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     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
 348     return FALSE;
 349 }
 350
 351 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
 352         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
 353 {
 354     unsigned int ret = 1;
 355     /* We use one PARAM for the pos fixup, and in some cases one to load
 356      * some immediate values into the shader. */
 357     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
 358     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
 359     return ret;
 360 }
 361
 362 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 363  * When constant_list == NULL, it will load all the constants.
 364  *
 365  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 366  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 367  */
 368 /* GL locking is done by the caller */
 369 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
 370         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
 371         const float *constants, char *dirty_consts)
 372 {
 373     struct wined3d_shader_lconst *lconst;
 374     DWORD i, j;
 375     unsigned int ret;
 376
 377     if (TRACE_ON(d3d_constants))
 378     {
 379         for(i = 0; i < max_constants; i++) {
 380             if(!dirty_consts[i]) continue;
 381             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
 382                         constants[i * 4 + 0], constants[i * 4 + 1],
 383                         constants[i * 4 + 2], constants[i * 4 + 3]);
 384         }
 385     }
 386
 387     i = 0;
 388
 389     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
 390     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
 391     {
 392         float lcl_const[4];
 393         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
 394          * shaders, the first 8 constants are marked dirty for reload
 395          */
 396         for(; i < min(8, max_constants); i++) {
 397             if(!dirty_consts[i]) continue;
 398             dirty_consts[i] = 0;
 399
 400             j = 4 * i;
 401             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
 402             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
 403             else lcl_const[0] = constants[j + 0];
 404
 405             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
 406             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
 407             else lcl_const[1] = constants[j + 1];
 408
 409             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
 410             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
 411             else lcl_const[2] = constants[j + 2];
 412
 413             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
 414             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
 415             else lcl_const[3] = constants[j + 3];
 416
 417             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
 418         }
 419
 420         /* If further constants are dirty, reload them without clamping.
 421          *
 422          * The alternative is not to touch them, but then we cannot reset the dirty constant count
 423          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
 424          * above would always re-check the first 8 constants since max_constant remains at the init
 425          * value
 426          */
 427     }
 428
 429     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
 430     {
 431         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
 432          * or just reloading *all* constants at once
 433          *
 434         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
 435          */
 436         for(; i < max_constants; i++) {
 437             if(!dirty_consts[i]) continue;
 438
 439             /* Find the next block of dirty constants */
 440             dirty_consts[i] = 0;
 441             j = i;
 442             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
 443                 dirty_consts[i] = 0;
 444             }
 445
 446             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
 447         }
 448     } else {
 449         for(; i < max_constants; i++) {
 450             if(dirty_consts[i]) {
 451                 dirty_consts[i] = 0;
 452                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
 453             }
 454         }
 455     }
 456     checkGLcall("glProgramEnvParameter4fvARB()");
 457
 458     /* Load immediate constants */
 459     if (shader->load_local_constsF)
 460     {
 461         if (TRACE_ON(d3d_shader))
 462         {
 463             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 464             {
 465                 GLfloat* values = (GLfloat*)lconst->value;
 466                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
 467                         values[0], values[1], values[2], values[3]);
 468             }
 469         }
 470         /* Immediate constants are clamped for 1.X shaders at loading times */
 471         ret = 0;
 472         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 473         {
 474             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
 475             ret = max(ret, lconst->idx + 1);
 476             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
 477         }
 478         checkGLcall("glProgramEnvParameter4fvARB()");
 479         return ret; /* The loaded immediate constants need reloading for the next shader */
 480     } else {
 481         return 0; /* No constants are dirty now */
 482     }
 483 }
 484
 485 /**
 486  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
 487  */
 488 static void shader_arb_load_np2fixup_constants(void *shader_priv,
 489         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
 490 {
 491     const struct shader_arb_priv * priv = shader_priv;
 492
 493     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
 494     if (!use_ps(state)) return;
 495
 496     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
 497         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
 498         UINT i;
 499         WORD active = fixup->super.active;
 500         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
 501
 502         for (i = 0; active; active >>= 1, ++i)
 503         {
 504             const struct wined3d_texture *tex = state->textures[i];
 505             const unsigned char idx = fixup->super.idx[i];
 506             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 507
 508             if (!(active & 1)) continue;
 509
 510             if (!tex) {
 511                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
 512                 continue;
 513             }
 514
 515             if (idx % 2)
 516             {
 517                 tex_dim[2] = tex->pow2_matrix[0];
 518                 tex_dim[3] = tex->pow2_matrix[5];
 519             }
 520             else
 521             {
 522                 tex_dim[0] = tex->pow2_matrix[0];
 523                 tex_dim[1] = tex->pow2_matrix[5];
 524             }
 525         }
 526
 527         for (i = 0; i < fixup->super.num_consts; ++i) {
 528             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 529                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
 530         }
 531     }
 532 }
 533
 534 /* GL locking is done by the caller. */
 535 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
 536         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
 537 {
 538     const struct wined3d_gl_info *gl_info = context->gl_info;
 539     unsigned char i;
 540
 541     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
 542     {
 543         int texunit = gl_shader->bumpenvmatconst[i].texunit;
 544
 545         /* The state manager takes care that this function is always called if the bump env matrix changes */
 546         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
 547         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 548                 gl_shader->bumpenvmatconst[i].const_num, data));
 549
 550         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
 551         {
 552             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
 553              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
 554              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
 555              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
 556             */
 557             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
 558             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
 559                     gl_shader->luminanceconst[i].const_num, scale));
 560         }
 561     }
 562     checkGLcall("Load bumpmap consts");
 563
 564     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
 565     {
 566         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
 567         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
 568         * ycorrection.z: 1.0
 569         * ycorrection.w: 0.0
 570         */
 571         float val[4];
 572         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
 573         val[1] = context->render_offscreen ? 1.0f : -1.0f;
 574         val[2] = 1.0f;
 575         val[3] = 0.0f;
 576         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
 577         checkGLcall("y correction loading");
 578     }
 579
 580     if (!gl_shader->num_int_consts) return;
 581
 582     for(i = 0; i < MAX_CONST_I; i++)
 583     {
 584         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 585         {
 586             float val[4];
 587             val[0] = (float)state->ps_consts_i[4 * i];
 588             val[1] = (float)state->ps_consts_i[4 * i + 1];
 589             val[2] = (float)state->ps_consts_i[4 * i + 2];
 590             val[3] = -1.0f;
 591
 592             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
 593         }
 594     }
 595     checkGLcall("Load ps int consts");
 596 }
 597
 598 /* GL locking is done by the caller. */
 599 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
 600         const struct wined3d_context *context, const struct wined3d_state *state)
 601 {
 602     const struct wined3d_gl_info *gl_info = context->gl_info;
 603     float position_fixup[4];
 604     unsigned char i;
 605
 606     /* Upload the position fixup */
 607     shader_get_position_fixup(context, state, position_fixup);
 608     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 609
 610     if (!gl_shader->num_int_consts) return;
 611
 612     for(i = 0; i < MAX_CONST_I; i++)
 613     {
 614         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
 615         {
 616             float val[4];
 617             val[0] = (float)state->vs_consts_i[4 * i];
 618             val[1] = (float)state->vs_consts_i[4 * i + 1];
 619             val[2] = (float)state->vs_consts_i[4 * i + 2];
 620             val[3] = -1.0f;
 621
 622             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
 623         }
 624     }
 625     checkGLcall("Load vs int consts");
 626 }
 627
 628 /**
 629  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
 630  *
 631  * We only support float constants in ARB at the moment, so don't
 632  * worry about the Integers or Booleans
 633  */
 634 /* GL locking is done by the caller (state handler) */
 635 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
 636 {
 637     struct wined3d_device *device = context->swapchain->device;
 638     const struct wined3d_state *state = &device->stateBlock->state;
 639     const struct wined3d_gl_info *gl_info = context->gl_info;
 640     struct shader_arb_priv *priv = device->shader_priv;
 641
 642     if (context != priv->last_context)
 643     {
 644         memset(priv->vshader_const_dirty, 1,
 645                 sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
 646         priv->highest_dirty_vs_const = device->d3d_vshader_constantF;
 647
 648         memset(priv->pshader_const_dirty, 1,
 649                 sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
 650         priv->highest_dirty_ps_const = device->d3d_pshader_constantF;
 651
 652         priv->last_context = context;
 653     }
 654
 655     if (useVertexShader)
 656     {
 657         struct wined3d_shader *vshader = state->vertex_shader;
 658         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 659
 660         /* Load DirectX 9 float constants for vertex shader */
 661         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
 662                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
 663         shader_arb_vs_local_constants(gl_shader, context, state);
 664     }
 665
 666     if (usePixelShader)
 667     {
 668         struct wined3d_shader *pshader = state->pixel_shader;
 669         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
 670         UINT rt_height = state->fb->render_targets[0]->resource.height;
 671
 672         /* Load DirectX 9 float constants for pixel shader */
 673         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
 674                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
 675         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 676     }
 677 }
 678
 679 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
 680 {
 681     struct wined3d_context *context = context_get_current();
 682     struct shader_arb_priv *priv = device->shader_priv;
 683
 684     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 685      * context. On a context switch the old context will be fully dirtified */
 686     if (!context || context->swapchain->device != device) return;
 687
 688     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
 689     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
 690 }
 691
 692 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
 693 {
 694     struct wined3d_context *context = context_get_current();
 695     struct shader_arb_priv *priv = device->shader_priv;
 696
 697     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
 698      * context. On a context switch the old context will be fully dirtified */
 699     if (!context || context->swapchain->device != device) return;
 700
 701     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
 702     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
 703 }
 704
 705 static DWORD *local_const_mapping(const struct wined3d_shader *shader)
 706 {
 707     const struct wined3d_shader_lconst *lconst;
 708     DWORD *ret;
 709     DWORD idx = 0;
 710
 711     if (shader->load_local_constsF || list_empty(&shader->constantsF))
 712         return NULL;
 713
 714     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * shader->limits.constant_float);
 715     if (!ret)
 716     {
 717         ERR("Out of memory\n");
 718         return NULL;
 719     }
 720
 721     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 722     {
 723         ret[lconst->idx] = idx++;
 724     }
 725     return ret;
 726 }
 727
 728 /* Generate the variable & register declarations for the ARB_vertex_program output target */
 729 static DWORD shader_generate_arb_declarations(const struct wined3d_shader *shader,
 730         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
 731         const struct wined3d_gl_info *gl_info, const DWORD *lconst_map,
 732         DWORD *num_clipplanes, const struct shader_arb_ctx_priv *ctx)
 733 {
 734     DWORD i, next_local = 0;
 735     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 736     const struct wined3d_shader_lconst *lconst;
 737     unsigned max_constantsF;
 738     DWORD map;
 739
 740     /* In pixel shaders, all private constants are program local, we don't need anything
 741      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
 742      * If we need a private constant the GL implementation will squeeze it in somewhere
 743      *
 744      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
 745      * immediate values. The posFixup is loaded using program.env for now, so always
 746      * subtract one from the number of constants. If the shader uses indirect addressing,
 747      * account for the helper const too because we have to declare all available d3d constants
 748      * and don't know which are actually used.
 749      */
 750     if (pshader)
 751     {
 752         max_constantsF = gl_info->limits.arb_ps_native_constants;
 753         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
 754         if (max_constantsF < 24)
 755             max_constantsF = gl_info->limits.arb_ps_float_constants;
 756     }
 757     else
 758     {
 759         const struct arb_vshader_private *shader_data = shader->backend_data;
 760         max_constantsF = gl_info->limits.arb_vs_native_constants;
 761         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
 762          * Also prevents max_constantsF from becoming less than 0 and
 763          * wrapping . */
 764         if (max_constantsF < 96)
 765             max_constantsF = gl_info->limits.arb_vs_float_constants;
 766
 767         if (reg_maps->usesrelconstF)
 768         {
 769             DWORD highest_constf = 0, clip_limit;
 770
 771             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
 772             max_constantsF -= count_bits(reg_maps->integer_constants);
 773
 774             for (i = 0; i < shader->limits.constant_float; ++i)
 775             {
 776                 DWORD idx = i >> 5;
 777                 DWORD shift = i & 0x1f;
 778                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
 779             }
 780
 781             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
 782             {
 783                 if(ctx->cur_vs_args->super.clip_enabled)
 784                     clip_limit = gl_info->limits.clipplanes;
 785                 else
 786                     clip_limit = 0;
 787             }
 788             else
 789             {
 790                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
 791                 clip_limit = min(count_bits(mask), 4);
 792             }
 793             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
 794             max_constantsF -= *num_clipplanes;
 795             if(*num_clipplanes < clip_limit)
 796             {
 797                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
 798             }
 799         }
 800         else
 801         {
 802             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
 803             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
 804         }
 805     }
 806
 807     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
 808     {
 809         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
 810     }
 811
 812     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
 813     {
 814         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
 815     }
 816
 817     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
 818     {
 819         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
 820         {
 821             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
 822         }
 823     }
 824
 825     /* Load local constants using the program-local space,
 826      * this avoids reloading them each time the shader is used
 827      */
 828     if (lconst_map)
 829     {
 830         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
 831         {
 832             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
 833                            lconst_map[lconst->idx]);
 834             next_local = max(next_local, lconst_map[lconst->idx] + 1);
 835         }
 836     }
 837
 838     /* After subtracting privately used constants from the hardware limit(they are loaded as
 839      * local constants), make sure the shader doesn't violate the env constant limit
 840      */
 841     if(pshader)
 842     {
 843         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
 844     }
 845     else
 846     {
 847         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
 848     }
 849
 850     /* Avoid declaring more constants than needed */
 851     max_constantsF = min(max_constantsF, shader->limits.constant_float);
 852
 853     /* we use the array-based constants array if the local constants are marked for loading,
 854      * because then we use indirect addressing, or when the local constant list is empty,
 855      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
 856      * local constants do not declare the loaded constants as an array because ARB compilers usually
 857      * do not optimize unused constants away
 858      */
 859     if (reg_maps->usesrelconstF)
 860     {
 861         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
 862         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
 863                     max_constantsF, max_constantsF - 1);
 864     } else {
 865         for(i = 0; i < max_constantsF; i++) {
 866             DWORD idx, mask;
 867             idx = i >> 5;
 868             mask = 1 << (i & 0x1f);
 869             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
 870             {
 871                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
 872             }
 873         }
 874     }
 875
 876     return next_local;
 877 }
 878
 879 static const char * const shift_tab[] = {
 880     "dummy",     /*  0 (none) */
 881     "coefmul.x", /*  1 (x2)   */
 882     "coefmul.y", /*  2 (x4)   */
 883     "coefmul.z", /*  3 (x8)   */
 884     "coefmul.w", /*  4 (x16)  */
 885     "dummy",     /*  5 (x32)  */
 886     "dummy",     /*  6 (x64)  */
 887     "dummy",     /*  7 (x128) */
 888     "dummy",     /*  8 (d256) */
 889     "dummy",     /*  9 (d128) */
 890     "dummy",     /* 10 (d64)  */
 891     "dummy",     /* 11 (d32)  */
 892     "coefdiv.w", /* 12 (d16)  */
 893     "coefdiv.z", /* 13 (d8)   */
 894     "coefdiv.y", /* 14 (d4)   */
 895     "coefdiv.x"  /* 15 (d2)   */
 896 };
 897
 898 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
 899         const struct wined3d_shader_dst_param *dst, char *write_mask)
 900 {
 901     char *ptr = write_mask;
 902
 903     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
 904     {
 905         *ptr++ = '.';
 906         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
 907         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
 908         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
 909         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
 910     }
 911
 912     *ptr = '\0';
 913 }
 914
 915 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
 916 {
 917     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
 918      * but addressed as "rgba". To fix this we need to swap the register's x
 919      * and z components. */
 920     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
 921     char *ptr = swizzle_str;
 922
 923     /* swizzle bits fields: wwzzyyxx */
 924     DWORD swizzle = param->swizzle;
 925     DWORD swizzle_x = swizzle & 0x03;
 926     DWORD swizzle_y = (swizzle >> 2) & 0x03;
 927     DWORD swizzle_z = (swizzle >> 4) & 0x03;
 928     DWORD swizzle_w = (swizzle >> 6) & 0x03;
 929
 930     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
 931      * generate a swizzle string. Unless we need to our own swizzling. */
 932     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
 933     {
 934         *ptr++ = '.';
 935         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
 936             *ptr++ = swizzle_chars[swizzle_x];
 937         } else {
 938             *ptr++ = swizzle_chars[swizzle_x];
 939             *ptr++ = swizzle_chars[swizzle_y];
 940             *ptr++ = swizzle_chars[swizzle_z];
 941             *ptr++ = swizzle_chars[swizzle_w];
 942         }
 943     }
 944
 945     *ptr = '\0';
 946 }
 947
 948 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
 949 {
 950     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 951     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
 952
 953     if (!strcmp(priv->addr_reg, src)) return;
 954
 955     strcpy(priv->addr_reg, src);
 956     shader_addline(buffer, "ARL A0.x, %s;\n", src);
 957 }
 958
 959 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
 960         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 961
 962 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
 963         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
 964 {
 965     /* oPos, oFog and oPts in D3D */
 966     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
 967     const struct wined3d_shader *shader = ins->ctx->shader;
 968     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
 969     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
 970     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 971
 972     *is_color = FALSE;
 973
 974     switch (reg->type)
 975     {
 976         case WINED3DSPR_TEMP:
 977             sprintf(register_name, "R%u", reg->idx);
 978             break;
 979
 980         case WINED3DSPR_INPUT:
 981             if (pshader)
 982             {
 983                 if (reg_maps->shader_version.major < 3)
 984                 {
 985                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
 986                     else strcpy(register_name, "fragment.color.secondary");
 987                 }
 988                 else
 989                 {
 990                     if(reg->rel_addr)
 991                     {
 992                         char rel_reg[50];
 993                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
 994
 995                         if (!strcmp(rel_reg, "**aL_emul**"))
 996                         {
 997                             DWORD idx = ctx->aL + reg->idx;
 998                             if(idx < MAX_REG_INPUT)
 999                             {
1000                                 strcpy(register_name, ctx->ps_input[idx]);
1001                             }
1002                             else
1003                             {
1004                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1005                                 sprintf(register_name, "out_of_bounds_%u", idx);
1006                             }
1007                         }
1008                         else if (reg_maps->input_registers & 0x0300)
1009                         {
1010                             /* There are two ways basically:
1011                              *
1012                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1013                              *    That means trouble if the loop also contains a breakc or if the control values
1014                              *    aren't local constants.
