */
2531         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2532         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2533
2534         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2535         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2536         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2537                     src_name);
2538     }
2539 }
2540
2541 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2542 {
2543     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2544     char dst_name[50];
2545     char src_name[3][50];
2546
2547     /* ARB_fragment_program has a convenient LRP instruction */
2548     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2549         shader_hw_map2gl(ins);
2550         return;
2551     }
2552
2553     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2554     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2555     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2556     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2557
2558     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2559     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2560                    dst_name, src_name[0], src_name[2]);
2561 }
2562
2563 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2564 {
2565     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2566      * must contain fixed constants. So we need a separate function to filter those constants and
2567      * can't use map2gl
2568      */
2569     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2570     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2571     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2572     char dst_name[50];
2573     char src_name0[50], src_name1[50], src_name2[50];
2574     BOOL is_color;
2575
2576     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2577     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2578         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2579         /* No modifiers are supported on SCS */
2580         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2581
2582         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2583         {
2584             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2585             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2586         }
2587     } else if(priv->target_version >= NV2) {
2588         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2589
2590         /* Sincos writemask must be .x, .y or .xy */
2591         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2592             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2593         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2594             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2595     } else {
2596         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2597          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2598          *
2599          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2600          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2601          *
2602          * The constants we get are:
2603          *
2604          *  +1   +1,     -1     -1     +1      +1      -1       -1
2605          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2606          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2607          *
2608          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2609          *
2610          * (x/2)^2 = x^2 / 4
2611          * (x/2)^3 = x^3 / 8
2612          * (x/2)^4 = x^4 / 16
2613          * (x/2)^5 = x^5 / 32
2614          * etc
2615          *
2616          * To get the final result:
2617          * sin(x) = 2 * sin(x/2) * cos(x/2)
2618          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2619          * (from sin(x+y) and cos(x+y) rules)
2620          *
2621          * As per MSDN, dst.z is undefined after the operation, and so is
2622          * dst.x and dst.y if they're masked out by the writemask. Ie
2623          * sincos dst.y, src1, c0, c1
2624          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2625          * vsa.exe also stops with an error if the dest register is the same register as the source
2626          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2627          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2628          */
2629         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2630         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2631         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2632
2633         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2634         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2635         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2636         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2637         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2638         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2639
2640         /* sin(x/2)
2641          *
2642          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2643          * properly merge that with MULs in the code above?
2644          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2645          * we can merge the sine and cosine MAD rows to calculate them together.
2646          */
2647         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2648         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2649         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2650         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2651
2652         /* cos(x/2) */
2653         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2654         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2655         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2656
2657         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2658             /* cos x */
2659             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2660             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2661         }
2662         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2663             /* sin x */
2664             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2665             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2666         }
2667     }
2668 }
2669
2670 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2671 {
2672     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2673     char dst_name[50];
2674     char src_name[50];
2675     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2676
2677     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2678     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2679
2680     /* SGN is only valid in vertex shaders */
2681     if(ctx->target_version >= NV2) {
2682         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2683         return;
2684     }
2685
2686     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2687      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2688      */
2689     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2690         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2691     } else {
2692         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2693          * Then use TA, and calculate the final result
2694          *
2695          * Not reading from TA? Store the first result in TA to avoid overwriting the
2696          * destination if src reg = dst reg
2697          */
2698         if(strstr(src_name, "TA"))
2699         {
2700             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2701             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2702             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2703         }
2704         else
2705         {
2706             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2707             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2708             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2709         }
2710     }
2711 }
2712
2713 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2714 {
2715     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2716     char src[50];
2717     char dst[50];
2718     char dst_name[50];
2719     BOOL is_color;
2720
2721     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2722     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2723     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2724
2725     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2726     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2727 }
2728
2729 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2730 {
2731     *need_abs = FALSE;
2732
2733     switch(mod)
2734     {
2735         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2736         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2737         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2738         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2739         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2740         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2741         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2742         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2743         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2744         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2745         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2746         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2747         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2748     }
2749     FIXME("Unknown modifier %u\n", mod);
2750     return mod;
2751 }
2752
2753 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2754 {
2755     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2756     char src0[50], dst[50];
2757     struct wined3d_shader_src_param src0_copy = ins->src[0];
2758     BOOL need_abs = FALSE;
2759     const char *instr;
2760
2761     switch(ins->handler_idx)
2762     {
2763         case WINED3DSIH_LOG:  instr = "LG2"; break;
2764         case WINED3DSIH_LOGP: instr = "LOG"; break;
2765         default:
2766             ERR("Unexpected instruction %d\n", ins->handler_idx);
2767             return;
2768     }
2769
2770     /* LOG and LOGP operate on the absolute value of the input */
2771     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2772
2773     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2774     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2775
2776     if(need_abs)
2777     {
2778         shader_addline(buffer, "ABS TA, %s;\n", src0);
2779         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2780     }
2781     else
2782     {
2783         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2784     }
2785 }
2786
2787 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2788 {
2789     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2790     char src0[50], src1[50], dst[50];
2791     struct wined3d_shader_src_param src0_copy = ins->src[0];
2792     BOOL need_abs = FALSE;
2793     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2794     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2795
2796     /* POW operates on the absolute value of the input */
2797     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2798
2799     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2800     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2801     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2802
2803     if (need_abs)
2804         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2805     else
2806         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2807
2808     if (priv->target_version >= NV2)
2809     {
2810         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2811         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2812         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2813     }
2814     else
2815     {
2816         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2817         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2818
2819         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2820         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2821         /* Possibly add flt_eps to avoid getting float special values */
2822         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2823         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2824         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2825         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2826     }
2827 }
2828
2829 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2830 {
2831     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2832     char src_name[50];
2833     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2834
2835     /* src0 is aL */
2836     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2837
2838     if(vshader)
2839     {
2840         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2841         struct list *e = list_head(&priv->control_frames);
2842         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2843
2844         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2845         /* The constant loader makes sure to load -1 into iX.w */
2846         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2847         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2848         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2849     }
2850     else
2851     {
2852         shader_addline(buffer, "LOOP %s;\n", src_name);
2853     }
2854 }
2855
2856 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2857 {
2858     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2859     char src_name[50];
2860     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2861
2862     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2863
2864     /* The constant loader makes sure to load -1 into iX.w */
2865     if(vshader)
2866     {
2867         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2868         struct list *e = list_head(&priv->control_frames);
2869         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2870
2871         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2872
2873         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2874         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2875         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2876     }
2877     else
2878     {
2879         shader_addline(buffer, "REP %s;\n", src_name);
2880     }
2881 }
2882
2883 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2884 {
2885     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2886     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2887
2888     if(vshader)
2889     {
2890         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2891         struct list *e = list_head(&priv->control_frames);
2892         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2893
2894         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2895         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2896         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2897
2898         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2899     }
2900     else
2901     {
2902         shader_addline(buffer, "ENDLOOP;\n");
2903     }
2904 }
2905
2906 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2907 {
2908     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2909     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2910
2911     if(vshader)
2912     {
2913         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2914         struct list *e = list_head(&priv->control_frames);
2915         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2916
2917         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2918         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2919         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2920
2921         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2922     }
2923     else
2924     {
2925         shader_addline(buffer, "ENDREP;\n");
2926     }
2927 }
2928
2929 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2930 {
2931     struct control_frame *control_frame;
2932
2933     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2934     {
2935         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2936     }
2937     ERR("Could not find loop for break\n");
2938     return NULL;
2939 }
2940
2941 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2942 {
2943     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2944     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2945     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2946
2947     if(vshader)
2948     {
2949         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2950     }
2951     else
2952     {
2953         shader_addline(buffer, "BRK;\n");
2954     }
2955 }
2956
2957 static const char *get_compare(enum wined3d_shader_rel_op op)
2958 {
2959     switch (op)
2960     {
2961         case WINED3D_SHADER_REL_OP_GT: return "GT";
2962         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2963         case WINED3D_SHADER_REL_OP_GE: return "GE";
2964         case WINED3D_SHADER_REL_OP_LT: return "LT";
2965         case WINED3D_SHADER_REL_OP_NE: return "NE";
2966         case WINED3D_SHADER_REL_OP_LE: return "LE";
2967         default:
2968             FIXME("Unrecognized operator %#x.\n", op);
2969             return "(\?\?)";
2970     }
2971 }
2972
2973 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2974 {
2975     switch (op)
2976     {
2977         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2978         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2979         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2980         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2981         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2982         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2983         default:
2984             FIXME("Unrecognized operator %#x.\n", op);
2985             return -1;
2986     }
2987 }
2988
2989 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2990 {
2991     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2992     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2993     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2994     char src_name0[50];
2995     char src_name1[50];
2996     const char *comp = get_compare(ins->flags);
2997
2998     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2999     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3000
3001     if(vshader)
3002     {
3003         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3004          * away the subtraction result
3005          */
3006         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3007         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3008     }
3009     else
3010     {
3011         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3012         shader_addline(buffer, "BRK (%s.x);\n", comp);
3013     }
3014 }
3015
3016 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3017 {
3018     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3019     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3020     struct list *e = list_head(&priv->control_frames);
3021     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3022     const char *comp;
3023     char src_name0[50];
3024     char src_name1[50];
3025     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3026
3027     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3028     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3029
3030     if(vshader)
3031     {
3032         /* Invert the flag. We jump to the else label if the condition is NOT true */
3033         comp = get_compare(invert_compare(ins->flags));
3034         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3035         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3036     }
3037     else
3038     {
3039         comp = get_compare(ins->flags);
3040         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3041         shader_addline(buffer, "IF %s.x;\n", comp);
3042     }
3043 }
3044
3045 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3046 {
3047     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3048     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3049     struct list *e = list_head(&priv->control_frames);
3050     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3051     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3052
3053     if(vshader)
3054     {
3055         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3056         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3057         control_frame->had_else = TRUE;
3058     }
3059     else
3060     {
3061         shader_addline(buffer, "ELSE;\n");
3062     }
3063 }
3064
3065 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3066 {
3067     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3068     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3069     struct list *e = list_head(&priv->control_frames);
3070     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3071     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3072
3073     if(vshader)
3074     {
3075         if(control_frame->had_else)
3076         {
3077             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3078         }
3079         else
3080         {
3081             shader_addline(buffer, "#No else branch. else is endif\n");
3082             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3083         }
3084     }
3085     else
3086     {
3087         shader_addline(buffer, "ENDIF;\n");
3088     }
3089 }
3090
3091 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3092 {
3093     DWORD sampler_idx = ins->src[1].reg.idx;
3094     char reg_dest[40];
3095     char reg_src[3][40];
3096     WORD flags = TEX_DERIV;
3097
3098     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3099     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3100     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3101     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3102
3103     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3104     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3105
3106     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3107 }
3108
3109 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3110 {
3111     DWORD sampler_idx = ins->src[1].reg.idx;
3112     char reg_dest[40];
3113     char reg_coord[40];
3114     WORD flags = TEX_LOD;
3115
3116     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3117     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3118
3119     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3120     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3121
3122     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3123 }
3124
3125 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3126 {
3127     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3128     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3129
3130     priv->in_main_func = FALSE;
3131     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3132      * subroutine, don't generate a label that will make GL complain
3133      */
3134     if(priv->target_version == ARB) return;
3135
3136     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3137 }
3138
3139 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3140         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3141         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3142         struct wined3d_shader_buffer *buffer)
3143 {
3144     unsigned int i;
3145
3146     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3147      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3148      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3149      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3150      */
3151     if (args->super.fog_src == VS_FOG_Z)
3152         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3153     else if (!reg_maps->fog)
3154         /* posFixup.x is always 1.0, so we can safely use it */
3155         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3156
3157     /* Clipplanes are always stored without y inversion */
3158     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3159     {
3160         if (args->super.clip_enabled)
3161         {
3162             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3163             {
3164                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3165             }
3166         }
3167     }
3168     else if (args->clip.boolclip.clip_texcoord)
3169     {
3170         unsigned int cur_clip = 0;
3171         char component[4] = {'x', 'y', 'z', 'w'};
3172         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3173
3174         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3175         {
3176             if (args->clip.boolclip.clipplane_mask & (1 << i))
3177             {
3178                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3179                                component[cur_clip++], i);
3180             }
3181         }
3182         switch (cur_clip)
3183         {
3184             case 0:
3185                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3186                 break;
3187             case 1:
3188                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3189                 break;
3190             case 2:
3191                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3192                 break;
3193             case 3:
3194                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3195                 break;
3196         }
3197         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3198                        args->clip.boolclip.clip_texcoord - 1);
3199     }
3200
3201     /* Write the final position.
