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
1917     /* All versions have a destination register */
1918     shader_arb_get_dst_param(ins, dst, reg_dest);
1919
1920     /* 1.0-1.4: Use destination register number as texture code.
1921        2.0+: Use provided sampler number as texure code. */
1922     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1923         reg_sampler_code = dst->reg.idx;
1924     else
1925         reg_sampler_code = ins->src[1].reg.idx;
1926
1927     /* 1.0-1.3: Use the texcoord varying.
1928        1.4+: Use provided coordinate source register. */
1929     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1930         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1931     else {
1932         /* TEX is the only instruction that can handle DW and DZ natively */
1933         src = ins->src[0];
1934         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1935         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1936         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1937     }
1938
1939     /* projection flag:
1940      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1941      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1942      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1943      */
1944     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1945     {
1946         DWORD flags = 0;
1947         if (reg_sampler_code < MAX_TEXTURES)
1948             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1949         if (flags & WINED3D_PSARGS_PROJECTED)
1950             myflags |= TEX_PROJ;
1951     }
1952     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1953     {
1954         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1955         if (src_mod == WINED3DSPSM_DZ)
1956         {
1957             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1958              * varying register, so we need a temp reg
1959              */
1960             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1961             strcpy(reg_coord, "TA");
1962             myflags |= TEX_PROJ;
1963         } else if(src_mod == WINED3DSPSM_DW) {
1964             myflags |= TEX_PROJ;
1965         }
1966     } else {
1967         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1968         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1969     }
1970     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1971 }
1972
1973 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1974 {
1975     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1976     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1977     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1978             ins->ctx->reg_maps->shader_version.minor);
1979     char dst_str[50];
1980
1981     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1982     {
1983         DWORD reg = dst->reg.idx;
1984
1985         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1986         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1987     } else {
1988         char reg_src[40];
1989
1990         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1991         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1992         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1993    }
1994 }
1995
1996 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1997 {
1998      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1999      DWORD flags = 0;
2000
2001      DWORD reg1 = ins->dst[0].reg.idx;
2002      char dst_str[50];
2003      char src_str[50];
2004
2005      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2006      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2007      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2008      /* Move .x first in case src_str is "TA" */
2009      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2010      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2011      if (reg1 < MAX_TEXTURES)
2012      {
2013          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2014          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2015      }
2016      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2017 }
2018
2019 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2020 {
2021      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2022
2023      DWORD reg1 = ins->dst[0].reg.idx;
2024      char dst_str[50];
2025      char src_str[50];
2026
2027      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2028      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2029      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2030      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2031      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2032      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2033 }
2034
2035 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2036 {
2037     DWORD reg1 = ins->dst[0].reg.idx;
2038     char dst_str[50];
2039     char src_str[50];
2040
2041     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2042     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2043     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2044     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2045 }
2046
2047 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2048 {
2049     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2050     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2051     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2052     char reg_coord[40], dst_reg[50], src_reg[50];
2053     DWORD reg_dest_code;
2054
2055     /* All versions have a destination register. The Tx where the texture coordinates come
2056      * from is the varying incarnation of the texture register
2057      */
2058     reg_dest_code = dst->reg.idx;
2059     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2060     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2061     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2062
2063     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2064      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2065      *
2066      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2067      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2068      *
2069      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2070      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2071      * extension.
2072      */
2073     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2074     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2075     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2076     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2077
2078     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2079      * so we can't let the GL handle this.
2080      */
2081     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2082             & WINED3D_PSARGS_PROJECTED)
2083     {
2084         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2085         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2086         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2087     } else {
2088         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2089     }
2090
2091     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2092
2093     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2094     {
2095         /* No src swizzles are allowed, so this is ok */
2096         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2097                        src_reg, reg_dest_code, reg_dest_code);
2098         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2099     }
2100 }
2101
2102 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2103 {
2104     DWORD reg = ins->dst[0].reg.idx;
2105     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2106     char src0_name[50], dst_name[50];
2107     BOOL is_color;
2108     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2109
2110     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2111     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2112      * T<reg+1> register. Use this register to store the calculated vector
2113      */
2114     tmp_reg.idx = reg + 1;
2115     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2116     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2117 }
2118
2119 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2120 {
2121     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2122     DWORD flags;
2123     DWORD reg = ins->dst[0].reg.idx;
2124     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2125     char dst_str[50];
2126     char src0_name[50];
2127     char dst_reg[50];
2128     BOOL is_color;
2129
2130     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2131     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2132
2133     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2134     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2135     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2136     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2137     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2138 }
2139
2140 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2141 {
2142     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2143     DWORD reg = ins->dst[0].reg.idx;
2144     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2145     char src0_name[50], dst_name[50];
2146     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2147     BOOL is_color;
2148
2149     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2150      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2151      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2152      */
2153     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2154     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2155
2156     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2157     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2158                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2159     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2160 }
2161
2162 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2163 {
2164     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2165     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2166     DWORD flags;
2167     DWORD reg = ins->dst[0].reg.idx;
2168     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2169     char dst_str[50];
2170     char src0_name[50], dst_name[50];
2171     BOOL is_color;
2172
2173     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2174     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2175     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2176
2177     /* Sample the texture using the calculated coordinates */
2178     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2179     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2180     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2181     tex_mx->current_row = 0;
2182 }
2183
2184 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2185 {
2186     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2187     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2188     DWORD flags;
2189     DWORD reg = ins->dst[0].reg.idx;
2190     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2191     char dst_str[50];
2192     char src0_name[50];
2193     char dst_reg[50];
2194     BOOL is_color;
2195
2196     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2197      * components for temporary data storage
2198      */
2199     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2200     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2201     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2202
2203     /* Construct the eye-ray vector from w coordinates */
2204     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2205     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2206     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2207
2208     /* Calculate reflection vector
2209      */
2210     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2211     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2212     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2213     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2214     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2215     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2216     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2217
2218     /* Sample the texture using the calculated coordinates */
2219     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2220     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2221     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2222     tex_mx->current_row = 0;
2223 }
2224
2225 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2226 {
2227     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2228     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2229     DWORD flags;
2230     DWORD reg = ins->dst[0].reg.idx;
2231     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2232     char dst_str[50];
2233     char src0_name[50];
2234     char src1_name[50];
2235     char dst_reg[50];
2236     BOOL is_color;
2237
2238     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2239     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2240     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2241     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2242     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2243
2244     /* Calculate reflection vector.
2245      *
2246      *                   dot(N, E)
2247      * dst_reg.xyz = 2 * --------- * N - E
2248      *                   dot(N, N)
2249      *
2250      * Which normalizes the normal vector
2251      */
2252     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2253     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2254     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2255     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2256     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2257     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2258
2259     /* Sample the texture using the calculated coordinates */
2260     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2261     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2262     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2263     tex_mx->current_row = 0;
2264 }
2265
2266 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2267 {
2268     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2269     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2270     char dst_name[50];
2271     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2272     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2273
2274     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2275      * which is essentially an input, is the destination register because it is the first
2276      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2277      * here(writemasks/swizzles are not valid on texdepth)
2278      */
2279     shader_arb_get_dst_param(ins, dst, dst_name);
2280
2281     /* According to the msdn, the source register(must be r5) is unusable after
2282      * the texdepth instruction, so we're free to modify it
2283      */
2284     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2285
2286     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2287      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2288      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2289      */
2290     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2291     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2292     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2293     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2294 }
2295
2296 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2297  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2298  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2299 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2300 {
2301     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2302     DWORD sampler_idx = ins->dst[0].reg.idx;
2303     char src0[50];
2304     char dst_str[50];
2305
2306     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2307     shader_addline(buffer, "MOV TB, 0.0;\n");
2308     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2309
2310     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2311     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2312 }
2313
2314 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2315  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2316 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2317 {
2318     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2319     char src0[50];
2320     char dst_str[50];
2321     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2322
2323     /* Handle output register */
2324     shader_arb_get_dst_param(ins, dst, dst_str);
2325     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2326     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2327 }
2328
2329 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2330  * Perform the 3rd row of a 3x3 matrix multiply */
2331 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2332 {
2333     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2334     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2335     char dst_str[50], dst_name[50];
2336     char src0[50];
2337     BOOL is_color;
2338
2339     shader_arb_get_dst_param(ins, dst, dst_str);
2340     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2341     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2342     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2343     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2344 }
2345
2346 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2347  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2348  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2349  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2350  */
2351 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2352 {
2353     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2354     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2355     char src0[50], dst_name[50];
2356     BOOL is_color;
2357     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2358     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2359
2360     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2361     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2362     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2363
2364     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2365      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2366      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2367      */
2368     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2369     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2370     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2371     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2372 }
2373
2374 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2375     Vertex/Pixel shaders to ARB_vertex_program codes */
2376 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2377 {
2378     int i;
2379     int nComponents = 0;
2380     struct wined3d_shader_dst_param tmp_dst = {{0}};
2381     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2382     struct wined3d_shader_instruction tmp_ins;
2383
2384     memset(&tmp_ins, 0, sizeof(tmp_ins));
2385
2386     /* Set constants for the temporary argument */
2387     tmp_ins.ctx = ins->ctx;
2388     tmp_ins.dst_count = 1;
2389     tmp_ins.dst = &tmp_dst;
2390     tmp_ins.src_count = 2;
2391     tmp_ins.src = tmp_src;
2392
2393     switch(ins->handler_idx)
2394     {
2395         case WINED3DSIH_M4x4:
2396             nComponents = 4;
2397             tmp_ins.handler_idx = WINED3DSIH_DP4;
2398             break;
2399         case WINED3DSIH_M4x3:
2400             nComponents = 3;
2401             tmp_ins.handler_idx = WINED3DSIH_DP4;
2402             break;
2403         case WINED3DSIH_M3x4:
2404             nComponents = 4;
2405             tmp_ins.handler_idx = WINED3DSIH_DP3;
2406             break;
2407         case WINED3DSIH_M3x3:
2408             nComponents = 3;
2409             tmp_ins.handler_idx = WINED3DSIH_DP3;
2410             break;
2411         case WINED3DSIH_M3x2:
2412             nComponents = 2;
2413             tmp_ins.handler_idx = WINED3DSIH_DP3;
2414             break;
2415         default:
2416             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2417             break;
2418     }
2419
2420     tmp_dst = ins->dst[0];
2421     tmp_src[0] = ins->src[0];
2422     tmp_src[1] = ins->src[1];
2423     for (i = 0; i < nComponents; i++) {
2424         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2425         shader_hw_map2gl(&tmp_ins);
2426         ++tmp_src[1].reg.idx;
2427     }
2428 }
2429
2430 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2431 {
2432     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2433
2434     char dst[50];
2435     char src[50];
2436
2437     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2438     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2439     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2440     {
2441         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2442          * .w is used
2443          */
2444         strcat(src, ".w");
2445     }
2446
2447     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2448 }
2449
2450 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2451 {
2452     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2453     const char *instruction;
2454
2455     char dst[50];
2456     char src[50];
2457
2458     switch(ins->handler_idx)
2459     {
2460         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2461         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2462         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2463         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2464         default: instruction = "";
2465             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2466             break;
2467     }
2468
2469     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2470     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2471     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2472     {
2473         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2474          * .w is used
2475          */
2476         strcat(src, ".w");
2477     }
2478
2479     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2480 }
2481
2482 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2483 {
2484     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2485     char dst_name[50];
2486     char src_name[50];
2487     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2488     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2489     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2490
2491     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2492     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2493
2494     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2495      * otherwise NRM or RSQ would return NaN */
2496     if(pshader && priv->target_version >= NV3)
2497     {
2498         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2499          *
2500          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2501          */
2502         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2503         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2504         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2505     }
2506     else if(priv->target_version >= NV2)
2507     {
2508         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2509         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2510         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2511                        src_name);
2512     }
2513     else
2514     {
2515         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2516
2517         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2518         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2519          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2520          */
2521         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2522         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2523
2524         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2525         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2526         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2527                     src_name);
2528     }
2529 }
2530
2531 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2532 {
2533     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2534     char dst_name[50];
2535     char src_name[3][50];
2536
2537     /* ARB_fragment_program has a convenient LRP instruction */
2538     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2539         shader_hw_map2gl(ins);
2540         return;
2541     }
2542
2543     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2544     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2545     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2546     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2547
2548     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2549     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2550                    dst_name, src_name[0], src_name[2]);
2551 }
2552
2553 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2554 {
2555     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2556      * must contain fixed constants. So we need a separate function to filter those constants and
2557      * can't use map2gl
2558      */
2559     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2560     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2561     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2562     char dst_name[50];
2563     char src_name0[50], src_name1[50], src_name2[50];
2564     BOOL is_color;
2565
2566     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2567     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2568         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2569         /* No modifiers are supported on SCS */
2570         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2571
2572         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2573         {
2574             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2575             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2576         }
2577     } else if(priv->target_version >= NV2) {
2578         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2579
2580         /* Sincos writemask must be .x, .y or .xy */
2581         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2582             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2583         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2584             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2585     } else {
2586         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2587          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2588          *
2589          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2590          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2591          *
2592          * The constants we get are:
2593          *
2594          *  +1   +1,     -1     -1     +1      +1      -1       -1
2595          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2596          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2597          *
2598          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2599          *
2600          * (x/2)^2 = x^2 / 4
2601          * (x/2)^3 = x^3 / 8
2602          * (x/2)^4 = x^4 / 16
2603          * (x/2)^5 = x^5 / 32
2604          * etc
2605          *
2606          * To get the final result:
2607          * sin(x) = 2 * sin(x/2) * cos(x/2)
2608          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2609          * (from sin(x+y) and cos(x+y) rules)
2610          *
2611          * As per MSDN, dst.z is undefined after the operation, and so is
2612          * dst.x and dst.y if they're masked out by the writemask. Ie
2613          * sincos dst.y, src1, c0, c1
2614          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2615          * vsa.exe also stops with an error if the dest register is the same register as the source
2616          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2617          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2618          */
2619         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2620         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2621         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2622
2623         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2624         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2625         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2626         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2627         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2628         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2629
2630         /* sin(x/2)
2631          *
2632          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2633          * properly merge that with MULs in the code above?
2634          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2635          * we can merge the sine and cosine MAD rows to calculate them together.