1015                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1016                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1017                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1018                              *    ADAC to load the condition code register and pop it again afterwards
1019                              */
1020                             FIXME("Relative input register addressing with more than 8 registers\n");
1021
1022                             /* This is better than nothing for now */
1023                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1024                         }
1025                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1026                         {
1027                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1028                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1029                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1030                              * proper varyings, or loop unrolling
1031                              *
1032                              * For now use the texcoords and hope for the best
1033                              */
1034                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1035                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1036                         }
1037                         else
1038                         {
1039                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1040                              * pulls GL_NV_fragment_program2 in
1041                              */
1042                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1043                         }
1044                     }
1045                     else
1046                     {
1047                         if(reg->idx < MAX_REG_INPUT)
1048                         {
1049                             strcpy(register_name, ctx->ps_input[reg->idx]);
1050                         }
1051                         else
1052                         {
1053                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1054                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1055                         }
1056                     }
1057                 }
1058             }
1059             else
1060             {
1061                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1062                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1063             }
1064             break;
1065
1066         case WINED3DSPR_CONST:
1067             if (!pshader && reg->rel_addr)
1068             {
1069                 const struct arb_vshader_private *shader_data = shader->backend_data;
1070                 UINT rel_offset = shader_data->rel_offset;
1071                 BOOL aL = FALSE;
1072                 char rel_reg[50];
1073                 if (reg_maps->shader_version.major < 2)
1074                 {
1075                     sprintf(rel_reg, "A0.x");
1076                 } else {
1077                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1078                     if(ctx->target_version == ARB) {
1079                         if (!strcmp(rel_reg, "**aL_emul**"))
1080                         {
1081                             aL = TRUE;
1082                         } else {
1083                             shader_arb_request_a0(ins, rel_reg);
1084                             sprintf(rel_reg, "A0.x");
1085                         }
1086                     }
1087                 }
1088                 if(aL)
1089                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1090                 else if (reg->idx >= rel_offset)
1091                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1092                 else
1093                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1094             }
1095             else
1096             {
1097                 if (reg_maps->usesrelconstF)
1098                     sprintf(register_name, "C[%u]", reg->idx);
1099                 else
1100                     sprintf(register_name, "C%u", reg->idx);
1101             }
1102             break;
1103
1104         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1105             if (pshader)
1106             {
1107                 if (reg_maps->shader_version.major == 1
1108                         && reg_maps->shader_version.minor <= 3)
1109                 {
1110                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1111                      * and as source to most instructions. For some instructions it is the texcoord
1112                      * input. Those instructions know about the special use
1113                      */
1114                     sprintf(register_name, "T%u", reg->idx);
1115                 } else {
1116                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1117                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1118                 }
1119             }
1120             else
1121             {
1122                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1123                 {
1124                     sprintf(register_name, "A%u", reg->idx);
1125                 }
1126                 else
1127                 {
1128                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1129                 }
1130             }
1131             break;
1132
1133         case WINED3DSPR_COLOROUT:
1134             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1135             {
1136                 strcpy(register_name, "TMP_COLOR");
1137             }
1138             else
1139             {
1140                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1141                 if (reg_maps->rt_mask > 1)
1142                 {
1143                     sprintf(register_name, "result.color[%u]", reg->idx);
1144                 }
1145                 else
1146                 {
1147                     strcpy(register_name, "result.color");
1148                 }
1149             }
1150             break;
1151
1152         case WINED3DSPR_RASTOUT:
1153             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1154             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1155             break;
1156
1157         case WINED3DSPR_DEPTHOUT:
1158             strcpy(register_name, "result.depth");
1159             break;
1160
1161         case WINED3DSPR_ATTROUT:
1162         /* case WINED3DSPR_OUTPUT: */
1163             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1164             else strcpy(register_name, ctx->color_output[reg->idx]);
1165             break;
1166
1167         case WINED3DSPR_TEXCRDOUT:
1168             if (pshader)
1169             {
1170                 sprintf(register_name, "oT[%u]", reg->idx);
1171             }
1172             else
1173             {
1174                 if (reg_maps->shader_version.major < 3)
1175                 {
1176                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1177                 }
1178                 else
1179                 {
1180                     strcpy(register_name, ctx->vs_output[reg->idx]);
1181                 }
1182             }
1183             break;
1184
1185         case WINED3DSPR_LOOP:
1186             if(ctx->target_version >= NV2)
1187             {
1188                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1189                 if(pshader) sprintf(register_name, "A0.x");
1190                 else sprintf(register_name, "aL.y");
1191             }
1192             else
1193             {
1194                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1195                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1196                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1197                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1198                  * indexing
1199                  */
1200                 sprintf(register_name, "**aL_emul**");
1201             }
1202
1203             break;
1204
1205         case WINED3DSPR_CONSTINT:
1206             sprintf(register_name, "I%u", reg->idx);
1207             break;
1208
1209         case WINED3DSPR_MISCTYPE:
1210             if (!reg->idx)
1211             {
1212                 sprintf(register_name, "vpos");
1213             }
1214             else if(reg->idx == 1)
1215             {
1216                 sprintf(register_name, "fragment.facing.x");
1217             }
1218             else
1219             {
1220                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1221             }
1222             break;
1223
1224         default:
1225             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1226             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1227             break;
1228     }
1229 }
1230
1231 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1232         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1233 {
1234     char register_name[255];
1235     char write_mask[6];
1236     BOOL is_color;
1237
1238     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1239     strcpy(str, register_name);
1240
1241     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1242     strcat(str, write_mask);
1243 }
1244
1245 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1246 {
1247     switch(channel_source)
1248     {
1249         case CHANNEL_SOURCE_ZERO: return "0";
1250         case CHANNEL_SOURCE_ONE: return "1";
1251         case CHANNEL_SOURCE_X: return "x";
1252         case CHANNEL_SOURCE_Y: return "y";
1253         case CHANNEL_SOURCE_Z: return "z";
1254         case CHANNEL_SOURCE_W: return "w";
1255         default:
1256             FIXME("Unhandled channel source %#x\n", channel_source);
1257             return "undefined";
1258     }
1259 }
1260
1261 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1262         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1263 {
1264     DWORD mask;
1265
1266     if (is_complex_fixup(fixup))
1267     {
1268         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1269         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1270         return;
1271     }
1272
1273     mask = 0;
1274     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1275     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1276     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1277     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1278     mask &= dst_mask;
1279
1280     if (mask)
1281     {
1282         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1283                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1284                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1285     }
1286
1287     mask = 0;
1288     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1289     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1290     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1291     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1292     mask &= dst_mask;
1293
1294     if (mask)
1295     {
1296         char reg_mask[6];
1297         char *ptr = reg_mask;
1298
1299         if (mask != WINED3DSP_WRITEMASK_ALL)
1300         {
1301             *ptr++ = '.';
1302             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1303             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1304             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1305             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1306         }
1307         *ptr = '\0';
1308
1309         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1310     }
1311 }
1312
1313 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1314 {
1315     DWORD mod;
1316     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1317     if (!ins->dst_count) return "";
1318
1319     mod = ins->dst[0].modifiers;
1320
1321     /* Silently ignore PARTIALPRECISION if its not supported */
1322     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1323
1324     if(mod & WINED3DSPDM_MSAMPCENTROID)
1325     {
1326         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1327         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1328     }
1329
1330     switch(mod)
1331     {
1332         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1333             return "H_SAT";
1334
1335         case WINED3DSPDM_SATURATE:
1336             return "_SAT";
1337
1338         case WINED3DSPDM_PARTIALPRECISION:
1339             return "H";
1340
1341         case 0:
1342             return "";
1343
1344         default:
1345             FIXME("Unknown modifiers 0x%08x\n", mod);
1346             return "";
1347     }
1348 }
1349
1350 #define TEX_PROJ        0x1
1351 #define TEX_BIAS        0x2
1352 #define TEX_LOD         0x4
1353 #define TEX_DERIV       0x10
1354
1355 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1356         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1357 {
1358     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1359     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1360     const struct wined3d_shader *shader = ins->ctx->shader;
1361     const struct wined3d_texture *texture;
1362     const char *tex_type;
1363     BOOL np2_fixup = FALSE;
1364     struct wined3d_device *device = shader->device;
1365     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1366     const char *mod;
1367     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1368
1369     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1370     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1371
1372     switch(sampler_type) {
1373         case WINED3DSTT_1D:
1374             tex_type = "1D";
1375             break;
1376
1377         case WINED3DSTT_2D:
1378             texture = device->stateBlock->state.textures[sampler_idx];
1379             if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1380             {
1381                 tex_type = "RECT";
1382             } else {
1383                 tex_type = "2D";
1384             }
1385             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1386             {
1387                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1388                 {
1389                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1390                     else np2_fixup = TRUE;
1391                 }
1392             }
1393             break;
1394
1395         case WINED3DSTT_VOLUME:
1396             tex_type = "3D";
1397             break;
1398
1399         case WINED3DSTT_CUBE:
1400             tex_type = "CUBE";
1401             break;
1402
1403         default:
1404             ERR("Unexpected texture type %d\n", sampler_type);
1405             tex_type = "";
1406     }
1407
1408     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1409      * so don't use shader_arb_get_modifier
1410      */
1411     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1412     else mod = "";
1413
1414     /* Fragment samplers always have indentity mapping */
1415     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1416     {
1417         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1418     }
1419
1420     if (flags & TEX_DERIV)
1421     {
1422         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1423         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1424         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1425                        dsx, dsy,sampler_idx, tex_type);
1426     }
1427     else if(flags & TEX_LOD)
1428     {
1429         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1430         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1431         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1432                        sampler_idx, tex_type);
1433     }
1434     else if (flags & TEX_BIAS)
1435     {
1436         /* Shouldn't be possible, but let's check for it */
1437         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1438         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1439         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1440     }
1441     else if (flags & TEX_PROJ)
1442     {
1443         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1444     }
1445     else
1446     {
1447         if (np2_fixup)
1448         {
1449             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1450             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1451                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1452
1453             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1454         }
1455         else
1456             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1457     }
1458
1459     if (pshader)
1460     {
1461         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1462                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1463                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1464                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1465     }
1466 }
1467
1468 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1469         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1470 {
1471     /* Generate a line that does the input modifier computation and return the input register to use */
1472     BOOL is_color = FALSE;
1473     char regstr[256];
1474     char swzstr[20];
1475     int insert_line;
1476     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1477     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1478     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1479     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1480
1481     /* Assume a new line will be added */
1482     insert_line = 1;
1483
1484     /* Get register name */
1485     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1486     shader_arb_get_swizzle(src, is_color, swzstr);
1487
1488     switch (src->modifiers)
1489     {
1490     case WINED3DSPSM_NONE:
1491         sprintf(outregstr, "%s%s", regstr, swzstr);
1492         insert_line = 0;
1493         break;
1494     case WINED3DSPSM_NEG:
1495         sprintf(outregstr, "-%s%s", regstr, swzstr);
1496         insert_line = 0;
1497         break;
1498     case WINED3DSPSM_BIAS:
1499         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1500         break;
1501     case WINED3DSPSM_BIASNEG:
1502         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1503         break;
1504     case WINED3DSPSM_SIGN:
1505         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1506         break;
1507     case WINED3DSPSM_SIGNNEG:
1508         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1509         break;
1510     case WINED3DSPSM_COMP:
1511         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1512         break;
1513     case WINED3DSPSM_X2:
1514         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1515         break;
1516     case WINED3DSPSM_X2NEG:
1517         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1518         break;
1519     case WINED3DSPSM_DZ:
1520         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1521         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1522         break;
1523     case WINED3DSPSM_DW:
1524         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1525         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1526         break;
1527     case WINED3DSPSM_ABS:
1528         if(ctx->target_version >= NV2) {
1529             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1530             insert_line = 0;
1531         } else {
1532             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1533         }
1534         break;
1535     case WINED3DSPSM_ABSNEG:
1536         if(ctx->target_version >= NV2) {
1537             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1538         } else {
1539             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1540             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1541         }
1542         insert_line = 0;
1543         break;
1544     default:
1545         sprintf(outregstr, "%s%s", regstr, swzstr);
1546         insert_line = 0;
1547     }
1548
1549     /* Return modified or original register, with swizzle */
1550     if (insert_line)
1551         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1552 }
1553
1554 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1555 {
1556     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1557     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1558     char dst_name[50];
1559     char src_name[2][50];
1560     DWORD sampler_code = dst->reg.idx;
1561
1562     shader_arb_get_dst_param(ins, dst, dst_name);
1563
1564     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1565      *
1566      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1567      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1568      * temps is done.
1569      */
1570     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1571     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1572     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1573     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1574     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1575
1576     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1577     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1578 }
1579
1580 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1581 {
1582     *extra_char = ' ';
1583     switch(mod)
1584     {
1585         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1586         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1587         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1588         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1589         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1590         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1591         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1592         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1593         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1594         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1595         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1596         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1597         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1598     }
1599     FIXME("Unknown modifier %u\n", mod);
1600     return mod;
1601 }
1602
1603 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1604 {
1605     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1606     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1607     char dst_name[50];
1608     char src_name[3][50];
1609     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1610             ins->ctx->reg_maps->shader_version.minor);
1611
1612     shader_arb_get_dst_param(ins, dst, dst_name);
1613     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1614
1615     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1616     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1617     {
1618         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1619     }
1620     else
1621     {
1622         struct wined3d_shader_src_param src0_copy = ins->src[0];
1623         char extra_neg;
1624
1625         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1626         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1627
1628         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1629         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1630         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1631         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1632                 dst_name, src_name[1], src_name[2]);
1633     }
1634 }
1635
1636 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1637 {
1638     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1639     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1640     char dst_name[50];
1641     char src_name[3][50];
1642
1643     shader_arb_get_dst_param(ins, dst, dst_name);
1644
1645     /* Generate input register names (with modifiers) */
1646     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1647     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1648     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1649
1650     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1651             dst_name, src_name[0], src_name[2], src_name[1]);
1652 }
1653
1654 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1655  * dst = dot2(src0, src1) + src2 */
1656 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1657 {
1658     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1659     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1660     char dst_name[50];
1661     char src_name[3][50];
1662     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1663
1664     shader_arb_get_dst_param(ins, dst, dst_name);
1665     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1666     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1667
1668     if(ctx->target_version >= NV3)
1669     {
1670         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1671         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1672         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1673                        dst_name, src_name[0], src_name[1], src_name[2]);
1674     }
1675     else if(ctx->target_version >= NV2)
1676     {
1677         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1678          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1679          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1680          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1681          *
1682          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1683          *
1684          * .xyxy and other swizzles that we could get with this are not valid in
1685          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1686          */
1687         struct wined3d_shader_src_param tmp_param = ins->src[1];
1688         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1689         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1690
1691         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1692
1693         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1694                        dst_name, src_name[2], src_name[0], src_name[1]);
1695     }
1696     else
1697     {
1698         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1699         /* 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]
1700         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1701         */
1702         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1703         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1704         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1705         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1706     }
1707 }
1708
1709 /* Map the opcode 1-to-1 to the GL code */
1710 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1711 {
1712     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1713     const char *instruction;
1714     char arguments[256], dst_str[50];
1715     unsigned int i;
1716     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1717
1718     switch (ins->handler_idx)
1719     {
1720         case WINED3DSIH_ABS: instruction = "ABS"; break;
1721         case WINED3DSIH_ADD: instruction = "ADD"; break;
1722         case WINED3DSIH_CRS: instruction = "XPD"; break;
1723         case WINED3DSIH_DP3: instruction = "DP3"; break;
1724         case WINED3DSIH_DP4: instruction = "DP4"; break;
1725         case WINED3DSIH_DST: instruction = "DST"; break;
1726         case WINED3DSIH_FRC: instruction = "FRC"; break;
1727         case WINED3DSIH_LIT: instruction = "LIT"; break;
1728         case WINED3DSIH_LRP: instruction = "LRP"; break;
1729         case WINED3DSIH_MAD: instruction = "MAD"; break;
1730         case WINED3DSIH_MAX: instruction = "MAX"; break;
1731         case WINED3DSIH_MIN: instruction = "MIN"; break;
1732         case WINED3DSIH_MOV: instruction = "MOV"; break;
1733         case WINED3DSIH_MUL: instruction = "MUL"; break;
1734         case WINED3DSIH_SGE: instruction = "SGE"; break;
1735         case WINED3DSIH_SLT: instruction = "SLT"; break;
1736         case WINED3DSIH_SUB: instruction = "SUB"; break;
1737         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1738         case WINED3DSIH_DSX: instruction = "DDX"; break;
1739         default: instruction = "";
1740             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1741             break;
1742     }
1743
1744     /* Note that shader_arb_add_dst_param() adds spaces. */
1745     arguments[0] = '\0';
1746     shader_arb_get_dst_param(ins, dst, dst_str);
1747     for (i = 0; i < ins->src_count; ++i)
1748     {
1749         char operand[100];
1750         strcat(arguments, ", ");
1751         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1752         strcat(arguments, operand);
1753     }
1754     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1755 }
1756
1757 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1758 {
1759     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1760     shader_addline(buffer, "NOP;\n");
1761 }
1762
1763 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1764 {
1765     const struct wined3d_shader *shader = ins->ctx->shader;
1766     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1767     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1768     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1769     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1770     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1771     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1772
1773     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1774     char src0_param[256];
1775
1776     if (ins->handler_idx == WINED3DSIH_MOVA)
1777     {
1778         const struct arb_vshader_private *shader_data = shader->backend_data;
1779         char write_mask[6];
1780         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1781
1782         if(ctx->target_version >= NV2) {
1783             shader_hw_map2gl(ins);
1784             return;
1785         }
1786         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1787         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1788
1789         /* This implements the mova formula used in GLSL. The first two instructions
1790          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1791          * in this case:
1792          * mova A0.x, 0.0
1793          *
1794          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1795          *
1796          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1797          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1798          */
1799         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1800         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1801
1802         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1803         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1804         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1805         if (shader_data->rel_offset)
1806         {
1807             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1808         }
1809         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1810
1811         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1812     }
1813     else if (reg_maps->shader_version.major == 1
1814           && !shader_is_pshader_version(reg_maps->shader_version.type)
1815           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1816     {
1817         const struct arb_vshader_private *shader_data = shader->backend_data;
1818         src0_param[0] = '\0';
1819
1820         if (shader_data->rel_offset)
1821         {
1822             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1823             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1824             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1825             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1826         }
1827         else
1828         {
1829             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1830              * with more than one component. Thus replicate the first source argument over all
1831              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1832             struct wined3d_shader_src_param tmp_src = ins->src[0];
1833             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1834             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1835             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1836         }
1837     }
1838     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1839     {
1840         if (ctx->cur_ps_args->super.srgb_correction && shader->u.ps.color0_mov)
1841         {
1842             shader_addline(buffer, "#mov handled in srgb write code\n");
1843             return;
1844         }
1845         shader_hw_map2gl(ins);
1846     }
1847     else
1848     {
1849         shader_hw_map2gl(ins);
1850     }
1851 }
1852
1853 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1854 {
1855     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1856     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1857     char reg_dest[40];
1858
1859     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1860      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1861      */
1862     shader_arb_get_dst_param(ins, dst, reg_dest);
1863
1864     if (ins->ctx->reg_maps->shader_version.major >= 2)
1865     {
1866         const char *kilsrc = "TA";
1867         BOOL is_color;
1868
1869         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1870         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1871         {
1872             kilsrc = reg_dest;
1873         }
1874         else
1875         {
1876             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1877              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1878              * masked out components to 0(won't kill)
1879              */
1880             char x = '0', y = '0', z = '0', w = '0';
1881             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1882             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1883             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1884             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1885             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1886         }
1887         shader_addline(buffer, "KIL %s;\n", kilsrc);
1888     } else {
1889         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1890          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1891          *
1892          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1893          * or pass in any temporary register(in shader phase 2)
1894          */
1895         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1896             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1897         } else {
1898             shader_arb_get_dst_param(ins, dst, reg_dest);
1899         }
1900         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1901         shader_addline(buffer, "KIL TA;\n");
1902     }
1903 }
1904
1905 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1906 {
1907     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1908     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1909     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1910             ins->ctx->reg_maps->shader_version.minor);
1911     struct wined3d_shader_src_param src;
1912
1913     char reg_dest[40];
1914     char reg_coord[40];
1915     DWORD reg_sampler_code;
1916     WORD myflags = 0;
1917
1918     /* All versions have a destination register */
1919     shader_arb_get_dst_param(ins, dst, reg_dest);
1920
1921     /* 1.0-1.4: Use destination register number as texture code.
1922        2.0+: Use provided sampler number as texure code. */
1923     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1924         reg_sampler_code = dst->reg.idx;
1925     else
1926         reg_sampler_code = ins->src[1].reg.idx;
1927
1928     /* 1.0-1.3: Use the texcoord varying.
1929        1.4+: Use provided coordinate source register. */
1930     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1931         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1932     else {
1933         /* TEX is the only instruction that can handle DW and DZ natively */
1934         src = ins->src[0];
1935         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1936         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1937         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1938     }
1939
1940     /* projection flag:
1941      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1942      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1943      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1944      */
1945     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1946     {
1947         DWORD flags = 0;
1948         if (reg_sampler_code < MAX_TEXTURES)
1949             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1950         if (flags & WINED3D_PSARGS_PROJECTED)
1951             myflags |= TEX_PROJ;
1952     }
1953     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1954     {
1955         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1956         if (src_mod == WINED3DSPSM_DZ)
1957         {
1958             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1959              * varying register, so we need a temp reg
1960              */
1961             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1962             strcpy(reg_coord, "TA");
1963             myflags |= TEX_PROJ;
1964         } else if(src_mod == WINED3DSPSM_DW) {
1965             myflags |= TEX_PROJ;
1966         }
1967     } else {
1968         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1969         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1970     }
1971     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1972 }
1973
1974 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1975 {
1976     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1977     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1978     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1979             ins->ctx->reg_maps->shader_version.minor);
1980     char dst_str[50];
1981
1982     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1983     {
1984         DWORD reg = dst->reg.idx;
1985
1986         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1987         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1988     } else {
1989         char reg_src[40];
1990
1991         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1992         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1993         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1994    }
1995 }
1996
1997 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1998 {
1999      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2000      DWORD flags = 0;
2001
2002      DWORD reg1 = ins->dst[0].reg.idx;
2003      char dst_str[50];
2004      char src_str[50];
2005
2006      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2007      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2008      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2009      /* Move .x first in case src_str is "TA" */
2010      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2011      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2012      if (reg1 < MAX_TEXTURES)
2013      {
2014          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2015          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2016      }
2017      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2018 }
2019
2020 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2021 {
2022      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2023
2024      DWORD reg1 = ins->dst[0].reg.idx;
2025      char dst_str[50];
2026      char src_str[50];
2027
2028      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2029      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2030      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2031      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2032      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2033      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2034 }
2035
2036 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2037 {
2038     DWORD reg1 = ins->dst[0].reg.idx;
2039     char dst_str[50];
2040     char src_str[50];
2041
2042     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2043     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2044     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2045     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2046 }
2047
2048 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2049 {
2050     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2051     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2052     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2053     char reg_coord[40], dst_reg[50], src_reg[50];
2054     DWORD reg_dest_code;
2055
2056     /* All versions have a destination register. The Tx where the texture coordinates come
2057      * from is the varying incarnation of the texture register
2058      */
2059     reg_dest_code = dst->reg.idx;
2060     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2061     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2062     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2063
2064     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2065      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2066      *
2067      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2068      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2069      *
2070      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2071      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2072      * extension.