3202      *
3203      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3204      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3205      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3206      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3207      */
3208     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3209     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3210     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3211
3212     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3213      * and the glsl equivalent
3214      */
3215     if (need_helper_const(shader_data, reg_maps, gl_info))
3216     {
3217         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3218         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3219     }
3220     else
3221     {
3222         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3223         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3224     }
3225
3226     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3227
3228     priv_ctx->footer_written = TRUE;
3229 }
3230
3231 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3232 {
3233     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3234     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3235     const struct wined3d_shader *shader = ins->ctx->shader;
3236     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3237
3238     if(priv->target_version == ARB) return;
3239
3240     if(vshader)
3241     {
3242         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3243                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3244     }
3245
3246     shader_addline(buffer, "RET;\n");
3247 }
3248
3249 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3250 {
3251     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3252     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3253 }
3254
3255 /* GL locking is done by the caller */
3256 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3257 {
3258     GLuint program_id = 0;
3259     GLint pos;
3260
3261     const char *blt_vprogram =
3262         "!!ARBvp1.0\n"
3263         "PARAM c[1] = { { 1, 0.5 } };\n"
3264         "MOV result.position, vertex.position;\n"
3265         "MOV result.color, c[0].x;\n"
3266         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3267         "END\n";
3268
3269     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3270     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3271     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3272             strlen(blt_vprogram), blt_vprogram));
3273     checkGLcall("glProgramStringARB()");
3274
3275     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3276     if (pos != -1)
3277     {
3278         FIXME("Vertex program error at position %d: %s\n\n", pos,
3279             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3280         shader_arb_dump_program_source(blt_vprogram);
3281     }
3282     else
3283     {
3284         GLint native;
3285
3286         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3287         checkGLcall("glGetProgramivARB()");
3288         if (!native) WARN("Program exceeds native resource limits.\n");
3289     }
3290
3291     return program_id;
3292 }
3293
3294 /* GL locking is done by the caller */
3295 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3296         enum tex_types tex_type, BOOL masked)
3297 {
3298     GLuint program_id = 0;
3299     const char *fprogram;
3300     GLint pos;
3301
3302     static const char * const blt_fprograms_full[tex_type_count] =
3303     {
3304         /* tex_1d */
3305         NULL,
3306         /* tex_2d */
3307         "!!ARBfp1.0\n"
3308         "TEMP R0;\n"
3309         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3310         "MOV result.depth.z, R0.x;\n"
3311         "END\n",
3312         /* tex_3d */
3313         NULL,
3314         /* tex_cube */
3315         "!!ARBfp1.0\n"
3316         "TEMP R0;\n"
3317         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3318         "MOV result.depth.z, R0.x;\n"
3319         "END\n",
3320         /* tex_rect */
3321         "!!ARBfp1.0\n"
3322         "TEMP R0;\n"
3323         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3324         "MOV result.depth.z, R0.x;\n"
3325         "END\n",
3326     };
3327
3328     static const char * const blt_fprograms_masked[tex_type_count] =
3329     {
3330         /* tex_1d */
3331         NULL,
3332         /* tex_2d */
3333         "!!ARBfp1.0\n"
3334         "PARAM mask = program.local[0];\n"
3335         "TEMP R0;\n"
3336         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3337         "MUL R0.x, R0.x, R0.y;\n"
3338         "KIL -R0.x;\n"
3339         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3340         "MOV result.depth.z, R0.x;\n"
3341         "END\n",
3342         /* tex_3d */
3343         NULL,
3344         /* tex_cube */
3345         "!!ARBfp1.0\n"
3346         "PARAM mask = program.local[0];\n"
3347         "TEMP R0;\n"
3348         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3349         "MUL R0.x, R0.x, R0.y;\n"
3350         "KIL -R0.x;\n"
3351         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3352         "MOV result.depth.z, R0.x;\n"
3353         "END\n",
3354         /* tex_rect */
3355         "!!ARBfp1.0\n"
3356         "PARAM mask = program.local[0];\n"
3357         "TEMP R0;\n"
3358         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3359         "MUL R0.x, R0.x, R0.y;\n"
3360         "KIL -R0.x;\n"
3361         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3362         "MOV result.depth.z, R0.x;\n"
3363         "END\n",
3364     };
3365
3366     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3367     if (!fprogram)
3368     {
3369         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3370         tex_type = tex_2d;
3371         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3372     }
3373
3374     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3375     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3376     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3377     checkGLcall("glProgramStringARB()");
3378
3379     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3380     if (pos != -1)
3381     {
3382         FIXME("Fragment program error at position %d: %s\n\n", pos,
3383             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3384         shader_arb_dump_program_source(fprogram);
3385     }
3386     else
3387     {
3388         GLint native;
3389
3390         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3391         checkGLcall("glGetProgramivARB()");
3392         if (!native) WARN("Program exceeds native resource limits.\n");
3393     }
3394
3395     return program_id;
3396 }
3397
3398 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3399         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3400 {
3401     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3402
3403     if(condcode)
3404     {
3405         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3406         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3407         /* Calculate the > 0.0031308 case */
3408         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3409         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3410         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3411         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3412         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3413         /* Calculate the < case */
3414         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3415     }
3416     else
3417     {
3418         /* Calculate the > 0.0031308 case */
3419         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3420         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3421         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3422         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3423         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3424         /* Calculate the < case */
3425         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3426         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3427         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3428         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3429         /* Store the components > 0.0031308 in the destination */
3430         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3431         /* Add the components that are < 0.0031308 */
3432         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3433         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3434         * result.color writes(.rgb first, then .a), or handle overwriting already written
3435         * components. The assembler uses a temporary register in this case, which is usually
3436         * not allocated from one of our registers that were used earlier.
3437         */
3438     }
3439     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3440 }
3441
3442 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3443 {
3444     const struct wined3d_shader_lconst *constant;
3445
3446     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3447     {
3448         if (constant->idx == idx)
3449         {
3450             return constant->value;
3451         }
3452     }
3453     return NULL;
3454 }
3455
3456 static void init_ps_input(const struct wined3d_shader *shader,
3457         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3458 {
3459     static const char * const texcoords[8] =
3460     {
3461         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3462         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3463     };
3464     unsigned int i;
3465     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3466     const char *semantic_name;
3467     DWORD semantic_idx;
3468
3469     switch(args->super.vp_mode)
3470     {
3471         case pretransformed:
3472         case fixedfunction:
3473             /* The pixelshader has to collect the varyings on its own. In any case properly load
3474              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3475              * other attribs to 0.0.
3476              *
3477              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3478              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3479              * load the texcoord attrib pointers to match the pixel shader signature
3480              */
3481             for(i = 0; i < MAX_REG_INPUT; i++)
3482             {
3483                 semantic_name = sig[i].semantic_name;
3484                 semantic_idx = sig[i].semantic_idx;
3485                 if (!semantic_name) continue;
3486
3487                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3488                 {
3489                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3490                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3491                     else priv->ps_input[i] = "0.0";
3492                 }
3493                 else if(args->super.vp_mode == fixedfunction)
3494                 {
3495                     priv->ps_input[i] = "0.0";
3496                 }
3497                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3498                 {
3499                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3500                     else priv->ps_input[i] = "0.0";
3501                 }
3502                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3503                 {
3504                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3505                     else priv->ps_input[i] = "0.0";
3506                 }
3507                 else
3508                 {
3509                     priv->ps_input[i] = "0.0";
3510                 }
3511
3512                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3513             }
3514             break;
3515
3516         case vertexshader:
3517             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3518              * fragment.color
3519              */
3520             for(i = 0; i < 8; i++)
3521             {
3522                 priv->ps_input[i] = texcoords[i];
3523             }
3524             priv->ps_input[8] = "fragment.color.primary";
3525             priv->ps_input[9] = "fragment.color.secondary";
3526             break;
3527     }
3528 }
3529
3530 /* GL locking is done by the caller */
3531 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3532         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3533         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3534 {
3535     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3536     const struct wined3d_shader_lconst *lconst;
3537     const DWORD *function = shader->function;
3538     GLuint retval;
3539     char fragcolor[16];
3540     DWORD *lconst_map = local_const_mapping(shader), next_local;
3541     struct shader_arb_ctx_priv priv_ctx;
3542     BOOL dcl_td = FALSE;
3543     BOOL want_nv_prog = FALSE;
3544     struct arb_pshader_private *shader_priv = shader->backend_data;
3545     GLint errPos;
3546     DWORD map;
3547
3548     char srgbtmp[4][4];
3549     unsigned int i, found = 0;
3550
3551     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3552     {
3553         if (!(map & 1)
3554                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3555                 || (reg_maps->shader_version.major < 2 && !i))
3556             continue;
3557
3558         sprintf(srgbtmp[found], "R%u", i);
3559         ++found;
3560         if (found == 4) break;
3561     }
3562
3563     switch(found) {
3564         case 0:
3565             sprintf(srgbtmp[0], "TA");
3566             sprintf(srgbtmp[1], "TB");
3567             sprintf(srgbtmp[2], "TC");
3568             sprintf(srgbtmp[3], "TD");
3569             dcl_td = TRUE;
3570             break;
3571         case 1:
3572             sprintf(srgbtmp[1], "TA");
3573             sprintf(srgbtmp[2], "TB");
3574             sprintf(srgbtmp[3], "TC");
3575             break;
3576         case 2:
3577             sprintf(srgbtmp[2], "TA");
3578             sprintf(srgbtmp[3], "TB");
3579             break;
3580         case 3:
3581             sprintf(srgbtmp[3], "TA");
3582             break;
3583         case 4:
3584             break;
3585     }
3586
3587     /*  Create the hw ARB shader */
3588     memset(&priv_ctx, 0, sizeof(priv_ctx));
3589     priv_ctx.cur_ps_args = args;
3590     priv_ctx.compiled_fprog = compiled;
3591     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3592     init_ps_input(shader, args, &priv_ctx);
3593     list_init(&priv_ctx.control_frames);
3594
3595     /* Avoid enabling NV_fragment_program* if we do not need it.
3596      *
3597      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3598      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3599      * is faster than what we gain from using higher native instructions. There are some things though
3600      * that cannot be emulated. In that case enable the extensions.
3601      * If the extension is enabled, instruction handlers that support both ways will use it.
3602      *
3603      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3604      * So enable the best we can get.
3605      */
3606     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3607        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3608     {
3609         want_nv_prog = TRUE;
3610     }
3611
3612     shader_addline(buffer, "!!ARBfp1.0\n");
3613     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3614     {
3615         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3616         priv_ctx.target_version = NV3;
3617     }
3618     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3619     {
3620         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3621         priv_ctx.target_version = NV2;
3622     } else {
3623         if(want_nv_prog)
3624         {
3625             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3626              * limits properly
3627              */
3628             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3629             ERR("Try GLSL\n");
3630         }
3631         priv_ctx.target_version = ARB;
3632     }
3633
3634     if (reg_maps->rt_mask > 1)
3635     {
3636         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3637     }
3638
3639     if (reg_maps->shader_version.major < 3)
3640     {
3641         switch(args->super.fog) {
3642             case FOG_OFF:
3643                 break;
3644             case FOG_LINEAR:
3645                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3646                 break;
3647             case FOG_EXP:
3648                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3649                 break;
3650             case FOG_EXP2:
3651                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3652                 break;
3653         }
3654     }
3655
3656     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3657      * unused temps away(but occupies them for the whole shader if they're used once). Always
3658      * declaring them avoids tricky bookkeeping work
3659      */
3660     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3661     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3662     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3663     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3664     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3665     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3666     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3667
3668     if (reg_maps->shader_version.major < 2)
3669     {
3670         strcpy(fragcolor, "R0");
3671     }
3672     else
3673     {
3674         if (args->super.srgb_correction)
3675         {
3676             if (shader->u.ps.color0_mov)
3677             {
3678                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3679             }
3680             else
3681             {
3682                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3683                 strcpy(fragcolor, "TMP_COLOR");
3684             }
3685         } else {
3686             strcpy(fragcolor, "result.color");
3687         }
3688     }
3689
3690     if(args->super.srgb_correction) {
3691         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3692                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3693         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3694                        srgb_sub_high, 0.0, 0.0, 0.0);
3695     }
3696
3697     /* Base Declarations */
3698     next_local = shader_generate_arb_declarations(shader, reg_maps,
3699             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3700
3701     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3702     {
3703         unsigned char bump_const;
3704
3705         if (!(map & 1)) continue;
3706
3707         bump_const = compiled->numbumpenvmatconsts;
3708         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3709         compiled->bumpenvmatconst[bump_const].texunit = i;
3710         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3711         compiled->luminanceconst[bump_const].texunit = i;
3712
3713         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3714          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3715          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3716          * textures due to conditional NP2 restrictions)
3717          *
3718          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3719          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3720          * their location is shader dependent anyway and they cannot be loaded globally.
