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][WINED3DTSS_BUMPENVMAT00];
 546         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 547                 gl_shader->bumpenvmatconst[i].const_num, data));
 548
 549         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 550         {
 551             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
 552              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 553              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 554              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
 555             */
 556             const float *scale = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVLSCALE];
 557             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 558                     gl_shader->luminanceconst[i].const_num, scale));
 559         }
 560     }
 561     checkGLcall("Load bumpmap consts");
 562
 563     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 564     {
 565         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 566         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 567         * ycorrection.z: 1.0
 568         * ycorrection.w: 0.0
 569         */
 570         float val[4];
 571         val[0] = context->render_offscreen ? 0.0f : (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 WINED3DTSS_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         DWORD src_mod = ins->src[0].modifiers;
1955         if (src_mod == WINED3DSPSM_DZ) {
1956             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1957              * varying register, so we need a temp reg
1958              */
1959             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1960             strcpy(reg_coord, "TA");
1961             myflags |= TEX_PROJ;
1962         } else if(src_mod == WINED3DSPSM_DW) {
1963             myflags |= TEX_PROJ;
1964         }
1965     } else {
1966         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1967         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1968     }
1969     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1970 }
1971
1972 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1973 {
1974     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1975     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1976     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1977             ins->ctx->reg_maps->shader_version.minor);
1978     char dst_str[50];
1979
1980     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1981     {
1982         DWORD reg = dst->reg.idx;
1983
1984         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1985         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1986     } else {
1987         char reg_src[40];
1988
1989         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1990         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1991         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1992    }
1993 }
1994
1995 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1996 {
1997      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1998      DWORD flags = 0;
1999
2000      DWORD reg1 = ins->dst[0].reg.idx;
2001      char dst_str[50];
2002      char src_str[50];
2003
2004      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2005      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2006      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2007      /* Move .x first in case src_str is "TA" */
2008      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2009      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2010      if (reg1 < MAX_TEXTURES)
2011      {
2012          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2013          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2014      }
2015      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2016 }
2017
2018 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2019 {
2020      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2021
2022      DWORD reg1 = ins->dst[0].reg.idx;
2023      char dst_str[50];
2024      char src_str[50];
2025
2026      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2027      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2028      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2029      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2030      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2031      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2032 }
2033
2034 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2035 {
2036     DWORD reg1 = ins->dst[0].reg.idx;
2037     char dst_str[50];
2038     char src_str[50];
2039
2040     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2041     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2042     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2043     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2044 }
2045
2046 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2047 {
2048     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2049     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2050     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2051     char reg_coord[40], dst_reg[50], src_reg[50];
2052     DWORD reg_dest_code;
2053
2054     /* All versions have a destination register. The Tx where the texture coordinates come
2055      * from is the varying incarnation of the texture register
2056      */
2057     reg_dest_code = dst->reg.idx;
2058     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2059     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2060     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2061
2062     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2063      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2064      *
2065      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2066      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2067      *
2068      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2069      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2070      * extension.
2071      */
2072     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2073     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2074     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2075     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2076
2077     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2078      * so we can't let the GL handle this.
2079      */
2080     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2081             & WINED3D_PSARGS_PROJECTED)
2082     {
2083         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2084         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2085         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2086     } else {
2087         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2088     }
2089
2090     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2091
2092     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2093     {
2094         /* No src swizzles are allowed, so this is ok */
2095         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2096                        src_reg, reg_dest_code, reg_dest_code);
2097         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2098     }
2099 }
2100
2101 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2102 {
2103     DWORD reg = ins->dst[0].reg.idx;
2104     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2105     char src0_name[50], dst_name[50];
2106     BOOL is_color;
2107     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2108
2109     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2110     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2111      * T<reg+1> register. Use this register to store the calculated vector
2112      */
2113     tmp_reg.idx = reg + 1;
2114     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2115     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2116 }
2117
2118 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2119 {
2120     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2121     DWORD flags;
2122     DWORD reg = ins->dst[0].reg.idx;
2123     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2124     char dst_str[50];
2125     char src0_name[50];
2126     char dst_reg[50];
2127     BOOL is_color;
2128
2129     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2130     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2131
2132     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2133     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2134     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2135     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2136     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2137 }
2138
2139 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2140 {
2141     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2142     DWORD reg = ins->dst[0].reg.idx;
2143     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2144     char src0_name[50], dst_name[50];
2145     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2146     BOOL is_color;
2147
2148     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2149      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2150      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2151      */
2152     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2153     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2154
2155     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2156     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2157                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2158     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2159 }
2160
2161 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2162 {
2163     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2164     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2165     DWORD flags;
2166     DWORD reg = ins->dst[0].reg.idx;
2167     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2168     char dst_str[50];
2169     char src0_name[50], dst_name[50];
2170     BOOL is_color;
2171
2172     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2173     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2174     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2175
2176     /* Sample the texture using the calculated coordinates */
2177     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2178     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2179     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2180     tex_mx->current_row = 0;
2181 }
2182
2183 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2184 {
2185     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2186     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2187     DWORD flags;
2188     DWORD reg = ins->dst[0].reg.idx;
2189     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2190     char dst_str[50];
2191     char src0_name[50];
2192     char dst_reg[50];
2193     BOOL is_color;
2194
2195     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2196      * components for temporary data storage
2197      */
2198     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2199     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2200     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2201
2202     /* Construct the eye-ray vector from w coordinates */
2203     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2204     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2205     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2206
2207     /* Calculate reflection vector
2208      */
2209     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2210     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2211     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2212     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2213     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2214     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2215     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2216
2217     /* Sample the texture using the calculated coordinates */
2218     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2219     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2220     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2221     tex_mx->current_row = 0;
2222 }
2223
2224 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2225 {
2226     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2227     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2228     DWORD flags;
2229     DWORD reg = ins->dst[0].reg.idx;
2230     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2231     char dst_str[50];
2232     char src0_name[50];
2233     char src1_name[50];
2234     char dst_reg[50];
2235     BOOL is_color;
2236
2237     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2238     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2239     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2240     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2241     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2242
2243     /* Calculate reflection vector.
2244      *
2245      *                   dot(N, E)
2246      * dst_reg.xyz = 2 * --------- * N - E
2247      *                   dot(N, N)
2248      *
2249      * Which normalizes the normal vector
2250      */
2251     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2252     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2253     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2254     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2255     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2256     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2257
2258     /* Sample the texture using the calculated coordinates */
2259     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2260     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2261     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2262     tex_mx->current_row = 0;
2263 }
2264
2265 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2266 {
2267     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2268     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2269     char dst_name[50];
2270     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2271     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2272
2273     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2274      * which is essentially an input, is the destination register because it is the first
2275      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2276      * here(writemasks/swizzles are not valid on texdepth)
2277      */
2278     shader_arb_get_dst_param(ins, dst, dst_name);
2279
2280     /* According to the msdn, the source register(must be r5) is unusable after
2281      * the texdepth instruction, so we're free to modify it
2282      */
2283     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2284
2285     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2286      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2287      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2288      */
2289     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2290     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2291     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2292     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2293 }
2294
2295 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2296  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2297  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2298 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2299 {
2300     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2301     DWORD sampler_idx = ins->dst[0].reg.idx;
2302     char src0[50];
2303     char dst_str[50];
2304
2305     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2306     shader_addline(buffer, "MOV TB, 0.0;\n");
2307     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2308
2309     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2310     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2311 }
2312
2313 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2314  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2315 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2316 {
2317     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2318     char src0[50];
2319     char dst_str[50];
2320     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2321
2322     /* Handle output register */
2323     shader_arb_get_dst_param(ins, dst, dst_str);
2324     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2325     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2326 }
2327
2328 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2329  * Perform the 3rd row of a 3x3 matrix multiply */
2330 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2331 {
2332     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2333     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2334     char dst_str[50], dst_name[50];
2335     char src0[50];
2336     BOOL is_color;
2337
2338     shader_arb_get_dst_param(ins, dst, dst_str);
2339     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2340     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2341     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2342     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2343 }
2344
2345 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2346  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2347  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2348  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2349  */
2350 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2351 {
2352     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2353     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2354     char src0[50], dst_name[50];
2355     BOOL is_color;
2356     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2357     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2358
2359     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2360     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2361     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2362
2363     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2364      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2365      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2366      */
2367     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2368     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2369     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2370     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2371 }
2372
2373 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2374     Vertex/Pixel shaders to ARB_vertex_program codes */
2375 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2376 {
2377     int i;
2378     int nComponents = 0;
2379     struct wined3d_shader_dst_param tmp_dst = {{0}};
2380     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2381     struct wined3d_shader_instruction tmp_ins;
2382
2383     memset(&tmp_ins, 0, sizeof(tmp_ins));
2384
2385     /* Set constants for the temporary argument */
2386     tmp_ins.ctx = ins->ctx;
2387     tmp_ins.dst_count = 1;
2388     tmp_ins.dst = &tmp_dst;
2389     tmp_ins.src_count = 2;
2390     tmp_ins.src = tmp_src;
2391
2392     switch(ins->handler_idx)
2393     {
2394         case WINED3DSIH_M4x4:
2395             nComponents = 4;
2396             tmp_ins.handler_idx = WINED3DSIH_DP4;
2397             break;
2398         case WINED3DSIH_M4x3:
2399             nComponents = 3;
2400             tmp_ins.handler_idx = WINED3DSIH_DP4;
2401             break;
2402         case WINED3DSIH_M3x4:
2403             nComponents = 4;
2404             tmp_ins.handler_idx = WINED3DSIH_DP3;
2405             break;
2406         case WINED3DSIH_M3x3:
2407             nComponents = 3;
2408             tmp_ins.handler_idx = WINED3DSIH_DP3;
2409             break;
2410         case WINED3DSIH_M3x2:
2411             nComponents = 2;
2412             tmp_ins.handler_idx = WINED3DSIH_DP3;
2413             break;
2414         default:
2415             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2416             break;
2417     }
2418
2419     tmp_dst = ins->dst[0];
2420     tmp_src[0] = ins->src[0];
2421     tmp_src[1] = ins->src[1];
2422     for (i = 0; i < nComponents; i++) {
2423         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2424         shader_hw_map2gl(&tmp_ins);
2425         ++tmp_src[1].reg.idx;
2426     }
2427 }
2428
2429 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2430 {
2431     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2432
2433     char dst[50];
2434     char src[50];
2435
2436     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2437     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2438     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2439     {
2440         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2441          * .w is used
2442          */
2443         strcat(src, ".w");
2444     }
2445
2446     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2447 }
2448
2449 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2450 {
2451     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2452     const char *instruction;
2453
2454     char dst[50];
2455     char src[50];
2456
2457     switch(ins->handler_idx)
2458     {
2459         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2460         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2461         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2462         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2463         default: instruction = "";
2464             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2465             break;
2466     }
2467
2468     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2469     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2470     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2471     {
2472         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2473          * .w is used
2474          */
2475         strcat(src, ".w");
2476     }
2477
2478     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2479 }
2480
2481 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2482 {
2483     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2484     char dst_name[50];
2485     char src_name[50];
2486     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2487     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2488     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2489
2490     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2491     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2492
2493     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2494      * otherwise NRM or RSQ would return NaN */
2495     if(pshader && priv->target_version >= NV3)
2496     {
2497         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2498          *
2499          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2500          */
2501         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2502         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2503         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2504     }
2505     else if(priv->target_version >= NV2)
2506     {
2507         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2508         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2509         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2510                        src_name);
2511     }
2512     else
2513     {
2514         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2515
2516         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2517         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2518          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2519          */
2520         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2521         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2522
2523         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2524         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2525         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2526                     src_name);
2527     }
2528 }
2529
2530 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2531 {
2532     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2533     char dst_name[50];
2534     char src_name[3][50];
2535
2536     /* ARB_fragment_program has a convenient LRP instruction */
2537     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2538         shader_hw_map2gl(ins);
2539         return;
2540     }
2541
2542     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2543     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2544     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2545     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2546
2547     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2548     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2549                    dst_name, src_name[0], src_name[2]);
2550 }
2551
2552 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2553 {
2554     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2555      * must contain fixed constants. So we need a separate function to filter those constants and
2556      * can't use map2gl
2557      */
2558     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2559     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2560     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2561     char dst_name[50];
2562     char src_name0[50], src_name1[50], src_name2[50];
2563     BOOL is_color;
2564
2565     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2566     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2567         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2568         /* No modifiers are supported on SCS */
2569         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2570
2571         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2572         {
2573             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2574             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2575         }
2576     } else if(priv->target_version >= NV2) {
2577         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2578
2579         /* Sincos writemask must be .x, .y or .xy */
2580         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2581             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2582         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2583             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2584     } else {
2585         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2586          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2587          *
2588          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2589          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2590          *
2591          * The constants we get are:
2592          *
2593          *  +1   +1,     -1     -1     +1      +1      -1       -1
2594          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2595          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2596          *
2597          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2598          *
2599          * (x/2)^2 = x^2 / 4
2600          * (x/2)^3 = x^3 / 8
2601          * (x/2)^4 = x^4 / 16
2602          * (x/2)^5 = x^5 / 32
2603          * etc
2604          *
2605          * To get the final result:
2606          * sin(x) = 2 * sin(x/2) * cos(x/2)
2607          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2608          * (from sin(x+y) and cos(x+y) rules)
2609          *
2610          * As per MSDN, dst.z is undefined after the operation, and so is
2611          * dst.x and dst.y if they're masked out by the writemask. Ie
2612          * sincos dst.y, src1, c0, c1
2613          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2614          * vsa.exe also stops with an error if the dest register is the same register as the source
2615          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2616          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2617          */
2618         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2619         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2620         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2621
2622         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2623         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2624         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2625         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2626         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2627         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2628
2629         /* sin(x/2)
2630          *
2631          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2632          * properly merge that with MULs in the code above?
2633          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2634          * we can merge the sine and cosine MAD rows to calculate them together.