2073      */
2074     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2075     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2076     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2077     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2078
2079     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2080      * so we can't let the GL handle this.
2081      */
2082     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2083             & WINED3D_PSARGS_PROJECTED)
2084     {
2085         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2086         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2087         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2088     } else {
2089         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2090     }
2091
2092     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2093
2094     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2095     {
2096         /* No src swizzles are allowed, so this is ok */
2097         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2098                        src_reg, reg_dest_code, reg_dest_code);
2099         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2100     }
2101 }
2102
2103 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2104 {
2105     DWORD reg = ins->dst[0].reg.idx;
2106     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2107     char src0_name[50], dst_name[50];
2108     BOOL is_color;
2109     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2110
2111     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2112     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2113      * T<reg+1> register. Use this register to store the calculated vector
2114      */
2115     tmp_reg.idx = reg + 1;
2116     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2117     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2118 }
2119
2120 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2121 {
2122     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2123     DWORD flags;
2124     DWORD reg = ins->dst[0].reg.idx;
2125     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2126     char dst_str[50];
2127     char src0_name[50];
2128     char dst_reg[50];
2129     BOOL is_color;
2130
2131     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2132     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2133
2134     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2135     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2136     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2137     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2138     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2139 }
2140
2141 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2142 {
2143     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2144     DWORD reg = ins->dst[0].reg.idx;
2145     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2146     char src0_name[50], dst_name[50];
2147     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2148     BOOL is_color;
2149
2150     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2151      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2152      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2153      */
2154     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2155     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2156
2157     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2158     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2159                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2160     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2161 }
2162
2163 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2164 {
2165     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2166     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2167     DWORD flags;
2168     DWORD reg = ins->dst[0].reg.idx;
2169     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2170     char dst_str[50];
2171     char src0_name[50], dst_name[50];
2172     BOOL is_color;
2173
2174     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2175     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2176     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2177
2178     /* Sample the texture using the calculated coordinates */
2179     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2180     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2181     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2182     tex_mx->current_row = 0;
2183 }
2184
2185 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2186 {
2187     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2188     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2189     DWORD flags;
2190     DWORD reg = ins->dst[0].reg.idx;
2191     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2192     char dst_str[50];
2193     char src0_name[50];
2194     char dst_reg[50];
2195     BOOL is_color;
2196
2197     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2198      * components for temporary data storage
2199      */
2200     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2201     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2202     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2203
2204     /* Construct the eye-ray vector from w coordinates */
2205     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2206     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2207     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2208
2209     /* Calculate reflection vector
2210      */
2211     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2212     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2213     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2214     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2215     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2216     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2217     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2218
2219     /* Sample the texture using the calculated coordinates */
2220     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2221     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2222     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2223     tex_mx->current_row = 0;
2224 }
2225
2226 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2227 {
2228     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2229     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2230     DWORD flags;
2231     DWORD reg = ins->dst[0].reg.idx;
2232     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2233     char dst_str[50];
2234     char src0_name[50];
2235     char src1_name[50];
2236     char dst_reg[50];
2237     BOOL is_color;
2238
2239     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2240     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2241     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2242     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2243     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2244
2245     /* Calculate reflection vector.
2246      *
2247      *                   dot(N, E)
2248      * dst_reg.xyz = 2 * --------- * N - E
2249      *                   dot(N, N)
2250      *
2251      * Which normalizes the normal vector
2252      */
2253     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2254     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2255     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2256     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2257     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2258     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2259
2260     /* Sample the texture using the calculated coordinates */
2261     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2262     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2263     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2264     tex_mx->current_row = 0;
2265 }
2266
2267 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2268 {
2269     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2270     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2271     char dst_name[50];
2272     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2273     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2274
2275     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2276      * which is essentially an input, is the destination register because it is the first
2277      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2278      * here(writemasks/swizzles are not valid on texdepth)
2279      */
2280     shader_arb_get_dst_param(ins, dst, dst_name);
2281
2282     /* According to the msdn, the source register(must be r5) is unusable after
2283      * the texdepth instruction, so we're free to modify it
2284      */
2285     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2286
2287     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2288      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2289      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2290      */
2291     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2292     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2293     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2294     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2295 }
2296
2297 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2298  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2299  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2300 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2301 {
2302     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2303     DWORD sampler_idx = ins->dst[0].reg.idx;
2304     char src0[50];
2305     char dst_str[50];
2306
2307     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2308     shader_addline(buffer, "MOV TB, 0.0;\n");
2309     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2310
2311     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2312     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2313 }
2314
2315 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2316  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2317 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2318 {
2319     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2320     char src0[50];
2321     char dst_str[50];
2322     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2323
2324     /* Handle output register */
2325     shader_arb_get_dst_param(ins, dst, dst_str);
2326     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2327     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2328 }
2329
2330 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2331  * Perform the 3rd row of a 3x3 matrix multiply */
2332 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2333 {
2334     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2335     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2336     char dst_str[50], dst_name[50];
2337     char src0[50];
2338     BOOL is_color;
2339
2340     shader_arb_get_dst_param(ins, dst, dst_str);
2341     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2342     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2343     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2344     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2345 }
2346
2347 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2348  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2349  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2350  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2351  */
2352 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2353 {
2354     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2355     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2356     char src0[50], dst_name[50];
2357     BOOL is_color;
2358     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2359     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2360
2361     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2362     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2363     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2364
2365     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2366      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2367      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2368      */
2369     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2370     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2371     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2372     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2373 }
2374
2375 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2376     Vertex/Pixel shaders to ARB_vertex_program codes */
2377 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2378 {
2379     int i;
2380     int nComponents = 0;
2381     struct wined3d_shader_dst_param tmp_dst = {{0}};
2382     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2383     struct wined3d_shader_instruction tmp_ins;
2384
2385     memset(&tmp_ins, 0, sizeof(tmp_ins));
2386
2387     /* Set constants for the temporary argument */
2388     tmp_ins.ctx = ins->ctx;
2389     tmp_ins.dst_count = 1;
2390     tmp_ins.dst = &tmp_dst;
2391     tmp_ins.src_count = 2;
2392     tmp_ins.src = tmp_src;
2393
2394     switch(ins->handler_idx)
2395     {
2396         case WINED3DSIH_M4x4:
2397             nComponents = 4;
2398             tmp_ins.handler_idx = WINED3DSIH_DP4;
2399             break;
2400         case WINED3DSIH_M4x3:
2401             nComponents = 3;
2402             tmp_ins.handler_idx = WINED3DSIH_DP4;
2403             break;
2404         case WINED3DSIH_M3x4:
2405             nComponents = 4;
2406             tmp_ins.handler_idx = WINED3DSIH_DP3;
2407             break;
2408         case WINED3DSIH_M3x3:
2409             nComponents = 3;
2410             tmp_ins.handler_idx = WINED3DSIH_DP3;
2411             break;
2412         case WINED3DSIH_M3x2:
2413             nComponents = 2;
2414             tmp_ins.handler_idx = WINED3DSIH_DP3;
2415             break;
2416         default:
2417             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2418             break;
2419     }
2420
2421     tmp_dst = ins->dst[0];
2422     tmp_src[0] = ins->src[0];
2423     tmp_src[1] = ins->src[1];
2424     for (i = 0; i < nComponents; i++) {
2425         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2426         shader_hw_map2gl(&tmp_ins);
2427         ++tmp_src[1].reg.idx;
2428     }
2429 }
2430
2431 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2432 {
2433     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2434
2435     char dst[50];
2436     char src[50];
2437
2438     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2439     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2440     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2441     {
2442         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2443          * .w is used
2444          */
2445         strcat(src, ".w");
2446     }
2447
2448     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2449 }
2450
2451 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2452 {
2453     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2454     const char *instruction;
2455
2456     char dst[50];
2457     char src[50];
2458
2459     switch(ins->handler_idx)
2460     {
2461         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2462         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2463         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2464         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2465         default: instruction = "";
2466             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2467             break;
2468     }
2469
2470     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2471     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2472     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2473     {
2474         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2475          * .w is used
2476          */
2477         strcat(src, ".w");
2478     }
2479
2480     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2481 }
2482
2483 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2484 {
2485     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2486     char dst_name[50];
2487     char src_name[50];
2488     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2489     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2490     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2491
2492     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2493     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2494
2495     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2496      * otherwise NRM or RSQ would return NaN */
2497     if(pshader && priv->target_version >= NV3)
2498     {
2499         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2500          *
2501          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2502          */
2503         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2504         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2505         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2506     }
2507     else if(priv->target_version >= NV2)
2508     {
2509         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2510         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2511         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2512                        src_name);
2513     }
2514     else
2515     {
2516         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2517
2518         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2519         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2520          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2521          */
2522         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2523         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2524
2525         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2526         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2527         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2528                     src_name);
2529     }
2530 }
2531
2532 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2533 {
2534     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2535     char dst_name[50];
2536     char src_name[3][50];
2537
2538     /* ARB_fragment_program has a convenient LRP instruction */
2539     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2540         shader_hw_map2gl(ins);
2541         return;
2542     }
2543
2544     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2545     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2546     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2547     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2548
2549     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2550     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2551                    dst_name, src_name[0], src_name[2]);
2552 }
2553
2554 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2555 {
2556     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2557      * must contain fixed constants. So we need a separate function to filter those constants and
2558      * can't use map2gl
2559      */
2560     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2561     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2562     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2563     char dst_name[50];
2564     char src_name0[50], src_name1[50], src_name2[50];
2565     BOOL is_color;
2566
2567     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2568     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2569         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2570         /* No modifiers are supported on SCS */
2571         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2572
2573         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2574         {
2575             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2576             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2577         }
2578     } else if(priv->target_version >= NV2) {
2579         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2580
2581         /* Sincos writemask must be .x, .y or .xy */
2582         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2583             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2584         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2585             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2586     } else {
2587         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2588          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2589          *
2590          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2591          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2592          *
2593          * The constants we get are:
2594          *
2595          *  +1   +1,     -1     -1     +1      +1      -1       -1
2596          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2597          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2598          *
2599          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2600          *
2601          * (x/2)^2 = x^2 / 4
2602          * (x/2)^3 = x^3 / 8
2603          * (x/2)^4 = x^4 / 16
2604          * (x/2)^5 = x^5 / 32
2605          * etc
2606          *
2607          * To get the final result:
2608          * sin(x) = 2 * sin(x/2) * cos(x/2)
2609          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2610          * (from sin(x+y) and cos(x+y) rules)
2611          *
2612          * As per MSDN, dst.z is undefined after the operation, and so is
2613          * dst.x and dst.y if they're masked out by the writemask. Ie
2614          * sincos dst.y, src1, c0, c1
2615          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2616          * vsa.exe also stops with an error if the dest register is the same register as the source
2617          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2618          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2619          */
2620         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2621         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2622         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2623
2624         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2625         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2626         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2627         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2628         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2629         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2630
2631         /* sin(x/2)
2632          *
2633          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2634          * properly merge that with MULs in the code above?
2635          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2636          * we can merge the sine and cosine MAD rows to calculate them together.
2637          */
2638         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2639         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2640         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2641         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2642
2643         /* cos(x/2) */
2644         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2645         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2646         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2647
2648         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2649             /* cos x */
2650             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2651             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2652         }
2653         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2654             /* sin x */
2655             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2656             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2657         }
2658     }
2659 }
2660
2661 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2662 {
2663     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2664     char dst_name[50];
2665     char src_name[50];
2666     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2667
2668     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2669     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2670
2671     /* SGN is only valid in vertex shaders */
2672     if(ctx->target_version >= NV2) {
2673         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2674         return;
2675     }
2676
2677     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2678      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2679      */
2680     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2681         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2682     } else {
2683         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2684          * Then use TA, and calculate the final result
2685          *
2686          * Not reading from TA? Store the first result in TA to avoid overwriting the
2687          * destination if src reg = dst reg
2688          */
2689         if(strstr(src_name, "TA"))
2690         {
2691             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2692             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2693             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2694         }
2695         else
2696         {
2697             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2698             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2699             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2700         }
2701     }
2702 }
2703
2704 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2705 {
2706     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2707     char src[50];
2708     char dst[50];
2709     char dst_name[50];
2710     BOOL is_color;
2711
2712     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2713     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2714     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2715
2716     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2717     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2718 }
2719
2720 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2721 {
2722     *need_abs = FALSE;
2723
2724     switch(mod)
2725     {
2726         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2727         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2728         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2729         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2730         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2731         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2732         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2733         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2734         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2735         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2736         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2737         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2738         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2739     }
2740     FIXME("Unknown modifier %u\n", mod);
2741     return mod;
2742 }
2743
2744 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2745 {
2746     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2747     char src0[50], dst[50];
2748     struct wined3d_shader_src_param src0_copy = ins->src[0];
2749     BOOL need_abs = FALSE;
2750     const char *instr;
2751
2752     switch(ins->handler_idx)
2753     {
2754         case WINED3DSIH_LOG:  instr = "LG2"; break;
2755         case WINED3DSIH_LOGP: instr = "LOG"; break;
2756         default:
2757             ERR("Unexpected instruction %d\n", ins->handler_idx);
2758             return;
2759     }
2760
2761     /* LOG and LOGP operate on the absolute value of the input */
2762     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2763
2764     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2765     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2766
2767     if(need_abs)
2768     {
2769         shader_addline(buffer, "ABS TA, %s;\n", src0);
2770         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2771     }
2772     else
2773     {
2774         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2775     }
2776 }
2777
2778 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2779 {
2780     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2781     char src0[50], src1[50], dst[50];
2782     struct wined3d_shader_src_param src0_copy = ins->src[0];
2783     BOOL need_abs = FALSE;
2784     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2785     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2786
2787     /* POW operates on the absolute value of the input */
2788     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2789
2790     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2791     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2792     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2793
2794     if (need_abs)
2795         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2796     else
2797         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2798
2799     if (priv->target_version >= NV2)
2800     {
2801         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2802         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2803         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2804     }
2805     else
2806     {
2807         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2808         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2809
2810         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2811         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2812         /* Possibly add flt_eps to avoid getting float special values */
2813         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2814         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2815         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2816         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2817     }
2818 }
2819
2820 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2821 {
2822     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2823     char src_name[50];
2824     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2825
2826     /* src0 is aL */
2827     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2828
2829     if(vshader)
2830     {
2831         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2832         struct list *e = list_head(&priv->control_frames);
2833         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2834
2835         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2836         /* The constant loader makes sure to load -1 into iX.w */
2837         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2838         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2839         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2840     }
2841     else
2842     {
2843         shader_addline(buffer, "LOOP %s;\n", src_name);
2844     }
2845 }
2846
2847 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2848 {
2849     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2850     char src_name[50];
2851     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2852
2853     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2854
2855     /* The constant loader makes sure to load -1 into iX.w */
2856     if(vshader)
2857     {
2858         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2859         struct list *e = list_head(&priv->control_frames);
2860         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2861
2862         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2863
2864         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2865         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2866         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2867     }
2868     else
2869     {
2870         shader_addline(buffer, "REP %s;\n", src_name);
2871     }
2872 }
2873
2874 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2875 {
2876     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2877     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2878
2879     if(vshader)
2880     {
2881         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2882         struct list *e = list_head(&priv->control_frames);
2883         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2884
2885         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2886         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2887         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2888
2889         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2890     }
2891     else
2892     {
2893         shader_addline(buffer, "ENDLOOP;\n");
2894     }
2895 }
2896
2897 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2898 {
2899     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2900     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2901
2902     if(vshader)
2903     {
2904         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2905         struct list *e = list_head(&priv->control_frames);
2906         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2907
2908         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2909         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2910         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2911
2912         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2913     }
2914     else
2915     {
2916         shader_addline(buffer, "ENDREP;\n");
2917     }
2918 }
2919
2920 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2921 {
2922     struct control_frame *control_frame;
2923
2924     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2925     {
2926         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2927     }
2928     ERR("Could not find loop for break\n");
2929     return NULL;
2930 }
2931
2932 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2933 {
2934     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2935     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2936     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2937
2938     if(vshader)
2939     {
2940         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2941     }
2942     else
2943     {
2944         shader_addline(buffer, "BRK;\n");
2945     }
2946 }
2947
2948 static const char *get_compare(enum wined3d_shader_rel_op op)
2949 {
2950     switch (op)
2951     {
2952         case WINED3D_SHADER_REL_OP_GT: return "GT";
2953         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2954         case WINED3D_SHADER_REL_OP_GE: return "GE";
2955         case WINED3D_SHADER_REL_OP_LT: return "LT";
2956         case WINED3D_SHADER_REL_OP_NE: return "NE";
2957         case WINED3D_SHADER_REL_OP_LE: return "LE";
2958         default:
2959             FIXME("Unrecognized operator %#x.\n", op);
2960             return "(\?\?)";
2961     }
2962 }
2963
2964 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
2965 {
2966     switch (op)
2967     {
2968         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
2969         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
2970         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
2971         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
2972         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
2973         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
2974         default:
2975             FIXME("Unrecognized operator %#x.\n", op);
2976             return -1;
2977     }
2978 }
2979
2980 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2981 {
2982     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2983     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2984     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2985     char src_name0[50];
2986     char src_name1[50];
2987     const char *comp = get_compare(ins->flags);
2988
2989     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2990     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2991
2992     if(vshader)
2993     {
2994         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2995          * away the subtraction result
2996          */
2997         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2998         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2999     }
3000     else
3001     {
3002         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3003         shader_addline(buffer, "BRK (%s.x);\n", comp);
3004     }
3005 }
3006
3007 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3008 {
3009     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3010     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3011     struct list *e = list_head(&priv->control_frames);
3012     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3013     const char *comp;
3014     char src_name0[50];
3015     char src_name1[50];
3016     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3017
3018     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3019     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3020
3021     if(vshader)
3022     {
3023         /* Invert the flag. We jump to the else label if the condition is NOT true */
3024         comp = get_compare(invert_compare(ins->flags));
3025         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3026         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3027     }
3028     else
3029     {
3030         comp = get_compare(ins->flags);
3031         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3032         shader_addline(buffer, "IF %s.x;\n", comp);
3033     }
3034 }
3035
3036 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3037 {
3038     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3039     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3040     struct list *e = list_head(&priv->control_frames);
3041     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3042     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3043
3044     if(vshader)
3045     {
3046         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3047         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3048         control_frame->had_else = TRUE;
3049     }
3050     else
3051     {
3052         shader_addline(buffer, "ELSE;\n");
3053     }
3054 }
3055
3056 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3057 {
3058     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3059     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3060     struct list *e = list_head(&priv->control_frames);
3061     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3062     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3063
3064     if(vshader)
3065     {
3066         if(control_frame->had_else)
3067         {
3068             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3069         }
3070         else
3071         {
3072             shader_addline(buffer, "#No else branch. else is endif\n");
3073             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3074         }
3075     }
3076     else
3077     {
3078         shader_addline(buffer, "ENDIF;\n");
3079     }
3080 }
3081
3082 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3083 {
3084     DWORD sampler_idx = ins->src[1].reg.idx;
3085     char reg_dest[40];
3086     char reg_src[3][40];
3087     WORD flags = TEX_DERIV;
3088
3089     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3090     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3091     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3092     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3093
3094     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3095     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3096
3097     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3098 }
3099
3100 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3101 {
3102     DWORD sampler_idx = ins->src[1].reg.idx;
3103     char reg_dest[40];
3104     char reg_coord[40];
3105     WORD flags = TEX_LOD;
3106
3107     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3108     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3109
3110     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3111     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3112
3113     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3114 }
3115
3116 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3117 {
3118     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3119     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3120
3121     priv->in_main_func = FALSE;
3122     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3123      * subroutine, don't generate a label that will make GL complain
3124      */
3125     if(priv->target_version == ARB) return;
3126
3127     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3128 }
3129
3130 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3131         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3132         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3133         struct wined3d_shader_buffer *buffer)
3134 {
3135     unsigned int i;
3136
3137     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3138      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3139      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3140      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3141      */
3142     if (args->super.fog_src == VS_FOG_Z)
3143     {
3144         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3145     }
3146     else
3147     {
3148         if (!reg_maps->fog)
3149         {
3150             /* posFixup.x is always 1.0, so we can safely use it */
3151             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3152         }
3153         else
3154         {
3155             /* Clamp fogcoord */
3156             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3157             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3158
3159             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3160             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3161         }
3162     }
3163
3164     /* Clipplanes are always stored without y inversion */
3165     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3166     {
3167         if (args->super.clip_enabled)
3168         {
3169             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3170             {
3171                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3172             }
3173         }
3174     }
3175     else if (args->clip.boolclip.clip_texcoord)
3176     {
3177         unsigned int cur_clip = 0;
3178         char component[4] = {'x', 'y', 'z', 'w'};
3179         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3180
3181         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3182         {
3183             if (args->clip.boolclip.clipplane_mask & (1 << i))
3184             {
3185                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3186                                component[cur_clip++], i);
3187             }
3188         }
3189         switch (cur_clip)
3190         {
3191             case 0:
3192                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3193                 break;
3194             case 1:
3195                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3196                 break;
3197             case 2:
3198                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3199                 break;
3200             case 3:
3201                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3202                 break;
3203         }
3204         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3205                        args->clip.boolclip.clip_texcoord - 1);
3206     }
3207
3208     /* Write the final position.