3721          */
3722         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3723         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3724                        i, compiled->bumpenvmatconst[bump_const].const_num);
3725         compiled->numbumpenvmatconsts = bump_const + 1;
3726
3727         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3728
3729         compiled->luminanceconst[bump_const].const_num = next_local++;
3730         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3731                        i, compiled->luminanceconst[bump_const].const_num);
3732     }
3733
3734     for(i = 0; i < MAX_CONST_I; i++)
3735     {
3736         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3737         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3738         {
3739             const DWORD *control_values = find_loop_control_values(shader, i);
3740
3741             if(control_values)
3742             {
3743                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3744                                 control_values[0], control_values[1], control_values[2]);
3745             }
3746             else
3747             {
3748                 compiled->int_consts[i] = next_local;
3749                 compiled->num_int_consts++;
3750                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3751             }
3752         }
3753     }
3754
3755     if(reg_maps->vpos || reg_maps->usesdsy)
3756     {
3757         compiled->ycorrection = next_local;
3758         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3759
3760         if(reg_maps->vpos)
3761         {
3762             shader_addline(buffer, "TEMP vpos;\n");
3763             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3764              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3765              * ycorrection.z: 1.0
3766              * ycorrection.w: 0.0
3767              */
3768             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3769             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3770         }
3771     }
3772     else
3773     {
3774         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3775     }
3776
3777     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3778      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3779      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3780      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3781      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3782      * shader compilation errors and the subsequent errors when drawing with this shader. */
3783     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3784         unsigned char cur_fixup_sampler = 0;
3785
3786         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3787         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3788         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3789
3790         fixup->offset = next_local;
3791         fixup->super.active = 0;
3792
3793         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3794             if (!(map & (1 << i))) continue;
3795
3796             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3797                 fixup->super.active |= (1 << i);
3798                 fixup->super.idx[i] = cur_fixup_sampler++;
3799             } else {
3800                 FIXME("No free constant found to load NP2 fixup data into shader. "
3801                       "Sampling from this texture will probably look wrong.\n");
3802                 break;
3803             }
3804         }
3805
3806         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3807         if (fixup->super.num_consts) {
3808             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3809                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3810         }
3811     }
3812
3813     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3814     {
3815         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3816     }
3817
3818     /* Base Shader Body */
3819     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3820
3821     if(args->super.srgb_correction) {
3822         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3823                                   priv_ctx.target_version >= NV2);
3824     }
3825
3826     if(strcmp(fragcolor, "result.color")) {
3827         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3828     }
3829     shader_addline(buffer, "END\n");
3830
3831     /* TODO: change to resource.glObjectHandle or something like that */
3832     GL_EXTCALL(glGenProgramsARB(1, &retval));
3833
3834     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3835     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3836
3837     TRACE("Created hw pixel shader, prg=%d\n", retval);
3838     /* Create the program and check for errors */
3839     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3840                buffer->bsize, buffer->buffer));
3841     checkGLcall("glProgramStringARB()");
3842
3843     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3844     if (errPos != -1)
3845     {
3846         FIXME("HW PixelShader Error at position %d: %s\n\n",
3847               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3848         shader_arb_dump_program_source(buffer->buffer);
3849         retval = 0;
3850     }
3851     else
3852     {
3853         GLint native;
3854
3855         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3856         checkGLcall("glGetProgramivARB()");
3857         if (!native) WARN("Program exceeds native resource limits.\n");
3858     }
3859
3860     /* Load immediate constants */
3861     if (lconst_map)
3862     {
3863         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3864         {
3865             const float *value = (const float *)lconst->value;
3866             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3867             checkGLcall("glProgramLocalParameter4fvARB");
3868         }
3869         HeapFree(GetProcessHeap(), 0, lconst_map);
3870     }
3871
3872     return retval;
3873 }
3874
3875 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3876 {
3877     unsigned int i;
3878     int ret;
3879
3880     for(i = 0; i < MAX_REG_INPUT; i++)
3881     {
3882         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3883         {
3884             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3885             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3886             continue;
3887         }
3888
3889         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3890         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3891         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3892         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3893         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3894         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3895     }
3896     return 0;
3897 }
3898
3899 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3900 {
3901     struct wined3d_shader_signature_element *new;
3902     int i;
3903     char *name;
3904
3905     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3906     for(i = 0; i < MAX_REG_INPUT; i++)
3907     {
3908         if (!sig[i].semantic_name) continue;
3909
3910         new[i] = sig[i];
3911         /* Clone the semantic string */
3912         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3913         strcpy(name, sig[i].semantic_name);
3914         new[i].semantic_name = name;
3915     }
3916     return new;
3917 }
3918
3919 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3920 {
3921     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3922     struct ps_signature *found_sig;
3923
3924     if (entry)
3925     {
3926         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3927         TRACE("Found existing signature %u\n", found_sig->idx);
3928         return found_sig->idx;
3929     }
3930     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3931     found_sig->sig = clone_sig(sig);
3932     found_sig->idx = priv->ps_sig_number++;
3933     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3934     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3935     {
3936         ERR("Failed to insert program entry.\n");
3937     }
3938     return found_sig->idx;
3939 }
3940
3941 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3942         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3943 {
3944     unsigned int i, j;
3945     static const char * const texcoords[8] =
3946     {
3947         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3948         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3949     };
3950     struct wined3d_device *device = shader->device;
3951     const struct wined3d_shader_signature_element *sig;
3952     const char *semantic_name;
3953     DWORD semantic_idx, reg_idx;
3954
3955     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3956      * and varying 9 to result.color.secondary
3957      */
3958     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3959     {
3960         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3961         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3962         "result.color.primary", "result.color.secondary"
3963     };
3964
3965     if(sig_num == ~0)
3966     {
3967         TRACE("Pixel shader uses builtin varyings\n");
3968         /* Map builtins to builtins */
3969         for(i = 0; i < 8; i++)
3970         {
3971             priv_ctx->texcrd_output[i] = texcoords[i];
3972         }
3973         priv_ctx->color_output[0] = "result.color.primary";
3974         priv_ctx->color_output[1] = "result.color.secondary";
3975         priv_ctx->fog_output = "result.fogcoord";
3976
3977         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3978         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3979         {
3980             semantic_name = shader->output_signature[i].semantic_name;
3981             if (!semantic_name) continue;
3982
3983             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
3984             {
3985                 TRACE("o%u is TMP_OUT\n", i);
3986                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3987                 else priv_ctx->vs_output[i] = "TA";
3988             }
3989             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
3990             {
3991                 TRACE("o%u is result.pointsize\n", i);
3992                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3993                 else priv_ctx->vs_output[i] = "TA";
3994             }
3995             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3996             {
3997                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3998                 if (!shader->output_signature[i].semantic_idx)
3999                     priv_ctx->vs_output[i] = "result.color.primary";
4000                 else if (shader->output_signature[i].semantic_idx == 1)
4001                     priv_ctx->vs_output[i] = "result.color.secondary";
4002                 else priv_ctx->vs_output[i] = "TA";
4003             }
4004             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4005             {
4006                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4007                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4008                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4009             }
4010             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4011             {
4012                 TRACE("o%u is result.fogcoord\n", i);
4013                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4014                 else priv_ctx->vs_output[i] = "result.fogcoord";
4015             }
4016             else
4017             {
4018                 priv_ctx->vs_output[i] = "TA";
4019             }
4020         }
4021         return;
4022     }
4023
4024     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4025      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4026      */
4027     sig = device->stateBlock->state.pixel_shader->input_signature;
4028     TRACE("Pixel shader uses declared varyings\n");
4029
4030     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4031     for(i = 0; i < 8; i++)
4032     {
4033         priv_ctx->texcrd_output[i] = "TA";
4034     }
4035     priv_ctx->color_output[0] = "TA";
4036     priv_ctx->color_output[1] = "TA";
4037     priv_ctx->fog_output = "TA";
4038
4039     for(i = 0; i < MAX_REG_INPUT; i++)
4040     {
4041         semantic_name = sig[i].semantic_name;
4042         semantic_idx = sig[i].semantic_idx;
4043         reg_idx = sig[i].register_idx;
4044         if (!semantic_name) continue;
4045
4046         /* If a declared input register is not written by builtin arguments, don't write to it.
4047          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4048          *
4049          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4050          * to TMP_OUT in any case
4051          */
4052         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4053         {
4054             if (semantic_idx < 8)
4055                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4056         }
4057         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4058         {
4059             if (semantic_idx < 2)
4060                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4061         }
4062         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4063         {
4064             if (!semantic_idx)
4065                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4066         }
4067         else
4068         {
4069             continue;
4070         }
4071
4072         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4073                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4074         {
4075             compiled->need_color_unclamp = TRUE;
4076         }
4077     }
4078
4079     /* Map declared to declared */
4080     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4081     {
4082         /* Write unread output to TA to throw them away */
4083         priv_ctx->vs_output[i] = "TA";
4084         semantic_name = shader->output_signature[i].semantic_name;
4085         if (!semantic_name) continue;
4086
4087         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4088                 && !shader->output_signature[i].semantic_idx)
4089         {
4090             priv_ctx->vs_output[i] = "TMP_OUT";
4091             continue;
4092         }
4093         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4094                 && !shader->output_signature[i].semantic_idx)
4095         {
4096             priv_ctx->vs_output[i] = "result.pointsize";
4097             continue;
4098         }
4099
4100         for(j = 0; j < MAX_REG_INPUT; j++)
4101         {
4102             if (!sig[j].semantic_name) continue;
4103
4104             if (!strcmp(sig[j].semantic_name, semantic_name)
4105                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4106             {
4107                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4108
4109                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4110                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4111                 {
4112                     compiled->need_color_unclamp = TRUE;
4113                 }
4114             }
4115         }
4116     }
4117 }
4118
4119 /* GL locking is done by the caller */
4120 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4121         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4122         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4123 {
4124     const struct arb_vshader_private *shader_data = shader->backend_data;
4125     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4126     const struct wined3d_shader_lconst *lconst;
4127     const DWORD *function = shader->function;
4128     GLuint ret;
4129     DWORD next_local, *lconst_map = local_const_mapping(shader);
4130     struct shader_arb_ctx_priv priv_ctx;
4131     unsigned int i;
4132     GLint errPos;
4133
4134     memset(&priv_ctx, 0, sizeof(priv_ctx));
4135     priv_ctx.cur_vs_args = args;
4136     list_init(&priv_ctx.control_frames);
4137     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4138
4139     /*  Create the hw ARB shader */
4140     shader_addline(buffer, "!!ARBvp1.0\n");
4141
4142     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4143      * mesurable performance penalty, and we can always make use of it for clipplanes.
4144      */
4145     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4146     {
4147         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4148         priv_ctx.target_version = NV3;
4149         shader_addline(buffer, "ADDRESS aL;\n");
4150     }
4151     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4152     {
4153         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4154         priv_ctx.target_version = NV2;
4155         shader_addline(buffer, "ADDRESS aL;\n");
4156     } else {
4157         priv_ctx.target_version = ARB;
4158     }
4159
4160     shader_addline(buffer, "TEMP TMP_OUT;\n");
4161     if (need_helper_const(shader_data, reg_maps, gl_info))
4162     {
4163         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4164     }
4165     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4166     {
4167         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4168         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4169     }
4170
4171     shader_addline(buffer, "TEMP TA;\n");
4172     shader_addline(buffer, "TEMP TB;\n");
4173
4174     /* Base Declarations */
4175     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4176             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4177
4178     for(i = 0; i < MAX_CONST_I; i++)
4179     {
4180         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4181         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4182         {
4183             const DWORD *control_values = find_loop_control_values(shader, i);
4184
4185             if(control_values)
4186             {
4187                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4188                                 control_values[0], control_values[1], control_values[2]);
4189             }
4190             else
4191             {
4192                 compiled->int_consts[i] = next_local;
4193                 compiled->num_int_consts++;
4194                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4195             }
4196         }
4197     }
4198
4199     /* We need a constant to fixup the final position */
4200     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4201     compiled->pos_fixup = next_local++;
4202
4203     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4204      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4205      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4206      * a replacement shader depend on the texcoord.w being set properly.
4207      *
4208      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4209      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4210      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4211      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4212      * this can eat a number of instructions, so skip it unless this cap is set as well
4213      */
4214     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4215     {
4216         struct wined3d_device *device = shader->device;
4217         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4218         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4219
4220         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4221         {
4222             int i;
4223             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4224             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4225             {
4226                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4227                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4228             }
4229         }
4230     }
4231
4232     /* The shader starts with the main function */
4233     priv_ctx.in_main_func = TRUE;
4234     /* Base Shader Body */
4235     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4236
4237     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4238             shader_data, args, reg_maps, gl_info, buffer);
4239
4240     shader_addline(buffer, "END\n");
4241
4242     /* TODO: change to resource.glObjectHandle or something like that */
4243     GL_EXTCALL(glGenProgramsARB(1, &ret));
4244
4245     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4246     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4247
4248     TRACE("Created hw vertex shader, prg=%d\n", ret);
4249     /* Create the program and check for errors */
4250     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4251                buffer->bsize, buffer->buffer));
4252     checkGLcall("glProgramStringARB()");
4253
4254     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4255     if (errPos != -1)
4256     {
4257         FIXME("HW VertexShader Error at position %d: %s\n\n",
4258               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4259         shader_arb_dump_program_source(buffer->buffer);
4260         ret = -1;
4261     }
4262     else
4263     {
4264         GLint native;
4265
4266         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4267         checkGLcall("glGetProgramivARB()");
4268         if (!native) WARN("Program exceeds native resource limits.\n");
4269
4270         /* Load immediate constants */
4271         if (lconst_map)
4272         {
4273             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4274             {
4275                 const float *value = (const float *)lconst->value;
4276                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4277             }
4278         }
4279     }
4280     HeapFree(GetProcessHeap(), 0, lconst_map);
4281
4282     return ret;
4283 }
4284
4285 /* GL locking is done by the caller */
4286 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4287         const struct arb_ps_compile_args *args)
4288 {
4289     struct wined3d_device *device = shader->device;
4290     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4291     UINT i;
4292     DWORD new_size;
4293     struct arb_ps_compiled_shader *new_array;
4294     struct wined3d_shader_buffer buffer;
4295     struct arb_pshader_private *shader_data;
4296     GLuint ret;
4297
4298     if (!shader->backend_data)
4299     {
4300         struct shader_arb_priv *priv = device->shader_priv;
4301
4302         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4303         shader_data = shader->backend_data;
4304         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4305
4306         if (shader->reg_maps.shader_version.major < 3)
4307             shader_data->input_signature_idx = ~0;
4308         else
4309             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4310
4311         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4312
4313         if (!device->vs_clipping)
4314             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4315                     gl_info->limits.texture_stages - 1);