2636          */
2637         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2638         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2639         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2640         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2641
2642         /* cos(x/2) */
2643         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2644         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2645         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2646
2647         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2648             /* cos x */
2649             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2650             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2651         }
2652         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2653             /* sin x */
2654             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2655             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2656         }
2657     }
2658 }
2659
2660 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2661 {
2662     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2663     char dst_name[50];
2664     char src_name[50];
2665     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2666
2667     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2668     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2669
2670     /* SGN is only valid in vertex shaders */
2671     if(ctx->target_version >= NV2) {
2672         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2673         return;
2674     }
2675
2676     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2677      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2678      */
2679     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2680         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2681     } else {
2682         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2683          * Then use TA, and calculate the final result
2684          *
2685          * Not reading from TA? Store the first result in TA to avoid overwriting the
2686          * destination if src reg = dst reg
2687          */
2688         if(strstr(src_name, "TA"))
2689         {
2690             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2691             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2692             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2693         }
2694         else
2695         {
2696             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2697             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2698             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2699         }
2700     }
2701 }
2702
2703 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2704 {
2705     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2706     char src[50];
2707     char dst[50];
2708     char dst_name[50];
2709     BOOL is_color;
2710
2711     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2712     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2713     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2714
2715     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2716     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2717 }
2718
2719 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2720 {
2721     *need_abs = FALSE;
2722
2723     switch(mod)
2724     {
2725         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2726         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2727         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2728         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2729         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2730         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2731         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2732         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2733         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2734         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2735         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2736         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2737         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2738     }
2739     FIXME("Unknown modifier %u\n", mod);
2740     return mod;
2741 }
2742
2743 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2744 {
2745     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2746     char src0[50], dst[50];
2747     struct wined3d_shader_src_param src0_copy = ins->src[0];
2748     BOOL need_abs = FALSE;
2749     const char *instr;
2750
2751     switch(ins->handler_idx)
2752     {
2753         case WINED3DSIH_LOG:  instr = "LG2"; break;
2754         case WINED3DSIH_LOGP: instr = "LOG"; break;
2755         default:
2756             ERR("Unexpected instruction %d\n", ins->handler_idx);
2757             return;
2758     }
2759
2760     /* LOG and LOGP operate on the absolute value of the input */
2761     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2762
2763     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2764     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2765
2766     if(need_abs)
2767     {
2768         shader_addline(buffer, "ABS TA, %s;\n", src0);
2769         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2770     }
2771     else
2772     {
2773         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2774     }
2775 }
2776
2777 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2778 {
2779     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2780     char src0[50], src1[50], dst[50];
2781     struct wined3d_shader_src_param src0_copy = ins->src[0];
2782     BOOL need_abs = FALSE;
2783     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2784     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2785
2786     /* POW operates on the absolute value of the input */
2787     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2788
2789     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2790     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2791     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2792
2793     if (need_abs)
2794         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2795     else
2796         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2797
2798     if (priv->target_version >= NV2)
2799     {
2800         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2801         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2802         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2803     }
2804     else
2805     {
2806         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2807         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2808
2809         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2810         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2811         /* Possibly add flt_eps to avoid getting float special values */
2812         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2813         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2814         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2815         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2816     }
2817 }
2818
2819 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2820 {
2821     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2822     char src_name[50];
2823     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2824
2825     /* src0 is aL */
2826     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2827
2828     if(vshader)
2829     {
2830         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2831         struct list *e = list_head(&priv->control_frames);
2832         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2833
2834         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2835         /* The constant loader makes sure to load -1 into iX.w */
2836         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2837         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2838         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2839     }
2840     else
2841     {
2842         shader_addline(buffer, "LOOP %s;\n", src_name);
2843     }
2844 }
2845
2846 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2847 {
2848     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2849     char src_name[50];
2850     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2851
2852     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2853
2854     /* The constant loader makes sure to load -1 into iX.w */
2855     if(vshader)
2856     {
2857         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2858         struct list *e = list_head(&priv->control_frames);
2859         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2860
2861         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2862
2863         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2864         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2865         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2866     }
2867     else
2868     {
2869         shader_addline(buffer, "REP %s;\n", src_name);
2870     }
2871 }
2872
2873 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2874 {
2875     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2876     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2877
2878     if(vshader)
2879     {
2880         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2881         struct list *e = list_head(&priv->control_frames);
2882         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2883
2884         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2885         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2886         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2887
2888         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2889     }
2890     else
2891     {
2892         shader_addline(buffer, "ENDLOOP;\n");
2893     }
2894 }
2895
2896 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2897 {
2898     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2899     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2900
2901     if(vshader)
2902     {
2903         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2904         struct list *e = list_head(&priv->control_frames);
2905         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2906
2907         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2908         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2909         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2910
2911         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2912     }
2913     else
2914     {
2915         shader_addline(buffer, "ENDREP;\n");
2916     }
2917 }
2918
2919 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2920 {
2921     struct control_frame *control_frame;
2922
2923     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2924     {
2925         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2926     }
2927     ERR("Could not find loop for break\n");
2928     return NULL;
2929 }
2930
2931 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2932 {
2933     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2934     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2935     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2936
2937     if(vshader)
2938     {
2939         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2940     }
2941     else
2942     {
2943         shader_addline(buffer, "BRK;\n");
2944     }
2945 }
2946
2947 static const char *get_compare(enum wined3d_shader_rel_op op)
2948 {
2949     switch (op)
2950     {
2951         case WINED3D_SHADER_REL_OP_GT: return "GT";
2952         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2953         case WINED3D_SHADER_REL_OP_GE: return "GE";
2954         case WINED3D_SHADER_REL_OP_LT: return "LT";
2955         case WINED3D_SHADER_REL_OP_NE: return "NE";
2956         case WINED3D_SHADER_REL_OP_LE: return "LE";
2957         default:
2958             FIXME("Unrecognized operator %#x.\n", op);
2959             return "(\?\?)";
2960     }
2961 }
2962
2963 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2964 {
2965     switch (op)
2966     {
2967         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2968         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2969         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2970         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2971         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2972         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2973         default:
2974             FIXME("Unrecognized operator %#x.\n", op);
2975             return -1;
2976     }
2977 }
2978
2979 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2980 {
2981     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2982     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2983     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2984     char src_name0[50];
2985     char src_name1[50];
2986     const char *comp = get_compare(ins->flags);
2987
2988     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2989     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2990
2991     if(vshader)
2992     {
2993         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2994          * away the subtraction result
2995          */
2996         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2997         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2998     }
2999     else
3000     {
3001         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3002         shader_addline(buffer, "BRK (%s.x);\n", comp);
3003     }
3004 }
3005
3006 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3007 {
3008     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3009     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3010     struct list *e = list_head(&priv->control_frames);
3011     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3012     const char *comp;
3013     char src_name0[50];
3014     char src_name1[50];
3015     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3016
3017     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3018     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3019
3020     if(vshader)
3021     {
3022         /* Invert the flag. We jump to the else label if the condition is NOT true */
3023         comp = get_compare(invert_compare(ins->flags));
3024         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3025         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3026     }
3027     else
3028     {
3029         comp = get_compare(ins->flags);
3030         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3031         shader_addline(buffer, "IF %s.x;\n", comp);
3032     }
3033 }
3034
3035 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3036 {
3037     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3038     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3039     struct list *e = list_head(&priv->control_frames);
3040     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3041     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3042
3043     if(vshader)
3044     {
3045         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3046         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3047         control_frame->had_else = TRUE;
3048     }
3049     else
3050     {
3051         shader_addline(buffer, "ELSE;\n");
3052     }
3053 }
3054
3055 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3056 {
3057     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3058     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3059     struct list *e = list_head(&priv->control_frames);
3060     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3061     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3062
3063     if(vshader)
3064     {
3065         if(control_frame->had_else)
3066         {
3067             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3068         }
3069         else
3070         {
3071             shader_addline(buffer, "#No else branch. else is endif\n");
3072             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3073         }
3074     }
3075     else
3076     {
3077         shader_addline(buffer, "ENDIF;\n");
3078     }
3079 }
3080
3081 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3082 {
3083     DWORD sampler_idx = ins->src[1].reg.idx;
3084     char reg_dest[40];
3085     char reg_src[3][40];
3086     WORD flags = TEX_DERIV;
3087
3088     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3089     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3090     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3091     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3092
3093     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3094     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3095
3096     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3097 }
3098
3099 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3100 {
3101     DWORD sampler_idx = ins->src[1].reg.idx;
3102     char reg_dest[40];
3103     char reg_coord[40];
3104     WORD flags = TEX_LOD;
3105
3106     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3107     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3108
3109     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3110     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3111
3112     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3113 }
3114
3115 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3116 {
3117     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3118     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3119
3120     priv->in_main_func = FALSE;
3121     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3122      * subroutine, don't generate a label that will make GL complain
3123      */
3124     if(priv->target_version == ARB) return;
3125
3126     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3127 }
3128
3129 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3130         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3131         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3132         struct wined3d_shader_buffer *buffer)
3133 {
3134     unsigned int i;
3135
3136     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3137      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3138      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3139      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3140      */
3141     if (args->super.fog_src == VS_FOG_Z)
3142         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3143     else if (!reg_maps->fog)
3144         /* posFixup.x is always 1.0, so we can safely use it */
3145         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3146
3147     /* Clipplanes are always stored without y inversion */
3148     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3149     {
3150         if (args->super.clip_enabled)
3151         {
3152             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3153             {
3154                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3155             }
3156         }
3157     }
3158     else if (args->clip.boolclip.clip_texcoord)
3159     {
3160         unsigned int cur_clip = 0;
3161         char component[4] = {'x', 'y', 'z', 'w'};
3162         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3163
3164         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3165         {
3166             if (args->clip.boolclip.clipplane_mask & (1 << i))
3167             {
3168                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3169                                component[cur_clip++], i);
3170             }
3171         }
3172         switch (cur_clip)
3173         {
3174             case 0:
3175                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3176                 break;
3177             case 1:
3178                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3179                 break;
3180             case 2:
3181                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3182                 break;
3183             case 3:
3184                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3185                 break;
3186         }
3187         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3188                        args->clip.boolclip.clip_texcoord - 1);
3189     }
3190
3191     /* Write the final position.
3192      *
3193      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3194      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3195      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3196      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3197      */
3198     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3199     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3200     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3201
3202     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3203      * and the glsl equivalent
3204      */
3205     if (need_helper_const(shader_data, reg_maps, gl_info))
3206     {
3207         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3208         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3209     }
3210     else
3211     {
3212         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3213         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3214     }
3215
3216     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3217
3218     priv_ctx->footer_written = TRUE;
3219 }
3220
3221 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3222 {
3223     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3224     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3225     const struct wined3d_shader *shader = ins->ctx->shader;
3226     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3227
3228     if(priv->target_version == ARB) return;
3229
3230     if(vshader)
3231     {
3232         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3233                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3234     }
3235
3236     shader_addline(buffer, "RET;\n");
3237 }
3238
3239 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3240 {
3241     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3242     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3243 }
3244
3245 /* GL locking is done by the caller */
3246 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3247 {
3248     GLuint program_id = 0;
3249     GLint pos;
3250
3251     const char *blt_vprogram =
3252         "!!ARBvp1.0\n"
3253         "PARAM c[1] = { { 1, 0.5 } };\n"
3254         "MOV result.position, vertex.position;\n"
3255         "MOV result.color, c[0].x;\n"
3256         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3257         "END\n";
3258
3259     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3260     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3261     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3262             strlen(blt_vprogram), blt_vprogram));
3263     checkGLcall("glProgramStringARB()");
3264
3265     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3266     if (pos != -1)
3267     {
3268         FIXME("Vertex program error at position %d: %s\n\n", pos,
3269             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3270         shader_arb_dump_program_source(blt_vprogram);
3271     }
3272     else
3273     {
3274         GLint native;
3275
3276         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3277         checkGLcall("glGetProgramivARB()");
3278         if (!native) WARN("Program exceeds native resource limits.\n");
3279     }
3280
3281     return program_id;
3282 }
3283
3284 /* GL locking is done by the caller */
3285 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3286         enum tex_types tex_type, BOOL masked)
3287 {
3288     GLuint program_id = 0;
3289     const char *fprogram;
3290     GLint pos;
3291
3292     static const char * const blt_fprograms_full[tex_type_count] =
3293     {
3294         /* tex_1d */
3295         NULL,
3296         /* tex_2d */
3297         "!!ARBfp1.0\n"
3298         "TEMP R0;\n"
3299         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3300         "MOV result.depth.z, R0.x;\n"
3301         "END\n",
3302         /* tex_3d */
3303         NULL,
3304         /* tex_cube */
3305         "!!ARBfp1.0\n"
3306         "TEMP R0;\n"
3307         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3308         "MOV result.depth.z, R0.x;\n"
3309         "END\n",
3310         /* tex_rect */
3311         "!!ARBfp1.0\n"
3312         "TEMP R0;\n"
3313         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3314         "MOV result.depth.z, R0.x;\n"
3315         "END\n",
3316     };
3317
3318     static const char * const blt_fprograms_masked[tex_type_count] =
3319     {
3320         /* tex_1d */
3321         NULL,
3322         /* tex_2d */
3323         "!!ARBfp1.0\n"
3324         "PARAM mask = program.local[0];\n"
3325         "TEMP R0;\n"
3326         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3327         "MUL R0.x, R0.x, R0.y;\n"
3328         "KIL -R0.x;\n"
3329         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3330         "MOV result.depth.z, R0.x;\n"
3331         "END\n",
3332         /* tex_3d */
3333         NULL,
3334         /* tex_cube */
3335         "!!ARBfp1.0\n"
3336         "PARAM mask = program.local[0];\n"
3337         "TEMP R0;\n"
3338         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3339         "MUL R0.x, R0.x, R0.y;\n"
3340         "KIL -R0.x;\n"
3341         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3342         "MOV result.depth.z, R0.x;\n"
3343         "END\n",
3344         /* tex_rect */
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], RECT;\n"
3352         "MOV result.depth.z, R0.x;\n"
3353         "END\n",
3354     };
3355
3356     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3357     if (!fprogram)
3358     {
3359         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3360         tex_type = tex_2d;
3361         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3362     }
3363
3364     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3365     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3366     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3367     checkGLcall("glProgramStringARB()");
3368
3369     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3370     if (pos != -1)
3371     {
3372         FIXME("Fragment program error at position %d: %s\n\n", pos,
3373             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3374         shader_arb_dump_program_source(fprogram);
3375     }
3376     else
3377     {
3378         GLint native;
3379
3380         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3381         checkGLcall("glGetProgramivARB()");
3382         if (!native) WARN("Program exceeds native resource limits.\n");
3383     }
3384
3385     return program_id;
3386 }
3387
3388 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3389         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3390 {
3391     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3392
3393     if(condcode)
3394     {
3395         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3396         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3397         /* Calculate the > 0.0031308 case */
3398         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3399         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3400         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3401         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3402         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3403         /* Calculate the < case */
3404         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3405     }
3406     else
3407     {
3408         /* Calculate the > 0.0031308 case */
3409         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3410         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3411         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3412         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3413         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3414         /* Calculate the < case */
3415         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3416         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3417         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3418         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3419         /* Store the components > 0.0031308 in the destination */
3420         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3421         /* Add the components that are < 0.0031308 */
3422         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3423         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3424         * result.color writes(.rgb first, then .a), or handle overwriting already written
3425         * components. The assembler uses a temporary register in this case, which is usually
3426         * not allocated from one of our registers that were used earlier.
3427         */
3428     }
3429     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3430 }
3431
3432 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3433 {
3434     const struct wined3d_shader_lconst *constant;
3435
3436     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3437     {
3438         if (constant->idx == idx)
3439         {
3440             return constant->value;
3441         }
3442     }
3443     return NULL;
3444 }
3445
3446 static void init_ps_input(const struct wined3d_shader *shader,
3447         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3448 {
3449     static const char * const texcoords[8] =
3450     {
3451         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3452         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3453     };
3454     unsigned int i;
3455     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3456     const char *semantic_name;
3457     DWORD semantic_idx;
3458
3459     switch(args->super.vp_mode)
3460     {
3461         case pretransformed:
3462         case fixedfunction:
3463             /* The pixelshader has to collect the varyings on its own. In any case properly load
3464              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3465              * other attribs to 0.0.