2635          */
2636         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2637         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2638         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2639         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2640
2641         /* cos(x/2) */
2642         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2643         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2644         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2645
2646         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2647             /* cos x */
2648             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2649             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2650         }
2651         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2652             /* sin x */
2653             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2654             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2655         }
2656     }
2657 }
2658
2659 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2660 {
2661     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2662     char dst_name[50];
2663     char src_name[50];
2664     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2665
2666     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2667     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2668
2669     /* SGN is only valid in vertex shaders */
2670     if(ctx->target_version >= NV2) {
2671         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2672         return;
2673     }
2674
2675     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2676      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2677      */
2678     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2679         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2680     } else {
2681         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2682          * Then use TA, and calculate the final result
2683          *
2684          * Not reading from TA? Store the first result in TA to avoid overwriting the
2685          * destination if src reg = dst reg
2686          */
2687         if(strstr(src_name, "TA"))
2688         {
2689             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2690             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2691             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2692         }
2693         else
2694         {
2695             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2696             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2697             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2698         }
2699     }
2700 }
2701
2702 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2703 {
2704     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2705     char src[50];
2706     char dst[50];
2707     char dst_name[50];
2708     BOOL is_color;
2709
2710     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2711     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2712     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2713
2714     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2715     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2716 }
2717
2718 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2719 {
2720     *need_abs = FALSE;
2721
2722     switch(mod)
2723     {
2724         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2725         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2726         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2727         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2728         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2729         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2730         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2731         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2732         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2733         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2734         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2735         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2736         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2737     }
2738     FIXME("Unknown modifier %u\n", mod);
2739     return mod;
2740 }
2741
2742 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2743 {
2744     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2745     char src0[50], dst[50];
2746     struct wined3d_shader_src_param src0_copy = ins->src[0];
2747     BOOL need_abs = FALSE;
2748     const char *instr;
2749
2750     switch(ins->handler_idx)
2751     {
2752         case WINED3DSIH_LOG:  instr = "LG2"; break;
2753         case WINED3DSIH_LOGP: instr = "LOG"; break;
2754         default:
2755             ERR("Unexpected instruction %d\n", ins->handler_idx);
2756             return;
2757     }
2758
2759     /* LOG and LOGP operate on the absolute value of the input */
2760     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2761
2762     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2763     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2764
2765     if(need_abs)
2766     {
2767         shader_addline(buffer, "ABS TA, %s;\n", src0);
2768         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2769     }
2770     else
2771     {
2772         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2773     }
2774 }
2775
2776 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2777 {
2778     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2779     char src0[50], src1[50], dst[50];
2780     struct wined3d_shader_src_param src0_copy = ins->src[0];
2781     BOOL need_abs = FALSE;
2782     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2783     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2784
2785     /* POW operates on the absolute value of the input */
2786     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2787
2788     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2789     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2790     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2791
2792     if (need_abs)
2793         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2794     else
2795         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2796
2797     if (priv->target_version >= NV2)
2798     {
2799         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2800         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2801         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2802     }
2803     else
2804     {
2805         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2806         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2807
2808         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2809         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2810         /* Possibly add flt_eps to avoid getting float special values */
2811         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2812         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2813         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2814         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2815     }
2816 }
2817
2818 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2819 {
2820     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2821     char src_name[50];
2822     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2823
2824     /* src0 is aL */
2825     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2826
2827     if(vshader)
2828     {
2829         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2830         struct list *e = list_head(&priv->control_frames);
2831         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2832
2833         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2834         /* The constant loader makes sure to load -1 into iX.w */
2835         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2836         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2837         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2838     }
2839     else
2840     {
2841         shader_addline(buffer, "LOOP %s;\n", src_name);
2842     }
2843 }
2844
2845 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2846 {
2847     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2848     char src_name[50];
2849     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2850
2851     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2852
2853     /* The constant loader makes sure to load -1 into iX.w */
2854     if(vshader)
2855     {
2856         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2857         struct list *e = list_head(&priv->control_frames);
2858         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2859
2860         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2861
2862         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2863         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2864         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2865     }
2866     else
2867     {
2868         shader_addline(buffer, "REP %s;\n", src_name);
2869     }
2870 }
2871
2872 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2873 {
2874     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2875     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2876
2877     if(vshader)
2878     {
2879         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2880         struct list *e = list_head(&priv->control_frames);
2881         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2882
2883         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2884         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2885         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2886
2887         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2888     }
2889     else
2890     {
2891         shader_addline(buffer, "ENDLOOP;\n");
2892     }
2893 }
2894
2895 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2896 {
2897     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2898     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2899
2900     if(vshader)
2901     {
2902         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2903         struct list *e = list_head(&priv->control_frames);
2904         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2905
2906         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2907         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2908         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2909
2910         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2911     }
2912     else
2913     {
2914         shader_addline(buffer, "ENDREP;\n");
2915     }
2916 }
2917
2918 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2919 {
2920     struct control_frame *control_frame;
2921
2922     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2923     {
2924         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2925     }
2926     ERR("Could not find loop for break\n");
2927     return NULL;
2928 }
2929
2930 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2931 {
2932     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2933     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2934     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2935
2936     if(vshader)
2937     {
2938         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2939     }
2940     else
2941     {
2942         shader_addline(buffer, "BRK;\n");
2943     }
2944 }
2945
2946 static const char *get_compare(enum wined3d_shader_rel_op op)
2947 {
2948     switch (op)
2949     {
2950         case WINED3D_SHADER_REL_OP_GT: return "GT";
2951         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2952         case WINED3D_SHADER_REL_OP_GE: return "GE";
2953         case WINED3D_SHADER_REL_OP_LT: return "LT";
2954         case WINED3D_SHADER_REL_OP_NE: return "NE";
2955         case WINED3D_SHADER_REL_OP_LE: return "LE";
2956         default:
2957             FIXME("Unrecognized operator %#x.\n", op);
2958             return "(\?\?)";
2959     }
2960 }
2961
2962 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2963 {
2964     switch (op)
2965     {
2966         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2967         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2968         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2969         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2970         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2971         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2972         default:
2973             FIXME("Unrecognized operator %#x.\n", op);
2974             return -1;
2975     }
2976 }
2977
2978 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2979 {
2980     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2981     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2982     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2983     char src_name0[50];
2984     char src_name1[50];
2985     const char *comp = get_compare(ins->flags);
2986
2987     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2988     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2989
2990     if(vshader)
2991     {
2992         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2993          * away the subtraction result
2994          */
2995         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2996         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2997     }
2998     else
2999     {
3000         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3001         shader_addline(buffer, "BRK (%s.x);\n", comp);
3002     }
3003 }
3004
3005 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3006 {
3007     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3008     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3009     struct list *e = list_head(&priv->control_frames);
3010     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3011     const char *comp;
3012     char src_name0[50];
3013     char src_name1[50];
3014     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3015
3016     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3017     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3018
3019     if(vshader)
3020     {
3021         /* Invert the flag. We jump to the else label if the condition is NOT true */
3022         comp = get_compare(invert_compare(ins->flags));
3023         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3024         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3025     }
3026     else
3027     {
3028         comp = get_compare(ins->flags);
3029         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3030         shader_addline(buffer, "IF %s.x;\n", comp);
3031     }
3032 }
3033
3034 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3035 {
3036     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3037     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3038     struct list *e = list_head(&priv->control_frames);
3039     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3040     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3041
3042     if(vshader)
3043     {
3044         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3045         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3046         control_frame->had_else = TRUE;
3047     }
3048     else
3049     {
3050         shader_addline(buffer, "ELSE;\n");
3051     }
3052 }
3053
3054 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3055 {
3056     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3057     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3058     struct list *e = list_head(&priv->control_frames);
3059     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3060     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3061
3062     if(vshader)
3063     {
3064         if(control_frame->had_else)
3065         {
3066             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3067         }
3068         else
3069         {
3070             shader_addline(buffer, "#No else branch. else is endif\n");
3071             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3072         }
3073     }
3074     else
3075     {
3076         shader_addline(buffer, "ENDIF;\n");
3077     }
3078 }
3079
3080 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3081 {
3082     DWORD sampler_idx = ins->src[1].reg.idx;
3083     char reg_dest[40];
3084     char reg_src[3][40];
3085     WORD flags = TEX_DERIV;
3086
3087     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3088     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3089     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3090     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3091
3092     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3093     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3094
3095     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3096 }
3097
3098 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3099 {
3100     DWORD sampler_idx = ins->src[1].reg.idx;
3101     char reg_dest[40];
3102     char reg_coord[40];
3103     WORD flags = TEX_LOD;
3104
3105     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3106     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3107
3108     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3109     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3110
3111     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3112 }
3113
3114 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3115 {
3116     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3117     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3118
3119     priv->in_main_func = FALSE;
3120     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3121      * subroutine, don't generate a label that will make GL complain
3122      */
3123     if(priv->target_version == ARB) return;
3124
3125     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3126 }
3127
3128 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3129         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3130         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3131         struct wined3d_shader_buffer *buffer)
3132 {
3133     unsigned int i;
3134
3135     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3136      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3137      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3138      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3139      */
3140     if (args->super.fog_src == VS_FOG_Z)
3141         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3142     else if (!reg_maps->fog)
3143         /* posFixup.x is always 1.0, so we can safely use it */
3144         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3145
3146     /* Clipplanes are always stored without y inversion */
3147     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3148     {
3149         if (args->super.clip_enabled)
3150         {
3151             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3152             {
3153                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3154             }
3155         }
3156     }
3157     else if (args->clip.boolclip.clip_texcoord)
3158     {
3159         unsigned int cur_clip = 0;
3160         char component[4] = {'x', 'y', 'z', 'w'};
3161         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3162
3163         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3164         {
3165             if (args->clip.boolclip.clipplane_mask & (1 << i))
3166             {
3167                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3168                                component[cur_clip++], i);
3169             }
3170         }
3171         switch (cur_clip)
3172         {
3173             case 0:
3174                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3175                 break;
3176             case 1:
3177                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3178                 break;
3179             case 2:
3180                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3181                 break;
3182             case 3:
3183                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3184                 break;
3185         }
3186         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3187                        args->clip.boolclip.clip_texcoord - 1);
3188     }
3189
3190     /* Write the final position.
3191      *
3192      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3193      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3194      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3195      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3196      */
3197     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3198     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3199     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3200
3201     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3202      * and the glsl equivalent
3203      */
3204     if (need_helper_const(shader_data, reg_maps, gl_info))
3205     {
3206         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3207         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3208     }
3209     else
3210     {
3211         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3212         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3213     }
3214
3215     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3216
3217     priv_ctx->footer_written = TRUE;
3218 }
3219
3220 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3221 {
3222     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3223     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3224     const struct wined3d_shader *shader = ins->ctx->shader;
3225     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3226
3227     if(priv->target_version == ARB) return;
3228
3229     if(vshader)
3230     {
3231         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3232                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3233     }
3234
3235     shader_addline(buffer, "RET;\n");
3236 }
3237
3238 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3239 {
3240     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3241     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3242 }
3243
3244 /* GL locking is done by the caller */
3245 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3246 {
3247     GLuint program_id = 0;
3248     GLint pos;
3249
3250     const char *blt_vprogram =
3251         "!!ARBvp1.0\n"
3252         "PARAM c[1] = { { 1, 0.5 } };\n"
3253         "MOV result.position, vertex.position;\n"
3254         "MOV result.color, c[0].x;\n"
3255         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3256         "END\n";
3257
3258     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3259     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3260     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3261             strlen(blt_vprogram), blt_vprogram));
3262     checkGLcall("glProgramStringARB()");
3263
3264     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3265     if (pos != -1)
3266     {
3267         FIXME("Vertex program error at position %d: %s\n\n", pos,
3268             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3269         shader_arb_dump_program_source(blt_vprogram);
3270     }
3271     else
3272     {
3273         GLint native;
3274
3275         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3276         checkGLcall("glGetProgramivARB()");
3277         if (!native) WARN("Program exceeds native resource limits.\n");
3278     }
3279
3280     return program_id;
3281 }
3282
3283 /* GL locking is done by the caller */
3284 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3285         enum tex_types tex_type, BOOL masked)
3286 {
3287     GLuint program_id = 0;
3288     const char *fprogram;
3289     GLint pos;
3290
3291     static const char * const blt_fprograms_full[tex_type_count] =
3292     {
3293         /* tex_1d */
3294         NULL,
3295         /* tex_2d */
3296         "!!ARBfp1.0\n"
3297         "TEMP R0;\n"
3298         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3299         "MOV result.depth.z, R0.x;\n"
3300         "END\n",
3301         /* tex_3d */
3302         NULL,
3303         /* tex_cube */
3304         "!!ARBfp1.0\n"
3305         "TEMP R0;\n"
3306         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3307         "MOV result.depth.z, R0.x;\n"
3308         "END\n",
3309         /* tex_rect */
3310         "!!ARBfp1.0\n"
3311         "TEMP R0;\n"
3312         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3313         "MOV result.depth.z, R0.x;\n"
3314         "END\n",
3315     };
3316
3317     static const char * const blt_fprograms_masked[tex_type_count] =
3318     {
3319         /* tex_1d */
3320         NULL,
3321         /* tex_2d */
3322         "!!ARBfp1.0\n"
3323         "PARAM mask = program.local[0];\n"
3324         "TEMP R0;\n"
3325         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3326         "MUL R0.x, R0.x, R0.y;\n"
3327         "KIL -R0.x;\n"
3328         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3329         "MOV result.depth.z, R0.x;\n"
3330         "END\n",
3331         /* tex_3d */
3332         NULL,
3333         /* tex_cube */
3334         "!!ARBfp1.0\n"
3335         "PARAM mask = program.local[0];\n"
3336         "TEMP R0;\n"
3337         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3338         "MUL R0.x, R0.x, R0.y;\n"
3339         "KIL -R0.x;\n"
3340         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3341         "MOV result.depth.z, R0.x;\n"
3342         "END\n",
3343         /* tex_rect */
3344         "!!ARBfp1.0\n"
3345         "PARAM mask = program.local[0];\n"
3346         "TEMP R0;\n"
3347         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3348         "MUL R0.x, R0.x, R0.y;\n"
3349         "KIL -R0.x;\n"
3350         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3351         "MOV result.depth.z, R0.x;\n"
3352         "END\n",
3353     };
3354
3355     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3356     if (!fprogram)
3357     {
3358         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3359         tex_type = tex_2d;
3360         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3361     }
3362
3363     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3364     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3365     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3366     checkGLcall("glProgramStringARB()");
3367
3368     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3369     if (pos != -1)
3370     {
3371         FIXME("Fragment program error at position %d: %s\n\n", pos,
3372             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3373         shader_arb_dump_program_source(fprogram);
3374     }
3375     else
3376     {
3377         GLint native;
3378
3379         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3380         checkGLcall("glGetProgramivARB()");
3381         if (!native) WARN("Program exceeds native resource limits.\n");
3382     }
3383
3384     return program_id;
3385 }
3386
3387 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3388         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3389 {
3390     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3391
3392     if(condcode)
3393     {
3394         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3395         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3396         /* Calculate the > 0.0031308 case */
3397         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3398         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3399         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3400         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3401         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3402         /* Calculate the < case */
3403         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3404     }
3405     else
3406     {
3407         /* Calculate the > 0.0031308 case */
3408         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3409         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3410         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3411         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3412         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3413         /* Calculate the < case */
3414         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3415         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3416         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3417         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3418         /* Store the components > 0.0031308 in the destination */
3419         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3420         /* Add the components that are < 0.0031308 */
3421         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3422         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3423         * result.color writes(.rgb first, then .a), or handle overwriting already written
3424         * components. The assembler uses a temporary register in this case, which is usually
3425         * not allocated from one of our registers that were used earlier.
3426         */
3427     }
3428     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3429 }
3430
3431 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3432 {
3433     const struct wined3d_shader_lconst *constant;
3434
3435     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3436     {
3437         if (constant->idx == idx)
3438         {
3439             return constant->value;
3440         }
3441     }
3442     return NULL;
3443 }
3444
3445 static void init_ps_input(const struct wined3d_shader *shader,
3446         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3447 {
3448     static const char * const texcoords[8] =
3449     {
3450         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3451         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3452     };
3453     unsigned int i;
3454     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3455     const char *semantic_name;
3456     DWORD semantic_idx;
3457
3458     switch(args->super.vp_mode)
3459     {
3460         case pretransformed:
3461         case fixedfunction:
3462             /* The pixelshader has to collect the varyings on its own. In any case properly load
3463              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3464              * other attribs to 0.0.