4316         else
4317             shader_data->clipplane_emulation = ~0U;
4318     }
4319     shader_data = shader->backend_data;
4320
4321     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4322      * so a linear search is more performant than a hashmap or a binary search
4323      * (cache coherency etc)
4324      */
4325     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4326     {
4327         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4328             return &shader_data->gl_shaders[i];
4329     }
4330
4331     TRACE("No matching GL shader found, compiling a new shader\n");
4332     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4333         if (shader_data->num_gl_shaders)
4334         {
4335             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4336             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4337                                     new_size * sizeof(*shader_data->gl_shaders));
4338         } else {
4339             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4340             new_size = 1;
4341         }
4342
4343         if(!new_array) {
4344             ERR("Out of memory\n");
4345             return 0;
4346         }
4347         shader_data->gl_shaders = new_array;
4348         shader_data->shader_array_size = new_size;
4349     }
4350
4351     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4352
4353     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4354
4355     if (!shader_buffer_init(&buffer))
4356     {
4357         ERR("Failed to initialize shader buffer.\n");
4358         return 0;
4359     }
4360
4361     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4362             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4363     shader_buffer_free(&buffer);
4364     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4365
4366     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4367 }
4368
4369 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4370                                  const DWORD use_map, BOOL skip_int) {
4371     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4372     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4373     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4374     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4375     if(stored->ps_signature != new->ps_signature) return FALSE;
4376     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4377     if(skip_int) return TRUE;
4378
4379     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4380 }
4381
4382 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4383         const struct arb_vs_compile_args *args)
4384 {
4385     struct wined3d_device *device = shader->device;
4386     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4387     DWORD use_map = device->strided_streams.use_map;
4388     UINT i;
4389     DWORD new_size;
4390     struct arb_vs_compiled_shader *new_array;
4391     struct wined3d_shader_buffer buffer;
4392     struct arb_vshader_private *shader_data;
4393     GLuint ret;
4394
4395     if (!shader->backend_data)
4396     {
4397         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4398
4399         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4400         shader_data = shader->backend_data;
4401
4402         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4403                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4404         {
4405             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4406             {
4407                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4408                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4409                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4410             }
4411             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4412                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4413             else if (reg_maps->max_rel_offset > 63)
4414                 shader_data->rel_offset = reg_maps->min_rel_offset;
4415         }
4416     }
4417     shader_data = shader->backend_data;
4418
4419     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4420      * so a linear search is more performant than a hashmap or a binary search
4421      * (cache coherency etc)
4422      */
4423     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4424         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4425                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4426         {
4427             return &shader_data->gl_shaders[i];
4428         }
4429     }
4430
4431     TRACE("No matching GL shader found, compiling a new shader\n");
4432
4433     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4434         if (shader_data->num_gl_shaders)
4435         {
4436             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4437             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4438                                     new_size * sizeof(*shader_data->gl_shaders));
4439         } else {
4440             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4441             new_size = 1;
4442         }
4443
4444         if(!new_array) {
4445             ERR("Out of memory\n");
4446             return 0;
4447         }
4448         shader_data->gl_shaders = new_array;
4449         shader_data->shader_array_size = new_size;
4450     }
4451
4452     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4453
4454     if (!shader_buffer_init(&buffer))
4455     {
4456         ERR("Failed to initialize shader buffer.\n");
4457         return 0;
4458     }
4459
4460     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4461             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4462     shader_buffer_free(&buffer);
4463     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4464
4465     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4466 }
4467
4468 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4469         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4470 {
4471     struct wined3d_device *device = shader->device;
4472     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4473     int i;
4474     WORD int_skip;
4475
4476     find_ps_compile_args(state, shader, &args->super);
4477
4478     /* This forces all local boolean constants to 1 to make them stateblock independent */
4479     args->bools = shader->reg_maps.local_bool_consts;
4480
4481     for(i = 0; i < MAX_CONST_B; i++)
4482     {
4483         if (state->ps_consts_b[i])
4484             args->bools |= ( 1 << i);
4485     }
4486
4487     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4488      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4489      * duplicate the shader than have a no-op KIL instruction in every shader
4490      */
4491     if (!device->vs_clipping && use_vs(state)
4492             && state->render_states[WINED3D_RS_CLIPPING]
4493             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4494         args->clip = 1;
4495     else
4496         args->clip = 0;
4497
4498     /* Skip if unused or local, or supported natively */
4499     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4500     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4501     {
4502         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4503         return;
4504     }
4505
4506     for(i = 0; i < MAX_CONST_I; i++)
4507     {
4508         if(int_skip & (1 << i))
4509         {
4510             args->loop_ctrl[i][0] = 0;
4511             args->loop_ctrl[i][1] = 0;
4512             args->loop_ctrl[i][2] = 0;
4513         }
4514         else
4515         {
4516             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4517             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4518             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4519         }
4520     }
4521 }
4522
4523 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4524         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4525 {
4526     struct wined3d_device *device = shader->device;
4527     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4528     int i;
4529     WORD int_skip;
4530
4531     find_vs_compile_args(state, shader, &args->super);
4532
4533     args->clip.boolclip_compare = 0;
4534     if (use_ps(state))
4535     {
4536         const struct wined3d_shader *ps = state->pixel_shader;
4537         const struct arb_pshader_private *shader_priv = ps->backend_data;
4538         args->ps_signature = shader_priv->input_signature_idx;
4539
4540         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4541     }
4542     else
4543     {
4544         args->ps_signature = ~0;
4545         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4546         {
4547             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4548         }
4549         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4550     }
4551
4552     if (args->clip.boolclip.clip_texcoord)
4553     {
4554         if (state->render_states[WINED3D_RS_CLIPPING])
4555             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4556         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4557     }
4558
4559     /* This forces all local boolean constants to 1 to make them stateblock independent */
4560     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4561     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4562     for(i = 0; i < MAX_CONST_B; i++)
4563     {
4564         if (state->vs_consts_b[i])
4565             args->clip.boolclip.bools |= ( 1 << i);
4566     }
4567
4568     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4569     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4570     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4571     args->vertex.samplers[3] = 0;
4572
4573     /* Skip if unused or local */
4574     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4575     /* This is about flow control, not clipping. */
4576     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4577     {
4578         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4579         return;
4580     }
4581
4582     for(i = 0; i < MAX_CONST_I; i++)
4583     {
4584         if(int_skip & (1 << i))
4585         {
4586             args->loop_ctrl[i][0] = 0;
4587             args->loop_ctrl[i][1] = 0;
4588             args->loop_ctrl[i][2] = 0;
4589         }
4590         else
4591         {
4592             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4593             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4594             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4595         }
4596     }
4597 }
4598
4599 /* GL locking is done by the caller */
4600 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4601 {
4602     struct wined3d_device *device = context->swapchain->device;
4603     struct shader_arb_priv *priv = device->shader_priv;
4604     const struct wined3d_gl_info *gl_info = context->gl_info;
4605     const struct wined3d_state *state = &device->stateBlock->state;
4606     int i;
4607
4608     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4609     if (usePS)
4610     {
4611         struct wined3d_shader *ps = state->pixel_shader;
4612         struct arb_ps_compile_args compile_args;
4613         struct arb_ps_compiled_shader *compiled;
4614
4615         TRACE("Using pixel shader %p.\n", ps);
4616         find_arb_ps_compile_args(state, ps, &compile_args);
4617         compiled = find_arb_pshader(ps, &compile_args);
4618         priv->current_fprogram_id = compiled->prgId;
4619         priv->compiled_fprog = compiled;
4620
4621         /* Bind the fragment program */
4622         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4623         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4624
4625         if(!priv->use_arbfp_fixed_func) {
4626             /* Enable OpenGL fragment programs */
4627             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4628             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4629         }
4630         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4631                 device, priv->current_fprogram_id);
4632
4633         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4634          * a 1.x and newer shader, reload the first 8 constants
4635          */
4636         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4637         {
4638             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4639             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4640             for(i = 0; i < 8; i++)
4641             {
4642                 priv->pshader_const_dirty[i] = 1;
4643             }
4644             /* Also takes care of loading local constants */
4645             shader_arb_load_constants(context, TRUE, FALSE);
4646         }
4647         else
4648         {
4649             UINT rt_height = state->fb->render_targets[0]->resource.height;
4650             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4651         }
4652
4653         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4654         if (compiled->np2fixup_info.super.active)
4655             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4656     }
4657     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4658     {
4659         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4660         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4661         * replacement shader
4662         */
4663         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4664         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4665         priv->current_fprogram_id = 0;
4666     }
4667
4668     if (useVS)
4669     {
4670         struct wined3d_shader *vs = state->vertex_shader;
4671         struct arb_vs_compile_args compile_args;
4672         struct arb_vs_compiled_shader *compiled;
4673
4674         TRACE("Using vertex shader %p\n", vs);
4675         find_arb_vs_compile_args(state, vs, &compile_args);
4676         compiled = find_arb_vshader(vs, &compile_args);
4677         priv->current_vprogram_id = compiled->prgId;
4678         priv->compiled_vprog = compiled;
4679
4680         /* Bind the vertex program */
4681         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4682         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4683
4684         /* Enable OpenGL vertex programs */
4685         glEnable(GL_VERTEX_PROGRAM_ARB);
4686         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4687         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4688         shader_arb_vs_local_constants(compiled, context, state);
4689
4690         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4691             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4692
4693             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4694             {
4695                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4696                 checkGLcall("glClampColorARB");
4697             } else {
4698                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4699             }
4700         }
4701     }
4702     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4703     {
4704         priv->current_vprogram_id = 0;
4705         glDisable(GL_VERTEX_PROGRAM_ARB);
4706         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4707     }
4708 }
4709
4710 /* GL locking is done by the caller */
4711 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4712         enum tex_types tex_type, const SIZE *ds_mask_size)
4713 {
4714     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4715     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4716     struct shader_arb_priv *priv = shader_priv;
4717     GLuint *blt_fprogram;
4718
4719     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4720     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4721     glEnable(GL_VERTEX_PROGRAM_ARB);
4722
4723     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4724     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4725     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4726     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4727     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4728 }
4729
4730 /* GL locking is done by the caller */
4731 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4732 {
4733     struct shader_arb_priv *priv = shader_priv;
4734
4735     if (priv->current_vprogram_id) {
4736         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4737         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4738
4739         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4740     }
4741     else
4742     {
4743         glDisable(GL_VERTEX_PROGRAM_ARB);
4744         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4745     }
4746
4747     if (priv->current_fprogram_id) {
4748         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4749         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4750
4751         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4752     }
4753     else if(!priv->use_arbfp_fixed_func)
4754     {
4755         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4756         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4757     }
4758 }
4759
4760 static void shader_arb_destroy(struct wined3d_shader *shader)
4761 {
4762     struct wined3d_device *device = shader->device;
4763     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4764
4765     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4766     {
4767         struct arb_pshader_private *shader_data = shader->backend_data;
4768         UINT i;
4769
4770         if(!shader_data) return; /* This can happen if a shader was never compiled */
4771
4772         if (shader_data->num_gl_shaders)
4773         {
4774             struct wined3d_context *context = context_acquire(device, NULL);
4775
4776             ENTER_GL();
4777             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4778             {
4779                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4780                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4781             }
4782             LEAVE_GL();
4783
4784             context_release(context);
4785         }
4786
4787         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4788         HeapFree(GetProcessHeap(), 0, shader_data);
4789         shader->backend_data = NULL;
4790     }
4791     else
4792     {
4793         struct arb_vshader_private *shader_data = shader->backend_data;
4794         UINT i;
4795
4796         if(!shader_data) return; /* This can happen if a shader was never compiled */
4797
4798         if (shader_data->num_gl_shaders)
4799         {
4800             struct wined3d_context *context = context_acquire(device, NULL);
4801
4802             ENTER_GL();
4803             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4804             {
4805                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4806                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4807             }
4808             LEAVE_GL();
4809
4810             context_release(context);
4811         }
4812
4813         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4814         HeapFree(GetProcessHeap(), 0, shader_data);
4815         shader->backend_data = NULL;
4816     }
4817 }
4818
4819 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4820 {
4821     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4822     return compare_sig(key, e->sig);
4823 }
4824
4825 static const struct wine_rb_functions sig_tree_functions =
4826 {
4827     wined3d_rb_alloc,
4828     wined3d_rb_realloc,
4829     wined3d_rb_free,
4830     sig_tree_compare
4831 };
4832
4833 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4834 {
4835     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4836
4837     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4838             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4839     if (!priv->vshader_const_dirty)
4840         goto fail;
4841     memset(priv->vshader_const_dirty, 1,
4842            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4843
4844     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4845             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4846     if (!priv->pshader_const_dirty)
4847         goto fail;
4848     memset(priv->pshader_const_dirty, 1,
4849             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4850
4851     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4852     {
4853         ERR("RB tree init failed\n");
4854         goto fail;
4855     }
4856     device->shader_priv = priv;
4857     return WINED3D_OK;
4858
4859 fail:
4860     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4861     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4862     HeapFree(GetProcessHeap(), 0, priv);
4863     return E_OUTOFMEMORY;
4864 }
4865
4866 static void release_signature(struct wine_rb_entry *entry, void *context)
4867 {
4868     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4869     int i;
4870     for(i = 0; i < MAX_REG_INPUT; i++)
4871     {
4872         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4873     }
4874     HeapFree(GetProcessHeap(), 0, sig->sig);
4875     HeapFree(GetProcessHeap(), 0, sig);
4876 }
4877
4878 /* Context activation is done by the caller. */
4879 static void shader_arb_free(struct wined3d_device *device)
4880 {
4881     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4882     struct shader_arb_priv *priv = device->shader_priv;
4883     int i;
4884
4885     ENTER_GL();
4886     if(priv->depth_blt_vprogram_id) {
4887         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4888     }
4889     for (i = 0; i < tex_type_count; ++i)
4890     {
4891         if (priv->depth_blt_fprogram_id_full[i])
4892         {
4893             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4894         }
4895         if (priv->depth_blt_fprogram_id_masked[i])
4896         {
4897             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4898         }
4899     }
4900     LEAVE_GL();
4901
4902     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4903     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4904     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4905     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4906 }
4907
4908 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4909 {
4910     struct shader_arb_priv *priv = shader_priv;
4911
4912     if (priv->last_context == context)
4913         priv->last_context = NULL;
4914 }
4915
4916 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4917 {
4918     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4919     {
4920         DWORD vs_consts;
4921
4922         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4923          * for vertex programs. If the native limit is less than that it's
4924          * not very useful, and e.g. Mesa swrast returns 0, probably to
4925          * indicate it's a software implementation. */
4926         if (gl_info->limits.arb_vs_native_constants < 96)
4927             vs_consts = gl_info->limits.arb_vs_float_constants;
4928         else
4929             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4930