3209      *
3210      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3211      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3212      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3213      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3214      */
3215     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3216     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3217     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3218
3219     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3220      * and the glsl equivalent
3221      */
3222     if (need_helper_const(shader_data, reg_maps, gl_info))
3223     {
3224         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3225         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3226     }
3227     else
3228     {
3229         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3230         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3231     }
3232
3233     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3234
3235     priv_ctx->footer_written = TRUE;
3236 }
3237
3238 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3239 {
3240     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3241     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3242     const struct wined3d_shader *shader = ins->ctx->shader;
3243     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3244
3245     if(priv->target_version == ARB) return;
3246
3247     if(vshader)
3248     {
3249         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3250                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3251     }
3252
3253     shader_addline(buffer, "RET;\n");
3254 }
3255
3256 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3257 {
3258     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3259     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3260 }
3261
3262 /* GL locking is done by the caller */
3263 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3264 {
3265     GLuint program_id = 0;
3266     GLint pos;
3267
3268     const char *blt_vprogram =
3269         "!!ARBvp1.0\n"
3270         "PARAM c[1] = { { 1, 0.5 } };\n"
3271         "MOV result.position, vertex.position;\n"
3272         "MOV result.color, c[0].x;\n"
3273         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3274         "END\n";
3275
3276     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3277     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3278     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3279             strlen(blt_vprogram), blt_vprogram));
3280     checkGLcall("glProgramStringARB()");
3281
3282     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3283     if (pos != -1)
3284     {
3285         FIXME("Vertex program error at position %d: %s\n\n", pos,
3286             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3287         shader_arb_dump_program_source(blt_vprogram);
3288     }
3289     else
3290     {
3291         GLint native;
3292
3293         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3294         checkGLcall("glGetProgramivARB()");
3295         if (!native) WARN("Program exceeds native resource limits.\n");
3296     }
3297
3298     return program_id;
3299 }
3300
3301 /* GL locking is done by the caller */
3302 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3303         enum tex_types tex_type, BOOL masked)
3304 {
3305     GLuint program_id = 0;
3306     const char *fprogram;
3307     GLint pos;
3308
3309     static const char * const blt_fprograms_full[tex_type_count] =
3310     {
3311         /* tex_1d */
3312         NULL,
3313         /* tex_2d */
3314         "!!ARBfp1.0\n"
3315         "TEMP R0;\n"
3316         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3317         "MOV result.depth.z, R0.x;\n"
3318         "END\n",
3319         /* tex_3d */
3320         NULL,
3321         /* tex_cube */
3322         "!!ARBfp1.0\n"
3323         "TEMP R0;\n"
3324         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3325         "MOV result.depth.z, R0.x;\n"
3326         "END\n",
3327         /* tex_rect */
3328         "!!ARBfp1.0\n"
3329         "TEMP R0;\n"
3330         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3331         "MOV result.depth.z, R0.x;\n"
3332         "END\n",
3333     };
3334
3335     static const char * const blt_fprograms_masked[tex_type_count] =
3336     {
3337         /* tex_1d */
3338         NULL,
3339         /* tex_2d */
3340         "!!ARBfp1.0\n"
3341         "PARAM mask = program.local[0];\n"
3342         "TEMP R0;\n"
3343         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3344         "MUL R0.x, R0.x, R0.y;\n"
3345         "KIL -R0.x;\n"
3346         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3347         "MOV result.depth.z, R0.x;\n"
3348         "END\n",
3349         /* tex_3d */
3350         NULL,
3351         /* tex_cube */
3352         "!!ARBfp1.0\n"
3353         "PARAM mask = program.local[0];\n"
3354         "TEMP R0;\n"
3355         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3356         "MUL R0.x, R0.x, R0.y;\n"
3357         "KIL -R0.x;\n"
3358         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3359         "MOV result.depth.z, R0.x;\n"
3360         "END\n",
3361         /* tex_rect */
3362         "!!ARBfp1.0\n"
3363         "PARAM mask = program.local[0];\n"
3364         "TEMP R0;\n"
3365         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3366         "MUL R0.x, R0.x, R0.y;\n"
3367         "KIL -R0.x;\n"
3368         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3369         "MOV result.depth.z, R0.x;\n"
3370         "END\n",
3371     };
3372
3373     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3374     if (!fprogram)
3375     {
3376         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3377         tex_type = tex_2d;
3378         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3379     }
3380
3381     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3382     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3383     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3384     checkGLcall("glProgramStringARB()");
3385
3386     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3387     if (pos != -1)
3388     {
3389         FIXME("Fragment program error at position %d: %s\n\n", pos,
3390             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3391         shader_arb_dump_program_source(fprogram);
3392     }
3393     else
3394     {
3395         GLint native;
3396
3397         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3398         checkGLcall("glGetProgramivARB()");
3399         if (!native) WARN("Program exceeds native resource limits.\n");
3400     }
3401
3402     return program_id;
3403 }
3404
3405 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3406         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3407 {
3408     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3409
3410     if(condcode)
3411     {
3412         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3413         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3414         /* Calculate the > 0.0031308 case */
3415         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3416         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3417         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3418         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3419         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3420         /* Calculate the < case */
3421         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3422     }
3423     else
3424     {
3425         /* Calculate the > 0.0031308 case */
3426         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3427         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3428         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3429         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3430         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3431         /* Calculate the < case */
3432         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3433         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3434         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3435         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3436         /* Store the components > 0.0031308 in the destination */
3437         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3438         /* Add the components that are < 0.0031308 */
3439         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3440         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3441         * result.color writes(.rgb first, then .a), or handle overwriting already written
3442         * components. The assembler uses a temporary register in this case, which is usually
3443         * not allocated from one of our registers that were used earlier.
3444         */
3445     }
3446     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3447 }
3448
3449 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3450 {
3451     const struct wined3d_shader_lconst *constant;
3452
3453     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3454     {
3455         if (constant->idx == idx)
3456         {
3457             return constant->value;
3458         }
3459     }
3460     return NULL;
3461 }
3462
3463 static void init_ps_input(const struct wined3d_shader *shader,
3464         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3465 {
3466     static const char * const texcoords[8] =
3467     {
3468         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3469         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3470     };
3471     unsigned int i;
3472     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3473     const char *semantic_name;
3474     DWORD semantic_idx;
3475
3476     switch(args->super.vp_mode)
3477     {
3478         case pretransformed:
3479         case fixedfunction:
3480             /* The pixelshader has to collect the varyings on its own. In any case properly load
3481              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3482              * other attribs to 0.0.
3483              *
3484              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3485              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3486              * load the texcoord attrib pointers to match the pixel shader signature
3487              */
3488             for(i = 0; i < MAX_REG_INPUT; i++)
3489             {
3490                 semantic_name = sig[i].semantic_name;
3491                 semantic_idx = sig[i].semantic_idx;
3492                 if (!semantic_name) continue;
3493
3494                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3495                 {
3496                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3497                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3498                     else priv->ps_input[i] = "0.0";
3499                 }
3500                 else if(args->super.vp_mode == fixedfunction)
3501                 {
3502                     priv->ps_input[i] = "0.0";
3503                 }
3504                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3505                 {
3506                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3507                     else priv->ps_input[i] = "0.0";
3508                 }
3509                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3510                 {
3511                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3512                     else priv->ps_input[i] = "0.0";
3513                 }
3514                 else
3515                 {
3516                     priv->ps_input[i] = "0.0";
3517                 }
3518
3519                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3520             }
3521             break;
3522
3523         case vertexshader:
3524             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3525              * fragment.color
3526              */
3527             for(i = 0; i < 8; i++)
3528             {
3529                 priv->ps_input[i] = texcoords[i];
3530             }
3531             priv->ps_input[8] = "fragment.color.primary";
3532             priv->ps_input[9] = "fragment.color.secondary";
3533             break;
3534     }
3535 }
3536
3537 /* GL locking is done by the caller */
3538 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3539         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3540         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3541 {
3542     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3543     const struct wined3d_shader_lconst *lconst;
3544     const DWORD *function = shader->function;
3545     GLuint retval;
3546     char fragcolor[16];
3547     DWORD *lconst_map = local_const_mapping(shader), next_local;
3548     struct shader_arb_ctx_priv priv_ctx;
3549     BOOL dcl_td = FALSE;
3550     BOOL want_nv_prog = FALSE;
3551     struct arb_pshader_private *shader_priv = shader->backend_data;
3552     GLint errPos;
3553     DWORD map;
3554
3555     char srgbtmp[4][4];
3556     unsigned int i, found = 0;
3557
3558     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3559     {
3560         if (!(map & 1)
3561                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3562                 || (reg_maps->shader_version.major < 2 && !i))
3563             continue;
3564
3565         sprintf(srgbtmp[found], "R%u", i);
3566         ++found;
3567         if (found == 4) break;
3568     }
3569
3570     switch(found) {
3571         case 0:
3572             sprintf(srgbtmp[0], "TA");
3573             sprintf(srgbtmp[1], "TB");
3574             sprintf(srgbtmp[2], "TC");
3575             sprintf(srgbtmp[3], "TD");
3576             dcl_td = TRUE;
3577             break;
3578         case 1:
3579             sprintf(srgbtmp[1], "TA");
3580             sprintf(srgbtmp[2], "TB");
3581             sprintf(srgbtmp[3], "TC");
3582             break;
3583         case 2:
3584             sprintf(srgbtmp[2], "TA");
3585             sprintf(srgbtmp[3], "TB");
3586             break;
3587         case 3:
3588             sprintf(srgbtmp[3], "TA");
3589             break;
3590         case 4:
3591             break;
3592     }
3593
3594     /*  Create the hw ARB shader */
3595     memset(&priv_ctx, 0, sizeof(priv_ctx));
3596     priv_ctx.cur_ps_args = args;
3597     priv_ctx.compiled_fprog = compiled;
3598     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3599     init_ps_input(shader, args, &priv_ctx);
3600     list_init(&priv_ctx.control_frames);
3601
3602     /* Avoid enabling NV_fragment_program* if we do not need it.
3603      *
3604      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3605      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3606      * is faster than what we gain from using higher native instructions. There are some things though
3607      * that cannot be emulated. In that case enable the extensions.
3608      * If the extension is enabled, instruction handlers that support both ways will use it.
3609      *
3610      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3611      * So enable the best we can get.
3612      */
3613     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3614        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3615     {
3616         want_nv_prog = TRUE;
3617     }
3618
3619     shader_addline(buffer, "!!ARBfp1.0\n");
3620     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3621     {
3622         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3623         priv_ctx.target_version = NV3;
3624     }
3625     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3626     {
3627         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3628         priv_ctx.target_version = NV2;
3629     } else {
3630         if(want_nv_prog)
3631         {
3632             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3633              * limits properly
3634              */
3635             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3636             ERR("Try GLSL\n");
3637         }
3638         priv_ctx.target_version = ARB;
3639     }
3640
3641     if (reg_maps->rt_mask > 1)
3642     {
3643         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3644     }
3645
3646     if (reg_maps->shader_version.major < 3)
3647     {
3648         switch(args->super.fog) {
3649             case FOG_OFF:
3650                 break;
3651             case FOG_LINEAR:
3652                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3653                 break;
3654             case FOG_EXP:
3655                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3656                 break;
3657             case FOG_EXP2:
3658                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3659                 break;
3660         }
3661     }
3662
3663     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3664      * unused temps away(but occupies them for the whole shader if they're used once). Always
3665      * declaring them avoids tricky bookkeeping work
3666      */
3667     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3668     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3669     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3670     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3671     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3672     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3673     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3674
3675     if (reg_maps->shader_version.major < 2)
3676     {
3677         strcpy(fragcolor, "R0");
3678     }
3679     else
3680     {
3681         if (args->super.srgb_correction)
3682         {
3683             if (shader->u.ps.color0_mov)
3684             {
3685                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3686             }
3687             else
3688             {
3689                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3690                 strcpy(fragcolor, "TMP_COLOR");
3691             }
3692         } else {
3693             strcpy(fragcolor, "result.color");
3694         }
3695     }
3696
3697     if(args->super.srgb_correction) {
3698         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3699                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3700         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3701                        srgb_sub_high, 0.0, 0.0, 0.0);
3702     }
3703
3704     /* Base Declarations */
3705     next_local = shader_generate_arb_declarations(shader, reg_maps,
3706             buffer, gl_info, lconst_map, NULL, &priv_ctx);
3707
3708     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3709     {
3710         unsigned char bump_const;
3711
3712         if (!(map & 1)) continue;
3713
3714         bump_const = compiled->numbumpenvmatconsts;
3715         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3716         compiled->bumpenvmatconst[bump_const].texunit = i;
3717         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3718         compiled->luminanceconst[bump_const].texunit = i;
3719
3720         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3721          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3722          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3723          * textures due to conditional NP2 restrictions)
3724          *
3725          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3726          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3727          * their location is shader dependent anyway and they cannot be loaded globally.
3728          */
3729         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3730         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3731                        i, compiled->bumpenvmatconst[bump_const].const_num);
3732         compiled->numbumpenvmatconsts = bump_const + 1;
3733
3734         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3735
3736         compiled->luminanceconst[bump_const].const_num = next_local++;
3737         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3738                        i, compiled->luminanceconst[bump_const].const_num);
3739     }
3740
3741     for(i = 0; i < MAX_CONST_I; i++)
3742     {
3743         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3744         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3745         {
3746             const DWORD *control_values = find_loop_control_values(shader, i);
3747
3748             if(control_values)
3749             {
3750                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3751                                 control_values[0], control_values[1], control_values[2]);
3752             }
3753             else
3754             {
3755                 compiled->int_consts[i] = next_local;
3756                 compiled->num_int_consts++;
3757                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3758             }
3759         }
3760     }
3761
3762     if(reg_maps->vpos || reg_maps->usesdsy)
3763     {
3764         compiled->ycorrection = next_local;
3765         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3766
3767         if(reg_maps->vpos)
3768         {
3769             shader_addline(buffer, "TEMP vpos;\n");
3770             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3771              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3772              * ycorrection.z: 1.0
3773              * ycorrection.w: 0.0
3774              */
3775             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3776             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3777         }
3778     }
3779     else
3780     {
3781         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3782     }
3783
3784     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3785      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3786      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3787      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3788      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3789      * shader compilation errors and the subsequent errors when drawing with this shader. */
3790     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3791         unsigned char cur_fixup_sampler = 0;
3792
3793         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3794         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3795         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3796
3797         fixup->offset = next_local;
3798         fixup->super.active = 0;
3799
3800         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3801             if (!(map & (1 << i))) continue;
3802
3803             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3804                 fixup->super.active |= (1 << i);
3805                 fixup->super.idx[i] = cur_fixup_sampler++;
3806             } else {
3807                 FIXME("No free constant found to load NP2 fixup data into shader. "
3808                       "Sampling from this texture will probably look wrong.\n");
3809                 break;
3810             }
3811         }
3812
3813         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3814         if (fixup->super.num_consts) {
3815             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3816                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3817         }
3818     }
3819
3820     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3821     {
3822         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3823     }
3824
3825     /* Base Shader Body */
3826     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3827
3828     if(args->super.srgb_correction) {
3829         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3830                                   priv_ctx.target_version >= NV2);
3831     }
3832
3833     if(strcmp(fragcolor, "result.color")) {
3834         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3835     }
3836     shader_addline(buffer, "END\n");
3837
3838     /* TODO: change to resource.glObjectHandle or something like that */
3839     GL_EXTCALL(glGenProgramsARB(1, &retval));
3840
3841     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3842     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3843
3844     TRACE("Created hw pixel shader, prg=%d\n", retval);
3845     /* Create the program and check for errors */
3846     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3847                buffer->bsize, buffer->buffer));
3848     checkGLcall("glProgramStringARB()");
3849
3850     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3851     if (errPos != -1)
3852     {
3853         FIXME("HW PixelShader Error at position %d: %s\n\n",
3854               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3855         shader_arb_dump_program_source(buffer->buffer);
3856         retval = 0;
3857     }
3858     else
3859     {
3860         GLint native;
3861
3862         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3863         checkGLcall("glGetProgramivARB()");
3864         if (!native) WARN("Program exceeds native resource limits.\n");
3865     }
3866
3867     /* Load immediate constants */
3868     if (lconst_map)
3869     {
3870         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3871         {
3872             const float *value = (const float *)lconst->value;
3873             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3874             checkGLcall("glProgramLocalParameter4fvARB");
3875         }
3876         HeapFree(GetProcessHeap(), 0, lconst_map);
3877     }
3878
3879     return retval;
3880 }
3881
3882 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3883 {
3884     unsigned int i;
3885     int ret;
3886
3887     for(i = 0; i < MAX_REG_INPUT; i++)
3888     {
3889         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3890         {
3891             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3892             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3893             continue;
3894         }
3895
3896         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3897         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3898         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3899         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3900         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3901         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3902     }
3903     return 0;
3904 }
3905
3906 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3907 {
3908     struct wined3d_shader_signature_element *new;
3909     int i;
3910     char *name;
3911
3912     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3913     for(i = 0; i < MAX_REG_INPUT; i++)
3914     {
3915         if (!sig[i].semantic_name) continue;
3916
3917         new[i] = sig[i];
3918         /* Clone the semantic string */
3919         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3920         strcpy(name, sig[i].semantic_name);
3921         new[i].semantic_name = name;
3922     }
3923     return new;
3924 }
3925
3926 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3927 {
3928     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3929     struct ps_signature *found_sig;
3930
3931     if (entry)
3932     {
3933         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3934         TRACE("Found existing signature %u\n", found_sig->idx);
3935         return found_sig->idx;
3936     }
3937     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3938     found_sig->sig = clone_sig(sig);
3939     found_sig->idx = priv->ps_sig_number++;
3940     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3941     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3942     {
3943         ERR("Failed to insert program entry.\n");
3944     }
3945     return found_sig->idx;
3946 }
3947
3948 static void init_output_registers(const struct wined3d_shader *shader, DWORD sig_num,
3949         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3950 {
3951     unsigned int i, j;
3952     static const char * const texcoords[8] =
3953     {
3954         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3955         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3956     };
3957     struct wined3d_device *device = shader->device;
3958     const struct wined3d_shader_signature_element *sig;
3959     const char *semantic_name;
3960     DWORD semantic_idx, reg_idx;
3961
3962     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3963      * and varying 9 to result.color.secondary
3964      */
3965     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3966     {
3967         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3968         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3969         "result.color.primary", "result.color.secondary"
3970     };
3971
3972     if(sig_num == ~0)
3973     {
3974         TRACE("Pixel shader uses builtin varyings\n");
3975         /* Map builtins to builtins */
3976         for(i = 0; i < 8; i++)
3977         {
3978             priv_ctx->texcrd_output[i] = texcoords[i];
3979         }
3980         priv_ctx->color_output[0] = "result.color.primary";
3981         priv_ctx->color_output[1] = "result.color.secondary";
3982         priv_ctx->fog_output = "TMP_FOGCOORD";
3983
3984         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3985         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
3986         {
3987             semantic_name = shader->output_signature[i].semantic_name;
3988             if (!semantic_name) continue;
3989
3990             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
3991             {
3992                 TRACE("o%u is TMP_OUT\n", i);
3993                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3994                 else priv_ctx->vs_output[i] = "TA";
3995             }
3996             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
3997             {
3998                 TRACE("o%u is result.pointsize\n", i);
3999                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4000                 else priv_ctx->vs_output[i] = "TA";
4001             }
4002             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4003             {
4004                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4005                 if (!shader->output_signature[i].semantic_idx)
4006                     priv_ctx->vs_output[i] = "result.color.primary";
4007                 else if (shader->output_signature[i].semantic_idx == 1)
4008                     priv_ctx->vs_output[i] = "result.color.secondary";
4009                 else priv_ctx->vs_output[i] = "TA";
4010             }
4011             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4012             {
4013                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4014                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4015                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4016             }
4017             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4018             {
4019                 TRACE("o%u is result.fogcoord\n", i);
4020                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4021                 else priv_ctx->vs_output[i] = "result.fogcoord";
4022             }
4023             else
4024             {
4025                 priv_ctx->vs_output[i] = "TA";
4026             }
4027         }
4028         return;
4029     }
4030
4031     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
4032      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
4033      */
4034     sig = device->stateBlock->state.pixel_shader->input_signature;
4035     TRACE("Pixel shader uses declared varyings\n");
4036
4037     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4038     for(i = 0; i < 8; i++)
4039     {
4040         priv_ctx->texcrd_output[i] = "TA";
4041     }
4042     priv_ctx->color_output[0] = "TA";
4043     priv_ctx->color_output[1] = "TA";
4044     priv_ctx->fog_output = "TA";
4045
4046     for(i = 0; i < MAX_REG_INPUT; i++)
4047     {
4048         semantic_name = sig[i].semantic_name;
4049         semantic_idx = sig[i].semantic_idx;
4050         reg_idx = sig[i].register_idx;
4051         if (!semantic_name) continue;
4052
4053         /* If a declared input register is not written by builtin arguments, don't write to it.