3466              *
3467              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3468              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3469              * load the texcoord attrib pointers to match the pixel shader signature
3470              */
3471             for(i = 0; i < MAX_REG_INPUT; i++)
3472             {
3473                 semantic_name = sig[i].semantic_name;
3474                 semantic_idx = sig[i].semantic_idx;
3475                 if (!semantic_name) continue;
3476
3477                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3478                 {
3479                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3480                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3481                     else priv->ps_input[i] = "0.0";
3482                 }
3483                 else if(args->super.vp_mode == fixedfunction)
3484                 {
3485                     priv->ps_input[i] = "0.0";
3486                 }
3487                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3488                 {
3489                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3490                     else priv->ps_input[i] = "0.0";
3491                 }
3492                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3493                 {
3494                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3495                     else priv->ps_input[i] = "0.0";
3496                 }
3497                 else
3498                 {
3499                     priv->ps_input[i] = "0.0";
3500                 }
3501
3502                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3503             }
3504             break;
3505
3506         case vertexshader:
3507             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3508              * fragment.color
3509              */
3510             for(i = 0; i < 8; i++)
3511             {
3512                 priv->ps_input[i] = texcoords[i];
3513             }
3514             priv->ps_input[8] = "fragment.color.primary";
3515             priv->ps_input[9] = "fragment.color.secondary";
3516             break;
3517     }
3518 }
3519
3520 /* GL locking is done by the caller */
3521 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3522         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3523         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3524 {
3525     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3526     const struct wined3d_shader_lconst *lconst;
3527     const DWORD *function = shader->function;
3528     GLuint retval;
3529     char fragcolor[16];
3530     DWORD *lconst_map = local_const_mapping(shader), next_local;
3531     struct shader_arb_ctx_priv priv_ctx;
3532     BOOL dcl_td = FALSE;
3533     BOOL want_nv_prog = FALSE;
3534     struct arb_pshader_private *shader_priv = shader->backend_data;
3535     GLint errPos;
3536     DWORD map;
3537
3538     char srgbtmp[4][4];
3539     unsigned int i, found = 0;
3540
3541     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3542     {
3543         if (!(map & 1)
3544                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3545                 || (reg_maps->shader_version.major < 2 && !i))
3546             continue;
3547
3548         sprintf(srgbtmp[found], "R%u", i);
3549         ++found;
3550         if (found == 4) break;
3551     }
3552
3553     switch(found) {
3554         case 0:
3555             sprintf(srgbtmp[0], "TA");
3556             sprintf(srgbtmp[1], "TB");
3557             sprintf(srgbtmp[2], "TC");
3558             sprintf(srgbtmp[3], "TD");
3559             dcl_td = TRUE;
3560             break;
3561         case 1:
3562             sprintf(srgbtmp[1], "TA");
3563             sprintf(srgbtmp[2], "TB");
3564             sprintf(srgbtmp[3], "TC");
3565             break;
3566         case 2:
3567             sprintf(srgbtmp[2], "TA");
3568             sprintf(srgbtmp[3], "TB");
3569             break;
3570         case 3:
3571             sprintf(srgbtmp[3], "TA");
3572             break;
3573         case 4:
3574             break;
3575     }
3576
3577     /*  Create the hw ARB shader */
3578     memset(&priv_ctx, 0, sizeof(priv_ctx));
3579     priv_ctx.cur_ps_args = args;
3580     priv_ctx.compiled_fprog = compiled;
3581     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3582     init_ps_input(shader, args, &priv_ctx);
3583     list_init(&priv_ctx.control_frames);
3584
3585     /* Avoid enabling NV_fragment_program* if we do not need it.
3586      *
3587      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3588      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3589      * is faster than what we gain from using higher native instructions. There are some things though
3590      * that cannot be emulated. In that case enable the extensions.
3591      * If the extension is enabled, instruction handlers that support both ways will use it.
3592      *
3593      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3594      * So enable the best we can get.
3595      */
3596     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3597        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3598     {
3599         want_nv_prog = TRUE;
3600     }
3601
3602     shader_addline(buffer, "!!ARBfp1.0\n");
3603     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3604     {
3605         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3606         priv_ctx.target_version = NV3;
3607     }
3608     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3609     {
3610         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3611         priv_ctx.target_version = NV2;
3612     } else {
3613         if(want_nv_prog)
3614         {
3615             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3616              * limits properly
3617              */
3618             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3619             ERR("Try GLSL\n");
3620         }
3621         priv_ctx.target_version = ARB;
3622     }
3623
3624     if (reg_maps->rt_mask > 1)
3625     {
3626         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3627     }
3628
3629     if (reg_maps->shader_version.major < 3)
3630     {
3631         switch(args->super.fog) {
3632             case FOG_OFF:
3633                 break;
3634             case FOG_LINEAR:
3635                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3636                 break;
3637             case FOG_EXP:
3638                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3639                 break;
3640             case FOG_EXP2:
3641                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3642                 break;
3643         }
3644     }
3645
3646     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3647      * unused temps away(but occupies them for the whole shader if they're used once). Always
3648      * declaring them avoids tricky bookkeeping work
3649      */
3650     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3651     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3652     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3653     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3654     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3655     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3656     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3657
3658     if (reg_maps->shader_version.major < 2)
3659     {
3660         strcpy(fragcolor, "R0");
3661     }
3662     else
3663     {
3664         if (args->super.srgb_correction)
3665         {
3666             if (shader->u.ps.color0_mov)
3667             {
3668                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3669             }
3670             else
3671             {
3672                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3673                 strcpy(fragcolor, "TMP_COLOR");
3674             }
3675         } else {
3676             strcpy(fragcolor, "result.color");
3677         }
3678     }
3679
3680     if(args->super.srgb_correction) {
3681         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3682                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3683         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3684                        srgb_sub_high, 0.0, 0.0, 0.0);
3685     }
3686
3687     /* Base Declarations */
3688     next_local = shader_generate_arb_declarations(shader, reg_maps,
3689             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3690
3691     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3692     {
3693         unsigned char bump_const;
3694
3695         if (!(map & 1)) continue;
3696
3697         bump_const = compiled->numbumpenvmatconsts;
3698         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3699         compiled->bumpenvmatconst[bump_const].texunit = i;
3700         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3701         compiled->luminanceconst[bump_const].texunit = i;
3702
3703         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3704          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3705          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3706          * textures due to conditional NP2 restrictions)
3707          *
3708          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3709          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3710          * their location is shader dependent anyway and they cannot be loaded globally.
3711          */
3712         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3713         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3714                        i, compiled->bumpenvmatconst[bump_const].const_num);
3715         compiled->numbumpenvmatconsts = bump_const + 1;
3716
3717         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3718
3719         compiled->luminanceconst[bump_const].const_num = next_local++;
3720         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3721                        i, compiled->luminanceconst[bump_const].const_num);
3722     }
3723
3724     for(i = 0; i < MAX_CONST_I; i++)
3725     {
3726         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3727         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3728         {
3729             const DWORD *control_values = find_loop_control_values(shader, i);
3730
3731             if(control_values)
3732             {
3733                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3734                                 control_values[0], control_values[1], control_values[2]);
3735             }
3736             else
3737             {
3738                 compiled->int_consts[i] = next_local;
3739                 compiled->num_int_consts++;
3740                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3741             }
3742         }
3743     }
3744
3745     if(reg_maps->vpos || reg_maps->usesdsy)
3746     {
3747         compiled->ycorrection = next_local;
3748         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3749
3750         if(reg_maps->vpos)
3751         {
3752             shader_addline(buffer, "TEMP vpos;\n");
3753             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3754              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3755              * ycorrection.z: 1.0
3756              * ycorrection.w: 0.0
3757              */
3758             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3759             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3760         }
3761     }
3762     else
3763     {
3764         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3765     }
3766
3767     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3768      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3769      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3770      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3771      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3772      * shader compilation errors and the subsequent errors when drawing with this shader. */
3773     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3774         unsigned char cur_fixup_sampler = 0;
3775
3776         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3777         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3778         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3779
3780         fixup->offset = next_local;
3781         fixup->super.active = 0;
3782
3783         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3784             if (!(map & (1 << i))) continue;
3785
3786             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3787                 fixup->super.active |= (1 << i);
3788                 fixup->super.idx[i] = cur_fixup_sampler++;
3789             } else {
3790                 FIXME("No free constant found to load NP2 fixup data into shader. "
3791                       "Sampling from this texture will probably look wrong.\n");
3792                 break;
3793             }
3794         }
3795
3796         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3797         if (fixup->super.num_consts) {
3798             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3799                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3800         }
3801     }
3802
3803     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3804     {
3805         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3806     }
3807
3808     /* Base Shader Body */
3809     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3810
3811     if(args->super.srgb_correction) {
3812         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3813                                   priv_ctx.target_version >= NV2);
3814     }
3815
3816     if(strcmp(fragcolor, "result.color")) {
3817         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3818     }
3819     shader_addline(buffer, "END\n");
3820
3821     /* TODO: change to resource.glObjectHandle or something like that */
3822     GL_EXTCALL(glGenProgramsARB(1, &retval));
3823
3824     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3825     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3826
3827     TRACE("Created hw pixel shader, prg=%d\n", retval);
3828     /* Create the program and check for errors */
3829     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3830                buffer->bsize, buffer->buffer));
3831     checkGLcall("glProgramStringARB()");
3832
3833     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3834     if (errPos != -1)
3835     {
3836         FIXME("HW PixelShader Error at position %d: %s\n\n",
3837               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3838         shader_arb_dump_program_source(buffer->buffer);
3839         retval = 0;
3840     }
3841     else
3842     {
3843         GLint native;
3844
3845         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3846         checkGLcall("glGetProgramivARB()");
3847         if (!native) WARN("Program exceeds native resource limits.\n");
3848     }
3849
3850     /* Load immediate constants */
3851     if (lconst_map)
3852     {
3853         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3854         {
3855             const float *value = (const float *)lconst->value;
3856             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3857             checkGLcall("glProgramLocalParameter4fvARB");
3858         }
3859         HeapFree(GetProcessHeap(), 0, lconst_map);
3860     }
3861
3862     return retval;
3863 }
3864
3865 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3866 {
3867     unsigned int i;
3868     int ret;
3869
3870     for(i = 0; i < MAX_REG_INPUT; i++)
3871     {
3872         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3873         {
3874             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3875             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3876             continue;
3877         }
3878
3879         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3880         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3881         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3882         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3883         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3884         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3885     }
3886     return 0;
3887 }
3888
3889 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3890 {
3891     struct wined3d_shader_signature_element *new;
3892     int i;
3893     char *name;
3894
3895     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3896     for(i = 0; i < MAX_REG_INPUT; i++)
3897     {
3898         if (!sig[i].semantic_name) continue;
3899
3900         new[i] = sig[i];
3901         /* Clone the semantic string */
3902         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3903         strcpy(name, sig[i].semantic_name);
3904         new[i].semantic_name = name;
3905     }
3906     return new;
3907 }
3908
3909 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3910 {
3911     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3912     struct ps_signature *found_sig;
3913
3914     if (entry)
3915     {
3916         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3917         TRACE("Found existing signature %u\n", found_sig->idx);
3918         return found_sig->idx;
3919     }
3920     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3921     found_sig->sig = clone_sig(sig);
3922     found_sig->idx = priv->ps_sig_number++;
3923     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3924     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3925     {
3926         ERR("Failed to insert program entry.\n");
3927     }
3928     return found_sig->idx;
3929 }
3930
3931 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3932         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3933 {
3934     unsigned int i, j;
3935     static const char * const texcoords[8] =
3936     {
3937         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3938         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3939     };
3940     struct wined3d_device *device = shader->device;
3941     const struct wined3d_shader_signature_element *sig;
3942     const char *semantic_name;
3943     DWORD semantic_idx, reg_idx;
3944
3945     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3946      * and varying 9 to result.color.secondary
3947      */
3948     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3949     {
3950         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3951         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3952         "result.color.primary", "result.color.secondary"
3953     };
3954
3955     if(sig_num == ~0)
3956     {
3957         TRACE("Pixel shader uses builtin varyings\n");
3958         /* Map builtins to builtins */
3959         for(i = 0; i < 8; i++)
3960         {
3961             priv_ctx->texcrd_output[i] = texcoords[i];
3962         }
3963         priv_ctx->color_output[0] = "result.color.primary";
3964         priv_ctx->color_output[1] = "result.color.secondary";
3965         priv_ctx->fog_output = "result.fogcoord";
3966
3967         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3968         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3969         {
3970             semantic_name = shader->output_signature[i].semantic_name;
3971             if (!semantic_name) continue;
3972
3973             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
3974             {
3975                 TRACE("o%u is TMP_OUT\n", i);
3976                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3977                 else priv_ctx->vs_output[i] = "TA";
3978             }
3979             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
3980             {
3981                 TRACE("o%u is result.pointsize\n", i);
3982                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3983                 else priv_ctx->vs_output[i] = "TA";
3984             }
3985             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3986             {
3987                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3988                 if (!shader->output_signature[i].semantic_idx)
3989                     priv_ctx->vs_output[i] = "result.color.primary";
3990                 else if (shader->output_signature[i].semantic_idx == 1)
3991                     priv_ctx->vs_output[i] = "result.color.secondary";
3992                 else priv_ctx->vs_output[i] = "TA";
3993             }
3994             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3995             {
3996                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3997                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3998                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3999             }
4000             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4001             {
4002                 TRACE("o%u is result.fogcoord\n", i);
4003                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4004                 else priv_ctx->vs_output[i] = "result.fogcoord";
4005             }
4006             else
4007             {
4008                 priv_ctx->vs_output[i] = "TA";
4009             }
4010         }
4011         return;
4012     }
4013
4014     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4015      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4016      */
4017     sig = device->stateBlock->state.pixel_shader->input_signature;
4018     TRACE("Pixel shader uses declared varyings\n");
4019
4020     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4021     for(i = 0; i < 8; i++)
4022     {
4023         priv_ctx->texcrd_output[i] = "TA";
4024     }
4025     priv_ctx->color_output[0] = "TA";
4026     priv_ctx->color_output[1] = "TA";
4027     priv_ctx->fog_output = "TA";
4028
4029     for(i = 0; i < MAX_REG_INPUT; i++)
4030     {
4031         semantic_name = sig[i].semantic_name;
4032         semantic_idx = sig[i].semantic_idx;
4033         reg_idx = sig[i].register_idx;
4034         if (!semantic_name) continue;
4035
4036         /* If a declared input register is not written by builtin arguments, don't write to it.
4037          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4038          *
4039          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4040          * to TMP_OUT in any case
4041          */
4042         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4043         {
4044             if (semantic_idx < 8)
4045                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4046         }
4047         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4048         {
4049             if (semantic_idx < 2)
4050                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4051         }
4052         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4053         {
4054             if (!semantic_idx)
4055                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4056         }
4057         else
4058         {
4059             continue;
4060         }
4061
4062         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4063                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4064         {
4065             compiled->need_color_unclamp = TRUE;
4066         }
4067     }
4068
4069     /* Map declared to declared */
4070     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4071     {
4072         /* Write unread output to TA to throw them away */
4073         priv_ctx->vs_output[i] = "TA";
4074         semantic_name = shader->output_signature[i].semantic_name;
4075         if (!semantic_name) continue;
4076
4077         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4078                 && !shader->output_signature[i].semantic_idx)
4079         {
4080             priv_ctx->vs_output[i] = "TMP_OUT";
4081             continue;
4082         }
4083         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4084                 && !shader->output_signature[i].semantic_idx)
4085         {
4086             priv_ctx->vs_output[i] = "result.pointsize";
4087             continue;
4088         }
4089
4090         for(j = 0; j < MAX_REG_INPUT; j++)
4091         {
4092             if (!sig[j].semantic_name) continue;
4093
4094             if (!strcmp(sig[j].semantic_name, semantic_name)
4095                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4096             {
4097                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4098
4099                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4100                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4101                 {
4102                     compiled->need_color_unclamp = TRUE;
4103                 }
4104             }
4105         }
4106     }
4107 }
4108
4109 /* GL locking is done by the caller */
4110 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4111         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4112         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4113 {
4114     const struct arb_vshader_private *shader_data = shader->backend_data;
4115     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4116     const struct wined3d_shader_lconst *lconst;
4117     const DWORD *function = shader->function;
4118     GLuint ret;
4119     DWORD next_local, *lconst_map = local_const_mapping(shader);
4120     struct shader_arb_ctx_priv priv_ctx;
4121     unsigned int i;
4122     GLint errPos;
4123
4124     memset(&priv_ctx, 0, sizeof(priv_ctx));
4125     priv_ctx.cur_vs_args = args;
4126     list_init(&priv_ctx.control_frames);
4127     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4128
4129     /*  Create the hw ARB shader */
4130     shader_addline(buffer, "!!ARBvp1.0\n");
4131
4132     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4133      * mesurable performance penalty, and we can always make use of it for clipplanes.