4931         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4932         {
4933             caps->VertexShaderVersion = 3;
4934             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4935         }
4936         else if (vs_consts >= 256)
4937         {
4938             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4939             caps->VertexShaderVersion = 2;
4940             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4941         }
4942         else
4943         {
4944             caps->VertexShaderVersion = 1;
4945             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4946         }
4947         caps->MaxVertexShaderConst = vs_consts;
4948     }
4949     else
4950     {
4951         caps->VertexShaderVersion = 0;
4952         caps->MaxVertexShaderConst = 0;
4953     }
4954
4955     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4956     {
4957         DWORD ps_consts;
4958
4959         /* Similar as above for vertex programs, but the minimum for fragment
4960          * programs is 24. */
4961         if (gl_info->limits.arb_ps_native_constants < 24)
4962             ps_consts = gl_info->limits.arb_ps_float_constants;
4963         else
4964             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4965
4966         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4967         {
4968             caps->PixelShaderVersion = 3;
4969             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4970         }
4971         else if (ps_consts >= 32)
4972         {
4973             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4974             caps->PixelShaderVersion = 2;
4975             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4976         }
4977         else
4978         {
4979             caps->PixelShaderVersion = 1;
4980             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4981         }
4982         caps->PixelShader1xMaxValue = 8.0f;
4983         caps->MaxPixelShaderConst = ps_consts;
4984     }
4985     else
4986     {
4987         caps->PixelShaderVersion = 0;
4988         caps->PixelShader1xMaxValue = 0.0f;
4989         caps->MaxPixelShaderConst = 0;
4990     }
4991
4992     caps->VSClipping = use_nv_clip(gl_info);
4993 }
4994
4995 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4996 {
4997     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4998     {
4999         TRACE("Checking support for color_fixup:\n");
5000         dump_color_fixup_desc(fixup);
5001     }
5002
5003     /* We support everything except complex conversions. */
5004     if (!is_complex_fixup(fixup))
5005     {
5006         TRACE("[OK]\n");
5007         return TRUE;
5008     }
5009
5010     TRACE("[FAILED]\n");
5011     return FALSE;
5012 }
5013
5014 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5015     DWORD shift;
5016     char write_mask[20], regstr[50];
5017     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5018     BOOL is_color = FALSE;
5019     const struct wined3d_shader_dst_param *dst;
5020
5021     if (!ins->dst_count) return;
5022
5023     dst = &ins->dst[0];
5024     shift = dst->shift;
5025     if (!shift) return; /* Saturate alone is handled by the instructions */
5026
5027     shader_arb_get_write_mask(ins, dst, write_mask);
5028     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5029
5030     /* Generate a line that does the output modifier computation
5031      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5032      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5033      */
5034     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5035                    regstr, write_mask, regstr, shift_tab[shift]);
5036 }
5037
5038 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5039 {
5040     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
5041     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
5042     /* WINED3DSIH_AND           */ NULL,
5043     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5044     /* WINED3DSIH_BREAK         */ shader_hw_break,
5045     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5046     /* WINED3DSIH_BREAKP        */ NULL,
5047     /* WINED3DSIH_CALL          */ shader_hw_call,
5048     /* WINED3DSIH_CALLNZ        */ NULL,
5049     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5050     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5051     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5052     /* WINED3DSIH_CUT           */ NULL,
5053     /* WINED3DSIH_DCL           */ NULL,
5054     /* WINED3DSIH_DEF           */ NULL,
5055     /* WINED3DSIH_DEFB          */ NULL,
5056     /* WINED3DSIH_DEFI          */ NULL,
5057     /* WINED3DSIH_DIV           */ NULL,
5058     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5059     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5060     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5061     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5062     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5063     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5064     /* WINED3DSIH_ELSE          */ shader_hw_else,
5065     /* WINED3DSIH_EMIT          */ NULL,
5066     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5067     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5068     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5069     /* WINED3DSIH_EQ            */ NULL,
5070     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5071     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5072     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5073     /* WINED3DSIH_FTOI          */ NULL,
5074     /* WINED3DSIH_GE            */ NULL,
5075     /* WINED3DSIH_IADD          */ NULL,
5076     /* WINED3DSIH_IEQ           */ NULL,
5077     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5078     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5079     /* WINED3DSIH_IGE           */ NULL,
5080     /* WINED3DSIH_IMUL          */ NULL,
5081     /* WINED3DSIH_ITOF          */ NULL,
5082     /* WINED3DSIH_LABEL         */ shader_hw_label,
5083     /* WINED3DSIH_LD            */ NULL,
5084     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5085     /* WINED3DSIH_LOG           */ shader_hw_log,
5086     /* WINED3DSIH_LOGP          */ shader_hw_log,
5087     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5088     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5089     /* WINED3DSIH_LT            */ NULL,
5090     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5091     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5092     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5093     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5094     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5095     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5096     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5097     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5098     /* WINED3DSIH_MOV           */ shader_hw_mov,
5099     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5100     /* WINED3DSIH_MOVC          */ NULL,
5101     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5102     /* WINED3DSIH_NOP           */ shader_hw_nop,
5103     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5104     /* WINED3DSIH_PHASE         */ NULL,
5105     /* WINED3DSIH_POW           */ shader_hw_pow,
5106     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5107     /* WINED3DSIH_REP           */ shader_hw_rep,
5108     /* WINED3DSIH_RET           */ shader_hw_ret,
5109     /* WINED3DSIH_ROUND_NI      */ NULL,
5110     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5111     /* WINED3DSIH_SAMPLE        */ NULL,
5112     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5113     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5114     /* WINED3DSIH_SETP          */ NULL,
5115     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5116     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5117     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5118     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5119     /* WINED3DSIH_SQRT          */ NULL,
5120     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5121     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5122     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5123     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5124     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5125     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5126     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5127     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5128     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5129     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5130     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5131     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5132     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5133     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5134     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5135     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5136     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5137     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5138     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5139     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5140     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5141     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5142     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5143     /* WINED3DSIH_UDIV          */ NULL,
5144     /* WINED3DSIH_USHR          */ NULL,
5145     /* WINED3DSIH_UTOF          */ NULL,
5146     /* WINED3DSIH_XOR           */ NULL,
5147 };
5148
5149 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5150         const struct wined3d_shader *shader, DWORD idx)
5151 {
5152     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5153     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5154     const struct wined3d_shader_lconst *constant;
5155     WORD bools = 0;
5156     WORD flag = (1 << idx);
5157     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5158
5159     if (reg_maps->local_bool_consts & flag)
5160     {
5161         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5162         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5163         {
5164             if (constant->idx == idx)
5165             {
5166                 return constant->value[0];
5167             }
5168         }
5169         ERR("Local constant not found\n");
5170         return FALSE;
5171     }
5172     else
5173     {
5174         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5175         else bools = priv->cur_ps_args->bools;
5176         return bools & flag;
5177     }
5178 }
5179
5180 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5181         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5182 {
5183     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5184     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5185
5186     /* Integer constants can either be a local constant, or they can be stored in the shader
5187      * type specific compile args. */
5188     if (reg_maps->local_int_consts & (1 << idx))
5189     {
5190         const struct wined3d_shader_lconst *constant;
5191
5192         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5193         {
5194             if (constant->idx == idx)
5195             {
5196                 loop_control->count = constant->value[0];
5197                 loop_control->start = constant->value[1];
5198                 /* Step is signed. */
5199                 loop_control->step = (int)constant->value[2];
5200                 return;
5201             }
5202         }
5203         /* If this happens the flag was set incorrectly */
5204         ERR("Local constant not found\n");
5205         loop_control->count = 0;
5206         loop_control->start = 0;
5207         loop_control->step = 0;
5208         return;
5209     }
5210
5211     switch (reg_maps->shader_version.type)
5212     {
5213         case WINED3D_SHADER_TYPE_VERTEX:
5214             /* Count and aL start value are unsigned */
5215             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5216             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5217             /* Step is signed. */
5218             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5219             break;
5220
5221         case WINED3D_SHADER_TYPE_PIXEL:
5222             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5223             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5224             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5225             break;
5226
5227         default:
5228             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5229             break;
5230     }
5231 }
5232
5233 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5234 {
5235     unsigned int i;
5236     struct wined3d_shader_dst_param *dst_param = NULL;
5237     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5238     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5239     if(!rec)
5240     {
5241         ERR("Out of memory\n");
5242         return;
5243     }
5244
5245     rec->ins = *ins;
5246     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5247     if(!dst_param) goto free;
5248     *dst_param = *ins->dst;
5249     if(ins->dst->reg.rel_addr)
5250     {
5251         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5252         if(!rel_addr) goto free;
5253         *rel_addr = *ins->dst->reg.rel_addr;
5254         dst_param->reg.rel_addr = rel_addr;
5255     }
5256     rec->ins.dst = dst_param;
5257
5258     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5259     if(!src_param) goto free;
5260     for(i = 0; i < ins->src_count; i++)
5261     {
5262         src_param[i] = ins->src[i];
5263         if(ins->src[i].reg.rel_addr)
5264         {
5265             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5266             if(!rel_addr) goto free;
5267             *rel_addr = *ins->src[i].reg.rel_addr;
5268             src_param[i].reg.rel_addr = rel_addr;
5269         }
5270     }
5271     rec->ins.src = src_param;
5272     list_add_tail(list, &rec->entry);
5273     return;
5274
5275 free:
5276     ERR("Out of memory\n");
5277     if(dst_param)
5278     {
5279         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5280         HeapFree(GetProcessHeap(), 0, dst_param);
5281     }
5282     if(src_param)
5283     {
5284         for(i = 0; i < ins->src_count; i++)
5285         {
5286             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5287         }
5288         HeapFree(GetProcessHeap(), 0, src_param);
5289     }
5290     HeapFree(GetProcessHeap(), 0, rec);
5291 }
5292
5293 static void free_recorded_instruction(struct list *list)
5294 {
5295     struct recorded_instruction *rec_ins, *entry2;
5296     unsigned int i;
5297
5298     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5299     {
5300         list_remove(&rec_ins->entry);
5301         if(rec_ins->ins.dst)
5302         {
5303             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5304             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5305         }
5306         if(rec_ins->ins.src)
5307         {
5308             for(i = 0; i < rec_ins->ins.src_count; i++)
5309             {
5310                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5311             }
5312             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5313         }
5314         HeapFree(GetProcessHeap(), 0, rec_ins);
5315     }
5316 }
5317
5318 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5319     SHADER_HANDLER hw_fct;
5320     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5321     const struct wined3d_shader *shader = ins->ctx->shader;
5322     struct control_frame *control_frame;
5323     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5324     BOOL bool_const;
5325
5326     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5327     {
5328         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5329         list_add_head(&priv->control_frames, &control_frame->entry);
5330
5331         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5332         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5333
5334         if(priv->target_version >= NV2)
5335         {
5336             control_frame->no.loop = priv->num_loops++;
5337             priv->loop_depth++;
5338         }
5339         else
5340         {
5341             /* Don't bother recording when we're in a not used if branch */
5342             if(priv->muted)
5343             {
5344                 return;
5345             }
5346
5347             if(!priv->recording)
5348             {
5349                 list_init(&priv->record);
5350                 priv->recording = TRUE;
5351                 control_frame->outer_loop = TRUE;
5352                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5353                 return; /* Instruction is handled */
5354             }
5355             /* Record this loop in the outer loop's recording */
5356         }
5357     }
5358     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5359     {
5360         if(priv->target_version >= NV2)
5361         {
5362             /* Nothing to do. The control frame is popped after the HW instr handler */
5363         }
5364         else
5365         {
5366             struct list *e = list_head(&priv->control_frames);
5367             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5368             list_remove(&control_frame->entry);
5369
5370             if(control_frame->outer_loop)
5371             {
5372                 unsigned int iteration;
5373                 int aL = 0;
5374                 struct list copy;
5375
5376                 /* Turn off recording before playback */
5377                 priv->recording = FALSE;
5378
5379                 /* Move the recorded instructions to a separate list and get them out of the private data
5380                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5381                  * be recorded again, thus priv->record might be overwritten
5382                  */
5383                 list_init(&copy);
5384                 list_move_tail(&copy, &priv->record);
5385                 list_init(&priv->record);
5386
5387                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5388                 {
5389                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5390                                    control_frame->loop_control.count, control_frame->loop_control.start,
5391                                    control_frame->loop_control.step);
5392                     aL = control_frame->loop_control.start;
5393                 }
5394                 else
5395                 {
5396                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5397                 }
5398
5399                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5400                 {
5401                     struct recorded_instruction *rec_ins;
5402                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5403                     {
5404                         priv->aL = aL;
5405                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5406                     }
5407                     else
5408                     {
5409                         shader_addline(buffer, "#Iteration %u\n", iteration);
5410                     }
5411
5412                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5413                     {
5414                         shader_arb_handle_instruction(&rec_ins->ins);
5415                     }
5416
5417                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5418                     {
5419                         aL += control_frame->loop_control.step;
5420                     }
5421                 }
5422                 shader_addline(buffer, "#end loop/rep\n");
5423
5424                 free_recorded_instruction(&copy);
5425                 HeapFree(GetProcessHeap(), 0, control_frame);
5426                 return; /* Instruction is handled */
5427             }
5428             else
5429             {
5430                 /* This is a nested loop. Proceed to the normal recording function */
5431                 HeapFree(GetProcessHeap(), 0, control_frame);
5432             }
5433         }
5434     }
5435
5436     if(priv->recording)
5437     {
5438         record_instruction(&priv->record, ins);
5439         return;
5440     }
5441
5442     /* boolean if */
5443     if(ins->handler_idx == WINED3DSIH_IF)
5444     {
5445         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5446         list_add_head(&priv->control_frames, &control_frame->entry);
5447         control_frame->type = IF;
5448
5449         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5450         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5451         if (!priv->muted && !bool_const)
5452         {
5453             shader_addline(buffer, "#if(FALSE){\n");
5454             priv->muted = TRUE;
5455             control_frame->muting = TRUE;
5456         }
5457         else shader_addline(buffer, "#if(TRUE) {\n");
5458
5459         return; /* Instruction is handled */
5460     }
5461     else if(ins->handler_idx == WINED3DSIH_IFC)
5462     {
5463         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5464         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5465         control_frame->type = IFC;
5466         control_frame->no.ifc = priv->num_ifcs++;
5467         list_add_head(&priv->control_frames, &control_frame->entry);
5468     }
5469     else if(ins->handler_idx == WINED3DSIH_ELSE)
5470     {
5471         struct list *e = list_head(&priv->control_frames);
5472         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5473
5474         if(control_frame->type == IF)
5475         {
5476             shader_addline(buffer, "#} else {\n");
5477             if(!priv->muted && !control_frame->muting)
5478             {
5479                 priv->muted = TRUE;
5480                 control_frame->muting = TRUE;
5481             }
5482             else if(control_frame->muting) priv->muted = FALSE;
5483             return; /* Instruction is handled. */
5484         }
5485         /* In case of an ifc, generate a HW shader instruction */
5486     }
5487     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5488     {
5489         struct list *e = list_head(&priv->control_frames);
5490         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5491
5492         if(control_frame->type == IF)
5493         {
5494             shader_addline(buffer, "#} endif\n");
5495             if(control_frame->muting) priv->muted = FALSE;
5496             list_remove(&control_frame->entry);
5497             HeapFree(GetProcessHeap(), 0, control_frame);
5498             return; /* Instruction is handled */
5499         }
5500     }
5501
5502     if(priv->muted) return;
5503
5504     /* Select handler */
5505     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5506
5507     /* Unhandled opcode */
5508     if (!hw_fct)
5509     {