3465              *
3466              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3467              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3468              * load the texcoord attrib pointers to match the pixel shader signature
3469              */
3470             for(i = 0; i < MAX_REG_INPUT; i++)
3471             {
3472                 semantic_name = sig[i].semantic_name;
3473                 semantic_idx = sig[i].semantic_idx;
3474                 if (!semantic_name) continue;
3475
3476                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3477                 {
3478                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3479                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3480                     else priv->ps_input[i] = "0.0";
3481                 }
3482                 else if(args->super.vp_mode == fixedfunction)
3483                 {
3484                     priv->ps_input[i] = "0.0";
3485                 }
3486                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3487                 {
3488                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3489                     else priv->ps_input[i] = "0.0";
3490                 }
3491                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3492                 {
3493                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3494                     else priv->ps_input[i] = "0.0";
3495                 }
3496                 else
3497                 {
3498                     priv->ps_input[i] = "0.0";
3499                 }
3500
3501                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3502             }
3503             break;
3504
3505         case vertexshader:
3506             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3507              * fragment.color
3508              */
3509             for(i = 0; i < 8; i++)
3510             {
3511                 priv->ps_input[i] = texcoords[i];
3512             }
3513             priv->ps_input[8] = "fragment.color.primary";
3514             priv->ps_input[9] = "fragment.color.secondary";
3515             break;
3516     }
3517 }
3518
3519 /* GL locking is done by the caller */
3520 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3521         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3522         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3523 {
3524     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3525     const struct wined3d_shader_lconst *lconst;
3526     const DWORD *function = shader->function;
3527     GLuint retval;
3528     char fragcolor[16];
3529     DWORD *lconst_map = local_const_mapping(shader), next_local;
3530     struct shader_arb_ctx_priv priv_ctx;
3531     BOOL dcl_td = FALSE;
3532     BOOL want_nv_prog = FALSE;
3533     struct arb_pshader_private *shader_priv = shader->backend_data;
3534     GLint errPos;
3535     DWORD map;
3536
3537     char srgbtmp[4][4];
3538     unsigned int i, found = 0;
3539
3540     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3541     {
3542         if (!(map & 1)
3543                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3544                 || (reg_maps->shader_version.major < 2 && !i))
3545             continue;
3546
3547         sprintf(srgbtmp[found], "R%u", i);
3548         ++found;
3549         if (found == 4) break;
3550     }
3551
3552     switch(found) {
3553         case 0:
3554             sprintf(srgbtmp[0], "TA");
3555             sprintf(srgbtmp[1], "TB");
3556             sprintf(srgbtmp[2], "TC");
3557             sprintf(srgbtmp[3], "TD");
3558             dcl_td = TRUE;
3559             break;
3560         case 1:
3561             sprintf(srgbtmp[1], "TA");
3562             sprintf(srgbtmp[2], "TB");
3563             sprintf(srgbtmp[3], "TC");
3564             break;
3565         case 2:
3566             sprintf(srgbtmp[2], "TA");
3567             sprintf(srgbtmp[3], "TB");
3568             break;
3569         case 3:
3570             sprintf(srgbtmp[3], "TA");
3571             break;
3572         case 4:
3573             break;
3574     }
3575
3576     /*  Create the hw ARB shader */
3577     memset(&priv_ctx, 0, sizeof(priv_ctx));
3578     priv_ctx.cur_ps_args = args;
3579     priv_ctx.compiled_fprog = compiled;
3580     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3581     init_ps_input(shader, args, &priv_ctx);
3582     list_init(&priv_ctx.control_frames);
3583
3584     /* Avoid enabling NV_fragment_program* if we do not need it.
3585      *
3586      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3587      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3588      * is faster than what we gain from using higher native instructions. There are some things though
3589      * that cannot be emulated. In that case enable the extensions.
3590      * If the extension is enabled, instruction handlers that support both ways will use it.
3591      *
3592      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3593      * So enable the best we can get.
3594      */
3595     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3596        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3597     {
3598         want_nv_prog = TRUE;
3599     }
3600
3601     shader_addline(buffer, "!!ARBfp1.0\n");
3602     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3603     {
3604         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3605         priv_ctx.target_version = NV3;
3606     }
3607     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3608     {
3609         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3610         priv_ctx.target_version = NV2;
3611     } else {
3612         if(want_nv_prog)
3613         {
3614             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3615              * limits properly
3616              */
3617             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3618             ERR("Try GLSL\n");
3619         }
3620         priv_ctx.target_version = ARB;
3621     }
3622
3623     if (reg_maps->rt_mask > 1)
3624     {
3625         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3626     }
3627
3628     if (reg_maps->shader_version.major < 3)
3629     {
3630         switch(args->super.fog) {
3631             case FOG_OFF:
3632                 break;
3633             case FOG_LINEAR:
3634                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3635                 break;
3636             case FOG_EXP:
3637                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3638                 break;
3639             case FOG_EXP2:
3640                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3641                 break;
3642         }
3643     }
3644
3645     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3646      * unused temps away(but occupies them for the whole shader if they're used once). Always
3647      * declaring them avoids tricky bookkeeping work
3648      */
3649     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3650     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3651     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3652     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3653     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3654     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3655     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3656
3657     if (reg_maps->shader_version.major < 2)
3658     {
3659         strcpy(fragcolor, "R0");
3660     }
3661     else
3662     {
3663         if (args->super.srgb_correction)
3664         {
3665             if (shader->u.ps.color0_mov)
3666             {
3667                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3668             }
3669             else
3670             {
3671                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3672                 strcpy(fragcolor, "TMP_COLOR");
3673             }
3674         } else {
3675             strcpy(fragcolor, "result.color");
3676         }
3677     }
3678
3679     if(args->super.srgb_correction) {
3680         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3681                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3682         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3683                        srgb_sub_high, 0.0, 0.0, 0.0);
3684     }
3685
3686     /* Base Declarations */
3687     next_local = shader_generate_arb_declarations(shader, reg_maps,
3688             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3689
3690     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3691     {
3692         unsigned char bump_const;
3693
3694         if (!(map & 1)) continue;
3695
3696         bump_const = compiled->numbumpenvmatconsts;
3697         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3698         compiled->bumpenvmatconst[bump_const].texunit = i;
3699         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3700         compiled->luminanceconst[bump_const].texunit = i;
3701
3702         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3703          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3704          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3705          * textures due to conditional NP2 restrictions)
3706          *
3707          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3708          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3709          * their location is shader dependent anyway and they cannot be loaded globally.
3710          */
3711         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3712         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3713                        i, compiled->bumpenvmatconst[bump_const].const_num);
3714         compiled->numbumpenvmatconsts = bump_const + 1;
3715
3716         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3717
3718         compiled->luminanceconst[bump_const].const_num = next_local++;
3719         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3720                        i, compiled->luminanceconst[bump_const].const_num);
3721     }
3722
3723     for(i = 0; i < MAX_CONST_I; i++)
3724     {
3725         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3726         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3727         {
3728             const DWORD *control_values = find_loop_control_values(shader, i);
3729
3730             if(control_values)
3731             {
3732                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3733                                 control_values[0], control_values[1], control_values[2]);
3734             }
3735             else
3736             {
3737                 compiled->int_consts[i] = next_local;
3738                 compiled->num_int_consts++;
3739                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3740             }
3741         }
3742     }
3743
3744     if(reg_maps->vpos || reg_maps->usesdsy)
3745     {
3746         compiled->ycorrection = next_local;
3747         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3748
3749         if(reg_maps->vpos)
3750         {
3751             shader_addline(buffer, "TEMP vpos;\n");
3752             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3753              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3754              * ycorrection.z: 1.0
3755              * ycorrection.w: 0.0
3756              */
3757             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3758             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3759         }
3760     }
3761     else
3762     {
3763         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3764     }
3765
3766     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3767      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3768      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3769      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3770      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3771      * shader compilation errors and the subsequent errors when drawing with this shader. */
3772     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3773         unsigned char cur_fixup_sampler = 0;
3774
3775         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3776         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3777         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3778
3779         fixup->offset = next_local;
3780         fixup->super.active = 0;
3781
3782         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3783             if (!(map & (1 << i))) continue;
3784
3785             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3786                 fixup->super.active |= (1 << i);
3787                 fixup->super.idx[i] = cur_fixup_sampler++;
3788             } else {
3789                 FIXME("No free constant found to load NP2 fixup data into shader. "
3790                       "Sampling from this texture will probably look wrong.\n");
3791                 break;
3792             }
3793         }
3794
3795         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3796         if (fixup->super.num_consts) {
3797             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3798                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3799         }
3800     }
3801
3802     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3803     {
3804         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3805     }
3806
3807     /* Base Shader Body */
3808     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3809
3810     if(args->super.srgb_correction) {
3811         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3812                                   priv_ctx.target_version >= NV2);
3813     }
3814
3815     if(strcmp(fragcolor, "result.color")) {
3816         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3817     }
3818     shader_addline(buffer, "END\n");
3819
3820     /* TODO: change to resource.glObjectHandle or something like that */
3821     GL_EXTCALL(glGenProgramsARB(1, &retval));
3822
3823     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3824     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3825
3826     TRACE("Created hw pixel shader, prg=%d\n", retval);
3827     /* Create the program and check for errors */
3828     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3829                buffer->bsize, buffer->buffer));
3830     checkGLcall("glProgramStringARB()");
3831
3832     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3833     if (errPos != -1)
3834     {
3835         FIXME("HW PixelShader Error at position %d: %s\n\n",
3836               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3837         shader_arb_dump_program_source(buffer->buffer);
3838         retval = 0;
3839     }
3840     else
3841     {
3842         GLint native;
3843
3844         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3845         checkGLcall("glGetProgramivARB()");
3846         if (!native) WARN("Program exceeds native resource limits.\n");
3847     }
3848
3849     /* Load immediate constants */
3850     if (lconst_map)
3851     {
3852         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3853         {
3854             const float *value = (const float *)lconst->value;
3855             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3856             checkGLcall("glProgramLocalParameter4fvARB");
3857         }
3858         HeapFree(GetProcessHeap(), 0, lconst_map);
3859     }
3860
3861     return retval;
3862 }
3863
3864 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3865 {
3866     unsigned int i;
3867     int ret;
3868
3869     for(i = 0; i < MAX_REG_INPUT; i++)
3870     {
3871         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3872         {
3873             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3874             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3875             continue;
3876         }
3877
3878         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3879         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3880         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3881         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3882         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3883         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3884     }
3885     return 0;
3886 }
3887
3888 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3889 {
3890     struct wined3d_shader_signature_element *new;
3891     int i;
3892     char *name;
3893
3894     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3895     for(i = 0; i < MAX_REG_INPUT; i++)
3896     {
3897         if (!sig[i].semantic_name) continue;
3898
3899         new[i] = sig[i];
3900         /* Clone the semantic string */
3901         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3902         strcpy(name, sig[i].semantic_name);
3903         new[i].semantic_name = name;
3904     }
3905     return new;
3906 }
3907
3908 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3909 {
3910     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3911     struct ps_signature *found_sig;
3912
3913     if (entry)
3914     {
3915         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3916         TRACE("Found existing signature %u\n", found_sig->idx);
3917         return found_sig->idx;
3918     }
3919     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3920     found_sig->sig = clone_sig(sig);
3921     found_sig->idx = priv->ps_sig_number++;
3922     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3923     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3924     {
3925         ERR("Failed to insert program entry.\n");
3926     }
3927     return found_sig->idx;
3928 }
3929
3930 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3931         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3932 {
3933     unsigned int i, j;
3934     static const char * const texcoords[8] =
3935     {
3936         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3937         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3938     };
3939     struct wined3d_device *device = shader->device;
3940     const struct wined3d_shader_signature_element *sig;
3941     const char *semantic_name;
3942     DWORD semantic_idx, reg_idx;
3943
3944     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3945      * and varying 9 to result.color.secondary
3946      */
3947     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3948     {
3949         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3950         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3951         "result.color.primary", "result.color.secondary"
3952     };
3953
3954     if(sig_num == ~0)
3955     {
3956         TRACE("Pixel shader uses builtin varyings\n");
3957         /* Map builtins to builtins */
3958         for(i = 0; i < 8; i++)
3959         {
3960             priv_ctx->texcrd_output[i] = texcoords[i];
3961         }
3962         priv_ctx->color_output[0] = "result.color.primary";
3963         priv_ctx->color_output[1] = "result.color.secondary";
3964         priv_ctx->fog_output = "result.fogcoord";
3965
3966         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3967         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3968         {
3969             semantic_name = shader->output_signature[i].semantic_name;
3970             if (!semantic_name) continue;
3971
3972             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3973             {
3974                 TRACE("o%u is TMP_OUT\n", i);
3975                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3976                 else priv_ctx->vs_output[i] = "TA";
3977             }
3978             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3979             {
3980                 TRACE("o%u is result.pointsize\n", i);
3981                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3982                 else priv_ctx->vs_output[i] = "TA";
3983             }
3984             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3985             {
3986                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
3987                 if (!shader->output_signature[i].semantic_idx)
3988                     priv_ctx->vs_output[i] = "result.color.primary";
3989                 else if (shader->output_signature[i].semantic_idx == 1)
3990                     priv_ctx->vs_output[i] = "result.color.secondary";
3991                 else priv_ctx->vs_output[i] = "TA";
3992             }
3993             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3994             {
3995                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
3996                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3997                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
3998             }
3999             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4000             {
4001                 TRACE("o%u is result.fogcoord\n", i);
4002                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4003                 else priv_ctx->vs_output[i] = "result.fogcoord";
4004             }
4005             else
4006             {
4007                 priv_ctx->vs_output[i] = "TA";
4008             }
4009         }
4010         return;
4011     }
4012
4013     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4014      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4015      */
4016     sig = device->stateBlock->state.pixel_shader->input_signature;
4017     TRACE("Pixel shader uses declared varyings\n");
4018
4019     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4020     for(i = 0; i < 8; i++)
4021     {
4022         priv_ctx->texcrd_output[i] = "TA";
4023     }
4024     priv_ctx->color_output[0] = "TA";
4025     priv_ctx->color_output[1] = "TA";
4026     priv_ctx->fog_output = "TA";
4027
4028     for(i = 0; i < MAX_REG_INPUT; i++)
4029     {
4030         semantic_name = sig[i].semantic_name;
4031         semantic_idx = sig[i].semantic_idx;
4032         reg_idx = sig[i].register_idx;
4033         if (!semantic_name) continue;
4034
4035         /* If a declared input register is not written by builtin arguments, don't write to it.
4036          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4037          *
4038          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4039          * to TMP_OUT in any case
4040          */
4041         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4042         {
4043             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4044         }
4045         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4046         {
4047             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4048         }
4049         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4050         {
4051             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4052         }
4053         else
4054         {
4055             continue;
4056         }
4057
4058         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4059                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4060         {
4061             compiled->need_color_unclamp = TRUE;
4062         }
4063     }
4064
4065     /* Map declared to declared */
4066     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4067     {
4068         /* Write unread output to TA to throw them away */
4069         priv_ctx->vs_output[i] = "TA";
4070         semantic_name = shader->output_signature[i].semantic_name;
4071         if (!semantic_name) continue;
4072
4073         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4074                 && !shader->output_signature[i].semantic_idx)
4075         {
4076             priv_ctx->vs_output[i] = "TMP_OUT";
4077             continue;
4078         }
4079         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4080                 && !shader->output_signature[i].semantic_idx)
4081         {
4082             priv_ctx->vs_output[i] = "result.pointsize";
4083             continue;
4084         }
4085
4086         for(j = 0; j < MAX_REG_INPUT; j++)
4087         {
4088             if (!sig[j].semantic_name) continue;
4089
4090             if (!strcmp(sig[j].semantic_name, semantic_name)
4091                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4092             {
4093                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4094
4095                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4096                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4097                 {
4098                     compiled->need_color_unclamp = TRUE;
4099                 }
4100             }
4101         }
4102     }
4103 }
4104
4105 /* GL locking is done by the caller */
4106 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4107         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4108         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4109 {
4110     const struct arb_vshader_private *shader_data = shader->backend_data;
4111     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4112     const struct wined3d_shader_lconst *lconst;
4113     const DWORD *function = shader->function;
4114     GLuint ret;
4115     DWORD next_local, *lconst_map = local_const_mapping(shader);
4116     struct shader_arb_ctx_priv priv_ctx;
4117     unsigned int i;
4118     GLint errPos;
4119
4120     memset(&priv_ctx, 0, sizeof(priv_ctx));
4121     priv_ctx.cur_vs_args = args;
4122     list_init(&priv_ctx.control_frames);
4123     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4124
4125     /*  Create the hw ARB shader */
4126     shader_addline(buffer, "!!ARBvp1.0\n");
4127
4128     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4129      * mesurable performance penalty, and we can always make use of it for clipplanes.