4054          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4055          *
4056          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4057          * to TMP_OUT in any case
4058          */
4059         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4060         {
4061             if (semantic_idx < 8)
4062                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4063         }
4064         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4065         {
4066             if (semantic_idx < 2)
4067                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4068         }
4069         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4070         {
4071             if (!semantic_idx)
4072                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4073         }
4074         else
4075         {
4076             continue;
4077         }
4078
4079         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4080                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4081         {
4082             compiled->need_color_unclamp = TRUE;
4083         }
4084     }
4085
4086     /* Map declared to declared */
4087     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4088     {
4089         /* Write unread output to TA to throw them away */
4090         priv_ctx->vs_output[i] = "TA";
4091         semantic_name = shader->output_signature[i].semantic_name;
4092         if (!semantic_name) continue;
4093
4094         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4095                 && !shader->output_signature[i].semantic_idx)
4096         {
4097             priv_ctx->vs_output[i] = "TMP_OUT";
4098             continue;
4099         }
4100         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4101                 && !shader->output_signature[i].semantic_idx)
4102         {
4103             priv_ctx->vs_output[i] = "result.pointsize";
4104             continue;
4105         }
4106
4107         for(j = 0; j < MAX_REG_INPUT; j++)
4108         {
4109             if (!sig[j].semantic_name) continue;
4110
4111             if (!strcmp(sig[j].semantic_name, semantic_name)
4112                     && sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4113             {
4114                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4115
4116                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4117                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4118                 {
4119                     compiled->need_color_unclamp = TRUE;
4120                 }
4121             }
4122         }
4123     }
4124 }
4125
4126 /* GL locking is done by the caller */
4127 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4128         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4129         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4130 {
4131     const struct arb_vshader_private *shader_data = shader->backend_data;
4132     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4133     const struct wined3d_shader_lconst *lconst;
4134     const DWORD *function = shader->function;
4135     GLuint ret;
4136     DWORD next_local, *lconst_map = local_const_mapping(shader);
4137     struct shader_arb_ctx_priv priv_ctx;
4138     unsigned int i;
4139     GLint errPos;
4140
4141     memset(&priv_ctx, 0, sizeof(priv_ctx));
4142     priv_ctx.cur_vs_args = args;
4143     list_init(&priv_ctx.control_frames);
4144     init_output_registers(shader, args->ps_signature, &priv_ctx, compiled);
4145
4146     /*  Create the hw ARB shader */
4147     shader_addline(buffer, "!!ARBvp1.0\n");
4148
4149     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4150      * mesurable performance penalty, and we can always make use of it for clipplanes.
4151      */
4152     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4153     {
4154         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4155         priv_ctx.target_version = NV3;
4156         shader_addline(buffer, "ADDRESS aL;\n");
4157     }
4158     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4159     {
4160         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4161         priv_ctx.target_version = NV2;
4162         shader_addline(buffer, "ADDRESS aL;\n");
4163     } else {
4164         priv_ctx.target_version = ARB;
4165     }
4166
4167     shader_addline(buffer, "TEMP TMP_OUT;\n");
4168     if (reg_maps->fog)
4169         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4170     if (need_helper_const(shader_data, reg_maps, gl_info))
4171     {
4172         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4173     }
4174     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4175     {
4176         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4177         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4178     }
4179
4180     shader_addline(buffer, "TEMP TA;\n");
4181     shader_addline(buffer, "TEMP TB;\n");
4182
4183     /* Base Declarations */
4184     next_local = shader_generate_arb_declarations(shader, reg_maps, buffer,
4185             gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4186
4187     for(i = 0; i < MAX_CONST_I; i++)
4188     {
4189         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4190         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4191         {
4192             const DWORD *control_values = find_loop_control_values(shader, i);
4193
4194             if(control_values)
4195             {
4196                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4197                                 control_values[0], control_values[1], control_values[2]);
4198             }
4199             else
4200             {
4201                 compiled->int_consts[i] = next_local;
4202                 compiled->num_int_consts++;
4203                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4204             }
4205         }
4206     }
4207
4208     /* We need a constant to fixup the final position */
4209     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4210     compiled->pos_fixup = next_local++;
4211
4212     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4213      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4214      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4215      * a replacement shader depend on the texcoord.w being set properly.
4216      *
4217      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4218      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4219      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4220      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4221      * this can eat a number of instructions, so skip it unless this cap is set as well
4222      */
4223     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4224     {
4225         struct wined3d_device *device = shader->device;
4226         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4227         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4228
4229         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4230         {
4231             int i;
4232             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4233             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4234             {
4235                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4236                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4237             }
4238         }
4239     }
4240
4241     /* The shader starts with the main function */
4242     priv_ctx.in_main_func = TRUE;
4243     /* Base Shader Body */
4244     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4245
4246     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4247             shader_data, args, reg_maps, gl_info, buffer);
4248
4249     shader_addline(buffer, "END\n");
4250
4251     /* TODO: change to resource.glObjectHandle or something like that */
4252     GL_EXTCALL(glGenProgramsARB(1, &ret));
4253
4254     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4255     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4256
4257     TRACE("Created hw vertex shader, prg=%d\n", ret);
4258     /* Create the program and check for errors */
4259     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4260                buffer->bsize, buffer->buffer));
4261     checkGLcall("glProgramStringARB()");
4262
4263     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4264     if (errPos != -1)
4265     {
4266         FIXME("HW VertexShader Error at position %d: %s\n\n",
4267               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4268         shader_arb_dump_program_source(buffer->buffer);
4269         ret = -1;
4270     }
4271     else
4272     {
4273         GLint native;
4274
4275         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4276         checkGLcall("glGetProgramivARB()");
4277         if (!native) WARN("Program exceeds native resource limits.\n");
4278
4279         /* Load immediate constants */
4280         if (lconst_map)
4281         {
4282             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
4283             {
4284                 const float *value = (const float *)lconst->value;
4285                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4286             }
4287         }
4288     }
4289     HeapFree(GetProcessHeap(), 0, lconst_map);
4290
4291     return ret;
4292 }
4293
4294 /* GL locking is done by the caller */
4295 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4296         const struct arb_ps_compile_args *args)
4297 {
4298     struct wined3d_device *device = shader->device;
4299     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4300     UINT i;
4301     DWORD new_size;
4302     struct arb_ps_compiled_shader *new_array;
4303     struct wined3d_shader_buffer buffer;
4304     struct arb_pshader_private *shader_data;
4305     GLuint ret;
4306
4307     if (!shader->backend_data)
4308     {
4309         struct shader_arb_priv *priv = device->shader_priv;
4310
4311         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4312         shader_data = shader->backend_data;
4313         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4314
4315         if (shader->reg_maps.shader_version.major < 3)
4316             shader_data->input_signature_idx = ~0;
4317         else
4318             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4319
4320         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4321
4322         if (!device->vs_clipping)
4323             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4324                     gl_info->limits.texture_stages - 1);
4325         else
4326             shader_data->clipplane_emulation = ~0U;
4327     }
4328     shader_data = shader->backend_data;
4329
4330     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4331      * so a linear search is more performant than a hashmap or a binary search
4332      * (cache coherency etc)
4333      */
4334     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4335     {
4336         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4337             return &shader_data->gl_shaders[i];
4338     }
4339
4340     TRACE("No matching GL shader found, compiling a new shader\n");
4341     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4342         if (shader_data->num_gl_shaders)
4343         {
4344             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4345             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4346                                     new_size * sizeof(*shader_data->gl_shaders));
4347         } else {
4348             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4349             new_size = 1;
4350         }
4351
4352         if(!new_array) {
4353             ERR("Out of memory\n");
4354             return 0;
4355         }
4356         shader_data->gl_shaders = new_array;
4357         shader_data->shader_array_size = new_size;
4358     }
4359
4360     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4361
4362     pixelshader_update_samplers(&shader->reg_maps, device->stateBlock->state.textures);
4363
4364     if (!shader_buffer_init(&buffer))
4365     {
4366         ERR("Failed to initialize shader buffer.\n");
4367         return 0;
4368     }
4369
4370     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4371             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4372     shader_buffer_free(&buffer);
4373     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4374
4375     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4376 }
4377
4378 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4379                                  const DWORD use_map, BOOL skip_int) {
4380     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4381     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4382     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4383     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4384     if(stored->ps_signature != new->ps_signature) return FALSE;
4385     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4386     if(skip_int) return TRUE;
4387
4388     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4389 }
4390
4391 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4392         const struct arb_vs_compile_args *args)
4393 {
4394     struct wined3d_device *device = shader->device;
4395     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4396     DWORD use_map = device->strided_streams.use_map;
4397     UINT i;
4398     DWORD new_size;
4399     struct arb_vs_compiled_shader *new_array;
4400     struct wined3d_shader_buffer buffer;
4401     struct arb_vshader_private *shader_data;
4402     GLuint ret;
4403
4404     if (!shader->backend_data)
4405     {
4406         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4407
4408         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4409         shader_data = shader->backend_data;
4410
4411         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4412                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4413         {
4414             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4415             {
4416                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4417                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4418                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4419             }
4420             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4421                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4422             else if (reg_maps->max_rel_offset > 63)
4423                 shader_data->rel_offset = reg_maps->min_rel_offset;
4424         }
4425     }
4426     shader_data = shader->backend_data;
4427
4428     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4429      * so a linear search is more performant than a hashmap or a binary search
4430      * (cache coherency etc)
4431      */
4432     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4433         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4434                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4435         {
4436             return &shader_data->gl_shaders[i];
4437         }
4438     }
4439
4440     TRACE("No matching GL shader found, compiling a new shader\n");
4441
4442     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4443         if (shader_data->num_gl_shaders)
4444         {
4445             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4446             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4447                                     new_size * sizeof(*shader_data->gl_shaders));
4448         } else {
4449             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4450             new_size = 1;
4451         }
4452
4453         if(!new_array) {
4454             ERR("Out of memory\n");
4455             return 0;
4456         }
4457         shader_data->gl_shaders = new_array;
4458         shader_data->shader_array_size = new_size;
4459     }
4460
4461     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4462
4463     if (!shader_buffer_init(&buffer))
4464     {
4465         ERR("Failed to initialize shader buffer.\n");
4466         return 0;
4467     }
4468
4469     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4470             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4471     shader_buffer_free(&buffer);
4472     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4473
4474     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4475 }
4476
4477 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4478         const struct wined3d_shader *shader, struct arb_ps_compile_args *args)
4479 {
4480     struct wined3d_device *device = shader->device;
4481     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4482     int i;
4483     WORD int_skip;
4484
4485     find_ps_compile_args(state, shader, &args->super);
4486
4487     /* This forces all local boolean constants to 1 to make them stateblock independent */
4488     args->bools = shader->reg_maps.local_bool_consts;
4489
4490     for(i = 0; i < MAX_CONST_B; i++)
4491     {
4492         if (state->ps_consts_b[i])
4493             args->bools |= ( 1 << i);
4494     }
4495
4496     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4497      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4498      * duplicate the shader than have a no-op KIL instruction in every shader
4499      */
4500     if (!device->vs_clipping && use_vs(state)
4501             && state->render_states[WINED3D_RS_CLIPPING]
4502             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4503         args->clip = 1;
4504     else
4505         args->clip = 0;
4506
4507     /* Skip if unused or local, or supported natively */
4508     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4509     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4510     {
4511         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4512         return;
4513     }
4514
4515     for(i = 0; i < MAX_CONST_I; i++)
4516     {
4517         if(int_skip & (1 << i))
4518         {
4519             args->loop_ctrl[i][0] = 0;
4520             args->loop_ctrl[i][1] = 0;
4521             args->loop_ctrl[i][2] = 0;
4522         }
4523         else
4524         {
4525             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4526             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4527             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4528         }
4529     }
4530 }
4531
4532 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4533         const struct wined3d_shader *shader, struct arb_vs_compile_args *args)
4534 {
4535     struct wined3d_device *device = shader->device;
4536     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4537     int i;
4538     WORD int_skip;
4539
4540     find_vs_compile_args(state, shader, &args->super);
4541
4542     args->clip.boolclip_compare = 0;
4543     if (use_ps(state))
4544     {
4545         const struct wined3d_shader *ps = state->pixel_shader;
4546         const struct arb_pshader_private *shader_priv = ps->backend_data;
4547         args->ps_signature = shader_priv->input_signature_idx;
4548
4549         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4550     }
4551     else
4552     {
4553         args->ps_signature = ~0;
4554         if (!device->vs_clipping && device->adapter->fragment_pipe == &arbfp_fragment_pipeline)
4555         {
4556             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4557         }
4558         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4559     }
4560
4561     if (args->clip.boolclip.clip_texcoord)
4562     {
4563         if (state->render_states[WINED3D_RS_CLIPPING])
4564             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4565         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4566     }
4567
4568     /* This forces all local boolean constants to 1 to make them stateblock independent */
4569     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4570     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4571     for(i = 0; i < MAX_CONST_B; i++)
4572     {
4573         if (state->vs_consts_b[i])
4574             args->clip.boolclip.bools |= ( 1 << i);
4575     }
4576
4577     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4578     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4579     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4580     args->vertex.samplers[3] = 0;
4581
4582     /* Skip if unused or local */
4583     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4584     /* This is about flow control, not clipping. */
4585     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4586     {
4587         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4588         return;
4589     }
4590
4591     for(i = 0; i < MAX_CONST_I; i++)
4592     {
4593         if(int_skip & (1 << i))
4594         {
4595             args->loop_ctrl[i][0] = 0;
4596             args->loop_ctrl[i][1] = 0;
4597             args->loop_ctrl[i][2] = 0;
4598         }
4599         else
4600         {
4601             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4602             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4603             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4604         }
4605     }
4606 }
4607
4608 /* GL locking is done by the caller */
4609 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4610 {
4611     struct wined3d_device *device = context->swapchain->device;
4612     struct shader_arb_priv *priv = device->shader_priv;
4613     const struct wined3d_gl_info *gl_info = context->gl_info;
4614     const struct wined3d_state *state = &device->stateBlock->state;
4615     int i;
4616
4617     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4618     if (usePS)
4619     {
4620         struct wined3d_shader *ps = state->pixel_shader;
4621         struct arb_ps_compile_args compile_args;
4622         struct arb_ps_compiled_shader *compiled;
4623
4624         TRACE("Using pixel shader %p.\n", ps);
4625         find_arb_ps_compile_args(state, ps, &compile_args);
4626         compiled = find_arb_pshader(ps, &compile_args);
4627         priv->current_fprogram_id = compiled->prgId;
4628         priv->compiled_fprog = compiled;
4629
4630         /* Bind the fragment program */
4631         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4632         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4633
4634         if(!priv->use_arbfp_fixed_func) {
4635             /* Enable OpenGL fragment programs */
4636             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4637             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4638         }
4639         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n",
4640                 device, priv->current_fprogram_id);
4641
4642         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4643          * a 1.x and newer shader, reload the first 8 constants
4644          */
4645         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4646         {
4647             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4648             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4649             for(i = 0; i < 8; i++)
4650             {
4651                 priv->pshader_const_dirty[i] = 1;
4652             }
4653             /* Also takes care of loading local constants */
4654             shader_arb_load_constants(context, TRUE, FALSE);
4655         }
4656         else
4657         {
4658             UINT rt_height = state->fb->render_targets[0]->resource.height;
4659             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4660         }
4661
4662         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4663         if (compiled->np2fixup_info.super.active)
4664             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4665     }
4666     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4667     {
4668         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4669         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4670         * replacement shader
4671         */
4672         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4673         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4674         priv->current_fprogram_id = 0;
4675     }
4676
4677     if (useVS)
4678     {
4679         struct wined3d_shader *vs = state->vertex_shader;
4680         struct arb_vs_compile_args compile_args;
4681         struct arb_vs_compiled_shader *compiled;
4682
4683         TRACE("Using vertex shader %p\n", vs);
4684         find_arb_vs_compile_args(state, vs, &compile_args);
4685         compiled = find_arb_vshader(vs, &compile_args);
4686         priv->current_vprogram_id = compiled->prgId;
4687         priv->compiled_vprog = compiled;
4688
4689         /* Bind the vertex program */
4690         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4691         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4692
4693         /* Enable OpenGL vertex programs */
4694         glEnable(GL_VERTEX_PROGRAM_ARB);
4695         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4696         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", device, priv->current_vprogram_id);
4697         shader_arb_vs_local_constants(compiled, context, state);
4698
4699         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4700             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4701
4702             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4703             {
4704                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4705                 checkGLcall("glClampColorARB");
4706             } else {
4707                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4708             }
4709         }
4710     }
4711     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4712     {
4713         priv->current_vprogram_id = 0;
4714         glDisable(GL_VERTEX_PROGRAM_ARB);
4715         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4716     }
4717 }
4718
4719 /* GL locking is done by the caller */
4720 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4721         enum tex_types tex_type, const SIZE *ds_mask_size)
4722 {
4723     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4724     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4725     struct shader_arb_priv *priv = shader_priv;
4726     GLuint *blt_fprogram;
4727
4728     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4729     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4730     glEnable(GL_VERTEX_PROGRAM_ARB);
4731
4732     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4733     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4734     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4735     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4736     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4737 }
4738
4739 /* GL locking is done by the caller */
4740 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4741 {
4742     struct shader_arb_priv *priv = shader_priv;
4743
4744     if (priv->current_vprogram_id) {
4745         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4746         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4747
4748         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4749     }
4750     else
4751     {
4752         glDisable(GL_VERTEX_PROGRAM_ARB);
4753         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4754     }
4755
4756     if (priv->current_fprogram_id) {
4757         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4758         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4759
4760         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4761     }
4762     else if(!priv->use_arbfp_fixed_func)
4763     {
4764         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4765         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4766     }
4767 }
4768
4769 static void shader_arb_destroy(struct wined3d_shader *shader)
4770 {
4771     struct wined3d_device *device = shader->device;
4772     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4773
4774     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4775     {
4776         struct arb_pshader_private *shader_data = shader->backend_data;
4777         UINT i;
4778
4779         if(!shader_data) return; /* This can happen if a shader was never compiled */
4780
4781         if (shader_data->num_gl_shaders)
4782         {
4783             struct wined3d_context *context = context_acquire(device, NULL);
4784
4785             ENTER_GL();
4786             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4787             {
4788                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4789                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4790             }
4791             LEAVE_GL();
4792
4793             context_release(context);
4794         }
4795
4796         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4797         HeapFree(GetProcessHeap(), 0, shader_data);
4798         shader->backend_data = NULL;
4799     }
4800     else
4801     {
4802         struct arb_vshader_private *shader_data = shader->backend_data;
4803         UINT i;
4804
4805         if(!shader_data) return; /* This can happen if a shader was never compiled */
4806
4807         if (shader_data->num_gl_shaders)
4808         {
4809             struct wined3d_context *context = context_acquire(device, NULL);
4810
4811             ENTER_GL();
4812             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4813             {
4814                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4815                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4816             }
4817             LEAVE_GL();
4818
4819             context_release(context);
4820         }
4821
4822         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4823         HeapFree(GetProcessHeap(), 0, shader_data);
4824         shader->backend_data = NULL;
4825     }
4826 }
4827
4828 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4829 {
4830     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4831     return compare_sig(key, e->sig);
4832 }
4833
4834 static const struct wine_rb_functions sig_tree_functions =
4835 {
4836     wined3d_rb_alloc,
4837     wined3d_rb_realloc,
4838     wined3d_rb_free,
4839     sig_tree_compare
4840 };
4841
4842 static HRESULT shader_arb_alloc(struct wined3d_device *device)
4843 {
4844     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4845
4846     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4847             sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4848     if (!priv->vshader_const_dirty)
4849         goto fail;
4850     memset(priv->vshader_const_dirty, 1,
4851            sizeof(*priv->vshader_const_dirty) * device->d3d_vshader_constantF);
4852
4853     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4854             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4855     if (!priv->pshader_const_dirty)
4856         goto fail;