5510         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5511         return;
5512     }
5513     hw_fct(ins);
5514
5515     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5516     {
5517         struct list *e = list_head(&priv->control_frames);
5518         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5519         list_remove(&control_frame->entry);
5520         HeapFree(GetProcessHeap(), 0, control_frame);
5521         priv->loop_depth--;
5522     }
5523     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5524     {
5525         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5526         struct list *e = list_head(&priv->control_frames);
5527         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5528         list_remove(&control_frame->entry);
5529         HeapFree(GetProcessHeap(), 0, control_frame);
5530     }
5531
5532
5533     shader_arb_add_instruction_modifiers(ins);
5534 }
5535
5536 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5537 {
5538     shader_arb_handle_instruction,
5539     shader_arb_select,
5540     shader_arb_select_depth_blt,
5541     shader_arb_deselect_depth_blt,
5542     shader_arb_update_float_vertex_constants,
5543     shader_arb_update_float_pixel_constants,
5544     shader_arb_load_constants,
5545     shader_arb_load_np2fixup_constants,
5546     shader_arb_destroy,
5547     shader_arb_alloc,
5548     shader_arb_free,
5549     shader_arb_context_destroyed,
5550     shader_arb_get_caps,
5551     shader_arb_color_fixup_supported,
5552 };
5553
5554 /* ARB_fragment_program fixed function pipeline replacement definitions */
5555 #define ARB_FFP_CONST_TFACTOR           0
5556 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5557 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5558 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5559 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5560
5561 struct arbfp_ffp_desc
5562 {
5563     struct ffp_frag_desc parent;
5564     GLuint shader;
5565     unsigned int num_textures_used;
5566 };
5567
5568 /* Context activation and GL locking are done by the caller. */
5569 static void arbfp_enable(BOOL enable)
5570 {
5571     if(enable) {
5572         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5573         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5574     } else {
5575         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5576         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5577     }
5578 }
5579
5580 static HRESULT arbfp_alloc(struct wined3d_device *device)
5581 {
5582     struct shader_arb_priv *priv;
5583     /* Share private data between the shader backend and the pipeline replacement, if both
5584      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5585      * if no pixel shader is bound or not
5586      */
5587     if (device->shader_backend == &arb_program_shader_backend)
5588     {
5589         device->fragment_priv = device->shader_priv;
5590     }
5591     else
5592     {
5593         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5594         if (!device->fragment_priv) return E_OUTOFMEMORY;
5595     }
5596     priv = device->fragment_priv;
5597     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5598     {
5599         ERR("Failed to initialize rbtree.\n");
5600         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5601         return E_OUTOFMEMORY;
5602     }
5603     priv->use_arbfp_fixed_func = TRUE;
5604     return WINED3D_OK;
5605 }
5606
5607 /* Context activation is done by the caller. */
5608 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5609 {
5610     const struct wined3d_gl_info *gl_info = context;
5611     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5612
5613     ENTER_GL();
5614     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5615     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5616     HeapFree(GetProcessHeap(), 0, entry_arb);
5617     LEAVE_GL();
5618 }
5619
5620 /* Context activation is done by the caller. */
5621 static void arbfp_free(struct wined3d_device *device)
5622 {
5623     struct shader_arb_priv *priv = device->fragment_priv;
5624
5625     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5626     priv->use_arbfp_fixed_func = FALSE;
5627
5628     if (device->shader_backend != &arb_program_shader_backend)
5629     {
5630         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5631     }
5632 }
5633
5634 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5635 {
5636     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5637     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5638                            WINED3DTEXOPCAPS_SELECTARG1                  |
5639                            WINED3DTEXOPCAPS_SELECTARG2                  |
5640                            WINED3DTEXOPCAPS_MODULATE4X                  |
5641                            WINED3DTEXOPCAPS_MODULATE2X                  |
5642                            WINED3DTEXOPCAPS_MODULATE                    |
5643                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5644                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5645                            WINED3DTEXOPCAPS_ADD                         |
5646                            WINED3DTEXOPCAPS_SUBTRACT                    |
5647                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5648                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5649                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5650                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5651                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5652                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5653                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5654                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5655                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5656                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5657                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5658                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5659                            WINED3DTEXOPCAPS_LERP                        |
5660                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5661                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5662
5663     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5664
5665     caps->MaxTextureBlendStages   = 8;
5666     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5667 }
5668
5669 static void state_texfactor_arbfp(struct wined3d_context *context,
5670         const struct wined3d_state *state, DWORD state_id)
5671 {
5672     struct wined3d_device *device = context->swapchain->device;
5673     const struct wined3d_gl_info *gl_info = context->gl_info;
5674     float col[4];
5675
5676     /* Don't load the parameter if we're using an arbfp pixel shader,
5677      * otherwise we'll overwrite application provided constants. */
5678     if (device->shader_backend == &arb_program_shader_backend)
5679     {
5680         struct shader_arb_priv *priv;
5681
5682         if (use_ps(state)) return;
5683
5684         priv = device->shader_priv;
5685         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5686         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5687     }
5688
5689     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5690     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5691     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5692 }
5693
5694 static void state_arb_specularenable(struct wined3d_context *context,
5695         const struct wined3d_state *state, DWORD state_id)
5696 {
5697     struct wined3d_device *device = context->swapchain->device;
5698     const struct wined3d_gl_info *gl_info = context->gl_info;
5699     float col[4];
5700
5701     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5702      * application provided constants
5703      */
5704     if (device->shader_backend == &arb_program_shader_backend)
5705     {
5706         struct shader_arb_priv *priv;
5707
5708         if (use_ps(state)) return;
5709
5710         priv = device->shader_priv;
5711         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5712         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5713     }
5714
5715     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5716     {
5717         /* The specular color has no alpha */
5718         col[0] = 1.0f; col[1] = 1.0f;
5719         col[2] = 1.0f; col[3] = 0.0f;
5720     } else {
5721         col[0] = 0.0f; col[1] = 0.0f;
5722         col[2] = 0.0f; col[3] = 0.0f;
5723     }
5724     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5725     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5726 }
5727
5728 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5729 {
5730     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5731     struct wined3d_device *device = context->swapchain->device;
5732     const struct wined3d_gl_info *gl_info = context->gl_info;
5733     float mat[2][2];
5734
5735     if (use_ps(state))
5736     {
5737         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5738         {
5739             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5740              * anyway
5741              */
5742             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5743                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5744         }
5745
5746         if(device->shader_backend == &arb_program_shader_backend) {
5747             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5748             return;
5749         }
5750     }
5751     else if (device->shader_backend == &arb_program_shader_backend)
5752     {
5753         struct shader_arb_priv *priv = device->shader_priv;
5754         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5755         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5756     }
5757
5758     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5759     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5760     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5761     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5762
5763     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5764     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5765 }
5766
5767 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5768         const struct wined3d_state *state, DWORD state_id)
5769 {
5770     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5771     struct wined3d_device *device = context->swapchain->device;
5772     const struct wined3d_gl_info *gl_info = context->gl_info;
5773     float param[4];
5774
5775     if (use_ps(state))
5776     {
5777         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5778         {
5779             /* The pixel shader has to know the luminance offset. Do a constants update if it
5780              * isn't scheduled anyway
5781              */
5782             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5783                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5784         }
5785
5786         if(device->shader_backend == &arb_program_shader_backend) {
5787             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5788             return;
5789         }
5790     }
5791     else if (device->shader_backend == &arb_program_shader_backend)
5792     {
5793         struct shader_arb_priv *priv = device->shader_priv;
5794         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5795         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5796     }
5797
5798     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5799     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5800     param[2] = 0.0f;
5801     param[3] = 0.0f;
5802
5803     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5804     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5805 }
5806
5807 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5808 {
5809     const char *ret;
5810
5811     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5812
5813     switch(arg & WINED3DTA_SELECTMASK) {
5814         case WINED3DTA_DIFFUSE:
5815             ret = "fragment.color.primary"; break;
5816
5817         case WINED3DTA_CURRENT:
5818             if (!stage) ret = "fragment.color.primary";
5819             else ret = "ret";
5820             break;
5821
5822         case WINED3DTA_TEXTURE:
5823             switch(stage) {
5824                 case 0: ret = "tex0"; break;
5825                 case 1: ret = "tex1"; break;
5826                 case 2: ret = "tex2"; break;
5827                 case 3: ret = "tex3"; break;
5828                 case 4: ret = "tex4"; break;
5829                 case 5: ret = "tex5"; break;
5830                 case 6: ret = "tex6"; break;
5831                 case 7: ret = "tex7"; break;
5832                 default: ret = "unknown texture";
5833             }
5834             break;
5835
5836         case WINED3DTA_TFACTOR:
5837             ret = "tfactor"; break;
5838
5839         case WINED3DTA_SPECULAR:
5840             ret = "fragment.color.secondary"; break;
5841
5842         case WINED3DTA_TEMP:
5843             ret = "tempreg"; break;
5844
5845         case WINED3DTA_CONSTANT:
5846             FIXME("Implement perstage constants\n");
5847             switch(stage) {
5848                 case 0: ret = "const0"; break;
5849                 case 1: ret = "const1"; break;
5850                 case 2: ret = "const2"; break;
5851                 case 3: ret = "const3"; break;
5852                 case 4: ret = "const4"; break;
5853                 case 5: ret = "const5"; break;
5854                 case 6: ret = "const6"; break;
5855                 case 7: ret = "const7"; break;
5856                 default: ret = "unknown constant";
5857             }
5858             break;
5859
5860         default:
5861             return "unknown";
5862     }
5863
5864     if(arg & WINED3DTA_COMPLEMENT) {
5865         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5866         if(argnum == 0) ret = "arg0";
5867         if(argnum == 1) ret = "arg1";
5868         if(argnum == 2) ret = "arg2";
5869     }
5870     if(arg & WINED3DTA_ALPHAREPLICATE) {
5871         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5872         if(argnum == 0) ret = "arg0";
5873         if(argnum == 1) ret = "arg1";
5874         if(argnum == 2) ret = "arg2";
5875     }
5876     return ret;
5877 }
5878
5879 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5880         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5881 {
5882     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5883     unsigned int mul = 1;
5884     BOOL mul_final_dest = FALSE;
5885
5886     if(color && alpha) dstmask = "";
5887     else if(color) dstmask = ".xyz";
5888     else dstmask = ".w";
5889
5890     if(dst == tempreg) dstreg = "tempreg";
5891     else dstreg = "ret";
5892
5893     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5894     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5895     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5896
5897     switch (op)
5898     {
5899         case WINED3D_TOP_DISABLE:
5900             if (!stage)
5901                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5902             break;
5903
5904         case WINED3D_TOP_SELECT_ARG2:
5905             arg1 = arg2;
5906             /* FALLTHROUGH */
5907         case WINED3D_TOP_SELECT_ARG1:
5908             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5909             break;
5910
5911         case WINED3D_TOP_MODULATE_4X:
5912             mul = 2;
5913             /* FALLTHROUGH */
5914         case WINED3D_TOP_MODULATE_2X:
5915             mul *= 2;
5916             if (!strcmp(dstreg, "result.color"))
5917             {
5918                 dstreg = "ret";
5919                 mul_final_dest = TRUE;
5920             }
5921             /* FALLTHROUGH */
5922         case WINED3D_TOP_MODULATE:
5923             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5924             break;
5925
5926         case WINED3D_TOP_ADD_SIGNED_2X:
5927             mul = 2;
5928             if (!strcmp(dstreg, "result.color"))
5929             {
5930                 dstreg = "ret";
5931                 mul_final_dest = TRUE;
5932             }
5933             /* FALLTHROUGH */
5934         case WINED3D_TOP_ADD_SIGNED:
5935             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5936             arg2 = "arg2";
5937             /* FALLTHROUGH */
5938         case WINED3D_TOP_ADD:
5939             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5940             break;
5941
5942         case WINED3D_TOP_SUBTRACT:
5943             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5944             break;
5945
5946         case WINED3D_TOP_ADD_SMOOTH:
5947             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5948             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5949             break;
5950
5951         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
5952             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5953             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5954             break;
5955         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
5956             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5957             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5958             break;
5959         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
5960             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5961             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5962             break;
5963         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
5964             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5965             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5966             break;
5967
5968         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
5969             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5970             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5971             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5972             break;
5973
5974         /* D3DTOP_PREMODULATE ???? */
5975
5976         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
5977             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5978             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5979             break;
5980         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
5981             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5982             break;
5983         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
5984             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5985             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5986             break;
5987         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
5988             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5989             break;
5990
5991         case WINED3D_TOP_DOTPRODUCT3:
5992             mul = 4;
5993             if (!strcmp(dstreg, "result.color"))
5994             {
5995                 dstreg = "ret";
5996                 mul_final_dest = TRUE;
5997             }
5998             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5999             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6000             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6001             break;
6002
6003         case WINED3D_TOP_MULTIPLY_ADD:
6004             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6005             break;
6006
6007         case WINED3D_TOP_LERP:
6008             /* The msdn is not quite right here */
6009             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6010             break;
6011
6012         case WINED3D_TOP_BUMPENVMAP:
6013         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6014             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6015             break;
6016
6017         default:
6018             FIXME("Unhandled texture op %08x\n", op);
6019     }
6020
6021     if(mul == 2) {
6022         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6023     } else if(mul == 4) {
6024         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6025     }
6026 }
6027
6028 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6029 {
6030     unsigned int stage;
6031     struct wined3d_shader_buffer buffer;
6032     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6033     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6034     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6035     UINT lowest_disabled_stage;
6036     const char *textype;
6037     const char *instr, *sat;
6038     char colorcor_dst[8];
6039     GLuint ret;
6040     DWORD arg0, arg1, arg2;
6041     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6042     BOOL op_equal;
6043     const char *final_combiner_src = "ret";
6044     GLint pos;
6045
6046     /* Find out which textures are read */
6047     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6048     {
6049         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6050             break;
6051         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6052         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6053         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6054         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6055         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6056         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6057
6058         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6059             tex_read[stage] = TRUE;
6060         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6061             tex_read[stage] = TRUE;
6062         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6063         {
6064             bump_used[stage] = TRUE;
6065             tex_read[stage] = TRUE;
6066         }
6067         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6068         {
6069             bump_used[stage] = TRUE;
6070             tex_read[stage] = TRUE;
6071             luminance_used[stage] = TRUE;
6072         }
6073         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6074         {
6075             tfactor_used = TRUE;
6076         }
6077
6078         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6079             tfactor_used = TRUE;
6080         }
6081
6082         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6083         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6084             tempreg_used = TRUE;
6085         }
6086
6087         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6088             continue;
6089         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6090         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6091         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6092         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6093         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6094         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6095
6096         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6097             tempreg_used = TRUE;