4130      */
4131     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4132     {
4133         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4134         priv_ctx.target_version = NV3;
4135         shader_addline(buffer, "ADDRESS aL;\n");
4136     }
4137     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4138     {
4139         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4140         priv_ctx.target_version = NV2;
4141         shader_addline(buffer, "ADDRESS aL;\n");
4142     } else {
4143         priv_ctx.target_version = ARB;
4144     }
4145
4146     shader_addline(buffer, "TEMP TMP_OUT;\n");
4147     if (need_helper_const(shader_data, reg_maps, gl_info))
4148     {
4149         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4150     }
4151     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4152     {
4153         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4154         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4155     }
4156
4157     shader_addline(buffer, "TEMP TA;\n");
4158     shader_addline(buffer, "TEMP TB;\n");
4159
4160     /* Base Declarations */
4161     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4162             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4163
4164     for(i = 0; i < MAX_CONST_I; i++)
4165     {
4166         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4167         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4168         {
4169             const DWORD *control_values = find_loop_control_values(shader, i);
4170
4171             if(control_values)
4172             {
4173                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4174                                 control_values[0], control_values[1], control_values[2]);
4175             }
4176             else
4177             {
4178                 compiled->int_consts[i] = next_local;
4179                 compiled->num_int_consts++;
4180                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4181             }
4182         }
4183     }
4184
4185     /* We need a constant to fixup the final position */
4186     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4187     compiled->pos_fixup = next_local++;
4188
4189     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4190      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4191      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4192      * a replacement shader depend on the texcoord.w being set properly.
4193      *
4194      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4195      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4196      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4197      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4198      * this can eat a number of instructions, so skip it unless this cap is set as well
4199      */
4200     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4201     {
4202         struct wined3d_device *device = shader->device;
4203         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4204         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4205
4206         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4207         {
4208             int i;
4209             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4210             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4211             {
4212                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4213                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4214             }
4215         }
4216     }
4217
4218     /* The shader starts with the main function */
4219     priv_ctx.in_main_func = TRUE;
4220     /* Base Shader Body */
4221     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4222
4223     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4224             shader_data, args, reg_maps, gl_info, buffer);
4225
4226     shader_addline(buffer, "END\n");
4227
4228     /* TODO: change to resource.glObjectHandle or something like that */
4229     GL_EXTCALL(glGenProgramsARB(1, &ret));
4230
4231     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4232     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4233
4234     TRACE("Created hw vertex shader, prg=%d\n", ret);
4235     /* Create the program and check for errors */
4236     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4237                buffer->bsize, buffer->buffer));
4238     checkGLcall("glProgramStringARB()");
4239
4240     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4241     if (errPos != -1)
4242     {
4243         FIXME("HW VertexShader Error at position %d: %s\n\n",
4244               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4245         shader_arb_dump_program_source(buffer->buffer);
4246         ret = -1;
4247     }
4248     else
4249     {
4250         GLint native;
4251
4252         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4253         checkGLcall("glGetProgramivARB()");
4254         if (!native) WARN("Program exceeds native resource limits.\n");
4255
4256         /* Load immediate constants */
4257         if (lconst_map)
4258         {
4259             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4260             {
4261                 const float *value = (const float *)lconst->value;
4262                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4263             }
4264         }
4265     }
4266     HeapFree(GetProcessHeap(), 0, lconst_map);
4267
4268     return ret;
4269 }
4270
4271 /* GL locking is done by the caller */
4272 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4273         const struct arb_ps_compile_args *args)
4274 {
4275     struct wined3d_device *device = shader->device;
4276     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4277     UINT i;
4278     DWORD new_size;
4279     struct arb_ps_compiled_shader *new_array;
4280     struct wined3d_shader_buffer buffer;
4281     struct arb_pshader_private *shader_data;
4282     GLuint ret;
4283
4284     if (!shader->backend_data)
4285     {
4286         struct shader_arb_priv *priv = device->shader_priv;
4287
4288         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4289         shader_data = shader->backend_data;
4290         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4291
4292         if (shader->reg_maps.shader_version.major < 3)
4293             shader_data->input_signature_idx = ~0;
4294         else
4295             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4296
4297         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4298
4299         if (!device->vs_clipping)
4300             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4301                     gl_info->limits.texture_stages - 1);
4302         else
4303             shader_data->clipplane_emulation = ~0U;
4304     }
4305     shader_data = shader->backend_data;
4306
4307     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4308      * so a linear search is more performant than a hashmap or a binary search
4309      * (cache coherency etc)
4310      */
4311     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4312     {
4313         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4314             return &shader_data->gl_shaders[i];
4315     }
4316
4317     TRACE("No matching GL shader found, compiling a new shader\n");
4318     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4319         if (shader_data->num_gl_shaders)
4320         {
4321             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4322             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4323                                     new_size * sizeof(*shader_data->gl_shaders));
4324         } else {
4325             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4326             new_size = 1;
4327         }
4328
4329         if(!new_array) {
4330             ERR("Out of memory\n");
4331             return 0;
4332         }
4333         shader_data->gl_shaders = new_array;
4334         shader_data->shader_array_size = new_size;
4335     }
4336
4337     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4338
4339     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4340
4341     if (!shader_buffer_init(&buffer))
4342     {
4343         ERR("Failed to initialize shader buffer.\n");
4344         return 0;
4345     }
4346
4347     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4348             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4349     shader_buffer_free(&buffer);
4350     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4351
4352     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4353 }
4354
4355 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4356                                  const DWORD use_map, BOOL skip_int) {
4357     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4358     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4359     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4360     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4361     if(stored->ps_signature != new->ps_signature) return FALSE;
4362     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4363     if(skip_int) return TRUE;
4364
4365     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4366 }
4367
4368 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4369         const struct arb_vs_compile_args *args)
4370 {
4371     struct wined3d_device *device = shader->device;
4372     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4373     DWORD use_map = device->strided_streams.use_map;
4374     UINT i;
4375     DWORD new_size;
4376     struct arb_vs_compiled_shader *new_array;
4377     struct wined3d_shader_buffer buffer;
4378     struct arb_vshader_private *shader_data;
4379     GLuint ret;
4380
4381     if (!shader->backend_data)
4382     {
4383         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4384
4385         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4386         shader_data = shader->backend_data;
4387
4388         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4389                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4390         {
4391             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4392             {
4393                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4394                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4395                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4396             }
4397             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4398                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4399             else if (reg_maps->max_rel_offset > 63)
4400                 shader_data->rel_offset = reg_maps->min_rel_offset;
4401         }
4402     }
4403     shader_data = shader->backend_data;
4404
4405     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4406      * so a linear search is more performant than a hashmap or a binary search
4407      * (cache coherency etc)
4408      */
4409     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4410         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4411                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4412         {
4413             return &shader_data->gl_shaders[i];
4414         }
4415     }
4416
4417     TRACE("No matching GL shader found, compiling a new shader\n");
4418
4419     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4420         if (shader_data->num_gl_shaders)
4421         {
4422             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4423             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4424                                     new_size * sizeof(*shader_data->gl_shaders));
4425         } else {
4426             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4427             new_size = 1;
4428         }
4429
4430         if(!new_array) {
4431             ERR("Out of memory\n");
4432             return 0;
4433         }
4434         shader_data->gl_shaders = new_array;
4435         shader_data->shader_array_size = new_size;
4436     }
4437
4438     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4439
4440     if (!shader_buffer_init(&buffer))
4441     {
4442         ERR("Failed to initialize shader buffer.\n");
4443         return 0;
4444     }
4445
4446     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4447             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4448     shader_buffer_free(&buffer);
4449     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4450
4451     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4452 }
4453
4454 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4455         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4456 {
4457     struct wined3d_device *device = shader->device;
4458     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4459     int i;
4460     WORD int_skip;
4461
4462     find_ps_compile_args(state, shader, &args->super);
4463
4464     /* This forces all local boolean constants to 1 to make them stateblock independent */
4465     args->bools = shader->reg_maps.local_bool_consts;
4466
4467     for(i = 0; i < MAX_CONST_B; i++)
4468     {
4469         if (state->ps_consts_b[i])
4470             args->bools |= ( 1 << i);
4471     }
4472
4473     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4474      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4475      * duplicate the shader than have a no-op KIL instruction in every shader
4476      */
4477     if (!device->vs_clipping && use_vs(state)
4478             && state->render_states[WINED3DRS_CLIPPING]
4479             && state->render_states[WINED3DRS_CLIPPLANEENABLE])
4480         args->clip = 1;
4481     else
4482         args->clip = 0;
4483
4484     /* Skip if unused or local, or supported natively */
4485     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4486     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4487     {
4488         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4489         return;
4490     }
4491
4492     for(i = 0; i < MAX_CONST_I; i++)
4493     {
4494         if(int_skip & (1 << i))
4495         {
4496             args->loop_ctrl[i][0] = 0;
4497             args->loop_ctrl[i][1] = 0;
4498             args->loop_ctrl[i][2] = 0;
4499         }
4500         else
4501         {
4502             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4503             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4504             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4505         }
4506     }
4507 }
4508
4509 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4510         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4511 {
4512     struct wined3d_device *device = shader->device;
4513     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4514     int i;
4515     WORD int_skip;
4516
4517     find_vs_compile_args(state, shader, &args->super);
4518
4519     args->clip.boolclip_compare = 0;
4520     if (use_ps(state))
4521     {
4522         const struct wined3d_shader *ps = state->pixel_shader;
4523         const struct arb_pshader_private *shader_priv = ps->backend_data;
4524         args->ps_signature = shader_priv->input_signature_idx;
4525
4526         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4527     }
4528     else
4529     {
4530         args->ps_signature = ~0;
4531         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4532         {
4533             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4534         }
4535         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4536     }
4537
4538     if (args->clip.boolclip.clip_texcoord)
4539     {
4540         if (state->render_states[WINED3DRS_CLIPPING])
4541             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3DRS_CLIPPLANEENABLE];
4542         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4543     }
4544
4545     /* This forces all local boolean constants to 1 to make them stateblock independent */
4546     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4547     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4548     for(i = 0; i < MAX_CONST_B; i++)
4549     {
4550         if (state->vs_consts_b[i])
4551             args->clip.boolclip.bools |= ( 1 << i);
4552     }
4553
4554     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4555     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4556     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4557     args->vertex.samplers[3] = 0;
4558
4559     /* Skip if unused or local */
4560     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4561     /* This is about flow control, not clipping. */
4562     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4563     {
4564         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4565         return;
4566     }
4567
4568     for(i = 0; i < MAX_CONST_I; i++)
4569     {
4570         if(int_skip & (1 << i))
4571         {
4572             args->loop_ctrl[i][0] = 0;
4573             args->loop_ctrl[i][1] = 0;
4574             args->loop_ctrl[i][2] = 0;
4575         }
4576         else
4577         {
4578             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4579             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4580             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4581         }
4582     }
4583 }
4584
4585 /* GL locking is done by the caller */
4586 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4587 {
4588     struct wined3d_device *device = context->swapchain->device;
4589     struct shader_arb_priv *priv = device->shader_priv;
4590     const struct wined3d_gl_info *gl_info = context->gl_info;
4591     const struct wined3d_state *state = &device->stateBlock->state;
4592     int i;
4593
4594     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4595     if (usePS)
4596     {
4597         struct wined3d_shader *ps = state->pixel_shader;
4598         struct arb_ps_compile_args compile_args;
4599         struct arb_ps_compiled_shader *compiled;
4600
4601         TRACE("Using pixel shader %p.\n", ps);
4602         find_arb_ps_compile_args(state, ps, &compile_args);
4603         compiled = find_arb_pshader(ps, &compile_args);
4604         priv->current_fprogram_id = compiled->prgId;
4605         priv->compiled_fprog = compiled;
4606
4607         /* Bind the fragment program */
4608         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4609         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4610
4611         if(!priv->use_arbfp_fixed_func) {
4612             /* Enable OpenGL fragment programs */
4613             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4614             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4615         }
4616         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4617                 device, priv->current_fprogram_id);
4618
4619         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4620          * a 1.x and newer shader, reload the first 8 constants
4621          */
4622         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4623         {
4624             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4625             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4626             for(i = 0; i < 8; i++)
4627             {
4628                 priv->pshader_const_dirty[i] = 1;
4629             }
4630             /* Also takes care of loading local constants */
4631             shader_arb_load_constants(context, TRUE, FALSE);
4632         }
4633         else
4634         {
4635             UINT rt_height = state->fb->render_targets[0]->resource.height;
4636             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4637         }
4638
4639         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4640         if (compiled->np2fixup_info.super.active)
4641             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4642     }
4643     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4644     {
4645         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4646         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4647         * replacement shader
4648         */
4649         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4650         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4651         priv->current_fprogram_id = 0;
4652     }
4653
4654     if (useVS)
4655     {
4656         struct wined3d_shader *vs = state->vertex_shader;
4657         struct arb_vs_compile_args compile_args;
4658         struct arb_vs_compiled_shader *compiled;
4659
4660         TRACE("Using vertex shader %p\n", vs);
4661         find_arb_vs_compile_args(state, vs, &compile_args);
4662         compiled = find_arb_vshader(vs, &compile_args);
4663         priv->current_vprogram_id = compiled->prgId;
4664         priv->compiled_vprog = compiled;
4665
4666         /* Bind the vertex program */
4667         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4668         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4669
4670         /* Enable OpenGL vertex programs */
4671         glEnable(GL_VERTEX_PROGRAM_ARB);
4672         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4673         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4674         shader_arb_vs_local_constants(compiled, context, state);
4675
4676         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4677             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4678
4679             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4680             {
4681                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4682                 checkGLcall("glClampColorARB");
4683             } else {
4684                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4685             }
4686         }
4687     }
4688     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4689     {
4690         priv->current_vprogram_id = 0;
4691         glDisable(GL_VERTEX_PROGRAM_ARB);
4692         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4693     }
4694 }
4695
4696 /* GL locking is done by the caller */
4697 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4698         enum tex_types tex_type, const SIZE *ds_mask_size)
4699 {
4700     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4701     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4702     struct shader_arb_priv *priv = shader_priv;
4703     GLuint *blt_fprogram;
4704
4705     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4706     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4707     glEnable(GL_VERTEX_PROGRAM_ARB);
4708
4709     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4710     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4711     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4712     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4713     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4714 }
4715
4716 /* GL locking is done by the caller */
4717 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4718 {
4719     struct shader_arb_priv *priv = shader_priv;
4720
4721     if (priv->current_vprogram_id) {
4722         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4723         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4724
4725         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4726     }
4727     else
4728     {
4729         glDisable(GL_VERTEX_PROGRAM_ARB);
4730         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4731     }
4732
4733     if (priv->current_fprogram_id) {
4734         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4735         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4736
4737         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4738     }
4739     else if(!priv->use_arbfp_fixed_func)
4740     {
4741         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4742         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4743     }
4744 }
4745
4746 static void shader_arb_destroy(struct wined3d_shader *shader)
4747 {
4748     struct wined3d_device *device = shader->device;
4749     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4750
4751     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4752     {
4753         struct arb_pshader_private *shader_data = shader->backend_data;
4754         UINT i;
4755
4756         if(!shader_data) return; /* This can happen if a shader was never compiled */
4757
4758         if (shader_data->num_gl_shaders)
4759         {
4760             struct wined3d_context *context = context_acquire(device, NULL);
4761
4762             ENTER_GL();
4763             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4764             {
4765                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4766                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4767             }
4768             LEAVE_GL();
4769
4770             context_release(context);
4771         }
4772
4773         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4774         HeapFree(GetProcessHeap(), 0, shader_data);
4775         shader->backend_data = NULL;
4776     }
4777     else
4778     {
4779         struct arb_vshader_private *shader_data = shader->backend_data;