4134      */
4135     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4136     {
4137         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4138         priv_ctx.target_version = NV3;
4139         shader_addline(buffer, "ADDRESS aL;\n");
4140     }
4141     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4142     {
4143         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4144         priv_ctx.target_version = NV2;
4145         shader_addline(buffer, "ADDRESS aL;\n");
4146     } else {
4147         priv_ctx.target_version = ARB;
4148     }
4149
4150     shader_addline(buffer, "TEMP TMP_OUT;\n");
4151     if (need_helper_const(shader_data, reg_maps, gl_info))
4152     {
4153         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4154     }
4155     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4156     {
4157         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4158         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4159     }
4160
4161     shader_addline(buffer, "TEMP TA;\n");
4162     shader_addline(buffer, "TEMP TB;\n");
4163
4164     /* Base Declarations */
4165     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4166             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4167
4168     for(i = 0; i < MAX_CONST_I; i++)
4169     {
4170         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4171         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4172         {
4173             const DWORD *control_values = find_loop_control_values(shader, i);
4174
4175             if(control_values)
4176             {
4177                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4178                                 control_values[0], control_values[1], control_values[2]);
4179             }
4180             else
4181             {
4182                 compiled->int_consts[i] = next_local;
4183                 compiled->num_int_consts++;
4184                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4185             }
4186         }
4187     }
4188
4189     /* We need a constant to fixup the final position */
4190     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4191     compiled->pos_fixup = next_local++;
4192
4193     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4194      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4195      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4196      * a replacement shader depend on the texcoord.w being set properly.
4197      *
4198      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4199      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4200      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4201      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4202      * this can eat a number of instructions, so skip it unless this cap is set as well
4203      */
4204     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4205     {
4206         struct wined3d_device *device = shader->device;
4207         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4208         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4209
4210         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4211         {
4212             int i;
4213             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4214             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4215             {
4216                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4217                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4218             }
4219         }
4220     }
4221
4222     /* The shader starts with the main function */
4223     priv_ctx.in_main_func = TRUE;
4224     /* Base Shader Body */
4225     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4226
4227     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4228             shader_data, args, reg_maps, gl_info, buffer);
4229
4230     shader_addline(buffer, "END\n");
4231
4232     /* TODO: change to resource.glObjectHandle or something like that */
4233     GL_EXTCALL(glGenProgramsARB(1, &ret));
4234
4235     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4236     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4237
4238     TRACE("Created hw vertex shader, prg=%d\n", ret);
4239     /* Create the program and check for errors */
4240     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4241                buffer->bsize, buffer->buffer));
4242     checkGLcall("glProgramStringARB()");
4243
4244     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4245     if (errPos != -1)
4246     {
4247         FIXME("HW VertexShader Error at position %d: %s\n\n",
4248               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4249         shader_arb_dump_program_source(buffer->buffer);
4250         ret = -1;
4251     }
4252     else
4253     {
4254         GLint native;
4255
4256         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4257         checkGLcall("glGetProgramivARB()");
4258         if (!native) WARN("Program exceeds native resource limits.\n");
4259
4260         /* Load immediate constants */
4261         if (lconst_map)
4262         {
4263             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4264             {
4265                 const float *value = (const float *)lconst->value;
4266                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4267             }
4268         }
4269     }
4270     HeapFree(GetProcessHeap(), 0, lconst_map);
4271
4272     return ret;
4273 }
4274
4275 /* GL locking is done by the caller */
4276 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4277         const struct arb_ps_compile_args *args)
4278 {
4279     struct wined3d_device *device = shader->device;
4280     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4281     UINT i;
4282     DWORD new_size;
4283     struct arb_ps_compiled_shader *new_array;
4284     struct wined3d_shader_buffer buffer;
4285     struct arb_pshader_private *shader_data;
4286     GLuint ret;
4287
4288     if (!shader->backend_data)
4289     {
4290         struct shader_arb_priv *priv = device->shader_priv;
4291
4292         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4293         shader_data = shader->backend_data;
4294         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4295
4296         if (shader->reg_maps.shader_version.major < 3)
4297             shader_data->input_signature_idx = ~0;
4298         else
4299             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4300
4301         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4302
4303         if (!device->vs_clipping)
4304             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4305                     gl_info->limits.texture_stages - 1);
4306         else
4307             shader_data->clipplane_emulation = ~0U;
4308     }
4309     shader_data = shader->backend_data;
4310
4311     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4312      * so a linear search is more performant than a hashmap or a binary search
4313      * (cache coherency etc)
4314      */
4315     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4316     {
4317         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4318             return &shader_data->gl_shaders[i];
4319     }
4320
4321     TRACE("No matching GL shader found, compiling a new shader\n");
4322     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4323         if (shader_data->num_gl_shaders)
4324         {
4325             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4326             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4327                                     new_size * sizeof(*shader_data->gl_shaders));
4328         } else {
4329             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4330             new_size = 1;
4331         }
4332
4333         if(!new_array) {
4334             ERR("Out of memory\n");
4335             return 0;
4336         }
4337         shader_data->gl_shaders = new_array;
4338         shader_data->shader_array_size = new_size;
4339     }
4340
4341     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4342
4343     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4344
4345     if (!shader_buffer_init(&buffer))
4346     {
4347         ERR("Failed to initialize shader buffer.\n");
4348         return 0;
4349     }
4350
4351     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4352             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4353     shader_buffer_free(&buffer);
4354     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4355
4356     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4357 }
4358
4359 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4360                                  const DWORD use_map, BOOL skip_int) {
4361     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4362     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4363     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4364     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4365     if(stored->ps_signature != new->ps_signature) return FALSE;
4366     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4367     if(skip_int) return TRUE;
4368
4369     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4370 }
4371
4372 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4373         const struct arb_vs_compile_args *args)
4374 {
4375     struct wined3d_device *device = shader->device;
4376     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4377     DWORD use_map = device->strided_streams.use_map;
4378     UINT i;
4379     DWORD new_size;
4380     struct arb_vs_compiled_shader *new_array;
4381     struct wined3d_shader_buffer buffer;
4382     struct arb_vshader_private *shader_data;
4383     GLuint ret;
4384
4385     if (!shader->backend_data)
4386     {
4387         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4388
4389         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4390         shader_data = shader->backend_data;
4391
4392         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4393                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4394         {
4395             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4396             {
4397                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4398                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4399                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4400             }
4401             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4402                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4403             else if (reg_maps->max_rel_offset > 63)
4404                 shader_data->rel_offset = reg_maps->min_rel_offset;
4405         }
4406     }
4407     shader_data = shader->backend_data;
4408
4409     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4410      * so a linear search is more performant than a hashmap or a binary search
4411      * (cache coherency etc)
4412      */
4413     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4414         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4415                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4416         {
4417             return &shader_data->gl_shaders[i];
4418         }
4419     }
4420
4421     TRACE("No matching GL shader found, compiling a new shader\n");
4422
4423     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4424         if (shader_data->num_gl_shaders)
4425         {
4426             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4427             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4428                                     new_size * sizeof(*shader_data->gl_shaders));
4429         } else {
4430             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4431             new_size = 1;
4432         }
4433
4434         if(!new_array) {
4435             ERR("Out of memory\n");
4436             return 0;
4437         }
4438         shader_data->gl_shaders = new_array;
4439         shader_data->shader_array_size = new_size;
4440     }
4441
4442     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4443
4444     if (!shader_buffer_init(&buffer))
4445     {
4446         ERR("Failed to initialize shader buffer.\n");
4447         return 0;
4448     }
4449
4450     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4451             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4452     shader_buffer_free(&buffer);
4453     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4454
4455     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4456 }
4457
4458 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4459         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4460 {
4461     struct wined3d_device *device = shader->device;
4462     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4463     int i;
4464     WORD int_skip;
4465
4466     find_ps_compile_args(state, shader, &args->super);
4467
4468     /* This forces all local boolean constants to 1 to make them stateblock independent */
4469     args->bools = shader->reg_maps.local_bool_consts;
4470
4471     for(i = 0; i < MAX_CONST_B; i++)
4472     {
4473         if (state->ps_consts_b[i])
4474             args->bools |= ( 1 << i);
4475     }
4476
4477     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4478      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4479      * duplicate the shader than have a no-op KIL instruction in every shader
4480      */
4481     if (!device->vs_clipping && use_vs(state)
4482             && state->render_states[WINED3D_RS_CLIPPING]
4483             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4484         args->clip = 1;
4485     else
4486         args->clip = 0;
4487
4488     /* Skip if unused or local, or supported natively */
4489     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4490     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4491     {
4492         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4493         return;
4494     }
4495
4496     for(i = 0; i < MAX_CONST_I; i++)
4497     {
4498         if(int_skip & (1 << i))
4499         {
4500             args->loop_ctrl[i][0] = 0;
4501             args->loop_ctrl[i][1] = 0;
4502             args->loop_ctrl[i][2] = 0;
4503         }
4504         else
4505         {
4506             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4507             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4508             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4509         }
4510     }
4511 }
4512
4513 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4514         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4515 {
4516     struct wined3d_device *device = shader->device;
4517     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4518     int i;
4519     WORD int_skip;
4520
4521     find_vs_compile_args(state, shader, &args->super);
4522
4523     args->clip.boolclip_compare = 0;
4524     if (use_ps(state))
4525     {
4526         const struct wined3d_shader *ps = state->pixel_shader;
4527         const struct arb_pshader_private *shader_priv = ps->backend_data;
4528         args->ps_signature = shader_priv->input_signature_idx;
4529
4530         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4531     }
4532     else
4533     {
4534         args->ps_signature = ~0;
4535         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4536         {
4537             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4538         }
4539         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4540     }
4541
4542     if (args->clip.boolclip.clip_texcoord)
4543     {
4544         if (state->render_states[WINED3D_RS_CLIPPING])
4545             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4546         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4547     }
4548
4549     /* This forces all local boolean constants to 1 to make them stateblock independent */
4550     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4551     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4552     for(i = 0; i < MAX_CONST_B; i++)
4553     {
4554         if (state->vs_consts_b[i])
4555             args->clip.boolclip.bools |= ( 1 << i);
4556     }
4557
4558     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4559     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4560     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4561     args->vertex.samplers[3] = 0;
4562
4563     /* Skip if unused or local */
4564     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4565     /* This is about flow control, not clipping. */
4566     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4567     {
4568         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4569         return;
4570     }
4571
4572     for(i = 0; i < MAX_CONST_I; i++)
4573     {
4574         if(int_skip & (1 << i))
4575         {
4576             args->loop_ctrl[i][0] = 0;
4577             args->loop_ctrl[i][1] = 0;
4578             args->loop_ctrl[i][2] = 0;
4579         }
4580         else
4581         {
4582             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4583             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4584             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4585         }
4586     }
4587 }
4588
4589 /* GL locking is done by the caller */
4590 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4591 {
4592     struct wined3d_device *device = context->swapchain->device;
4593     struct shader_arb_priv *priv = device->shader_priv;
4594     const struct wined3d_gl_info *gl_info = context->gl_info;
4595     const struct wined3d_state *state = &device->stateBlock->state;
4596     int i;
4597
4598     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4599     if (usePS)
4600     {
4601         struct wined3d_shader *ps = state->pixel_shader;
4602         struct arb_ps_compile_args compile_args;
4603         struct arb_ps_compiled_shader *compiled;
4604
4605         TRACE("Using pixel shader %p.\n", ps);
4606         find_arb_ps_compile_args(state, ps, &compile_args);
4607         compiled = find_arb_pshader(ps, &compile_args);
4608         priv->current_fprogram_id = compiled->prgId;
4609         priv->compiled_fprog = compiled;
4610
4611         /* Bind the fragment program */
4612         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4613         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4614
4615         if(!priv->use_arbfp_fixed_func) {
4616             /* Enable OpenGL fragment programs */
4617             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4618             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4619         }
4620         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4621                 device, priv->current_fprogram_id);
4622
4623         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4624          * a 1.x and newer shader, reload the first 8 constants
4625          */
4626         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4627         {
4628             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4629             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4630             for(i = 0; i < 8; i++)
4631             {
4632                 priv->pshader_const_dirty[i] = 1;
4633             }
4634             /* Also takes care of loading local constants */
4635             shader_arb_load_constants(context, TRUE, FALSE);
4636         }
4637         else
4638         {
4639             UINT rt_height = state->fb->render_targets[0]->resource.height;
4640             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4641         }
4642
4643         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4644         if (compiled->np2fixup_info.super.active)
4645             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4646     }
4647     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4648     {
4649         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4650         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4651         * replacement shader
4652         */
4653         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4654         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4655         priv->current_fprogram_id = 0;
4656     }
4657
4658     if (useVS)
4659     {
4660         struct wined3d_shader *vs = state->vertex_shader;
4661         struct arb_vs_compile_args compile_args;
4662         struct arb_vs_compiled_shader *compiled;
4663
4664         TRACE("Using vertex shader %p\n", vs);
4665         find_arb_vs_compile_args(state, vs, &compile_args);
4666         compiled = find_arb_vshader(vs, &compile_args);
4667         priv->current_vprogram_id = compiled->prgId;
4668         priv->compiled_vprog = compiled;
4669
4670         /* Bind the vertex program */
4671         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4672         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4673
4674         /* Enable OpenGL vertex programs */
4675         glEnable(GL_VERTEX_PROGRAM_ARB);
4676         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4677         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4678         shader_arb_vs_local_constants(compiled, context, state);
4679
4680         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4681             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4682
4683             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4684             {
4685                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4686                 checkGLcall("glClampColorARB");
4687             } else {
4688                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4689             }
4690         }
4691     }
4692     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4693     {
4694         priv->current_vprogram_id = 0;
4695         glDisable(GL_VERTEX_PROGRAM_ARB);
4696         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4697     }
4698 }
4699
4700 /* GL locking is done by the caller */
4701 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4702         enum tex_types tex_type, const SIZE *ds_mask_size)
4703 {
4704     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4705     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4706     struct shader_arb_priv *priv = shader_priv;
4707     GLuint *blt_fprogram;
4708
4709     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4710     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4711     glEnable(GL_VERTEX_PROGRAM_ARB);
4712
4713     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4714     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4715     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4716     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4717     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4718 }
4719
4720 /* GL locking is done by the caller */
4721 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4722 {
4723     struct shader_arb_priv *priv = shader_priv;
4724
4725     if (priv->current_vprogram_id) {
4726         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4727         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4728
4729         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4730     }
4731     else
4732     {
4733         glDisable(GL_VERTEX_PROGRAM_ARB);
4734         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4735     }
4736
4737     if (priv->current_fprogram_id) {
4738         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4739         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4740
4741         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4742     }
4743     else if(!priv->use_arbfp_fixed_func)
4744     {
4745         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4746         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4747     }
4748 }
4749
4750 static void shader_arb_destroy(struct wined3d_shader *shader)
4751 {
4752     struct wined3d_device *device = shader->device;
4753     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4754
4755     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4756     {
4757         struct arb_pshader_private *shader_data = shader->backend_data;
4758         UINT i;
4759
4760         if(!shader_data) return; /* This can happen if a shader was never compiled */
4761
4762         if (shader_data->num_gl_shaders)
4763         {
4764             struct wined3d_context *context = context_acquire(device, NULL);
4765
4766             ENTER_GL();
4767             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4768             {
4769                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4770                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4771             }
4772             LEAVE_GL();
4773
4774             context_release(context);
4775         }