4857     memset(priv->pshader_const_dirty, 1,
4858             sizeof(*priv->pshader_const_dirty) * device->d3d_pshader_constantF);
4859
4860     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4861     {
4862         ERR("RB tree init failed\n");
4863         goto fail;
4864     }
4865     device->shader_priv = priv;
4866     return WINED3D_OK;
4867
4868 fail:
4869     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4870     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4871     HeapFree(GetProcessHeap(), 0, priv);
4872     return E_OUTOFMEMORY;
4873 }
4874
4875 static void release_signature(struct wine_rb_entry *entry, void *context)
4876 {
4877     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4878     int i;
4879     for(i = 0; i < MAX_REG_INPUT; i++)
4880     {
4881         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4882     }
4883     HeapFree(GetProcessHeap(), 0, sig->sig);
4884     HeapFree(GetProcessHeap(), 0, sig);
4885 }
4886
4887 /* Context activation is done by the caller. */
4888 static void shader_arb_free(struct wined3d_device *device)
4889 {
4890     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4891     struct shader_arb_priv *priv = device->shader_priv;
4892     int i;
4893
4894     ENTER_GL();
4895     if(priv->depth_blt_vprogram_id) {
4896         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4897     }
4898     for (i = 0; i < tex_type_count; ++i)
4899     {
4900         if (priv->depth_blt_fprogram_id_full[i])
4901         {
4902             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4903         }
4904         if (priv->depth_blt_fprogram_id_masked[i])
4905         {
4906             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4907         }
4908     }
4909     LEAVE_GL();
4910
4911     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4912     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4913     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4914     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4915 }
4916
4917 static void shader_arb_context_destroyed(void *shader_priv, const struct wined3d_context *context)
4918 {
4919     struct shader_arb_priv *priv = shader_priv;
4920
4921     if (priv->last_context == context)
4922         priv->last_context = NULL;
4923 }
4924
4925 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4926 {
4927     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4928     {
4929         DWORD vs_consts;
4930
4931         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4932          * for vertex programs. If the native limit is less than that it's
4933          * not very useful, and e.g. Mesa swrast returns 0, probably to
4934          * indicate it's a software implementation. */
4935         if (gl_info->limits.arb_vs_native_constants < 96)
4936             vs_consts = gl_info->limits.arb_vs_float_constants;
4937         else
4938             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4939
4940         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4941         {
4942             caps->VertexShaderVersion = 3;
4943             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4944         }
4945         else if (vs_consts >= 256)
4946         {
4947             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4948             caps->VertexShaderVersion = 2;
4949             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4950         }
4951         else
4952         {
4953             caps->VertexShaderVersion = 1;
4954             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4955         }
4956         caps->MaxVertexShaderConst = vs_consts;
4957     }
4958     else
4959     {
4960         caps->VertexShaderVersion = 0;
4961         caps->MaxVertexShaderConst = 0;
4962     }
4963
4964     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4965     {
4966         DWORD ps_consts;
4967
4968         /* Similar as above for vertex programs, but the minimum for fragment
4969          * programs is 24. */
4970         if (gl_info->limits.arb_ps_native_constants < 24)
4971             ps_consts = gl_info->limits.arb_ps_float_constants;
4972         else
4973             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4974
4975         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4976         {
4977             caps->PixelShaderVersion = 3;
4978             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4979         }
4980         else if (ps_consts >= 32)
4981         {
4982             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4983             caps->PixelShaderVersion = 2;
4984             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4985         }
4986         else
4987         {
4988             caps->PixelShaderVersion = 1;
4989             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4990         }
4991         caps->PixelShader1xMaxValue = 8.0f;
4992         caps->MaxPixelShaderConst = ps_consts;
4993     }
4994     else
4995     {
4996         caps->PixelShaderVersion = 0;
4997         caps->PixelShader1xMaxValue = 0.0f;
4998         caps->MaxPixelShaderConst = 0;
4999     }
5000
5001     caps->VSClipping = use_nv_clip(gl_info);
5002 }
5003
5004 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5005 {
5006     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5007     {
5008         TRACE("Checking support for color_fixup:\n");
5009         dump_color_fixup_desc(fixup);
5010     }
5011
5012     /* We support everything except complex conversions. */
5013     if (!is_complex_fixup(fixup))
5014     {
5015         TRACE("[OK]\n");
5016         return TRUE;
5017     }
5018
5019     TRACE("[FAILED]\n");
5020     return FALSE;
5021 }
5022
5023 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5024     DWORD shift;
5025     char write_mask[20], regstr[50];
5026     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5027     BOOL is_color = FALSE;
5028     const struct wined3d_shader_dst_param *dst;
5029
5030     if (!ins->dst_count) return;
5031
5032     dst = &ins->dst[0];
5033     shift = dst->shift;
5034     if (!shift) return; /* Saturate alone is handled by the instructions */
5035
5036     shader_arb_get_write_mask(ins, dst, write_mask);
5037     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5038
5039     /* Generate a line that does the output modifier computation
5040      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5041      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5042      */
5043     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5044                    regstr, write_mask, regstr, shift_tab[shift]);
5045 }
5046
5047 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5048 {
5049     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
5050     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
5051     /* WINED3DSIH_AND           */ NULL,
5052     /* WINED3DSIH_BEM           */ pshader_hw_bem,
5053     /* WINED3DSIH_BREAK         */ shader_hw_break,
5054     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
5055     /* WINED3DSIH_BREAKP        */ NULL,
5056     /* WINED3DSIH_CALL          */ shader_hw_call,
5057     /* WINED3DSIH_CALLNZ        */ NULL,
5058     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
5059     /* WINED3DSIH_CND           */ pshader_hw_cnd,
5060     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
5061     /* WINED3DSIH_CUT           */ NULL,
5062     /* WINED3DSIH_DCL           */ NULL,
5063     /* WINED3DSIH_DEF           */ NULL,
5064     /* WINED3DSIH_DEFB          */ NULL,
5065     /* WINED3DSIH_DEFI          */ NULL,
5066     /* WINED3DSIH_DIV           */ NULL,
5067     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
5068     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
5069     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
5070     /* WINED3DSIH_DST           */ shader_hw_map2gl,
5071     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
5072     /* WINED3DSIH_DSY           */ shader_hw_dsy,
5073     /* WINED3DSIH_ELSE          */ shader_hw_else,
5074     /* WINED3DSIH_EMIT          */ NULL,
5075     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
5076     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
5077     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
5078     /* WINED3DSIH_EQ            */ NULL,
5079     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
5080     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
5081     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
5082     /* WINED3DSIH_FTOI          */ NULL,
5083     /* WINED3DSIH_GE            */ NULL,
5084     /* WINED3DSIH_IADD          */ NULL,
5085     /* WINED3DSIH_IEQ           */ NULL,
5086     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
5087     /* WINED3DSIH_IFC           */ shader_hw_ifc,
5088     /* WINED3DSIH_IGE           */ NULL,
5089     /* WINED3DSIH_IMUL          */ NULL,
5090     /* WINED3DSIH_ITOF          */ NULL,
5091     /* WINED3DSIH_LABEL         */ shader_hw_label,
5092     /* WINED3DSIH_LD            */ NULL,
5093     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5094     /* WINED3DSIH_LOG           */ shader_hw_log,
5095     /* WINED3DSIH_LOGP          */ shader_hw_log,
5096     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5097     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5098     /* WINED3DSIH_LT            */ NULL,
5099     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5100     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5101     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5102     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5103     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5104     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5105     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5106     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5107     /* WINED3DSIH_MOV           */ shader_hw_mov,
5108     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5109     /* WINED3DSIH_MOVC          */ NULL,
5110     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5111     /* WINED3DSIH_NOP           */ shader_hw_nop,
5112     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5113     /* WINED3DSIH_PHASE         */ NULL,
5114     /* WINED3DSIH_POW           */ shader_hw_pow,
5115     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5116     /* WINED3DSIH_REP           */ shader_hw_rep,
5117     /* WINED3DSIH_RET           */ shader_hw_ret,
5118     /* WINED3DSIH_ROUND_NI      */ NULL,
5119     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5120     /* WINED3DSIH_SAMPLE        */ NULL,
5121     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5122     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5123     /* WINED3DSIH_SETP          */ NULL,
5124     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5125     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5126     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5127     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5128     /* WINED3DSIH_SQRT          */ NULL,
5129     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5130     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5131     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5132     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5133     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5134     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5135     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5136     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5137     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5138     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5139     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5140     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5141     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5142     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5143     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5144     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5145     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5146     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5147     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5148     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5149     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5150     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5151     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5152     /* WINED3DSIH_UDIV          */ NULL,
5153     /* WINED3DSIH_USHR          */ NULL,
5154     /* WINED3DSIH_UTOF          */ NULL,
5155     /* WINED3DSIH_XOR           */ NULL,
5156 };
5157
5158 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5159         const struct wined3d_shader *shader, DWORD idx)
5160 {
5161     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5162     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5163     const struct wined3d_shader_lconst *constant;
5164     WORD bools = 0;
5165     WORD flag = (1 << idx);
5166     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5167
5168     if (reg_maps->local_bool_consts & flag)
5169     {
5170         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5171         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5172         {
5173             if (constant->idx == idx)
5174             {
5175                 return constant->value[0];
5176             }
5177         }
5178         ERR("Local constant not found\n");
5179         return FALSE;
5180     }
5181     else
5182     {
5183         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5184         else bools = priv->cur_ps_args->bools;
5185         return bools & flag;
5186     }
5187 }
5188
5189 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5190         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5191 {
5192     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5193     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5194
5195     /* Integer constants can either be a local constant, or they can be stored in the shader
5196      * type specific compile args. */
5197     if (reg_maps->local_int_consts & (1 << idx))
5198     {
5199         const struct wined3d_shader_lconst *constant;
5200
5201         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5202         {
5203             if (constant->idx == idx)
5204             {
5205                 loop_control->count = constant->value[0];
5206                 loop_control->start = constant->value[1];
5207                 /* Step is signed. */
5208                 loop_control->step = (int)constant->value[2];
5209                 return;
5210             }
5211         }
5212         /* If this happens the flag was set incorrectly */
5213         ERR("Local constant not found\n");
5214         loop_control->count = 0;
5215         loop_control->start = 0;
5216         loop_control->step = 0;
5217         return;
5218     }
5219
5220     switch (reg_maps->shader_version.type)
5221     {
5222         case WINED3D_SHADER_TYPE_VERTEX:
5223             /* Count and aL start value are unsigned */
5224             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5225             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5226             /* Step is signed. */
5227             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5228             break;
5229
5230         case WINED3D_SHADER_TYPE_PIXEL:
5231             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5232             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5233             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5234             break;
5235
5236         default:
5237             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5238             break;
5239     }
5240 }
5241
5242 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5243 {
5244     unsigned int i;
5245     struct wined3d_shader_dst_param *dst_param = NULL;
5246     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5247     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5248     if(!rec)
5249     {
5250         ERR("Out of memory\n");
5251         return;
5252     }
5253
5254     rec->ins = *ins;
5255     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5256     if(!dst_param) goto free;
5257     *dst_param = *ins->dst;
5258     if(ins->dst->reg.rel_addr)
5259     {
5260         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5261         if(!rel_addr) goto free;
5262         *rel_addr = *ins->dst->reg.rel_addr;
5263         dst_param->reg.rel_addr = rel_addr;
5264     }
5265     rec->ins.dst = dst_param;
5266
5267     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5268     if(!src_param) goto free;
5269     for(i = 0; i < ins->src_count; i++)
5270     {
5271         src_param[i] = ins->src[i];
5272         if(ins->src[i].reg.rel_addr)
5273         {
5274             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5275             if(!rel_addr) goto free;
5276             *rel_addr = *ins->src[i].reg.rel_addr;
5277             src_param[i].reg.rel_addr = rel_addr;
5278         }
5279     }
5280     rec->ins.src = src_param;
5281     list_add_tail(list, &rec->entry);
5282     return;
5283
5284 free:
5285     ERR("Out of memory\n");
5286     if(dst_param)
5287     {
5288         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5289         HeapFree(GetProcessHeap(), 0, dst_param);
5290     }
5291     if(src_param)
5292     {
5293         for(i = 0; i < ins->src_count; i++)
5294         {
5295             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5296         }
5297         HeapFree(GetProcessHeap(), 0, src_param);
5298     }
5299     HeapFree(GetProcessHeap(), 0, rec);
5300 }
5301
5302 static void free_recorded_instruction(struct list *list)
5303 {
5304     struct recorded_instruction *rec_ins, *entry2;
5305     unsigned int i;
5306
5307     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5308     {
5309         list_remove(&rec_ins->entry);
5310         if(rec_ins->ins.dst)
5311         {
5312             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5313             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5314         }
5315         if(rec_ins->ins.src)
5316         {
5317             for(i = 0; i < rec_ins->ins.src_count; i++)
5318             {
5319                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5320             }
5321             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5322         }
5323         HeapFree(GetProcessHeap(), 0, rec_ins);
5324     }
5325 }
5326
5327 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5328     SHADER_HANDLER hw_fct;
5329     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5330     const struct wined3d_shader *shader = ins->ctx->shader;
5331     struct control_frame *control_frame;
5332     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5333     BOOL bool_const;
5334
5335     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5336     {
5337         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5338         list_add_head(&priv->control_frames, &control_frame->entry);
5339
5340         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5341         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5342
5343         if(priv->target_version >= NV2)
5344         {
5345             control_frame->no.loop = priv->num_loops++;
5346             priv->loop_depth++;
5347         }
5348         else
5349         {
5350             /* Don't bother recording when we're in a not used if branch */
5351             if(priv->muted)
5352             {
5353                 return;
5354             }
5355
5356             if(!priv->recording)
5357             {
5358                 list_init(&priv->record);
5359                 priv->recording = TRUE;
5360                 control_frame->outer_loop = TRUE;
5361                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5362                 return; /* Instruction is handled */
5363             }
5364             /* Record this loop in the outer loop's recording */
5365         }
5366     }
5367     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5368     {
5369         if(priv->target_version >= NV2)
5370         {
5371             /* Nothing to do. The control frame is popped after the HW instr handler */
5372         }
5373         else
5374         {
5375             struct list *e = list_head(&priv->control_frames);
5376             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5377             list_remove(&control_frame->entry);
5378
5379             if(control_frame->outer_loop)
5380             {
5381                 unsigned int iteration;
5382                 int aL = 0;
5383                 struct list copy;
5384
5385                 /* Turn off recording before playback */
5386                 priv->recording = FALSE;
5387
5388                 /* Move the recorded instructions to a separate list and get them out of the private data
5389                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5390                  * be recorded again, thus priv->record might be overwritten
5391                  */
5392                 list_init(&copy);
5393                 list_move_tail(&copy, &priv->record);
5394                 list_init(&priv->record);
5395
5396                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5397                 {
5398                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5399                                    control_frame->loop_control.count, control_frame->loop_control.start,
5400                                    control_frame->loop_control.step);
5401                     aL = control_frame->loop_control.start;
5402                 }
5403                 else
5404                 {
5405                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5406                 }
5407
5408                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5409                 {
5410                     struct recorded_instruction *rec_ins;
5411                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5412                     {
5413                         priv->aL = aL;
5414                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5415                     }
5416                     else
5417                     {
5418                         shader_addline(buffer, "#Iteration %u\n", iteration);
5419                     }
5420
5421                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5422                     {
5423                         shader_arb_handle_instruction(&rec_ins->ins);
5424                     }
5425
5426                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5427                     {
5428                         aL += control_frame->loop_control.step;
5429                     }
5430                 }
5431                 shader_addline(buffer, "#end loop/rep\n");
5432
5433                 free_recorded_instruction(&copy);
5434                 HeapFree(GetProcessHeap(), 0, control_frame);
5435                 return; /* Instruction is handled */
5436             }
5437             else
5438             {
5439                 /* This is a nested loop. Proceed to the normal recording function */
5440                 HeapFree(GetProcessHeap(), 0, control_frame);
5441             }
5442         }
5443     }
5444
5445     if(priv->recording)
5446     {
5447         record_instruction(&priv->record, ins);
5448         return;
5449     }
5450
5451     /* boolean if */
5452     if(ins->handler_idx == WINED3DSIH_IF)
5453     {
5454         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5455         list_add_head(&priv->control_frames, &control_frame->entry);
5456         control_frame->type = IF;
5457
5458         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5459         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5460         if (!priv->muted && !bool_const)
5461         {
5462             shader_addline(buffer, "#if(FALSE){\n");
5463             priv->muted = TRUE;
5464             control_frame->muting = TRUE;
5465         }
5466         else shader_addline(buffer, "#if(TRUE) {\n");
5467
5468         return; /* Instruction is handled */
5469     }
5470     else if(ins->handler_idx == WINED3DSIH_IFC)
5471     {
5472         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5473         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5474         control_frame->type = IFC;
5475         control_frame->no.ifc = priv->num_ifcs++;
5476         list_add_head(&priv->control_frames, &control_frame->entry);
5477     }
5478     else if(ins->handler_idx == WINED3DSIH_ELSE)
5479     {
5480         struct list *e = list_head(&priv->control_frames);
5481         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5482
5483         if(control_frame->type == IF)
5484         {
5485             shader_addline(buffer, "#} else {\n");
5486             if(!priv->muted && !control_frame->muting)
5487             {
5488                 priv->muted = TRUE;
5489                 control_frame->muting = TRUE;
5490             }
5491             else if(control_frame->muting) priv->muted = FALSE;
5492             return; /* Instruction is handled. */
5493         }
5494         /* In case of an ifc, generate a HW shader instruction */
5495     }
5496     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5497     {
5498         struct list *e = list_head(&priv->control_frames);
5499         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5500
5501         if(control_frame->type == IF)
5502         {
5503             shader_addline(buffer, "#} endif\n");
5504             if(control_frame->muting) priv->muted = FALSE;
5505             list_remove(&control_frame->entry);
5506             HeapFree(GetProcessHeap(), 0, control_frame);
5507             return; /* Instruction is handled */
5508         }
5509     }
5510
5511     if(priv->muted) return;
5512
5513     /* Select handler */
5514     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5515
5516     /* Unhandled opcode */
5517     if (!hw_fct)
5518     {
5519         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5520         return;
5521     }
5522     hw_fct(ins);
5523
5524     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5525     {
5526         struct list *e = list_head(&priv->control_frames);
5527         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5528         list_remove(&control_frame->entry);
5529         HeapFree(GetProcessHeap(), 0, control_frame);
5530         priv->loop_depth--;
5531     }
5532     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5533     {
5534         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5535         struct list *e = list_head(&priv->control_frames);
5536         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5537         list_remove(&control_frame->entry);
5538         HeapFree(GetProcessHeap(), 0, control_frame);
5539     }
5540
5541
5542     shader_arb_add_instruction_modifiers(ins);
5543 }
5544
5545 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5546 {
5547     shader_arb_handle_instruction,
5548     shader_arb_select,
5549     shader_arb_select_depth_blt,
5550     shader_arb_deselect_depth_blt,
5551     shader_arb_update_float_vertex_constants,
5552     shader_arb_update_float_pixel_constants,
5553     shader_arb_load_constants,
5554     shader_arb_load_np2fixup_constants,
5555     shader_arb_destroy,
5556     shader_arb_alloc,
5557     shader_arb_free,
5558     shader_arb_context_destroyed,
5559     shader_arb_get_caps,
5560     shader_arb_color_fixup_supported,
5561 };
5562
5563 /* ARB_fragment_program fixed function pipeline replacement definitions */
5564 #define ARB_FFP_CONST_TFACTOR           0
5565 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5566 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5567 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5568 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5569
5570 struct arbfp_ffp_desc
5571 {
5572     struct ffp_frag_desc parent;
5573     GLuint shader;
5574     unsigned int num_textures_used;
5575 };
5576
5577 /* Context activation and GL locking are done by the caller. */
5578 static void arbfp_enable(BOOL enable)
5579 {
5580     if(enable) {
5581         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5582         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5583     } else {
5584         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5585         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5586     }
5587 }
5588
5589 static HRESULT arbfp_alloc(struct wined3d_device *device)
5590 {
5591     struct shader_arb_priv *priv;
5592     /* Share private data between the shader backend and the pipeline replacement, if both
5593      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5594      * if no pixel shader is bound or not
5595      */
5596     if (device->shader_backend == &arb_program_shader_backend)
5597     {
5598         device->fragment_priv = device->shader_priv;
5599     }
5600     else
5601     {
5602         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5603         if (!device->fragment_priv) return E_OUTOFMEMORY;
5604     }
5605     priv = device->fragment_priv;
5606     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5607     {
5608         ERR("Failed to initialize rbtree.\n");
5609         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5610         return E_OUTOFMEMORY;
5611     }
5612     priv->use_arbfp_fixed_func = TRUE;
5613     return WINED3D_OK;
5614 }
5615
5616 /* Context activation is done by the caller. */
5617 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5618 {
5619     const struct wined3d_gl_info *gl_info = context;
5620     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5621
5622     ENTER_GL();
5623     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5624     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5625     HeapFree(GetProcessHeap(), 0, entry_arb);
5626     LEAVE_GL();
5627 }
5628
5629 /* Context activation is done by the caller. */
5630 static void arbfp_free(struct wined3d_device *device)
5631 {
5632     struct shader_arb_priv *priv = device->fragment_priv;
5633
5634     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5635     priv->use_arbfp_fixed_func = FALSE;
5636
5637     if (device->shader_backend != &arb_program_shader_backend)
5638     {
5639         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5640     }
5641 }
5642
5643 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5644 {
5645     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5646     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5647                            WINED3DTEXOPCAPS_SELECTARG1                  |
5648                            WINED3DTEXOPCAPS_SELECTARG2                  |
5649                            WINED3DTEXOPCAPS_MODULATE4X                  |
5650                            WINED3DTEXOPCAPS_MODULATE2X                  |
5651                            WINED3DTEXOPCAPS_MODULATE                    |