6098         }
6099         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6100             tfactor_used = TRUE;
6101         }
6102     }
6103     lowest_disabled_stage = stage;
6104
6105     /* Shader header */
6106     if (!shader_buffer_init(&buffer))
6107     {
6108         ERR("Failed to initialize shader buffer.\n");
6109         return 0;
6110     }
6111
6112     shader_addline(&buffer, "!!ARBfp1.0\n");
6113
6114     switch(settings->fog) {
6115         case FOG_OFF:                                                         break;
6116         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6117         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6118         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6119         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6120     }
6121
6122     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6123     shader_addline(&buffer, "TEMP TMP;\n");
6124     shader_addline(&buffer, "TEMP ret;\n");
6125     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6126     shader_addline(&buffer, "TEMP arg0;\n");
6127     shader_addline(&buffer, "TEMP arg1;\n");
6128     shader_addline(&buffer, "TEMP arg2;\n");
6129     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6130         if(!tex_read[stage]) continue;
6131         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6132         if(!bump_used[stage]) continue;
6133         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6134         if(!luminance_used[stage]) continue;
6135         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6136     }
6137     if(tfactor_used) {
6138         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6139     }
6140         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6141
6142     if(settings->sRGB_write) {
6143         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6144                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6145         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6146                        srgb_sub_high, 0.0, 0.0, 0.0);
6147     }
6148
6149     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6150         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6151
6152     /* Generate texture sampling instructions) */
6153     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6154     {
6155         if (!tex_read[stage])
6156             continue;
6157
6158         switch(settings->op[stage].tex_type) {
6159             case tex_1d:                    textype = "1D";     break;
6160             case tex_2d:                    textype = "2D";     break;
6161             case tex_3d:                    textype = "3D";     break;
6162             case tex_cube:                  textype = "CUBE";   break;
6163             case tex_rect:                  textype = "RECT";   break;
6164             default: textype = "unexpected_textype";   break;
6165         }
6166
6167         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6168                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6169             sat = "";
6170         else
6171             sat = "_SAT";
6172
6173         if(settings->op[stage].projected == proj_none) {
6174             instr = "TEX";
6175         } else if(settings->op[stage].projected == proj_count4 ||
6176                   settings->op[stage].projected == proj_count3) {
6177             instr = "TXP";
6178         } else {
6179             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6180             instr = "TXP";
6181         }
6182
6183         if (stage > 0
6184                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6185                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6186         {
6187             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6188             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6189             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6190             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6191
6192             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6193              * so multiply the displacement with the dividing parameter before passing it to TXP
6194              */
6195             if (settings->op[stage].projected != proj_none) {
6196                 if(settings->op[stage].projected == proj_count4) {
6197                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6198                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6199                 } else {
6200                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6201                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6202                 }
6203             } else {
6204                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6205             }
6206
6207             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6208                     instr, sat, stage, stage, textype);
6209             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6210             {
6211                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6212                                stage - 1, stage - 1, stage - 1);
6213                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6214             }
6215         } else if(settings->op[stage].projected == proj_count3) {
6216             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6217             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6218             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6219                             instr, sat, stage, stage, textype);
6220         } else {
6221             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6222                             instr, sat, stage, stage, stage, textype);
6223         }
6224
6225         sprintf(colorcor_dst, "tex%u", stage);
6226         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6227                 settings->op[stage].color_fixup);
6228     }
6229
6230     /* Generate the main shader */
6231     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6232     {
6233         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6234         {
6235             if (!stage)
6236                 final_combiner_src = "fragment.color.primary";
6237             break;
6238         }
6239
6240         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6241                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6242             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6243         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6244                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6245             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6246         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6247                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6248             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6249         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6250                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6251             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6252         else
6253             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6254                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6255                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6256                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6257
6258         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6259         {
6260             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6261                           settings->op[stage].cop, settings->op[stage].carg0,
6262                           settings->op[stage].carg1, settings->op[stage].carg2);
6263             if (!stage)
6264                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6265         }
6266         else if (op_equal)
6267         {
6268             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6269                           settings->op[stage].cop, settings->op[stage].carg0,
6270                           settings->op[stage].carg1, settings->op[stage].carg2);
6271         } else {
6272             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6273                           settings->op[stage].cop, settings->op[stage].carg0,
6274                           settings->op[stage].carg1, settings->op[stage].carg2);
6275             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6276                           settings->op[stage].aop, settings->op[stage].aarg0,
6277                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6278         }
6279     }
6280
6281     if(settings->sRGB_write) {
6282         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6283         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6284         shader_addline(&buffer, "MOV result.color, ret;\n");
6285     } else {
6286         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6287     }
6288
6289     /* Footer */
6290     shader_addline(&buffer, "END\n");
6291
6292     /* Generate the shader */
6293     GL_EXTCALL(glGenProgramsARB(1, &ret));
6294     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6295     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6296             strlen(buffer.buffer), buffer.buffer));
6297     checkGLcall("glProgramStringARB()");
6298
6299     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6300     if (pos != -1)
6301     {
6302         FIXME("Fragment program error at position %d: %s\n\n", pos,
6303               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6304         shader_arb_dump_program_source(buffer.buffer);
6305     }
6306     else
6307     {
6308         GLint native;
6309
6310         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6311         checkGLcall("glGetProgramivARB()");
6312         if (!native) WARN("Program exceeds native resource limits.\n");
6313     }
6314
6315     shader_buffer_free(&buffer);
6316     return ret;
6317 }
6318
6319 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6320 {
6321     const struct wined3d_device *device = context->swapchain->device;
6322     const struct wined3d_gl_info *gl_info = context->gl_info;
6323     struct shader_arb_priv *priv = device->fragment_priv;
6324     BOOL use_vshader = use_vs(state);
6325     BOOL use_pshader = use_ps(state);
6326     struct ffp_frag_settings settings;
6327     const struct arbfp_ffp_desc *desc;
6328     unsigned int i;
6329
6330     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6331
6332     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6333     {
6334         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6335         {
6336             /* Reload fixed function constants since they collide with the
6337              * pixel shader constants. */
6338             for (i = 0; i < MAX_TEXTURES; ++i)
6339             {
6340                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6341             }
6342             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6343             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6344         }
6345         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6346         {
6347             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6348         }
6349         return;
6350     }
6351
6352     if (!use_pshader)
6353     {
6354         /* Find or create a shader implementing the fixed function pipeline
6355          * settings, then activate it. */
6356         gen_ffp_frag_op(device, state, &settings, FALSE);
6357         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6358         if(!desc) {
6359             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6360             if (!new_desc)
6361             {
6362                 ERR("Out of memory\n");
6363                 return;
6364             }
6365             new_desc->num_textures_used = 0;
6366             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6367             {
6368                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6369                     break;
6370                 new_desc->num_textures_used = i;
6371             }
6372
6373             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6374             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6375             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6376             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6377             desc = new_desc;
6378         }
6379
6380         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6381          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6382          * deactivate it.
6383          */
6384         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6385         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6386         priv->current_fprogram_id = desc->shader;
6387
6388         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6389         {
6390             /* Reload fixed function constants since they collide with the
6391              * pixel shader constants. */
6392             for (i = 0; i < MAX_TEXTURES; ++i)
6393             {
6394                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6395             }
6396             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6397             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6398         }
6399         context->last_was_pshader = FALSE;
6400     } else {
6401         context->last_was_pshader = TRUE;
6402     }
6403
6404     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6405      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6406      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6407      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6408      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6409      *
6410      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6411      * shader handler
6412      */
6413     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6414     {
6415         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6416
6417         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6418             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6419     }
6420     if (use_pshader)
6421         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6422 }
6423
6424 /* We can't link the fog states to the fragment state directly since the
6425  * vertex pipeline links them to FOGENABLE. A different linking in different
6426  * pipeline parts can't be expressed in the combined state table, so we need
6427  * to handle that with a forwarding function. The other invisible side effect
6428  * is that changing the fog start and fog end (which links to FOGENABLE in
6429  * vertex) results in the fragment_prog_arbfp function being called because
6430  * FOGENABLE is dirty, which calls this function here. */
6431 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6432 {
6433     enum fogsource new_source;
6434
6435     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6436
6437     if (!isStateDirty(context, STATE_PIXELSHADER))
6438         fragment_prog_arbfp(context, state, state_id);
6439
6440     if (!state->render_states[WINED3D_RS_FOGENABLE])
6441         return;
6442
6443     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6444     {
6445         if (use_vs(state))
6446         {
6447             new_source = FOGSOURCE_VS;
6448         }
6449         else
6450         {
6451             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6452                 new_source = FOGSOURCE_COORD;
6453             else
6454                 new_source = FOGSOURCE_FFP;
6455         }
6456     }
6457     else
6458     {
6459         new_source = FOGSOURCE_FFP;
6460     }
6461
6462     if (new_source != context->fog_source)
6463     {
6464         context->fog_source = new_source;
6465         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6466     }
6467 }
6468
6469 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6470 {
6471     if (!isStateDirty(context, STATE_PIXELSHADER))
6472         fragment_prog_arbfp(context, state, state_id);
6473 }
6474
6475 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6476 {
6477     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6478     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6479     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6480     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6481     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6482     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6483     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6484     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6485     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6486     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6487     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6488     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6489     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6490     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6491     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6492     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6493     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6494     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6495     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6496     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6497     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6498     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6499     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6502     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6503     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6505     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6599     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6600     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6602     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6603     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6606     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6607     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6608     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6609     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6610     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6611     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6612     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6613     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6615     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6616     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6617 };
6618
6619 const struct fragment_pipeline arbfp_fragment_pipeline = {
6620     arbfp_enable,
6621     arbfp_get_caps,
6622     arbfp_alloc,
6623     arbfp_free,
6624     shader_arb_color_fixup_supported,
6625     arbfp_fragmentstate_template,
6626     TRUE /* We can disable projected textures */
6627 };
6628
6629 struct arbfp_blit_priv {
6630     GLenum yuy2_rect_shader, yuy2_2d_shader;
6631     GLenum uyvy_rect_shader, uyvy_2d_shader;
6632     GLenum yv12_rect_shader, yv12_2d_shader;
6633     GLenum p8_rect_shader, p8_2d_shader;
6634     GLuint palette_texture;
6635 };
6636
6637 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6638 {
6639     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6640     if(!device->blit_priv) {
6641         ERR("Out of memory\n");
6642         return E_OUTOFMEMORY;
6643     }
6644     return WINED3D_OK;
6645 }
6646
6647 /* Context activation is done by the caller. */
6648 static void arbfp_blit_free(struct wined3d_device *device)
6649 {
6650     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6651     struct arbfp_blit_priv *priv = device->blit_priv;
6652
6653     ENTER_GL();
6654     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6655     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6656     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6657     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6658     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6659     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6660     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6661     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6662     checkGLcall("Delete yuv and p8 programs");
6663
6664     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6665     LEAVE_GL();
6666
6667     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6668     device->blit_priv = NULL;
6669 }
6670
6671 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6672         GLenum textype, char *luminance)
6673 {
6674     char chroma;
6675     const char *tex, *texinstr;
6676
6677     if (fixup == COMPLEX_FIXUP_UYVY) {
6678         chroma = 'x';
6679         *luminance = 'w';
6680     } else {
6681         chroma = 'w';
6682         *luminance = 'x';
6683     }
6684     switch(textype) {
6685         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6686         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6687         default:
6688             /* This is more tricky than just replacing the texture type - we have to navigate
6689              * properly in the texture to find the correct chroma values
6690              */
6691             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6692             return FALSE;
6693     }
6694
6695     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6696      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6697      * filtering when we sample the texture.
6698      *
6699      * These are the rules for reading the chroma:
6700      *
6701      * Even pixel: Cr
6702      * Even pixel: U
6703      * Odd pixel: V
6704      *
6705      * So we have to get the sampling x position in non-normalized coordinates in integers
6706      */
6707     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6708         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6709         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6710     } else {
6711         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6712     }
6713     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6714      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6715      * 0.5, so add 0.5.
6716      */
6717     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6718     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6719
6720     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6721      * even and odd pixels respectively
6722      */
6723     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6724     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6725
6726     /* Sample Pixel 1 */
6727     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6728
6729     /* Put the value into either of the chroma values */
6730     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6731     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6732     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6733     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6734
6735     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6736      * the pixel right to the current one. Otherwise, sample the left pixel.