4776
4777         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4778         HeapFree(GetProcessHeap(), 0, shader_data);
4779         shader->backend_data = NULL;
4780     }
4781     else
4782     {
4783         struct arb_vshader_private *shader_data = shader->backend_data;
4784         UINT i;
4785
4786         if(!shader_data) return; /* This can happen if a shader was never compiled */
4787
4788         if (shader_data->num_gl_shaders)
4789         {
4790             struct wined3d_context *context = context_acquire(device, NULL);
4791
4792             ENTER_GL();
4793             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4794             {
4795                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4796                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4797             }
4798             LEAVE_GL();
4799
4800             context_release(context);
4801         }
4802
4803         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4804         HeapFree(GetProcessHeap(), 0, shader_data);
4805         shader->backend_data = NULL;
4806     }
4807 }
4808
4809 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4810 {
4811     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4812     return compare_sig(key, e->sig);
4813 }
4814
4815 static const struct wine_rb_functions sig_tree_functions =
4816 {
4817     wined3d_rb_alloc,
4818     wined3d_rb_realloc,
4819     wined3d_rb_free,
4820     sig_tree_compare
4821 };
4822
4823 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4824 {
4825     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4826
4827     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4828             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4829     if (!priv->vshader_const_dirty)
4830         goto fail;
4831     memset(priv->vshader_const_dirty, 1,
4832            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4833
4834     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4835             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4836     if (!priv->pshader_const_dirty)
4837         goto fail;
4838     memset(priv->pshader_const_dirty, 1,
4839             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4840
4841     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4842     {
4843         ERR("RB tree init failed\n");
4844         goto fail;
4845     }
4846     device->shader_priv = priv;
4847     return WINED3D_OK;
4848
4849 fail:
4850     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4851     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4852     HeapFree(GetProcessHeap(), 0, priv);
4853     return E_OUTOFMEMORY;
4854 }
4855
4856 static void release_signature(struct wine_rb_entry *entry, void *context)
4857 {
4858     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4859     int i;
4860     for(i = 0; i < MAX_REG_INPUT; i++)
4861     {
4862         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4863     }
4864     HeapFree(GetProcessHeap(), 0, sig->sig);
4865     HeapFree(GetProcessHeap(), 0, sig);
4866 }
4867
4868 /* Context activation is done by the caller. */
4869 static void shader_arb_free(struct wined3d_device *device)
4870 {
4871     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4872     struct shader_arb_priv *priv = device->shader_priv;
4873     int i;
4874
4875     ENTER_GL();
4876     if(priv->depth_blt_vprogram_id) {
4877         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4878     }
4879     for (i = 0; i < tex_type_count; ++i)
4880     {
4881         if (priv->depth_blt_fprogram_id_full[i])
4882         {
4883             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4884         }
4885         if (priv->depth_blt_fprogram_id_masked[i])
4886         {
4887             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4888         }
4889     }
4890     LEAVE_GL();
4891
4892     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4893     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4894     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4895     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4896 }
4897
4898 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4899 {
4900     struct shader_arb_priv *priv = shader_priv;
4901
4902     if (priv->last_context == context)
4903         priv->last_context = NULL;
4904 }
4905
4906 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4907 {
4908     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4909     {
4910         DWORD vs_consts;
4911
4912         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4913          * for vertex programs. If the native limit is less than that it's
4914          * not very useful, and e.g. Mesa swrast returns 0, probably to
4915          * indicate it's a software implementation. */
4916         if (gl_info->limits.arb_vs_native_constants < 96)
4917             vs_consts = gl_info->limits.arb_vs_float_constants;
4918         else
4919             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4920
4921         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4922         {
4923             caps->VertexShaderVersion = 3;
4924             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4925         }
4926         else if (vs_consts >= 256)
4927         {
4928             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4929             caps->VertexShaderVersion = 2;
4930             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4931         }
4932         else
4933         {
4934             caps->VertexShaderVersion = 1;
4935             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4936         }
4937         caps->MaxVertexShaderConst = vs_consts;
4938     }
4939     else
4940     {
4941         caps->VertexShaderVersion = 0;
4942         caps->MaxVertexShaderConst = 0;
4943     }
4944
4945     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4946     {
4947         DWORD ps_consts;
4948
4949         /* Similar as above for vertex programs, but the minimum for fragment
4950          * programs is 24. */
4951         if (gl_info->limits.arb_ps_native_constants < 24)
4952             ps_consts = gl_info->limits.arb_ps_float_constants;
4953         else
4954             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4955
4956         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4957         {
4958             caps->PixelShaderVersion = 3;
4959             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4960         }
4961         else if (ps_consts >= 32)
4962         {
4963             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4964             caps->PixelShaderVersion = 2;
4965             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4966         }
4967         else
4968         {
4969             caps->PixelShaderVersion = 1;
4970             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4971         }
4972         caps->PixelShader1xMaxValue = 8.0f;
4973         caps->MaxPixelShaderConst = ps_consts;
4974     }
4975     else
4976     {
4977         caps->PixelShaderVersion = 0;
4978         caps->PixelShader1xMaxValue = 0.0f;
4979         caps->MaxPixelShaderConst = 0;
4980     }
4981
4982     caps->VSClipping = use_nv_clip(gl_info);
4983 }
4984
4985 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4986 {
4987     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4988     {
4989         TRACE("Checking support for color_fixup:\n");
4990         dump_color_fixup_desc(fixup);
4991     }
4992
4993     /* We support everything except complex conversions. */
4994     if (!is_complex_fixup(fixup))
4995     {
4996         TRACE("[OK]\n");
4997         return TRUE;
4998     }
4999
5000     TRACE("[FAILED]\n");
5001     return FALSE;
5002 }
5003
5004 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5005     DWORD shift;
5006     char write_mask[20], regstr[50];
5007     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5008     BOOL is_color = FALSE;
5009     const struct wined3d_shader_dst_param *dst;
5010
5011     if (!ins->dst_count) return;
5012
5013     dst = &ins->dst[0];
5014     shift = dst->shift;
5015     if (!shift) return; /* Saturate alone is handled by the instructions */
5016
5017     shader_arb_get_write_mask(ins, dst, write_mask);
5018     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5019
5020     /* Generate a line that does the output modifier computation
5021      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5022      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5023      */
5024     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5025                    regstr, write_mask, regstr, shift_tab[shift]);
5026 }
5027
5028 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5029 {
5030     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
5031     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
5032     /* WINED3DSIH_AND           */ NULL,
5033     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5034     /* WINED3DSIH_BREAK         */ shader_hw_break,
5035     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5036     /* WINED3DSIH_BREAKP        */ NULL,
5037     /* WINED3DSIH_CALL          */ shader_hw_call,
5038     /* WINED3DSIH_CALLNZ        */ NULL,
5039     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5040     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5041     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5042     /* WINED3DSIH_CUT           */ NULL,
5043     /* WINED3DSIH_DCL           */ NULL,
5044     /* WINED3DSIH_DEF           */ NULL,
5045     /* WINED3DSIH_DEFB          */ NULL,
5046     /* WINED3DSIH_DEFI          */ NULL,
5047     /* WINED3DSIH_DIV           */ NULL,
5048     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5049     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5050     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5051     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5052     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5053     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5054     /* WINED3DSIH_ELSE          */ shader_hw_else,
5055     /* WINED3DSIH_EMIT          */ NULL,
5056     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5057     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5058     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5059     /* WINED3DSIH_EQ            */ NULL,
5060     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5061     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5062     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5063     /* WINED3DSIH_FTOI          */ NULL,
5064     /* WINED3DSIH_GE            */ NULL,
5065     /* WINED3DSIH_IADD          */ NULL,
5066     /* WINED3DSIH_IEQ           */ NULL,
5067     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5068     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5069     /* WINED3DSIH_IGE           */ NULL,
5070     /* WINED3DSIH_IMUL          */ NULL,
5071     /* WINED3DSIH_ITOF          */ NULL,
5072     /* WINED3DSIH_LABEL         */ shader_hw_label,
5073     /* WINED3DSIH_LD            */ NULL,
5074     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5075     /* WINED3DSIH_LOG           */ shader_hw_log,
5076     /* WINED3DSIH_LOGP          */ shader_hw_log,
5077     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5078     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5079     /* WINED3DSIH_LT            */ NULL,
5080     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5081     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5082     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5083     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5084     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5085     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5086     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5087     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5088     /* WINED3DSIH_MOV           */ shader_hw_mov,
5089     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5090     /* WINED3DSIH_MOVC          */ NULL,
5091     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5092     /* WINED3DSIH_NOP           */ shader_hw_nop,
5093     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5094     /* WINED3DSIH_PHASE         */ NULL,
5095     /* WINED3DSIH_POW           */ shader_hw_pow,
5096     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5097     /* WINED3DSIH_REP           */ shader_hw_rep,
5098     /* WINED3DSIH_RET           */ shader_hw_ret,
5099     /* WINED3DSIH_ROUND_NI      */ NULL,
5100     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5101     /* WINED3DSIH_SAMPLE        */ NULL,
5102     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5103     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5104     /* WINED3DSIH_SETP          */ NULL,
5105     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5106     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5107     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5108     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5109     /* WINED3DSIH_SQRT          */ NULL,
5110     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5111     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5112     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5113     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5114     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5115     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5116     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5117     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5118     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5119     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5120     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5121     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5122     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5123     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5124     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5125     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5126     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5127     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5128     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5129     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5130     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5131     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5132     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5133     /* WINED3DSIH_UDIV          */ NULL,
5134     /* WINED3DSIH_USHR          */ NULL,
5135     /* WINED3DSIH_UTOF          */ NULL,
5136     /* WINED3DSIH_XOR           */ NULL,
5137 };
5138
5139 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5140         const struct wined3d_shader *shader, DWORD idx)
5141 {
5142     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5143     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5144     const struct wined3d_shader_lconst *constant;
5145     WORD bools = 0;
5146     WORD flag = (1 << idx);
5147     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5148
5149     if (reg_maps->local_bool_consts & flag)
5150     {
5151         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5152         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5153         {
5154             if (constant->idx == idx)
5155             {
5156                 return constant->value[0];
5157             }
5158         }
5159         ERR("Local constant not found\n");
5160         return FALSE;
5161     }
5162     else
5163     {
5164         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5165         else bools = priv->cur_ps_args->bools;
5166         return bools & flag;
5167     }
5168 }
5169
5170 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5171         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5172 {
5173     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5174     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5175
5176     /* Integer constants can either be a local constant, or they can be stored in the shader
5177      * type specific compile args. */
5178     if (reg_maps->local_int_consts & (1 << idx))
5179     {
5180         const struct wined3d_shader_lconst *constant;
5181
5182         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5183         {
5184             if (constant->idx == idx)
5185             {
5186                 loop_control->count = constant->value[0];
5187                 loop_control->start = constant->value[1];
5188                 /* Step is signed. */
5189                 loop_control->step = (int)constant->value[2];
5190                 return;
5191             }
5192         }
5193         /* If this happens the flag was set incorrectly */
5194         ERR("Local constant not found\n");
5195         loop_control->count = 0;
5196         loop_control->start = 0;
5197         loop_control->step = 0;
5198         return;
5199     }
5200
5201     switch (reg_maps->shader_version.type)
5202     {
5203         case WINED3D_SHADER_TYPE_VERTEX:
5204             /* Count and aL start value are unsigned */
5205             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5206             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5207             /* Step is signed. */
5208             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5209             break;
5210
5211         case WINED3D_SHADER_TYPE_PIXEL:
5212             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5213             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5214             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5215             break;
5216
5217         default:
5218             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5219             break;
5220     }
5221 }
5222
5223 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5224 {
5225     unsigned int i;
5226     struct wined3d_shader_dst_param *dst_param = NULL;
5227     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5228     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5229     if(!rec)
5230     {
5231         ERR("Out of memory\n");
5232         return;
5233     }
5234
5235     rec->ins = *ins;
5236     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5237     if(!dst_param) goto free;
5238     *dst_param = *ins->dst;
5239     if(ins->dst->reg.rel_addr)
5240     {
5241         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5242         if(!rel_addr) goto free;
5243         *rel_addr = *ins->dst->reg.rel_addr;
5244         dst_param->reg.rel_addr = rel_addr;
5245     }
5246     rec->ins.dst = dst_param;
5247
5248     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5249     if(!src_param) goto free;
5250     for(i = 0; i < ins->src_count; i++)
5251     {
5252         src_param[i] = ins->src[i];
5253         if(ins->src[i].reg.rel_addr)
5254         {
5255             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5256             if(!rel_addr) goto free;
5257             *rel_addr = *ins->src[i].reg.rel_addr;
5258             src_param[i].reg.rel_addr = rel_addr;
5259         }
5260     }
5261     rec->ins.src = src_param;
5262     list_add_tail(list, &rec->entry);
5263     return;
5264
5265 free:
5266     ERR("Out of memory\n");
5267     if(dst_param)
5268     {
5269         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5270         HeapFree(GetProcessHeap(), 0, dst_param);
5271     }
5272     if(src_param)
5273     {
5274         for(i = 0; i < ins->src_count; i++)
5275         {
5276             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5277         }
5278         HeapFree(GetProcessHeap(), 0, src_param);
5279     }
5280     HeapFree(GetProcessHeap(), 0, rec);
5281 }
5282
5283 static void free_recorded_instruction(struct list *list)
5284 {
5285     struct recorded_instruction *rec_ins, *entry2;
5286     unsigned int i;
5287
5288     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5289     {
5290         list_remove(&rec_ins->entry);
5291         if(rec_ins->ins.dst)
5292         {
5293             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5294             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5295         }
5296         if(rec_ins->ins.src)
5297         {
5298             for(i = 0; i < rec_ins->ins.src_count; i++)
5299             {
5300                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5301             }
5302             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5303         }
5304         HeapFree(GetProcessHeap(), 0, rec_ins);
5305     }
5306 }
5307
5308 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5309     SHADER_HANDLER hw_fct;
5310     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5311     const struct wined3d_shader *shader = ins->ctx->shader;
5312     struct control_frame *control_frame;
5313     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5314     BOOL bool_const;
5315
5316     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5317     {
5318         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5319         list_add_head(&priv->control_frames, &control_frame->entry);
5320
5321         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5322         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5323
5324         if(priv->target_version >= NV2)
5325         {
5326             control_frame->no.loop = priv->num_loops++;
5327             priv->loop_depth++;
5328         }
5329         else
5330         {
5331             /* Don't bother recording when we're in a not used if branch */
5332             if(priv->muted)
5333             {
5334                 return;
5335             }
5336
5337             if(!priv->recording)
5338             {
5339                 list_init(&priv->record);
5340                 priv->recording = TRUE;
5341                 control_frame->outer_loop = TRUE;
5342                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5343                 return; /* Instruction is handled */
5344             }
5345             /* Record this loop in the outer loop's recording */
5346         }
5347     }
5348     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5349     {
5350         if(priv->target_version >= NV2)
5351         {
5352             /* Nothing to do. The control frame is popped after the HW instr handler */
5353         }
5354         else
5355         {
5356             struct list *e = list_head(&priv->control_frames);
5357             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5358             list_remove(&control_frame->entry);
5359
5360             if(control_frame->outer_loop)
5361             {
5362                 unsigned int iteration;
5363                 int aL = 0;
5364                 struct list copy;
5365
5366                 /* Turn off recording before playback */
5367                 priv->recording = FALSE;
5368
5369                 /* Move the recorded instructions to a separate list and get them out of the private data
5370                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5371                  * be recorded again, thus priv->record might be overwritten
5372                  */
5373                 list_init(&copy);
5374                 list_move_tail(&copy, &priv->record);
5375                 list_init(&priv->record);
5376
5377                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5378                 {
5379                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5380                                    control_frame->loop_control.count, control_frame->loop_control.start,
5381                                    control_frame->loop_control.step);
5382                     aL = control_frame->loop_control.start;
5383                 }
5384                 else
5385                 {
5386                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5387                 }
5388
5389                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5390                 {
5391                     struct recorded_instruction *rec_ins;
5392                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5393                     {
5394                         priv->aL = aL;
5395                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5396                     }
5397                     else
5398                     {
5399                         shader_addline(buffer, "#Iteration %u\n", iteration);
5400                     }
5401
5402                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5403                     {
5404                         shader_arb_handle_instruction(&rec_ins->ins);
5405                     }
5406
5407                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5408                     {
5409                         aL += control_frame->loop_control.step;