5652                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5653                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5654                            WINED3DTEXOPCAPS_ADD                         |
5655                            WINED3DTEXOPCAPS_SUBTRACT                    |
5656                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5657                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5658                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5659                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5660                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5661                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5662                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5663                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5664                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5665                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5666                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5667                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5668                            WINED3DTEXOPCAPS_LERP                        |
5669                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5670                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5671
5672     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5673
5674     caps->MaxTextureBlendStages   = 8;
5675     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5676 }
5677
5678 static void state_texfactor_arbfp(struct wined3d_context *context,
5679         const struct wined3d_state *state, DWORD state_id)
5680 {
5681     struct wined3d_device *device = context->swapchain->device;
5682     const struct wined3d_gl_info *gl_info = context->gl_info;
5683     float col[4];
5684
5685     /* Don't load the parameter if we're using an arbfp pixel shader,
5686      * otherwise we'll overwrite application provided constants. */
5687     if (device->shader_backend == &arb_program_shader_backend)
5688     {
5689         struct shader_arb_priv *priv;
5690
5691         if (use_ps(state)) return;
5692
5693         priv = device->shader_priv;
5694         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5695         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5696     }
5697
5698     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5699     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5700     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5701 }
5702
5703 static void state_arb_specularenable(struct wined3d_context *context,
5704         const struct wined3d_state *state, DWORD state_id)
5705 {
5706     struct wined3d_device *device = context->swapchain->device;
5707     const struct wined3d_gl_info *gl_info = context->gl_info;
5708     float col[4];
5709
5710     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5711      * application provided constants
5712      */
5713     if (device->shader_backend == &arb_program_shader_backend)
5714     {
5715         struct shader_arb_priv *priv;
5716
5717         if (use_ps(state)) return;
5718
5719         priv = device->shader_priv;
5720         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5721         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5722     }
5723
5724     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5725     {
5726         /* The specular color has no alpha */
5727         col[0] = 1.0f; col[1] = 1.0f;
5728         col[2] = 1.0f; col[3] = 0.0f;
5729     } else {
5730         col[0] = 0.0f; col[1] = 0.0f;
5731         col[2] = 0.0f; col[3] = 0.0f;
5732     }
5733     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5734     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5735 }
5736
5737 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5738 {
5739     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5740     struct wined3d_device *device = context->swapchain->device;
5741     const struct wined3d_gl_info *gl_info = context->gl_info;
5742     float mat[2][2];
5743
5744     if (use_ps(state))
5745     {
5746         if (stage && (state->pixel_shader->reg_maps.bumpmat & (1 << stage)))
5747         {
5748             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5749              * anyway
5750              */
5751             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5752                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5753         }
5754
5755         if(device->shader_backend == &arb_program_shader_backend) {
5756             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5757             return;
5758         }
5759     }
5760     else if (device->shader_backend == &arb_program_shader_backend)
5761     {
5762         struct shader_arb_priv *priv = device->shader_priv;
5763         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5764         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5765     }
5766
5767     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5768     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5769     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5770     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5771
5772     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5773     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5774 }
5775
5776 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5777         const struct wined3d_state *state, DWORD state_id)
5778 {
5779     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5780     struct wined3d_device *device = context->swapchain->device;
5781     const struct wined3d_gl_info *gl_info = context->gl_info;
5782     float param[4];
5783
5784     if (use_ps(state))
5785     {
5786         if (stage && (state->pixel_shader->reg_maps.luminanceparams & (1 << stage)))
5787         {
5788             /* The pixel shader has to know the luminance offset. Do a constants update if it
5789              * isn't scheduled anyway
5790              */
5791             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5792                 context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
5793         }
5794
5795         if(device->shader_backend == &arb_program_shader_backend) {
5796             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5797             return;
5798         }
5799     }
5800     else if (device->shader_backend == &arb_program_shader_backend)
5801     {
5802         struct shader_arb_priv *priv = device->shader_priv;
5803         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5804         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5805     }
5806
5807     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5808     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5809     param[2] = 0.0f;
5810     param[3] = 0.0f;
5811
5812     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5813     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5814 }
5815
5816 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5817 {
5818     const char *ret;
5819
5820     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5821
5822     switch(arg & WINED3DTA_SELECTMASK) {
5823         case WINED3DTA_DIFFUSE:
5824             ret = "fragment.color.primary"; break;
5825
5826         case WINED3DTA_CURRENT:
5827             if (!stage) ret = "fragment.color.primary";
5828             else ret = "ret";
5829             break;
5830
5831         case WINED3DTA_TEXTURE:
5832             switch(stage) {
5833                 case 0: ret = "tex0"; break;
5834                 case 1: ret = "tex1"; break;
5835                 case 2: ret = "tex2"; break;
5836                 case 3: ret = "tex3"; break;
5837                 case 4: ret = "tex4"; break;
5838                 case 5: ret = "tex5"; break;
5839                 case 6: ret = "tex6"; break;
5840                 case 7: ret = "tex7"; break;
5841                 default: ret = "unknown texture";
5842             }
5843             break;
5844
5845         case WINED3DTA_TFACTOR:
5846             ret = "tfactor"; break;
5847
5848         case WINED3DTA_SPECULAR:
5849             ret = "fragment.color.secondary"; break;
5850
5851         case WINED3DTA_TEMP:
5852             ret = "tempreg"; break;
5853
5854         case WINED3DTA_CONSTANT:
5855             FIXME("Implement perstage constants\n");
5856             switch(stage) {
5857                 case 0: ret = "const0"; break;
5858                 case 1: ret = "const1"; break;
5859                 case 2: ret = "const2"; break;
5860                 case 3: ret = "const3"; break;
5861                 case 4: ret = "const4"; break;
5862                 case 5: ret = "const5"; break;
5863                 case 6: ret = "const6"; break;
5864                 case 7: ret = "const7"; break;
5865                 default: ret = "unknown constant";
5866             }
5867             break;
5868
5869         default:
5870             return "unknown";
5871     }
5872
5873     if(arg & WINED3DTA_COMPLEMENT) {
5874         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5875         if(argnum == 0) ret = "arg0";
5876         if(argnum == 1) ret = "arg1";
5877         if(argnum == 2) ret = "arg2";
5878     }
5879     if(arg & WINED3DTA_ALPHAREPLICATE) {
5880         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5881         if(argnum == 0) ret = "arg0";
5882         if(argnum == 1) ret = "arg1";
5883         if(argnum == 2) ret = "arg2";
5884     }
5885     return ret;
5886 }
5887
5888 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5889         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5890 {
5891     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5892     unsigned int mul = 1;
5893     BOOL mul_final_dest = FALSE;
5894
5895     if(color && alpha) dstmask = "";
5896     else if(color) dstmask = ".xyz";
5897     else dstmask = ".w";
5898
5899     if(dst == tempreg) dstreg = "tempreg";
5900     else dstreg = "ret";
5901
5902     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5903     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5904     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5905
5906     switch (op)
5907     {
5908         case WINED3D_TOP_DISABLE:
5909             if (!stage)
5910                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5911             break;
5912
5913         case WINED3D_TOP_SELECT_ARG2:
5914             arg1 = arg2;
5915             /* FALLTHROUGH */
5916         case WINED3D_TOP_SELECT_ARG1:
5917             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5918             break;
5919
5920         case WINED3D_TOP_MODULATE_4X:
5921             mul = 2;
5922             /* FALLTHROUGH */
5923         case WINED3D_TOP_MODULATE_2X:
5924             mul *= 2;
5925             if (!strcmp(dstreg, "result.color"))
5926             {
5927                 dstreg = "ret";
5928                 mul_final_dest = TRUE;
5929             }
5930             /* FALLTHROUGH */
5931         case WINED3D_TOP_MODULATE:
5932             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5933             break;
5934
5935         case WINED3D_TOP_ADD_SIGNED_2X:
5936             mul = 2;
5937             if (!strcmp(dstreg, "result.color"))
5938             {
5939                 dstreg = "ret";
5940                 mul_final_dest = TRUE;
5941             }
5942             /* FALLTHROUGH */
5943         case WINED3D_TOP_ADD_SIGNED:
5944             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5945             arg2 = "arg2";
5946             /* FALLTHROUGH */
5947         case WINED3D_TOP_ADD:
5948             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5949             break;
5950
5951         case WINED3D_TOP_SUBTRACT:
5952             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5953             break;
5954
5955         case WINED3D_TOP_ADD_SMOOTH:
5956             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5957             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5958             break;
5959
5960         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
5961             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5962             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5963             break;
5964         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
5965             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5966             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5967             break;
5968         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
5969             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5970             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5971             break;
5972         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
5973             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5974             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5975             break;
5976
5977         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
5978             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5979             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5980             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5981             break;
5982
5983         /* D3DTOP_PREMODULATE ???? */
5984
5985         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
5986             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5987             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5988             break;
5989         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
5990             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5991             break;
5992         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
5993             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5994             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5995             break;
5996         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
5997             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5998             break;
5999
6000         case WINED3D_TOP_DOTPRODUCT3:
6001             mul = 4;
6002             if (!strcmp(dstreg, "result.color"))
6003             {
6004                 dstreg = "ret";
6005                 mul_final_dest = TRUE;
6006             }
6007             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6008             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6009             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6010             break;
6011
6012         case WINED3D_TOP_MULTIPLY_ADD:
6013             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6014             break;
6015
6016         case WINED3D_TOP_LERP:
6017             /* The msdn is not quite right here */
6018             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6019             break;
6020
6021         case WINED3D_TOP_BUMPENVMAP:
6022         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6023             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6024             break;
6025
6026         default:
6027             FIXME("Unhandled texture op %08x\n", op);
6028     }
6029
6030     if(mul == 2) {
6031         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6032     } else if(mul == 4) {
6033         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
6034     }
6035 }
6036
6037 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6038 {
6039     unsigned int stage;
6040     struct wined3d_shader_buffer buffer;
6041     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6042     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6043     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6044     UINT lowest_disabled_stage;
6045     const char *textype;
6046     const char *instr, *sat;
6047     char colorcor_dst[8];
6048     GLuint ret;
6049     DWORD arg0, arg1, arg2;
6050     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6051     BOOL op_equal;
6052     const char *final_combiner_src = "ret";
6053     GLint pos;
6054
6055     /* Find out which textures are read */
6056     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6057     {
6058         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6059             break;
6060         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6061         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6062         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6063         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6064         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6065         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6066
6067         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6068             tex_read[stage] = TRUE;
6069         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6070             tex_read[stage] = TRUE;
6071         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6072         {
6073             bump_used[stage] = TRUE;
6074             tex_read[stage] = TRUE;
6075         }
6076         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6077         {
6078             bump_used[stage] = TRUE;
6079             tex_read[stage] = TRUE;
6080             luminance_used[stage] = TRUE;
6081         }
6082         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6083         {
6084             tfactor_used = TRUE;
6085         }
6086
6087         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6088             tfactor_used = TRUE;
6089         }
6090
6091         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6092         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6093             tempreg_used = TRUE;
6094         }
6095
6096         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6097             continue;
6098         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6099         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6100         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6101         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6102         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6103         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6104
6105         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6106             tempreg_used = TRUE;
6107         }
6108         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6109             tfactor_used = TRUE;
6110         }
6111     }
6112     lowest_disabled_stage = stage;
6113
6114     /* Shader header */
6115     if (!shader_buffer_init(&buffer))
6116     {
6117         ERR("Failed to initialize shader buffer.\n");
6118         return 0;
6119     }
6120
6121     shader_addline(&buffer, "!!ARBfp1.0\n");
6122
6123     switch(settings->fog) {
6124         case FOG_OFF:                                                         break;
6125         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6126         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6127         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6128         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6129     }
6130
6131     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6132     shader_addline(&buffer, "TEMP TMP;\n");
6133     shader_addline(&buffer, "TEMP ret;\n");
6134     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6135     shader_addline(&buffer, "TEMP arg0;\n");
6136     shader_addline(&buffer, "TEMP arg1;\n");
6137     shader_addline(&buffer, "TEMP arg2;\n");
6138     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6139         if(!tex_read[stage]) continue;
6140         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6141         if(!bump_used[stage]) continue;
6142         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6143         if(!luminance_used[stage]) continue;
6144         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6145     }
6146     if(tfactor_used) {
6147         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6148     }
6149         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6150
6151     if(settings->sRGB_write) {
6152         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6153                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6154         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6155                        srgb_sub_high, 0.0, 0.0, 0.0);
6156     }
6157
6158     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6159         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6160
6161     /* Generate texture sampling instructions) */
6162     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6163     {
6164         if (!tex_read[stage])
6165             continue;
6166
6167         switch(settings->op[stage].tex_type) {
6168             case tex_1d:                    textype = "1D";     break;
6169             case tex_2d:                    textype = "2D";     break;
6170             case tex_3d:                    textype = "3D";     break;
6171             case tex_cube:                  textype = "CUBE";   break;
6172             case tex_rect:                  textype = "RECT";   break;
6173             default: textype = "unexpected_textype";   break;
6174         }
6175
6176         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP
6177                 || settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6178             sat = "";
6179         else
6180             sat = "_SAT";
6181
6182         if(settings->op[stage].projected == proj_none) {
6183             instr = "TEX";
6184         } else if(settings->op[stage].projected == proj_count4 ||
6185                   settings->op[stage].projected == proj_count3) {
6186             instr = "TXP";
6187         } else {
6188             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6189             instr = "TXP";
6190         }
6191
6192         if (stage > 0
6193                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6194                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6195         {
6196             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6197             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6198             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6199             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6200
6201             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6202              * so multiply the displacement with the dividing parameter before passing it to TXP
6203              */
6204             if (settings->op[stage].projected != proj_none) {
6205                 if(settings->op[stage].projected == proj_count4) {
6206                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6207                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6208                 } else {
6209                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6210                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6211                 }
6212             } else {
6213                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6214             }
6215
6216             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6217                     instr, sat, stage, stage, textype);
6218             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6219             {
6220                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6221                                stage - 1, stage - 1, stage - 1);
6222                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6223             }
6224         } else if(settings->op[stage].projected == proj_count3) {
6225             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6226             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6227             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6228                             instr, sat, stage, stage, textype);
6229         } else {
6230             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6231                             instr, sat, stage, stage, stage, textype);
6232         }
6233
6234         sprintf(colorcor_dst, "tex%u", stage);
6235         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6236                 settings->op[stage].color_fixup);
6237     }
6238
6239     /* Generate the main shader */
6240     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6241     {
6242         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6243         {
6244             if (!stage)
6245                 final_combiner_src = "fragment.color.primary";
6246             break;
6247         }
6248
6249         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6250                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6251             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6252         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6253                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6254             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6255         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6256                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6257             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6258         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6259                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6260             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6261         else
6262             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6263                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6264                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6265                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6266
6267         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6268         {
6269             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6270                           settings->op[stage].cop, settings->op[stage].carg0,
6271                           settings->op[stage].carg1, settings->op[stage].carg2);
6272             if (!stage)
6273                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6274         }
6275         else if (op_equal)
6276         {
6277             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6278                           settings->op[stage].cop, settings->op[stage].carg0,
6279                           settings->op[stage].carg1, settings->op[stage].carg2);
6280         } else {
6281             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6282                           settings->op[stage].cop, settings->op[stage].carg0,
6283                           settings->op[stage].carg1, settings->op[stage].carg2);
6284             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6285                           settings->op[stage].aop, settings->op[stage].aarg0,
6286                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6287         }
6288     }
6289
6290     if(settings->sRGB_write) {
6291         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6292         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6293         shader_addline(&buffer, "MOV result.color, ret;\n");
6294     } else {
6295         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6296     }
6297
6298     /* Footer */
6299     shader_addline(&buffer, "END\n");
6300
6301     /* Generate the shader */
6302     GL_EXTCALL(glGenProgramsARB(1, &ret));
6303     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6304     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6305             strlen(buffer.buffer), buffer.buffer));
6306     checkGLcall("glProgramStringARB()");
6307
6308     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6309     if (pos != -1)
6310     {
6311         FIXME("Fragment program error at position %d: %s\n\n", pos,
6312               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6313         shader_arb_dump_program_source(buffer.buffer);
6314     }
6315     else
6316     {
6317         GLint native;
6318
6319         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6320         checkGLcall("glGetProgramivARB()");
6321         if (!native) WARN("Program exceeds native resource limits.\n");
6322     }
6323
6324     shader_buffer_free(&buffer);
6325     return ret;
6326 }
6327
6328 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6329 {
6330     const struct wined3d_device *device = context->swapchain->device;
6331     const struct wined3d_gl_info *gl_info = context->gl_info;
6332     struct shader_arb_priv *priv = device->fragment_priv;
6333     BOOL use_vshader = use_vs(state);
6334     BOOL use_pshader = use_ps(state);
6335     struct ffp_frag_settings settings;
6336     const struct arbfp_ffp_desc *desc;
6337     unsigned int i;
6338
6339     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6340
6341     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6342     {
6343         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6344         {
6345             /* Reload fixed function constants since they collide with the
6346              * pixel shader constants. */
6347             for (i = 0; i < MAX_TEXTURES; ++i)
6348             {
6349                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6350             }
6351             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6352             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6353         }
6354         else if (use_pshader && !isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6355         {
6356             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6357         }
6358         return;
6359     }
6360
6361     if (!use_pshader)
6362     {
6363         /* Find or create a shader implementing the fixed function pipeline
6364          * settings, then activate it. */
6365         gen_ffp_frag_op(device, state, &settings, FALSE);
6366         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6367         if(!desc) {
6368             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6369             if (!new_desc)
6370             {
6371                 ERR("Out of memory\n");
6372                 return;
6373             }
6374             new_desc->num_textures_used = 0;
6375             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6376             {
6377                 if (settings.op[i].cop == WINED3D_TOP_DISABLE)
6378                     break;
6379                 new_desc->num_textures_used = i;
6380             }
6381
6382             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6383             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6384             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6385             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6386             desc = new_desc;
6387         }
6388
6389         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6390          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6391          * deactivate it.