4780         UINT i;
4781
4782         if(!shader_data) return; /* This can happen if a shader was never compiled */
4783
4784         if (shader_data->num_gl_shaders)
4785         {
4786             struct wined3d_context *context = context_acquire(device, NULL);
4787
4788             ENTER_GL();
4789             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4790             {
4791                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4792                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4793             }
4794             LEAVE_GL();
4795
4796             context_release(context);
4797         }
4798
4799         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4800         HeapFree(GetProcessHeap(), 0, shader_data);
4801         shader->backend_data = NULL;
4802     }
4803 }
4804
4805 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4806 {
4807     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4808     return compare_sig(key, e->sig);
4809 }
4810
4811 static const struct wine_rb_functions sig_tree_functions =
4812 {
4813     wined3d_rb_alloc,
4814     wined3d_rb_realloc,
4815     wined3d_rb_free,
4816     sig_tree_compare
4817 };
4818
4819 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4820 {
4821     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4822
4823     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4824             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4825     if (!priv->vshader_const_dirty)
4826         goto fail;
4827     memset(priv->vshader_const_dirty, 1,
4828            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4829
4830     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4831             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4832     if (!priv->pshader_const_dirty)
4833         goto fail;
4834     memset(priv->pshader_const_dirty, 1,
4835             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4836
4837     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4838     {
4839         ERR("RB tree init failed\n");
4840         goto fail;
4841     }
4842     device->shader_priv = priv;
4843     return WINED3D_OK;
4844
4845 fail:
4846     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4847     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4848     HeapFree(GetProcessHeap(), 0, priv);
4849     return E_OUTOFMEMORY;
4850 }
4851
4852 static void release_signature(struct wine_rb_entry *entry, void *context)
4853 {
4854     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4855     int i;
4856     for(i = 0; i < MAX_REG_INPUT; i++)
4857     {
4858         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4859     }
4860     HeapFree(GetProcessHeap(), 0, sig->sig);
4861     HeapFree(GetProcessHeap(), 0, sig);
4862 }
4863
4864 /* Context activation is done by the caller. */
4865 static void shader_arb_free(struct wined3d_device *device)
4866 {
4867     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4868     struct shader_arb_priv *priv = device->shader_priv;
4869     int i;
4870
4871     ENTER_GL();
4872     if(priv->depth_blt_vprogram_id) {
4873         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4874     }
4875     for (i = 0; i < tex_type_count; ++i)
4876     {
4877         if (priv->depth_blt_fprogram_id_full[i])
4878         {
4879             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4880         }
4881         if (priv->depth_blt_fprogram_id_masked[i])
4882         {
4883             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4884         }
4885     }
4886     LEAVE_GL();
4887
4888     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4889     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4890     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4891     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4892 }
4893
4894 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4895 {
4896     struct shader_arb_priv *priv = shader_priv;
4897
4898     if (priv->last_context == context)
4899         priv->last_context = NULL;
4900 }
4901
4902 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4903 {
4904     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4905     {
4906         DWORD vs_consts;
4907
4908         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4909          * for vertex programs. If the native limit is less than that it's
4910          * not very useful, and e.g. Mesa swrast returns 0, probably to
4911          * indicate it's a software implementation. */
4912         if (gl_info->limits.arb_vs_native_constants < 96)
4913             vs_consts = gl_info->limits.arb_vs_float_constants;
4914         else
4915             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4916
4917         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4918         {
4919             caps->VertexShaderVersion = 3;
4920             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4921         }
4922         else if (vs_consts >= 256)
4923         {
4924             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4925             caps->VertexShaderVersion = 2;
4926             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4927         }
4928         else
4929         {
4930             caps->VertexShaderVersion = 1;
4931             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4932         }
4933         caps->MaxVertexShaderConst = vs_consts;
4934     }
4935     else
4936     {
4937         caps->VertexShaderVersion = 0;
4938         caps->MaxVertexShaderConst = 0;
4939     }
4940
4941     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4942     {
4943         DWORD ps_consts;
4944
4945         /* Similar as above for vertex programs, but the minimum for fragment
4946          * programs is 24. */
4947         if (gl_info->limits.arb_ps_native_constants < 24)
4948             ps_consts = gl_info->limits.arb_ps_float_constants;
4949         else
4950             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4951
4952         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4953         {
4954             caps->PixelShaderVersion = 3;
4955             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4956         }
4957         else if (ps_consts >= 32)
4958         {
4959             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4960             caps->PixelShaderVersion = 2;
4961             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4962         }
4963         else
4964         {
4965             caps->PixelShaderVersion = 1;
4966             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4967         }
4968         caps->PixelShader1xMaxValue = 8.0f;
4969         caps->MaxPixelShaderConst = ps_consts;
4970     }
4971     else
4972     {
4973         caps->PixelShaderVersion = 0;
4974         caps->PixelShader1xMaxValue = 0.0f;
4975         caps->MaxPixelShaderConst = 0;
4976     }
4977
4978     caps->VSClipping = use_nv_clip(gl_info);
4979 }
4980
4981 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4982 {
4983     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4984     {
4985         TRACE("Checking support for color_fixup:\n");
4986         dump_color_fixup_desc(fixup);
4987     }
4988
4989     /* We support everything except complex conversions. */
4990     if (!is_complex_fixup(fixup))
4991     {
4992         TRACE("[OK]\n");
4993         return TRUE;
4994     }
4995
4996     TRACE("[FAILED]\n");
4997     return FALSE;
4998 }
4999
5000 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5001     DWORD shift;
5002     char write_mask[20], regstr[50];
5003     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5004     BOOL is_color = FALSE;
5005     const struct wined3d_shader_dst_param *dst;
5006
5007     if (!ins->dst_count) return;
5008
5009     dst = &ins->dst[0];
5010     shift = dst->shift;
5011     if (!shift) return; /* Saturate alone is handled by the instructions */
5012
5013     shader_arb_get_write_mask(ins, dst, write_mask);
5014     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5015
5016     /* Generate a line that does the output modifier computation
5017      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5018      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5019      */
5020     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5021                    regstr, write_mask, regstr, shift_tab[shift]);
5022 }
5023
5024 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5025 {
5026     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
5027     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
5028     /* WINED3DSIH_AND           */ NULL,
5029     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5030     /* WINED3DSIH_BREAK         */ shader_hw_break,
5031     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5032     /* WINED3DSIH_BREAKP        */ NULL,
5033     /* WINED3DSIH_CALL          */ shader_hw_call,
5034     /* WINED3DSIH_CALLNZ        */ NULL,
5035     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5036     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5037     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5038     /* WINED3DSIH_CUT           */ NULL,
5039     /* WINED3DSIH_DCL           */ NULL,
5040     /* WINED3DSIH_DEF           */ NULL,
5041     /* WINED3DSIH_DEFB          */ NULL,
5042     /* WINED3DSIH_DEFI          */ NULL,
5043     /* WINED3DSIH_DIV           */ NULL,
5044     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5045     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5046     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5047     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5048     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5049     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5050     /* WINED3DSIH_ELSE          */ shader_hw_else,
5051     /* WINED3DSIH_EMIT          */ NULL,
5052     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5053     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5054     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5055     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5056     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5057     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5058     /* WINED3DSIH_FTOI          */ NULL,
5059     /* WINED3DSIH_IADD          */ NULL,
5060     /* WINED3DSIH_IEQ           */ NULL,
5061     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5062     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5063     /* WINED3DSIH_IGE           */ NULL,
5064     /* WINED3DSIH_IMUL          */ NULL,
5065     /* WINED3DSIH_ITOF          */ NULL,
5066     /* WINED3DSIH_LABEL         */ shader_hw_label,
5067     /* WINED3DSIH_LD            */ NULL,
5068     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5069     /* WINED3DSIH_LOG           */ shader_hw_log,
5070     /* WINED3DSIH_LOGP          */ shader_hw_log,
5071     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5072     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5073     /* WINED3DSIH_LT            */ NULL,
5074     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5075     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5076     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5077     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5078     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5079     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5080     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5081     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5082     /* WINED3DSIH_MOV           */ shader_hw_mov,
5083     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5084     /* WINED3DSIH_MOVC          */ NULL,
5085     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5086     /* WINED3DSIH_NOP           */ shader_hw_nop,
5087     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5088     /* WINED3DSIH_PHASE         */ NULL,
5089     /* WINED3DSIH_POW           */ shader_hw_pow,
5090     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5091     /* WINED3DSIH_REP           */ shader_hw_rep,
5092     /* WINED3DSIH_RET           */ shader_hw_ret,
5093     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5094     /* WINED3DSIH_SAMPLE        */ NULL,
5095     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5096     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5097     /* WINED3DSIH_SETP          */ NULL,
5098     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5099     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5100     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5101     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5102     /* WINED3DSIH_SQRT          */ NULL,
5103     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5104     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5105     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5106     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5107     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5108     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5109     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5110     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5111     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5112     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5113     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5114     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5115     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5116     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5117     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5118     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5119     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5120     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5121     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5122     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5123     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5124     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5125     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5126     /* WINED3DSIH_UTOF          */ NULL,
5127 };
5128
5129 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5130         const struct wined3d_shader *shader, DWORD idx)
5131 {
5132     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5133     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5134     const struct wined3d_shader_lconst *constant;
5135     WORD bools = 0;
5136     WORD flag = (1 << idx);
5137     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5138
5139     if (reg_maps->local_bool_consts & flag)
5140     {
5141         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5142         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5143         {
5144             if (constant->idx == idx)
5145             {
5146                 return constant->value[0];
5147             }
5148         }
5149         ERR("Local constant not found\n");
5150         return FALSE;
5151     }
5152     else
5153     {
5154         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5155         else bools = priv->cur_ps_args->bools;
5156         return bools & flag;
5157     }
5158 }
5159
5160 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5161         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5162 {
5163     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5164     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5165
5166     /* Integer constants can either be a local constant, or they can be stored in the shader
5167      * type specific compile args. */
5168     if (reg_maps->local_int_consts & (1 << idx))
5169     {
5170         const struct wined3d_shader_lconst *constant;
5171
5172         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5173         {
5174             if (constant->idx == idx)
5175             {
5176                 loop_control->count = constant->value[0];
5177                 loop_control->start = constant->value[1];
5178                 /* Step is signed. */
5179                 loop_control->step = (int)constant->value[2];
5180                 return;
5181             }
5182         }
5183         /* If this happens the flag was set incorrectly */
5184         ERR("Local constant not found\n");
5185         loop_control->count = 0;
5186         loop_control->start = 0;
5187         loop_control->step = 0;
5188         return;
5189     }
5190
5191     switch (reg_maps->shader_version.type)
5192     {
5193         case WINED3D_SHADER_TYPE_VERTEX:
5194             /* Count and aL start value are unsigned */
5195             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5196             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5197             /* Step is signed. */
5198             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5199             break;
5200
5201         case WINED3D_SHADER_TYPE_PIXEL:
5202             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5203             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5204             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5205             break;
5206
5207         default:
5208             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5209             break;
5210     }
5211 }
5212
5213 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5214 {
5215     unsigned int i;
5216     struct wined3d_shader_dst_param *dst_param = NULL;
5217     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5218     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5219     if(!rec)
5220     {
5221         ERR("Out of memory\n");
5222         return;
5223     }
5224
5225     rec->ins = *ins;
5226     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5227     if(!dst_param) goto free;
5228     *dst_param = *ins->dst;
5229     if(ins->dst->reg.rel_addr)
5230     {
5231         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5232         if(!rel_addr) goto free;
5233         *rel_addr = *ins->dst->reg.rel_addr;
5234         dst_param->reg.rel_addr = rel_addr;
5235     }
5236     rec->ins.dst = dst_param;
5237
5238     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5239     if(!src_param) goto free;
5240     for(i = 0; i < ins->src_count; i++)
5241     {
5242         src_param[i] = ins->src[i];
5243         if(ins->src[i].reg.rel_addr)
5244         {
5245             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5246             if(!rel_addr) goto free;
5247             *rel_addr = *ins->src[i].reg.rel_addr;
5248             src_param[i].reg.rel_addr = rel_addr;
5249         }
5250     }
5251     rec->ins.src = src_param;
5252     list_add_tail(list, &rec->entry);
5253     return;
5254
5255 free:
5256     ERR("Out of memory\n");
5257     if(dst_param)
5258     {
5259         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5260         HeapFree(GetProcessHeap(), 0, dst_param);
5261     }
5262     if(src_param)
5263     {
5264         for(i = 0; i < ins->src_count; i++)
5265         {
5266             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5267         }
5268         HeapFree(GetProcessHeap(), 0, src_param);
5269     }
5270     HeapFree(GetProcessHeap(), 0, rec);
5271 }
5272
5273 static void free_recorded_instruction(struct list *list)
5274 {
5275     struct recorded_instruction *rec_ins, *entry2;
5276     unsigned int i;
5277
5278     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5279     {
5280         list_remove(&rec_ins->entry);
5281         if(rec_ins->ins.dst)
5282         {
5283             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5284             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5285         }
5286         if(rec_ins->ins.src)
5287         {
5288             for(i = 0; i < rec_ins->ins.src_count; i++)
5289             {
5290                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5291             }
5292             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5293         }
5294         HeapFree(GetProcessHeap(), 0, rec_ins);
5295     }
5296 }
5297
5298 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5299     SHADER_HANDLER hw_fct;
5300     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5301     const struct wined3d_shader *shader = ins->ctx->shader;
5302     struct control_frame *control_frame;
5303     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5304     BOOL bool_const;
5305
5306     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5307     {
5308         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5309         list_add_head(&priv->control_frames, &control_frame->entry);
5310
5311         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5312         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5313
5314         if(priv->target_version >= NV2)
5315         {
5316             control_frame->no.loop = priv->num_loops++;
5317             priv->loop_depth++;
5318         }
5319         else
5320         {
5321             /* Don't bother recording when we're in a not used if branch */
5322             if(priv->muted)
5323             {
5324                 return;
5325             }
5326
5327             if(!priv->recording)
5328             {
5329                 list_init(&priv->record);
5330                 priv->recording = TRUE;
5331                 control_frame->outer_loop = TRUE;
5332                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5333                 return; /* Instruction is handled */
5334             }
5335             /* Record this loop in the outer loop's recording */
5336         }
5337     }
5338     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5339     {
5340         if(priv->target_version >= NV2)
5341         {
5342             /* Nothing to do. The control frame is popped after the HW instr handler */
5343         }
5344         else
5345         {
5346             struct list *e = list_head(&priv->control_frames);
5347             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5348             list_remove(&control_frame->entry);
5349
5350             if(control_frame->outer_loop)
5351             {
5352                 unsigned int iteration;
5353                 int aL = 0;
5354                 struct list copy;
5355
5356                 /* Turn off recording before playback */
5357                 priv->recording = FALSE;
5358
5359                 /* Move the recorded instructions to a separate list and get them out of the private data
5360                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5361                  * be recorded again, thus priv->record might be overwritten
5362                  */
5363                 list_init(&copy);
5364                 list_move_tail(&copy, &priv->record);
5365                 list_init(&priv->record);
5366
5367                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5368                 {
5369                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5370                                    control_frame->loop_control.count, control_frame->loop_control.start,
5371                                    control_frame->loop_control.step);
5372                     aL = control_frame->loop_control.start;
5373                 }
5374                 else
5375                 {
5376                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5377                 }
5378
5379                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5380                 {
5381                     struct recorded_instruction *rec_ins;
5382                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5383                     {
5384                         priv->aL = aL;
5385                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5386                     }
5387                     else
5388                     {
5389                         shader_addline(buffer, "#Iteration %u\n", iteration);
5390                     }
5391
5392                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5393                     {
5394                         shader_arb_handle_instruction(&rec_ins->ins);
5395                     }
5396
5397                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5398                     {
5399                         aL += control_frame->loop_control.step;
5400                     }
5401                 }
5402                 shader_addline(buffer, "#end loop/rep\n");
5403
5404                 free_recorded_instruction(&copy);
5405                 HeapFree(GetProcessHeap(), 0, control_frame);
5406                 return; /* Instruction is handled */
5407             }
5408             else
5409             {
5410                 /* This is a nested loop. Proceed to the normal recording function */
5411                 HeapFree(GetProcessHeap(), 0, control_frame);
5412             }
5413         }
5414     }
5415
5416     if(priv->recording)
5417     {
5418         record_instruction(&priv->record, ins);
5419         return;
5420     }
5421
5422     /* boolean if */
5423     if(ins->handler_idx == WINED3DSIH_IF)
5424     {
5425         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5426         list_add_head(&priv->control_frames, &control_frame->entry);