5410                     }
5411                 }
5412                 shader_addline(buffer, "#end loop/rep\n");
5413
5414                 free_recorded_instruction(&copy);
5415                 HeapFree(GetProcessHeap(), 0, control_frame);
5416                 return; /* Instruction is handled */
5417             }
5418             else
5419             {
5420                 /* This is a nested loop. Proceed to the normal recording function */
5421                 HeapFree(GetProcessHeap(), 0, control_frame);
5422             }
5423         }
5424     }
5425
5426     if(priv->recording)
5427     {
5428         record_instruction(&priv->record, ins);
5429         return;
5430     }
5431
5432     /* boolean if */
5433     if(ins->handler_idx == WINED3DSIH_IF)
5434     {
5435         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5436         list_add_head(&priv->control_frames, &control_frame->entry);
5437         control_frame->type = IF;
5438
5439         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5440         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5441         if (!priv->muted && !bool_const)
5442         {
5443             shader_addline(buffer, "#if(FALSE){\n");
5444             priv->muted = TRUE;
5445             control_frame->muting = TRUE;
5446         }
5447         else shader_addline(buffer, "#if(TRUE) {\n");
5448
5449         return; /* Instruction is handled */
5450     }
5451     else if(ins->handler_idx == WINED3DSIH_IFC)
5452     {
5453         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5454         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5455         control_frame->type = IFC;
5456         control_frame->no.ifc = priv->num_ifcs++;
5457         list_add_head(&priv->control_frames, &control_frame->entry);
5458     }
5459     else if(ins->handler_idx == WINED3DSIH_ELSE)
5460     {
5461         struct list *e = list_head(&priv->control_frames);
5462         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5463
5464         if(control_frame->type == IF)
5465         {
5466             shader_addline(buffer, "#} else {\n");
5467             if(!priv->muted && !control_frame->muting)
5468             {
5469                 priv->muted = TRUE;
5470                 control_frame->muting = TRUE;
5471             }
5472             else if(control_frame->muting) priv->muted = FALSE;
5473             return; /* Instruction is handled. */
5474         }
5475         /* In case of an ifc, generate a HW shader instruction */
5476     }
5477     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5478     {
5479         struct list *e = list_head(&priv->control_frames);
5480         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5481
5482         if(control_frame->type == IF)
5483         {
5484             shader_addline(buffer, "#} endif\n");
5485             if(control_frame->muting) priv->muted = FALSE;
5486             list_remove(&control_frame->entry);
5487             HeapFree(GetProcessHeap(), 0, control_frame);
5488             return; /* Instruction is handled */
5489         }
5490     }
5491
5492     if(priv->muted) return;
5493
5494     /* Select handler */
5495     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5496
5497     /* Unhandled opcode */
5498     if (!hw_fct)
5499     {
5500         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5501         return;
5502     }
5503     hw_fct(ins);
5504
5505     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5506     {
5507         struct list *e = list_head(&priv->control_frames);
5508         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5509         list_remove(&control_frame->entry);
5510         HeapFree(GetProcessHeap(), 0, control_frame);
5511         priv->loop_depth--;
5512     }
5513     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5514     {
5515         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5516         struct list *e = list_head(&priv->control_frames);
5517         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5518         list_remove(&control_frame->entry);
5519         HeapFree(GetProcessHeap(), 0, control_frame);
5520     }
5521
5522
5523     shader_arb_add_instruction_modifiers(ins);
5524 }
5525
5526 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5527 {
5528     shader_arb_handle_instruction,
5529     shader_arb_select,
5530     shader_arb_select_depth_blt,
5531     shader_arb_deselect_depth_blt,
5532     shader_arb_update_float_vertex_constants,
5533     shader_arb_update_float_pixel_constants,
5534     shader_arb_load_constants,
5535     shader_arb_load_np2fixup_constants,
5536     shader_arb_destroy,
5537     shader_arb_alloc,
5538     shader_arb_free,
5539     shader_arb_context_destroyed,
5540     shader_arb_get_caps,
5541     shader_arb_color_fixup_supported,
5542 };
5543
5544 /* ARB_fragment_program fixed function pipeline replacement definitions */
5545 #define ARB_FFP_CONST_TFACTOR           0
5546 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5547 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5548 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5549 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5550
5551 struct arbfp_ffp_desc
5552 {
5553     struct ffp_frag_desc parent;
5554     GLuint shader;
5555     unsigned int num_textures_used;
5556 };
5557
5558 /* Context activation and GL locking are done by the caller. */
5559 static void arbfp_enable(BOOL enable)
5560 {
5561     if(enable) {
5562         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5563         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5564     } else {
5565         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5566         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5567     }
5568 }
5569
5570 static HRESULT arbfp_alloc(struct wined3d_device *device)
5571 {
5572     struct shader_arb_priv *priv;
5573     /* Share private data between the shader backend and the pipeline replacement, if both
5574      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5575      * if no pixel shader is bound or not
5576      */
5577     if (device->shader_backend == &arb_program_shader_backend)
5578     {
5579         device->fragment_priv = device->shader_priv;
5580     }
5581     else
5582     {
5583         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5584         if (!device->fragment_priv) return E_OUTOFMEMORY;
5585     }
5586     priv = device->fragment_priv;
5587     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5588     {
5589         ERR("Failed to initialize rbtree.\n");
5590         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5591         return E_OUTOFMEMORY;
5592     }
5593     priv->use_arbfp_fixed_func = TRUE;
5594     return WINED3D_OK;
5595 }
5596
5597 /* Context activation is done by the caller. */
5598 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5599 {
5600     const struct wined3d_gl_info *gl_info = context;
5601     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5602
5603     ENTER_GL();
5604     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5605     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5606     HeapFree(GetProcessHeap(), 0, entry_arb);
5607     LEAVE_GL();
5608 }
5609
5610 /* Context activation is done by the caller. */
5611 static void arbfp_free(struct wined3d_device *device)
5612 {
5613     struct shader_arb_priv *priv = device->fragment_priv;
5614
5615     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5616     priv->use_arbfp_fixed_func = FALSE;
5617
5618     if (device->shader_backend != &arb_program_shader_backend)
5619     {
5620         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5621     }
5622 }
5623
5624 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5625 {
5626     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5627     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5628                            WINED3DTEXOPCAPS_SELECTARG1                  |
5629                            WINED3DTEXOPCAPS_SELECTARG2                  |
5630                            WINED3DTEXOPCAPS_MODULATE4X                  |
5631                            WINED3DTEXOPCAPS_MODULATE2X                  |
5632                            WINED3DTEXOPCAPS_MODULATE                    |
5633                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5634                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5635                            WINED3DTEXOPCAPS_ADD                         |
5636                            WINED3DTEXOPCAPS_SUBTRACT                    |
5637                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5638                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5639                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5640                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5641                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5642                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5643                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5644                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5645                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5646                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5647                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5648                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5649                            WINED3DTEXOPCAPS_LERP                        |
5650                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5651                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5652
5653     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5654
5655     caps->MaxTextureBlendStages   = 8;
5656     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5657 }
5658
5659 static void state_texfactor_arbfp(struct wined3d_context *context,
5660         const struct wined3d_state *state, DWORD state_id)
5661 {
5662     struct wined3d_device *device = context->swapchain->device;
5663     const struct wined3d_gl_info *gl_info = context->gl_info;
5664     float col[4];
5665
5666     /* Don't load the parameter if we're using an arbfp pixel shader,
5667      * otherwise we'll overwrite application provided constants. */
5668     if (device->shader_backend == &arb_program_shader_backend)
5669     {
5670         struct shader_arb_priv *priv;
5671
5672         if (use_ps(state)) return;
5673
5674         priv = device->shader_priv;
5675         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5676         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5677     }
5678
5679     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5680     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5681     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5682 }
5683
5684 static void state_arb_specularenable(struct wined3d_context *context,
5685         const struct wined3d_state *state, DWORD state_id)
5686 {
5687     struct wined3d_device *device = context->swapchain->device;
5688     const struct wined3d_gl_info *gl_info = context->gl_info;
5689     float col[4];
5690
5691     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5692      * application provided constants
5693      */
5694     if (device->shader_backend == &arb_program_shader_backend)
5695     {
5696         struct shader_arb_priv *priv;
5697
5698         if (use_ps(state)) return;
5699
5700         priv = device->shader_priv;
5701         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5702         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5703     }
5704
5705     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5706     {
5707         /* The specular color has no alpha */
5708         col[0] = 1.0f; col[1] = 1.0f;
5709         col[2] = 1.0f; col[3] = 0.0f;
5710     } else {
5711         col[0] = 0.0f; col[1] = 0.0f;
5712         col[2] = 0.0f; col[3] = 0.0f;
5713     }
5714     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5715     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5716 }
5717
5718 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5719 {
5720     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5721     struct wined3d_device *device = context->swapchain->device;
5722     const struct wined3d_gl_info *gl_info = context->gl_info;
5723     float mat[2][2];
5724
5725     if (use_ps(state))
5726     {
5727         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5728         {
5729             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5730              * anyway
5731              */
5732             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5733                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5734         }
5735
5736         if(device->shader_backend == &arb_program_shader_backend) {
5737             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5738             return;
5739         }
5740     }
5741     else if (device->shader_backend == &arb_program_shader_backend)
5742     {
5743         struct shader_arb_priv *priv = device->shader_priv;
5744         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5745         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5746     }
5747
5748     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5749     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5750     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5751     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5752
5753     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5754     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5755 }
5756
5757 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5758         const struct wined3d_state *state, DWORD state_id)
5759 {
5760     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5761     struct wined3d_device *device = context->swapchain->device;
5762     const struct wined3d_gl_info *gl_info = context->gl_info;
5763     float param[4];
5764
5765     if (use_ps(state))
5766     {
5767         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5768         {
5769             /* The pixel shader has to know the luminance offset. Do a constants update if it
5770              * isn't scheduled anyway
5771              */
5772             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5773                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5774         }
5775
5776         if(device->shader_backend == &arb_program_shader_backend) {
5777             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5778             return;
5779         }
5780     }
5781     else if (device->shader_backend == &arb_program_shader_backend)
5782     {
5783         struct shader_arb_priv *priv = device->shader_priv;
5784         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5785         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5786     }
5787
5788     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5789     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5790     param[2] = 0.0f;
5791     param[3] = 0.0f;
5792
5793     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5794     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5795 }
5796
5797 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5798 {
5799     const char *ret;
5800
5801     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5802
5803     switch(arg & WINED3DTA_SELECTMASK) {
5804         case WINED3DTA_DIFFUSE:
5805             ret = "fragment.color.primary"; break;
5806
5807         case WINED3DTA_CURRENT:
5808             if (!stage) ret = "fragment.color.primary";
5809             else ret = "ret";
5810             break;
5811
5812         case WINED3DTA_TEXTURE:
5813             switch(stage) {
5814                 case 0: ret = "tex0"; break;
5815                 case 1: ret = "tex1"; break;
5816                 case 2: ret = "tex2"; break;
5817                 case 3: ret = "tex3"; break;
5818                 case 4: ret = "tex4"; break;
5819                 case 5: ret = "tex5"; break;
5820                 case 6: ret = "tex6"; break;
5821                 case 7: ret = "tex7"; break;
5822                 default: ret = "unknown texture";
5823             }
5824             break;
5825
5826         case WINED3DTA_TFACTOR:
5827             ret = "tfactor"; break;
5828
5829         case WINED3DTA_SPECULAR:
5830             ret = "fragment.color.secondary"; break;
5831
5832         case WINED3DTA_TEMP:
5833             ret = "tempreg"; break;
5834
5835         case WINED3DTA_CONSTANT:
5836             FIXME("Implement perstage constants\n");
5837             switch(stage) {
5838                 case 0: ret = "const0"; break;
5839                 case 1: ret = "const1"; break;
5840                 case 2: ret = "const2"; break;
5841                 case 3: ret = "const3"; break;
5842                 case 4: ret = "const4"; break;
5843                 case 5: ret = "const5"; break;
5844                 case 6: ret = "const6"; break;
5845                 case 7: ret = "const7"; break;
5846                 default: ret = "unknown constant";
5847             }
5848             break;
5849
5850         default:
5851             return "unknown";
5852     }
5853
5854     if(arg & WINED3DTA_COMPLEMENT) {
5855         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5856         if(argnum == 0) ret = "arg0";
5857         if(argnum == 1) ret = "arg1";
5858         if(argnum == 2) ret = "arg2";
5859     }
5860     if(arg & WINED3DTA_ALPHAREPLICATE) {
5861         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5862         if(argnum == 0) ret = "arg0";
5863         if(argnum == 1) ret = "arg1";
5864         if(argnum == 2) ret = "arg2";
5865     }
5866     return ret;
5867 }
5868
5869 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5870         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5871 {
5872     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5873     unsigned int mul = 1;
5874     BOOL mul_final_dest = FALSE;
5875
5876     if(color && alpha) dstmask = "";
5877     else if(color) dstmask = ".xyz";
5878     else dstmask = ".w";
5879
5880     if(dst == tempreg) dstreg = "tempreg";
5881     else dstreg = "ret";
5882
5883     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5884     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5885     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5886
5887     switch (op)
5888     {
5889         case WINED3D_TOP_DISABLE:
5890             if (!stage)
5891                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5892             break;
5893
5894         case WINED3D_TOP_SELECT_ARG2:
5895             arg1 = arg2;
5896             /* FALLTHROUGH */
5897         case WINED3D_TOP_SELECT_ARG1:
5898             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5899             break;
5900
5901         case WINED3D_TOP_MODULATE_4X:
5902             mul = 2;
5903             /* FALLTHROUGH */
5904         case WINED3D_TOP_MODULATE_2X:
5905             mul *= 2;
5906             if (!strcmp(dstreg, "result.color"))
5907             {
5908                 dstreg = "ret";
5909                 mul_final_dest = TRUE;
5910             }
5911             /* FALLTHROUGH */
5912         case WINED3D_TOP_MODULATE:
5913             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5914             break;
5915
5916         case WINED3D_TOP_ADD_SIGNED_2X:
5917             mul = 2;
5918             if (!strcmp(dstreg, "result.color"))
5919             {
5920                 dstreg = "ret";
5921                 mul_final_dest = TRUE;
5922             }
5923             /* FALLTHROUGH */
5924         case WINED3D_TOP_ADD_SIGNED:
5925             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5926             arg2 = "arg2";
5927             /* FALLTHROUGH */
5928         case WINED3D_TOP_ADD:
5929             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5930             break;
5931
5932         case WINED3D_TOP_SUBTRACT:
5933             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5934             break;
5935
5936         case WINED3D_TOP_ADD_SMOOTH:
5937             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5938             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5939             break;
5940
5941         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
5942             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5943             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5944             break;
5945         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
5946             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5947             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5948             break;
5949         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
5950             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5951             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5952             break;
5953         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
5954             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5955             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5956             break;
5957
5958         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
5959             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5960             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5961             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5962             break;
5963
5964         /* D3DTOP_PREMODULATE ???? */
5965
5966         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
5967             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5968             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5969             break;
5970         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
5971             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5972             break;
5973         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
5974             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5975             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5976             break;
5977         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
5978             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5979             break;
5980
5981         case WINED3D_TOP_DOTPRODUCT3:
5982             mul = 4;
5983             if (!strcmp(dstreg, "result.color"))
5984             {
5985                 dstreg = "ret";
5986                 mul_final_dest = TRUE;
5987             }
5988             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5989             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5990             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5991             break;
5992
5993         case WINED3D_TOP_MULTIPLY_ADD:
5994             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5995             break;
5996
5997         case WINED3D_TOP_LERP:
5998             /* The msdn is not quite right here */
5999             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6000             break;
6001
6002         case WINED3D_TOP_BUMPENVMAP:
6003         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6004             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6005             break;
6006
6007         default:
6008             FIXME("Unhandled texture op %08x\n", op);
6009     }
6010
6011     if(mul == 2) {
6012         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6013     } else if(mul == 4) {
6014         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6015     }
6016 }
6017
6018 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6019 {
6020     unsigned int stage;
6021     struct wined3d_shader_buffer buffer;
6022     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6023     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6024     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6025     UINT lowest_disabled_stage;
6026     const char *textype;
6027     const char *instr, *sat;
6028     char colorcor_dst[8];
6029     GLuint ret;
6030     DWORD arg0, arg1, arg2;
6031     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6032     BOOL op_equal;
6033     const char *final_combiner_src = "ret";
6034     GLint pos;
6035
6036     /* Find out which textures are read */
6037     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6038     {
6039         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6040             break;
6041         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6042         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6043         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6044         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6045         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6046         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6047