6737      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6738      */
6739     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6740     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6741     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6742
6743     /* Put the value into the other chroma */
6744     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6745     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6746     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6747     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6748
6749     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6750      * the current one and lerp the two U and V values
6751      */
6752
6753     /* This gives the correctly filtered luminance value */
6754     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6755
6756     return TRUE;
6757 }
6758
6759 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6760 {
6761     const char *tex;
6762
6763     switch(textype) {
6764         case GL_TEXTURE_2D:             tex = "2D";     break;
6765         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6766         default:
6767             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6768             return FALSE;
6769     }
6770
6771     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6772      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6773      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6774      * pitch of the luminance plane, the packing into the gl texture is a bit
6775      * unfortunate. If the whole texture is interpreted as luminance data it looks
6776      * approximately like this:
6777      *
6778      *        +----------------------------------+----
6779      *        |                                  |
6780      *        |                                  |
6781      *        |                                  |
6782      *        |                                  |
6783      *        |                                  |   2
6784      *        |            LUMINANCE             |   -
6785      *        |                                  |   3
6786      *        |                                  |
6787      *        |                                  |
6788      *        |                                  |
6789      *        |                                  |
6790      *        +----------------+-----------------+----
6791      *        |                |                 |
6792      *        |  U even rows   |  U odd rows     |
6793      *        |                |                 |   1
6794      *        +----------------+------------------   -
6795      *        |                |                 |   3
6796      *        |  V even rows   |  V odd rows     |
6797      *        |                |                 |
6798      *        +----------------+-----------------+----
6799      *        |                |                 |
6800      *        |     0.5        |       0.5       |
6801      *
6802      * So it appears as if there are 4 chroma images, but in fact the odd rows
6803      * in the chroma images are in the same row as the even ones. So its is
6804      * kinda tricky to read
6805      *
6806      * When reading from rectangle textures, keep in mind that the input y coordinates
6807      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6808      */
6809     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6810             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6811
6812     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6813     /* the chroma planes have only half the width */
6814     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6815
6816     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6817      * the coordinate. Also read the right side of the image when reading odd lines
6818      *
6819      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6820      * bleeding
6821      */
6822     if(textype == GL_TEXTURE_2D) {
6823
6824         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6825
6826         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6827
6828         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6829         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6830
6831         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6832         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6833         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6834         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6835         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6836
6837         /* clamp, keep the half pixel origin in mind */
6838         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6839         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6840         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6841         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6842     } else {
6843         /* Read from [size - size+size/4] */
6844         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6845         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6846
6847         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6848         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6849         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6850         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6851         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6852         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6853
6854         /* Make sure to read exactly from the pixel center */
6855         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6856         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6857
6858         /* Clamp */
6859         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6860         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6861         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6862         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6863         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6864     }
6865     /* Read the texture, put the result into the output register */
6866     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6867     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6868
6869     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6870      * No need to clamp because we're just reusing the already clamped value from above
6871      */
6872     if(textype == GL_TEXTURE_2D) {
6873         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6874     } else {
6875         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6876     }
6877     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6878     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6879
6880     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6881      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6882      * values due to filtering
6883      */
6884     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6885     if(textype == GL_TEXTURE_2D) {
6886         /* Multiply the y coordinate by 2/3 and clamp it */
6887         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6888         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6889         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6890         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6891     } else {
6892         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6893          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6894          * is bigger
6895          */
6896         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6897         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6898         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6899     }
6900     *luminance = 'a';
6901
6902     return TRUE;
6903 }
6904
6905 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6906         const struct wined3d_gl_info *gl_info, GLenum textype)
6907 {
6908     GLenum shader;
6909     struct wined3d_shader_buffer buffer;
6910     GLint pos;
6911
6912     /* Shader header */
6913     if (!shader_buffer_init(&buffer))
6914     {
6915         ERR("Failed to initialize shader buffer.\n");
6916         return 0;
6917     }
6918
6919     ENTER_GL();
6920     GL_EXTCALL(glGenProgramsARB(1, &shader));
6921     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6922     LEAVE_GL();
6923     if(!shader) {
6924         shader_buffer_free(&buffer);
6925         return 0;
6926     }
6927
6928     shader_addline(&buffer, "!!ARBfp1.0\n");
6929     shader_addline(&buffer, "TEMP index;\n");
6930
6931     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6932     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6933
6934     /* The alpha-component contains the palette index */
6935     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6936         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6937     else
6938         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6939
6940     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6941     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6942
6943     /* Use the alpha-component as an index in the palette to get the final color */
6944     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6945     shader_addline(&buffer, "END\n");
6946
6947     ENTER_GL();
6948     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6949             strlen(buffer.buffer), buffer.buffer));
6950     checkGLcall("glProgramStringARB()");
6951
6952     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6953     if (pos != -1)
6954     {
6955         FIXME("Fragment program error at position %d: %s\n\n", pos,
6956               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6957         shader_arb_dump_program_source(buffer.buffer);
6958     }
6959
6960     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6961         priv->p8_rect_shader = shader;
6962     else
6963         priv->p8_2d_shader = shader;
6964
6965     shader_buffer_free(&buffer);
6966     LEAVE_GL();
6967
6968     return shader;
6969 }
6970
6971 /* Context activation is done by the caller. */
6972 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
6973 {
6974     BYTE table[256][4];
6975     struct wined3d_device *device = surface->resource.device;
6976     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6977     struct arbfp_blit_priv *priv = device->blit_priv;
6978     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6979
6980     d3dfmt_p8_init_palette(surface, table, colorkey);
6981
6982     ENTER_GL();
6983
6984     if (gl_info->supported[APPLE_CLIENT_STORAGE])
6985     {
6986         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
6987         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
6988     }
6989
6990     if (!priv->palette_texture)
6991         glGenTextures(1, &priv->palette_texture);
6992
6993     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6994     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6995
6996     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6997
6998     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6999     /* Make sure we have discrete color levels. */
7000     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7001     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7002     /* Upload the palette */
7003     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7004     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7005
7006     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7007     {
7008         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7009         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7010     }
7011
7012     /* Switch back to unit 0 in which the 2D texture will be stored. */
7013     context_active_texture(context, gl_info, 0);
7014     LEAVE_GL();
7015 }
7016
7017 /* Context activation is done by the caller. */
7018 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7019         enum complex_fixup yuv_fixup, GLenum textype)
7020 {
7021     GLenum shader;
7022     struct wined3d_shader_buffer buffer;
7023     char luminance_component;
7024     GLint pos;
7025
7026     /* Shader header */
7027     if (!shader_buffer_init(&buffer))
7028     {
7029         ERR("Failed to initialize shader buffer.\n");
7030         return 0;
7031     }
7032
7033     ENTER_GL();
7034     GL_EXTCALL(glGenProgramsARB(1, &shader));
7035     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7036     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7037     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7038     LEAVE_GL();
7039     if(!shader) {
7040         shader_buffer_free(&buffer);
7041         return 0;
7042     }
7043
7044     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7045      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7046      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7047      * each single pixel it contains, and one U and one V value shared between both
7048      * pixels.
7049      *
7050      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7051      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7052      * take the format into account when generating the read swizzles
7053      *
7054      * Reading the Y value is straightforward - just sample the texture. The hardware
7055      * takes care of filtering in the horizontal and vertical direction.
7056      *
7057      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7058      * because that would mix the U and V values of one pixel or two adjacent pixels.
7059      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7060      * regardless of the filtering setting. Vertical filtering works automatically
7061      * though - the U and V values of two rows are mixed nicely.
7062      *
7063      * Appart of avoiding filtering issues, the code has to know which value it just
7064      * read, and where it can find the other one. To determine this, it checks if
7065      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7066      *
7067      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7068      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7069      *
7070      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7071      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7072      * in an unfiltered situation. Finding the luminance on the other hand requires
7073      * finding out if it is an odd or even pixel. The real drawback of this approach
7074      * is filtering. This would have to be emulated completely in the shader, reading
7075      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7076      * vertically. Beyond that it would require adjustments to the texture handling
7077      * code to deal with the width scaling
7078      */
7079     shader_addline(&buffer, "!!ARBfp1.0\n");
7080     shader_addline(&buffer, "TEMP luminance;\n");
7081     shader_addline(&buffer, "TEMP temp;\n");
7082     shader_addline(&buffer, "TEMP chroma;\n");
7083     shader_addline(&buffer, "TEMP texcrd;\n");
7084     shader_addline(&buffer, "TEMP texcrd2;\n");
7085     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7086     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7087     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7088
7089     switch (yuv_fixup)
7090     {
7091         case COMPLEX_FIXUP_UYVY:
7092         case COMPLEX_FIXUP_YUY2:
7093             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7094             {
7095                 shader_buffer_free(&buffer);
7096                 return 0;
7097             }
7098             break;
7099
7100         case COMPLEX_FIXUP_YV12:
7101             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7102             {
7103                 shader_buffer_free(&buffer);
7104                 return 0;
7105             }
7106             break;
7107
7108         default:
7109             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7110             shader_buffer_free(&buffer);
7111             return 0;
7112     }
7113
7114     /* Calculate the final result. Formula is taken from
7115      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7116      * ranges from -0.5 to 0.5
7117      */
7118     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7119
7120     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7121     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7122     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7123     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7124     shader_addline(&buffer, "END\n");
7125
7126     ENTER_GL();
7127     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7128             strlen(buffer.buffer), buffer.buffer));
7129     checkGLcall("glProgramStringARB()");
7130
7131     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7132     if (pos != -1)
7133     {
7134         FIXME("Fragment program error at position %d: %s\n\n", pos,
7135               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7136         shader_arb_dump_program_source(buffer.buffer);
7137     }
7138     else
7139     {
7140         GLint native;
7141
7142         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7143         checkGLcall("glGetProgramivARB()");
7144         if (!native) WARN("Program exceeds native resource limits.\n");
7145     }
7146
7147     shader_buffer_free(&buffer);
7148     LEAVE_GL();
7149
7150     switch (yuv_fixup)
7151     {
7152         case COMPLEX_FIXUP_YUY2:
7153             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7154             else priv->yuy2_2d_shader = shader;
7155             break;
7156
7157         case COMPLEX_FIXUP_UYVY:
7158             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7159             else priv->uyvy_2d_shader = shader;
7160             break;
7161
7162         case COMPLEX_FIXUP_YV12:
7163             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7164             else priv->yv12_2d_shader = shader;
7165             break;
7166         default:
7167             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7168     }
7169
7170     return shader;
7171 }
7172
7173 /* Context activation is done by the caller. */
7174 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7175 {
7176     GLenum shader;
7177     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7178     struct arbfp_blit_priv *priv = blit_priv;
7179     enum complex_fixup fixup;
7180     GLenum textype = surface->texture_target;
7181     const struct wined3d_gl_info *gl_info = context->gl_info;
7182
7183     if (surface->flags & SFLAG_CONVERTED)
7184     {
7185         ENTER_GL();
7186         glEnable(textype);
7187         checkGLcall("glEnable(textype)");
7188         LEAVE_GL();
7189         return WINED3D_OK;
7190     }
7191
7192     if (!is_complex_fixup(surface->resource.format->color_fixup))
7193     {
7194         TRACE("Fixup:\n");
7195         dump_color_fixup_desc(surface->resource.format->color_fixup);
7196         /* Don't bother setting up a shader for unconverted formats */
7197         ENTER_GL();
7198         glEnable(textype);
7199         checkGLcall("glEnable(textype)");
7200         LEAVE_GL();
7201         return WINED3D_OK;
7202     }
7203
7204     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7205
7206     switch(fixup)
7207     {
7208         case COMPLEX_FIXUP_YUY2:
7209             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7210             break;
7211
7212         case COMPLEX_FIXUP_UYVY:
7213             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7214             break;
7215
7216         case COMPLEX_FIXUP_YV12:
7217             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7218             break;
7219
7220         case COMPLEX_FIXUP_P8:
7221             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7222             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7223
7224             upload_palette(surface, context);
7225             break;
7226
7227         default:
7228             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7229             ENTER_GL();
7230             glEnable(textype);
7231             checkGLcall("glEnable(textype)");
7232             LEAVE_GL();
7233             return E_NOTIMPL;
7234     }
7235
7236     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7237
7238     ENTER_GL();
7239     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7240     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7241     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7242     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7243     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7244     checkGLcall("glProgramLocalParameter4fvARB");
7245     LEAVE_GL();
7246
7247     return WINED3D_OK;
7248 }
7249
7250 /* Context activation is done by the caller. */
7251 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7252 {
7253     ENTER_GL();
7254     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7255     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7256     glDisable(GL_TEXTURE_2D);
7257     checkGLcall("glDisable(GL_TEXTURE_2D)");
7258     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7259     {
7260         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7261         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7262     }
7263     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7264     {
7265         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7266         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7267     }
7268     LEAVE_GL();
7269 }
7270
7271 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7272         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7273         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7274 {
7275     enum complex_fixup src_fixup;
7276
7277     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7278         return FALSE;
7279
7280     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7281     {
7282         TRACE("Unsupported blit_op=%d\n", blit_op);
7283         return FALSE;
7284     }
7285
7286     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7287         return FALSE;
7288
7289     src_fixup = get_complex_fixup(src_format->color_fixup);
7290     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7291     {
7292         TRACE("Checking support for fixup:\n");
7293         dump_color_fixup_desc(src_format->color_fixup);
7294     }
7295
7296     if (!is_identity_fixup(dst_format->color_fixup))
7297     {
7298         TRACE("Destination fixups are not supported\n");
7299         return FALSE;
7300     }
7301
7302     if (is_identity_fixup(src_format->color_fixup))
7303     {
7304         TRACE("[OK]\n");
7305         return TRUE;
7306     }
7307
7308      /* We only support YUV conversions. */
7309     if (!is_complex_fixup(src_format->color_fixup))
7310     {
7311         TRACE("[FAILED]\n");
7312         return FALSE;
7313     }
7314
7315     switch(src_fixup)
7316     {
7317         case COMPLEX_FIXUP_YUY2:
7318         case COMPLEX_FIXUP_UYVY:
7319         case COMPLEX_FIXUP_YV12:
7320         case COMPLEX_FIXUP_P8:
7321             TRACE("[OK]\n");
7322             return TRUE;
7323
7324         default:
7325             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7326             TRACE("[FAILED]\n");
7327             return FALSE;
7328     }
7329 }
7330
7331 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7332         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7333         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7334 {
7335     struct wined3d_context *context;
7336     RECT src_rect = *src_rect_in;
7337     RECT dst_rect = *dst_rect_in;
7338
7339     /* Now load the surface */
7340     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7341             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7342     {
7343         /* Without FBO blits transferring from the drawable to the texture is
7344          * expensive, because we have to flip the data in sysmem. Since we can
7345          * flip in the blitter, we don't actually need that flip anyway. So we
7346          * use the surface's texture as scratch texture, and flip the source
7347          * rectangle instead. */
7348         surface_load_fb_texture(src_surface, FALSE);
7349
7350         src_rect.top = src_surface->resource.height - src_rect.top;
7351         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7352     }
7353     else
7354         surface_internal_preload(src_surface, SRGB_RGB);
7355
7356     /* Activate the destination context, set it up for blitting */
7357     context = context_acquire(device, dst_surface);
7358     context_apply_blit_state(context, device);
7359
7360     if (!surface_is_offscreen(dst_surface))
7361         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7362
7363     arbfp_blit_set(device->blit_priv, context, src_surface);
7364
7365     ENTER_GL();
7366
7367     /* Draw a textured quad */
7368     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7369
7370     LEAVE_GL();
7371
7372     /* Leave the opengl state valid for blitting */
7373     arbfp_blit_unset(context->gl_info);
7374
7375     if (wined3d_settings.strict_draw_ordering
7376             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7377             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7378         wglFlush(); /* Flush to ensure ordering across contexts. */
7379
7380     context_release(context);
7381
7382     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7383     return WINED3D_OK;
7384 }
7385
7386 /* Do not call while under the GL lock. */
7387 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7388         const RECT *dst_rect, const struct wined3d_color *color)
7389 {
7390     FIXME("Color filling not implemented by arbfp_blit\n");
7391     return WINED3DERR_INVALIDCALL;
7392 }
7393
7394 /* Do not call while under the GL lock. */
7395 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7396         struct wined3d_surface *surface, const RECT *rect, float depth)
7397 {
7398     FIXME("Depth filling not implemented by arbfp_blit.\n");
7399     return WINED3DERR_INVALIDCALL;
7400 }
7401
7402 const struct blit_shader arbfp_blit = {
7403     arbfp_blit_alloc,
7404     arbfp_blit_free,
7405     arbfp_blit_set,
7406     arbfp_blit_unset,
7407     arbfp_blit_supported,
7408     arbfp_blit_color_fill,
7409     arbfp_blit_depth_fill,
7410 };