5427         control_frame->type = IF;
5428
5429         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5430         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5431         if (!priv->muted && !bool_const)
5432         {
5433             shader_addline(buffer, "#if(FALSE){\n");
5434             priv->muted = TRUE;
5435             control_frame->muting = TRUE;
5436         }
5437         else shader_addline(buffer, "#if(TRUE) {\n");
5438
5439         return; /* Instruction is handled */
5440     }
5441     else if(ins->handler_idx == WINED3DSIH_IFC)
5442     {
5443         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5444         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5445         control_frame->type = IFC;
5446         control_frame->no.ifc = priv->num_ifcs++;
5447         list_add_head(&priv->control_frames, &control_frame->entry);
5448     }
5449     else if(ins->handler_idx == WINED3DSIH_ELSE)
5450     {
5451         struct list *e = list_head(&priv->control_frames);
5452         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5453
5454         if(control_frame->type == IF)
5455         {
5456             shader_addline(buffer, "#} else {\n");
5457             if(!priv->muted && !control_frame->muting)
5458             {
5459                 priv->muted = TRUE;
5460                 control_frame->muting = TRUE;
5461             }
5462             else if(control_frame->muting) priv->muted = FALSE;
5463             return; /* Instruction is handled. */
5464         }
5465         /* In case of an ifc, generate a HW shader instruction */
5466     }
5467     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5468     {
5469         struct list *e = list_head(&priv->control_frames);
5470         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5471
5472         if(control_frame->type == IF)
5473         {
5474             shader_addline(buffer, "#} endif\n");
5475             if(control_frame->muting) priv->muted = FALSE;
5476             list_remove(&control_frame->entry);
5477             HeapFree(GetProcessHeap(), 0, control_frame);
5478             return; /* Instruction is handled */
5479         }
5480     }
5481
5482     if(priv->muted) return;
5483
5484     /* Select handler */
5485     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5486
5487     /* Unhandled opcode */
5488     if (!hw_fct)
5489     {
5490         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5491         return;
5492     }
5493     hw_fct(ins);
5494
5495     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5496     {
5497         struct list *e = list_head(&priv->control_frames);
5498         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5499         list_remove(&control_frame->entry);
5500         HeapFree(GetProcessHeap(), 0, control_frame);
5501         priv->loop_depth--;
5502     }
5503     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5504     {
5505         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5506         struct list *e = list_head(&priv->control_frames);
5507         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5508         list_remove(&control_frame->entry);
5509         HeapFree(GetProcessHeap(), 0, control_frame);
5510     }
5511
5512
5513     shader_arb_add_instruction_modifiers(ins);
5514 }
5515
5516 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5517 {
5518     shader_arb_handle_instruction,
5519     shader_arb_select,
5520     shader_arb_select_depth_blt,
5521     shader_arb_deselect_depth_blt,
5522     shader_arb_update_float_vertex_constants,
5523     shader_arb_update_float_pixel_constants,
5524     shader_arb_load_constants,
5525     shader_arb_load_np2fixup_constants,
5526     shader_arb_destroy,
5527     shader_arb_alloc,
5528     shader_arb_free,
5529     shader_arb_context_destroyed,
5530     shader_arb_get_caps,
5531     shader_arb_color_fixup_supported,
5532 };
5533
5534 /* ARB_fragment_program fixed function pipeline replacement definitions */
5535 #define ARB_FFP_CONST_TFACTOR           0
5536 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5537 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5538 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5539 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5540
5541 struct arbfp_ffp_desc
5542 {
5543     struct ffp_frag_desc parent;
5544     GLuint shader;
5545     unsigned int num_textures_used;
5546 };
5547
5548 /* Context activation and GL locking are done by the caller. */
5549 static void arbfp_enable(BOOL enable)
5550 {
5551     if(enable) {
5552         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5553         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5554     } else {
5555         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5556         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5557     }
5558 }
5559
5560 static HRESULT arbfp_alloc(struct wined3d_device *device)
5561 {
5562     struct shader_arb_priv *priv;
5563     /* Share private data between the shader backend and the pipeline replacement, if both
5564      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5565      * if no pixel shader is bound or not
5566      */
5567     if (device->shader_backend == &arb_program_shader_backend)
5568     {
5569         device->fragment_priv = device->shader_priv;
5570     }
5571     else
5572     {
5573         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5574         if (!device->fragment_priv) return E_OUTOFMEMORY;
5575     }
5576     priv = device->fragment_priv;
5577     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5578     {
5579         ERR("Failed to initialize rbtree.\n");
5580         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5581         return E_OUTOFMEMORY;
5582     }
5583     priv->use_arbfp_fixed_func = TRUE;
5584     return WINED3D_OK;
5585 }
5586
5587 /* Context activation is done by the caller. */
5588 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5589 {
5590     const struct wined3d_gl_info *gl_info = context;
5591     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5592
5593     ENTER_GL();
5594     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5595     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5596     HeapFree(GetProcessHeap(), 0, entry_arb);
5597     LEAVE_GL();
5598 }
5599
5600 /* Context activation is done by the caller. */
5601 static void arbfp_free(struct wined3d_device *device)
5602 {
5603     struct shader_arb_priv *priv = device->fragment_priv;
5604
5605     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5606     priv->use_arbfp_fixed_func = FALSE;
5607
5608     if (device->shader_backend != &arb_program_shader_backend)
5609     {
5610         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5611     }
5612 }
5613
5614 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5615 {
5616     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5617     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5618                            WINED3DTEXOPCAPS_SELECTARG1                  |
5619                            WINED3DTEXOPCAPS_SELECTARG2                  |
5620                            WINED3DTEXOPCAPS_MODULATE4X                  |
5621                            WINED3DTEXOPCAPS_MODULATE2X                  |
5622                            WINED3DTEXOPCAPS_MODULATE                    |
5623                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5624                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5625                            WINED3DTEXOPCAPS_ADD                         |
5626                            WINED3DTEXOPCAPS_SUBTRACT                    |
5627                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5628                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5629                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5630                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5631                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5632                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5633                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5634                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5635                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5636                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5637                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5638                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5639                            WINED3DTEXOPCAPS_LERP                        |
5640                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5641                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5642
5643     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5644
5645     caps->MaxTextureBlendStages   = 8;
5646     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5647 }
5648
5649 static void state_texfactor_arbfp(struct wined3d_context *context,
5650         const struct wined3d_state *state, DWORD state_id)
5651 {
5652     struct wined3d_device *device = context->swapchain->device;
5653     const struct wined3d_gl_info *gl_info = context->gl_info;
5654     float col[4];
5655
5656     /* Don't load the parameter if we're using an arbfp pixel shader,
5657      * otherwise we'll overwrite application provided constants. */
5658     if (device->shader_backend == &arb_program_shader_backend)
5659     {
5660         struct shader_arb_priv *priv;
5661
5662         if (use_ps(state)) return;
5663
5664         priv = device->shader_priv;
5665         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5666         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5667     }
5668
5669     D3DCOLORTOGLFLOAT4(state->render_states[WINED3DRS_TEXTUREFACTOR], col);
5670     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5671     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5672 }
5673
5674 static void state_arb_specularenable(struct wined3d_context *context,
5675         const struct wined3d_state *state, DWORD state_id)
5676 {
5677     struct wined3d_device *device = context->swapchain->device;
5678     const struct wined3d_gl_info *gl_info = context->gl_info;
5679     float col[4];
5680
5681     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5682      * application provided constants
5683      */
5684     if (device->shader_backend == &arb_program_shader_backend)
5685     {
5686         struct shader_arb_priv *priv;
5687
5688         if (use_ps(state)) return;
5689
5690         priv = device->shader_priv;
5691         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5692         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5693     }
5694
5695     if (state->render_states[WINED3DRS_SPECULARENABLE])
5696     {
5697         /* The specular color has no alpha */
5698         col[0] = 1.0f; col[1] = 1.0f;
5699         col[2] = 1.0f; col[3] = 0.0f;
5700     } else {
5701         col[0] = 0.0f; col[1] = 0.0f;
5702         col[2] = 0.0f; col[3] = 0.0f;
5703     }
5704     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5705     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5706 }
5707
5708 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5709 {
5710     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5711     struct wined3d_device *device = context->swapchain->device;
5712     const struct wined3d_gl_info *gl_info = context->gl_info;
5713     float mat[2][2];
5714
5715     if (use_ps(state))
5716     {
5717         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5718         {
5719             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5720              * anyway
5721              */
5722             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5723                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5724         }
5725
5726         if(device->shader_backend == &arb_program_shader_backend) {
5727             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5728             return;
5729         }
5730     }
5731     else if (device->shader_backend == &arb_program_shader_backend)
5732     {
5733         struct shader_arb_priv *priv = device->shader_priv;
5734         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5735         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5736     }
5737
5738     mat[0][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5739     mat[0][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5740     mat[1][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5741     mat[1][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5742
5743     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5744     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5745 }
5746
5747 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5748         const struct wined3d_state *state, DWORD state_id)
5749 {
5750     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5751     struct wined3d_device *device = context->swapchain->device;
5752     const struct wined3d_gl_info *gl_info = context->gl_info;
5753     float param[4];
5754
5755     if (use_ps(state))
5756     {
5757         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5758         {
5759             /* The pixel shader has to know the luminance offset. Do a constants update if it
5760              * isn't scheduled anyway
5761              */
5762             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5763                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5764         }
5765
5766         if(device->shader_backend == &arb_program_shader_backend) {
5767             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5768             return;
5769         }
5770     }
5771     else if (device->shader_backend == &arb_program_shader_backend)
5772     {
5773         struct shader_arb_priv *priv = device->shader_priv;
5774         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5775         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5776     }
5777
5778     param[0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5779     param[1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5780     param[2] = 0.0f;
5781     param[3] = 0.0f;
5782
5783     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5784     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5785 }
5786
5787 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5788 {
5789     const char *ret;
5790
5791     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5792
5793     switch(arg & WINED3DTA_SELECTMASK) {
5794         case WINED3DTA_DIFFUSE:
5795             ret = "fragment.color.primary"; break;
5796
5797         case WINED3DTA_CURRENT:
5798             if (!stage) ret = "fragment.color.primary";
5799             else ret = "ret";
5800             break;
5801
5802         case WINED3DTA_TEXTURE:
5803             switch(stage) {
5804                 case 0: ret = "tex0"; break;
5805                 case 1: ret = "tex1"; break;
5806                 case 2: ret = "tex2"; break;
5807                 case 3: ret = "tex3"; break;
5808                 case 4: ret = "tex4"; break;
5809                 case 5: ret = "tex5"; break;
5810                 case 6: ret = "tex6"; break;
5811                 case 7: ret = "tex7"; break;
5812                 default: ret = "unknown texture";
5813             }
5814             break;
5815
5816         case WINED3DTA_TFACTOR:
5817             ret = "tfactor"; break;
5818
5819         case WINED3DTA_SPECULAR:
5820             ret = "fragment.color.secondary"; break;
5821
5822         case WINED3DTA_TEMP:
5823             ret = "tempreg"; break;
5824
5825         case WINED3DTA_CONSTANT:
5826             FIXME("Implement perstage constants\n");
5827             switch(stage) {
5828                 case 0: ret = "const0"; break;
5829                 case 1: ret = "const1"; break;
5830                 case 2: ret = "const2"; break;
5831                 case 3: ret = "const3"; break;
5832                 case 4: ret = "const4"; break;
5833                 case 5: ret = "const5"; break;
5834                 case 6: ret = "const6"; break;
5835                 case 7: ret = "const7"; break;
5836                 default: ret = "unknown constant";
5837             }
5838             break;
5839
5840         default:
5841             return "unknown";
5842     }
5843
5844     if(arg & WINED3DTA_COMPLEMENT) {
5845         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5846         if(argnum == 0) ret = "arg0";
5847         if(argnum == 1) ret = "arg1";
5848         if(argnum == 2) ret = "arg2";
5849     }
5850     if(arg & WINED3DTA_ALPHAREPLICATE) {
5851         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5852         if(argnum == 0) ret = "arg0";
5853         if(argnum == 1) ret = "arg1";
5854         if(argnum == 2) ret = "arg2";
5855     }
5856     return ret;
5857 }
5858
5859 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5860         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5861 {
5862     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5863     unsigned int mul = 1;
5864     BOOL mul_final_dest = FALSE;
5865
5866     if(color && alpha) dstmask = "";
5867     else if(color) dstmask = ".xyz";
5868     else dstmask = ".w";
5869
5870     if(dst == tempreg) dstreg = "tempreg";
5871     else dstreg = "ret";
5872
5873     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5874     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5875     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5876
5877     switch(op) {
5878         case WINED3DTOP_DISABLE:
5879             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5880             break;
5881
5882         case WINED3DTOP_SELECTARG2:
5883             arg1 = arg2;
5884             /* FALLTHROUGH */
5885         case WINED3DTOP_SELECTARG1:
5886             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5887             break;
5888
5889         case WINED3DTOP_MODULATE4X:
5890             mul = 2;
5891             /* FALLTHROUGH */
5892         case WINED3DTOP_MODULATE2X:
5893             mul *= 2;
5894             if (!strcmp(dstreg, "result.color"))
5895             {
5896                 dstreg = "ret";
5897                 mul_final_dest = TRUE;
5898             }
5899             /* FALLTHROUGH */
5900         case WINED3DTOP_MODULATE:
5901             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5902             break;
5903
5904         case WINED3DTOP_ADDSIGNED2X:
5905             mul = 2;
5906             if (!strcmp(dstreg, "result.color"))
5907             {
5908                 dstreg = "ret";
5909                 mul_final_dest = TRUE;
5910             }
5911             /* FALLTHROUGH */
5912         case WINED3DTOP_ADDSIGNED:
5913             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5914             arg2 = "arg2";
5915             /* FALLTHROUGH */
5916         case WINED3DTOP_ADD:
5917             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5918             break;
5919
5920         case WINED3DTOP_SUBTRACT:
5921             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5922             break;
5923
5924         case WINED3DTOP_ADDSMOOTH:
5925             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5926             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5927             break;
5928
5929         case WINED3DTOP_BLENDCURRENTALPHA:
5930             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5931             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5932             break;
5933         case WINED3DTOP_BLENDFACTORALPHA:
5934             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5935             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5936             break;
5937         case WINED3DTOP_BLENDTEXTUREALPHA:
5938             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5939             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5940             break;
5941         case WINED3DTOP_BLENDDIFFUSEALPHA:
5942             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5943             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5944             break;
5945
5946         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5947             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5948             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5949             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5950             break;
5951
5952         /* D3DTOP_PREMODULATE ???? */
5953
5954         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5955             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5956             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5957             break;
5958         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5959             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5960             break;
5961         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5962             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5963             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5964             break;
5965         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5966             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5967             break;
5968
5969         case WINED3DTOP_DOTPRODUCT3:
5970             mul = 4;
5971             if (!strcmp(dstreg, "result.color"))
5972             {
5973                 dstreg = "ret";
5974                 mul_final_dest = TRUE;
5975             }
5976             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5977             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5978             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5979             break;
5980
5981         case WINED3DTOP_MULTIPLYADD:
5982             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5983             break;
5984
5985         case WINED3DTOP_LERP:
5986             /* The msdn is not quite right here */
5987             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5988             break;
5989
5990         case WINED3DTOP_BUMPENVMAP:
5991         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5992             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5993             break;
5994
5995         default:
5996             FIXME("Unhandled texture op %08x\n", op);
5997     }
5998
5999     if(mul == 2) {
6000         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6001     } else if(mul == 4) {
6002         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6003     }
6004 }
6005
6006 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6007 {
6008     unsigned int stage;
6009     struct wined3d_shader_buffer buffer;
6010     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6011     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6012     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6013     UINT lowest_disabled_stage;
6014     const char *textype;
6015     const char *instr, *sat;
6016     char colorcor_dst[8];
6017     GLuint ret;
6018     DWORD arg0, arg1, arg2;
6019     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6020     BOOL op_equal;
6021     const char *final_combiner_src = "ret";
6022     GLint pos;
6023
6024     /* Find out which textures are read */
6025     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6026         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
6027         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6028         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6029         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6030         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6031         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6032         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6033
6034         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
6035         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
6036         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
6037             bump_used[stage] = TRUE;
6038             tex_read[stage] = TRUE;
6039         }
6040         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6041             bump_used[stage] = TRUE;
6042             tex_read[stage] = TRUE;
6043             luminance_used[stage] = TRUE;
6044         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
6045             tfactor_used = TRUE;
6046         }
6047
6048         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6049             tfactor_used = TRUE;
6050         }
6051
6052         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6053         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6054             tempreg_used = TRUE;
6055         }
6056
6057         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
6058         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6059         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6060         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6061         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6062         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6063         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6064