6048         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6049             tex_read[stage] = TRUE;
6050         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6051             tex_read[stage] = TRUE;
6052         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6053         {
6054             bump_used[stage] = TRUE;
6055             tex_read[stage] = TRUE;
6056         }
6057         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6058         {
6059             bump_used[stage] = TRUE;
6060             tex_read[stage] = TRUE;
6061             luminance_used[stage] = TRUE;
6062         }
6063         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6064         {
6065             tfactor_used = TRUE;
6066         }
6067
6068         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6069             tfactor_used = TRUE;
6070         }
6071
6072         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6073         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6074             tempreg_used = TRUE;
6075         }
6076
6077         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6078             continue;
6079         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6080         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6081         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6082         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6083         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6084         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6085
6086         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6087             tempreg_used = TRUE;
6088         }
6089         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6090             tfactor_used = TRUE;
6091         }
6092     }
6093     lowest_disabled_stage = stage;
6094
6095     /* Shader header */
6096     if (!shader_buffer_init(&buffer))
6097     {
6098         ERR("Failed to initialize shader buffer.\n");
6099         return 0;
6100     }
6101
6102     shader_addline(&buffer, "!!ARBfp1.0\n");
6103
6104     switch(settings->fog) {
6105         case FOG_OFF:                                                         break;
6106         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6107         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6108         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6109         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6110     }
6111
6112     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6113     shader_addline(&buffer, "TEMP TMP;\n");
6114     shader_addline(&buffer, "TEMP ret;\n");
6115     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6116     shader_addline(&buffer, "TEMP arg0;\n");
6117     shader_addline(&buffer, "TEMP arg1;\n");
6118     shader_addline(&buffer, "TEMP arg2;\n");
6119     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6120         if(!tex_read[stage]) continue;
6121         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6122         if(!bump_used[stage]) continue;
6123         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6124         if(!luminance_used[stage]) continue;
6125         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6126     }
6127     if(tfactor_used) {
6128         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6129     }
6130         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6131
6132     if(settings->sRGB_write) {
6133         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6134                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6135         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6136                        srgb_sub_high, 0.0, 0.0, 0.0);
6137     }
6138
6139     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6140         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6141
6142     /* Generate texture sampling instructions) */
6143     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6144     {
6145         if (!tex_read[stage])
6146             continue;
6147
6148         switch(settings->op[stage].tex_type) {
6149             case tex_1d:                    textype = "1D";     break;
6150             case tex_2d:                    textype = "2D";     break;
6151             case tex_3d:                    textype = "3D";     break;
6152             case tex_cube:                  textype = "CUBE";   break;
6153             case tex_rect:                  textype = "RECT";   break;
6154             default: textype = "unexpected_textype";   break;
6155         }
6156
6157         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6158                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6159             sat = "";
6160         else
6161             sat = "_SAT";
6162
6163         if(settings->op[stage].projected == proj_none) {
6164             instr = "TEX";
6165         } else if(settings->op[stage].projected == proj_count4 ||
6166                   settings->op[stage].projected == proj_count3) {
6167             instr = "TXP";
6168         } else {
6169             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6170             instr = "TXP";
6171         }
6172
6173         if (stage > 0
6174                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6175                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6176         {
6177             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6178             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6179             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6180             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6181
6182             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6183              * so multiply the displacement with the dividing parameter before passing it to TXP
6184              */
6185             if (settings->op[stage].projected != proj_none) {
6186                 if(settings->op[stage].projected == proj_count4) {
6187                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6188                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6189                 } else {
6190                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6191                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6192                 }
6193             } else {
6194                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6195             }
6196
6197             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6198                     instr, sat, stage, stage, textype);
6199             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6200             {
6201                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6202                                stage - 1, stage - 1, stage - 1);
6203                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6204             }
6205         } else if(settings->op[stage].projected == proj_count3) {
6206             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6207             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6208             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6209                             instr, sat, stage, stage, textype);
6210         } else {
6211             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6212                             instr, sat, stage, stage, stage, textype);
6213         }
6214
6215         sprintf(colorcor_dst, "tex%u", stage);
6216         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6217                 settings->op[stage].color_fixup);
6218     }
6219
6220     /* Generate the main shader */
6221     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6222     {
6223         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6224         {
6225             if (!stage)
6226                 final_combiner_src = "fragment.color.primary";
6227             break;
6228         }
6229
6230         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6231                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6232             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6233         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6234                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6235             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6236         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6237                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6238             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6239         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6240                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6241             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6242         else
6243             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6244                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6245                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6246                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6247
6248         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6249         {
6250             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6251                           settings->op[stage].cop, settings->op[stage].carg0,
6252                           settings->op[stage].carg1, settings->op[stage].carg2);
6253             if (!stage)
6254                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6255         }
6256         else if (op_equal)
6257         {
6258             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6259                           settings->op[stage].cop, settings->op[stage].carg0,
6260                           settings->op[stage].carg1, settings->op[stage].carg2);
6261         } else {
6262             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6263                           settings->op[stage].cop, settings->op[stage].carg0,
6264                           settings->op[stage].carg1, settings->op[stage].carg2);
6265             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6266                           settings->op[stage].aop, settings->op[stage].aarg0,
6267                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6268         }
6269     }
6270
6271     if(settings->sRGB_write) {
6272         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6273         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6274         shader_addline(&buffer, "MOV result.color, ret;\n");
6275     } else {
6276         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6277     }
6278
6279     /* Footer */
6280     shader_addline(&buffer, "END\n");
6281
6282     /* Generate the shader */
6283     GL_EXTCALL(glGenProgramsARB(1, &ret));
6284     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6285     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6286             strlen(buffer.buffer), buffer.buffer));
6287     checkGLcall("glProgramStringARB()");
6288
6289     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6290     if (pos != -1)
6291     {
6292         FIXME("Fragment program error at position %d: %s\n\n", pos,
6293               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6294         shader_arb_dump_program_source(buffer.buffer);
6295     }
6296     else
6297     {
6298         GLint native;
6299
6300         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6301         checkGLcall("glGetProgramivARB()");
6302         if (!native) WARN("Program exceeds native resource limits.\n");
6303     }
6304
6305     shader_buffer_free(&buffer);
6306     return ret;
6307 }
6308
6309 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6310 {
6311     const struct wined3d_device *device = context->swapchain->device;
6312     const struct wined3d_gl_info *gl_info = context->gl_info;
6313     struct shader_arb_priv *priv = device->fragment_priv;
6314     BOOL use_vshader = use_vs(state);
6315     BOOL use_pshader = use_ps(state);
6316     struct ffp_frag_settings settings;
6317     const struct arbfp_ffp_desc *desc;
6318     unsigned int i;
6319
6320     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6321
6322     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6323     {
6324         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6325         {
6326             /* Reload fixed function constants since they collide with the
6327              * pixel shader constants. */
6328             for (i = 0; i < MAX_TEXTURES; ++i)
6329             {
6330                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6331             }
6332             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6333             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6334         }
6335         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6336         {
6337             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6338         }
6339         return;
6340     }
6341
6342     if (!use_pshader)
6343     {
6344         /* Find or create a shader implementing the fixed function pipeline
6345          * settings, then activate it. */
6346         gen_ffp_frag_op(device, state, &settings, FALSE);
6347         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6348         if(!desc) {
6349             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6350             if (!new_desc)
6351             {
6352                 ERR("Out of memory\n");
6353                 return;
6354             }
6355             new_desc->num_textures_used = 0;
6356             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6357             {
6358                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6359                     break;
6360                 new_desc->num_textures_used = i;
6361             }
6362
6363             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6364             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6365             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6366             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6367             desc = new_desc;
6368         }
6369
6370         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6371          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6372          * deactivate it.
6373          */
6374         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6375         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6376         priv->current_fprogram_id = desc->shader;
6377
6378         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6379         {
6380             /* Reload fixed function constants since they collide with the
6381              * pixel shader constants. */
6382             for (i = 0; i < MAX_TEXTURES; ++i)
6383             {
6384                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6385             }
6386             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6387             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6388         }
6389         context->last_was_pshader = FALSE;
6390     } else {
6391         context->last_was_pshader = TRUE;
6392     }
6393
6394     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6395      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6396      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6397      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6398      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6399      *
6400      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6401      * shader handler
6402      */
6403     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6404     {
6405         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6406
6407         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6408             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6409     }
6410     if (use_pshader)
6411         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6412 }
6413
6414 /* We can't link the fog states to the fragment state directly since the
6415  * vertex pipeline links them to FOGENABLE. A different linking in different
6416  * pipeline parts can't be expressed in the combined state table, so we need
6417  * to handle that with a forwarding function. The other invisible side effect
6418  * is that changing the fog start and fog end (which links to FOGENABLE in
6419  * vertex) results in the fragment_prog_arbfp function being called because
6420  * FOGENABLE is dirty, which calls this function here. */
6421 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6422 {
6423     enum fogsource new_source;
6424
6425     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6426
6427     if (!isStateDirty(context, STATE_PIXELSHADER))
6428         fragment_prog_arbfp(context, state, state_id);
6429
6430     if (!state->render_states[WINED3D_RS_FOGENABLE])
6431         return;
6432
6433     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6434     {
6435         if (use_vs(state))
6436         {
6437             new_source = FOGSOURCE_VS;
6438         }
6439         else
6440         {
6441             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6442                 new_source = FOGSOURCE_COORD;
6443             else
6444                 new_source = FOGSOURCE_FFP;
6445         }
6446     }
6447     else
6448     {
6449         new_source = FOGSOURCE_FFP;
6450     }
6451
6452     if (new_source != context->fog_source)
6453     {
6454         context->fog_source = new_source;
6455         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6456     }
6457 }
6458
6459 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6460 {
6461     if (!isStateDirty(context, STATE_PIXELSHADER))
6462         fragment_prog_arbfp(context, state, state_id);
6463 }
6464
6465 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6466 {
6467     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6468     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6469     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6470     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6471     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6472     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6473     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6474     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6475     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6476     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6477     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6478     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6479     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6480     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6481     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6482     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6483     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6484     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6485     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6486     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6487     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6488     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6489     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6490     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6491     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6492     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6493     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6494     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6495     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6496     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6497     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6498     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6499     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6502     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6503     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6505     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6589     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6590     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6593     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6596     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6605     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6606     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6607 };
6608
6609 const struct fragment_pipeline arbfp_fragment_pipeline = {
6610     arbfp_enable,
6611     arbfp_get_caps,
6612     arbfp_alloc,
6613     arbfp_free,
6614     shader_arb_color_fixup_supported,
6615     arbfp_fragmentstate_template,
6616     TRUE /* We can disable projected textures */
6617 };
6618
6619 struct arbfp_blit_priv {
6620     GLenum yuy2_rect_shader, yuy2_2d_shader;
6621     GLenum uyvy_rect_shader, uyvy_2d_shader;
6622     GLenum yv12_rect_shader, yv12_2d_shader;
6623     GLenum p8_rect_shader, p8_2d_shader;
6624     GLuint palette_texture;
6625 };
6626
6627 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6628 {
6629     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6630     if(!device->blit_priv) {
6631         ERR("Out of memory\n");
6632         return E_OUTOFMEMORY;
6633     }
6634     return WINED3D_OK;
6635 }
6636
6637 /* Context activation is done by the caller. */
6638 static void arbfp_blit_free(struct wined3d_device *device)
6639 {
6640     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6641     struct arbfp_blit_priv *priv = device->blit_priv;
6642
6643     ENTER_GL();
6644     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6645     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6646     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6647     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6648     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6649     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6650     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6651     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6652     checkGLcall("Delete yuv and p8 programs");
6653
6654     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6655     LEAVE_GL();
6656
6657     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6658     device->blit_priv = NULL;
6659 }
6660
6661 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6662         GLenum textype, char *luminance)
6663 {
6664     char chroma;
6665     const char *tex, *texinstr;
6666
6667     if (fixup == COMPLEX_FIXUP_UYVY) {
6668         chroma = 'x';
6669         *luminance = 'w';
6670     } else {
6671         chroma = 'w';
6672         *luminance = 'x';
6673     }
6674     switch(textype) {
6675         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6676         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6677         default:
6678             /* This is more tricky than just replacing the texture type - we have to navigate
6679              * properly in the texture to find the correct chroma values
6680              */
6681             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6682             return FALSE;
6683     }
6684
6685     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6686      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6687      * filtering when we sample the texture.
6688      *
6689      * These are the rules for reading the chroma:
6690      *
6691      * Even pixel: Cr
6692      * Even pixel: U
6693      * Odd pixel: V
6694      *
6695      * So we have to get the sampling x position in non-normalized coordinates in integers
6696      */
6697     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6698         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6699         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6700     } else {
6701         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6702     }
6703     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6704      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6705      * 0.5, so add 0.5.
6706      */
6707     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6708     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6709
6710     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6711      * even and odd pixels respectively
6712      */
6713     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6714     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6715
6716     /* Sample Pixel 1 */
6717     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6718
6719     /* Put the value into either of the chroma values */
6720     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6721     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6722     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6723     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6724
6725     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6726      * the pixel right to the current one. Otherwise, sample the left pixel.