6392          */
6393         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6394         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6395         priv->current_fprogram_id = desc->shader;
6396
6397         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6398         {
6399             /* Reload fixed function constants since they collide with the
6400              * pixel shader constants. */
6401             for (i = 0; i < MAX_TEXTURES; ++i)
6402             {
6403                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6404             }
6405             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6406             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6407         }
6408         context->last_was_pshader = FALSE;
6409     } else {
6410         context->last_was_pshader = TRUE;
6411     }
6412
6413     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6414      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6415      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6416      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6417      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6418      *
6419      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6420      * shader handler
6421      */
6422     if (!isStateDirty(context, context->state_table[STATE_VSHADER].representative))
6423     {
6424         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6425
6426         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6427             context_apply_state(context, state, STATE_VERTEXSHADERCONSTANT);
6428     }
6429     if (use_pshader)
6430         context_apply_state(context, state, STATE_PIXELSHADERCONSTANT);
6431 }
6432
6433 /* We can't link the fog states to the fragment state directly since the
6434  * vertex pipeline links them to FOGENABLE. A different linking in different
6435  * pipeline parts can't be expressed in the combined state table, so we need
6436  * to handle that with a forwarding function. The other invisible side effect
6437  * is that changing the fog start and fog end (which links to FOGENABLE in
6438  * vertex) results in the fragment_prog_arbfp function being called because
6439  * FOGENABLE is dirty, which calls this function here. */
6440 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6441 {
6442     enum fogsource new_source;
6443
6444     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6445
6446     if (!isStateDirty(context, STATE_PIXELSHADER))
6447         fragment_prog_arbfp(context, state, state_id);
6448
6449     if (!state->render_states[WINED3D_RS_FOGENABLE])
6450         return;
6451
6452     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6453     {
6454         if (use_vs(state))
6455         {
6456             new_source = FOGSOURCE_VS;
6457         }
6458         else
6459         {
6460             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6461                 new_source = FOGSOURCE_COORD;
6462             else
6463                 new_source = FOGSOURCE_FFP;
6464         }
6465     }
6466     else
6467     {
6468         new_source = FOGSOURCE_FFP;
6469     }
6470
6471     if (new_source != context->fog_source)
6472     {
6473         context->fog_source = new_source;
6474         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6475     }
6476 }
6477
6478 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6479 {
6480     if (!isStateDirty(context, STATE_PIXELSHADER))
6481         fragment_prog_arbfp(context, state, state_id);
6482 }
6483
6484 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6485 {
6486     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6487     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6488     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6489     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6490     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6491     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6492     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6493     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6494     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6495     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6496     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6497     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6498     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6499     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6500     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6501     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6502     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6503     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6504     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6505     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6506     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6507     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6508     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6509     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6510     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6511     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6512     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6513     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6514     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6515     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6516     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6517     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6518     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6519     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6520     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6521     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6522     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6523     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6524     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6525     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6526     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6527     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6528     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6529     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6530     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6531     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6532     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6533     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6534     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6535     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6536     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6537     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6538     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6539     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6540     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6541     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6542     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6543     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6544     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6545     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6546     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6547     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6548     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6549     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6550     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6551     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6552     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6553     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6554     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6555     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6556     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6557     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6558     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6559     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6560     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6561     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6562     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6563     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6564     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6565     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6566     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6567     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6568     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6569     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6570     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6571     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6572     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6573     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6574     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6575     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6576     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6577     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6578     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6579     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6580     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6581     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6582     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6583     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6584     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6585     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6586     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6587     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6588     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6589     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6590     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6591     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6592     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6593     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6594     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6595     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6596     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6597     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6598     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6599     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6600     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6601     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6602     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6603     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6604     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6605     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6606     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6607     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6608     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6609     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6610     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6611     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6612     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6613     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6614     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6615     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6624     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6625     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6626 };
6627
6628 const struct fragment_pipeline arbfp_fragment_pipeline = {
6629     arbfp_enable,
6630     arbfp_get_caps,
6631     arbfp_alloc,
6632     arbfp_free,
6633     shader_arb_color_fixup_supported,
6634     arbfp_fragmentstate_template,
6635     TRUE /* We can disable projected textures */
6636 };
6637
6638 struct arbfp_blit_priv {
6639     GLenum yuy2_rect_shader, yuy2_2d_shader;
6640     GLenum uyvy_rect_shader, uyvy_2d_shader;
6641     GLenum yv12_rect_shader, yv12_2d_shader;
6642     GLenum p8_rect_shader, p8_2d_shader;
6643     GLuint palette_texture;
6644 };
6645
6646 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6647 {
6648     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6649     if(!device->blit_priv) {
6650         ERR("Out of memory\n");
6651         return E_OUTOFMEMORY;
6652     }
6653     return WINED3D_OK;
6654 }
6655
6656 /* Context activation is done by the caller. */
6657 static void arbfp_blit_free(struct wined3d_device *device)
6658 {
6659     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6660     struct arbfp_blit_priv *priv = device->blit_priv;
6661
6662     ENTER_GL();
6663     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6664     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6665     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6666     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6667     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6668     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6669     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6670     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6671     checkGLcall("Delete yuv and p8 programs");
6672
6673     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6674     LEAVE_GL();
6675
6676     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6677     device->blit_priv = NULL;
6678 }
6679
6680 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6681         GLenum textype, char *luminance)
6682 {
6683     char chroma;
6684     const char *tex, *texinstr;
6685
6686     if (fixup == COMPLEX_FIXUP_UYVY) {
6687         chroma = 'x';
6688         *luminance = 'w';
6689     } else {
6690         chroma = 'w';
6691         *luminance = 'x';
6692     }
6693     switch(textype) {
6694         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6695         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6696         default:
6697             /* This is more tricky than just replacing the texture type - we have to navigate
6698              * properly in the texture to find the correct chroma values
6699              */
6700             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6701             return FALSE;
6702     }
6703
6704     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6705      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6706      * filtering when we sample the texture.
6707      *
6708      * These are the rules for reading the chroma:
6709      *
6710      * Even pixel: Cr
6711      * Even pixel: U
6712      * Odd pixel: V
6713      *
6714      * So we have to get the sampling x position in non-normalized coordinates in integers
6715      */
6716     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6717         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6718         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6719     } else {
6720         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6721     }
6722     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6723      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6724      * 0.5, so add 0.5.
6725      */
6726     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6727     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6728
6729     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6730      * even and odd pixels respectively
6731      */
6732     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6733     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6734
6735     /* Sample Pixel 1 */
6736     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6737
6738     /* Put the value into either of the chroma values */
6739     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6740     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6741     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6742     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6743
6744     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6745      * the pixel right to the current one. Otherwise, sample the left pixel.
6746      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6747      */
6748     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6749     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6750     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6751
6752     /* Put the value into the other chroma */
6753     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6754     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6755     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6756     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6757
6758     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6759      * the current one and lerp the two U and V values
6760      */
6761
6762     /* This gives the correctly filtered luminance value */
6763     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6764
6765     return TRUE;
6766 }
6767
6768 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6769 {
6770     const char *tex;
6771
6772     switch(textype) {
6773         case GL_TEXTURE_2D:             tex = "2D";     break;
6774         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6775         default:
6776             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6777             return FALSE;
6778     }
6779
6780     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6781      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6782      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6783      * pitch of the luminance plane, the packing into the gl texture is a bit
6784      * unfortunate. If the whole texture is interpreted as luminance data it looks
6785      * approximately like this:
6786      *
6787      *        +----------------------------------+----
6788      *        |                                  |
6789      *        |                                  |
6790      *        |                                  |
6791      *        |                                  |
6792      *        |                                  |   2
6793      *        |            LUMINANCE             |   -
6794      *        |                                  |   3
6795      *        |                                  |
6796      *        |                                  |
6797      *        |                                  |
6798      *        |                                  |
6799      *        +----------------+-----------------+----
6800      *        |                |                 |
6801      *        |  U even rows   |  U odd rows     |
6802      *        |                |                 |   1
6803      *        +----------------+------------------   -
6804      *        |                |                 |   3
6805      *        |  V even rows   |  V odd rows     |
6806      *        |                |                 |
6807      *        +----------------+-----------------+----
6808      *        |                |                 |
6809      *        |     0.5        |       0.5       |
6810      *
6811      * So it appears as if there are 4 chroma images, but in fact the odd rows
6812      * in the chroma images are in the same row as the even ones. So its is
6813      * kinda tricky to read
6814      *
6815      * When reading from rectangle textures, keep in mind that the input y coordinates
6816      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6817      */
6818     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6819             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6820
6821     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6822     /* the chroma planes have only half the width */
6823     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6824
6825     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6826      * the coordinate. Also read the right side of the image when reading odd lines
6827      *
6828      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6829      * bleeding
6830      */
6831     if(textype == GL_TEXTURE_2D) {
6832
6833         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6834
6835         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6836
6837         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6838         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6839
6840         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6841         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6842         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6843         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6844         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6845
6846         /* clamp, keep the half pixel origin in mind */
6847         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6848         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6849         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6850         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6851     } else {
6852         /* Read from [size - size+size/4] */
6853         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6854         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6855
6856         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6857         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6858         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6859         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6860         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6861         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6862
6863         /* Make sure to read exactly from the pixel center */
6864         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6865         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6866
6867         /* Clamp */
6868         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6869         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6870         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6871         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6872         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6873     }
6874     /* Read the texture, put the result into the output register */
6875     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6876     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6877
6878     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6879      * No need to clamp because we're just reusing the already clamped value from above
6880      */
6881     if(textype == GL_TEXTURE_2D) {
6882         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6883     } else {
6884         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6885     }
6886     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6887     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6888
6889     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6890      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6891      * values due to filtering
6892      */
6893     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6894     if(textype == GL_TEXTURE_2D) {
6895         /* Multiply the y coordinate by 2/3 and clamp it */
6896         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6897         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6898         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6899         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6900     } else {
6901         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6902          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6903          * is bigger
6904          */
6905         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6906         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6907         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6908     }
6909     *luminance = 'a';
6910
6911     return TRUE;
6912 }
6913
6914 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6915         const struct wined3d_gl_info *gl_info, GLenum textype)
6916 {
6917     GLenum shader;
6918     struct wined3d_shader_buffer buffer;
6919     GLint pos;
6920
6921     /* Shader header */
6922     if (!shader_buffer_init(&buffer))
6923     {
6924         ERR("Failed to initialize shader buffer.\n");
6925         return 0;
6926     }
6927
6928     ENTER_GL();
6929     GL_EXTCALL(glGenProgramsARB(1, &shader));
6930     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6931     LEAVE_GL();
6932     if(!shader) {
6933         shader_buffer_free(&buffer);
6934         return 0;
6935     }
6936
6937     shader_addline(&buffer, "!!ARBfp1.0\n");
6938     shader_addline(&buffer, "TEMP index;\n");
6939
6940     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6941     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6942
6943     /* The alpha-component contains the palette index */
6944     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6945         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6946     else
6947         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6948
6949     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6950     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6951
6952     /* Use the alpha-component as an index in the palette to get the final color */
6953     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6954     shader_addline(&buffer, "END\n");
6955
6956     ENTER_GL();
6957     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6958             strlen(buffer.buffer), buffer.buffer));
6959     checkGLcall("glProgramStringARB()");
6960
6961     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6962     if (pos != -1)
6963     {
6964         FIXME("Fragment program error at position %d: %s\n\n", pos,
6965               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6966         shader_arb_dump_program_source(buffer.buffer);
6967     }
6968
6969     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6970         priv->p8_rect_shader = shader;
6971     else
6972         priv->p8_2d_shader = shader;
6973
6974     shader_buffer_free(&buffer);
6975     LEAVE_GL();
6976
6977     return shader;
6978 }
6979
6980 /* Context activation is done by the caller. */
6981 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
6982 {
6983     BYTE table[256][4];
6984     struct wined3d_device *device = surface->resource.device;
6985     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6986     struct arbfp_blit_priv *priv = device->blit_priv;
6987     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6988
6989     d3dfmt_p8_init_palette(surface, table, colorkey);
6990
6991     ENTER_GL();
6992
6993     if (gl_info->supported[APPLE_CLIENT_STORAGE])
6994     {
6995         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE);
6996         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_FALSE)");
6997     }
6998
6999     if (!priv->palette_texture)
7000         glGenTextures(1, &priv->palette_texture);
7001
7002     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7003     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7004
7005     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7006
7007     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7008     /* Make sure we have discrete color levels. */
7009     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7010     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7011     /* Upload the palette */
7012     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7013     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7014
7015     if (gl_info->supported[APPLE_CLIENT_STORAGE])
7016     {
7017         glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE);
7018         checkGLcall("glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, GL_TRUE)");
7019     }
7020
7021     /* Switch back to unit 0 in which the 2D texture will be stored. */
7022     context_active_texture(context, gl_info, 0);
7023     LEAVE_GL();
7024 }
7025
7026 /* Context activation is done by the caller. */
7027 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7028         enum complex_fixup yuv_fixup, GLenum textype)
7029 {
7030     GLenum shader;
7031     struct wined3d_shader_buffer buffer;
7032     char luminance_component;
7033     GLint pos;
7034
7035     /* Shader header */
7036     if (!shader_buffer_init(&buffer))
7037     {
7038         ERR("Failed to initialize shader buffer.\n");
7039         return 0;
7040     }
7041
7042     ENTER_GL();
7043     GL_EXTCALL(glGenProgramsARB(1, &shader));
7044     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7045     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7046     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7047     LEAVE_GL();
7048     if(!shader) {
7049         shader_buffer_free(&buffer);
7050         return 0;
7051     }
7052
7053     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7054      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7055      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7056      * each single pixel it contains, and one U and one V value shared between both
7057      * pixels.
7058      *
7059      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7060      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7061      * take the format into account when generating the read swizzles
7062      *
7063      * Reading the Y value is straightforward - just sample the texture. The hardware
7064      * takes care of filtering in the horizontal and vertical direction.
7065      *
7066      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7067      * because that would mix the U and V values of one pixel or two adjacent pixels.
7068      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7069      * regardless of the filtering setting. Vertical filtering works automatically
7070      * though - the U and V values of two rows are mixed nicely.
7071      *
7072      * Appart of avoiding filtering issues, the code has to know which value it just
7073      * read, and where it can find the other one. To determine this, it checks if
7074      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7075      *
7076      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7077      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7078      *
7079      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7080      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7081      * in an unfiltered situation. Finding the luminance on the other hand requires
7082      * finding out if it is an odd or even pixel. The real drawback of this approach
7083      * is filtering. This would have to be emulated completely in the shader, reading
7084      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7085      * vertically. Beyond that it would require adjustments to the texture handling
7086      * code to deal with the width scaling
7087      */
7088     shader_addline(&buffer, "!!ARBfp1.0\n");
7089     shader_addline(&buffer, "TEMP luminance;\n");
7090     shader_addline(&buffer, "TEMP temp;\n");
7091     shader_addline(&buffer, "TEMP chroma;\n");
7092     shader_addline(&buffer, "TEMP texcrd;\n");
7093     shader_addline(&buffer, "TEMP texcrd2;\n");
7094     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7095     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7096     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7097
7098     switch (yuv_fixup)
7099     {
7100         case COMPLEX_FIXUP_UYVY:
7101         case COMPLEX_FIXUP_YUY2:
7102             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7103             {
7104                 shader_buffer_free(&buffer);
7105                 return 0;
7106             }
7107             break;
7108
7109         case COMPLEX_FIXUP_YV12:
7110             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7111             {
7112                 shader_buffer_free(&buffer);
7113                 return 0;
7114             }
7115             break;
7116
7117         default:
7118             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7119             shader_buffer_free(&buffer);
7120             return 0;
7121     }
7122
7123     /* Calculate the final result. Formula is taken from
7124      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7125      * ranges from -0.5 to 0.5
7126      */
7127     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7128
7129     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7130     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7131     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7132     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7133     shader_addline(&buffer, "END\n");
7134
7135     ENTER_GL();
7136     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7137             strlen(buffer.buffer), buffer.buffer));
7138     checkGLcall("glProgramStringARB()");
7139
7140     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7141     if (pos != -1)
7142     {
7143         FIXME("Fragment program error at position %d: %s\n\n", pos,
7144               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7145         shader_arb_dump_program_source(buffer.buffer);
7146     }
7147     else
7148     {
7149         GLint native;
7150
7151         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7152         checkGLcall("glGetProgramivARB()");
7153         if (!native) WARN("Program exceeds native resource limits.\n");
7154     }
7155
7156     shader_buffer_free(&buffer);
7157     LEAVE_GL();
7158
7159     switch (yuv_fixup)
7160     {
7161         case COMPLEX_FIXUP_YUY2:
7162             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7163             else priv->yuy2_2d_shader = shader;
7164             break;
7165
7166         case COMPLEX_FIXUP_UYVY:
7167             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7168             else priv->uyvy_2d_shader = shader;
7169             break;
7170
7171         case COMPLEX_FIXUP_YV12:
7172             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7173             else priv->yv12_2d_shader = shader;
7174             break;
7175         default:
7176             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7177     }
7178
7179     return shader;
7180 }
7181
7182 /* Context activation is done by the caller. */
7183 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7184 {
7185     GLenum shader;
7186     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7187     struct arbfp_blit_priv *priv = blit_priv;
7188     enum complex_fixup fixup;
7189     GLenum textype = surface->texture_target;
7190     const struct wined3d_gl_info *gl_info = context->gl_info;
7191
7192     if (surface->flags & SFLAG_CONVERTED)
7193     {
7194         ENTER_GL();
7195         glEnable(textype);
7196         checkGLcall("glEnable(textype)");
7197         LEAVE_GL();
7198         return WINED3D_OK;
7199     }
7200
7201     if (!is_complex_fixup(surface->resource.format->color_fixup))
7202     {
7203         TRACE("Fixup:\n");
7204         dump_color_fixup_desc(surface->resource.format->color_fixup);
7205         /* Don't bother setting up a shader for unconverted formats */
7206         ENTER_GL();
7207         glEnable(textype);
7208         checkGLcall("glEnable(textype)");
7209         LEAVE_GL();
7210         return WINED3D_OK;
7211     }
7212
7213     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7214
7215     switch(fixup)
7216     {
7217         case COMPLEX_FIXUP_YUY2:
7218             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7219             break;
7220
7221         case COMPLEX_FIXUP_UYVY:
7222             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7223             break;
7224
7225         case COMPLEX_FIXUP_YV12:
7226             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7227             break;
7228
7229         case COMPLEX_FIXUP_P8:
7230             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7231             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7232
7233             upload_palette(surface, context);
7234             break;
7235
7236         default:
7237             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7238             ENTER_GL();
7239             glEnable(textype);
7240             checkGLcall("glEnable(textype)");
7241             LEAVE_GL();
7242             return E_NOTIMPL;
7243     }
7244
7245     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7246
7247     ENTER_GL();
7248     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7249     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7250     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7251     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7252     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7253     checkGLcall("glProgramLocalParameter4fvARB");
7254     LEAVE_GL();
7255
7256     return WINED3D_OK;
7257 }
7258
7259 /* Context activation is done by the caller. */
7260 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7261 {
7262     ENTER_GL();
7263     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7264     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7265     glDisable(GL_TEXTURE_2D);
7266     checkGLcall("glDisable(GL_TEXTURE_2D)");
7267     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7268     {
7269         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7270         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7271     }
7272     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7273     {
7274         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7275         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7276     }
7277     LEAVE_GL();
7278 }
7279
7280 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7281         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7282         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7283 {
7284     enum complex_fixup src_fixup;
7285
7286     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7287         return FALSE;
7288
7289     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7290     {
7291         TRACE("Unsupported blit_op=%d\n", blit_op);
7292         return FALSE;
7293     }
7294
7295     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7296         return FALSE;
7297
7298     src_fixup = get_complex_fixup(src_format->color_fixup);
7299     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7300     {
7301         TRACE("Checking support for fixup:\n");
7302         dump_color_fixup_desc(src_format->color_fixup);
7303     }
7304
7305     if (!is_identity_fixup(dst_format->color_fixup))
7306     {
7307         TRACE("Destination fixups are not supported\n");
7308         return FALSE;
7309     }
7310
7311     if (is_identity_fixup(src_format->color_fixup))
7312     {
7313         TRACE("[OK]\n");
7314         return TRUE;
7315     }
7316
7317      /* We only support YUV conversions. */
7318     if (!is_complex_fixup(src_format->color_fixup))
7319     {
7320         TRACE("[FAILED]\n");
7321         return FALSE;
7322     }
7323
7324     switch(src_fixup)
7325     {
7326         case COMPLEX_FIXUP_YUY2:
7327         case COMPLEX_FIXUP_UYVY:
7328         case COMPLEX_FIXUP_YV12:
7329         case COMPLEX_FIXUP_P8:
7330             TRACE("[OK]\n");
7331             return TRUE;
7332
7333         default:
7334             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7335             TRACE("[FAILED]\n");
7336             return FALSE;
7337     }
7338 }
7339
7340 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7341         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7342         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7343 {
7344     struct wined3d_context *context;
7345     RECT src_rect = *src_rect_in;
7346     RECT dst_rect = *dst_rect_in;
7347
7348     /* Now load the surface */
7349     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7350             && (src_surface->flags & (SFLAG_INTEXTURE | SFLAG_INDRAWABLE)) == SFLAG_INDRAWABLE)
7351     {
7352         /* Without FBO blits transferring from the drawable to the texture is
7353          * expensive, because we have to flip the data in sysmem. Since we can
7354          * flip in the blitter, we don't actually need that flip anyway. So we
7355          * use the surface's texture as scratch texture, and flip the source
7356          * rectangle instead. */
7357         surface_load_fb_texture(src_surface, FALSE);
7358
7359         src_rect.top = src_surface->resource.height - src_rect.top;
7360         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7361     }
7362     else
7363         surface_internal_preload(src_surface, SRGB_RGB);
7364
7365     /* Activate the destination context, set it up for blitting */
7366     context = context_acquire(device, dst_surface);
7367     context_apply_blit_state(context, device);
7368
7369     if (!surface_is_offscreen(dst_surface))
7370         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7371
7372     arbfp_blit_set(device->blit_priv, context, src_surface);
7373
7374     ENTER_GL();
7375
7376     /* Draw a textured quad */
7377     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7378
7379     LEAVE_GL();
7380
7381     /* Leave the opengl state valid for blitting */
7382     arbfp_blit_unset(context->gl_info);
7383
7384     if (wined3d_settings.strict_draw_ordering
7385             || (dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7386             && (dst_surface->container.u.swapchain->front_buffer == dst_surface)))
7387         wglFlush(); /* Flush to ensure ordering across contexts. */
7388
7389     context_release(context);
7390
7391     surface_modify_location(dst_surface, dst_surface->draw_binding, TRUE);
7392     return WINED3D_OK;
7393 }
7394
7395 /* Do not call while under the GL lock. */
7396 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7397         const RECT *dst_rect, const struct wined3d_color *color)
7398 {
7399     FIXME("Color filling not implemented by arbfp_blit\n");
7400     return WINED3DERR_INVALIDCALL;
7401 }
7402
7403 /* Do not call while under the GL lock. */
7404 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7405         struct wined3d_surface *surface, const RECT *rect, float depth)
7406 {
7407     FIXME("Depth filling not implemented by arbfp_blit.\n");
7408     return WINED3DERR_INVALIDCALL;
7409 }
7410
7411 const struct blit_shader arbfp_blit = {
7412     arbfp_blit_alloc,
7413     arbfp_blit_free,
7414     arbfp_blit_set,
7415     arbfp_blit_unset,
7416     arbfp_blit_supported,
7417     arbfp_blit_color_fill,
7418     arbfp_blit_depth_fill,
7419 };