6065         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6066             tempreg_used = TRUE;
6067         }
6068         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6069             tfactor_used = TRUE;
6070         }
6071     }
6072     lowest_disabled_stage = stage;
6073
6074     /* Shader header */
6075     if (!shader_buffer_init(&buffer))
6076     {
6077         ERR("Failed to initialize shader buffer.\n");
6078         return 0;
6079     }
6080
6081     shader_addline(&buffer, "!!ARBfp1.0\n");
6082
6083     switch(settings->fog) {
6084         case FOG_OFF:                                                         break;
6085         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6086         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6087         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6088         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6089     }
6090
6091     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6092     shader_addline(&buffer, "TEMP TMP;\n");
6093     shader_addline(&buffer, "TEMP ret;\n");
6094     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6095     shader_addline(&buffer, "TEMP arg0;\n");
6096     shader_addline(&buffer, "TEMP arg1;\n");
6097     shader_addline(&buffer, "TEMP arg2;\n");
6098     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6099         if(!tex_read[stage]) continue;
6100         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6101         if(!bump_used[stage]) continue;
6102         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6103         if(!luminance_used[stage]) continue;
6104         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6105     }
6106     if(tfactor_used) {
6107         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6108     }
6109         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6110
6111     if(settings->sRGB_write) {
6112         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6113                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6114         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6115                        srgb_sub_high, 0.0, 0.0, 0.0);
6116     }
6117
6118     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6119         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6120
6121     /* Generate texture sampling instructions) */
6122     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
6123         if(!tex_read[stage]) continue;
6124
6125         switch(settings->op[stage].tex_type) {
6126             case tex_1d:                    textype = "1D";     break;
6127             case tex_2d:                    textype = "2D";     break;
6128             case tex_3d:                    textype = "3D";     break;
6129             case tex_cube:                  textype = "CUBE";   break;
6130             case tex_rect:                  textype = "RECT";   break;
6131             default: textype = "unexpected_textype";   break;
6132         }
6133
6134         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
6135            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6136             sat = "";
6137         } else {
6138             sat = "_SAT";
6139         }
6140
6141         if(settings->op[stage].projected == proj_none) {
6142             instr = "TEX";
6143         } else if(settings->op[stage].projected == proj_count4 ||
6144                   settings->op[stage].projected == proj_count3) {
6145             instr = "TXP";
6146         } else {
6147             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6148             instr = "TXP";
6149         }
6150
6151         if(stage > 0 &&
6152            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6153             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6154             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6155             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6156             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6157             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6158
6159             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6160              * so multiply the displacement with the dividing parameter before passing it to TXP
6161              */
6162             if (settings->op[stage].projected != proj_none) {
6163                 if(settings->op[stage].projected == proj_count4) {
6164                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6165                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6166                 } else {
6167                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6168                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6169                 }
6170             } else {
6171                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6172             }
6173
6174             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6175                            instr, sat, stage, stage, textype);
6176             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6177                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6178                                stage - 1, stage - 1, stage - 1);
6179                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6180             }
6181         } else if(settings->op[stage].projected == proj_count3) {
6182             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6183             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6184             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6185                             instr, sat, stage, stage, textype);
6186         } else {
6187             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6188                             instr, sat, stage, stage, stage, textype);
6189         }
6190
6191         sprintf(colorcor_dst, "tex%u", stage);
6192         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6193                 settings->op[stage].color_fixup);
6194     }
6195
6196     /* Generate the main shader */
6197     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6198     {
6199         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6200         {
6201             if (!stage) final_combiner_src = "fragment.color.primary";
6202             break;
6203         }
6204
6205         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6206            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6207             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6208         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6209                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6210             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6211         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6212                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6213             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6214         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6215                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6216             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6217         } else {
6218             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6219                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6220                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6221                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6222         }
6223
6224         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6225             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6226                           settings->op[stage].cop, settings->op[stage].carg0,
6227                           settings->op[stage].carg1, settings->op[stage].carg2);
6228             if (!stage)
6229                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6230         }
6231         else if (op_equal)
6232         {
6233             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6234                           settings->op[stage].cop, settings->op[stage].carg0,
6235                           settings->op[stage].carg1, settings->op[stage].carg2);
6236         } else {
6237             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6238                           settings->op[stage].cop, settings->op[stage].carg0,
6239                           settings->op[stage].carg1, settings->op[stage].carg2);
6240             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6241                           settings->op[stage].aop, settings->op[stage].aarg0,
6242                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6243         }
6244     }
6245
6246     if(settings->sRGB_write) {
6247         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6248         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6249         shader_addline(&buffer, "MOV result.color, ret;\n");
6250     } else {
6251         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6252     }
6253
6254     /* Footer */
6255     shader_addline(&buffer, "END\n");
6256
6257     /* Generate the shader */
6258     GL_EXTCALL(glGenProgramsARB(1, &ret));
6259     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6260     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6261             strlen(buffer.buffer), buffer.buffer));
6262     checkGLcall("glProgramStringARB()");
6263
6264     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6265     if (pos != -1)
6266     {
6267         FIXME("Fragment program error at position %d: %s\n\n", pos,
6268               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6269         shader_arb_dump_program_source(buffer.buffer);
6270     }
6271     else
6272     {
6273         GLint native;
6274
6275         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6276         checkGLcall("glGetProgramivARB()");
6277         if (!native) WARN("Program exceeds native resource limits.\n");
6278     }
6279
6280     shader_buffer_free(&buffer);
6281     return ret;
6282 }
6283
6284 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6285 {
6286     const struct wined3d_device *device = context->swapchain->device;
6287     const struct wined3d_gl_info *gl_info = context->gl_info;
6288     struct shader_arb_priv *priv = device->fragment_priv;
6289     BOOL use_vshader = use_vs(state);
6290     BOOL use_pshader = use_ps(state);
6291     struct ffp_frag_settings settings;
6292     const struct arbfp_ffp_desc *desc;
6293     unsigned int i;
6294
6295     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6296
6297     if (isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE)))
6298     {
6299         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6300         {
6301             /* Reload fixed function constants since they collide with the
6302              * pixel shader constants. */
6303             for (i = 0; i < MAX_TEXTURES; ++i)
6304             {
6305                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00));
6306             }
6307             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3DRS_TEXTUREFACTOR));
6308             state_arb_specularenable(context, state, STATE_RENDER(WINED3DRS_SPECULARENABLE));
6309         }
6310         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6311         {
6312             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6313         }
6314         return;
6315     }
6316
6317     if (!use_pshader)
6318     {
6319         /* Find or create a shader implementing the fixed function pipeline
6320          * settings, then activate it. */
6321         gen_ffp_frag_op(device, state, &settings, FALSE);
6322         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6323         if(!desc) {
6324             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6325             if (!new_desc)
6326             {
6327                 ERR("Out of memory\n");
6328                 return;
6329             }
6330             new_desc->num_textures_used = 0;
6331             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6332             {
6333                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6334                 new_desc->num_textures_used = i;
6335             }
6336
6337             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6338             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6339             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6340             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6341             desc = new_desc;
6342         }
6343
6344         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6345          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6346          * deactivate it.
6347          */
6348         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6349         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6350         priv->current_fprogram_id = desc->shader;
6351
6352         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6353         {
6354             /* Reload fixed function constants since they collide with the
6355              * pixel shader constants. */
6356             for (i = 0; i < MAX_TEXTURES; ++i)
6357             {
6358                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00));
6359             }
6360             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3DRS_TEXTUREFACTOR));
6361             state_arb_specularenable(context, state, STATE_RENDER(WINED3DRS_SPECULARENABLE));
6362         }
6363         context->last_was_pshader = FALSE;
6364     } else {
6365         context->last_was_pshader = TRUE;
6366     }
6367
6368     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6369      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6370      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6371      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6372      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6373      *
6374      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6375      * shader handler
6376      */
6377     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6378     {
6379         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6380
6381         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6382             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6383     }
6384     if (use_pshader)
6385         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6386 }
6387
6388 /* We can't link the fog states to the fragment state directly since the
6389  * vertex pipeline links them to FOGENABLE. A different linking in different
6390  * pipeline parts can't be expressed in the combined state table, so we need
6391  * to handle that with a forwarding function. The other invisible side effect
6392  * is that changing the fog start and fog end (which links to FOGENABLE in
6393  * vertex) results in the fragment_prog_arbfp function being called because
6394  * FOGENABLE is dirty, which calls this function here. */
6395 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6396 {
6397     enum fogsource new_source;
6398
6399     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6400
6401     if (!isStateDirty(context, STATE_PIXELSHADER))
6402         fragment_prog_arbfp(context, state, state_id);
6403
6404     if (!state->render_states[WINED3DRS_FOGENABLE])
6405         return;
6406
6407     if (state->render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6408     {
6409         if (use_vs(state))
6410         {
6411             new_source = FOGSOURCE_VS;
6412         }
6413         else
6414         {
6415             if (state->render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6416                 new_source = FOGSOURCE_COORD;
6417             else
6418                 new_source = FOGSOURCE_FFP;
6419         }
6420     }
6421     else
6422     {
6423         new_source = FOGSOURCE_FFP;
6424     }
6425
6426     if (new_source != context->fog_source)
6427     {
6428         context->fog_source = new_source;
6429         state_fogstartend(context, state, STATE_RENDER(WINED3DRS_FOGSTART));
6430     }
6431 }
6432
6433 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6434 {
6435     if (!isStateDirty(context, STATE_PIXELSHADER))
6436         fragment_prog_arbfp(context, state, state_id);
6437 }
6438
6439 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6440     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6441     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6442     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6443     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6444     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6445     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6446     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6447     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6448     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6449     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6450     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6451     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6452     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6453     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6454     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6455     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6456     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6457     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6458     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6459     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6460     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6461     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6462     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6463     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6464     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6465     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6466     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6467     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6468     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6469     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6470     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6471     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6472     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6473     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6474     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6475     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6476     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6477     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6478     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6479     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6480     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6481     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6482     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6483     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6484     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6485     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6486     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6487     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6488     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6489     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6490     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6491     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6492     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6493     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6494     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6495     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6496     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6497     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6498     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6499     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6502     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6503     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6505     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6562     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6563     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6566     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6569     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6578     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6579     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6580 };
6581
6582 const struct fragment_pipeline arbfp_fragment_pipeline = {
6583     arbfp_enable,
6584     arbfp_get_caps,
6585     arbfp_alloc,
6586     arbfp_free,
6587     shader_arb_color_fixup_supported,
6588     arbfp_fragmentstate_template,
6589     TRUE /* We can disable projected textures */
6590 };
6591
6592 struct arbfp_blit_priv {
6593     GLenum yuy2_rect_shader, yuy2_2d_shader;
6594     GLenum uyvy_rect_shader, uyvy_2d_shader;
6595     GLenum yv12_rect_shader, yv12_2d_shader;
6596     GLenum p8_rect_shader, p8_2d_shader;
6597     GLuint palette_texture;
6598 };
6599
6600 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6601 {
6602     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6603     if(!device->blit_priv) {
6604         ERR("Out of memory\n");
6605         return E_OUTOFMEMORY;
6606     }
6607     return WINED3D_OK;
6608 }
6609
6610 /* Context activation is done by the caller. */
6611 static void arbfp_blit_free(struct wined3d_device *device)
6612 {
6613     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6614     struct arbfp_blit_priv *priv = device->blit_priv;
6615
6616     ENTER_GL();
6617     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6618     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6619     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6620     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6621     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6622     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6623     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6624     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6625     checkGLcall("Delete yuv and p8 programs");
6626
6627     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6628     LEAVE_GL();
6629
6630     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6631     device->blit_priv = NULL;
6632 }
6633
6634 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6635         GLenum textype, char *luminance)
6636 {
6637     char chroma;
6638     const char *tex, *texinstr;
6639
6640     if (fixup == COMPLEX_FIXUP_UYVY) {
6641         chroma = 'x';
6642         *luminance = 'w';
6643     } else {
6644         chroma = 'w';
6645         *luminance = 'x';
6646     }
6647     switch(textype) {
6648         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6649         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6650         default:
6651             /* This is more tricky than just replacing the texture type - we have to navigate
6652              * properly in the texture to find the correct chroma values
6653              */
6654             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6655             return FALSE;
6656     }
6657
6658     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6659      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6660      * filtering when we sample the texture.
6661      *
6662      * These are the rules for reading the chroma:
6663      *
6664      * Even pixel: Cr
6665      * Even pixel: U
6666      * Odd pixel: V
6667      *
6668      * So we have to get the sampling x position in non-normalized coordinates in integers
6669      */
6670     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6671         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6672         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6673     } else {
6674         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6675     }
6676     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6677      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6678      * 0.5, so add 0.5.
6679      */
6680     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6681     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6682
6683     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6684      * even and odd pixels respectively
6685      */
6686     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6687     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6688
6689     /* Sample Pixel 1 */
6690     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6691
6692     /* Put the value into either of the chroma values */
6693     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6694     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6695     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6696     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6697
6698     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6699      * the pixel right to the current one. Otherwise, sample the left pixel.