6727      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6728      */
6729     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6730     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6731     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6732
6733     /* Put the value into the other chroma */
6734     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6735     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6736     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6737     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6738
6739     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6740      * the current one and lerp the two U and V values
6741      */
6742
6743     /* This gives the correctly filtered luminance value */
6744     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6745
6746     return TRUE;
6747 }
6748
6749 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6750 {
6751     const char *tex;
6752
6753     switch(textype) {
6754         case GL_TEXTURE_2D:             tex = "2D";     break;
6755         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6756         default:
6757             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6758             return FALSE;
6759     }
6760
6761     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6762      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6763      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6764      * pitch of the luminance plane, the packing into the gl texture is a bit
6765      * unfortunate. If the whole texture is interpreted as luminance data it looks
6766      * approximately like this:
6767      *
6768      *        +----------------------------------+----
6769      *        |                                  |
6770      *        |                                  |
6771      *        |                                  |
6772      *        |                                  |
6773      *        |                                  |   2
6774      *        |            LUMINANCE             |   -
6775      *        |                                  |   3
6776      *        |                                  |
6777      *        |                                  |
6778      *        |                                  |
6779      *        |                                  |
6780      *        +----------------+-----------------+----
6781      *        |                |                 |
6782      *        |  U even rows   |  U odd rows     |
6783      *        |                |                 |   1
6784      *        +----------------+------------------   -
6785      *        |                |                 |   3
6786      *        |  V even rows   |  V odd rows     |
6787      *        |                |                 |
6788      *        +----------------+-----------------+----
6789      *        |                |                 |
6790      *        |     0.5        |       0.5       |
6791      *
6792      * So it appears as if there are 4 chroma images, but in fact the odd rows
6793      * in the chroma images are in the same row as the even ones. So its is
6794      * kinda tricky to read
6795      *
6796      * When reading from rectangle textures, keep in mind that the input y coordinates
6797      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6798      */
6799     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6800             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6801
6802     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6803     /* the chroma planes have only half the width */
6804     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6805
6806     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6807      * the coordinate. Also read the right side of the image when reading odd lines
6808      *
6809      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6810      * bleeding
6811      */
6812     if(textype == GL_TEXTURE_2D) {
6813
6814         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6815
6816         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6817
6818         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6819         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6820
6821         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6822         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6823         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6824         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6825         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6826
6827         /* clamp, keep the half pixel origin in mind */
6828         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6829         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6830         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6831         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6832     } else {
6833         /* Read from [size - size+size/4] */
6834         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6835         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6836
6837         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6838         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6839         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6840         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6841         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6842         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6843
6844         /* Make sure to read exactly from the pixel center */
6845         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6846         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6847
6848         /* Clamp */
6849         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6850         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6851         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6852         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6853         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6854     }
6855     /* Read the texture, put the result into the output register */
6856     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6857     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6858
6859     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6860      * No need to clamp because we're just reusing the already clamped value from above
6861      */
6862     if(textype == GL_TEXTURE_2D) {
6863         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6864     } else {
6865         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6866     }
6867     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6868     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6869
6870     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6871      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6872      * values due to filtering
6873      */
6874     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6875     if(textype == GL_TEXTURE_2D) {
6876         /* Multiply the y coordinate by 2/3 and clamp it */
6877         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6878         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6879         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6880         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6881     } else {
6882         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6883          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6884          * is bigger
6885          */
6886         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6887         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6888         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6889     }
6890     *luminance = 'a';
6891
6892     return TRUE;
6893 }
6894
6895 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6896         const struct wined3d_gl_info *gl_info, GLenum textype)
6897 {
6898     GLenum shader;
6899     struct wined3d_shader_buffer buffer;
6900     GLint pos;
6901
6902     /* Shader header */
6903     if (!shader_buffer_init(&buffer))
6904     {
6905         ERR("Failed to initialize shader buffer.\n");
6906         return 0;
6907     }
6908
6909     ENTER_GL();
6910     GL_EXTCALL(glGenProgramsARB(1, &shader));
6911     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6912     LEAVE_GL();
6913     if(!shader) {
6914         shader_buffer_free(&buffer);
6915         return 0;
6916     }
6917
6918     shader_addline(&buffer, "!!ARBfp1.0\n");
6919     shader_addline(&buffer, "TEMP index;\n");
6920
6921     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6922     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6923
6924     /* The alpha-component contains the palette index */
6925     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6926         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6927     else
6928         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6929
6930     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6931     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6932
6933     /* Use the alpha-component as an index in the palette to get the final color */
6934     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6935     shader_addline(&buffer, "END\n");
6936
6937     ENTER_GL();
6938     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6939             strlen(buffer.buffer), buffer.buffer));
6940     checkGLcall("glProgramStringARB()");
6941
6942     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6943     if (pos != -1)
6944     {
6945         FIXME("Fragment program error at position %d: %s\n\n", pos,
6946               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6947         shader_arb_dump_program_source(buffer.buffer);
6948     }
6949
6950     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6951         priv->p8_rect_shader = shader;
6952     else
6953         priv->p8_2d_shader = shader;
6954
6955     shader_buffer_free(&buffer);
6956     LEAVE_GL();
6957
6958     return shader;
6959 }
6960
6961 /* Context activation is done by the caller. */
6962 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
6963 {
6964     BYTE table[256][4];
6965     struct wined3d_device *device = surface->resource.device;
6966     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6967     struct arbfp_blit_priv *priv = device->blit_priv;
6968     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6969
6970     d3dfmt_p8_init_palette(surface, table, colorkey);
6971
6972     ENTER_GL();
6973
6974     if (gl_info->supported[APPLE_CLIENT_STORAGE])
6975     {
6976         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
6977         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
6978     }
6979
6980     if (!priv->palette_texture)
6981         glGenTextures(1, &priv->palette_texture);
6982
6983     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6984     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6985
6986     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6987
6988     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6989     /* Make sure we have discrete color levels. */
6990     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6991     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6992     /* Upload the palette */
6993     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6994     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6995
6996     if (gl_info->supported[APPLE_CLIENT_STORAGE])
6997     {
6998         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
6999         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7000     }
7001
7002     /* Switch back to unit 0 in which the 2D texture will be stored. */
7003     context_active_texture(context, gl_info, 0);
7004     LEAVE_GL();
7005 }
7006
7007 /* Context activation is done by the caller. */
7008 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7009         enum complex_fixup yuv_fixup, GLenum textype)
7010 {
7011     GLenum shader;
7012     struct wined3d_shader_buffer buffer;
7013     char luminance_component;
7014     GLint pos;
7015
7016     /* Shader header */
7017     if (!shader_buffer_init(&buffer))
7018     {
7019         ERR("Failed to initialize shader buffer.\n");
7020         return 0;
7021     }
7022
7023     ENTER_GL();
7024     GL_EXTCALL(glGenProgramsARB(1, &shader));
7025     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7026     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7027     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7028     LEAVE_GL();
7029     if(!shader) {
7030         shader_buffer_free(&buffer);
7031         return 0;
7032     }
7033
7034     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7035      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7036      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7037      * each single pixel it contains, and one U and one V value shared between both
7038      * pixels.
7039      *
7040      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7041      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7042      * take the format into account when generating the read swizzles
7043      *
7044      * Reading the Y value is straightforward - just sample the texture. The hardware
7045      * takes care of filtering in the horizontal and vertical direction.
7046      *
7047      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7048      * because that would mix the U and V values of one pixel or two adjacent pixels.
7049      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7050      * regardless of the filtering setting. Vertical filtering works automatically
7051      * though - the U and V values of two rows are mixed nicely.
7052      *
7053      * Appart of avoiding filtering issues, the code has to know which value it just
7054      * read, and where it can find the other one. To determine this, it checks if
7055      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7056      *
7057      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7058      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7059      *
7060      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7061      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7062      * in an unfiltered situation. Finding the luminance on the other hand requires
7063      * finding out if it is an odd or even pixel. The real drawback of this approach
7064      * is filtering. This would have to be emulated completely in the shader, reading
7065      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7066      * vertically. Beyond that it would require adjustments to the texture handling
7067      * code to deal with the width scaling
7068      */
7069     shader_addline(&buffer, "!!ARBfp1.0\n");
7070     shader_addline(&buffer, "TEMP luminance;\n");
7071     shader_addline(&buffer, "TEMP temp;\n");
7072     shader_addline(&buffer, "TEMP chroma;\n");
7073     shader_addline(&buffer, "TEMP texcrd;\n");
7074     shader_addline(&buffer, "TEMP texcrd2;\n");
7075     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7076     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7077     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7078
7079     switch (yuv_fixup)
7080     {
7081         case COMPLEX_FIXUP_UYVY:
7082         case COMPLEX_FIXUP_YUY2:
7083             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7084             {
7085                 shader_buffer_free(&buffer);
7086                 return 0;
7087             }
7088             break;
7089
7090         case COMPLEX_FIXUP_YV12:
7091             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7092             {
7093                 shader_buffer_free(&buffer);
7094                 return 0;
7095             }
7096             break;
7097
7098         default:
7099             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7100             shader_buffer_free(&buffer);
7101             return 0;
7102     }
7103
7104     /* Calculate the final result. Formula is taken from
7105      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7106      * ranges from -0.5 to 0.5
7107      */
7108     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7109
7110     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7111     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7112     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7113     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7114     shader_addline(&buffer, "END\n");
7115
7116     ENTER_GL();
7117     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7118             strlen(buffer.buffer), buffer.buffer));
7119     checkGLcall("glProgramStringARB()");
7120
7121     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7122     if (pos != -1)
7123     {
7124         FIXME("Fragment program error at position %d: %s\n\n", pos,
7125               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7126         shader_arb_dump_program_source(buffer.buffer);
7127     }
7128     else
7129     {
7130         GLint native;
7131
7132         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7133         checkGLcall("glGetProgramivARB()");
7134         if (!native) WARN("Program exceeds native resource limits.\n");
7135     }
7136
7137     shader_buffer_free(&buffer);
7138     LEAVE_GL();
7139
7140     switch (yuv_fixup)
7141     {
7142         case COMPLEX_FIXUP_YUY2:
7143             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7144             else priv->yuy2_2d_shader = shader;
7145             break;
7146
7147         case COMPLEX_FIXUP_UYVY:
7148             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7149             else priv->uyvy_2d_shader = shader;
7150             break;
7151
7152         case COMPLEX_FIXUP_YV12:
7153             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7154             else priv->yv12_2d_shader = shader;
7155             break;
7156         default:
7157             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7158     }
7159
7160     return shader;
7161 }
7162
7163 /* Context activation is done by the caller. */
7164 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7165 {
7166     GLenum shader;
7167     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7168     struct arbfp_blit_priv *priv = blit_priv;
7169     enum complex_fixup fixup;
7170     GLenum textype = surface->texture_target;
7171     const struct wined3d_gl_info *gl_info = context->gl_info;
7172
7173     if (surface->flags & SFLAG_CONVERTED)
7174     {
7175         ENTER_GL();
7176         glEnable(textype);
7177         checkGLcall("glEnable(textype)");
7178         LEAVE_GL();
7179         return WINED3D_OK;
7180     }
7181
7182     if (!is_complex_fixup(surface->resource.format->color_fixup))
7183     {
7184         TRACE("Fixup:\n");
7185         dump_color_fixup_desc(surface->resource.format->color_fixup);
7186         /* Don't bother setting up a shader for unconverted formats */
7187         ENTER_GL();
7188         glEnable(textype);
7189         checkGLcall("glEnable(textype)");
7190         LEAVE_GL();
7191         return WINED3D_OK;
7192     }
7193
7194     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7195
7196     switch(fixup)
7197     {
7198         case COMPLEX_FIXUP_YUY2:
7199             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7200             break;
7201
7202         case COMPLEX_FIXUP_UYVY:
7203             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7204             break;
7205
7206         case COMPLEX_FIXUP_YV12:
7207             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7208             break;
7209
7210         case COMPLEX_FIXUP_P8:
7211             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7212             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7213
7214             upload_palette(surface, context);
7215             break;
7216
7217         default:
7218             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7219             ENTER_GL();
7220             glEnable(textype);
7221             checkGLcall("glEnable(textype)");
7222             LEAVE_GL();
7223             return E_NOTIMPL;
7224     }
7225
7226     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7227
7228     ENTER_GL();
7229     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7230     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7231     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7232     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7233     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7234     checkGLcall("glProgramLocalParameter4fvARB");
7235     LEAVE_GL();
7236
7237     return WINED3D_OK;
7238 }
7239
7240 /* Context activation is done by the caller. */
7241 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7242 {
7243     ENTER_GL();
7244     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7245     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7246     glDisable(GL_TEXTURE_2D);
7247     checkGLcall("glDisable(GL_TEXTURE_2D)");
7248     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7249     {
7250         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7251         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7252     }
7253     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7254     {
7255         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7256         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7257     }
7258     LEAVE_GL();
7259 }
7260
7261 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7262         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7263         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7264 {
7265     enum complex_fixup src_fixup;
7266
7267     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7268         return FALSE;
7269
7270     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7271     {
7272         TRACE("Unsupported blit_op=%d\n", blit_op);
7273         return FALSE;
7274     }
7275
7276     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7277         return FALSE;
7278
7279     src_fixup = get_complex_fixup(src_format->color_fixup);
7280     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7281     {
7282         TRACE("Checking support for fixup:\n");
7283         dump_color_fixup_desc(src_format->color_fixup);
7284     }
7285
7286     if (!is_identity_fixup(dst_format->color_fixup))
7287     {
7288         TRACE("Destination fixups are not supported\n");
7289         return FALSE;
7290     }
7291
7292     if (is_identity_fixup(src_format->color_fixup))
7293     {
7294         TRACE("[OK]\n");
7295         return TRUE;
7296     }
7297
7298      /* We only support YUV conversions. */
7299     if (!is_complex_fixup(src_format->color_fixup))
7300     {
7301         TRACE("[FAILED]\n");
7302         return FALSE;
7303     }
7304
7305     switch(src_fixup)
7306     {
7307         case COMPLEX_FIXUP_YUY2:
7308         case COMPLEX_FIXUP_UYVY:
7309         case COMPLEX_FIXUP_YV12:
7310         case COMPLEX_FIXUP_P8:
7311             TRACE("[OK]\n");
7312             return TRUE;
7313
7314         default:
7315             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7316             TRACE("[FAILED]\n");
7317             return FALSE;
7318     }
7319 }
7320
7321 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7322         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7323         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7324 {
7325     struct wined3d_context *context;
7326     RECT src_rect = *src_rect_in;
7327     RECT dst_rect = *dst_rect_in;
7328
7329     /* Now load the surface */
7330     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7331             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7332     {
7333         /* Without FBO blits transferring from the drawable to the texture is
7334          * expensive, because we have to flip the data in sysmem. Since we can
7335          * flip in the blitter, we don't actually need that flip anyway. So we
7336          * use the surface's texture as scratch texture, and flip the source
7337          * rectangle instead. */
7338         surface_load_fb_texture(src_surface, FALSE);
7339
7340         src_rect.top = src_surface->resource.height - src_rect.top;
7341         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7342     }
7343     else
7344         surface_internal_preload(src_surface, SRGB_RGB);
7345
7346     /* Activate the destination context, set it up for blitting */
7347     context = context_acquire(device, dst_surface);
7348     context_apply_blit_state(context, device);
7349
7350     if (!surface_is_offscreen(dst_surface))
7351         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7352
7353     arbfp_blit_set(device->blit_priv, context, src_surface);
7354
7355     ENTER_GL();
7356
7357     /* Draw a textured quad */
7358     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7359
7360     LEAVE_GL();
7361
7362     /* Leave the opengl state valid for blitting */
7363     arbfp_blit_unset(context->gl_info);
7364
7365     if (wined3d_settings.strict_draw_ordering
7366             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7367             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7368         wglFlush(); /* Flush to ensure ordering across contexts. */
7369
7370     context_release(context);
7371
7372     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7373     return WINED3D_OK;
7374 }
7375
7376 /* Do not call while under the GL lock. */
7377 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7378         const RECT *dst_rect, const struct wined3d_color *color)
7379 {
7380     FIXME("Color filling not implemented by arbfp_blit\n");
7381     return WINED3DERR_INVALIDCALL;
7382 }
7383
7384 /* Do not call while under the GL lock. */
7385 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7386         struct wined3d_surface *surface, const RECT *rect, float depth)
7387 {
7388     FIXME("Depth filling not implemented by arbfp_blit.\n");
7389     return WINED3DERR_INVALIDCALL;
7390 }
7391
7392 const struct blit_shader arbfp_blit = {
7393     arbfp_blit_alloc,
7394     arbfp_blit_free,
7395     arbfp_blit_set,
7396     arbfp_blit_unset,
7397     arbfp_blit_supported,
7398     arbfp_blit_color_fill,
7399     arbfp_blit_depth_fill,
7400 };