6700      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6701      */
6702     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6703     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6704     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6705
6706     /* Put the value into the other chroma */
6707     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6708     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6709     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6710     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6711
6712     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6713      * the current one and lerp the two U and V values
6714      */
6715
6716     /* This gives the correctly filtered luminance value */
6717     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6718
6719     return TRUE;
6720 }
6721
6722 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6723 {
6724     const char *tex;
6725
6726     switch(textype) {
6727         case GL_TEXTURE_2D:             tex = "2D";     break;
6728         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6729         default:
6730             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6731             return FALSE;
6732     }
6733
6734     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6735      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6736      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6737      * pitch of the luminance plane, the packing into the gl texture is a bit
6738      * unfortunate. If the whole texture is interpreted as luminance data it looks
6739      * approximately like this:
6740      *
6741      *        +----------------------------------+----
6742      *        |                                  |
6743      *        |                                  |
6744      *        |                                  |
6745      *        |                                  |
6746      *        |                                  |   2
6747      *        |            LUMINANCE             |   -
6748      *        |                                  |   3
6749      *        |                                  |
6750      *        |                                  |
6751      *        |                                  |
6752      *        |                                  |
6753      *        +----------------+-----------------+----
6754      *        |                |                 |
6755      *        |  U even rows   |  U odd rows     |
6756      *        |                |                 |   1
6757      *        +----------------+------------------   -
6758      *        |                |                 |   3
6759      *        |  V even rows   |  V odd rows     |
6760      *        |                |                 |
6761      *        +----------------+-----------------+----
6762      *        |                |                 |
6763      *        |     0.5        |       0.5       |
6764      *
6765      * So it appears as if there are 4 chroma images, but in fact the odd rows
6766      * in the chroma images are in the same row as the even ones. So its is
6767      * kinda tricky to read
6768      *
6769      * When reading from rectangle textures, keep in mind that the input y coordinates
6770      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6771      */
6772     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6773             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6774
6775     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6776     /* the chroma planes have only half the width */
6777     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6778
6779     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6780      * the coordinate. Also read the right side of the image when reading odd lines
6781      *
6782      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6783      * bleeding
6784      */
6785     if(textype == GL_TEXTURE_2D) {
6786
6787         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6788
6789         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6790
6791         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6792         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6793
6794         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6795         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6796         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6797         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6798         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6799
6800         /* clamp, keep the half pixel origin in mind */
6801         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6802         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6803         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6804         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6805     } else {
6806         /* Read from [size - size+size/4] */
6807         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6808         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6809
6810         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6811         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6812         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6813         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6814         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6815         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6816
6817         /* Make sure to read exactly from the pixel center */
6818         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6819         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6820
6821         /* Clamp */
6822         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6823         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6824         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6825         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6826         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6827     }
6828     /* Read the texture, put the result into the output register */
6829     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6830     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6831
6832     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6833      * No need to clamp because we're just reusing the already clamped value from above
6834      */
6835     if(textype == GL_TEXTURE_2D) {
6836         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6837     } else {
6838         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6839     }
6840     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6841     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6842
6843     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6844      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6845      * values due to filtering
6846      */
6847     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6848     if(textype == GL_TEXTURE_2D) {
6849         /* Multiply the y coordinate by 2/3 and clamp it */
6850         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6851         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6852         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6853         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6854     } else {
6855         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6856          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6857          * is bigger
6858          */
6859         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6860         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6861         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6862     }
6863     *luminance = 'a';
6864
6865     return TRUE;
6866 }
6867
6868 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6869         const struct wined3d_gl_info *gl_info, GLenum textype)
6870 {
6871     GLenum shader;
6872     struct wined3d_shader_buffer buffer;
6873     GLint pos;
6874
6875     /* Shader header */
6876     if (!shader_buffer_init(&buffer))
6877     {
6878         ERR("Failed to initialize shader buffer.\n");
6879         return 0;
6880     }
6881
6882     ENTER_GL();
6883     GL_EXTCALL(glGenProgramsARB(1, &shader));
6884     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6885     LEAVE_GL();
6886     if(!shader) {
6887         shader_buffer_free(&buffer);
6888         return 0;
6889     }
6890
6891     shader_addline(&buffer, "!!ARBfp1.0\n");
6892     shader_addline(&buffer, "TEMP index;\n");
6893
6894     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6895     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6896
6897     /* The alpha-component contains the palette index */
6898     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6899         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6900     else
6901         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6902
6903     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6904     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6905
6906     /* Use the alpha-component as an index in the palette to get the final color */
6907     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6908     shader_addline(&buffer, "END\n");
6909
6910     ENTER_GL();
6911     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6912             strlen(buffer.buffer), buffer.buffer));
6913     checkGLcall("glProgramStringARB()");
6914
6915     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6916     if (pos != -1)
6917     {
6918         FIXME("Fragment program error at position %d: %s\n\n", pos,
6919               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6920         shader_arb_dump_program_source(buffer.buffer);
6921     }
6922
6923     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6924         priv->p8_rect_shader = shader;
6925     else
6926         priv->p8_2d_shader = shader;
6927
6928     shader_buffer_free(&buffer);
6929     LEAVE_GL();
6930
6931     return shader;
6932 }
6933
6934 /* Context activation is done by the caller. */
6935 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
6936 {
6937     BYTE table[256][4];
6938     struct wined3d_device *device = surface->resource.device;
6939     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6940     struct arbfp_blit_priv *priv = device->blit_priv;
6941     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6942
6943     d3dfmt_p8_init_palette(surface, table, colorkey);
6944
6945     ENTER_GL();
6946     if (!priv->palette_texture)
6947         glGenTextures(1, &priv->palette_texture);
6948
6949     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6950     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6951
6952     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6953
6954     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6955     /* Make sure we have discrete color levels. */
6956     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6957     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6958     /* Upload the palette */
6959     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6960     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6961
6962     /* Switch back to unit 0 in which the 2D texture will be stored. */
6963     context_active_texture(context, gl_info, 0);
6964     LEAVE_GL();
6965 }
6966
6967 /* Context activation is done by the caller. */
6968 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
6969         enum complex_fixup yuv_fixup, GLenum textype)
6970 {
6971     GLenum shader;
6972     struct wined3d_shader_buffer buffer;
6973     char luminance_component;
6974     GLint pos;
6975
6976     /* Shader header */
6977     if (!shader_buffer_init(&buffer))
6978     {
6979         ERR("Failed to initialize shader buffer.\n");
6980         return 0;
6981     }
6982
6983     ENTER_GL();
6984     GL_EXTCALL(glGenProgramsARB(1, &shader));
6985     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6986     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6987     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6988     LEAVE_GL();
6989     if(!shader) {
6990         shader_buffer_free(&buffer);
6991         return 0;
6992     }
6993
6994     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6995      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6996      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6997      * each single pixel it contains, and one U and one V value shared between both
6998      * pixels.
6999      *
7000      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7001      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7002      * take the format into account when generating the read swizzles
7003      *
7004      * Reading the Y value is straightforward - just sample the texture. The hardware
7005      * takes care of filtering in the horizontal and vertical direction.
7006      *
7007      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7008      * because that would mix the U and V values of one pixel or two adjacent pixels.
7009      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7010      * regardless of the filtering setting. Vertical filtering works automatically
7011      * though - the U and V values of two rows are mixed nicely.
7012      *
7013      * Appart of avoiding filtering issues, the code has to know which value it just
7014      * read, and where it can find the other one. To determine this, it checks if
7015      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7016      *
7017      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7018      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7019      *
7020      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7021      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7022      * in an unfiltered situation. Finding the luminance on the other hand requires
7023      * finding out if it is an odd or even pixel. The real drawback of this approach
7024      * is filtering. This would have to be emulated completely in the shader, reading
7025      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7026      * vertically. Beyond that it would require adjustments to the texture handling
7027      * code to deal with the width scaling
7028      */
7029     shader_addline(&buffer, "!!ARBfp1.0\n");
7030     shader_addline(&buffer, "TEMP luminance;\n");
7031     shader_addline(&buffer, "TEMP temp;\n");
7032     shader_addline(&buffer, "TEMP chroma;\n");
7033     shader_addline(&buffer, "TEMP texcrd;\n");
7034     shader_addline(&buffer, "TEMP texcrd2;\n");
7035     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7036     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7037     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7038
7039     switch (yuv_fixup)
7040     {
7041         case COMPLEX_FIXUP_UYVY:
7042         case COMPLEX_FIXUP_YUY2:
7043             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7044             {
7045                 shader_buffer_free(&buffer);
7046                 return 0;
7047             }
7048             break;
7049
7050         case COMPLEX_FIXUP_YV12:
7051             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7052             {
7053                 shader_buffer_free(&buffer);
7054                 return 0;
7055             }
7056             break;
7057
7058         default:
7059             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7060             shader_buffer_free(&buffer);
7061             return 0;
7062     }
7063
7064     /* Calculate the final result. Formula is taken from
7065      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7066      * ranges from -0.5 to 0.5
7067      */
7068     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7069
7070     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7071     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7072     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7073     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7074     shader_addline(&buffer, "END\n");
7075
7076     ENTER_GL();
7077     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7078             strlen(buffer.buffer), buffer.buffer));
7079     checkGLcall("glProgramStringARB()");
7080
7081     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7082     if (pos != -1)
7083     {
7084         FIXME("Fragment program error at position %d: %s\n\n", pos,
7085               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7086         shader_arb_dump_program_source(buffer.buffer);
7087     }
7088     else
7089     {
7090         GLint native;
7091
7092         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7093         checkGLcall("glGetProgramivARB()");
7094         if (!native) WARN("Program exceeds native resource limits.\n");
7095     }
7096
7097     shader_buffer_free(&buffer);
7098     LEAVE_GL();
7099
7100     switch (yuv_fixup)
7101     {
7102         case COMPLEX_FIXUP_YUY2:
7103             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7104             else priv->yuy2_2d_shader = shader;
7105             break;
7106
7107         case COMPLEX_FIXUP_UYVY:
7108             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7109             else priv->uyvy_2d_shader = shader;
7110             break;
7111
7112         case COMPLEX_FIXUP_YV12:
7113             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7114             else priv->yv12_2d_shader = shader;
7115             break;
7116         default:
7117             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7118     }
7119
7120     return shader;
7121 }
7122
7123 /* Context activation is done by the caller. */
7124 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7125 {
7126     GLenum shader;
7127     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7128     struct arbfp_blit_priv *priv = blit_priv;
7129     enum complex_fixup fixup;
7130     GLenum textype = surface->texture_target;
7131     const struct wined3d_gl_info *gl_info = context->gl_info;
7132
7133     if (!is_complex_fixup(surface->resource.format->color_fixup))
7134     {
7135         TRACE("Fixup:\n");
7136         dump_color_fixup_desc(surface->resource.format->color_fixup);
7137         /* Don't bother setting up a shader for unconverted formats */
7138         ENTER_GL();
7139         glEnable(textype);
7140         checkGLcall("glEnable(textype)");
7141         LEAVE_GL();
7142         return WINED3D_OK;
7143     }
7144
7145     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7146
7147     switch(fixup)
7148     {
7149         case COMPLEX_FIXUP_YUY2:
7150             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7151             break;
7152
7153         case COMPLEX_FIXUP_UYVY:
7154             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7155             break;
7156
7157         case COMPLEX_FIXUP_YV12:
7158             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7159             break;
7160
7161         case COMPLEX_FIXUP_P8:
7162             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7163             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7164
7165             upload_palette(surface, context);
7166             break;
7167
7168         default:
7169             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7170             ENTER_GL();
7171             glEnable(textype);
7172             checkGLcall("glEnable(textype)");
7173             LEAVE_GL();
7174             return E_NOTIMPL;
7175     }
7176
7177     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7178
7179     ENTER_GL();
7180     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7181     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7182     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7183     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7184     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7185     checkGLcall("glProgramLocalParameter4fvARB");
7186     LEAVE_GL();
7187
7188     return WINED3D_OK;
7189 }
7190
7191 /* Context activation is done by the caller. */
7192 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7193 {
7194     ENTER_GL();
7195     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7196     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7197     glDisable(GL_TEXTURE_2D);
7198     checkGLcall("glDisable(GL_TEXTURE_2D)");
7199     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7200     {
7201         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7202         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7203     }
7204     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7205     {
7206         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7207         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7208     }
7209     LEAVE_GL();
7210 }
7211
7212 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7213         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7214         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7215 {
7216     enum complex_fixup src_fixup;
7217
7218     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7219         return FALSE;
7220
7221     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7222     {
7223         TRACE("Unsupported blit_op=%d\n", blit_op);
7224         return FALSE;
7225     }
7226
7227     if (src_pool == WINED3DPOOL_SYSTEMMEM || dst_pool == WINED3DPOOL_SYSTEMMEM)
7228         return FALSE;
7229
7230     src_fixup = get_complex_fixup(src_format->color_fixup);
7231     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7232     {
7233         TRACE("Checking support for fixup:\n");
7234         dump_color_fixup_desc(src_format->color_fixup);
7235     }
7236
7237     if (!is_identity_fixup(dst_format->color_fixup))
7238     {
7239         TRACE("Destination fixups are not supported\n");
7240         return FALSE;
7241     }
7242
7243     if (is_identity_fixup(src_format->color_fixup))
7244     {
7245         TRACE("[OK]\n");
7246         return TRUE;
7247     }
7248
7249      /* We only support YUV conversions. */
7250     if (!is_complex_fixup(src_format->color_fixup))
7251     {
7252         TRACE("[FAILED]\n");
7253         return FALSE;
7254     }
7255
7256     switch(src_fixup)
7257     {
7258         case COMPLEX_FIXUP_YUY2:
7259         case COMPLEX_FIXUP_UYVY:
7260         case COMPLEX_FIXUP_YV12:
7261         case COMPLEX_FIXUP_P8:
7262             TRACE("[OK]\n");
7263             return TRUE;
7264
7265         default:
7266             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7267             TRACE("[FAILED]\n");
7268             return FALSE;
7269     }
7270 }
7271
7272 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7273         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7274         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7275 {
7276     struct wined3d_context *context;
7277     RECT src_rect = *src_rect_in;
7278     RECT dst_rect = *dst_rect_in;
7279
7280     /* Now load the surface */
7281     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7282             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7283     {
7284         /* Without FBO blits transfering from the drawable to the texture is
7285          * expensive, because we have to flip the data in sysmem. Since we can
7286          * flip in the blitter, we don't actually need that flip anyway. So we
7287          * use the surface's texture as scratch texture, and flip the source
7288          * rectangle instead. */
7289         surface_load_fb_texture(src_surface, FALSE);
7290
7291         src_rect.top = src_surface->resource.height - src_rect.top;
7292         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7293     }
7294     else
7295         surface_internal_preload(src_surface, SRGB_RGB);
7296
7297     /* Activate the destination context, set it up for blitting */
7298     context = context_acquire(device, dst_surface);
7299     context_apply_blit_state(context, device);
7300
7301     if (!surface_is_offscreen(dst_surface))
7302         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7303
7304     arbfp_blit_set(device->blit_priv, context, src_surface);
7305
7306     ENTER_GL();
7307
7308     /* Draw a textured quad */
7309     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7310
7311     LEAVE_GL();
7312
7313     /* Leave the opengl state valid for blitting */
7314     arbfp_blit_unset(context->gl_info);
7315
7316     if (wined3d_settings.strict_draw_ordering
7317             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7318             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7319         wglFlush(); /* Flush to ensure ordering across contexts. */
7320
7321     context_release(context);
7322
7323     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7324     return WINED3D_OK;
7325 }
7326
7327 /* Do not call while under the GL lock. */
7328 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7329         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7330 {
7331     FIXME("Color filling not implemented by arbfp_blit\n");
7332     return WINED3DERR_INVALIDCALL;
7333 }
7334
7335 /* Do not call while under the GL lock. */
7336 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7337         struct wined3d_surface *surface, const RECT *rect, float depth)
7338 {
7339     FIXME("Depth filling not implemented by arbfp_blit.\n");
7340     return WINED3DERR_INVALIDCALL;
7341 }
7342
7343 const struct blit_shader arbfp_blit = {
7344     arbfp_blit_alloc,
7345     arbfp_blit_free,
7346     arbfp_blit_set,
7347     arbfp_blit_unset,
7348     arbfp_blit_supported,
7349     arbfp_blit_color_fill,
7350     arbfp_blit_depth_fill,
7351 };